Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Process Control
Concepts, Dynamics and Applications
S.K. SINGH
Head
Maintenance Services Group (Electrical) and Telecommunication
Tata Steel Limited
Jamshedpur
New Delhi-110001
2009
PROCESS CONTROL: Concepts, Dynamics and Applications
S.K. Singh
© 2009 by PHI Learning Private Limited, New Delhi. All rights reserved. No part of this book may
be reproduced in any form, by mimeograph or any other means, without permission in writing from
the publisher.
ISBN-978-81-203-3678-0
The export rights of this book are vested solely with the publisher.
Published by Asoke K. Ghosh, PHI Learning Private Limited, M-97, Connaught Circus,
New Delhi-110001 and Printed by Jay Print Pack Private Limited, New Delhi-110015.
To
My wife MEENA,
children ABHISHEK, AVINASH and ALKA
and parents
for their love, understanding and support
Contents
H.M. Nerurkar
Chief Operating Officer
Tata Steel Limited
xvii
Preface
Intended Audience
· This book is designed to serve as a textbook for both four-year undergraduate (B.E.,
B.Tech) and two-year postgraduate (M.E., M.Tech.) students of mainly Chemical
engineering, but the information on process dynamics will be useful for mechanical,
instrumentation and electrical engineers who help design process control systems.
· It will also serve as a handbook for practising engineers and as a guide to research
and field personnel at all levels. It presents a thorough treatment of the principles
that govern the operation and control behaviour of industrial process.
xix
xx Contents
Preface
· This book will also serve as training material to various short-term and long-term
training courses of process control systems. A sincere attempt has been made to
cover the syllabus comprehensively that is taught in the courses mentioned earlier,
so as to meet everyones need in this single volume of the book.
Organization of the Book
The book is organized in the following five parts:
Part I: Process Control Concepts
Part II: Process Control Dynamics and Design
Part III: Advanced Process Control
Part IV: Computer-based Control
Part V: Case Studies
An attempt has been made to cover fundamental principles with clear ties to applications
and with guidelines on their putting into practice for implementing process control of
wide range of industrial processes.
The chapters and topics are organized in a systematic way in order that the
students get a thorough knowledge and understanding of the subject. As the name
suggests, this book covers the concepts, dynamics and applications of process control
systems.
Part I: Process Control Concepts (Chapters 1 through 3) First part of the book
provides a broad introduction to the concepts (including definition and classifications),
mathematical modelling and feedback controls of process control for a chemical process.
An attempt is made to define the process control, to identify the needs and incentives
for process control. Chapter 2 on modelling demonstrates how we can construct useful
models, starting from basic principles, and determines the scope and difficulties of
mathematical modelling for process control purposes. Chapter 3 provides a detailed
introduction to feedback control, developing the basic idea of a feedback system,
proportional, integral, derivative (PID) controllers, and methods of analyzing closed-loop
stability. It covers the analysis and design of feedback control systems which represent
the control schemes encounters most often in a chemical plant.
Part II: Process Control Dynamics and Design (Chapters 4 and 5) The second part of
the book introduces dynamic response analysis, stability criterion and design of process
control system. It covers specification of control system, controller design, real-time
process control software and design of more complex control systems. The emphasis is
on identifying those process characteristics which shape the dynamic response for a
variety of processing units. A detailed discussion has been provided to the subject of
controller tuning, root-locus techniques and frequency response analysis.
Part III: Advanced Process Control (Chapter 6) It explores the description, analysis
and design of more complex control systems aush as cascade control, feedforward
control, predictive and adaptive controls, statistical process control and intelligent
control.
Part IV: Computer-based Control (Chapters 7 through 11) Fourth part discusses the
basic concepts of computers for process control applications with their configurations,
Preface xxi
S.K. SINGH
Acknowledgements
I am indebted to my colleagues and friends for their help, moral support and
encouragement, in various ways, in completing this task.
I am thankful to the senior executives of Tata Steel for their encouragement
without which I would not have been able to complete this book.
I wish to acknowledge the assistance given by the editorial staff at PHI Learning,
New Delhi and their sustained interest in bringing this book in its present form.
Finally, I give immeasurable thanks to my familywife Meena and children Alka,
Avinash and Abhishekfor their sacrifices, patience, understanding and providing
encouragement during completion of the book. They endured many evenings and
weekends of solitude for the thrill of seeing a book cover hang on a den wall.
S.K. SINGH
xxiii
Part I
Introduction to Process
Control Systems
1.1 INTRODUCTION
The industrial revolution witnessed the development of process industries when human
physical power was substituted with machine. Early production processes were natural
scale-up versions of the traditional manual practices. They were designed as batch
process and later, replaced by continuous processes which resulted in economic and
technological benefits. There has been an increasing trends over the years towards
larger and more complex plants and processes to take advantage of the economies of
scale and technological developments.
The industrial process automation (or control) also came into the existence since
industries started. It is not a discovery of recent past, but is rather old as the industry
itself. Engineers and designers of process industries always tried to automate the
processes as much as possible, based on the available instruments. There has been
continuous development of industrial process which requires the development of
better instrumentation and automatic control. Conversely, the advancement in
instrumentation and control contributed to the development of larger and more complex
processes, bringing numerous new technologies and economic benefits. The industrial
process automation is now used for control of almost all kind of industries which include
chemical and petrochemical plants, paper pulp and paper mills, pharmaceuticals, foods
and beverages, water and sewage treatment plants, oil and gas fields, iron and steel,
other metal industries, etc.
devices, etc., each performing to achieve an overall objective of producing some product
or group of related products (chemical or physical). The performance of an industrial
process is influenced by internal and external conditions called process variables such
as temperature, pressure, flow, level, dimension, weight, volume, voltage, speed,
position, force, etc. The control of process variables is achieved by the control
equipment, e.g. electromagnetic valves, transformer tap positioners, etc. The process
control is, therefore, an engineering science of measuring one or more of these process
variables and controlling them to the desired level called set points or reference points
in spite of disturbances.
Chemical engineering views a process in the form of a process flow diagram (PFD)
as shown in Figure 1.1, which depicts the major plant equipment items and stream data
at the various points of interest in a process design candidate for a mixing process. The
stream data include the expected flow rates, temperatures and composition of material
and energy flow streams.
The construction engineers use the chemical engineers process design to develop
a detailed plant design. The design is represented by a set of schematics and other
drawings. The primary drawing used for construction is the piping and instrument (P&I)
diagram. Figures 1.2 and 1.3 show P&I diagram for a mixing and dryer processes
respectively with respect to PFD of Figure 1.1. This representation depicts the detailed
vessel configurations and the interconnected piping network. It also defines the control
valves and motor controls for the vessel and piping network. However, it is not to scale
and does not indicate how the equipment is installed. Sometimes the piping material
is shown on the P&I diagram. Traditionally, the P&I diagram defines the
instrumentation and controls for the process. However, only the basic single-loop
controls are shown.
When we talk of controlling a process we really mean controlling the state of a
process. One may wish to keep the process at its current state or may wish to change
the process from one state to another. The variables which uniquely define the state of
the process are called state variables and are traditionally represented by the vector x.
These are generally variables representing some accumulation of momentum, mass or
energy within the process. Let us consider a heating furnace, whose state is solely
determined by the accumulation of energy within the furnace. Hence the state vector
x is a one-dimensional vector representing the internal energy of the process.
x = (Internal energy)
The state of the process is affected by a number of inputs to the process. In control,
the inputs are defined as variables, which affect the state of a process, while in physical
terms they may be flows into or out of the process. In the example of heating furnace,
the energy in the furnace is affected by the power supplied to the heating elements, a
physical input, and also by the amount of heat loss to the surroundings, a physical
output. However, in a control engineering sense, these are both inputs affecting the
state of the process, the internal energy.
Process inputs are sometimes subdivided into those, which can be directly
manipulated, and those, which cannot. The former are known as manipulated variables
or sometimes simply as inputs and are represented by the vector u. The latter are
Flow
Material 11 Agitator
stream Flow 200
200
22
stream
Material
Mixing tank
Excess moisture
100 77
Exchanger Filter Material
Material
stream
stream
33 44 66
Material Material
stream 300 stream
400 Delumper
Filtrate 88
waste 500
500
656
Material
stream 600
Dryer
700
700
Product bin
Product
99
Definition of Industrial Processes and Control
FC Ratio
107
I1
1 EX
Agitator 106 2
Cascade
FC
I1
108 200
EX
LAH
1
TI 105
102
LAH
103
LC
Cascade
100
101 LSL
2
Mixing tank 103
Symbol legend
Discrete control XY
module 104
1 Material
stream
Regulatory control
module 23
3
Interlock cause
Interlock cause
Interlock action
Interlock action
known as disturbances or loads and are represented by the vector d. The relationship
between process components is shown in Figure 1.4.
In the case of heating furnace, the manipulated variable and disturbance vectors
are defined by
u = (Power) and d = (Heat loss)
FC
FC TI
TI 33
EX
EX
moisture
301
301 304
304 401
401 Excess
Excess Moisture
2 77
Material CWR Filter
stream
CWR Filter Material
TI
TI Stream
33 44 66 AC
AC
303
303
Material 501
501
Material Material
Material 55
300
300 stream EX
EX
CWR
CWR Stream Stream
stream
Exchanger 400 601
701
Exchanger 400 44
TI Filtrate
Filtrate 88
TI waste
302 Waste 500
500
302
66
Dryer
Dryer
Material
Material 4 Delumper
stream EX
EX
Stream 600
600
501
501 3
66
EX
EX
Product bin 700
700 701
701
Product Bin
Product Material
99 stream
Manipulated
variables u
Process States x
(e.g. a heating furnace)
Disturbances d
The relationship between these inputs and the states of the process can be
represented by a set of differential equations collectively known as the state equations:
FZ
H
Z WFV (1.1)
FV
where the scaler t represents time and f is some vector of linear or nonlinear functions.
These equations are mass, heat momentum balances, which define the rate of change
of the states, not the value of the states. This is an important distinction as it says that
processes are dynamic in nature, i.e. they do not react instantaneously to inputs.
The steady-state equations used for design calculations are
FZ
H
Z WFV (1.2)
FV
The actual state of the process is not always seen. Perhaps it is not required to
see the whole state or perhaps it is not possible or economic to measure the actual state.
We see the process outputs or measurements represented by the vector y, which are
related to one or more of the states. This relationship is represented by the output
equation:
y = g(x, u, d) (1.3)
In the case of the heating furnace the output variable, which we always think of
as representing the state is the temperature.
y = (Temperature)
This is a relationship representation of the internal energy given that the mass and
specific heat of the oven essentially remain constant.
Steam
supply
Steam
Steam
TTi i(t),
(t), ff(t)
(t) TT (t)
(t)
Process steam
Process steam Heated
Heatedsteam
steam
T
Condensate return
Condensate return
valve can be manipulated (by throttling and opening the steam flow) to correct the
deviation. The above can be done by manual control in which the operator has to
frequently monitor the temperature to take corrective action to maintain the desired
(set) value by opening or closing of steam flow valve. This type of manual control system
will be operator-dependent and differs from operator to operator. Also, it would be
extremely difficult for an operator to monitor hundreds of variables and take corrective
action to maintain the desired value. Therefore, this manual control can be replaced
with an automatic process control by designing and implementing a suitable control
system, as shown in Figure 1.7.
TC
T Temperature controller
TT
T Temperature transmitter
Ti(t), f (t)
TI(t), f(t) T (t)
T(t)
Process steam Heated steam
Heated Steam
T
Compensate return
Ti
Process fluid flow w
T
w
Heater
The process fluid flows into the tank from a pipe and flows out of the tank by
overflow. Thus, the volume of the tank is constant. The flow of heating fluid can be
changed by adjusting the opening of the valve in the heating medium line. The objective
of this stirred-tank heat exchanger is to keep the exit temperature T at its desired (set-
point) value when disturbances like change in inlet process fluid flow rate and
temperature, heating fluid temperature, pressure of the heating fluid upstream of the
valve, etc., occur. The automatic control can be achieved by measuring the outlet
temperature by putting a temperature sensor, comparing the value with the desired
(set-point) value and adjusting the heat input of the heater accordingly, as shown in
Figure 1.9.
TT TC
Ti
Temperature sensor
Q(t)
Heater
Fluid inlet
Fluid outlet
The process fluid flows into the tank from a pipe. A desired level of fluid is to be
maintained by opening or closing the outlet fluid valve. The objective of this tank is
to keep the fluid level at its desired (set point) value when disturbances like change in
inlet fluid flow rate and temperature, etc., occur. The automatic control can be achieved
by measuring the liquid level by putting a level sensor, comparing the value with
the desired (set point) value and adjusting the fluid outlet accordingly, as shown in
Figure 1.11.
Level sensor Level transmitter
LT
Fluid inlet
Level controller
LC Set-point (SP)
Fluid outlet
Error
e Final control
6 Amplifier Controller Process
element
Set-point
Set point
(SP)
(SP)
Feedback
Feedback Sensor /
Sensor/transducer
amplifier
amplifier transducer
Set-point
Set point
(SP)
(SP) e(t) p(t) Q(t) T(t)
S Controller Heater Stirred tank
Feedback
Sensor/transducer
to control all these variables at the same time. There are single-variable processes, in
which only one variable is to be controlled, as well as multi-variable processes, in which
many variables, perhaps interrelate, may require regulation.
Following terms are important in an automatic process control system:
(a) Controlled variable: Controlled variable or process variable is a variable that
must be controlled, or maintained at some desired value. The output
temperatures in the first-two examples of process control systems, discussed
earlier, and the liquid level in the third example are the controlled variables.
(b) Set-point (SP): Set-point or reference-point is desired value of the controlled
variable to be maintained at all times.
(c) Manipulated variable: It is a variable used to maintain the controlled variable
at its set point. It is the consequence of the manipulated variable. The valve
position in the above three examples are the manipulated variable.
(d) Disturbance: Any variable that causes variation in the controlled variable is
called disturbance or upset. There are number of disturbances in any process
loop. In the example of heat exchanger, inlet process temperature, inlet
process flow, the heat content, etc., are the disturbances.
Force,
displacement, Sensor output
Sensor Signal conditioner
Measured voltage, flow Transmitter
variable, C output Cout
flow, temperature devices, etc. But in some other cases, the transmitter output is
known as nonlinear function of the process variable such as thermocouples, orifice
flowmeters, etc.
There are three terms used frequently related to the sensor/transmitter
combination in an industrial process control systems:
(a) Range: The difference betwen the low and high values of the process variable,
that is measured, is called the range.
(b) Span: The difference between the high and low values of the range is called
the span.
(c) Zero: Zero is the low value of the range which does not have to be zero in
order to be called the zero of the instrument.
Consider a temperature sensor/transmitter that has been calibrated to measure a
process temperature between the values of 850°C to 1200°C. The range of this
sensor/transmitter combination is 850°C to 1200°C. The span of the instrument is 350°C
(1200°C850°C). The zero of the instrument in this example is 850°C.
The transfer function of the sensor/transmitter combination relates its output
signal to input signal, which is the process variable, as shown in Figure 1.14.
%O
U -V
*
U (1.4)
%
U U V U
where
Kt = Transmitter gain
tt = Transmitter time constant.
;
U
)
U (1.5)
/
U
The output Y(t) and the input M(t) are expressed as deviation variables, that is,
Y(t) = y(t) ysteady state
and
M(t) = m(t) msteady state
16 Introduction to Process Control Systems
Therefore, at the steady-state, operating levels Y and M will be zero. From the transfer
function, the response of Y(t) to a specified M(t) can be obtained by inverting the
equation
Y(s) = G(s)M(s) (1.6)
The transfer function consists of two parts, size relationship between the input
and the output signals and timing between the input and the output signals. In case
of linear component and sinusoidal input signal, the size relationship is measured by
gain and the timing is measured by phase difference as follows:
#ORNKVWFGQHVJGQWVRWVUKIPCN
)CKP
#ORNKVWFGQHVJGKPRWVUKIPCN
and
Phase difference = (Output phase angle) (Input phase angle)
Head, c
Flow, m
Capillary
(a) Physical system
m c
R
Metering device
Throttling device
u0 u1
m c
u2
(c) Physical system
u2
Kb
u0
Kb
m + + c
Ka
FJ S
J J
4
UGEHV (1.8)
FS IM # S
where
h1,h2 = Heads of liquid, ft
q = Liquid flow rate, ft3/sec
g = Acceleration due to gravity, ft/sec2
k = A flow coefficient (usually about 0.6) = Ö(FD/L)
F = Friction factor
A = Area of restriction, ft2
D = Inside diameter of pipe, ft
L = Equivalent pipe length, ft
18 Introduction to Process Control Systems
m %U
Flow, m
Head, c (d) Block diagram
(c) Physical process diagram
Figure 1.17 Capacitance-type processes.
Process Dynamics 19
The liquid capacitance of a liquid storage tank, as shown in Figure 1.17(a), is equal
to the cross-sectional area of the tank at the liquid surface and it can be calculated as
FX
E HV (1.10)
FJ
If the cross-sectional area is constant then the capacitance is also constant. The gas
capacitance of a tank, as shown in Figure 1.17((b), is constant and is analogous to
electric capacitance. The gas capacitance can be calculated as
FX 8
E HV (1.11)
FR P46
where
v = Weight of gas in the tank, lb
p = Pressure, lb/ft2
V = Volume of the tank, ft3
n = Polytropic exponent for uninsulated tanks; its value is between 1.0 and 1.2
R = Gas constant for a specific gas, ft/deg.
T = Temperature of gas, deg.
Ç ( /C (1.13)
where
S F = Net force acting on a mass
M = Total mass
a = Acceleration of the mass.
Flow, m
Head, c
m R c
Ts + 1
Substituting the value of q in Eq. (1.14) and multiplying both side of the equation by R,
we get
FE È EØ
4% 4 É O Ù 4O E
FV Ê 4Ú
or
FE
4% E 4O (1.15)
FV
Let
F
4% 6 CPF U
FV
Then
T(sc) + c = Rm
or
c(Ts + 1) = Rm
or
4O
E (1.16)
6U
where
R = Resistance, time/area
C = Capacitance, area
c = System output
RC = Process time constant, T
s = Differential operator = d/dt.
u2
Flow, m
h2
R2
u1
Head, c
R1
Outflow, q
u2 R
R22 u1 R
R11
T22 s + 11
T s T11s + 11
T s
R1
+ +
T1s 1
m RR22 + h2 RR11// RR22
TT22 ss + 11 TT11ss + 11 + c
The response of multiple time-constant processes is slower than that of single time-
constant processes, because the initial response of the system is retarded (the process
gain is reduced) by the second time-constant. The response of multiple time-constant
gets more retarded as the number of time constants increases and the overall response
gradually changes into S-shaped reaction curve.
Inflow Outflow, q
(a) Schematic
Input, X Output, Y
K
Kpp
F, c1
1
V c1 c2
11
Ts +
Ts
F, c2
The initial value of the output is again taken as zero, so that c2 represents the
deviation from the normal value. The equation is transformed and rearranged to give
the transfer function c2/c1. The time constant is V/F, the hold-up time in tank. Then
E
(1.19)
E 6U
where T = V/F. The assumption underlying Eq. (1.18) is that the concentration is
uniform throughout the tank or the input stream is instantaneously mixed with the
liquid in the tank. Actually a few seconds are required to mix the contents of a tank.
The output concentration is not the same as the average concentration in the tank when
both are changing.
If a stirred tank is used as a chemical reactor, the quantity reacting per unit of
time is an additional output term in the material balance for the reactant. For a first-
order reaction, the reaction rate is proportional to the concentration of reactant in the
24 Introduction to Process Control Systems
tank, which is same as the exit concentration for a well-mixed tank. The quantity
reacting is given as
r = kc2V (1.20)
The material balance for reactant is
FE
(E (E ME8 8UE 8 (1.21)
FV
Note that the reactor differs from the blending tank in that the output
concentration does not equal the input concentration at steady state. To solve for the
effect of changes in input concentration, it is convenient to define both input and output
variables as deviations from the normal values,
'E E E 'E E E
or
F 'E FE
FV FV
Substituting these definitions in Eq. (1.21) gives
F 'E
(
E 'E (
E 'E M
E 'E 8 8 (1.22)
FV
But, the normal quantity reacted is equal to the flow rate times the difference between
the normal inlet and outlet concentrations. That is,
(
E E ME8
Therefore, Eq. (1.22) gives
F 'E
( 'E ( 'E M'E 8 8 (1.23)
FV
Equation (1.23) is another form of Eq. (1.21) with each c replaced by Dc. Therefore, the
D is usually dropped, and the variables in Eq. (1.21) are understood to be deviations from
the normal values. With the initial values at zero, Eq. (1.21) is transformed and
rearranged as follows:
8UE
M8 ( E (E
or
8 (E
UE E (1.24)
M8 ( M8 (
After solving Eq. (1.24), the transfer function for the reactor is written by using the
time constant and the system gain. Let
(
- (1.25)
M8 (
Therefore,
8 (E
UE E -E
M8 ( M8 (
Process Dynamics 25
or
È 8U Ø
E É Ù -E
Ê M8 ( Ú
or
E -
E 8U
-X (
Let
8
6
M8 (
Therefore,
E -
E 6U
and
8 8(
6
M8 (
M8(
The chemical reaction keeps the time constant of the system less than the hold-
up time. Equation (1.25) shows that, as the rate constant becomes very large, the time
constant approaches 1/k. A fast reaction also means a low value of K, the steady-state
gain, and this may help explain the effect of k on the time constant. After a step change
in c1, the initial rate of change of c2 is shown by Eq. (1.21) as Fc1/V, with or without a
chemical reaction, since c2 is initially zero. With a large k, the eventual change in c2 is
small, and so less time is required to accomplish a given fraction of the change.
F F
E
V [X P E
V X P E
V - X P T
V (1.26)
FV FV
where
wn = The natural frequency of the system related to the speed of response for
particular value of x
z = The damping ratio of the system
K = The system gain
t = Time
r(t) = Input response of the system
c(t) = Output response of the system.
26 Introduction to Process Control Systems
h
L
#.S F J È IØ FJ
# É 2 JS Ù 4# (1.27)
IE FV Ê IE Ú FV
where
A = Cross-sectional area
r = Liquid density (density of gas above fluid is negligible)
P = Applied pressure
R = Frictional resistance
h = Head of liquid
g = Acceleration due to grevity.
With laminar flow, the resistance is given by the HagenPoiseuille equation
. F J .N FJ 2IE
J JK (1.29)
I FV S I& FV S I
Process Dynamics 27
F J [ FJ
JK J (1.30)
X P FV X P FV
The significance of wn and z becomes apparent after considering the solution to Eq.
(1.30) for a step-change to input pressure. With a damping coefficient less than 1, the
output overshoots the final value and oscillates before coming to equilibrium. The
system is said to be underdamped. For z < 1.0, we have
J
JK
GZR
[X P VUKP X P [ V G (1.31)
[
where
[
G VCP
[
With a damping coefficient of zero, the response is an undamped sine-wave of frequency
wn and amplitude 2hi. With a damping frequency of 1.0, the system is critically damped
and comes to equilibrium without overshooting. That is, for z = 1.0,
J
X P V GZR
X P V (1.32)
JK
If the damping coefficient is greater than 1.0, the system is overdamped and comes
to equilibrium slowly. When z > 1, the quadratic term can be factored and the solution
is given in terms of the two roots. That is, for z > 1.0,
U [
U
6C U
6D U
X X
and
J Ë È V Ø È V ØÛ
Ì 6C GZR ÉÊ 6 ÙÚ 6D GZR ÉÊ 6 ÙÚ Ü (1.33)
JK 6D 6C Í C D Ý
The damping coefficient is found either from the decay ratio, which is the ratio of
successive peak heights, or from the maximum overshoot, where
È Q[ Ø
Decay ratio = GZR É Ù (1.34)
Ê [ Ú
and
/CZKOWOQXGTUJQQV È Q[ Ø
GZR FGEC[TCVKQ
(KPCNXCNWG É Ù (1.35)
Ê [ Ú
The frequency of the damped oscillations is only slightly lower than the natural
frequency for values of z between 0 and about 0.5, where
Inlet Outlet
Flow rate, W Flow rate, Ww
Temperature, Twi Temperature, Two
Heater
m
u2 u1
c
u4 u3
The number of defining equations is only one that is the first law of
thermodynamics, which states the principle of conservation of energy. Therefore, the
degrees of freedom of this process are five. Hence, the maximum number of automatic
controllers that can be placed on this process is five. In the above example, c is a
controlled variable, m is a manipulated variable and u1 u4 are load variables, also called
uncontrolled independent variables. These uncontrolled independent variables, when
change, will upset the control system.
The process involves a phase change, for example, a boiler producing saturated
steam in which the number of components is one (that is water), and the number of
Transfer Functions of Physical Systems 31
phases is two (that is, water and steam). The degrees of freedom can be calculated using
Gibbs phase rule:
n = nc np + 2 (1.39)
where
n = Degrees of freedom of chemical process
nc = Number of components
np = Number of phases.
F K c F
m
m
x x1 x2 x
F = kx F = c( x 1 x 2) F = m
x
c J
K
T
T q1 q2
q
T = kq T = c( R1 R 2) T = J R
k, spring constant c, damping coefficient m, J, inertia
(a) Spring (b) Damper (c) Inertia
a b
d1 d2
T1
T2
x2
q1
y q2
x1 b a
y x1 x2
a b a b n = d2/d1 q2 = q1/n
(d) Lever T2 = nT1 n, speed di, diameter
(e) Gears
Figure 1.25 Mechanical system elements.
32 Introduction to Process Control Systems
rotational motion, together with the equations used to describe them. The arrows
identify the positive directions of the associated variables.
Spring, as shown in Figure 1.25(a), exists in many design configurations and
materials. For linearized models, the force F (torque T) is taken to be proportional to
linear deflection x (angular deflection q ). Note that in linearized models F and x
represent the variations of force and deflection about the operating point values. For
larger variations of deflections, the behaviour can sometimes be very nonlinear. The
spring constant k for nonlinear springs changes with the operating point.
Dampers or dashpots, as shown in Figure 1.25(b), generate a damping force F
(torque T) proportional to
Z Z
R R the difference of the velocities across the
damper, in the opposite direction. In practice, the friction in mechanical systems may
greatly differ from the viscous friction of this linear damper model. For dry friction, or
Coulomb friction, the direction of force or torque is opposite to the velocity difference,
and independent of its magnitude. The approximate linearization of such behaviour is
a subject in the study of nonlinear systems.
In Figure 1.25(c), by Newtons second law, force F (torque T) equals mass m
(inertia J) times acceleration Z
R . Lever, as shown in Figure 1.25(d), is a mechanical
implementation of a summing junction in a block diagram and is often used for this
purpose. For angles small enough from horizontal, the total motion y equals the sum
of the motion due to x1 with x2 = 0 and that due to x2 with x1 = 0. If a = b then y =
0.5 (x1 x2), where the direction of x2 is reversed. If the input and output of a system
are available in mechanical form and are applied to x1 and x2, respectively, then the
linkage with a = b and x2 opposite implements the feedback loop, as well as provides the
system error e = x1 x2.
Gear, as shown in Figure 1.25(e), is very common element often identified in terms
of its gear ratio n.
P
5RGGFQHFTKXKPIIGCT
X
R
R
R
X R (1.40)
5RGGFQHFTKXGPIGCT R R
where wi = qi is the angular velocity (in rad/sec) of the gear with diameter di. The
rotation T2 = nT1 between the torques arises because the two gears have a common
contact force, and the torque equals this force times the gear radius.
The above mechanical elements, as shown in Figure 1.25, can be broadly
categorized into two types of mechanical components: translational mechanical
components and rotational mechanical components. These are discussed now.
d2x
m
dt 2
x=0 c
k kx Mass, m
Mass, m
f (t)
dx
c
dt
Mass, m (b) Free-body diagram
x(t)
f (t)
(a) Spring-mass-damper
Figure 1.26 Spring-mass-damper mechanical system.
As shown in Figure 1.26(a), let us assume that the mass m is moving from x = 0
position. In this case, the spring is stretched, so the spring force kx is in upward
direction (as per Newtons second law), and hence opposes the downward acceleration.
It therefore receives a minus sign on the right side of Eq. (1.41) for mx. Since the mass
moves down, the damping force cx is upward, and this term must also have a minus
sign. The external force f (t) helps downward acceleration and therefore has a plus sign.
The resulting equation is
m Z = kx c Z + f(t) (1.41)
Rearranging the above equation, it gives the differential equation of motion in the usual
form
m Z + cZ + kx = f (t) (1.42)
With zero initial conditions, the Laplace transform and transfer function of
Eq. (1.42) can be written as
(ms2 + cs + k)X(s) = F(s)
or
:
U
(1.43)
(
U OU EU M
The two-mass system, as shown in Figure 1.27, can represent a dynamic absorber,
when a relatively small mass m1 is attached to a main mass m via spring k1 and damper
c1 to reduce vibrations x due to force f. Let us assume that m and m1 both move to the
right from the zero positions; m is farther and faster than m1. The spring force
k1(x x1) opposes m and helps m1, and damper force c1( Z Z 1) has the same effect.
Then the differential equations of motion become
mZ
= kx k1(x x1) c1( Z
Z
1) + f (1.44)
m1x1 = k1(x Z
1) + c1( Z
Z
1) (1.45)
34 Introduction to Process Control Systems
x x1
k1
f
m m1
k
c1
Figure 1.27 Two-mass system as dynamic absorber.
or
mZ
+ c1 Z
+ (k + k1)x = c1 Z 1 + k1x1 + f (1.46)
m1 Z1 + c1 Z1 + k1x1 = c1 Z + k1x
(1.47)
The transfer function of the above equation is given by
(ms2 + c1s + k + k1)X(s) = (c1s + k1)X1(s) + F(s) (1.48)
2
(m1s + c1s + k1)X1(s) = (c1s + k1)X(s) (1.49)
By substituting the value X1(s) from Eq. (1.49) into Eq. (1.48), we get
:
U O U E U M
(1.50)
(
U
OU E U M M
O U E U M
E U M
k c
J1 J2
q1 q2 q3
Figure 1.28 Rotating drive system.
Transfer Functions of Physical Systems 35
,
R
U
U R
U (1.57)
E
Substituting Eq. (1.57) into the Eq. (1.56) and rearranging give the transfer function
R
U M
(1.58)
R
U
, , EU
, , U
, MEU M
i R i C i L
i Cv v Li
v iR
V IZ V IZ V IZ
Z R 1 Z Ls
v (t ) Ri (t )
Z di
Cs v(t ) L
1 1 t
³
i (t ) v (t ) v(t ) i(t ) dt dt
R 1 t
i(t ) C
C f
dv
i(t ) ³
L f
v(t )dt
dt
R
i
+ vs = v iR
v(t) i(t)
R C L
v R C L
i
1
v iR
C ³
i dt Li
1
R Ls
³ v dt
1 dv 1
Z i v C
Cs R dt L
!
V IZ 1 1 1
Z Z1 Z 2 Z 3 Zn Cs
Z R Ls
(b) Electrical circuits V IZ
1
Z
1
Z1
1
Z2
1
Z3
! 1
Zn
Figure 1.29 Electrical systems.
The characteristics represented by the circuits of Figure 1.29 are used extensively
as controllers to improve the performance of feedback control systems. The transfer
function Eo(s)/Ei(s), between inputs ei(t) and outputs eo(t), can be found by use of the
results illustrated in Figure 1.29(b). With no current through the output terminals, all
are in effect of voltage dividers, in which eo is a fraction of the voltage ei, determined
by the current i through the input terminals caused by ei.
Transfer Functions of Physical Systems 37
Figure 1.30 illustrates various types of electrical networks, generally used for
realizing control function while implementing process control system.
R C R1
R
ei i C eo i R i
C
R1 R1 R2
i2
R2
C C
C1 C
i i i1 R1
C1
L1 L2
a b
R1 R2
È Ø
'K + É 4 Ù and 'Q +
Ê %U Ú %U
Then
'Q
(1.59)
'K 4%U
(b) Transient lead network (Figure 1.30b)
È Ø
'K + É 4 Ù and Eo = IR
Ê %U Ú
38 Introduction to Process Control Systems
Then
'Q 4%U
(1.60)
'K 4%U
È Ø È Ø
'K + É 4 4 Ù and 'Q + É 4 Ù
Ê %U Ú Ê %U Ú
Then
'Q U U
(1.61)
'K
U B U
where
4
t = RC and B
4 4
(d) Phase lead network (Figure 1.30d)
Ë Û
'K + Ì 4 Ü
Í
4 % U Ý
Since, from Figure 1.29(b) the equivalent impedance Z1 of the parallel impedances
Zr = R1 and Zc =
% U
is given by
QT<
< <T <E
<T
<E
Hence,
4 4 44% U
'K + and Eo = IR
4% U
Then
'Q U U
B (1.62)
'K BU U
where
4
tÿ = R1C1 and B
4 4
(e) Lag-lead network (Figure 1.30e)
Ë Û È Ø
'K + Ì 4 Ü and 'Q + É 4
Í
4 % U %U Ý Ê % U ÙÚ
Transfer Functions of Physical Systems 39
Then
'Q
U U
U U
(1.63)
'K U U U
U U U U
where
t1 = R1C1; t2 = R2C2; t12 = R1C2
'Q 4 4
(1.65)
'K . . U
. 4 . 4 . 4 U 4 4
40 Introduction to Process Control Systems
+
v ef Lf
if
ea em ec Lc
ic
ia Constant
reference
field
(c) Armature-controlled dc motor (d) Two-phase ac servomotor
Rf Rg Lg
ef eg eL ZL
Lf
if ig
(e) dc generator
ia
Terminals Speed, w
+
kw
Shaft
Housing
(f) Tachometer
Transfer Functions of Physical Systems 41
Back voltage
Plunger
position +
Magnetic force x
on plunger,
Soft iron L
fp = ki
plunger
i
Plunger
Spring mass, M
R
Spring Friction, B
(g) Linear actuator (Solenoid )
Figure 1.31 Electromechanical systems.
Voltage-driven potentiometer
A potentiometer contains a slider, which moves along a resistance element, as shown
in Figure 1.31(a). The potentiometer has a voltage v applied across the entire resistance
while a fraction of that voltage appears across the output. That fraction depends on the
ratio of the angle subtended by the slider compared to the maximum angle.
Field-controlled dc motor
Schematic diagram of this arrangement is shown in Figure 1.31(b). In this case, the
motor load is assumed to consists of an inertia J and a damper with damping constant
B. The developed motor torque T accelerates the inertia J and overcomes damping B.
Motor shaft position q and torque T are then related by
where
,
6O = Motor time constant
$
and
.H
6H = Field time constant
4H
Often Tf << Tm, and a satisfactory approximation in the operating range of interest is
R
U - V
4H $
(1.69)
'H
U U
6O U
Note that the transfer function was derived for the combination of the motor and
its load. This load affects motor speed, that is, it loads the motor, so a series connection
of two blocks with individual transfer function would be incorrect.
Armature-controlled dc motor
Figure 1.31(c) shows a model for an armature-controlled dc motor with a fixed field. An
input voltage causes a current to flow in the armature of the dc motor. If that current
creates a sufficiently large field to interact with the fixed field, the armature begins to
rotate. The armature rotates because a torque is applied both to the inertia and friction
of the motor, and the inertia and friction of the load, connected to the motor. That
torque is proportional to the armature current. As the armature spins, a reverse voltage
(also called a counter emf or electromotive force) is induced so as to oppose the input
voltage to the armature circuit. The armature loop, as shown in Figure 1.31(c), is
described by
ea = Raia + Laia + em
Ea(s) = (Ra + Las)Ia(s) + Em(s)
Here the counter-emf voltage can be taken to be proportional to shaft (armature) speed.
Then
em = Ke R , Em(s) = Kesq (s)
and the developed torque is proportional to current ia. Therefore,
T = Ktia, T(s) = KtIa(s)
Eliminating Ia, Em and T between the above equations and Eqs. (1.66) and (1.67) results
in the desired transfer function in the common form:
R
U - G
(1.70)
'C
U U=6C6O U
6O H 6C U H ?
where
,4C
Tm = = Motor time constant
-G -V
Transfer Functions of Physical Systems 43
.C
Ta = = Armature time constant
4C
$4C
g= = Damping factor
-G -V
Two-phase ac servomotor
In this application (Figure 1.31d), fixed and variable magnitude ac voltages are applied
to the reference and control fields, respectively. A 90° phase-shift arranged between
these voltages is made positive or negative depending on the desired direction of
rotation. The control field is described by
ec = Rcic + Lcic, Ec(s)= (Rc + Lcs)Ic(s)
The developed motor torque T can be taken to be proportional to ic and to decrease
proportionally with increasing speed, and it is described by
.E
Tc = = Electrical time constant
4E
-E
K= = Motor constant
4E
$ -X
dc generator
In this case (Figure 1.31e), the field loop equation is
'.
U <.I
(1.73)
'I
U .I U 4I <.
' .
U '. 'I
(1.74)
' H
U 'I'H
Tachometer
The tachometer is a special case of a dc generator in which the field is replaced by a
permanent magnet, which is equivalent to having a constant field current. In
tachometer, the output voltage is directly proportional to the tachometers rate of
angular rotation. Ideally, the armature supplies no current and the inertia is zero. It is
normally used with a very small electrical load, so that ia is nearly zero. Friction and
inertia are specially made as small as is practical. A tachometer is usually connected to
a rotating device so as to measure its rate of rotation without hindering the operation
of the device being monitored [Figure 1.31f].
Vi
Vo
Outflow
ca in the tank as close as possible to the desired (set or reference) level cd. Water flows
into the tank through valve Vi and flows out of the tank via valve Vo.
In this example, the valve is adjusted to make output ca equal to input cd but not
readjusted continually to keep the two equal. The limitation of an open-loop control
system is the difference between desired and actual level (error e = cd ca). This error
gets developed due to disturbances acting on the system and parameter variation of the
system. A block diagram of an open-loop system is shown in Figure 1.33.
Disturbance, d (t)
Plant or
Input Output
process
Disturbance
Desired plant Control Plant or
behaviour Controller process Output
Disturbance
Plant or
Desired plant process
Controller Control Output
behaviour
(reference)
Measured
Measurements
A strong motivation for using feedback control is to correct the error between
the controlled output and its desired value corresponding to the reference value.
The error is caused by disturbance inputs, parameter variations and imperfect
modelling. Feedback also affects other system performance characteristics such
as stability, sensitivity and overall gain of the process. The block diagram of a feedback
(or closed-loop) control system for regulating the temperature of a room is shown in
Figure 1.35(b).
The desired temperature, which is the reference input temperature (Tr), is set on
a calibrated dial, as shown in Figure 1.35(a). This positions the valve that admits hot
water for circulation through the radiator. Accordingly, heat flux (qi) flows into the room
and heat flux (qe) flows out of the room to the environment. The out-flux (qe) also
depends on the environment temperature (Te), which is usually a random variable. The
difference between heat in-flux and out-flux raises the room temperature. The valve dial
is calibrated when the environmental temperature (Te) and system parameters have
certain values. When these values change significantly, the controlled temperature (To)
will deviate from its reference value (Tr) by a large error and hence precise control will
not be realized in manual or open-loop system.
The block diagram of a feedback (or closed-loop) control system for controlling the
room temperature is shown in Figure 1.35(b). The controlled output c(t) is measured by
a sensor (or transducer) and compared with its desired value, which is the reference
input r(t), by a comparator or error detector. The error e(t) is used to produce a control
input u(t) to the actuator through an appropriate control algorithm with the objective
of correcting the error. The functional relationship, used to synthesize the control signal
u(t) as a function of the error e(t), is called control algorithm.
Techniques for Process Control 47
10°C
Tr
20°C Environment
30°C Calibrated dial
40°C Te
Room (qe)
Hot
water
Valve
Radiator
qi
Temperature, To
(a) Room temperature control
Controller Disturbance,
v(t) Controlled
Room output, c(t)
Input, r (t) Control u(t) temperature
e(t)
algorithm Actuator control
+
Sensors/transducers
Level controller
L
Set-point (SP), cd
Vo
Actual
level, ca
Fluid outlet, q
In closed-loop or feedback control system, the feedback loop causes the system to
take corrective action if the actual level ca deviates due to the disturbances from the
desired level cd. A block diagram of a closed-loop or feedback control system is shown
in Figure 1.37.
Disturbance, d(t)
Manipulated
Input, r (t) + Error, e (t) Feedback Output, c(t)
variable, m(t) Process
Reference controller
Controlled
or
set-point variable
Feedback
Sensor/transducer
Advantages
· It is very simple technique that compensates for all disturbances.
· The controller adjusts the controlled variable (affected by the disturbances) with
the set point (reference).
· The feedback control loop does not care for the disturbances entering into the
process. It tries only to maintain the controlled variable at set-point (reference)
by compensating for all disturbances.
Disadvantages
· It does not take corrective action until after the disturbance has upset the
process and generated an error signal.
· The feedback controllers continue changing its output until measurement and
set-point are in agreement.
· It solves the problems by trial and error method and thus causes oscillatory
response of a feedback loop
Feed forward
controller
Temperature
T
FT T
Condensate return
Disturbances
Sensor/transducer
d(t )
Input, r(t) Output
(set-point) Feed forward Process
controller Manipulated variable, m(t) Controlled
variable
(b) Block diagram
Figure 1.38 Heat exchanger feed forward control system.
In this example we have seen that the feed forward control technique compensates
for only two disturbances and if any other disturbance enters the system, it does not
compensate for this additional disturbance. This results in a permanent deviation of the
controlled variable from the set-point.
Therefore, to avoid such problem, feed forward control is used in combination with
feedback control system in all practical applications. In such systems, feed forward
control compensates for the major disturbances (reduces the effects of measurable
disturbances), while feedback control compensates for all other disturbances (such as
inaccuracies in the process model, measurement errors, and unmeasured disturbances),
as shown in Figure 1.39.
50 Introduction to Process Control Systems
Steam
Set-point
Feed forward
controller
Feedback
loop
TC
Temperature
FT TT transmitter
TT
Condensate return
Figure 1.39 Heat exchanger feed forward and feedback control system.
Advantages
· Feed forward control acts before the output is disturbed and is capable of very
good control performance with an accurate model.
· A stable feed forward controller cannot induce instability in a system that is
stable without feed forward control.
Disadvantages
· Feed forward control systems can lead to poor performance if improperly
designed and tuned.
· The major limitation to feed forward control is its instability to reduce steady-
state offset to zero. However, this limitation is easily overcome by combining
feed forward with feedback.
Process Stirrer
Process
fluid flow, w TT TC fluid flow, w
Stirrer
TC
TT
Ti T
T
w Ti
w Set-point
Q(t)
Q(t)
Heater Heater
(a) With single-loop control Set-point
TC
Disturbance-2 Disturbance-1
Controlled
Set-point
+ + Secondary + + variable
Primary
Primary Process-2 Process-1
controller
Controller controller
+ +
Sensor/
transducer
Secondary loop
Sensor/
transducer
Primary loop
ensures zero steady-state offset for step-like disturbances. Cascade control system
considers the likely disturbances and tailors the control system to the disturbance(s)
that strongly degrades the performance. It uses an additional secondary measured
process input variable that has the important characteristics of indicating occurrence of
the key disturbance.
As shown in Figure 1.40(a), the frequent and large disturbance in the stirred-tank
heat exchanger is the heating oil pressure. When there is increase in oil pressure, the
initial response of the oil flow and the heat transferred is to increase. Ultimately, the
tank-exit temperature increases, and the feedback controller reduces the control valve
opening to compensate for the increased pressure. While the effect of the disturbance
is ultimately compensated by the single-loop strategy, the response is slow because the
exit temperature must be disturbed before the feedback controller can respond. Now, in
cascade controller, the secondary variable is selected to be the heating oil flow, because
it responds in a predictable way to the disturbances in the oil pressure. The control
objective of tight control of the exit temperature and the final element are unchanged.
As shown in Figure 1.40(b), the cascade control system employs two feedback
controllers, with the output of the master (or primary) controller changing the set-point
52 Introduction to Process Control Systems
of the slave (or secondary) controller. The output of the exit temperature controller
adjusts the set-point of the flow controller in the cascade structure, i.e. the secondary
controller set-point is equal to the primary controller output. Thus, the secondary flow
control loop is essentially the manipulated variable for the primary temperature
controller. The net feedback effect is the same for single-loop or cascade control. In
either case, the heating oil valve is adjusted ultimately by the feedback. Therefore, the
ability to control the exit-temperature has not been changed with cascade.
The single-loop structure makes no correction for the oil pressure disturbance until
the tank exit-temperature is upset. The cascade structure makes a much faster
correction, which provides a better control. This can be seen by analyzing the initial
response of the cascade control system to an oil pressure increase. The valve position
is initially constant and, therefore, the flow of oil increases. The oil-flow sensor quickly
detects the increased flow. Since the flow controller set-point would be unchanged, the
controller would respond by closing the valve to return the flow to its desired value.
Because the sensor and valves constitute a very fast process, the flow controller can
rapidly achieve its desired flow of oil. By quickly responding to the pressure-increase and
compensating by closing the control valve, the secondary controller corrects for the
disturbance before the tank-exit temperature is significantly affected by the disturbance.
Following are main features of a cascade control system:
· The flow controller is much faster than the temperature controller due to much
shorter dead time in the secondary loop than in the original single-loop system.
· The temperature controller with an integral mode remains in the design to
ensure zero offset for all disturbance sources.
Load stream, L
FT
Lm Ratio controller
Divider
y¸ Rm
RC Rd
Ratio set-point
Mm
P
FT
Manipulated
stream, M
(a) Method 1
Load stream, L
FT Ratio station
Lm
RS Ratio set-point
Mm Ratio set-point
FC
P
FT
Manipulated
stream, M
(b) Method 2
Figure 1.41 Ratio control system.
In the second method, the flow rate of the load stream (L) is measured and the
value is transmitted to the ratio station (RS), which multiplies this signal by an
adjustable gain (KR). The output signal from the ratio station is then used as the set-
point for the flow controller, which adjusts the flow rate of manipulated stream (M).
Since divider is not used in this system, the open-loop gain remains constant.
Ratio control system is typically used in the industrial applications, such as
(i) blending operations, (ii) holding the fuel-air ratio to a furnace at the optimum value,
(iii) maintaining a stoichiometric ratio of reactants to a reactor, (iv) keeping a specified
reflux ratio for a distillation column, etc.
system shown in Figure 1.42, the input and output signals are represented as x(t) and
y(t), respectively, which are continuous function of time. Controllers or filters modelled
by continuous mathematics can be implemented by electrical and electronic devices such
as resistors, capacitors, inductors and operational amplifiers, and the resulting system
is called analog or continuous control system. Analog control systems use continuous
electronic or pneumatic signals. The controllers see transmitter signals continuously,
and control valves are changed continuously.
x y
System
System
Input Output
When the input and output signals of a process are defined as discrete instants of
time, represented as x(kT) and y(kT) respectively, then both the signals and the system
are called digital or discrete control. Digital control systems are discontinuous in
operation, looking at a number of loops sequentially. Each individual loop is polled every
sampling period.
FC
D PT
Mass-flow-rate
computer
TT
PT
Set
TT TC QC
Hot coil
FT
TT Q
Computer
TT
T PC
TT
Composition TC
computer
PT
Steam
where x is mole fraction of the more volatile component in the liquid. The partial
derivatives are usually assumed to be constants that are evaluated at the steady-state
operating level from the vapourliquid equilibrium data. Thus, pressure and
temperature on a tray can be measured, as shown in Figure 1.43(c), and a composition
signal or pressure-compensated temperature signal is generated and controlled.
DTPC = K1DP K2DT (1.78)
where
T PC = Pressure-compensated temperature signal
K1, K2 = Constants.
TC LS
TT
Steam
LT
LC
Bottoms
drop even with the bottoms flow at zero (vapour boil-up is greater than the liquid rate
from tray-1). If no corrective action is taken, the reboiler may boil dry (which could foul
the tubes) and the bottoms pump could lose suction.
Seeing this problem developed, an operator can switch off the temperature loop
into manual and cut off the steam flow. The control system in Figure 1.44 will perform
this override control automatically. The low selector (LS) sends the lower of the two
signals to the steam valve. If the steam valve is air-to-open, the valve will be pinched
back by either high temperature (through the reverse-acting temperature controller) or
low-base level (through the low-base-level override controller).
In level control applications, the override controller can be a simple fixed-gain relay
that acts like a proportional controller. The gain of the controller shown in Figure 1.44
is 5. It would be zeroed so that as the level transmitter dropped from 20 to 0 per cent
of full scale, the output of the relay would drop from 100 to 0 per cent of scale. This
means that under normal conditions when the level is above 20 per cent, the output of
the relay will be at 100 per cent. This will be higher than the signal from the
temperature controller, so the low selector will pass the temperature controller output
signal to the valve. However, when the base level drops 20 per cent and continues to
fall 0 per cent, the signal from the relay will drop and at some point will become lower
than the temperature controller output. At this point, the temperature controller is
overridden by the low-base-level override controller. Other variables might also take
over control of the steam valve. If the pressure in the column is too high, we may want
to pinch the steam valve. If the temperature in the base gets too high, we might want
to do the same. So there could be a number of inputs to the low selector from various
override controllers. The lowest signal will be the one that goes to the valve.
In temperature and pressure override applications, the override controller usually
must be a PI controller, not a P controller as used in the level override controller. This
is why the typical change in the transmitter signal over which we want to take override
action on these applications (high pressure, high temperature, etc.), is only a small part
of the total transmitter span. A very high-gain controller (P) would have to be used to
achieve the override control action, and the override control loop would probably be a
closed-loop unstable at this high gain. Therefore, a PI controller must be used with a
lower gain and a reasonably fast reset time to achieve the tightest control possible.
Override and selective controls are widely used to handle safety problems and
constraint problems.
temperature (as can occur during a thundershower or blue norther), the pressure in the
column can rapidly fall. This can cause flashing of the liquid on the trays, will upset the
composition and level controls on the column, and can even cause the column to flood.
To prevent this rapid drop, Shinskey developed a floating-pressure control system,
as shown in Figure 1.45. In this system, a conventional PI pressure controller is used.
The output of the pressure controller goes to the cooling-water valve, which is ac so that
it will fail open. The pressure controller output is also sent to another controller, the
valve position controller (VPC). This controller looks at the signal to the valve, compares
it with the VPC set-point signal, and sends out a signal that is the set-point of the
pressure controller. Since the valve is ac, the set-point of the VPC is about 5 per cent
of the scale to keep the cooling-water valve almost wide open.
SP
PT PC
PV CO
VPC
VPC SP = 5%
CO
PV
Column
CW
AC
123456789012345678
123456789012345678
123456789012345678
123456789012345678
123456789012345678
123456789012345678
Figure 1.45 Floating-pressure VPC.
The VPC is a different type of cascade control system. The primary control is the
position of the valve. The secondary control is the column pressure. The pressure
controller is PI and is tuned fairly tightly so that it can prevent the sudden drops in
pressure. Its set-point is slowly changed by the PVC to drive the cooling-water valve
nearly wide open. A slow-acting, integral-only controller should be used in the VPC.
Example 2. Figure 1.46 illustrates another application of VPC to optimize a
process in which temperature of a reactor is to be controlled. The reactor is cooled by
both cooling water flowing through a jacket surrounding the reactor and by condensing
60 Introduction to Process Control Systems
AC
Refrigerant
Vapour
VPC SP
Reactor
CW
AC
TT TC
vapour that boils off the reactor in the heat exchanger cooled by a refrigerant. This form
of cooling is called auto-refrigeration.
From an energy-cost perspective, using cooling water is preferred to refrigerant
because water is much cheaper. However, the dynamic response of the temperature to
a change in cooling water may be much slower than the response to a change in
refrigerant flow. This is because the change in water-flow must change the jacket
temperature, which changes the metal-wall temperature, which then begins to change
the reaction mass temperature. Changes in refrigerant-flow quickly raise or lower the
pressure in the condenser and change the amount of vaporization in the reactor, which
is reflected in reactor temperature almost immediately.
So from a control point of view, we use refrigerant to control temperature. Much
tighter control is achieved when a refrigerant is used rather than cooling water. The
VPC approach handles this optimization problem very nicely. We simply control
temperature with refrigerant, but send the signal that is going to the refrigerant valve
(the temperature controller output) into a valve position controller, which will slowly
move the cooling-water valve to keep the refrigerant valve nearly closed. Since the
refrigerant valve is ac, the set-point signal to the VPC will be about 5 to 10 per cent
of full scale.
Application-based Classification of Control Systems 61
rinse operation, and the spin-dry operation are all initiated and terminated by a timer.
Most batch process control systems are a time-driven sequential control systems. This
system is described by a schematic and timing diagrams in which the schematic diagram
depicts the physical configuration of the system whereas the timing diagram defines the
sequential operation of the system.
From
Feedback Measuring processor
amplifier element
Deviation. The difference between the measured value of the controlled condition and
the set value is known as deviation. A positive sign applies when a measured value is
greater than the set value and negative if it is less.
Controlling element. It is the element that provides the control signal, which is
dependent upon the deviation, to the correcting unit.
Automatic controller. It is a device which compares a signal from a detecting element
with a signal representing the set value and operates in such a way as to reduce the
deviation.
Offset. A sustained deviation due to an inherent characteristic of proportional
controller action is known as the offset.
On
Temperature
Ideal case
(No lags)
Temperature Set-point Off
Sensor
sensor Steam flow
FFi,i,Zw TT TC
Temperature Differential
Steam
Steam
FFoo, ,wZ
Actual case
Heater
Heater
Steam
Measured Differential
temperature
Steam
On On On On
Set-point
Time
On
Heater
Off
Set-point Temperature
(b) Transfer-function curve
Set-point M1
(SP) Error, e
m
M2
On On On On
Heater
Temperature
Set-point Differential
(SP) (hysteresis)
Time
(d) Onoff controller with differential (hysteresis) gap
Differential gap
Set-point (hysteresis)
M1
(SP) Error, e
+ m
M2
100% 100%
80% 80%
60% 60%
40% 40%
30% 30%
20% 20%
10% 10%
0% 0%
Set-point
(SP) Error, e Output, m To manipulated
+ Kp variable
From sensor
or controlled
(b) Block diagram variable
Figure 1.51 shows a typical response curve for a kettle with proportional control of
temperature. A sudden drop in feed temperature causes a gradual drop in kettle
temperature, which leads to a greater flow of steam through the control valve. The
temperature eventually becomes constant at a value slightly below the set-point, and
this steady-state error is called the offset. An offset is inevitable, since more steam is
needed to heat the colder feed and the controller will admit more steam only if the
temperature is below the set-point.
Controller
p output
Tp
TF
Time
Figure 1.51 Response of proportional control system to step change in load.
The main limitation of a plain proportional controller is that it cannot keep the
controlled variable on set-point. Therefore, controlled variable will increase or decrease
from the set-point. The difference between the value of controlled variable and the set
point is called offset.
Derivative (D) or rate control action
In derivative control action, the output signal of the controller is a function of the rate
at which the error is changing. Derivative control action provides a means of obtaining
a controller with high sensitivity. An advantage of using derivative control action is that
it responds to the rate of change of the actuating error and can produce a significant
correction before the magnitude of the actuating error becomes too large.
Therefore, the proportional mode considers the present state of the process error,
and the integral mode looks at its past history, while the derivative mode anticipates
its future state and acts on that prediction. Derivative control predicts (anticipates)
actuating (process) errors before they have evolved and takes corrective action in
advance of that occurrence and tends to increase the stability of the system.
Mathematical expression (the relationship between output of the controller to the
input error) of derivative-only controller is given as
FG (1.82)
O 6F
FV
where Td is derivative time of the controller. Since the derivative control mode acts on
the rate at which the error signal changes, it can also cause unnecessary upsets, such
70 Introduction to Process Control Systems
Set-point
(SP) Error, e Output, m
+ To manipulated
TTddss variable
From sensor or
controlled
variable
Figure 1.52 Block diagram of derivative (D) control action.
Since derivative control operates on the rate of change of the error and not the
error itself, it is always used in combination with proportional or proportional-plus-
integral control action.
From sensor
or controlled
variable
Figure 1.53 Block diagram of integral (I) control action.
Process Control Actions (or Modes) 71
È V
Ø
O - R É G
Ê 6K Ô G FVÚÙ (1.84)
The integral time is Ti varied by providing a knob. It adjusts the integral control action.
The inverse of Ti is called the reset rate which is the number of times per minute that
the proportional part of the control action is duplicated. Reset is measured in terms of
repeats per minute. A block diagram of proportional-plus-integral (PI) control action is
shown in Figure 1.54.
Set-point § 1 · Output, m To
(SP) +
Error, e
KKpp¨¨§¨11 1 ·¸ ¸¸ manipulated
Ts
©© Tii s¹ ¹ variable
From sensor
or controlled
variable
Figure 1.54 Block diagram of proportional-plus-integral (PI) control action.
È FG Ø
O - R É G 6F Ù (1.85)
Ê FV Ú
where Td is the derivative time which is the time interval by which the rate action
advances the offset of the proportional control action. A block diagram of proportional-
plus-derivative (PD) control action is shown in Figure 1.55.
Set-point To
Error, e Output, m manipulated
(SP)
+ KKpp (11
TTd ss)
d variable
From sensor
or controlled
variable
Figure 1.55 Block diagram of proportional-plus-derivative (PD) control action.
72 Introduction to Process Control Systems
È V FG Ø
O - R É G
Ê 6K Ô G FV 6 F
FV ÙÚ
(1.86)
Set-point
(SP) Error, e § § 11 · · Output, m To
+ K pK¨¨p1¨©1 Tis TdTs¸¹d s ¸¸ manipulated
© Ti s ¹ variable
From sensor or
controlled
variable
Figure 1.56 Block diagram of PID control action.
A comparison between P, PI, PD, and PID type of control actions is given in
Table 1.1
Parameters
attributes P-type PI-type PD-type PID-type
expression (m)
6
FG Ø
FV ÚÙ
F
Equipment
design
Process
control
To containment
Safety valve
Feed composition
Methane = 10%
Ethane = 20% PAH
Propane = 30%
i-Buten = 15% Vapour
Vapor
n-Buten = 20%
i-Pentane = 5% ‘A’ component
TI
6
PC
1
TI TI TI TI
1 2 4 5
LC
1
TI
FI 3
1
FI Liquid
2 FI
3 ‘B’ component
Process Steam
fluid AC
1
Figure 1.58 Flash separation process with control strategy: TI = Temperature indicator; FI = Flow
indicator; PC = Pressure controller; LC = Level controller; PAH = Pressure alarm high;
AC = Analyzer controller.
in the overhead vapour pipe. Following control strategies can be defined for the process
shown in Figure 1.58:
1. Safety of personnel: The maximum pressure of 1200 KPa must not be exceeded
under any (conceivable) circumstances.
2. Environmental protection: Material must not be vented to the atmosphere
under any circumstances.
3. Smooth plant operation and production rate:
(a) Control all unstable variables (liquid level).
(b) All process variables should remain within reasonable ranges without undue
operator actions.
Process Control Objectives 75
(c) One variable should control the production rate with little variation.
(d) Control loops should function well independent of manual/automatic status
of other loops; that is, the system should have good integrity.
4. Product quality:
(a) The steady-state value of the ethane in the liquid product should be
maintained at its target of 10 mole% for steady-state operating condition
changes of
· +20 to 25% feed flow
· 5 mole% changes in the ethane and propane in the feed
· 10 to +50oC in the feed temperature
(b) The ethane in the liquid product should not deviate more than ±1 mole%
from its set-point during transient responses for the following disturbances:
· The feed temperature experiences a step from 0 to 30oC.
· The feed composition experiences steps of +5 mole% ethane and 5 mole%
propane.
· The feed flow set-point changes 5% in a step.
5. Efficiency and optimization: The heat transferred should be maximized from
the process integration exchanger before using the more expensive steam utility
exchanger.
6. Monitoring and diagnosis:
(a) Sensors and displays needed to monitor the normal and upset condition of
the unit must be provided to the plant operator.
(b) Sensors and calculated variables required to monitor the product quality and
thermal efficiency of the unit should be provided for longer-term monitoring.
1.13.1 Safety
The safety of people in plant and in the surrounding community is of paramount
importance. While no human activity is without risk, the typical goal is that working
at an industrial plant should involve much less risk than any other activity in a persons
life. No compromise with sound equipment and control safety practices is acceptable.
Plants are designed to operate safely at expected temperatures and pressures.
However, improper operation can lead to equipment failure and release of potentially
hazardous materials. Therefore, the process control strategies contribute to the overall
plant safety by maintaining key variables near their desired values. Since these control
strategies are important, they are automated to ensure rapid and complete
implementation. In Figure 1.58, the equipment could operate at high pressures under
normal conditions. If the pressure were allowed to increase too far beyond the normal
value, the vessel might burst, resulting in injuries or death. Therefore, the control
strategy includes a controller PC-1 that controls the pressure by adjusting the valve
position (i.e. per cent opening) in the vapour line.
Another consideration in plant safety is the proper response to major incidents,
such as equipment failures and excursions of variables outside of their acceptable
bounds. Feedback strategies cannot guarantee safe operation. A very large disturbance
can lead to an unsafe condition. Therefore, an additional layer of control, termed as
76 Introduction to Process Control Systems
emergency system, is applied to enforce bounds on key variables. Typically, this layer
involves either safely diverting the flow of material or shutting down the process when
unacceptable conditions occur. The control strategies are usually not complicated, for
example, an emergency control might stop the feed to a vessel when liquid level is
nearly overflowing. In Figure 1.58, the pressure is controlled by the element pressure
control (PC). Normally, it maintains the pressure at or near its desired value.
However, the control strategy relies on the proper operation of equipment like the
pressure sensor and the valve. Suppose that the sensor stopped providing a reliable
measurement, the control strategy could improperly close the overhead valve, leading
to an unsafe pressure. The correct control design would include an additional strategy
using independent equipment to prevent a very high pressure. For example, the safety
valve shown in Figure 1.58 is closed unless the pressure rises above a specified
maximum, then, it opens to vent the excess vapour. It is important to recognize that
this safety relief system is called on to act infrequently, perhaps once a year or less
often, therefore, its design should include highly reliable components to ensure that it
performs properly when needed.
Profit optimization
The typical goal of the plant is to return a profit. In the case of a utility such as water
purification, in which no income from sales is involved, the equivalent goal is to provide
the product at the lowest cost. Before achieving the profit-oriented goal, selected
independent variables are adjusted to satisfy the first-five higher-priority control
objectives. Often, some independent operating variables are not specified after the
higher objectives (that is, including product quality but excepting profit) have been
satisfied. When additional variables (degree of freedom) exist, the control objective can
increase profit while satisfying all other objectives. In Figure 1.58, to heat the inlet
stream, the control strategy can select the correct exchanger based on the cost of the
two heating fluids.
78 Introduction to Process Control Systems
First law: The best control system is the simplest one that will do the job
Complex and elegant control systems look great on paper but soon end up on manual
(taken out of service) in an industrial environment. Bigger is definitely not better in
control system design.
Levels of Process Control System 79
Second law: You must understand the process before you can control it
No degree of sophistication in the control system (from adaptive control, to expert
systems, to Kalman filters, to nonlinear model predictive control) will work if you do not
know how your process works. Many people have tried to use complex controllers to
overcome ignorance about the process fundamentals, and they have failed. Learn how
the process works before you start designing its control systems.
Control system
structure
Level-3
Algorithm
Level-2
Controller tuning
Level-1
The third level (level-3) is control system structure, determining what to control,
what to manipulate, and how to match one controlled variable with one manipulated
variable (called pairing). The selection of the control structure for a plant is a vitally
important function. A good choice of structure makes it easy to select an appropriate
algorithm and to tune. No matter what algorithm or tuning is used, it is very unlikely
that a poor structure can be made to give effective control.
The top level (level-4) is process design, developing a process flow-sheet and using
design parameters that produce an easily controllable plant. The steady-state
economically optimal plant may be much more difficult to control than an alternative
plant that is perhaps only slightly more expensive to build and operate. At this level,
the economic impact of a good process control engineer can be enormous, potentially
resulting in the difference between a profitable process and an economic disaster.
ILLUSTRATIVE EXAMPLES
Liquid inlet
q(t)
or
FJ
S
V SQ
V # (1.88)
FV
Combining Eqs. (1.87) and (1.88) to eliminate qo(t) gives the following linear
differential equation:
J FJ
S # (1.89)
4 FV
We shall introduce deviation variables into the analysis before proceeding to the
transfer function. Initially, the process is operating at steady state, which means that
dh/dt = 0. We can write Eq. (1.89) as
JU
SU (1.90)
4
where the subscript s has been used to indicate the steady-state value of the variable.
Subtracting Eq. (1.90) from Eq. (1.89) gives
F
J JU
S SU J JU # (1.91)
4 FV
If we define the deviation variable as
Q = q qs and H = h hs
3
U *
U #U*
U (1.93)
4
Notice that H(0) is zero and therefore the transform of dH/dt is simply sH(s).
Equation (1.93) can be rearranged in the standard form of the first-order lag to give
*
U 4
(1.94)
3
U U U
where t = AR.
Example 1.2 A tank having a time constant of 1 min and a resistance of (1/9) ft/cfm
is operating at steady state with an inlet flow of 10 ft3/min. At time
t = 0, the flow is suddenly increased to 100 ft3/min for 0.1 min by adding
an additional 9 ft3 of water to the tank uniformly over a period of
0.1 min. Plot the response in tank level and compare with the impulse
response.
82 Introduction to Process Control Systems
Before proceeding with the details of the computation, we should observe that, as
the time interval over which the 9 ft3 of water is added to the tank is shortened, the
input approaches an impulse function having a magnitude of 9. From the data given in
this example, the transfer function of the process is
*
U
3
U U
The input may be expressed as the difference in step functions:
Q(t) = 90[u(t) u(t 0.1)] (1.95)
The transform of Eq. (1.95) is
3
U
U
GU
(1.96)
Combining Eq. (1.96) and the transfer function of the process, we obtain
Ë G U Û
*
U Ì Ü (1.97)
Í U
U U
U Ý
The first term in Eq. (1.97) can be inverted to give 10(1 et). The second term, which
includes e0.1s, must be inverted by use of the theorem on translation of functions.
According to this theorem, the inverse of exp (sto)f(s) is f(t to) with f(t) = 0 for
(t to) < 0 or t < to. The inverse of the second term in Eq. (1.97) is
Î GU Þ ÎÑ HQTV
. Ï ß Ï
ÑÐ U
U Ñà ÐÑ= G
V ? HQTV
The complete solution to this problem, which is the inverse of Eq. (1.97), is
ÎÑ
GV V Þ
*
V Ï ß (1.98)
V V ! à
ÐÑ ËÍ
G
G ÛÝ
V
Impulse response
(ideal)
1.0
Area = 9 ft3
H(t)
qft3/min
10
0 0.1 0.2 0 1 2
t(min) t(min)
(a) Pulse input (b) Response of tank level
Figure 1.61 Approximation of an impulse function in a liquid-level system.
Temperature
sensor
Fi, w TT TC
Steam
Fo, w
Heater
Steam
Figure 1.62 Process diagram.
Since we are concerned only with the steady-state changes, we can use a block
diagram, as shown in Figure 1.63(a), that shows only the steady-state characteristic or
84 Introduction to Process Control Systems
gain of each element. A consistent set of units must be used in defining the gains so
that all units introduced cancel in following the signals around the loop. In this case,
the individual gains are usually made dimensionless, and the numbers used appear in
the definition of two gains.
(TCEVKQPCNEJCPIGKPRTGUUWTG 'RUK
-R
(TCEVKQPCNEJCPIGKPOGCUWTGFVGORGTCVWTG 'R O (
Load
KL
+ +
qc Kp Kv Kkj q
Sensor
Km
KL
+
+ e +
qc K q
K = Kp Kv Kkj
(b) Modified block diagram
Figure 1.63 Block diagrams of steady-state changes.
Illustrative Examples 85
where (U is normal or initial value of steam flow. Then the measured gain is 1.0, and
so the diagram can be modified, as shown in Figure 1.63(b). By algebraic manipulation,
we have
Dqÿ = Ke + KLDqF
Since qc is constant, e = Dq and
Dq (1 + K) = KLDqF
The load gain (KL) is the change in process temperature for unit change in feed
temperature at a given steam rate. Then for this example,
KL = 1.0
All the steam entering the jacket condenses. So a 1 per cent increase in steam flow
is a 1 per cent increase in heat-transfer rate. Since the normal steam rate gives a
40°F temperature rise, a 1 per cent increase will give a 40.4°F temperature rise, or an
increase of 0.4°F. This increase is divided by the temperature range of the chart to get
the process gain. Then
u(u(
- ML
from problem statement, Kv = 4.0. For Kp = 1,
K = 4(0.445) = 1.78
and
-.
'R 'R ( 'R (
-
For DqF = 5o Dq = 1.8oF
For Kp = 5 K = 8.9
For DqF = 5o Dq = 0.51oF
Example 1.4 A process steam is heated from 100 to 150°F by injecting steam into the
flowing solution. The temperature is measured 40 feet downstream from
the injection point, and a proportional controller is used to adjust the
flow steam. Because the fluid velocity is 40 fpm, there is a time delay
of 1 min from the injection point to the measurement point. The other
lags in the control loop are assumed negligible. Show the response of the
system to a change in feed temperature when the overall gain K is 0.5,
1.0 and 1.2.
Suppose that the feed temperature drops to 99°F. One minute later, the measured
temperature drops to 149°F. With an overall gain of 0.5, a 1°F drop in process
temperature causes the steam flow to increase enough to raise the temperature 0.5°F.
The effect of the increase in steam flow is felt 1 min later when the measured
temperature jumps to 149.5°F. This causes the steam valve to close by an amount that
will change the temperature from 0.25°F to 149.25°F. Further changes occur each
minute, and the temperature finally approaches 149.33°F.
With an overall gain of 1.0, a 1°F drop in temperature causes a steam-flow change
equivalent to 1°F, and 1 min after the change in steam flow the error is zero. However,
86 Introduction to Process Control Systems
when the error is reduced to zero, the steam flow is decreased to the initial value and
the temperature drops to 149°F again. Therefore the process oscillates between 149 and
150°F with a period of 2 min.
With an overall gain of 1.2, an error of 1°F leads to a corrective action of 1.2°F,
and 1 min after the correction is made, the error becomes +0.2°F. A positive error
makes the steam flow less than the normal, and 1 min later the error is 1°F 1.2(0.2°F)
= 1.24°F. The error becomes larger with each cycle, and the system is unstable.
Example 1.5 Gases A and B are fed continuously to a tank with a volume of 30 ft3.
The normal tank conditions are 40 psia and 80°F, and the normal flow
rates are FA = 40 and FB = 10 cfm measured at tank conditions. If the
flow of B is suddenly increased to 12 cfm, when does the concentration
of B reach 90 per cent of the new steady-state value?
The time constant for small changes in flow rate is the normal hold-up time. Then
8
6 OKP
(
We also have
y = 0.9 = 1 et/T
or
V
6
Thus
t = 1.4 min
Example 1.6 A first-order reaction is carried out in a stirred tank with a hold-up time
of 1.6 hr and a rate constant of 2 hr1. Show the effect of a sudden
change in feed concentration from 0.50 to 0.48 mole/ft3.
Initial conditions are
( E E ME 8
where E and E are the normal inlet and exit concentrations. Then
E
E
M8(
or
E = 0.119
The transfer function is
E -
E 6U
or
-
M8(
or
8(
6 JT
M8(
Review Questions 87
REVIEW QUESTIONS
1.1 Describe important components of a process control system. Give few examples of
process control system.
1.2 Describe the four necessary components of a feedback control system.
1.3 Review the equipment sketched in Figures 1.64(a) and (b) and find out the process
Rod
Flow-in varies,
cannot be adjusted
and type of control system. In particular, identify the necessary components of the
control system.
In the figure, the flow-in is a function of the connecting rod position. The flow-
out is a function of the level (pressure at the bottom of the tank) and the
resistance to flow.
1.4 With neat sketches and examples, explain the dynamic and steady-state
responses of industrial process that contain resistance, capacitance and dead-
time elements.
1.5 Derive the differential equations to describe the dynamic behaviour of physical
components of a process control systems such as mechanical, electrical, hydraulic
and pneumatic systems.
1.6 What is cascade control system? When are such systems used in process control
applications? Illustrate with neat sketch a cascade control system for stirred-tank
heat exchanger temperature control.
1.7 The preliminary process designs have been prepared for the systems shown in
Figure 1.65. The key variable to be controlled for the systems are (i) flow rate,
temperature, composition and pressure for flash system and (ii) composition,
Drum
Liquid
(a) Pump
Ti
Process fluid flow w
Solvent
T
w
Reactant
Cooling
(b)
Figure 1.65 Review question 1.7.
Review Questions 89
temperature and liquid level for the continuous-flow stirred-tank chemical reactor.
For both the processes, disturbances occur in the feed temperature and
composition. Answer the following questions for both the processes:
(a) Determine which sensor and final elements are required so that the important
variables can be controlled. Sketch them on the figure where they should be
located.
(b) Describe how the equipment capacities should be determined.
(c) Select controller pairings, that is, select which measured variable should be
controlled by adjusting which manipulated variable.
1.8 Illustrate with examples and sketches the electromechanical devices that are
encountered in industrial process control applications.
1.9 (a) What are different types of techniques that are useful in analysis and design
of modern process control systems?
(b) Explain with neat sketch the purpose of using feed forward control system with
its advantages and disadvantages.
1.10 Explain with sketches override and optimizing control systems. Give one example
of each type.
1.11 Explain with illustration the application of valve position control (VPC) for energy
optimization.
1.12 Explain with neat sketches the actuator based control systems such as electrical,
hydraulic and pneumatic systems. Derive correlation equations and draw the block
diagram representation of each system. Describe the advantages and
disadvantages of each type of control systems.
1.13 For each of the following processes, identify at least one control objective in each
of the seven categories (introduced in Section 1.13):
(a) The reactor-separator system in Figure 1.66
(b) The boiler in Figure 1.67
(c) The fired heater in Figure 1.68.
Describe a feedback approach appropriate for achieving each objective.
1.14 Consider any of the control systems shown in Figures 1.69(a)(d). Suggest a
feedback control calculation that can be used to determine the proper value of the
manipulated valve position. The only values available for the calculation are the
desired value and the measured value of the controlled variable.
1.15 Review the processes, as sketched in Figures 1.69(a)(d), in which the control
variable is to be determined at its desired value. Also,
(a) From your control engineering background, suggest the physical principle
used by the sensor.
(b) Explain the casual relationship between the manipulated and controlled variables.
(c) Explain whether the control valve should be opened or closed to increase the
value of the controlled variable.
(d) Identify possible disturbances that could influence the controlled variable.
Also, describe how the process equipment would have to be sized to account
for the disturbances.
90
F5
F2 V
T5 P1
F3 T6 F4
T3
T1 T4
F1
L1
L2
V V V
Introduction to Process Control Systems
V V
T7
T2 T8
Hot oil
Hot oil
T9 F6
Feed tank Heat exchanger Chemical reactor Heat exchanger Flash separator
Figure 1.66 Integrated feed tank, reactor and separator with recycle.
Review Questions 91
Superheated
Flue steam
gasgas
Flue
TC
2
TC
1
Treated
Treated
water
water
Treated
Radiation water
Fuel
Figure 1.67 Cascade control design for boiler superheated steam temperature control.
Flue gas
Feed
PC
FC
TC
SP
MV CV
FC
DP X
FI FY FY
r
AI
Figure 1.68 Fired heater with inferential control for better performance as fuel gas composition
changes.
92 Introduction to Process Control Systems
Ti
AC
Heater
FC
LC
Heater
Feed A Feed B
Product
FC
1.16 Feedback control uses measurement of a system output variable to determine the
value of a system input variable. Suggest an alternative control approach that
uses measured (disturbance) input variable to determine the value of different
(manipulated) input variable, with the goal of maintaining a system output
variable at its desired value. Apply your approach to one of the systems in
Figure 1.69. Can you suggest a name for your approach?
1.17 Explain how the basic strategy of control is employed in an industrial furnace
heating system. What are controlled and manipulated variables? Is the system
self-regulating?
1.18 What are the relative advantages and disadvantages of the proportional, integral
and derivative control actions? What are their characteristic effects on the closed-
loop response of a process?
1.19 What is the order of the closed-loop dynamic response for a second-order process
with PI control? Can the PI control destabilize such a process?
1.20 Discuss the effects of Kp, Ti and Td on the closed-loop response of a process
controlled (i) with PI (ii) with PID.
1.21 (a) Consider a first-order process. Could you have almost the same closed-loop
responses with PI and PID controllers and appropriate values of their
adjustable parameters?
(b) Repeat (a) for second-order process.
1.22 Which one of the three controllers, P, PI and PID, would give more robust closed-
loop response to an under-damped second-order system?
1.23 (a) Integral control action makes a process (i) faster or slower; (ii) more oscillatory
or less; (iii) with larger deviations from the set-point or smaller? Explain your
answers.
(b) Repeat (a) for derivative control action.
1.24 The signal to a pneumatic proportional-integral controller is e = 0.05 sin w t. The
controller gain is 3, and the reset time is 2 min.
(a) Give the equation for the controller output in dimensionless terms and in
pounds per square inch for w = 0.1 and 1.0 rad/min. The normal output
pressure is 9 psi.
(b) Sketch the outputs and the inputs for the two cases, and determine the phase
angle.
1.25 A unit-step change error is introduced into a PID controller. If Kp = 10, Ti = 1
and Td = 0.5, plot the response of the controller, P(t).
1.26 A PI controller is tested by applying a step change in input equal to 10 or
20 per cent of the chart scale. For a gain of 2 and a reset time of 0.5 min, show
the output expected for the two step changes. If it takes a few seconds to make
the change in input, how would you determine the true gain and reset time from
the output curve?
94 Introduction to Process Control Systems
1.27 A pneumatic PI controller has an output pressure of 10 psi when the set-point and
pen-point are together. The set-point and pen-point are suddenly displaced by
0.5 in. (i.e. a step change in error is introduced) and the following are obtained:
Determine the actual gain (psig per inch displacement) and the integral time.
1.28 A proportional controller is used to control the flow of steam to a kettle which is
continuously heating a thin slurry from 80 to 160oF. If the throttling range or
proportional band is 25 per cent (controller pressure goes from 3 to 15 psi and
valve position from open to closed for a temperature interval equals to 25 per cent
of full chart scale), what offset results from a 10oF drop in an inlet temperature?
Following data are given:
Flow rate = 500 lb/hr
Cp = 1 Btu/(lb)(oF)
Normal valve position = Half open
Normal controller pressure = 9 psi
Chart scale = 50200°F
1.29 A liquid-level system, as shown in Figure 1.68, has a cross-sectional area of
3.0 ft2. The valve characteristics are S J where q is flow rate (cfm) and h
is level above the valve (ft). Calculate the time constant for this system if the
average operating level is (i) 3 ft (ii) 9 ft.
1.30 A U-tube manometer is made of 0.5-cm tubing and has a mercury column of
100-cm long. Each leg of the manometer is connected by 500 cm of the same sized
tubing to a pressure tap in a water pipe. Assuming laminar flow for both water
and mercury, how does the presence of water above the mercury affect the critical
frequency and the damping coefficient?
1.31 A mercury manometer is depicted in Figure 1.70. Assuming the flow in the
manometer to be laminar and the steady-state friction law for drag force in
laminar flow is applied at each instant, determine a transfer function between the
applied pressure (p1) and the manometer reading (h). It will simplify the
calculation if, for inertial terms, the velocity profile is assumed to be flat. Ramp
your transfer function, written in standard second-order form, list (i) the steady-
state gain, (ii) t and (iii) z. Comment on these parameters as they are related to
the physical nature of the problem.
1.32 The level in an open tank is automatically controlled by adjusting a valve in the
discharging line. The normal level is 10 ft, and the normal flow to and from the
tank is 50 gpm. For a given head, the flow through the valve is a linear function
Self-Check Quiz 95
P1
p = 0
h
L
of the signal to the valve, and for a given opening the flow varies with the square
root of the head. If the proportional controller changes the signal to the valve by
2 psi for a 1-ft change in level, what are the steady-state levels for input flows of
40 and 60 gpm? What would be the levels if the valve position were fixed?
1.33 A PID controller is at steady state with an output pressure of 9 psig. The set-point
and pen-point are initially together. At time t = 0, the set-point is moved away
from the pen-point at a rate of 0.5 in./min. The motion of the set-point is in the
direction of lower readings. Plot the out-pressure versus time, if the knob settings
are: Kp = 2 psig/in. of pen travel; Ti = 1.25 min; Td = 0.4 min.
1.34 The temperature in a laboratory bath is controlled by a thermometer set to
have a differential gap of 1oF. When the heater is on, the bath temperature rises
3oF/min, and when the heater is off, the bath cools 5oF/min. Sketch the actual and
the measured bath temperatures, assuming that the heater has no lag but that the
thermometer temperature changes according to the equation
F3O 3 RO u
(
FV
SELF-CHECK QUIZ
A State True/False:
1.1 The first known automatic control system is the flyball governor.
1.2 The flyball governor was installed on Watts steam engine in 1800.
1.3 The manipulated variable is the variable used to maintain the controlled variable
at its set point.
1.4 The sensor is a device that converts a measurand (variable to be measured) into
a signal in a different form.
96 Introduction to Process Control Systems
1.5 The output of the sensor may be force, displacement, voltage, electrical resistance
or other physical quantity.
1.6 The transfer function of a control system is the relationship between the outut
signal and the input signal and defined as the ratio of the input signal divided by
the output signal.
1.7 The dead-time of a system is the property of a physical system by which the
response to an applied force is delayed in its effect.
1.8 Dead time occurs in the transportation of mass or energy along a particular path,
i.e., the time taken for movement of a mass from one point to another.
1.9 Load disturbances are the flow rates, temperatures or compositions of streams
entering (but sometimes leaving) the process.
1.10 The degree of freedom of a process represents the maximum number of
independently acting automatic controllers that can be placed on that process.
1.11 In an open-loop control system, the input has good control over the output
(controlled variable).
1.12 Feed-forward control is used to maximize the deviation of the controlled variable.
1.13 Feed-forward control acts after the output is disturbed.
1.14 In a cascade control configuration, there is one manipulated variable and only one
measurement.
1.15 In computed variable control systems, the variable is controlled by computing its
value from other measurements.
1.16 In proportional control, the output of controller is proportional to the error.
1.17 In proportional-plus-derivative (PD) control, the derivative (or rate) action causes
the controller output to vary with respect to constant rate of change of the
actuating error signal.
1.1 The variable used to maintain the controlled variable at its set point is called
(a) Controlled variable (b) Manipulated variable
(c) Set-point (d) Process
1.2 The variable that must be controlled or maintained at some desired value is called
(a) Controlled variable (b) Manipulated variable
(c) Set-point (d) Process
1.3 The desired value of the controlled variable that must be maintained all time is
called
(a) Controlled variable (b) Manipulated variable
(c) Set-point (d) Process
1.4 The sensor is a device in which
(a) It converts a measurand into a signal in a different form
(b) The input to the sensor is the measurand and the output is the signal produced
by the sensor
(c) The output of the sensor may be force, displacement, voltage, electrical
resistance or other physical quantity
(d) All of these
Self-Check Quiz 97
1.5 The difference between the high and low value of the range of a parameter is
known as
(a) Range (b) Span
(c) Both (a) and (b) (d) None of these
1.6 The range of a temperature sensor/transmitter that has been calibrated to
measure a process temperature between the values of 850°C to 1200°C is
(a) 850°C to 1200°C (b) 350°C
(c) 850°C (d) 1200°C
1.7 The transfer function consists of
(a) Size relationship between the input and the output signals
(b) Timing relationship between the input and the output signals
(c) Both (a) and (b)
(d) None of these
1.8 Dead time is also called
(a) Pure delay (b) Transport lag
(c) Distance-velocity lag (d) All of these
1.9 In an industrial process, the typical process variables are
(a) Temperature (b) Pressure
(c) Level (d) All of these
1.10 In an open-loop control system, the input has
(a) No control over the output (controlled variable)
(b) Good control over the output (controlled variable)
(c) Optimized control over the output (controlled variable)
(d) All of these
1.11 In a closed-loop control system, the input has
(a) No control over the output (controlled variable)
(b) Good control over the output (controlled variable)
(c) Both (a) and (b)
(d) None of these
1.12 A controller is used to
(a) Reduce the error by generating a correction signal to the final control element
(control valve)
(b) Eliminate the disturbance by generating a correction signal to the final control
element (control valve)
(c) Both (a) and (b)
(d) None of these
1.13 A proportional (P) controller
(a) Continuously adjusts the manipulated variable so that the input to the process
is approximately balanced with the process demand
(b) Continuously adjusts the controlled variable so that the input to the process is
approximately balanced with the process demand
(c) Both (a) and (b)
(d) None of these
98 Introduction to Process Control Systems
1.1 The first known automatic control system was the_________, which was installed
on _________ steam engine in _________.
1.2 Three main examples of final control element used in process control systems are
(a) __________, (b) ____________ and (c) ___________.
1.3 The sensor is a device that converts a ___________ into a signal in a different form.
1.4 The output of the sensor may be any ________________.
1.5 Usually, a signal __________ is required to convert the sensor output into an
electrical or mechanical (or pneumatic) signal suitable for use by a controller or
display device.
1.6 The transfer function of a control system is the relationship between the _______
signal and the _______ signal.
1.7 The transfer function consists of two parts, (a) _______relationship between the
input and the output signals and (b) between the input and the output signals.
1.8 The dead-time of a system is the property of a physical system by which the
response to an applied force is ______ in its effect.
1.9 In a closed-loop (or feedback) control system, the input has control over the
_____________.
1.10 In an open-loop control system, the input has _____ control over the output
(controlled variable).
Self-Check Quiz 99
1.11 In feed-forward control system, __________ are measured and compensated for
them before the controlled variable deviates from set point.
1.12 Feed-forward control is used to ___________ the deviation of the controlled
variable.
1.13 The major limitation to feed-forward control is its instability to reduce steady-state
offset to ________.
1.14 Ratio control system is a special type of _________control system.
1.15 Override control (also called selective control) is a form of _________ control in
which a ___________ variable can be set at any time by one of a number of
different controlled variables.
1.16 In proportional-plus-integral (PI) control, the output is proportional to a linear,
combination of the _____ error and _______ of the input error.
1.17 The proportional-plus-integral-plus-derivative (PID) control is also called
_______controller.
2
2.1 INTRODUCTION
As discussed in Chapter 1, a process is a sequence of chemical, physical or biological
activities for the conversion, transport, or storage of material or energy. A process or
plant may be viewed from an equipment, product or operations viewpoint. Process
engineers make use of models of these process or plant during design and construction
to incorporate all viewpoints. As per the S88.01 standard (ISA 1995), there are three
different models of processes, which represent real-time aspects of a process: process
model, physical model and control model.
It is important that a framework be used for the development of a model, which
is adequate and accurate to depict all the viewpoints of a process.
This chapter reviews the procedures needed to obtain the models necessary for a
process control system design. Many types of models are available to the control system
designer, and here these types are compared and contrasted.
Process
Process stage
Process operation
Process action
for a continuous distillation process could be start-up, recycling, process feed, and
shutdown.
Each process operation is further subdivided into one or more process actions.
According to S88.01, process actions describe minor processing activities that are
combined to make up a process operation. An example of process actions for the start-
up operation of a distillation stage could be fill bottoms with material, initiate bottoms
heat exchanger temperature control, wait for bottoms heat exchanger to reach
minimum temperature, initiate condenser temperature control, wait for condenser
temperature to reach maximum temperature, increase bottoms temperature set-
points, and wait for column pressure to reach operating point.
Enterprise
Site
Plant area
Process cell
Unit
Equipment
module
Control
module
Process
equipment
hierarchy. In the case of the bottom two levels, a grouping in the level may be
incorporated into another grouping at that same level. The lower four layers refer to
specific process types.
A process type, as shown in Figure 2.2, is a collection of physical processing and
control process grouped for a specific purpose. During engineering activities on a project,
the process at a lower level is grouped to form a higher-level process grouping. This
grouping simplifies the operation of that process and treats it as a single larger piece
of process. Once created, the process cannot be split up except by reengineering the
process at that level.
Even though the physical assets of an enterprise seem to be organized in a
hierarchical manner. In reality (practically), they are organized on a geographical basis,
rather than in a strict hierarchical manner, as shown in Figure 2.3. The physical model
Control Model 103
Enterprise
Process unit 1
Process cell 1
Process unit 2
Process unit 2
Process unit 1
Process cell 2
Process unit 2
Procedure
Unit procedure
Operation
Phase
As for the physical model, the control procedures can be organized hierarchically,
as shown in Figure 2.5. Smaller groupings are combined to form larger entities in the
model. The control model describes the control procedures of an enterprise in terms of
enterprise management, site management, plant area management, process manage-
ment, unit supervision, process control and safety protection.
The control procedures may also be organized geographically in the same manner
as the physical model of Figure 2.3. However, below the unit supervision level, it may
be difficult to enforce a strict geographical division when some process control objects
are shared among units. The control activity model provides an overall perspective
of the major activities involved in a plant-wide process control and their relationships.
This model is focussed toward the lower layers in the hierarchical control model of
Figure 2.5 because these layers are the typical domain of the control engineers. The
relationships between the activities are defined by the information flowing between
them.
Enterprise
management
Site
management
Plant-area
management
Process
management
Unit
supervision
Process
control
Final control
not need to be complex. For many systems, a simple empirical model that describes the
gross behaviour of the system is more than sufficient to design the control system of
a process.
The development of a process model is not just a matter of writing down as many
or few equations that come into ones mind. It is important that a framework be used
106 Process Control Modelling
for the development of a process model which is adequate and accurate for the purpose.
While developing a process model following important questions should be answered:
(a) What are the controlled variables? Proper regulation of these variables is the
ultimate goal of the control system. These variables may not be directly
measurable but is inferred from other measurements.
(b) What are the measured variables? These variables are those that can be
directly measured and thus used by the control algorithm. Usually, they are
quantities like flow, pressure, temperature and concentration. These variables
are somehow related to the controlled variables.
(c) Which variable can be manipulated? Control inherently involves variables
that can be adjusted in order to affect a change in the measured variables.
(d) What is the expected effectiveness of the control system? In order to
implement the control properly, the model must determine the following
aspects of the process:
· Sign and magnitude changes: When a manipulated variable is increased,
does the controlled variable increase or decrease?
· Speed of response: Does the controlled variable change rapidly or slowly
when the given manipulated variable changes? The answer to this question
determines what kind of controller manipulations can be tolerated by the
process.
(e) How sensitive is the system to changes in the operating point? Industrial
control systems often experience changes in the operating conditions and
equipment performance.
Following additional questions may be asked for further refinement of the decision
about selecting a right process model for desired control problem:
· What should be the form of model?
· How can the model be solved?
· Is the model behaving correctly?
· Is the model fidelity adequate?
Optimal control
In the optimal control application, we are interested in the optimum way of moving from
one steady operating point to another with the least expenditure of energy or the least
amount of bad quality product being made. There are many criteria which can be used
to define the optimal path.
Mathematical models
A mathematical model of a process is a system of equations whose solution, given
specific input data, is representative of the response of the process to a corresponding
set of inputs. Mathematical models are simple or complex, as dictated by the intended
use.
A mathematical model is a mathematical expression that describes the important
relationships between the input and output of a system or component. Mathematical
modelling is one of the important steps in the analysis and design of control systems.
It is the foundation of practical control techniques. Applying the many methods of
controls to improve command response, stability and disturbance rejection requires a
thorough understanding of the objects under control. How does a disturbance couple into
the plant? What delay does the feedback device inject? How will the power converter
limit the responsiveness of the system? To answer these questions, we need an accurate
mathematical description (model) of the system.
Most techniques used in the analysis of control problems are dependent on the
existence of descriptive mathematical equations. The mathematical models are obtained
by application of the physical laws pertaining to the nature of the components of control
systems. These components can be mechanical, electrical, hydraulic, pneumatic and
thermal elements that are commonly encountered in control systems as components of
the system to be controlled and as components of actuators, sensors and controllers.
These models are based on fundamental theories or laws, such as conservation of mass,
energy and momentum.
FZ
H
Z WFV (2.1)
FV
becomes the steady-state equation:
0 = f (x, u, d, t) (2.2)
Steady-state models of the process are often used in process optimization and
certain control algorithms, while dynamic models predict the time-varying behaviour of
the process and find in nonlinear control algorithms.
Consider a simple tank system, as in Figure 2.6, where a fluid is being pumped into
a tank through a valve (volumetric flowrate F1) and the outlet flow F2 passes through
another valve. The pressure at the discharge of the pump (P1) is fixed, as well as the
outlet pressure (P3) of the discharge line. The liquid surface is at the atmospheric
pressure (Patm).
We can write down a set of equations which describes the dynamic behaviour of
the tank. A mass balance on the liquid gives:
Process Modelling 109
P1, F1
Patm
h
P2
P 3, F 2
F
S8 F
#SJ
S ( S ( (2.3)
FV FV
or
FJ
S# S ( S (
FV
or
FJ
# ( ( (2.4)
FV
plus the algebraic equation for inlet and outlet flows:
( % 2 2 (2.5)
( % 2 2 (2.6)
with the definition of tank pressure head:
2 2CVO S IJ (2.7)
It can be noted that in the above model, there are no control loops on the system,
or if there were, the loops are not connected. This type of model leads to an open-loop
simulation of the process. In this case, where the control loops are activated, is called
a closed-loop simulation.
When the accumulation is zero, the balance results in an algebraic equation. For
a non-zero accumulation, this balance results in a differential equation, which is
generally written as
#EEWOWNCVKQPQH
: YKVJKP U[UVGO
(NQYQH:
KPVQU[UVGO
(NQYQH:
QWVQHU[UVGO
6KOG RGTKQF 6KOG RGTKQF 6KOG RGTKQF
#OQWPV QH :
IGPGTCVGFKPU[UVGO
#OQWPVQH:
EQPUWOGFKPU[UVGO (2.8)
6KOG RGTKQF 6KOG RGTKQF
where X is one of the fundamental quantities such as total mass, mass of a chemical
component, energy or momentum.
Other constitutive equations (further discussed in detail in Section 2.13.3) may be
used to completely determine the model, such as
Q = hA(DT) (heat transfer)
rA = k0eE/RTCA (chemical reaction rate)
PV = nRT
'2
F = cv (fluid flow)
S
yi = ki xi (phase equilibrium)
The model has the proper number of equations when the behaviour of the system
can be predicted from the model. A correctly formulated model has no degrees of
freedom. The concept of degrees of freedom is expressed as
DOF = NV NE
where
DOF = Degrees of freedom
NV = Number of variables
NE = Number of equations
Note that NV represents the number of variables in the system and does not
include the constant parameters or external stimuli. If NV is greater than NE, then the
system is underspecified and the model must be corrected either by including more
appropriate equations or by correctly designing a variable as a specified parameter or
external stimuli. If NE is greater than NV, then the system is overspecified and in
general no unique solution exists. In this situation, there are one or more dependent
equations or constant parameters that ought to be designated as variables.
However, a system model expressed as a set of (usually nonlinear) differential
equations is not immediately useful for a control system design. One could use a
nonlinear control algorithms, but these are difficult to develop and are usually overly
complicated. In order to design a control system, it is generally useful to model the
system as a set of linear differential equations. The control system designer has three
options:
Process Modelling 111
(a) Simulate the nonlinear system on a computer and numerically compute its
solution
(b) Develop a linear model that approximates the dynamic behaviour of the system
in the neighbourhood of a specified operating point
(c) Transform the nonlinear system into a linear system by an approximate
transformation of variables.
The first-two options are generally always feasible. However, the second method
is most useful for a control system design, and it is the approach used by most control
system designers. The first option is still useful for a control system design testing and
checkout before trying it on the real process. The third option can be used in a few
cases.
Gain-only process. When the response of the process variable to a change in the
manipulated variable is essentially instantaneous (as shown in Figure 2.7), the process
is called a gain-only process.
Process
variable
Manipulated
variable
Time
Figure 2.7 Response of gain-only process.
deadtime in the controlled variable response. In this case, the process is called a gain-
plus-deadtime process. The transfer function of a gain-plus-deadtime process is given by
GGDT(s) = K exp (sqD) (2.10)
Examples of gain-only processes are speed control of centrifugal compressors, liquid
flow control with a fast actuators, etc.
First-order process. When the response of the process variable to a step change in the
manipulated variable is represented as in Figure 2.8, the process is called a first-order
lag process.
Process
variable
Manipulated
variable
Time
Figure 2.8 Response of a first-order lag process.
F, T, cA
-
)(1
U (2.11)
U U
Process
variable
Manipulated
variable
Time
where
K = Gain
t1,t2 = Time constants
qD = Process deadtime.
Examples of second-order processes are temperature process where the lag of the
temperature sensor and thermowell is significant when compared to the process lag,
heat exchanger, cascaded tanks, etc.
Process
variable
Manipulated
variable
Time
Figure 2.11 Response of a second-order underdamped process.
Manipulated
variable
Time
Figure 2.12 Response of a second-order underdamped process.
T (x, y, t)
x
Figure 2.13 Two-dimensional body.
6 6
- Z - [ (2.17)
Z [
One should be aware that to solve the dynamic distributed parameter problem, the
following information must be known:
Process Modelling 117
(a) The values of the temperature T(x, y, t) at t = 0 over the whole region
(b) The values of the temperature at the boundary of the body.
The first are the initial conditions of the problem, the second are known as the
boundary conditions of the problem. A number of different types of boundary
specifications can be given.
Typical examples of distributed parameter systems are tubular reactors, heat
exchangers, and packed columns. Here, the variation in the states in one or more
directions is significant and cannot generally be neglected.
Often it is not necessary to consider the spatial variation of a state variable, such
as temperature or concentration in a well-stirred tank, where the state is a uniform
value. In this case, we can lump together various values of the state and say that there
is only one representative value. In this case, we derive a lumped parameter model of
the process which is represented by ordinary differential equations. Example of a simple
tank system of Figure 2.6 is a lumped parameter model.
FZ
14243
Z #
VZ $
VW &
VF
FV (2.18)
y = C(t)x
(b) The linear time invariant (LTI) model is given by
123
Z = Ax + Bu + Dd (2.19)
y = Cx
Both the types of models frequently appear in process control applications.
digital controllers arises because the signals are discrete-time signals. In addition, the
accuracy of the discrete-time signal values is finite. However, the accuracy of the
discrete-time signals is smaller than the accuracy of the sensors and actuators, and so
the accuracy of discrete-time signals is considered to be identical to the accuracy of
continuous signals.
In the case of continuous models, we have sets of differential equations, whereas
with discrete models we have difference equations. These often arise when we are
dealing with digital systems which sample data at regular specified points in time. In the
discrete case, the form of the model equations is given by the difference equation
x(k + 1) = Ax(k) + Bu(k) + Dd(k)
123
(2.20)
y(k) = Cx(k)
Here the new value of the states at time (k + 1) is computed from the previous values
at time (k).
FP [ F P [
CP
V
FV P
CP
V
FV P
! C
V F[
FV
C
V [
V
T
V (2.22)
where the coefficients ai(t) are not functions of y(t). If these coefficients (ai) are
constants, the equation is linear with constant coefficients. Many engineering systems
can be modelled by these equations. Systems modelled by these equations are known
as linear time-invariant (LTI) systems.
A differential equation does not, in general, have a unique solution unless some
additional constraints (or conditions) on the solution are known. A function y(t) has a
unique derivative dy/dt, but for a given derivative dy/dt, there are infinite possible
functions y(t). If we are given dy/dt, it is impossible to uniquely determine y(t) unless
additional piece of information about y(t) is given. For example, the solution of a
differential equation
F[
(2.23)
FV
Differential Equations 119
= bm f (k + m) + bm1 f (k + m 1) + L + b f (k + 1) + b f (k)
1 0 (2.25)
= bn f (k + n) + bn1 f (k + n 1) + L + b f (k + 1) + b f (k)
1 0 (2.26)
where some of the coefficients on both the sides can be zero. However, the coefficient
120 Process Control Modelling
of y(k + n) is normalized to unity. Equation (2.26) is valid for all values of k. Therefore,
the equation is still valid if we replace k by k n throughout the equation. This yields
the alternative form (the delay operator form) of Eq. (2.26) as
y(k) + an1 y(k 1) + L + a y(k n + 1) + a y(k n)
1 0
Linearity theorem
It is very important to realize that the Laplace transform is a linear operator. It can be
easily verified from the defining Eq. (2.28) that
L{ f1(t) + f2(t)} = F1(s) + F2(s) (2.29)
and
L{kf (t)} = kF(s) (2.30)
where k is a constant. Hence, it is verified that the Laplace transformation is a linear
transformation.
Laplace Transform 121
. ^Ô V
`
H
V FV
(
U
U
(2.34)
NKO H
V NKO U(
U
V U (2.36)
122 Process Control Modelling
Q L Ô
H
V . ^ (
U` (
U G UV
FU V (2.40)
E L
F%
V
Z Z %
V [4
V (2.43)
FV
Laplace Transform 123
where
C(t) = c(t) c(0) and R(t) = r(t) r(0)
are the deviation variables.
Equation (2.43) is the general linear first-order differential equation in terms of the
deviations of the input and output variables from their initial steady-state values. It has
three coefficients, x1, x0 and y, but without loss of generality, we can divide the equation
by one of the three so that we can characterize the equation by just two parameters.
In process control, it is customary to divide by the coefficient of the output variable, x0,
provided it is not zero. It results in a standard form of the linear first-order differential
equation, given as
F
U %
V %
V -4
V (2.44)
FV
where
Z
t = = Time constant
Z
[
K = = Steady-state gain
Z
UV FV
. >W
V@ Ô
G
U
(2.45)
#
. ^ #G B V ` Ô #G B V G UV FV (2.47)
U B
124 Process Control Modelling
0
0
t t t
t = 0 t = 0 t = 0
(a) Unit step (b) Ramp (c) Decaying exponential
¥
H 1.0
0 0
1.0
1.0 t t t
t = 0 t = T t = 0 t = 0
(d) Pulse (e) Unit impulse (f) Sinusoidal
Figure 2.14 Common test inputs for Laplace transform.
GKX V G KX V
UKP
X V (2.50)
K
where i = Ö1 is the unit of imaginary numbers. By substituting this into Laplace
transform equation (2.28), the Laplace transform of sinusoidal function can be derived
as
GKX V G KX V UV
. >UKP
X V@ Ô G FV
K
U KX V
Ô
ËÍ G G
U KX V ÛÝ FV
K
Ë G
U KX V G
U KX V Û
Ì Ü
K Í U KX U KX Ý
È Ø
É
K Ê U KX U KX ÙÚ
KX
K U X
X
(2.51)
U X
F E
V FE
V
Z
Z Z E
V [T
V \ (2.52)
FV FV
where c(t) is the output or dependent variable, r(t) is the input variable, t is time (the
independent variable), and x1, x0, y, z are constants.
Assuming the initial conditions are at steady state, the equation at the initial
conditions is
x0 c(t) = yr(0) + z (2.53)
Subtracting Eq. (2.53) from Eq. (2.52), results in
F%
V F%
V
Z
Z Z%
V [4
V (2.54)
FV FV
where
C(t) = c(t) c(0) and R(t) = r(t) r(0)
are the deviation variables. By definition, the initial conditions of the deviation variables
are zero.
The four parameters in Eq. (2.54): x2, x1, x0 and y, can be reduced to three by
dividing the entire equation by any of them, provided it is not zero. In process control,
126 Process Control Modelling
the standard form of the second-order equation is obtained by dividing coefficient x0,
provided it is not zero. The resulting equation in the standard form is given as
F F
U
%
V [U %
V %
V -4
V (2.55)
FV FV
where
Z
t = = Characteristic time
Z
Z Z
z = = Damping ratio
U Z Z Z
[
K= = Steady-state gain
Z
and
D & D
)
& (2.59)
C & C & C & C
Transfer Function Representations 127
where G(D) is a transfer operator. The relationship of Eq. (2.59) can be represented in
a block diagram form, as shown in Figure 2.15.
r(t) c(t)
b1 D b0
a3 D a2 D 2 a1 D a0
3
R(s) b1 s b0 C(s)
a3 s a2 s 2 a1 s a0
3
The transfer function G(s) relating the output C(s) to the input R(s) in the Laplace
domain is given by
C(s) = G(s)R(s) (2.63)
or
%
U
)
U (2.64)
4
U
where
D U D
)
U (2.65)
C U C U C U C
NKO %
V NKO = U%
U?
V U
NKO %
V NKO = U)
U 4
U?
V U
NKO %
V NKO )
U NKO U4
U
V U U
NKO %
V NKO )
U NKO 4
V
V U U
This means that the change in the output variable after a very long time, if bound,
can be obtained by multiplying the transfer function (evaluated at s = 0) with the final
value of the change-in input.
X4(s)
X(s)
X1(s) +
+
X(s) X(s)
R(s) C(s)
G(s)
X5(s) X(s)
C(s) = G(s)R(s)
X2(s) X3(s)
(a) Block (b) Summer (c) Junction
Figure 2.17 Components of a block diagram.
Equation (2.66) has the transmittance (proportionality function), containing in the block,
as shown in Figure 2.18, as
%
U
6
U (2.67)
4
U U U
R(s) 4 C(s)
s2 4 s 13
This becomes one block in Figure 2.19. If the system compares the desired
temperature r to the actual temperature c, the error (e) results, where
e = r c and E(s) = R(s) C(s)
The difference is created by using junction to produce the measurement of c for
comparison with r and then using a summer with one negative sign to produce e, as
shown in Figure 2.19. Finally, suppose the furnace operates upon e so as to modify the
heat applied to the object, where
FZ
CZ CG
FV
It will create a transmittance
:
U C
6
U
(2.69)
'
U U C
Now this completes Figure 2.19 which contains two blocks, one summer, and one
junction.
2.9.3 Linearization
A major difficulty in analyzing the dynamic response of many processes is that they are
nonlinear. That is, they cannot be represented by linear differential equations. A linear
differential equation consists of a sum of terms each of which contains no more than one
variable or derivative, which must appear to the first power. Even Laplace transform
technique can be used for analyzing only linear systems. There is no comparable
technique by which we can analyze the dynamics of a nonlinear system and generalize
the results to represent similar physical systems.
Linearization is a technique which is used to approximate the response of nonlinear
systems with linear differential equations that can then be analyzed by Laplace
transforms. The linear approximation to the nonlinear equations is valid for a region
near some base point around which the linearization is made. It is widely used in the
study of process dynamics and design of control systems for the following reasons:
1. A closed-form, analytical solutions can be obtained for linear system. Thus a
complete and general picture of a processs behaviour can be obtained
independently of the particular values of the parameters and input variables.
This is not possible for nonlinear systems. A computer simulation provides us
only with the behaviour of the system at specified values of inputs and parameters.
2. All the significant developments toward the design of effective control systems
have been limited to linear processes.
Transfer Function Representations 131
Let us first understand, what is a linear and nonlinear system? A linear system is
one that contains variables only to the first power in any one term of the equation. If
square roots, squares, exponential, products of variables, etc., appear in the equation,
it is a nonlinear system. An example of a linear system is
FZ
C
C Z H
V (2.70)
FV
where a0 and a1 are constants or functions of time only, not of dependent variables or
their derivatives. Mathematically, a linear differential equation is one for which the
following two properties hold true:
1. If x(t) is a solution, then cx(t) is also a solution, where c is a constant.
2. If x1 is a solution and x2 is also a solution, then (x1 + x2) is a solution.
Consider some example of a nonlinear system as follows:
FZ
C C Z
H V (2.71)
FV
FZ
C C Z H
V (2.72)
FV
FZ
C C G Z H
V (2.73)
FV
FZ
C Z
V Z
V H
V
C (2.74)
FV
where both x1 and x2 are dependent variables.
Linearization is quite straightforward in which the nonlinear functions are
expanded in Taylor series around the steady-state operating level, and all terms after
the first partial derivatives are neglected.
Let us assume that we have a nonlinear function f of the process variable x1 and
x2: f(x1, x2) wherein x1 could be mole fraction or temperature or flow rate. The nonlinear
differential equation is given as
FZ
H
Z (2.75)
FV
Expanding the nonlinear function f(x) into a Taylor series around its steady-state value
x0, we get
H
Z Z H
Z
È FH Ø Z Z
H
Z É Ù
Ê FZ Ú Z
È F H Ø
Z Z
É Ù
Ê FZ Ú Z
!
È F P H Ø
Z Z P
É P Ù
Ê FZ Ú Z P
! (2.76)
Linearization consists of truncating the series after the first partial derivatives. If
we neglect all terms of order two and higher, we take the following approximation for
the value of f(x):
132 Process Control Modelling
È FH Ø
H
Z H
Z É Ù
Z Z (2.77)
Ê FZ Ú Z
It is well known that the error introduced in the approximation in Eq. (2.77) is of
the same order of magnitude as the terms
È F H Ø
Z Z
+ É Ù (2.78)
Ê FZ Ú Z
Consequently, the linear approximation in Eq. (2.77) is satisfactory only when x is
very close to x0, where the value of the terms I is very small.
The process is graphically sketched in Figure 2.20, wherein we can see the
nonlinear function f(x) and its linear approximation around x0. It is also clear from the
figure that the linear approximation depends on the location of the point x0 around
which we make the expansion into a Taylor series. As shown in Figure 2.20, the
approximation is exact only at the point of linearization.
f(x) § df ·
Slope = ¨ ¸
© dx ¹ x0
Nonlinear f(x)
x
x0
Figure 2.20 Linearization of a nonlinear function.
CCA0
A0
Upstream a Downstream
plant CA1
A1 plant
a
The feed composition to a reactor varies with an amplitude larger than acceptable
for the reactor. It is not possible to alter the upstream process to reduce the oscillation
in the feed, therefore, a drum is located before the reactor to reduce the feed
composition variation, as shown in Figure 2.21. If the input variation is slow, with a
period of once per year, the output response will be essentially at its steady-state value
(the same as the input), with the transient response being insignificant. If the input
changed very rapidly, say every nanosecond, the output would not be significantly
influenced, that is, its output amplitude would be insignificant. Finally, if the input
varies at some intermediate frequency near the response time of the process, the output
will continuously fluctuate at values significantly different from its mean value. The
behaviour at extreme frequencies is easily determined, but the method for determining
the system behaviour at intermediate frequencies is not obvious and is useful for the
design process equipment, selection of operating conditions and formulation of control
algorithms to give desired performance.
For general frequency response analysis, periodic inputs will be limited to sine
inputs, which will be a mathematically manageable problem. Also, only the long-time
response (i.e. after the initial transient, when the output is periodic) is considered. The
periodic behaviour after a long time is sometimes referred to as steady state. However,
it seems best to restrict the term steady state to describe the systems with zero-time
derivatives. The periodic behaviour of the input and output after a long time and the
frequency response is shown in Figure 2.22.
Y ¢ Outlet from system
P¢
P
X ¢ Inlet to system
Time
Figure 2.22 Frequency response for a linear system.
134 Process Control Modelling
1. Root locus and Bode plot techniques are commonly taught as part of Control Engineering
subject. Their details have discussed in Chapter 4 of this book.
Time-domain Modelling 135
either the on or the off position, and thus the state of the switch can assume one of two
possible values. Thus, if we know the present state (position) of the switch at t0 and if
an input is applied, we are able to determine the future value of the state of the
element.
State variables are very popular in electrical and mechanical engineering control
problems, which tend to be of lower order (fewer differential equations) than chemical
engineering control problems. Transfer function representation is more useful in
practical process control problems because the matrices are of lower order than would
be required by a state variable representation.
State variables appear naturally in the differential equations describing chemical
engineering systems because our mathematical models are based on a number of first-
order differential equations, component balances, energy equations, etc.
144424443
where Z FZFV. Thus, this set of simultaneous differential equations can be written
in matrix form as follows:
È Z Ø È C C ! C P Ø È Z Ø
!
F Z Ù
É
É Ù
ÉC
É
C ! ÈD
C P Ù É Z Ù É
ÙÉ Ù "
D O Ø È W Ø
" ÙÉ " Ù (2.81)
FV É " Ù É " " " ÙÉ " Ù É ÙÉ Ù
ÉÊ Z ÙÚ ÉÊ C Ù É Ù Ê DP ! DPO Ú Ê WP Ú
P P CP ! CPP Ú Ê Z P Ú
The column matrix consisting of the state variables is called the state vector and
is written as
È Z Ø
ÉZ Ù
Z É Ù
É "Ù (2.82)
ÉÊ Z ÙÚ
P
where the boldface indicates a vector. The vector of input signals is defined as u. Then
the system can be represented by the compact notation of the state differential equation
(or state equation) as
Z #Z $W (2.83)
Time-domain Modelling 137
The quantity S can be any of the fundamental quantities such as total mass, mass
of individual components, total energy, or momentum and are given as
F
S8
Total mass balance
FV
Ç S ( Ç
K KPNGV
K K
L QWVNGV
S L (L (2.86)
FP # F
E # 8
Mass balance on component A
FV FV
Ç E# (K Ç
K
E# L (L T8
(2.87)
K KPNGV L QWVNGV
F'
Total energy balance
FV
F
7 -' 2'
FV
Ç
K KPNGV
SK (K JK Ç
L QWVNGV
S L ( L J L v3 v9U (2.88)
where
r = Density of the material in the system
ri = Density of the material in the ith inlet system
rj = Density of the material in the jth outlet system
V = Total volume of the system
Fi = Volumetric flow rate of the ith inlet system
Fj = Volumetric flow rate of the jth outlet system
nA = Number of moles of component A in the system
cA = Molar concentration (moles/volume) of A in the system
cAi = Molar concentration of A in the ith inlet system
cAj = Molar concentration of A in the jth outlet system
r = Reaction rate per unit volume for component A in the system
hi = Specific enthalpy of the material in the ith inlet system
hj = Specific enthalpy of the material in the jth outlet system
U = Internal energy of the system
KE = Kinetic energy of the system
PE = Potential energy of the system
Q = Amount of heat exchanged between the system and its surroundings per
unit time
Ws = Shaft work exchanged between the system and its surroundings per unit time.
By convention, a quantity is considered positive if it flows in the system and
negative if it flows out of the system. The state equations with the associated state
variables constitute the mathematical model of a process, which yields the dynamic or
static behaviour of the process. The application of the conservation principle as defined
in Eqs. (2.86)(2.88) will yield a set of differential equations with the fundamental
z-Transform 139
quantities (as dependent variables) and time (as the independent variable). The solution
of the differential equations will determine how the fundamental quantities, or
equivalently, the state variables, change with time. That is, it will determine the
dynamic behaviour of the process.
If the state variables do not change with time, we say that the process is at steady
state. In this case, the rate of accumulation of a fundamental quantity S per unit of time
is zero, and the resulting balances yield a set of algebraic equations.
2.12 z-TRANSFORM
As discussed in Section 2.8, Laplace transform is a powerful tool for the solution of
linear differential equations and is particularly useful when these equations represent
the dynamic behaviour of continuous systems. The z-transformation is used for certain
kinds of discontinuous system (also called discrete-time systems), whose dynamic
behaviour can be defined by linear difference equations. z-Transform is particularly
applicable to the study and design of sampled data control.
Let us consider the operation of an ideal, periodic sampler, as shown in
Figure 2.24. The sampler converts a continuous signal f (t) into a discrete signal f *(t) at
f(t)
Dt
0 1 2 3 4 5 6 7 8 9
Time, n
f*(t)
Dt
0 1 2 3 4 5 6 7 8 9
Time, n
Figure 2.24 Sampled data impulse representation of continuous signal f (t).
140 Process Control Modelling
In practice, impulse sampling is not attainable because the sampler remains closed
for a small but finite amount of time. However, the time of closure is usually small (i.e.
microseconds) compared to the sampling period and, consequently, impulse sampling
provides a suitable idealization.
Now, let us consider the Laplace transform of equation (2.89), F*(s). The value of
f(nDt) is considered to be a constant in each term of the summation and thus is invariant
when transformed. Since L[d (t)] = 1, it follows from the Real Translation Theorem that
the Laplace transform of a delayed unit impulse is L{d (t nDt)} = enDts. Thus, the
Laplace transform of Eq. (2.89) is given by
(
U Ç H
P ' V G P '
P
VU
(2.91)
By introducing the change of variable, the z-transform, F(z), of both f *(t) and f (t)
can be defined as
(
\ < ] H
V_ Ç H
P 'V \ P (2.92)
P
To simplify the notation, denote f(nDt) by fn. Then Eq. (2.92) can be written as
(
\ < ] H
V _ Ç HP \ P (2.93)
P
NKO H
P'V NKO (
\ (2.97)
P \
NKO H
P'V NKO
\ (
\ (2.98)
P \
Long division
Long division provides a second method for obtaining a inverse z-transform. In most
cases, it is considerably easier to use this method to obtain the inverse z-transform than
to use partial fraction expansion. However, the result (an infinite series) may not be as
useful as an analytical expression. Inversion via long division is an operation unique to
discrete-time systems; no analogous method exists for continuous-time systems.
Contour integration
This method for inverting z-transform utilizes a contour integral, where the contour
must be appropriately specified. Although, the integral can be evaluated using the
residue theorem, this method is seldom used in practice.
Î V
H
V Ï
ÐW
V V
By definition,
<]W
V_ (
\ Ç W
P6 \ P Ç \ P
P P
or
F(z) = 1 + z1 + z2 + z3 +
(2.100)
For |z| > 1, this infinite series converges, and yields
(
\ (2.101)
\
Note that |z| > 1 corresponds to zsDt > 1 (or s > 0).
ÎÑ V
H
V Ï CV
ÑÐ G V
z-Transform 143
By definition, we have
< ^G CP'V ` (
\ Ç H
P'V \ P Ç %G CP'V \ P (2.102)
P P
Since Eq. (2.102) is a power series in eaDtz1 that converges for |eaDtz1| < 1 (which
implies that s > a), we get
%
F(z) = (2.103)
\ G C'V
Î V
H
V Ï
Ð MV V
By definition, we get
< ^MV` = F(z) Ç H
P'V \ P (2.104)
P
Ç M
P'V \ P
P
MV\
\ (2.105)
Î V
H
V Ï
ÐUKP
CV V
Because
f(t) = sin (at), f(nt) = sin (ant)
and
G LPCV G LPCV
UKP
CPV
L
144 Process Control Modelling
< ^UKP
CV` (
\ Ç H
CPV \ P
P
Ç UKP
CPV \ P
P
Ç
G LPCV G LPCV \ P
P L
Ë LCPV P
Û
Ç
Ì
G \
G LCPV \ P Ü
Ç (2.106)
L ÍÌ P P ÝÜ
Therefore,
È Ø
< ^UKP
CV` É LCV
LCV Ù ((2.107)
L Ê G \ G \ Ú
model must be verified in the laboratory. The assumptions should be checked and
testing must be done for programming errors. All models include assumptions, which,
if correct, increase the likelihood that the model will provide unreliable results. Another
reason to verify is that models are in a form of software, and like any software, models
can have programming errors. Any useful model must be verified against an actual
control system.
A general method as an aid to learning and applying modelling skills, is
represented in Table 2.1. However, the process control engineers are free to adapt the
procedure to the needs of particular problems. It is worth noting that the steps could
F0
CAO
F1
V CA
Let us assume that in the example of Figure 2.25, the system is the liquid in the
tank. Also, the tank has been designed well, with baffling and impeller size, shape and
speed such that the concentration should be uniform in the liquid. Therefore, the
assumption and data in this example could be as follows:
Assumptions:
(a) Well-mixed vessel (tank)
(b) Density is the same for A and solvent
(c) Constant flow-in C
Data:
(a) F0 = 0.085 m3/min
V = 2.1 m3
CAinit = 0.925 mol/m3
DCAO = 0.925 mol/m3 = 1.85 mol/m3 after the step
(b) The system is initially at steady state (CAO = CA = CAinit at t = 0).
Note that the inlet concentration (CAO) remains constant after the step-change has
been introduced to this two-component system.
Conservation balance
Conservation balances are relationships that are obeyed by all physical systems under
common assumptions valid for industrial processes. The conservation equations most
often used in process control are the conservation of material, energy and momentum
(as discussed in Section 2.11.3). These conservation balances are often written in the
following general form for a stirred-tank heater system shown in Figure 2.26.
W
F1
F0
Q
Heater
Condensate Steam
Figure 2.26 General lumped-parameter system for stirred-tank heater.
148 Process Control Modelling
Constitutive balance
The constitutive equations provide relationships that are not universally applicable but
are selected to be sufficiently accurate for the specific system being studied. We often
find that an insufficient number of balances exist to determine all variables. In such
cases, additional constitutive equations are included to provide sufficient equations for
a completely specified model. Some examples of constitutive equations are as follows:
Heat Transfer, Q= hA(DT)
Chemical reaction rate, rA = kO eE/RT CA
Fluid flow, F= Cv(DP/r)1/2
Equation of state, PV = nRT
Phase equilibrium, yi = Ki xi
Modelling Procedure 149
In this problem, the level is never below the overflow, and the height above the
overflow (L LW) is very small compared with the height of the liquid in the tank (L).
Therefore, we will assume that the liquid level in the tank is approximately constant,
and the flows in and out are equal. That is,
F8
F0 = F1 = F or = F0 F1 = 0 (2.117)
FV
Therefore, V is constant.
The next step is to formulate a material balance on component A. Since the tank
is well mixed, the tank and outlet concentrations are the same and therefore,
component material balance can be formulated from Eq. (2.111) as
(Accumulation of component A) = (Component A in) (Component A out)
+ (Generation of component A)
or
(MWAVCA)t+Dt (MWAVCA)t = (MWAFCA0 MWAFCA)Dt (2.118)
with CA being moles/volume of component A and MWA being its molecular weight
and the generation term being zero, because there is no chemical reaction. Dividing
Eq. (2.118) by Dt and taking the limit as Dt ® 0 gives
F% #
/9# 8 /9# (
% # % # (2.119)
FV
One might initially believe that another balance on the only other component,
solvent S, could be included in the model:
F%5
/95 8 /95 (
%5 %5 (2.120)
FV
However, Eq. (2.115) is the sum of Eqs. (2.119) and (2.120). Thus, only two of the three
equations are independent. Therefore, only Eqs. (2.117) and (2.119) are required for the
model and should be considered in determining the degrees of freedom. For variables
CA and F1, external variables F0 and CA0, and Eqs. (2.117) and (2.119), the following
analysis shows that the model using only independent equations is exactly specified:
DOF = NV NE = 2 2 = 0
Note that the variable t representing time must be specified to use the model for
predicting the concentration at a particular time.
Ê U ÙÚ
The resulting equation can be solved directly from Eq. (2.121)
È F% # Ø F% # FGV U
F
G V U % #
%
GV U É
%# Ù GV U
% # # GV U
(2.122)
Ê FV U Ú FV FV FV U
or
%# GV U
%
Ô F
%# GV U
Ô
FV #
ÔG
V U
FV
U U
or
% U
% # GV U # GV U
U
or
CA = CA0 + Iet/t
Note that the integration was simplified by the fact, that is, CA0 is constant after
the step change (i.e. for t > 0). The initial condition is CA(t) = CAinit at t = 0, which can
be used to evaluate the constant of integration I = CAinit CA0. This formulation implies
that the time t is measured from the introduction of the step change. Therefore,
CA = CA0 + (CAinit CA0)et/t (2.123)
and
CA CAinit = [CA0 (CA0)init](1 et/tx)
152 Process Control Modelling
The final equation has used the extra relationship that (CA0)init = CAinit. Substituting
the numerical values gives
CA 0.925 = (CA0 0.925)(1 et/24.7x)
Two important aspects of the dynamic behaviour can be determined from Eq. (2.123).
The first is the speed of the dynamic response, which is characterized by the time
constant (t ). The second is the steady-state gain, which is defined as
'QWVRWV '% #
Steady-state gain, Kp =
' KPRWV '% #
Note that, in this example, the time constant depends on the equipment (V) and
operation of the process (F), and the steady-state gain is independent of these design and
operating variables. These values are not generally applicable to other processes.
2
Limit
1.5
concentration
Outlet
0 20 40 60 80 100 120
t
Time (min)
2
concentration
1
Inlet
0 20 40 60 80 100 120
Time (min)
Figure 2.27 Dynamic result for continuous-flow stirred-tank.
to learn a few values of this curve. The values for the change in concentration for
several values of time after the step are noted in Table 2.2.
The specific quantitative question posed in the goal statement (Section 2.13.1)
involves determining the time until 90% of the change in outlet concentration has
occurred. From Eq. (2.123), this time can be calculated by setting
CA = CAinit + 0.9(CA0 CAinit)
On rearrangement, it gives
=
% # KPKV % # ?
V U NP
OKP
% # KPKV % #
Note that this is the time from the introduction of the step change, which, since
the step is introduced at t = 10, becomes 68.8 in Figure 2.27. One should ask how
important the specification is; if it is critical, a sensitivity analysis should be performed.
For example, if the volume and flow are not exactly known and change within ± 5%
of their base values, the time calculated above is not exact. The range for this time
can be estimated from the bounds on the parameters that influence the time
constant:
/CZKOWO
V
OKP
/KPKOWO
V
OKP
Given the estimated inaccuracy in the data, one should wait at least 62.8 (not 56.8)
minutes after the step to be sure that 90% of the concentration change has occurred.
plotted in Figure 2.28 along with the model prediction. By visual evaluation and
considering the accuracy of each data point, one would accept the model as valid (or,
more accurately, not invalid) for most engineering applications.
0.9
0.8
0.7
0.6
initial concentration
0.5
Change from
0.4
0.3
0.2
0.1
0 10 20 30 40 50 60 70 80
Time from input step (min)
Figure 2.28 Comparison of empirical data (square) and model (line) for stirred tank.
ILLUSTRATIVE EXAMPLES
Example 2.1 Derive a set of state variables that represents a dynamic system of the
spring-mass-damper system.
A set of state variables that represents a dynamic system is illustrated in terms
of the spring-mass-damper system, as shown in Figure 2.29. The number of state
variables chosen to represent this system should be as small as possible in order to avoid
redundant state variables. A set of state variables sufficient to describe this system
includes the position and the velocity of the mass. Therefore, a set of state variables can
be defined as (x1, x2), where
F[
V
x1(t) = y(t) and Z
V
FV
156 Process Control Modelling
Spring, k Damper, c
Wall friction, b
Mass, M
y(t) u(t)
Figure 2.29 Spring-mass-damper system.
The differential equation describes the behaviour of the system and is usually
written as
F [ F[ (2.124)
/
D
M[ W
V
FV FV
To write Eq. (2.124) in terms of the state variables, we substitute the state
variables as already defined, and obtain
FZ
/ DZ MZ W
V (2.125)
FV
Therefore, we can write the differential equations that describe the behaviour of the
spring-mass-damper system as a set of two first-order differential equations, given as
FZ
Z (2.126)
FV
FZ D M
Z Z W (2.127)
FV / / /
This set of differential equations describes the behaviour of the state of the system
in terms of the rate of change of each state variable.
Example 2.2 Derive a set of state variables that represents a dynamic system of the
RLC electrical circuit.
Let us consider the RLC electrical circuit, as shown in Figure 2.30. The state of
this system can be described in terms of a set of state variables (x1, x2), where x1 is the
capacitor voltage vc(t) and x2 is equal to the inductor current iL(t). This choice of state
variables is intuitively satisfactory because the stored energy of the network can be
described in terms of these variables as
Y .K. %XE (2.128)
Illustrative Examples 157
iL
+
u(t) +
Current C
vc R v0
source
ic
Therefore, x1(t0) and x2(t0) represent the total initial energy of the network and thus the
state of the system is at t = t0. For a passive RLC network, the number of state variables
required is equal to the number of independent energy-storage elements. Utilizing
Kirchhoffs current law at the junction, we obtain a first-order differential equation by
describing the rate of change of capacitor voltage as
FXE
KE % W
V K. (2.129)
FV
Kirchhoffs voltage law for the right-hand loop provides the equation describing the
rate of change of inductor current as
FK.
. 4K. XE (2.130)
FV
The output of this system is represented by the linear algebraic equation
v0 = RiL(t)
We can rewrite Eqs. (2.129) and (2.130) as a set of two first-order differential
equations in terms of the state variables x1 and x2 as follows:
FZ
Z W
V (2.131)
FV % %
and
FZ 4
Z Z (2.132)
FV . .
The output signal is then given as
[
V X
V 4Z (2.133)
Utilizing Eqs. (2.131) and (2.132), and the initial conditions of the network represented
by [x1(t0), x2(t0)], the systems future behaviour and its output can be determined.
The state variables that describe a system are not a unique set. Therefore, several
alternative sets of state variables can be chosen, for example, a second-order system,
such as the mass-spring-damper of Figure 2.29 or RLC electrical circuit of Figure 2.30.
The state variables may be any two independent linear combinations of x1(t) and x2(t).
Therefore, for RLC circuit we might choose the set of state variables as the two
158 Process Control Modelling
voltages, vc(t) and vL(t), where vL is the voltage drop across the inductor. Then the new
state variables, x*1 and x*2, are related to the old state variables x1 and x2, as
x*1 = vc = x1 (2.134)
and
x*2 = vL = vc RiL = x1 Rx2 (2.135)
Equation (2.135) represents the relation between the inductor voltage and the former
state variables vc and iL. In an actual system there are several choices of a set of state
variables that specify the energy stored in a system and, therefore, adequately describe
the dynamics of the system. A widely used choice is a set of state variables that can be
readily measured.
Example 2.3 Derive a set of state variables that represents static and dynamic
behaviours of a stirred-tank heater.
Let us consider an example of state variable for a chemical process, such as the
stirred-tank heater, as shown in Figure 2.31. The fundamental quantities whose values
provide the information about heater are:
· Total mass of the liquid in the tank
· Total energy of the material in the tank
· Momentum of the heater.
Fi, Ti
h
Q
Heater
Condensate Fst
Steam
Figure 2.31 Stirred-tank heater.
The momentum of the heater remains constant even when the disturbances
change value and therefore it is not important to be considered further. The state
variables for the tank heater are total mass in the tank and total energy of the liquid
in the tank. They are defined as follows:
Total mass in the tank = rV = rAh (2.136)
where
r = Density of liquid
V = Volume of liquid
Illustrative Examples 159
* S8E R
6 6TGH S #JE R
6 6TGH (2.138)
where
cp = Heat capacity of the liquid tank
Tref = Reference temperature, where the specific enthalpy of the liquid is assumed
to be zero.
From Eqs. (2.136) and (2.138) it is concluded that the state variables for the stirred-tank
heater are h and T, while the parameters r, A, cp and Tref are characteristics of the tank
system.
Now, to develop the state equations for the stirred-tank heater, conservation
principle is applied on the two fundamental quantities: the total mass and the total
energy:
#EEWOWNCVKQP QH VQVCN OCUU
Total mass balance =
6KOG
+PRWVQHVQVCNOCUU 1WVRWVQHVQVCNOCUU
6KOG 6KOG
or
F
S #J
S (K S ( (2.139)
FV
where Fi and F are the volumetric flow rates (i.e. volume per unit of time) for the inlet
and outlet streams, respectively. Assuming constant density (independent of
temperature), Eq. (2.139) becomes
FJ
# (K ( (2.140)
FV
160 Process Control Modelling
We also have
#EEWOWNCVKQPQHVQVCNGPGTI[
Total mass balance =
6KOG
+PRWVQHVQVCNGPGTI[ 1WVRWVQHVQVCNGPGTI[
6KOG 6KOG
'PGTI[UWRRNKGFD[UVGCO
6KOG
or
F= S #JE R
6 6TGH ?
S (K E R
6K 6TGH S (E R
6 6TGH 3 (2.141)
FV
where Q is the amount of heat supplied by the steam per unit of time. The equation
above can take the following simpler form assuming that Tref = 0:
F
J6 3
# (K 6K (6 (2.142)
FV SER
Additional algebraic manipulations on Eq. (2.142) yield
F
J6 F6 FJ F6 3
# # #6 #J 6
(K ( (K6K (6
FV FV FV FV SER
or
F6 3
#J (K
6K 6 (2.143)
FV SER
Summarizing the modelling steps above, the state equations and state variables are
as follows:
State equations:
FJ (2.144)
# (K (
FV
and
F6 3
#J (K
6K 6 (2.145)
FV SER
State variables: h, T
Output variables: h, T (both measured)
Input variables: Ti, Fi (as disturbances), and manipulated variables are Q, F (for
feedback control) and Fi (for feed-forward control).
Parameters: A, r, cp
The state equations (2.144) and (2.145), with the state variables, the inputs and the
parameters, constitute the mathematical model of the stirred-tank heater.
Illustrative Examples 161
Example 2.4 Linearize the following nonlinear ordinary differential equation (ODE) for
a gravity-flow tank, which is derived from a momentum balance around
the exit pipe:
FX I - I
J ( E X (2.146)
FV . S #R
where
v = Velocity of liquid in the pipe
h = Liquid height in the tank
L = Length of pipe
KF = Friction factor constant
r = Density
Ap = Cross-sectional area of pipe
g = Gravitational force
gc = Gravitational constant.
Linearizing the v2 term gives
X X X
X X (2.147)
Equation (2.146) becomes
FX È IØ È X-
( IE Ø È X - ( IE Ø
É Ù J X É Ù (2.148)
FV Ê .Ú ÉÊ S # R ÙÚ Ê S #R Ú
This ordinary differential equation is now linear. The terms in the parentheses are
constants; they depend, of course, on the steady state around which the system is
linearized.
Example 2.5 Linearize the component balance equation for an irreversible nth-order,
nonlinear isothermal reaction occurring in a constant-volume, variable-
throughput continuous stirred-tank reactor (CSTR), which is given as
F% #
8 ( % # (% # 8
%#P B G ' 46
(2.149)
FV
Linearizing gives
F% #
8 = ( % # (
% # % # % #
( ( ?
FV
(
= (%
% % %
( ( ?
# # # #
Ë P P'
8 Ì M% P
% %
PM%
M% #
6 6 Û
# # # # Ü (2.150)
Í 46 Ý
Example 2.6 Derive a mathematical model of a mixing process in which two streams
are being mixed in a well-stirred tank, producing a product stream 3, as
shown in Figure 2.32.
162 Process Control Modelling
Stream 1 Stream 2
Q (Heat added
or removed)
Stream 3
CA3, T3, F3
Figure 2.32 Mixing process in a well-stirred tank.
is equal to the density of the material in the tank, r (i.e. r3 = r). In general, the densities
r1, r2 and r3 depend on the corresponding concentrations and temperatures. Therefore,
S S Ô
% # %$ 6 S Ô
% # %$ 6 S Ô
% # %$ 6
Usually (but not always) the dependencies above are weak and we assume that the
densities are independent of the concentrations and temperatures. Therefore, we
assume that
S S S S
This transforms Eq. (2.151) to the following:
F8 FJ
#
( ( ( (2.152)
FV FV
Now the second fundamental quantity needed to describe the mixing process is
#EEWOWNCVKQPQHEQORQPGPV#
Balance on component A in the tank =
6KOG
+PRWVQHEQORQPGPV#
6KOG
1WVRWVQHEQORQPGPV#
6KOG
or
F
% # 8
% # ( % # ( % # (
FV
or
F% # F8
8 % #
% # ( % # ( % # ( (2.153)
FV FV
Substituting dV/dt from Eq. (2.152), we obtain
F% #
8 % # =
( ( ( ?
% # ( % # ( % # (
FV
and since CA = CA3 due to the well-stirred assumptions,
F% #
8
% # % # (
% # % # (
% # % # ( (2.154)
FV
We also have
#EEWOWNCVKQPQHVQVCNGPGTI[
Total energy balance =
6KOG
È +PRWVQHVQVCNGPGTI[ Ø È 1WVRWVQHVQVCNGPGTI[ Ø
ÉÊ YKVJHGGFUVTGCO ÙÚ ÉÊ YKVJRTQFWEVUVTGCO ÙÚ
6KOG 6KOG
*GCVCFFGFQTTGOQXGFYKVJVJGEQKN
6KOG
164 Process Control Modelling
F
S8J
S
( J ( J S ( J 3 (2.155)
FV
Now h1, h2 and h3 are to be characterized in terms of other variables (i.e.
temperatures, concentrations, etc.). We know that
J
6 J
6 % R
6 6
14243
J
6 J
6 % R
6 6 (2.156)
J
6 J
6 % R
6 6
where T0 is the reference temperature. At this temperature, we have
SJ
6 %
% # *
$ * $ % # '* 5
6
14243
#
SJ
6 % *
% # * #
$ $ % # '* 5
6 (2.157)
SJ
6 % # * # %$ * $ % # '*
5
6
where *
# and *$ are the molar enthalpies
(enthalpy per mole) of components A and B
5, '*5 and '*5 are the heats of solution for streams 1, 2
at temperatures T0. '*
and 3 per mole of A at temperature T0. Substituting Eqs. (2.156) and (2.157) into the
total energy balance equation (2.155), we take
$ * $ % # '*5 S8% R
6 6
%
F=8
% # *
#
FV
$ $ % # '* 5 S (% R
6 6
% *
(
% # *
#
(
% # * # %$ * $ % # '* 5 S ( % R
6 6
(
% # * # % $ * $ % # '* 5 S ( % R
6 6 3
Illustrative Examples 165
5 S ( % R
6 6 ( % # '*5 S ( % R
6 6
( %# '*
5 S ( % R
6 6 3
( % # '*
F6 F8
S % R 8 S% R
6 6 '*
5 =% # ( % # ( % # ( ?
FV FV
5 S ( % R
6 6
( % # '*
5 S ( % R
6 6
( % # '*
5 S ( % R
6 6 3
( % # '*
and finally,
F6
S% R 8 % # ( = '*
5 '* 5 ? % # ( = '* 5 '* 5 ?
FV
S ( =% R
6 6 % R
6 6 ?
S ( =% R
6 6 % R
6 6 ? 3
If we assume that % R % R % R % R, we have
F6
S % R 8
% # ( =
'*
5 '*5 % # (
'*5 '*5 ?
FV (2.158)
S ( % R =
6 6 S ( % R
6 6 ? 3
Summarizing the steps above, we have the following:
State variables: V, CA3, T3
State equations:
F8 FJ
#
( ( (
FV FV
F% # F8
8 % #
% # ( % # ( % # (
FV FV
F6
S % R 8 % # (
'* 5 '*
5 % # (
'* 5 '* 5
FV
S ( % R =
6 6 S ( % R
6 6 ? 3
Input variables: F1, CA1, T1, F2, CA2, T2, F3 (for feedback control)
Output variables: V (or equivalently the height of liquid level, h), CA3, T3
Parameters (constant): S % R '*
5 '*5 '*5
Observations:
(a) The momentum does not change under any operating conditions and it will be
neglected in further treatment.
(b) We only need to consider two of the following three quantities: total mass,
mass of A, and mass of B. The third can be computed from the other two.
166 Process Control Modelling
(c) Usually, a mixing tank is equipped with a cooling or heating coil or jacket
through which flows a coolant (if heat is released during the mixing of the two
solutions) or a heating medium (if heat is absorbed during mixing) in an
attempt to keep the mixing isothermal.
(d) If the heats of solution are strong functions of concentration, i.e. if
5 '*5 and
'*5 '*5 are not small quantities, then from the
'*
or
(MWAVCA)t+Dt (MWAVCA)t = (MWAFCA0 MWAFCA MWAVkCA)D t (2.160)
Illustrative Examples 167
F0
CA0
F1
V CA
È Ø
144444444424444444443
Ê U Ú
Then we get
F
% # GV U (
% # GV U V
FV 8
or
(% #
Ô F
%# G ÔG
VU V U
FV (2.162)
8
or
(% # U V U
% # GV U
G +
8
or
(U
%# % # +GV U
8
The data give the initial condition of the inlet concentration of 0.925 mol/m3 at the
time of the step, t = 0. The initial steady-state reactor concentration can be determined
from the data and Eq. (2.161) with dCA /dt = 0.
168 Process Control Modelling
(
% # KPKV % #KPKV OQNO
( 8M
(
% # KPKV % # (
'% #
+
( 8M ( 8M
This can be substituted in Eq. (2.162) to give
(% # (
'% # V U
%# G
( 8M ( 8M
( (2.163)
% #
% # % #
G V U
( 8M
KPKV KPKV
This can be rearranged with Kp = F/(F + Vk) to give the change-in reactor
concentration. Then
CA (CA)init = KpDCA0(1 et/t)
DCA = (0.503)(0.925)(1 et/t)
Again, the time constant determines the speed of the response. Note that in this
example, the time constant depends on the equipment (V), the operation (F), and the
chemical reaction (k). Also, by comparing Eqs. (2.121) and (2.161) and using the same
values for F and V, the time constant for the chemical reactor is always shorter than
the time constant for the mixer. Their numerical values are as follows:
8
U OKP
( 8M
and
8
-R OQNO
( 8M
Thus, the steady-state gain and time constant in this example depend on
equipment design and operating conditions.
Results analysis: First, the result from Eq. (2.163) is calculated and plotted. As
shown in Figure 2.34(a), the reactant concentration increases as an exponential function
to its final value without overshoot or oscillation. In this case, the concentration exceeds
its maximum limit. Therefore, a corrective action will be evaluated. The concentration
reaches the alarm limit in 19.6 minutes after the step (29.6 minutes in the figure) and
exceeds the maximum limit after 22.5 minutes. The sensitivity of this result can be
evaluated from the analytical solution. In particular, the dependence of the time
constant on variables and parameters is given in Eq. (2.161). The time difference
between the alarm and the dangerous condition is too short for a person to respond
reliably, because other important events may be occurring simultaneously.
Illustrative Examples 169
1 1
Limit Limit
Outlet concentration
Outlet concentration
0.8 0.8
Alarm Alarm
0.6 0.6
0.4 0.4
0 0
10 20 30 40 50 60 70 80 10 20 30 40 50 60 70 80
Time (min) Time (min)
2 2
Inlet concentration
Inlet concentration
1 1
0 0
0 0
10 20 30 40 50 60 70 80 10 20 30 40 50 60 70 80
Time (min) Time (min)
(a) Without action at the alarm value (b) With action at the alarm value
Figure 2.34 Results for Example 3.7 (for isothermal CSTR).
Example 2.8 Derive a mathematical model of an on/off room heating system using
general modelling steps, as mentioned in Table 2.1.
Goals: Determine the dynamic response of the room temperature. Also, ensure
that the furnace does not have to switch on or off frequently than once per 3 minutes,
to allow the combustion zero to be purged of gases before re-ignition.
Information: The system is taken to be the air inside the dwelling. A sketch of
the system is given in Figure 2.35. The important variables are the room temperature
and the furnace on/off status.
Assumptions:
(a) The air in the room is well mixed.
(b) No transfer of material to or from the dwelling occurs.
170 Process Control Modelling
(Sensor)
Thermostat Controller
(Final element)
Figure 2.35 Feedback control for controlling room temperature.
(c) The heat transferred depends only on the temperature difference between the
room and the outside environment.
(d) No heat is transferred from the floor or ceiling.
(e) Effects of kinetic and potential energies are negligible.
Data:
(a) The heat capacity of the air Cv = 0.17 cal/(g°C)
(b) The overall heat transfer coefficient UA = 45 ´ 103 cal/(°Ch)
(c) The size of the dwelling = 5 m ´ 5 m ´ 3 m
(d) The furnace heating capacity Qh = 0 (off) or 1.5 ´ 106 (on) cal/h
(e) The furnace heating switches instantaneously at the values of 17°C (on) and
23°C (off)
(f) The initial room temperature is 20°C and the initial furnace status is off .
(g) The outside temperature (Ta) is 10°C.
Formulation: The system is defined as the air inside the house. To determine the
temperature, an energy balance should be formulated. Since no material is transferred,
no material balance is required. The application of the energy balance in Eq. (2.112) to
this system gives
F7
3 9U (2.164)
FV
The shaft work is zero. From principles of thermodynamics and heat transfer, the
following expressions can be used for a system with negligible accumulation of potential
and kinetic energy:
F7 F6
S8%X Q = UA(T Ta) + Qh (2.165)
FV FV
Illustrative Examples 171
with
0 when T > 23°C
14243
Qh = 1.5 ´ 106 when T < 17°C
unchanged when 17 < T > 23°C
which gives
F6
S8%X 7#
6 6C 3J (2.166)
FV
The degree of freedom for this formulation is zero since the model has two
equations, two variables (T and Qh), four parameters (UA, Cv, V and r), and one external
variables (Ta). Thus, the system is exactly specified with Eq. (2.166), when the status of
the heating has been defined by Eq. (2.165).
Solution Rearranging Eq. (2.166) gives the following linear ordinary differential
equation:
F6 7#6C 3J 8 S%X
6 U (2.167)
FV U 8 S%X 7#
Equation (2.167) is a linear differential equation when the value of heat transferred (Qh),
is constant. As described in the example data, Qh has one of the two constant values,
depending on the status of the furnace heating. Thus, the equation can be solved using
the integrating factor with one value of Qh until the switching value of temperature is
reached. Then, the equation is solved with the appropriate value of Qh until the next
switch occurs. The solution of Eq. (2.167) is given as
6 6KPKV
6HKPCN
6
KPKV
G V U
(2.168)
where
t = Time from step in Qh
t = Time constant = 0.34 h 3J
Tfinal = Final value of T as t ®ÿ ¥ = Ta + = 10oC when Qh = 0
o 6 7#
= 43.3 C when Qh = 1.5 ´ 10
Tinit = Value of T when a step in Qh occurs.
Results analysis: First, the numerical result is determined and plotted, as shown
in Figure 2.36. From the initial condition with the furnace off, the temperature
decreases according to Eq. (2.168) until the switch value of 17°C is reached. Then the
furnace heating begins instantaneously (Qh changes from 0 to 1.5 ´ 106), and since the
system is first-order with no momentum, the temperature immediately begins to
increase. This procedure is repeated as the room temperature follows a periodic
trajectory between 17°C and 23°C.
The analytical solution provides insight into how to alter the behaviour of the
system. The time constant is proportional to the mass in the room, which seems
reasonable. Also, it is inversely proportional to the heat transfer coefficient, since the
faster the heat transfer, the more quickly the system reaches equilibrium with its
surroundings. Therefore, insulating the house will decrease UA and increase the time
constant. Finally, the time constant does not depend on the heating by the furnace,
which is the forcing function of the system. Therefore, increasing the capacity of the
172 Process Control Modelling
temperature, T
Room
20
1
Heater
0
0
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time (hr)
Figure 2.36 Dynamic response of room temperature control.
furnace will not affect the time constant, although it will affect the time between
switches.
The goals of the modelling exercise have been satisfied. The temperature has been
determined as a function of time, and the switching frequency of the furnace has been
determined to be over 3 minutes, that is, longer than the minimum limit. However, a
switch could occur much faster due to a sudden change in outside temperature or to a
disturbance such as a door being opened, which would allow a rapid exchange of warm
and cold air. Therefore, a special safety system would be included to ensure that the
furnace would not be restarted until a safe time period after shutting off.
Example 2.9 Transform the following equation, and obtain the ratio of output to input,
Y(s)/X(s):
F [ F[
[ Z
FV FV
We get
ËÍ U ;
U U[ [ ÛÝ > U;
U [ @ ;
U :
U
If the initial values are y0 = 0, [ = 0, then
;
U ;
:
U : U U
Example 2.10 Using Laplace transforms; solve the following equation:
F Z
Z x0 = 0, x¢0 = 0
FV
Transforming each term and solving for X gives
U : :
U
:
U
U
Then, we get
Ë Û
Z . Ì Ü
Í U
U Ý
ËÍU Z
U U ÛÝ ÍËU Z
U UÝÛ > UZ
U @ Z
U
U U
174 Process Control Modelling
U U U
Z
U
U
U
U U U
The cubic in the denominator may be factored, and x(s) expanded in partial fractions:
U U U # $ % & '
Z
U (1.169)
U
U
U
U
U U U U U U
To find A, multiply both sides of Eq. (1.169) by s and then set s = 0; the result is
#
The other constants are determined in the same way. The procedure and results are
summarized in the following table.
REVIEW QUESTIONS
2.1 What is a mathematical model of a physical process? What do we mean by when
we talk about mathematical modelling?
2.2 Why do we need to develop the mathematical model for a process we want to
control?
2.3 What are the general steps of modelling procedures? Explain each of them with
examples.
2.4 Develop a relationship for overall material balance, component material balance,
and energy balance for model formulation.
2.5 Formulate the model for a liquid-in-tank process shown in Figure 2.37. The tank
has been designed well, with baffling and impeller size, shape and speed, such that
the concentration should be uniform in the liquid.
Review Questions 175
F0
CA 0
F1
V CA
Assumptions:
(a) Well-mixed vessel
(b) Density is the same for A and solvent
(c) Constant flow in
(d) The system is initially at steady state
Data:
F0 = 0.85 m3/min; CAinit = 0.925 mol/m3; V = 2.1 m3; DCA0 = 0.925 mol/m3
2.6 The chemical reactor in illustrative example 2.7 is to be modelled with the goal
of determining the concentration of the product CB as a function of time for the
same input change. Extend the analytical solution to answer this question.
2.7 What are the state variables, and what are the state equations? What are they
used for?
2.8 (a) How many state variables do we need to describe a system that is composed
of M phases and N components?
(b) Write a relationship that will give the molar or the specific enthalpy of a multi-
component liquid at temperature T and pressure p, with known composition for
the N components.
2.9 (a) What are the principal control considerations that affect the scope of
mathematical modelling of a chemical process?
(b) In what sense do the control considerations affect the mathematical modelling
of a chemical process?
2.10 A batch reactor with the parameters in illustrative Example 2.7 is initially empty
and is filled at the inlet flow rate, with the outlet flow being zero. Determine the
concentration of A in the tank during the filling process. After the tank is full, the
outlet flow is set equal to the inlet flow; that is, the reactor is operated like a
continuous-flow CSTR. Determine the concentration of A in the steady state.
2.11 The system in question 2.6 has an input concentration that varies as a sine with
amplitude A and frequency w. Determine the outlet concentration for this input.
2.12 The level flow system in Figure 2.38 is to be analyzed. The flow F0 is constant.
The flow F3 depends on the valve opening but not on the levels, whereas flow F1
and F2 depend on the varying pressures (i.e. levels). The system is initially at
176 Process Control Modelling
F0
F3
L1 F1
L2 F2
steady state, and adjusting the valve makes a step-increase in F3. Determine the
dynamic response of the levels and flows using an approximate linear model.
Without specific numerical values, sketch the approximate dynamic behaviour of
the variables.
2.13 Outline the steps that you should take during the development of a mathematical
model for process control purposes.
2.14 The room heating in illustrative Example 2.8 is reconsidered, for the following
situations, each representing a single change from the base case. Reformulate the
model as needed and determine the dynamic behaviour of the temperature and
heating status.
(a) Due to leaks, a constant flow into and out of the room exists. Assume that the
volume of air in the room is changed every hour with entering air at the
outside temperature.
(b) A mass of material (e.g. furniture) is present in the room. Assume that this
mass is always in equilibrium with the air, that is, the heat exchange is at
quasi-steady state. The mass is equivalent to 200 kg of wood.
(c) The ambient temperature decreases to 10°C.
(d) The duty of the furnace is reduced to 0.60 ´ 106 when on.
(e) The heat transferred to the room does not change instantaneously when the
furnace status changes. The relationship between the heat generated in the
furnace (Qf), which changes immediately when the switch is activated, and the
heat to the room (Qh) is
F3J
U3 3H 3J
FV
with tQ = 0.10 h.
Review Questions 177
F Z FZ
(d) # UKP
X V where x¢0 = 2, x0 = 1
FV FV
Cooling water
1
123456789
123456789
123456789
123456789 Steam
123456789
123456789
123456789
Bottoms product
FB, xB
2.19 Derive a mathematical model of a tabular heat exchanger shown in Figure 2.40.
Steam
T1 T2
Liquid T |z T |z+D z
Dz Steam
Figure 2.40 Tubular heat exchanger.
2.20 Derive a set of state variables that represents a dynamic system of the pendulum
oscillator model.
2.21 Give examples to demonstrate how can a model of a process be simplified by
disregarding physical and chemical phenomena with a limited impact on the
behaviour of the process.
2.22 Consider the two stirred-tank heaters shown in Figure 2.41.
(a) Identify the state variables of the system.
(b) Determine what balances you should perform.
(c) Develop the state model that describes the dynamic behaviour of the system.
(d) How would you express the heats given by the two steam flows in terms of
other variables?
The flow rates of the effluent streams are assumed to be proportional to the liquid
static pressure that causes the flow of the liquid. The cross-sectional areas of the
two tanks are A1 and A2 (ft2) and the flow rates are volumetric. No vapour is
produced either in the first or the second tank. At1 and At2 are the heat exchange
areas for the two steam coils.
F1, Ti
A1
F2, T2
Steam
100 psi
A2
Q1 (lb/min)
F3, T3
Steam
300 psi
Q2 (lb/min)
Figure 2.41 Review question 2.22.
Self-Check Quiz 179
SELF-CHECK QUIZ
A State True/False:
2.1 Laplace transforms are used to convert the __________ that represent the dynamic
behaviour of process output variables into __________________ equations.
2.2 Transfer function is restricted to mathematical models that are described by
________________ equations with constant ___________coefficients.
2.3 The state variables of a system characterize the _______ behaviour of a system.
2.4 The z-transformation is used for certain kinds of discontinuous system (also called
discrete-time systems), whose dynamic behaviour can be defined by ___________
equations.
3
3.1 INTRODUCTION
The successful and efficient operation of a system requires some form of control. Control
systems influence every facet of modern life such as automatic washers, microwave
ovens, process plants, navigation and guidance systems, space satellites, pollution
control, mass transit, economic regulation, etc. The degree, quality and means of control
may vary widely. As explained in Chapter 1, control may be exercised over one or more
variables called controlled outputs. Feedback controls are employed when it is required
to hold a quantity at some specified value (called reference or set value) in spite of the
effect of disturbances. In a feedback (or closed-loop) control system, the controlled output
is measured and fed back for comparison with the reference value and any error is used
to make the controlled output correspond to the reference value.
The development of feedback control systems started many years ago. But, in the
third century BC, Ktesibios of Alexandria was credited to develop a feedback control
system for the development of a water clock, where the inflow of water in a tank is
maintained constant by a float operated valve. Later on, the use of windmills in Europe
for motive power led The Netherlands Andrew Meikle in 1750 to develop a feedback
control system for turning a windmill into the wind. Then came the advent of the steam
age, and in 1788 James Watt invented his flyball governor in Scotland for regulating the
speed of steam engines. The early engineering developments of feedback system
involved trial and error method. The early control systems were successfully developed
by the procedures of invention, construction, testing and modification. Therefore, the
developments were more an art than a science. The specific approach perhaps began
with Maxwell in 1868 with his analysis of governors.
The theory of control systems has now reached a high level of sophistication and
their design has become quite analytical. In this chapter, the topics such as reasons for
using feedback control systems, its basic principles and classifications, feedback control
requirements and its selection and tunings, are described.
181
182 Feedback Control System
Controller mechanism
e +
Controller FSP
Fm
DP = Differential pressure
DP cell
Controller mechanism
e +
Controller FSP
Fm
DP = Differential pressure
DP cell
F
F
pSP
pm +
e
DP cell
Controller
mechanism
p Controller
Fi Tank
Fo
(c) Feedback system for gas pressure control. Here DP = differential pressure.
h 12345678901234
12345678901234
DP cell 12345678901234
12345678901234
12345678901234
12345678901234
12345678901234
hm 12345678901234
12345678901234
12345678901234
12345678901234
12345678901234
+
hSP
e Controller
mechanism
Controller
Controller mechanism
e
F1, c1 Controller
+
cSP
cm
Chromatograph
F2, c2 c
Controller mechanism
e
TSP
+
Tm Controller
Thermocouple
Flow control. Two feedback systems are shown in Figures 3.1(a) and (b), which control
the flow rate F at the desired (set-point) value FSP.
Pressure control. Figure 3.1(c) shows a feedback system for control of gas pressure
in the tank at the desired (set-point) pressure value pSP.
Liquid-level control. Figure 3.1(d) shows a feedback system for the control of
liquid levels at the bottom of a distillation column and its condenser accumulation
tank.
Basic Principles 185
Composition control. Figure 3.1(e) shows a feedback control system for composition
control. Composition is the controlled variable in the blending system and the desired
value of composition is cSP.
Temperature control. The system in Figure 3.1(f) controls the temperature of the
existing hot stream at the desired value TSP.
To simplify the presentation of a feedback control system, the diagrammatic details
of a controller mechanism is usually replaced with a simple circle carrying one of the
following characterizations:
FC for flow control
PC for pressure control
LC for level control
CC for composition control
TC for temperature control
FT for flow measurement and transmission
PT for pressure measurement and transmission
LT for level measurement and transmission
CT for composition measurement and transmission
TT for temperature measurement and transmission
Figures 3.2(a) and (b) show simplified representation of Figures 3.1(b) and (d)
respectively.
FC LT 12345678901
12345678901
12345678901
12345678901
12345678901
12345678901
12345678901
12345678901
LC
FT
All the examples above indicate that the basic hardware components of a feedback
control loop are the following:
(a) Process: the material equipment along with the physical or chemical
operations which take place (tanks, heat exchangers, reactors, separators, etc.)
(b) Measuring devices or sensors: which measures the plant parameters for
example, thermocouples (for temperature), bellows and diaphragms (for
pressure or liquid-level), orifice plates (for flow), gas chromatographs and
various types of spectroscopic analyzers (for composition), and so on.
186 Feedback Control System
(c) Transmission lines: used to carry the measurement signal from the sensor to
the controller and the control signal from the controller to the final control
element. These lines can either be pneumatic (compressed air or liquid) or
hydraulic or electrical and electronics.
(d) Controller: also includes the function of the comparators. This unit is built
with logic that decides by how much the value of the manipulated variable is
to be changed? It requires the specification of the desired value (set-point).
(e) Final control element: usually, a control valve, solenoid valve, or a variable-
speed metering pump. This device receives the control signal from the
controller and implements it by physically adjusting the value of the
manipulated variable.
Each of the elements discussed should be viewed as a physical system with an input
and an output. Consequently, their behaviour can be described by a differential equation
or equivalently by a transfer function.
Flow Orifice plates · Venturi flow nozzle · Dahl flow tube · Kennison flow nozzle
· Turbine flow meters · Ultrasound flow meters · Hot-wire anemometry
Liquid level Float-actuated devices · Displacer devices · Liquid head pressure devices
· Conductivity measurement · Sonic resonance · Radiation devices · Air
purge devices · Optical level detectors
Process
Process Thermocouple temperature, T
temperature, T temperature, Tm
Tm
Tm
Thermowell
Thermocouple
T
Response with
single-film
resistance
A Response with
B two-film resistances
Thermowell wall t
(c) Typical thermocouple arrangement (d) Response of thermocouples
Figure 3.3 Temperature profiles of thermocouple.
In the second case, as shown in Figure 3.3(b), we have major heat transfer film
resistances inside and outside the thermowell casing. This is equivalent to two capacities
188 Feedback Control System
F 6O F6O
U [U 6O 6 (3.2)
FV FV
The parameters t and z depend on the constructional and material characteristics
of the temperature-sensing device (i.e. thermocouple, casing, materials of construction,
etc.). It is clear that the response of a thermocouple modelled by Eq. (3.2) is slower than
that of a thermocouple modelled by Eq. (3.1), as depicted in Figure 3.3(d).
Stationary plate
Pressure input
Sensing
Sensingdiaphragm
diaphragmplate
plate
Figure 3.4 Variable capacitance differential pressure transducer.
A force balance around the sensing diaphragm leads to the following second-order
model:
F \ F\
U
[U \ - R 'R (3.3)
FV FV
where
z = Displacement of the sensing diaphragm
Dp = Actuating pressure difference
t,z ,Kp = Three parameters of a second-order system, defined in this case by the
constructional characteristics.
Various other types of pressure sensors, all of them measuring the displacement
of a mechanical part under the influence of Dp, are also in use.
Measuring Devices (Sensors) 189
Flow sensors have very fast dynamics and they are usually modelled by simple
algebraic equations, given as
(NQY B 'R (3.4)
2Q
U G U FU
(3.5)
2K
U U R U
with U F U R .
Error
e Final control
6 A mplifier Controller Process
Set point element
(SP)
Feedback Sensor/
amplifier Transducer
A ir pressure
Diaphragm
A ctuator
Fluid flow
For most actuators, the motion of the stem is proportional to the pressure applied
on the diaphragm. In general, this type of actuator can be used for functions other than
moving a valve stem. For example, it can be used to adjust dampers, variable speed
drives, rheostats and other devices. As pressure to the valve varies over its normal
range of operation (3 to 15 psig), the range of motion of the stem varies from a fraction
of an inch to several inches depending on the size of the actuator. Manufacturers provide
a range of actuators for various valve sizes.
Figure 3.8 shows control valve symbols. When the air pressure (power) is removed
from the valve, it assumes its failure position. Three failure positions are shown in
Figure 3.8, namely, fail open, fail closed, and fail locked (or unchanged). These positions
are selected for safety. The typical control valve has a relatively large unrecoverable
Fail open
FO
Fail closed
FC
Fail locked
FL
Butterfly
Hand valve
A ngle valve
pressure drop. Thus, a butterfly valve or damper is sometimes used for control. Many
control valves in a process design are not automated and must be opened or closed
manually, for example, a hand valve as shown in Figure 3.8. An angle valve is used as
safety valve which opens without an external power (air pressure) source when the
process pressure exceeds a specified limit.
Control valves are designed to fail either in the completely open or the completely
shut position. Appropriateness of valve action depends on the effect of the manipulated
variable on the safety of the process. For example, in case of steam or fuel handling,
Final Control Elements 193
the flow is required to be cut off in an emergency (valve to fail closed). If the valve is
handling cooling water to a reactor, the flow is required to go to maximum in an
emergency (valve to fail open).
§( ·
(OCZ - X '2V '2J FGU ¨ OCZ ¸ (3.7)
© (FGU ¹
194 Feedback Control System
§( ·
(OKP HOKP - X '2V '2J FGU ¨ OKP ¸ (3.8)
© (FGU ¹
where
Fdes = Flow rate at design
DPt = Total pressure drop through the system at design flow rates
(DPh)des = Pressure drop through the fixed resistances in the system at design flow
fmin = Minimum valve opening
Solving Eqs. (3.7) and (3.8) for DPt gives
ª (OCZ (OKP º
« »
'2V «¬ (FGU »¼
J FGU
'2 §H ( ·
(3.9)
¨ OKP OCZ ¸
© (OKP ¹
It is clear from Eq. (3.9) that as the second term in the denominator approaches
unity, the required pressure drop goes to infinity. So there is a limit to the achievable
rangeability of a system, also called rangeability index of the system. Rangeability
index of the system may be defined as
HOKP (OCZ
{ (3.10)
(OKP
The right hand side parameter of Eq. (3.10) must be chosen such that is less
than unity. The value of fmin can be reduced below 0.1 if a large turndown ratio is
required. This is accomplished by using two control valves in parallel, one large and one
small, in a split-range configuration. The small valve opens first, and then the large
valve opens as the signal to the two valves changes over its full range.
Pressure
to valve
Pv R( s) C ( s)
G v( s)
Pressure Flow through
F Flow through to valve valve
valve
(b) Block diagram
(a) Physical system
Figure 3.9 Block diagram of control valve.
The block diagram for the valve is shown in Figure 3.9. The input signal is the
pressure to the valve top and the output is the flow rate of fluid through the valve.
of a change of pressure to the valve top. The relation between the flow through the
valve and the valve stem position (or lift) is called valve flow characteristics. The valve
flow characteristics f(x) depend on the geometrical shape of the plugs surface. By
changing the shape of the plug and the seat in the valve, different relationships between
stem position and flow area can be attained. Three common valve characteristics are
linear, equal-percentage, and quick-opening, which are presented as follows:
For a linear valve,
H
Z Z (3.11)
For an equal-percentage valve,
H
Z B Z (3.12)
For a quick-opening valve,
H
Z Z (3.13)
1.0
Valve characteristics f( x) (Fraction of flow)
0.8 Quick-opening
0.6
L inear
0.4
Equal-percentage
( B = 50)
0.2
In terms of valve characteristics, valves can be divided into three types namely
linear, equal-percentage, and quick-opening as shown in Fig. 3.10. If constant pressure
drop over the valve is assumed and if the stem position is 50 percent open, a linear valve
gives 50 percent of the maximum flow and an equal-percentage valve gives only 15
percent of the maximum flow. In quick-opening (also called decreasing sensitivity type),
the valve sensitivity at any flow decreases with increasing flow. The maximum port area
is sufficiently large that pressure losses elsewhere than at the valve port may restrict
the maximum flow. In linear type, the valve sensitivity is more or less constant at any
flow. Linear valves are used when the pressure drop over the control valve is fairly
constant and a linear relationship exists between the controlled variable and the flow
rate of the manipulated variable. In equal-percentage type, the valve sensitivity
increases with increasing flow rate. The valve sensitivity at any given flow rate is a
constant percentage of the given flow rate, thus the term equal-percentage. Equal-
percentage valves are often used when the pressure drop available over the control
valve is not constant. This occurs when there are other pieces of equipment in the
system that acts as fixed resistances. The pressure drops over these parts of the process
vary as the square of the flow rate. At low flow rates, most of the pressure drop is taken
over the control valve since the pressure drop over the rest of the process equipment
is low. At high flow rates, the pressure drop over the control valve is low. In this
situation the equal-percentage tends to give a more linear relationship between flow and
control valve position than does linear valve.
Available Typical Typical Typical Relative Relative Noise or Available Flow Application Pressure Best economic
Valve Main size ranges, standard standard pressure flow shutoff cavitation control range- temperature drop size range
style characteristics inches body end range capacity capability trim characteristic ability ability capability inches (mm)
mm matcrials connection option
Regular Heavy duty ½16 Cast iron, carbon Flanged, To ANSI 250 Medium Excellent Yes Equal percentage, Moderate Quite low to High 14
sliding stem versatile (15400) steel, alloy steel, welded, linear, quick opening very high (2550)
stainless steel screwed special
Bar stock Compactness ½3 Stainless steel, Flangeless, To ANSI 600 Low Excellent No Equal percentage, Moderate Moderate Moderate ½1
(1580) nickel alloys screwed linear (1225)
Economy Light duty, ½4 Bronze, Screwed, To ANSI 300 Low Good Yes Equal percentage, Moderate Moderate Moderate 12
sliding stem inexpensive (15100) cast iron, steel flanged linear (2550)
Through-bore On-off service 124 Carbon steel, Flangeless To ANSI 900 High Excellent Yes Equal percentage Low Moderate Moderate 48
ball (25600) stainless steel (100200)
Partial ball Characterized 124 Carbon steel, Flangeless, To ANSI 600 High Excellent Yes Equal percentage High Quite low to Moderate 48
for throttling (25600) stainless steel flanged quite high (100200)
Eccentric Erosion resistance 112 Carbon steel, Flageless, To ANSI 600 Medium Excellent No Linear Moderate Quite low to High 48
plug (25300) stainless steel flanged quite high (100200)
Swing- No seal 236 Carbon steel, cast Flangeless, To ANSI 2500 High Poor No Equal percentage Moderate Very low to Moderate 636
through (50900) iron, stainless lugged, quite high (150900)
butterfly steel welded
Lined Elastomer or TFE 224 Carbon steel, cast Flangeless To ANSI 150 High Good No Equal percentage Low Moderate Low 624
butterfly liner (50600) iron, stainless lugged (1500600)
steel
High Offset disk, 272 Carbon steel, Flangeless, To ANSI 600 High Excellent No Linear Low Very low to Moderate 672
performance flexible seals (501800) stainless steel lugged quite high (2501800)
butterfly
Special Custom to 224 Carbon steel, alloy Flanged, To ANSI 4500 Medium Excellent Yes Custom Moderate to Very low to High to very
sliding stem application (50600) steel, stainless welded high quite high high
Final Control Elements
197
198 Feedback Control System
Table 3.3 Control-Valve-Characteristic Recommendations for Liquid-Level, Pressure, and Flow Control
Liquid-level systems
Constant DP Linear
Decreasing DP with increasing load, DP at Linear
maximum load > 20% of minimum-load DP
Decreasing DP with increasing load, DP at Equal percentage
maximum load < 20% of minimum-load DP
Increasing DP with increasing load, DP at Linear
maximum load < 200% of minimum-load DP
Increasing DP with increasing load, DP at Quick opening
maximum load > 200% of minimum-load DP
Pressure control systems
3.4.2 Actuators
Actuators are the distinguishing elements between just valves and the control valves.
An actuator is a device that translates the output signal of the automatic controller into
a position of a member exerting large power. The actuators must provide an accurate
output position proportional to the input signal in spite of various forces acting on the
output member. The most important forces are:
· Inertia forces caused by the mass moving parts.
· Static friction forces during impending motion of two adjacent surfaces.
· Thrust forces caused by weight and unbalanced fluid pressure.
Thus, the actuator is often required to employ a power-amplifying mechanism. Actuators
When control valve closes, flow rate increases in measuring element.
Final Control Elements 199
are available with many designs, power sources, and capabilities. Proper selection of
actuators involves process knowledge, valve knowledge, and actuator knowledge. A
control valve can perform its function only as well as the actuator can handle the static
and dynamic loads placed on it by the valve. Proper selection and sizing of actuators are,
therefore, very important. The actuator represents a very significant portion of the total
control-valve package price, and careful selection can minimize costs. The following
parameters must be considered while selecting actuators:
· Power source availability
· Fail-safe characteristics
· Actuator capability (torque or thrust requirements)
· Control functions
· Economics
Like automatic controllers, the actuators may operate by pneumatic, hydraulic,
electrical or a combination of these means. Control systems may be classified as
electrical, hydraulic or pneumatic, depending on the type of actuator that is employed.
However, many control systems are of a hybrid type, such as electro-hydraulic and
electro-pneumatic. A control system that employs a hydraulic actuator may also have an
electrical sensor/transducer for feedback and a servo-valve. In addition, passive control
systems are also used in some simple applications, where no external source of power
is used for actuation.
Passive actuators
Regulators are the common passive control systems and are used for very simple
applications. Passive control systems are usually mechanical and employ components
such as springs, dash-pots, levers, linkages and gears to perform the control functions.
Figure 3.11 illustrates a passive mechanical-feedback control system for regulating
the liquid level in a tank. Let us assume that q1 and q2, and h are the deviations from
h
Control L
valve
qd
h
q 1e + q 1
q 2e + q 2
their equilibrium values and qd, be a disturbance flow. Here, q1e, q2e, and he are the
constant equilibrium values. In steady-state equilibrium, q1e = q2e, qd = 0, and he is a
constant. The control system is a regulator whose purpose is to maintain the head of
the liquid equal to its desired or reference value he when there is a disturbance flow qd.
A change in the liquid level h, is sensed by a float that is connected by a mechanical
lever to a control valve. A turn screw in the float-lever mechanism is used to change
the length L when a change hr in the set-point corresponding to the desired level he is
required. It is assumed that hr = 0, that is, there is no change in the desired value of
he. If the head h increases, the valve moves an amount z and reduces the flow to the
tank and vice-versa. For small displacements, the valve displacement z is related to the
float displacement h by
C
\
J (3.14)
D
where a and b are the lever lengths. For a small deviation, the linearized equation for
the flow control valve is given by
q1 = c1 z (3.15)
where c1 > 0. The negative sign in Eq. (3.15) indicates that when z increases, the flow
q1 decreases and vice versa. The continuity equation for the tank results in
FJ
S SF S # (3.16)
FV
where A is the tank cross-sectional area and the outflow q2 is obtained as
§ SI ·
S
¨ ¸J (3.17)
© 4 ¹
Here, R is the hydraulic resistance of the outlet orifice. Defining a time constant
t1 = AR/rg, Eq. (3.16) becomes
SI
S SF S
U & J (3.18)
4
The block diagram is obtained from Eqs. (3.14), (3.15) and (3.18), as shown in
Figure 3.12(a). Letting k1 = ac1/b, the block diagram of Figure 3.12(a) can be redrawn,
as shown in Figure 3.12(b), where hr, which is the reference or desired change in the
liquid level, has been set to zero and h becomes the error. When modelling a regulator,
all variables can be represented as deviations from the equilibrium state that is required
to be maintained in the presence of disturbances. When the set-point is not changed, the
reference input is then set to zero.
Advantages:
· A passive feedback control system does not use any external power source for
sensing, error detection, amplification or actuation.
· The energy available in the input to each and every component of the system
is only used to produce its output.
Final Control Elements 201
qd
h( s) z q1 + h( s)
a R/Sg
– c1
b + U1s + 1
(a)
qd
hr = 0 error q1 + h
a R/Sg
–c1
b + U1s + 1
(b)
Electrical actuators
In applications such as machine tools and robotics, dc electric motors are commonly
used as actuators for relatively low power requirements. For large loads, hydraulic
actuators are preferred because of size and weight considerations. The electrical control
systems most widely used in industry employ dc motors.
However, ac motors are relatively inexpensive and two-phase induction motors are
sometimes used for low-power (fractional horse-power) applications. Three-phase
induction motors with pulse-width modulated power amplifiers are currently very
popular in high-power control applications. A special dc motor used in digital control
systems is called a stepper or step motor. Its input consists of electrical pulses and the
motor converts each pulse into fixed angular displacement. Stepper motors, however,
produce a low torque.
The schematic diagram of a dc electrical position-control system, used in robotics
for arm position control, is shown in Figure 3.13. Usually, each degree of freedom
employs its own actuator. A load with an inertia I and viscous friction with coefficient
c is to be positioned at some desired angle qr. The desired angle qr may be set as shown
in Figure 3.13 on an input potentiometer, which provides the reference voltage Er.
Alternately, the reference voltage Er corresponding to qr may be provided by a computer
storage through a digital-to-analog converter. The controlled position q0 is measured by
a rotary variable-differential transformer (RVDT), which supplies a feedback voltage E0.
A rotary potentiometer could be employed to sense q0, but it is usually avoided for
continuous use because the contact of the sliding wiper reduces its life. In RVDT and
its rectilinear counterpart, the linear variable-differential transformer (LVDT), there is
no physical contact between the core and the coil and hence the mechanical components
do not wear out or deteriorate. The corresponding absence of friction leads to high
resolution, no hysteresis and high reliability.
202 Feedback Control System
Input potentiometer
Rr
ac supply
Ra La
Er +
T R0
ia
A mplifier Rectifier L oad
Ec Ea I
if
E0
RV DT
The error voltage (Er E0) is amplified by an electronic amplifier that supplies voltage
Ec to a rectifier, which is a power amplifier. The rectifier is supplied by an external
single-phase or three-phase ac power and it amplifies its input voltage Ec to produce an
output voltage Ea, which is supplied to the armature of a dc motor. The motor is
armature controlled with fixed field current. It produces a torque T to position a load
connected to its shaft so that the controlled position q0 tends to equal the command qr.
The reference voltage Er is proportional to the command position qr, and is given as
Er = c1qr (3.19)
The feedback voltage E0, provided by the RVDT, is given by
E0 = c2q0 (3.20)
The electronic amplifier amplifies the error voltage and produces a voltage Ec so that
Ec = ka(Er E0) (3.21)
where ka is the gain of the amplifier.
The voltage Ec is fed to the driver of the rectifier. The rectified output voltage Ea
depends on the firing angle of the pulses relative to the ac supply waveform. A linear
relationship between the input voltage Ec and output voltage Ea can be obtained. The
time constants associated with the rectifier are negligibly small. Neglecting the
dynamics of the rectifier, the output voltage can be given as
Ea = kr Ec (3.22)
Final Control Elements 203
8..
MT (3.23)
Q 8
where V0 corresponds to the zero firing angle of the rectifier and VLL is the ac line-to-
line rms voltage. The block diagram of Eqs. (3.19), (3.20), (3.21) and (3.22) is shown in
Figure 3.14.
Rr Er + Ec Ea
c1 ka kr
–
E0
R0
c2
Advantages:
· Electrical control systems are very versatile.
· Its characteristics can be altered and the control law modified by simple and
inexpensive resistance-capacitance networks.
· Electrical components have an advantage in applications where control signals
are transmitted over long distances as in remote control.
Disadvantage:
· The torque-speed characteristics of ac motors are highly nonlinear unlike those
of separately excited dc motors.
Hydraulic actuators
Hydraulic actuators are used for industrial process control. They accept a signal from
a pneumatic or an electric controller and employ hydraulic pressure to drive an output
member. The hydraulic actuator is used where high speed and large power are required.
The schematic diagram of a hydraulic actuator based control system for the speed
control of a prime mover is shown in Figure 3.15. The prime mover may be an engine
or a turbine. The desired change in reference speed wr is set by the throttle position on
an input potentiometer, which provides a reference command voltage Er.
The controlled speed w 0 is sensed by a tachometer which supplies the feedback
voltage E0. The tachometer is driven from the prime mover shaft. The servo-amplifier
determines the error voltage (Er E0) and supplies a current I to the servo-valve torque
motor. The servo-valve admits hydraulic fluid to one side of the actuator, depending on
the direction of the current i. The actuator positions a fuel control valve. The
change in the fuel flow qf causes a change in the speed of the prime mover. A load
consisting of inertia I and viscous friction with coefficient c is connected to the prime
mover shaft.
204 Feedback Control System
Input potentiometer
Xr Supply
Return
Er
+
Throttle
Servo-
Servo- valve
amplifier
– Fuel control
valve
A ctuator Ya
E0 qf
T X0
Tachometer Prime L oad
mover I
§ ·
¨ ¸
¨ M ¸K
ZU
¨ [ ¸ (3.27)
¨ X & X & ¸
© P P ¹
Since the actuator positions a valve, its load is negligible and the actuator
displacement ya is given as
MI
[C
ZU (3.28)
&
For a small displacement ya of the fuel-control valve from its equilibrium position,
a linear relationship for the charge qf in fuel flow is given by
qf = c3 ya (3.29)
where it is seen from Figure 3.15 that c3 > 0.
Xr Er + i k1 xs
kg
Ya
c1 ka 1 2 [ c3
D + D + 1 D
– X 2
n
Xn
E0
X0
c2
A partial block diagram, using Eqs. (3.24) to (3.29), is shown in Figure 3.16.
Advantages:
· A comparatively small-size hydraulic actuator using high-pressure fluid can
develop very large forces or torque to provide rapid acceleration or deceleration
of heavy loads.
· For the same horsepower, hydraulic actuators are lighter than electrical motors
and considerable reduction in size and weight can be achieved.
· Hydraulic components are more rugged and more resistant to vibrations and
shocks than electrical components.
· Availability of both rectilinear actuators and rotary hydraulic motors makes the
design versatile.
· Hydraulic fluid can be used to carry the heat generated in the system and it also
acts as a lubricant.
Disadvantages:
· A source of pressurized hydraulic fluid with supply and return lines is required.
206 Feedback Control System
Pneumatic actuators
Pneumatic actuators may operate directly from the pneumatic output signal from a
pneumatic controller, or they may employ a separate source of compressed air. The
schematic diagram of a pneumatic actuator based control system for temperature
control of a room is shown in Figure 3.17. The control systems is a regulator whose
objective is to maintain a constant room temperature T1 corresponding to the set-point,
Ta
Pneumatic thermostat
ps T2
N ozzle
10° T1
xf
20°
P2
Flapper
q
k
Tr
Bimetal
A ctuator 30°
Heat exchanger
Control valve
z
Figure 3.17 Schematic diagram of a pneumatic temperature-control system.
A change in the distance between the nozzle and the flapper causes a change in
the pressure p1 behind the nozzle and consequently a change in the pressure p2 inside
the chamber of the diaphragm-type actuator. The force thus generated, changes the
position z of the flow-control valve. The flow-control valve allows hot or chilled water
to the heat exchanger in the room such that the controlled temperature T1 corresponds
to the set-point temperature Tr.
To control the linear mathematical model, let us consider that all variables denote
deviations from their constant equilibrium values. The displacement xf of the flapper, as
shown in Figure 3.17, increases when the set-point is increased by Tr. Also, the bimetal
is mounted such that xf decreases when the room temperature increases by T1. Hence
for small deviations from equilibrium, a linear relationship is given by
xf = c1 T1 c2T1 (3.30)
where c1 > 0 and c2 > 0.
When the nozzle is completely blocked so that there is no flow through it, the
pressure behind the nozzle is the supply pressure ps. When the nozzle is completely
unblocked, the pressure behind it approaches the atmospheric pressure pa. When the
flapper displacement is xf, the distance between the nozzle and flapper decreases by xf.
Within the linear range, the increase in the pressure behind the nozzle is given by
p1 = k1 (xf) = k1xf (3.31)
where k1, with k1 > 0, is the slope of the curve at the equilibrium position. The
transfer function between p1 and pressure p2 in the actuator chamber is given by
(tD +1 ) p2 = p1 (3.32)
Let m, c and k be the mass, coefficient of viscous friction and spring constant,
respectively, and A be the area of the diaphragm. Force balance yields
(mD2 + cD + k )z = p2 A
i.e.
§ [ ·
M ¨¨ & & ¸¸ \ R # (3.33)
© XP XP ¹
The valve motion is relatively slow and the valve mass is small. Hence wn is quite
large and usually beyond the frequency range of interest. Hence, we assume that the
force is balanced only by the spring and approximate Eq. (3.33) by
kz = p2 A (3.34)
And the linear equation for the deviation of the flow rate through the control valve
is given by
qf = c3 z (3.35)
By combining Eqs. (3.32) to (3.35), we obtain
(tD +1) qf = k2 p1 (3.36)
where
qf k2 = Ac3/k
208 Feedback Control System
The coefficient of heat transfer in the heat exchanger increases with the fluid flow
rate qf . For small deviations from equilibrium, it is assumed that the heat flux q is
proportional to the fluid flow rate qf . Hence, it follows that
q = c4 qf (3.37)
By assuming that the room can be lumped into one lump at temperature T1 and
the walls into another lump at temperature T2. The state transfer function can be given
as
§ · § ·
¨ M IU & ¸ ¨ ¸
¨ ¸S ¨ ¸ 6C
6
¨ [ ¸ ¨ [ ¸ (3.38)
¨ X & X & ¸ ¨ X & X & ¸
© P P ¹ © P P ¹
where q and Ta are the disturbance inputs, respectively.
The block diagram of the pneumatic temperature control system is obtained from
Eqs. (3.30), (3.31), (3.36), (3.37), and (3.38) are shown in Figure 3.18.
Ta
1/ka
U1D + 1
+ kg( U1D + 1)
+ xf p1 k2c4 q+ T1
c1 c1 1 2 [
UD + 1 D + D+ 1
X n2 Xn
–
c2
Advantages:
· The working fluid commonly used in pneumatic systems is air. It is readily
available and return lines are not required.
· A leak in the system does not pose any fire hazard as flammable fluids and
electrical sparks are absent.
· Pneumatic components are easy to maintain and are quite rugged.
· The initial cost of pneumatic components is less than that of electrical and
hydraulic components.
Hence pneumatic actuator based control systems are preferred in chemical and
petrochemical process-control applications.
Control Requirements of Feedback Control Systems 209
Disadvantages:
· Air is compressible fluid when compared to hydraulic fluids and hence pneumatic
control systems are slow acting with large time constants.
· Air lacks any lubricating properties.
· The normal operating pressure of pneumatic systems is much less than that of
hydraulic systems and consequently the output power is also less for the same
size.
· A source of compressed air is required and it is not as readily available as
electrical power.
c
m Settling time
Tolerance band
Figure 3.20 Variation of the controlled variable after a step change of command (desired value).
Load
Figure 3.21 Variation of the controlled variable after a step change of load (disturbance).
As shown in the figure, the shorter the settling time and smaller the maximum
deviations from the new steady-state value of the controlled variable, the better is the
dynamic behaviour of the feedback system. The measure of the dynamic behaviour of
the feedback system is given by the shaded area which is the product of the settling
time and the maximum deviation from the new steady-state value of the controlled
variable after a step change in load (disturbance) condition. In general, the behaviour
of the feedback control system resulting from a disturbance differs from that resulting
from a change in command (different rise times, etc.).
212 Feedback Control System
e m c e c
k1 k2 k
(a) (b)
Figure 3.22 Series connection of control elements.
v + m1 Controlled
Controller, kc c
system, kcs
Figure 3.24 shows the block diagram of a feedback control system without a
disturbance input. In the feedback circuit, r and c are the reference input and the
controlled variable respectively.
r e m1 c
k1 k2
+
4 (3.42)
M
By means of R, the persisting effect of the disturbances acting on the feedback
circuit can be calculated. If the disturbance acts on control variable m1, then the block
diagram of the feedback control system with additive disturbances, is modified, as shown
in Figure 3.25(b). From the figure, we get
m2 = m 1 + n 1
and
c1 = k2(m1 + n1)
= k2[(r c1)k1 + n1]
= k1k2r k1k2c1 + k2n1
M
T M P (3.43)
M M
214 Feedback Control System
Disturbance
n
r e m1 m2 c
k1 k2
+
(a)
Disturbance
n1
r e m1 m2 c1
k1 k2
+
(b)
n1
k2
r e m1 c1
k1 k2
+
(c)
Figure 3.25 Feedback control system with additive disturbance.
In case of Eq. (3.43), the disturbance k2n1 acts on the controlled variable c1. To
avoid calculation, the block diagram of Figure 3.12(b) is replaced by that of Figure
3.25(c). Equation (3.41) may then be employed with k2n1 substituted for n.
Multiplicative disturbances. A block diagram of a feedback control system with
multiplicative disturbances is shown in Figure 3.26. For a closed-loop circuit without
disturbance, the value of the controlled variable (output) can be derived from Eq. (3.41)
as
M
E r (3.44)
M
Block Diagram and Closed-Loop Response of a Feedback Control Systems 215
Disturbance, n
r e m1 c1
k1 k2
+
If a disturbance affects one of the circuit elements, as shown in Figure 3.26, this
is called the multiplicative disturbance. Due to the effect of n, k2 changes to
K = k2n¢ with n¢ = f(n)
The controlled variable c changes to
M P
E T (3.45)
M P
È M P M Ø
E E É T (3.46)
Ê M P M ÙÚ
The expression in parentheses in Eq. (3.46), gives the deviation in the controlled
variable normalized with respect to the rated value of reference input.
In any closed-loop system, the following errors are unaffected by the control factor
and directly affect the accuracy:
· Errors in the command transducer
· Errors in the response transducer
· Errors in the controller amplifier.
where Gp, Gd, Gm, Gc and Gf are the transfer functions between the corresponding inputs
and outputs.
Controller mechanism d
Measuring
ym device
(a) Schematic F
U
Process
Gd (s)
Final control
Controller element
+
[52
U G
U E
U O
U + [
U
Gc(s) Gf (s) Gp(s)
+
[O
U
Gm(s)
Figure 3.27(b) shows the block diagram for the generalized closed-loop system and
it is nothing more than a pictorial representation of Eqs. (3.47), (3.48), (3.49a), (3.49b)
and (3.50). Notice the direct correspondence between the schematic of Figure 3.27(a) and
block diagram of Figure 3.27(b).
The series of blocks between the comparator and the controlled output (i.e. Gc, Gf,
and Gp) constitutes the forward path, while the block Gm is on the feedback path
between the controlled output and the comparator.
If G = GcGf Gp then Figure 3.28(a) shows a simplified but equivalent version of the
block diagram. Algebraic manipulation of the equation yields
O
U )H
U)E
U > [52
U )O
U [
U@ [using Eq. (3.48)]
)H
U)E
U > [52
U [O
U@ [using Eq. (3.49a)]
)H
U E
U
)H
U)E
UG
U [using Eq. (3.49b)]
Block Diagram and Closed-Loop Response of a Feedback Control Systems 217
F
U
Gd (s)
+
[52
U G
U + [
U
G(s)
+
Gm(s)
(a)
F
U
Gload(s)
+
[52
U
GSP(s) [
U
+
(b)
Figure 3.28 Simplified block diagram of generalized closed-loop system.
[
U ^ `
) R
U ) H
U)E
U > [52
U )O
U [
U@ )F
UF
U
Its readjustment gives
) R
U)H
U)E
U )F
U
[
U [52
U F
U (3.51)
) R
U)H
U)E
U)O
U ) R
U) H
U)E
U)O
U
Equation (3.51) gives the closed-loop response of the process. We notice that it is
composed of two terms. The first term shows the effect on the output of a change in
the set-point, while the second constitutes the effect on the output of a change in the
load (disturbance). The corresponding transfer functions are known as closed-loop
transfer functions. In particular,
) R)H )E )
)52 (3.52)
) R)H )E)O ))O
is the closed-loop transfer function for a change in the set-point and
)F )F
)NQCF (3.53)
) R) H )E)O ))O
218 Feedback Control System
is the closed-loop transfer function for a change in the load. Figure 3.28 (b) shows a block
diagram equivalent to that of Figure 3.28(a), but further simplified.
For every feedback control system we can distinguish two types of control
problems: servo problem and regulator problem.
Servo problem. The disturbance does not change [i.e. F
U while the set-point
undergoes a change. The feedback controller acts in such a way as to keep y close to
the changing ySP. In such a case,
[
U )52
U [52
U (3.54)
Regulator problem. The set-point remains the same [i.e. [52
U ] while the load
changes. Then
y(s) = Gload(s) F (s) (3.55)
and the feedback controller tries to eliminate the impact of the load changes and keep
y at the desired set-point.
From Eqs. (3.52) and (3.53), it can be easily seen that the closed-loop overall
transfer function GSP and Gload depend not only on the process dynamics but also on the
dynamics of the measuring sensor, controller and final control element.
To expedite the construction of the overall closed-loop transfer functions for any
feedback control loop, use the following rules:
(a) The denominator of the overall transfer functions for both the load and the set-
point changes is the same. It is given by
1 + (product of the transfer functions in the loop) or 1 + (GpGmGcGf)
(b) The numerator of an overall closed-loop transfer function is the product of the
transfer functions on the forward path between the set-point or the load and
the controlled output. Thus
· The transfer functions on the forward path between the set-point ySP and
output y are: Gc, Gf and Gp. Therefore, the numerator is Gc Gf Gp.
· The transfer functions on the forward path between the load d and the
output is only Gd. Thus corresponding numerator is Gd.
Verify these two rules with the overall closed-loop transfer functions GSP and Gload
[Equations (3.53) and (3.54)]. Also, these rules can be used to formulate the closed-loop
transfer function between an input anywhere in the loop and the output.
Sensor
(eyes)
Figure 3.29 Manual and automatic feedback control system for locomotive.
220 Feedback Control System
The early control system were predominantly manual with a human operator
acting as a closed-loop controller. When the human operator is properly trained, a
manual control system can provide a very high level of control and can easily adapt itself
to changing situations. However, complete knowledge concerning the mathematical
modelling and functioning of many sophisticated manual control systems are still
lacking.
Increasing demand for the improvements in the quality of control of monotonous,
repetitive or unpleasant tasks and other considerations, such as unsafe jobs or
remotely located jobs, have provided economic incentives for the development and
implementation of automatic feedback control systems. Recent developments in
sophisticated control and robotics systems have led to the replacement of almost all
manual control tasks in manufacturing and process industries with automatic control
systems.
Disturbance, v(t)
Controlled
Input e(t) Sensor and Inverted output, q
u(t)
actuator pendulum
r = 0 + Moment
(spring)
10°C Tr
Calibrated dial Environment
20°C qe
(potentiometer)
30°C
40°C Te
Hot Room
water
Valve Radiator
qi
Temperature, T0
Disturbance, v(t)
Controller
Controlled
Input, Room output, c(t)
r(t) e(t) Control u(t) temperature
Actuator
algorithm control
+
Sensors/
transducers
One example of a multivariable control system would be several boilers discharging into
a common steam header, whose pressure must be controlled by manipulating their
individual firing rates. Another example could be parallel compressors drawing suction
from, or discharging into, a common header whose pressure is to be controlled.
Figure 3.32 shows three examples of multivariable control systems. In the
multivariable mechanical systems, as shown in Figure 3.32(a), the positions y1 and y2
of masses m1 and m2, respectively, are to be controlled independently by two feedback
control systems. In the first feedback control system, the actual position y1 is sensed and
after it is compared with its command input, a force F1 is generated to correct the error.
Similarly, force F2 is generated to control y2. However, it is seen that F1 affects not only
y1 but also y2. Also, F2 affects not only y2 but also y1. This system is multivariable and
the two control systems cannot be designed independently of each other as the
interaction must be considered.
In another example, as shown in Figure 3.32(b), it is necessary to control both
the flow and the composition of the outlet stream. To accomplish this objective,
we manipulate the flow of each of the two inlet streams. Figure 3.32(c) shows a chemical
reactor for which it is necessary to control the outlet temperature and composition.
The manipulated variables in this process are the cooling water flow and the process
flow.
y1 y2
m1 F1 m2 F2
Stirrer
AC FC
m1
AT FT
V
w1, x1
m2
x
w
w2, x2
Feed
Cooling
water
TT AT
All multivariable controllers rely on a model that represents the process behaviour.
The accuracy of the model determines the performance of the controller.
One of the biggest benefits of multivariable control technology is the ability to
control a process in regions of operational constraints. When encountering these
constraints the controllers ability to continuously drive the process toward optimal
performance is the basis of this technology. Many multivariable controllers come with
linear or nonlinear programming ability to allow building an objective function for local
optimization. The ability of the multivariable controller to adapt to changing process
conditions allows the controller to operate longer without retuning. The multivariable
controllers have the ability to work with nonlinear processes.
increase the heat when the temperature was above the set-point and turn it off when
it was below. Loops with positive feedback lock at one extreme or the other.
Negative feedback, on the other hand, works toward restoring balance. If the
temperature is too high, the heat is reduced. The heating action taken is manipulated
negatively, in effect, to the direction of the controlled variable, temperature.
y(t )
Desired level
of operation
t
Figure 3.33 Alternative closed-loop responses.
But, if our criterion had been Keep the maximum deviation as small as possible
or return to the desired level of operation and stay close to it in the shortest time, we
would have selected the other controller, yielding the closed-loop response of type B.
Similar dilemmas will be encountered quite often during the design of a controller.
For every process control application, we can distinguish two criteria: steady-state
performance criteria and dynamic response performance criteria.
The principal steady-state performance criterion usually is zero error at steady
state. In most situations, the proportional controller cannot achieve zero steady-state
error, while a PI controller can. Also, we know that for proportional control, the steady-
state error (offset) tends to zero as Kc ®ÿ ¥.
The evaluation of dynamic performance of a closed-loop system is based on the
following two types of commonly used criteria:
(a) Criteria that use only a few points of the response. They are simpler, but only
approximate.
(b) Criteria that use the entire closed-loop response from time t = 0 until t is very
large. These are more precise but also more cumbersome to use.
In this section, simple performance criteria will be dealt with in detail, which are
based on some characteristic features of the closed-loop response of a system. Some of
the factors that affect the performance of control loop, is described below, and their
quantitative control performance measures are presented.
xA2
A
xA3
AC
streams, A and B, are mixed in three series tank, and the output concentration of
component A is controlled by manipulating the flow of stream A. Here, we consider step
changes to the set-point; these changes represent the situation in which the plant
operator occasionally changes the value and allows a considerable time for the control
system to respond. A typical dynamic response is given in Figure 3.35. This is somewhat
idealized, because there is no measurement noise or effect of disturbance, but these
effects will be considered later.
A B
Controlled
Controlled and manipulated variables
P
Tr
C Manipulated
Time, t
Figure 3.35 Typical transient response of a feedback control system to a step set-point change.
228 Feedback Control System
Several facets of the dynamic response are considered in evaluating the control
performance.
Offset. Offset is the difference between final, steady-state values of the set-point, and
of the controlled variable. In most cases, a zero steady-state offset is highly desired,
because the control system should achieve the desired values, at least after a very long
time.
Reset time. Reset time (Tr) is the time from the step change in the set-point until the
controlled variable first reaches the new set-point. A short rise time is usually desired.
Integral error measures. Integral error measure indicates the cumulative deviation
of the controlled variable from its set-point during the transient response. Several such
measures are used:
Integral of the absolute value of the error (IAE):
Ô Ô
+#' ^52
V %8
V^ FV ^ [52
V [
V^ FV ^ G
V^ FV (3.56)
Ô
Integral of square of the error (ISE):
Ô ^52
V %8
V^ FV Ô ^ [52
V [
V^ Ô G
V FV
+5' FV (3.57)
Integral of product of time and the absolute value of error (ITAE):
+6#' Ô V^52
V %8
VFV^ Ô V^ [52
V [
V FV^ Ô V^ G
V^ FV (3.58)
Note that e(t) = ySP(t) y(t) is the deviation (error) of the response from the desired
set-point (SP). The IAE is an easy value to analyze visually, because it is the sum of
areas above and below the set-point. It is an appropriate measure of control performance
when the effect on control performance is linear with the deviation magnitude. The ISE
is appropriate when large deviations cause greater performance degradation than small
deviations. The ITAE penalizes deviations that endure for a long time. Note that IE is
not normally used, because positive and negative errors cancel in the integral, resulting
in the possibility for large positive and negative errors to give a small IE. A small
integral error measure (e.g. IAE) is desired.
Decay ratio (B/A). The decay ratio is the ratio of neighbouring peaks in an
undamped controlled-variable response. Usually, a periodic behaviour with large
amplitudes is avoided in process variables. Therefore, a small decay ratio is usually
desired, and an over-damped response is sometimes desired.
Periodic oscillation (P). Period of oscillation depends on the process dynamics and
is an important characteristic of the closed-loop response. It is not specified as a control
performance goal.
Selection of Feedback Control Systems 229
Settling time. Settling time is the time the system takes to attain a nearly constant
value, usually ±5% of its final value. This measure is related to the rise time and decay
ratio. A short settling time is usually favoured.
Manipulated-variable overshoot (C/D). This quantity is of concern because the
manipulated variable is also a process variable that influences performance. There are
often reasons to prevent large variations in the manipulated variable. Some large
manipulations can cause long-term degradation in equipment performance; an example
is the fuel flow to a furnace or boiler, where frequent, large manipulations can cause
undue thermal stresses. In other cases manipulations can disturb an integrated process,
as when the manipulated stream is supplied by another process. On the other hand,
some manipulated variables can be adjusted without concern, such as cooling water flow.
We will use the overshoot of the manipulated variable as an indication of how
aggressively it has been adjusted. The overshoot is the maximum amount that the
manipulated variable exceeds its final steady-state value and is usually expressed as a
per cent of change in the manipulated variable from its initial to its final value. Some
overshoot is acceptable in many cases; little or no overshoot may be the best policy in
some cases.
Load disturbances. Load disturbances may be in many forms, both periodic and
aperiodic. The aperiodic are principally step, ramp and random (stochastic) functions.
The step is the most difficult test, because it contains all frequencies from zero (steady
state) to a maximum, based on the rate of rise of the step. By contrast, the pulse may
rise as fast and therefore may contain the high-frequency spectrum, but it returns to
its original starting point, having no steady-state component. A pulse is essentially the
derivative of a step. Being the integral of a step, the ramp contains principally low-
frequency components. While a controller having integral action can eliminate steady-
state offset, it cannot prevent offset from developing during a ramp. In a plant
environment many random disturbances are encountered. Noise is a random
disturbance, whose frequency is above the range to which a control loop can respond.
Step disturbance. Often, an important disturbance occurs infrequently and
in sudden manner. The causes of such disturbances usually change to other parts of
the plant that influence the process being considered. An example of a step upset in
Figure 3.34 would be the inlet concentration of stream B.
Responses of the outlet concentration, without and with control, to this disturbance
are given in Figures 3.36(a) and (b). We will often consider dynamic responses similar
to those in Figure 3.36 when evaluating ways to achieve good control that minimizes
the effects of step disturbances. The explanations for the measures are the same as
for set-point changes except for rise time, which is not applicable, and for the
following measure, which has meaning only for disturbance responses and is shown in
Figure 3.36(b).
Maximum deviation. The maximum deviation of the controlled variable from the set-
point is an important measure of the process degradation experienced due to the
disturbance. For example, the deviation in pressure must remain below a specified
value. Usually, a small value is desirable so that the process variable remains close to
its set-point.
Stochastic inputs. A process typically experiences a continual stream of small and
large disturbances, so that the process is never at an exact steady state. A process that
is subjected to such seemingly random upsets is termed a stochastic system. The
response of the example process to stochastic upsets in all flows and concentrations is
given in Figures 3.37(a) and (b) without and with control.
The major control performance measure is the variance (s 2CV), or standard
deviation (sCV), of the controlled variable, which is defined as follows for n data
points:
P
T %8
PK Ç
[ [K (3.60)
This variable is closely related to the ISE performance measure for step
disturbances. The relationship depends on the approximations that (a) the mean can be
Selection of Feedback Control Systems 231
Manipulated
Time, t
(a) Without feedback control
Maximum deviation
Controlled and manipulated variables
Controlled
Manipulated
Time, t
(b) With feedback control
Figure 3.36 Transient response of three-tank mixing process (Figure 3.34).
replaced with the set-point, which is normally valid for closed-loop data and (b) the
number of points is large.
P
6
P K Ç
[ [K
6 Ô
[52 [ FV (3.62)
Since the goal is usually to maintain controlled variables close to their set-point,
a small value of the variance is desired. In addition, the variance of the manipulated
variable is often of interest, because too large a variance could cause long-term damage
to equipment (fuel to a furnace) or cause upsets in plant sections providing the
manipulated stream (steam-generating boilers).
Sine inputs. An important aspect of stochastic systems in plants is that the
disturbances can be thought of as the sum of many sine waves with different amplitudes
and frequencies. In many cases, the disturbance is composed predominantly of one or
232 Feedback Control System
Controlled
Manipulated
Time, t
(a) Without feedback control
Controlled
Controlled and manipulated variables
Manipulated
Time, t
(b) With feedback control
Figure 3.37 Transient response of the example process to a stochastic disturbance.
a few sine waves. Therefore, the behaviour of the control system in response to sine
inputs is of great practical importance. Because through this analysis we learn how the
frequency of the disturbance influences the control performance. The responses of the
example system to a sine disturbance in the inlet concentration of stream B with and
without control are given in Figures 3.38(a) and (b). A control performance is measured
by the amplitude of the output sine, which is often expressed as the ratio of the output
to input sine amplitudes. Again, a small output amplitude is desired.
Selection of Feedback Control Systems 233
Controlled
Manipulated
Time, t
(a) Without feedback control
Controlled
Controlled and manipulated variables
Manipulated
Time, t
(b) With feedback control
Figure 3.38 Transient response of the example system to a sine disturbance.
Product-quality control
For most industrial products, selling price is independent of quality within a particular
grade. If a product fails to meet the quality specifications which define that grade, it
cannot be sold for the price commanded by that grade. Because economic performance
changes stepwise when a product specification is violated, operators try to place ample
margin between its measured quality and the specification. However, a higher-quality
product typically costs more to produce.
Therefore, the maximum profit will usually be realized when the product quality
meets its specification and no better. As a consequence, the quality set-point must be
positioned far enough on the acceptable side of the specification to reduce the likelihood
of violation. With better control, variation in quality will be less, allowing the set-point
to move closer to specifications, thereby reducing operating cost.
Error magnitude
As a result of a sudden load change, a controlled variable will be driven away from set-
point until the control action can reverse its direction and restore equilibrium. The
maximum deviation or error resulting from the disturbance could be enough to initiate
a shutdown command or to violate specification in product quality. The plant shutdown
is a costly event, which is irreversible. Each time it happens, an economic penalty is
immediately applied to the process. Therefore, error magnitude would be more
important for evaluating the probability of a shutdown. If the error magnitude per unit
load change can be calculated, then the size of the load change, which will drive the
controlled variable from its set point to the trip point can be determined. Increasing the
control effectiveness can reduce the likelihood of shutdown.
Statistical properties
The quality of a product is often evaluated in terms of mean and standard deviation for
a number of samples chosen from a lot. While the concept has been applied principally
to piece parts, it is equally applicable to any measured variable taken over time. The
principal objective of a control system is to minimize the deviation of the controlled
variable, and the standard deviation is simply a transformation of that deviation over a
statistically significant number of samples or time of operation. While responses to
individual load changes were evaluated to minimize IE and IAE, statistical significance
requires the elimination of skew, attributable to, for example, upsets in one direction
only. To achieve this, the number of samples or time of observation needs to be large
relative to the response to any individual disturbance.
The economic incentive behind the standard deviation criterion is essentially
the same as for the error magnitude criterion. The principal distinction is that
error magnitude expresses a binary situation, i.e. either the controlled variable
violates the specification limit or not. The standard deviation criterion estimates the
percentage of time the controlled variable violates the specification based on a normal
distribution curve.
Practical Applications of Feedback Control Systems 235
Proportional
Proportional
x Sign
· Sign
x Units
· Units
+ Set-point
(SP)
Integral
Integral
·x Wind-up
Windup –
Derivative
Derivative
· Filter
Output
Output processing
processing x Filter
·x Initialization
Initiaization InputInput processing
processing
·x Limits
Limits x· Validity
Validity
x· Linearization
Linearization
x· Filtering
Filtering
Sensor
Process
order of the four major topics given in Table 3.4. First, a few key equipment
specifications are presented, followed by input processing calculations, performed before
the control calculation. Then, modifications to the PID control calculation are explained.
Finally, a few issues related to the output processing are presented.
236 Feedback Control System
Level of liquid (or solid) in vessels are typically expressed as a per cent of the span
of the sensor rather than in length (metres). Flows are often measured by pressure-drop
across an orifice meter. Since orifice plates are supplied in a limited number of sizes,
the equipment is selected to be the smallest size that is (just) large enough to measure
the largest expected flow. The expected flow is always greater than the design flow; as
a result of the limited equipment and expected flow range, the flow sensor can usually
measure at least 120% of the design value; and its range is essentially never an even
number such as 0 to 100 m3/day.
These simple guidelines do not satisfy all situations, and therefore, two important
exceptions are introduced. The first special situation involves normal operations, such
as start-up and major disturbances, when the variable covers a much greater range.
Clearly, the suppressed ranges about normal operation will not be satisfactory in these
cases. The usual practice is to provide an additional sensor with a much larger range
to provide a measurement, with lower accuracy and reproducibility, for these special
cases. For example, the furnace outlet temperature, as shown in Figure 3.40, is
normally about 600°C, which varies from about 20 to 600°C during start-up and must be
monitored to ensure that the proper warm-up rate is attained. An additional sensor with
a range of 0 to 800°C could be used for this purpose. This could also be used for control
238 Feedback Control System
Flue gas
FC
Feed
PC
550650°C 0 800°C
TC TI
fc
Figure 3.40 Furnace outlet temperature controller (fired heater with simple control strategy):
TC = Temperature controller; PC = Pressure controller; TI = Temperature indicator;
fc = Fail-closed valve.
by providing a switch, which selects either of the sensors for control. Normally, the
controller tuning constants would have to be adapted for the two types of operation.
A second special situation occurs when the accuracy of a sensor varies over its
range. For example, a flow might be normally about 30 m3/h in one operating situation
and about 100 m3/h in the other. Since a pressure-drop across an orifice meter does not
measure the flow accurately for the lower one-third of its range, two pressure-drop
measurements are required with different ranges. For this example, the meter ranges
might be 0 to 40 and 0 to 120 m3/h, with the smaller range providing accuracy for
smaller flows.
Small valve
Large valve
Acid Acid
Feed
pH
AC
LC
achieve the desired pH. In this design, the position of the larger valve is changed
infrequently by the operator, and the position of the smaller valve is changed
automatically by the controller.
Therefore, the above discussion summarizes that sensors and final control
elements are sized to (just) accommodate the typical operating range of the variable.
Extreme over-sizing of a single element is to be avoided; a separate element with larger
range can be provided, if necessary.
Another important issue is the behaviour of control equipment when power is
interrupted. Naturally, a power interruption is an infrequent occurrence, but proper
equipment specification is critical so that the system responds safely in this situation.
Power is supplied to most final control elements (i.e. valves) as air pressure, and loss
of power results from the stoppage of air compressors or from the failure of pneumatic
lines. The response of the valve when the air pressure, which is normally 3 to 15 psig
is called its failure mode. Most valve fail-open or fail-closed, with the selection
determined by the engineer to give the safest process conditions after the failure.
Normally, the safest conditions involve the lowest temperature and pressure. As an
example, the flash drum in Figure 3.42 would have the valve failure modes shown in
the figure, with fo used to designate a fail-open valve and fc, a fail-closed valve. The
valve failure modes in the example set the feed to zero, the output liquid-flow to
maximum, the heating medium-flow to zero, and the vapour-flow to its maximum. All
of these actions tend to minimize the possibility of an unsafe condition by reducing the
pressure. However, the proper failure actions must consider the integrated plant. For
example, if a gas-flow to the process normally receiving the liquid could result in a
hazardous situation, the valve being adjusted by the level controller would be changed
to fail-closed.
The proper failure mode can be ensured through simple mechanical changes to the
valve, which can be made after installation in the process. Basically, the failure mode
is determined by the spring that directs the valve position when no external air pressure
240 Feedback Control System
PC
fo
FC
LC
fc
Heating
fc fo
medium
Figure 3.42 A flash separation unit with the valve failure modes.
provides a countering force. This spring can be arranged to ensure either a fully opened
or fully closed position. As the air pressure is increased, the force on the restraining
diaphragm increases, and the valve step (position) moves against the spring. Therefore,
the failure mode of the final control element is selected to reduce the possibility of
injury to personnel and of damage to plant equipment.
The selection of a failure mode also affects the normal control system, because the
controller output increases, a fail-open valve closes and a fail-closed valve opens. As a
result, the failure mode affects the sign of the process transfer function expressed as
CV(s)/MV(s), which is the response seen by the controller. As a consequence, the
controller gain used for negative feedback control is influenced by the failure mode. If
the gain for the process CV(s)/F(s), with F(s) representing the flow through the
manipulated valve, is K*p, the correct sign for the controller gain is given, as shown in
Table 3.6.
Table 3.6 Failure Mode verses Sign for the Controller Gain
and the actual process available, and (c) reduce the effects of high-frequency noise. In
this section, we will look more closely at the processing of the signal from the
completion of transmission to just before the control algorithm.
'2
( - (3.63)
S
where
F = Volumetric flow rate
r = Density
DP = Pressure difference across the orifice.
Typically, the sensor measures the pressure-drop so that
-
(
5 5
4 < (3.64)
S
242 Feedback Control System
Controllable disturbances
Uncontrollable disturbances
Measurement noise
Electrical interference
Sampling frequency
Thus, using the sensor signal directly (i.e. without taking the square root) introduces
an error in the control loop. The accuracy would be improved by using the square root
of the signal, as shown in Eq. (3.64), for control and also for process monitoring.
In addition, the accuracy could be improved further for important flow measurements
by automatically correcting for fluid density variations as follows:
5 5 4 :
( - (3.65)
5 5 4 <
with the subscript 1 for the pressure difference sensor signal and 2 for the density
sensor signal. By far, the most common flow measurement approach commercially used
in Eq. (3.64), with Eq. (3.65) used only when the accurate flow measurement is
important enough to justify the added cost of the density analyzer.
Another common example for sensor nonlinearity is the thermocouple temperature
sensor. A thermocouple generates a millivolt signal that depends on the temperature
difference between the two junctions of the bimetallic circuit. The signal transmitted for
control is either in millivolts or linearly converted to milliamps. However, the
relationship between millivolts and temperature is not linear. Usually, the relationship
can be represented by a polynomial or a piecewise linear approximation to achieve a
more accurate temperature value; the additional calculations are easily programmed as
a function in the input processing to achieve a more accurate temperature value.
Practical Applications of Feedback Control Systems 243
Engineering units
Another potential input calculation expresses the input in engineering units, which
greatly simplifies the analysis of data by operations personnel. This calculation is
possible only in digital systems, as analog systems perform calculations using voltage or
pressure. As already discussed, the result of the transmission and any correction for
nonlinearity in digital systems is a signal in terms of the instrument range expressed
in a per cent (0 to 100) or a fraction (0 to 1). The variable is expressed in engineering
units according to the following equation:
CV = Z + R(S3S30) (3.66)
with S3 the signal form of the sensor after correction for nonlinearity.
Filtering
An important feature in input processing is filtering. The transmitted signal represents
the result of many effects; some of these effects are due to the process, some are
due to the sensor, and some are due to the transmission. These contributions to the
signal received by the controller vary over a wide range of frequencies, as presented in
Figure 3.43. The control calculation should be based only on the responses that can be
affected by the manipulated variable, because very high-frequency components will
result in high-frequency variation of the manipulated variable, which will not improve
and may degrade the performance of the controlled system.
Some noise components are because of electrical interference and mechanical
vibration, which have a much higher frequency than the process response. Other noise
components are due to changes such as imperfect mixing and variations in process input
variables such as flows, temperatures and compositions. Some of these variations may
be closer to the critical frequency of the control loop. Finally, some measurement
variations, due to changes in flows and compositions, occur at frequencies much below
the critical frequencies. The effects of these slow disturbances can be effectively
attenuated by feedback control.
The very high-frequency component of the signal cannot be influenced by a process
control system, and thus is considered noise. The goal, therefore, is to remove the
unwanted components from the signal, as shown in Figures 3.44 and 3.45. The filter is
located in the feedback loop, and dynamics involved with the filter, like process
dynamics, will influence the stability and control performance of the closed-loop system.
This statement can be demonstrated by deriving the following transfer function, which
shows that the filter appears in the characteristic equation:
244 Feedback Control System
D(s)
GGdd(s)
(s)
Figure 3.44 Block diagram of a feedback loop with a filter on the measurement.
%8
U ) R
U)X
U)E
U
(3.67)
52
U ) R
U)X
U)E
U)H
U)U
U
If it were possible to separate the signal (true process variable) from the noise,
the perfect filter in Figure 3.45 would transmit the unaltered true process variable
value to the controller and reduce the noise amplitude to zero. In addition, the perfect
filter would do this without introducing phase lag. Unfortunately, there is no clear
distinction between the true process variable, which can be influenced by adjusting the
manipulated variable, and the noise, which cannot be influenced and should be filtered.
Also, no filter calculation exists that has features of a perfect filter in Figure 3.45.
1
Amplitude
ratio Signal Noise
Phase angle
Frequency
Figure 3.45 Amplitude ratio and phase angle of a perfect filter.
The filter calculation usually employed in the chemical process industries is a first-
order transfer lag and is given as
%8H
U %8O
U (3.68)
U H U
Practical Applications of Feedback Control Systems 245
where
CVf (s) = Value after filter
CVm(s) = Measured value before the filter
tf = Filter time constant
The gain is unity because the filter should not alter the actual signal at low
frequency, including the steady state. The frequency response of the continuous filter
is given as
#4 f = tan1(wtf ) (3.69)
X U H
100
Control performance expressed as
per cent increase in IAE without
80
filter
60
40
20
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
UH
R U
Figure 3.46 The effect of measurement filtering on feedback control performance [q/(q + t) = 0.33].
shows the per cent increase in IAE over control without the filter as a function of the
filter time constant. This case study has been calculated for a plant with fraction dead
time of 0.33 under a PI controller. Thus, the results are typical but not general; similar
trends can be expected for other systems.
246 Feedback Control System
Based on the goals of filtering, the guidelines in Table 3.7, are recommended for
reducing the effects of high-frequency noise in a typical situation. These steps should be
implemented in the order shown until the desired control performance is achieved.
1. Reduce the amplification Set derivative time to zero, Prevent amplification of high-
of noise by the control Td = 0 frequency component by
algorithm controller
2. Allow only a slight Select a small filter tf, e.g. Do not allow the filter to
increase in the IAE of the tf < 0.05(q + t) degrade control performance
controlled variable
3. Reduce the noise effects on Select filter time constant to Achieve a small amplitude
the manipulated variable eliminate noise, e.g. tf > 5/wn, ratio for the high-frequency
where wn is the noise frequency components
Normally, step-2 will take priority over step-3, because the controlled-variable
performance is of greater importance. If reducing the effects of high-frequency noise is
an overriding concern, the guidelines can be altered accordingly, such as achieving
step-3 while allowing some degradation of the controlled-variable control performance.
The final issue in filtering relates to digital implementation. A digital filter can be
developed by first expressing the continuous filter in the time domain as a differential
equation:
F%8H
V
UH %8H
V %8O
V (3.70)
FV
leading to the digital form of the first-order filter:
È 'V Ø
%8H P #
%8H P
#
%8O P with # GZR É Ù (3.71)
Ê UH Ú
This equation can be derived by solving the differential equation defined by
Eq. (3.70) and assuming that the measured value (CVm)n is constant over the filter
execution period Dt. The digital filter also has to be initialized when the calculations are
first performed or when the computer is restarted. The typical filter initialization sets
the initial filtered value to the value of the initial measurement.
(CVf)1 = (CVm)1 (3.72)
As is apparent, the first-order filter can be easily implemented in a digital
computer. However, the digital filter does not give exactly the same results as the
continuous version, because of the effects of sampling. Sampling a continuous signal
results in some loss of information.
Set-point limits
Limits are often placed on the set-point. Without a limit, the set-point can take any
value in the controlled-variable sensor range. Since the controlled-variable sensor range
Practical Applications of Feedback Control Systems 247
is selected to provide information during upsets and other atypical operations, it may
include values that are clearly undesirable but not entirely preventable. Limits on the
set-point prevent an incorrect value being introduced (a) inadvertently by the operator
or (b) by poor control of a primary in a cascade control strategy.
The sense switch has two possible positions, as defined in Table 3.8, using two
common technologies. This approach is not necessary, but is used so widely that control
engineers should be aware of the practice.
where
(Kc)s = Dimensionless (scaled) controller gain = Kc(CVr/MVr)
CVr = Range of the sensor measuring the controlled variable in engineering units
MVr = Range of the manipulated variable (100% for control valve)
The range of values for the unscaled controller gain (Kc) is essentially unlimited,
because the value can be altered by changing the units of measurement. For example,
a controller gain of 1.0 (weight %)/(% open) is the same as 1.0 ´ 106 (ppm)/(% open).
However, the scaled controller gain has a limited range of values, because properly
248 Feedback Control System
designed sensors and final control elements have ranges that give good accuracy. For
example, a very small dimensionless controller gain indicates that the final control
element would have to be moved very accurately for small changes to control the
process. A general guideline is that the scaled controller gain should have a value near
1.0. Scaled controller gains outside the range of 0.01 to 10 suggest that the sensor or
final control element may have been improperly selected.
Some commercial controller algorithms include a slight modification in the
proportional tuning constant term that does not influence the result of the controller
calculation. The controller gain is replaced with the term 100/PB, with the symbol
PB representing the proportional band, which is dimensionless. The proportional band
is calculated as PB = 100/(Kc)s.
The net PID controller calculation in Eq. (3.74) is unchanged because the controller
gain is calculated as (Kc)s = 100/PB. Thus, the use of gain or proportional band is
arbitrary; each gives the same control loop performance. However, the engineer must
know which convention is used in the controller and enter the appropriate value. Note
that, in fine tuning, the controller is modified to be less aggressive by decreasing the
controller gain or increasing the proportional band.
Reset wind-up causes a very poor control performance because of changes in plant
operation. The poor control performance occurs when the controller is again able to
adjust the final control element and achieve zero offset. Suppose that the reset wind-
up has caused a very large positive value of the calculated controller output because a
non-zero value of the error occurred for a long time. To reduce the integral term, the
error must be negative for a very long time. Thus, the controller maintains the final
control element for a long time simply to reduce the (improperly wound-up) value of the
integral mode.
The improper calculation can be prevented by many modifications in the standard
PID algorithm that do not affect its good performance during normal circumstances.
These modifications achieve anti-reset wind-up. The first modification explained here is
termed external feedback and is offered in many commercial analog and digital
algorithms. The external feedback PI controller is shown in Figure 3.47. The system
behaves exactly like the standard algorithm when the limitation is not active, as is
11
TIIss 1cc
T
demonstrated by the following transfer function, which can be derived by block diagram
manipulation based on Figure 3.47:
14243
/8
U È Ø
- E É
'
U Ê 6+ U ÙÚ (3.75)
MV*(s) = MV(s)
However, the system with external feedback behaves differently from the
standard PI controller when a limitation is encountered. When a limitation is active
in Figure 3.47, the following transfer function defines the behaviour:
MV*(s) = Constant
14243
/8
U (3.76)
/8
U - E '
U
6+ U
with MV*(s) being the upper or lower MV limit. In this case, the controller output
approaches a finite, reasonable limiting value of KcE(s) + MV*(s). Thus, external
feedback is successful in providing anti-reset wind-up. These calculations can be
implemented in either analog or digital systems.
250 Feedback Control System
14243
Ë 'V'P 6 Û
'/8P - E Ì 'P 'P F
%8P %8P %8P Ü
Í 6+ ' V Ý (3.77)
MVn = MVn1 + DMVn
This algorithm does not accumulate the integral as long as the past value of the
manipulated variable, MVn1, is evaluated after the potential limitation. When this
convention is observed, any difference between the previously calculated MV and the MV
actually implemented (final control element) is not accumulated.
Many other methods are employed to prevent reset wind-up. The two methods
described above are widely used and are representatives of the other methods. Here, the
key point to be noted is that anti-reset wind-up should be included in every control
algorithm that has integral mode, because limitations are encountered, perhaps
infrequently, by essentially all control strategies due to large changes in operating
conditions.
Reset wind-up is relatively simpler to recognize and correct for a single-loop
controller computing to a valve, but it takes on increasing importance in more complex
control strategies, such as cascade and variable-structure systems. Also, the general
issue of reset wind-up exists for any controller that provides zero offset when no
limitations exist.
Initialization
The PID controller requires special calculations for initialization. The specific
initialization required depends upon the particular form of the PID control algorithm.
Practical Applications of Feedback Control Systems 251
Typical initialization for the standard digital PID algorithm, from Eq. (3.77), is as
follows:
Ë 'V'P 6 Û
'/8P - E Ì 'P 'P F
%8P %8P %8P Ü
Í 6+ ' V Ý
where
MVn = MVn1 + DMVn
and
MV1 = MV0
That is DMV1 = 0 for n = 1 for initialization.
123
En1 = MV0 for n = 1
(3.79)
CVn2 = CVn1 = CVn for n = 1
The initialization strategy ensures that no large initial change in the manipulated
variable will result from outdated past values for the error or controlled variables.
The above discussion clearly shows that the simple, single PID equation, while
performing well under limited conditions, is not sufficient to provide feedback control
under various conditions experienced in realistic plant operation.
ILLUSTRATIVE EXAMPLES
Example 3.1 Derive closed-loop response of the liquid level in a tank.
Consider a liquid-level control system for the tank, as shown in Figure 3.48(a). The
level h is the controlled output while Fi is the load (disturbance) and Fo the manipulated
variable. The transfer functions for each component of the feedback loop are as follows:
Process: The material balance around the tank gives
FJ
# (K (Q
FV
hSP
Set-point
DP level sensor
hm
LT LC
Fluid inlet, Fi
e
Gc(s)
h
Fluid outlet, Fo
Process Fi (s)
1
Gd
As
+
e (s) c (s) Fo(s) +
hSP(s) § 1 1· · K pp 1
GG
c c KKpp ¨¨¨1 ¸ GGf f Gp h (s)
+ © t s¸ ¸
1s1 ¹ ¹ UWppss 1 1 As
hm(s) K ppDB
GG
mm
UW22ss22 22[U
]Ws 1 1
F \ F\
U [U \ - R 'R - RB J
FV FV
where z = hm (i.e. the value indicated by the measuring devices). Therefore, the transfer
function for the sensor is
- RB
JO
U J
U (3.83)
U U [U U
Control valve. Let us assume that for the control valve of this system, the
response is that of a first-order system. Then
-X
(Q
U E
U (3.85)
U X U
Figure 3.48(b) shows the block diagram for the closed-loop system with the transfer
function for each component of the loop. The closed-loop response of the liquid level will
be given by Eq. (3.51), where the transfer functions (Gp, Gd, Gm, Gc and Gf) are shown
in Figure 3.48(b). The servo problem arises when the inlet flow rate (Fi) remains
constant and we change the desired set-point. In this case, the controller acts in such
a way as to keep the liquid level (h) close to the changing desired value (hSP). On the
other hand, for the regulator problem, the set-point (hSP) remains the same and the
feedback controller acts in such a way as to eliminate the impact of the changing load
and keep h at the desired value (hSP).
Process. If T, Ti and Tst are deviation variables, then the transfer function of a
stirred-tank heater is given as
U -
6
U 6K
U 6UV
U (3.86)
U C U C
where
(K 7#V
C - -
U U 8 8 S %R
and
U = Overall heat transfer coefficient
At = Area of heat transfer
Tst = Temperature of the steam
T = Liquid temperature in the tank
V = Volume of tank. Set-point
TT TC
Ti
Temperature sensor
(thermocouple)
Heater
Steam Tst
(a) Closed-loop temperature control: TT = temperature transmitter; TC = temperature controller
Process Ti(s)
1/ W
sa
Control valve
Controller +
Tst(s)
TSP(S) e(s) Kp K
Kc T(s)
+ W ps 1 sa +
Thermocouple
Km
6O
U - O6
U (3.87)
G
U 652
U 6O
U (3.88a)
E
U - E G
U (3.88b)
Control valve. Assume first-order dynamics:
-X
6UV
U E
U (3.89)
U X U
Figure 3.49(b) shows a block diagram for the closed-loop system with the transfer
function for each component of the loop. The closed loop response is easily found to be
6
U )52
U652
U )NQCF
U6K
U (3.90)
where the closed-loop transfer functions GSP and Gload are defined as follows:
- -X
-E
U C U X U
)52
U (3.91)
- -X
-O -E
U C U X U
and
U
)NQCF
U U C
(3.92)
- -X
-O -E
U C U X U
Example 3.3 Three-tank mixing process (Figure 3.34), initially is operating in the
normal range. At a time of about 20 minutes, it experiences a large
increase in the inlet concentration that causes the control valve to close
and thus reach a limit. After about 140 minutes, the inlet concentration
returns to its original value. Determine the dynamic response of the
feedback control system with and without anti-reset wind-up.
The results of simulations are presented in Figure 3.50. In Figure 3.50(a), the
dynamic response of the system without anti-reset wind-up is shown. As usual, the
set-point, controlled variable and manipulated variable are plotted. In addition, the
calculated controller output is plotted for assistance in analysis, although this variable
is not normally retained for display in a control system. After the initial disturbances,
the valve position is quickly reduced to 0% open. Note that the calculated controller
output continues to decrease, although it has no additional effect on the valve. During
256 Feedback Control System
Offset
CV(t)
MV(t)
Valve
Controller
output
Time
(a) Without anti-reset wind-up
Offset
CV(t)
Time
(b) With anti-reset wind-up
Figure 3.50 Dynamic response of the three-tank mixing system.
Illustrative Examples 257
the time from 20 to 160 minutes, the controlled variable does not return to its set-point
because of the limitation in the range of the manipulated variable. When the inlet
concentration returns to its normal value, the outlet concentration initially falls below
its set-point. The controller detects this situation immediately, but it cannot adjust the
valve until the calculated controller output increases to the value of zero. This delay,
which would be longer had the initial disturbance been longer, is the cause of a rather
large disturbance. Finally, the PI controller returns the controlled variable to its set-
point, since the manipulated variable is no longer limited.
The case with anti-reset-wind-up is shown in Figure 3.50(b). The initial part of the
process response is the same. However, the calculated controller output does not fall
below the value of 0%; in fact, it remains essentially equal to the true valve position.
When the inlet concentration returns to its normal value, the controller output is 0%
and can rapidly respond to the new operating conditions. The second disturbance is much
smaller than that in Figure 3.50(a), showing the advantage of an anti-reset wind-up.
Example 3.4 What is the correct sense which positions for the temperature feedback
controller of a three-tank mixing process (Figure 3.34)?
Note that the process gain and failure mode of the control valve must be known
to determine the proper sense of the controller. In this example, the valve failure mode
is fail-closed. Therefore, an increase in the controller output signal results in (a) the
valve opening, (b) the fuel flow increasing, (c) the heat transferred increasing and (d) the
temperature increasing. The overall process loop gain is the product of all signs in the
system, which must be positive to provide the desired (negative) feedback control. Then
+1 +1
sign (loop gain) = sign (Kv) sign (Kp) sign (Ks) KsenseKc = +1
which gives
Ksense = sign (Kv) sign (Kp)
Example 3.5 For a three-tank mixing process (Figure 3.34), the concentration sensor
has a range of 5% A, and the control valve is fail-closed. Determine the
dimensionless controller gain, proportional band, controller sense and
reset time.
The process reaction curve identification was applied to determine values for the
controller gain in engineering units and the integral time, 30% opening/% A and 11
minutes, respectively. Therefore, the dimensionless controller gain and proportional
band are
%8T
QRGPKPI#
#
- E - E
/8T
QRGP
and
2$
- E U
258 Feedback Control System
As we know that the pressure drop through the coil DPc depends on the flow rate
F and can be calculated as
§( ·
u ¨ ¸
'2E (3.93)
© ¹
Similarly, the pressure drop over the control valve DPv is difference of the total pressure
drop available DPt and the pressure drop through the coil DPc and is calculated as
§( ·
'2X '2V u ¨ ¸
'2V '2E (3.94)
© ¹
Now, an equation for the maximum flow conditions can be written as
§ ·
- X
'2V ¨ ¸ (3.95)
© ¹
Similarly, an equation for the minimum flow conditions will be
§ ·
- X
'2V ¨ ¸ (3.96)
© ¹
By solving simultaneously for the two unknowns in Eqs. (3.95) and (3.96), it will yield
the control valve size
Kv = 21.3
and Pump head DPp = DPt + 2 = 139.2 + 2 = 141.2 psi
At design condition (50 gpm), the valve fraction open fdes will be given as
REVIEW QUESTIONS
3.1 Define a feedback control system. What are the advantages and disadvantages of
a feedback control system?
3.2 What are the basic hardware components of a feedback control loop? Identify the
hardware elements present in a feedback loop for the temperature control of a
stirred-tank heater.
3.3 Develop the block diagram of a generalized feedback control system with one
disturbance, incorporating in each block the appropriate transfer function and on
each stream the appropriate variable.
3.4 Define feedback control system for various process parameter controls, such as
(a) flow control, (b) pressure control, (c) temperature control, and (d) composition
control. Draw the appropriate block diagram for each of the processes.
3.5 What does tuning a feedback control system mean? What are different tuning
methods for feedback control system? Explain any one tuning method.
3.6 Discuss different methods for selecting a feedback controller. Draw and explain closed-
loop responses to unit-step change in a load for various types of feedback controllers.
3.7 Describe the four necessary components of a feedback control system.
3.8 What is the classification of a feedback control system? Discuss in detail about
single-input, single-output and multiple-input, multiple-output (multivariable)
feedback control system.
3.9 Discuss the application issues related to feedback control such that the equipment
and calculations provide accuracy and reliability and also overcome the
shortcomings of the basic PID control algorithm.
3.10 Derive feedback algorithm for P, I, D and PID control modes.
3.11 Derive closed-loop response of a heat exchanger shown in Figure 3.51.
Steam
F
W, Ti To
Process
fluid Condensate
Figure 3.51 Process schematic of the heat exchanger (Review Question 3.11).
3.12 Derive closed-loop characteristic equation and closed-loop transfer function for the
two-heated-tank process shown in Figure 3.52.
3.13 Find the ultimate gain and period of a closed loop system with a proportional
controller and open-loop transfer function
)/
U
È Ø
U
U É U Ù
Ê Ú
260 Feedback Control System
T0 T1 T2
T2
Thermocouple
Thermocouple
Steam
Steam
Temperature
Temperature
transmitter
II/P
/P Temperature TTTT transmitter
Temperature
Converter
Converter controller
controller
PV
PV
CO
CO
TC
TC
SP
Figure 3.52 Two-heated-tank process (Review
SP Question 3.12).
3.14 The system of Review Question 3.12 is modified by using the cascade control
system shown in Figure 3.53 below.
M Y1 Y2
11 11 11
0.5ss 11
0.5 s s 11 55ss 11
--1
1 -1
-1
Y 2set
K
Kss 6S KK
Ms 6S
(a) Find the value of gain (Ks) in the proportional controller that gives a 0.707
damping coefficient for the closed-loop slave loop.
(b) Using this value of Ks in the slave loop, find the maximum closed-loop-stable
value of the master controller gain (KM). Compare this with the ultimate gain
found without cascade control in Review Question 3.12. Also compare ultimate
periods.
Review Questions 261
3.15 Find the ultimate gain of the closed-loop three-CSTR system with a PI controller:
(a) For TI = 3.03 and (b) For TI = 4.5.
3.16 The liquid level h(t) in a tank is held by a PI controller that changes the flow-rate
F(t) out of the tank. The flow-rate F0(t) into the tank and the level set-point h(t)set are
disturbances. The vertical cylinder tank is 10 ft2 in a cross-sectional area. The
transfer function of the feedback controller plus the control valve is
(
U È Ø HV OKP
)%
U - E É
'
U Ê 6+ U ÙÚ HV
(a) Write the equations describing the open-loop system.
(b) Write the equations describing the closed-loop system.
(c) Derive the open-loop transfer functions of the system:
*
U *
U
)/
U and ).
U
(
U (
U
*
U *
U
and
*
U
UGV
(
U
(e) What value of gain Kc gives a closed-loop system with a damping coefficient
0.707? What is the closed-loop time constant at this gain?
(f) What gain gives critical damping? What is the time constant with this gain?
3.17 Find the ultimate gain and period of the closed-loop three-CSTR system with a PID
controller tuned at TI = TD = 1.
3.18 Repeat Review Question 3.16(b) and (d) using proportional feedback controller. Will
there be a steady-state error in the closed-loop system for (a) a step-change in
set-point hset or (b) a step-change in feed rate F0?
3.19 The liquid level in a tank is controlled by manipulating the flow-out of the tank,
using a PI controller. The outflow rate is a function of only the valve position. The
valve has linear installed characteristics and passes 20 ft3/min when wide open.
The tank is vertical and cylindrical with a cross-sectional area of 25 ft2 and a 2-ft
level transmitter span.
(a) Derive the relationship between the feedback controller gain (Kc) and the reset
time (TI) that gives a critically damped closed-loop system.
(b) For a critically damped system with TI = 5 minutes, calculate the closed-loop
time constant.
262 Feedback Control System
3.20 A two-tank system with recycle is shown in Figure 3.54. Liquid levels are held by
proportional controllers:
F1 = K1h1 and F2 = K2h2
Flow into the system F0 and recycle flow FR can be varied by the operator.
(a) Derive the four closed-loop transfer functions relating the two levels and the
two load disturbances:
È * Ø È * Ø È * Ø È * Ø
ÉÊ ( ÙÚ ÉÊ ( ÙÚ ÉÊ ( ÙÚ ÉÊ ( ÙÚ
U 4
U
U 4
U
(b) Does the steady-state level in the second tank vary with the recycle flow-rate
FR? Use the final value theorem of Laplace transforms.
F0 FR
LC
h1 LC
LC
F1 H2
F2
SELF-CHECK QUIZ
A State True/False:
3.1 Controlled system is the plant or process in which influence on the controlled
variable takes place.
3.2 The rise time is the interval that elapses between the step change of command
(desired value) and the initial entry of the controlled variable into the tolerance
band surrounding the new required steady-state value of the controlled variable.
Self-Check Quiz 263
3.3 When the power necessary to alter the controlled output is supplied primarily from
sources other than the command input, the feedback control system is called
passive.
3.4 When the control objective is to maintain the controlled output at its constant
equilibrium position in spite of disturbances, the system is called a regulatory
feedback control.
3.5 A servomechanism is a tracking control system where the controlled output is a
mechanical position, velocity or acceleration.
3.6 The decay ratio is the ratio of neighbouring peaks in an undamped controlled-
variable response.
3.7 Final control elements are the mechanism which alters the value of the controlled
variable in response to the output signal from the automatic control device.
3.8 The final control elements receive the output of a controller (actuating signal) and
adjust accordingly the value of the manipulated variable.
3.1 For developing first feedback control system concept was credited to
(a) B.C., Ktesibios of Alexandria
(b) Andrew Meikle of Netherland
(c) James Watt
(d) Maxwell
3.2 The accuracy of a feedback control system is given by
(a) Maximum persisting deviation of the controlled variable from the readable
value set on the command control in the presence of the most unfavourable
combination of disturbances
(b) Maximum persisting deviation of the controlled variable from the value initially
achieved in response to the command (desired value) setting, again in the
presence of the most unfavourable combination of disturbances
(c) Both (a) and (b)
(d) None of these
3.3 The steady-state stability of a feedback control system is given by the
(a) Maximum persisting deviation of the controlled variable from the readable
value set on the command control in the presence of the most unfavourable
combination of disturbances
(b) Maximum persisting deviation of the controlled variable from the value initially
achieved in response to the command (desired value) setting, again in the
presence of the most unfavourable combination of disturbances
(c) Both (a) and (b)
(d) None of these
264 Feedback Control System
3.1 Controller system is a part that influences the __________ variable in the desired
direction.
3.2 The accuracy and the steady-state stability may be expressed in terms of the
persisting deviation given as percent of the rated value of the ___________
variable.
Self-Check Quiz 265
3.3 The dynamic behaviour of a feedback control system is measured by its response
to a ___________ of the command (desired value) or a step change in disturbance.
3.4 When the power necessary to alter the controlled output is supplied primarily from
sources other than the command input, the feedback control system is called
______________.
3.5 When the control objective is to maintain the controlled output at its constant
equilibrium position in spite of disturbances, the system is called a
_____________________.
3.6 A servomechanism is a _______________ where the controlled output is a
mechanical position, velocity or acceleration.
3.7 The guidance system of an ________________ is an example of a servomechanism
type feedback control system since its objective is to track and home in on an
enemy aircraft that may be performing evasive maneuvers.
3.8 Integral error measure indicates the cumulative deviation of the ______________
from its set-point during the_____________.
3.9 Period of oscillation depends on the process dynamics and is an important
characteristic of the_________________.
3.10 Control valve is a device that adjusts the value of the _________ variable.
Part II
4.1 INTRODUCTION
Most chemical processes are modelled by equations that are quite complex and
nonlinear. Very few analytical mathematical techniques are currently available for
modelling nonlinear system. Most of the techniques are applicable to linear system only.
However, nonlinear equations describing most systems can be linearized around some
steady-state operating conditions. The resulting linear equations adequately describe the
dynamic response of the system in some region around the steady-state conditions. The
size of the region over which the linear model is valid varies with the degree of
nonlinearity of the process and the magnitude of the disturbances. In many processes
the linear model can be successfully used to study dynamics and to design controllers.
Therefore, linear techniques are of great importance, particularly for continuous
processes.
The presence of measuring devices, controllers and final control elements changes
the dynamic characteristics of an uncontrolled process. The non-oscillatory first-order
processes may acquire oscillatory behaviour with PI control. Oscillatory second-order
processes may become unstable with a PI controller.
Thus, while designing a feedback control system, its stability characteristics are
main concern. In this chapter, we study and analyze the response and performance of
a control system and its stability criterion in terms of the complex frequency variable
s and the location of the poles and zeros on the s-plane. Frequency response method,
an alternative approach to the analysis and design of a control system, has also been
discussed.
system is considered to be stable if for every bounded input it produces abounded output,
regardless of its initial state. A system exhibiting unbounded response to abounded input
is unstable. Bounded is an input that always remains between upper and a lower limit
(e.g., sinusoidal or step). Unbounded output exists only in theory and not in practice
because all physical quantities are limited. Thus, the term unbounded means very large.
A bounded input function is a function of time that always falls within certain bounds
during the course of time. For example, the step function and sinusoidal function are
bounded inputs. The function f(t) = t is unbounded. Figure 4.1 shows responses of stable
and unstable systems.
y y
The stability of any system is determined by the location of the roots of its
characteristic equation or the poles of its transfer functions. The characteristic equation
of a continuous system is a polynomial in the complex variable s. The system to be
stable, the poles of the transfer function must lie in the left half of the s-plan (LHP).
In other words, if all the roots of this polynomial are in the left hand of the s-plane, the
system is stable. This stability requirement applies to any system, open-loop or closed-
loop. The stability of an open-loop process depends on the location of the poles of its
open-loop transfer function. The stability of a closed-loop process depends on the location
of the poles of its closed-loop transfer function. These closed-loop poles will naturally be
different from the open-loop poles because of the introduction of the feedback loop with
the controller. For a continuous closed-loop system, all the roots of 1 + GM(s) GP(s) must
lie in the left hand of the s-plane.
Thus, criteria for open-loop and closed-loop stability are different. Most systems are
open-loop stabled, but can be either closed-loop stable or unstable, depending on the
values of the controller parameters. This type of stable/not stable characterization is
referred to as absolute stability. The degree of stability of the closed-loop stable system
is further referred to as relative stability.
There are many physical systems that are inherently open-loop unstable, and some
systems are even designed to be open-loop unstable. Most modern fighter aircraft are
open-loop unstable by design, and without active feedback control assisting the pilot they
cannot fly. Active control is introduced by engineers to stabilize the unstable plant, e.g.
Stability Analysis 271
+
+ E 1
M Kp G= Y
+ ( U1s + 1)( U2s + 1)
+
1
H=
( U1s + 1)
To obtain the transient response Y(t), it is necessary to find the inverse of Eq. (4.3).
This requires obtaining the roots of the denominator of Eq. (4.2), which is third-order.
These roots are not easier to find as applicable for the second-order system by use of
the quadratic formula. However, they can be obtained using algebraic methods.
The roots of the denominator depend on the particular values of the time constants
and Kp. These roots determine the nature of the transient response. Consider the step
response for U , U and U for several values of Kp. Figure 4.3 shows the
results of inversion of Eq. (4.3). It is seen from these curves that, as Kp increases, the
Y( t)
K p = 12
Kp = 9
Unit step
Kp = 6
1
Kp = 3
0
–1
0 2 4 6 8 10
Figure 4.3 Control system response of Figure 4.2 for unit-step change in set-point.
system response becomes more oscillatory. Beyond a certain value of Kp, the successive
amplitudes of the response grow rather than decay. This type of response is called
unstable. It can be seen from Figure 4.3 that there are values of Kp that give unstable
responses such as that corresponding to Kp = 12. If the order of Eq. (4.2) is higher than
three, the calculations necessary to obtain Figure 4.3 will become more difficult.
Let us consider a single-loop feedback control system as shown in Figure 4.4. The block
diagram of the control system may be described by
ª )/
U)2
U º UGV ª ).
U º
;
U « » ;
U « » .
U (4.4)
¬ )/
U)2
U ¼ ¬ )/
U)2
U ¼
Stability Analysis 273
L GL
M
GM 4 Y
–1
set
GP 4 Y
It should be kept in mind that the GM(s) transfer function in Eq. (4.4) is a combina-
tion of the process, transmitter and valve transfer function.
In order to simplify the nomenclature, let G(s) = GM(s)GP(s). Now to determine under
what conditions the system represented by Eq. (4.4) is stable, it is necessary to test the
response to a bounded input. Suppose a unit-step change in set-point is applied. Then
ª )/
U)2
U º )/
U)2
U (
U
;
U « )U » U (4.5)
¬ ¼ U
U T
U T
U TP
)/
U)2
U )
U (4.6)
F(s) is a function that arises in the rearrangement to the right-hand form of Eq. (4.5).
Equation (4.6) is called the closed-loop characteristic equation for the control system of
Figure 4.4. For example, for the control system of Figure 4.2 the step response is
-R
ª )/
U)2
U º U U U U (4.7)
;
U « )U » U
¬ ¼ ª -R º
« U
¬ U U U U U U »¼
Eq. (4.7) may be rearranged to
- R U U
;
U (4.8)
U ¬ªU U U U U U U U U U U U U U U - R º¼
- R U U
U U U
or ;
U (4.9)
U U T U T U T
274 Response Analysis of Control System and Stability Criterion
ªU U U U U U U U U U U U U U U - R º (4.10)
¬ ¼
Evidently, for this case, the function F(s) in Eq. (4.5) is
U U U U U U
(
U (4.11)
U U U
where a0 is positive. If a0 is originally negative, both sides are multiplied by 1. In this
form, it is necessary that all the coefficients a0, a1, a2, a3
, an1, an be positive if all
the roots are to lie in the left half plan. If any coefficient is negative, the system is
definitely unstable, and the Routh test will not be applicable for stability. However, the
Routh test will tell the number of roots in the right half plane. If all the coefficients are
positive, the system may be stable or unstable. It is then necessary to apply the Routh
Array procedure and Routh test theorems to determine the stability.
Stability Analysis 275
CP U T U T U T ! U TP (4.13)
where ri is the ith root of the characteristic equation. Multiplying the factors together,
we find that
S
U CP UP CP T T ! TP UP CP T T T T ! UP
! U P ! CP T T T ! TP
P
CP T T T T T T (4.14)
sn an an2 an4
sn1 an1 an3 an5
sn2 bn1 bn3 bn5
sn3 cn1 cn3 cn5
. . . .
. . . .
. . . .
s0 hn1 .
276 Response Analysis of Control System and Stability Criterion
where
CP CP CP CP !
CP CP
DP
CP !
CP CP CP
!
CP CP
DP
!
CP CP CP
CP ! CP
DP ! DP
and EP
DP
and so on. The algorithm for calculating the entries in the array can be followed on a
determinant basis or by using the form of the equation for bn1. The Routh array for
n = 7 by arranging the coefficients of Eq. (4.12) is given in Table 4.1. For any other value
of n, the array is prepared in the same manner. In general, there are (n + 1) rows. For
n even, the first row has one more element than the second row. The element in the
remaining rows are found from the formulas
C C C C C C C C
D D
C C
D C C D D C C D
E E
D D
and so on.
Table 4.1 Routh Array
Row
1 a0 a2 a4 a6
2 a1 a3 a5 a7
3 b1 b2 b3
4 c1 c2 c3
5 d1 d2
6 e1 e2
7 f1
M M
n + 1 g1
The elements for the other rows are found from formulas that correspond to those
just given. The elements in any row are always derived from the elements of the two
preceding rows. During the computation of the Routh array, any row can be divided by
a positive constant without changing the results of the test.
Thus, the Routh-Hurwitz criterion states that the number of roots of q(s) with
positive real parts is equal to the number of changes in sign of the first column of the
Routh array.
Stability Analysis 277
Steady-state performance
The steady-state performance specification is usually steady-state error. The steady-
state performance depends on both the system (process and controller) and the type of
disturbance. As we have discussed in Section 2.8, the error signal in the Laplace domain
E(s) is defined as the difference between the set-point Yset(s) and the process output Y(s).
It is given as
E(s) = Yset(s) Y(s) (4.18)
Suming that there is a change in the set-point Yset(s), but no change in the load
disturbance [L(s) = 0] and substituting for Y(s) from Eq. (4.4), it gives
ª )/
U)2
U º UGV
'
U ; UGV
U « » ;
U (4.19)
¬ )/
U)2
U ¼
'
U
UGV
)/
U)2
U (4.20)
;
U
278 Response Analysis of Control System and Stability Criterion
The steady-state value of the error can now be determined using final-value theorem (as
discussed in sec 2.8.1.5) of Laplace transform
' { NKO '
V NKO > U'
U@ (4.21)
V of U o
Now, let us consider a step and a ramp set-point inputs.
ª º ª º
' NKO > U'
U@ NKO « U » NKO « »
U o U o « )
U)
U U »
¬ O R ¼ U o )
U)
U
¬ / 2 ¼
If the steady-state error is to go to zero, the term must go to zero as
)/
U)2
U
s goes to zero. This means that the term GM(s)GP(s) must go to infinity as s goes to zero.
Thus, GM(s)GP(s) must contain a term, which is an integrator. If the process GM(s)
U
does not contain integration, we must put it into the controller GP(s). So we add reset
or integral action to eliminate steady-state error for step input changes in set-point.
If we use a proportional controller, the steady-state error is
ª º
' NKO « » (4.22)
U o )
U)
U \ \ \/
¬ / 2 ¼ -R
R R R0
where
zi = zeros of GM(s)
pi = poles of GM(s)
Thus the steady-state error is reduced by increasing zi = zeros of Kp, the controller gain.
Ramp input
; UGV
U
U
ª º ª º
' NKO « U » NKO « »
Uo Uo U= )
U)
U?
(4.23)
¬ )/
U)2
U U ¼ ¬ / 2 ¼
If the steady-state error is to zero, the term must go to zero as s goes
U= )/
U)2
U?
to zero. This requires that GM(s)GP(s) must contain a term. Double integration is
U
Stability Analysis 279
needed to drive the steady-state error to zero for a ramp input. This is done to make
the output track the changing set-point.
Dynamic performance
The dynamic performance of a system may be obtained by observing the location of the
roots of the system characteristic equation in the s-plane. The time-domain
specifications of time constants and damping coefficients for a closed-loop system can be
used directly in the Laplace domain. Dynamic specification of a system can be
determined using the following rules:
(a) If all the roots lie in the left half plane (LHP), the system is stable.
(b) If all the roots lie on the negative real axis, we know the system is over
damped or critically damped (all real roots).
(c) The farther out on the negative axis the roots lie, the faster the dynamics of
the system will be (the smaller the time constants).
(d) The roots that lie close to the imaginary axis will dominate the dynamic
response since the ones farther out will die out quickly.
(e) The farther any complex conjugate roots are from the real axis, the more
under-damped the system will be.
There is a quantitative relationship between the location of roots in the s-plane and
the damping coefficient. Let us assume that there is a second-order system or in case
s1
X
1[2
U
S-plane
[ = cos G
G
B
[
U
1 [2
s2 U
Figure 4.5 Dominant second-order root in the s-plane.
[ [
U r (4.24)
U U
280 Response Analysis of Control System and Stability Criterion
[ [
U (4.25)
U U
[ [
U (4.26)
U U
where z and t are the time constant and damping coefficient of the system respectively.
If the system is open-loop, these are the open-loop time constant and open-loop damping
coefficient. If the system is closed-loop, these are the closed-loop time constant and
closed-loop damping coefficient. The hypotenuse of the triangle shown in Figure 4.5 is
the distance from the origin out to the root s1.
§ [ · §[ ·
¨¨ ¸¸ ¨ ¸ (4.27)
© U ¹ ©U ¹ U
The angle f can be defined from the hypotenuse and the adjacent side of the triangle.
[
EQU G U [
(4.28)
U
Thus, the location of a complex root can be converted directly to a damping coefficient
and a time constant. The damping coefficient is equal to the cosine of the angle between
the negative real axis and a radial line from the origin to the root. The time constant
is equal to the reciprocal of the radial distance from the origin to the root. It can be
noticed that the lines of constant damping coefficient are radial lines in the s-plane.
Lines of constant time constant are circles.
Im( s)
s-plane
Kc md Kc m0
Ie( s)
– U1
0
For a first-order system, the closed-loop root is always real, so the system can
never be under-damped or oscillatory. The closed-loop damping coefficient of this system
is always greater than 1. The larger the value of controller gain, the smaller is the
closed-loop time constant because the root moves farther away from the origin (as we
know, the time constant is the reciprocal of the distance from the root to the origin).
If we want a closed-loop time constant of U (i.e., the closed-loop system is 10 times
faster than the open-loop system), we have to set Kc equal to . Eq. (4.32) shows that
-R
at this value of gain the closed-loop root is equal to . This first-order system can
U
never be closed-loop unstable because the root always lies in the left half plane (LHP).
No real system is only first-order. There are always small lags in the process, in the
control valve, or in the instrumentation that make all real systems of order higher than
first.
Second-order process
Now, let us consider a second-order process with proportional controller for which the
transfer function is given as
)/
U (4.33)
U U
The closed-loop characteristic equation is given by
)/
U)2
U -
U U E
5s2 + 6s + 1 + Kc = 0
The quadratic formula gives the two closed-loop roots:
r - E
U (4.34)
U r -E (4.35)
Figure 4.7(a) shows the locations of these roots for various values of Kc. When Kc
is zero, the closed-loop roots are at U
and s = 1. It may be noticed that these
values of s are the poles of the open-loop transfer function. The root locus plot always
starts at the poles of the open-loop transfer function. For Kc between zero and , the
two roots are real and lie on the negative real axis. The closed-loop system is critically
Stability Analysis 283
Kc d
X
s-plane
Kc
Kc 0 Kc 0
B
1
–1 5
Kc
Kc d
Kc
(a) G M ( s) GP ( s)
s 15s 1
X
K c 30
s-plane
Poles
K c 10
+1
K c 54.7 Zeros
Kc 5
K c 50 K c 1.25
Kc 3
–4 –3 –2 –1 +1
B
K c 55 K c 55 Kc d Kc 0
Kc 3
K c 50
Kc 5
K c 10 –1
K c 30
K c 0.5s 1
(b) G M ( s) GP ( s)
s 15s 1
284 Response Analysis of Control System and Stability Criterion
X
+2
s-plane K c K u 20
K c 10
Kc 5
+1
Kc 1
K c 20 Kc 5 Kc 0 Kc 0 Kc 0
B
–3 –2 –1 – 0.2
Kc 1
–1
Kc 5
K c 10
K c K u 20
Kc
(c) G M ( s) GP ( s)
s 15s 10.5s 1
damped (the closed-loop damping coefficient is 1) at - E since the roots are equal.
For values of gain greater than , the roots will be complex, and is given by
U rK -E (4.36)
As the gain goes infinity, the real parts of both roots are constant at and the
imaginary parts go to ± ¥ (infinity). Thus, the system becomes increasingly under-
damped. The closed-loop damping coefficient goes to zero as the gain becomes infinite.
However, this second-order system never becomes closed-loop unstable since the roots
are always in the left half of s-plane (LHP).
Suppose we want to design this system for a closed-loop damping coefficient of
0.707. From Eq. (4.28), the angle f can be calculated as f = cos (0.707) = 45°. Therefore,
we must find the value of gain on the root locus plot where it intersects a 45° line from
the origin. At the point of intersection the real and imaginary parts of the roots must
be equal. This occurs when - E . The closed-loop time constant U E of the system at
Stability Analysis 285
this value of gain can be calculated from the reciprocal of the radical distance from the
origin.
UE (4.37)
§· §·
¨ ¸ ¨ ¸
©¹ ©¹
Second-order process with a lead or a zero.
Let us add a lead or zero into the preceding example. The system transfer function is
given by
§ ·
-E ¨ U ¸
)/
U)2
U © ¹
U U
The closed-loop characteristic equation becomes
§ ·
-E ¨ U ¸
© ¹
)/
U)2
U (4.38)
U U
§ - ·
U ¨ E ¸ U - E (4.39)
© ¹
The roots are:
§ - ·
U ¨ E ¸ r (4.40)
© ¹ -E
- E
For low values of Kc, the term inside the square root will be positive, since +16 will
dominate; the two closed-loop roots are real and distinct. For very large values of gain,
the K c term will dominate and the roots will again be real. For intermediate values
2
of Kc, the term inside the square root will be negative and the roots will be complex.
The range of Kc values that give complex roots can be found from the roots of
- E (4.41)
- E
- E (4.42)
- E (4.43)
where
Kc1 = smaller value of Kc where the square-root term is zero
Kc2 = larger value of Kc where the square-root term is zero
Figure 4.7(b) shows the root locus plot. It may be noted that the effect of adding
a zero or a lead is to pull the root locus toward a more stable region of the s-plane. The
root locus starts at the poles of the open-loop transfer function. As the gain goes infinity,
the two paths of the root locus go to ¥ (infinity) and to the zero of the transfer function
286 Response Analysis of Control System and Stability Criterion
at s = 2. The system is closed-loop stable for all values of gain. The fastest-responding
system would be obtained with Kc = Kc1, where the two roots are equal and real.
Rule 3 The root loci end (Kc = ¥) at open-loop zeros of GP(s)GP(s) = 0. The
termination of the loci will occur at the zeros at infinity along
asymptotes.
Rule 4 The number of loci is equal to the order of the system, i.e., the number
of poles of GP(s)GP(s).
Rule 5 The complex parts of the curves always appear as complex conjugates.
Rule 6 The angle of the asymptotes of the loci (as s ® ¥) is equal to
q
r
0/
where
N = number of poles of GP(s)GP(s)
M = number of zeros of GP(s)GP(s)
Rule 6 comes from the fact that at a point on the root-locus plot the complex
number s must satisfy the equation
1 + GP(s)GP(s) = 0
GP(s)GP(s) = 1 + i0 (4.47)
Therefore, the argument of GP(s)GP(s)on a root-locus must always be
CTI )/
U)2
U rQ VCP (4.48)
Now GP(s)GP(s) is a ratio of polynomials, Mth order in the numerator and Nth order in
the denominator.
D/ U / D/ U / ! D U D
)/
U)2
U
C0 U 0 C0 U 0 ! C U C
(4.49)
On the asymptotes, s gets very big, so only the sN and sM terms remain significant.
D/
D/ U / C0
NKO >)/
U)2
U@ (4.50)
Uof C0 U 0 U0/
D/
C0
Uof > /
NKO )
U)2
U@ KR 0 /
T 0 / G
NKO >)/
U)2
U@ 0 / R
Uof
Eq. (4.48) must still be satisfied on the asymptote, and therefore,
0 / R rQ
288 Response Analysis of Control System and Stability Criterion
Im s = X Im z = y
s-plane +i z-plane
Stable
Re s B z eTss Re z x
–1 +1
B< 0
B> 0 Unit circle Unstable
–i
Stable
Unstable
Figure 4.8 Stability region in the s-plane and in the z-plane.
Figure 4.8 shows the s-plane. Let the real part of s be B and the imaginary part
of s be w.
s = a + iw (4.52)
The stability region in the s-plane is where a, the real part of s, is negative. Substituting
Eq. (4.52) into Eq. (4.51) gives
Frequency Response 289
\ G
6U B KX
GB6 GKX6
U U
(4.53)
The absolute value of z, |z| is GB 6U . When a is negative, |z| is less than 1. When
aÿ is positive, |z| is greater than 1. Therefore, the left hand of the s-plane maps into
the inside of the unit circle in the z-plane as shown in Figure 4.8. For finding stability
of an open-loop system, let us consider the open-loop transfer function given as
\ \ \ \ \ \/
*)/
\ (4.54)
\ R \ R \ R0
The stability of this open-loop system depends on the values of the poles of the
open-loop transfer function. If all the pi lie inside the unit circle, the system is open-loop
stable.
The equation describing the closed-loop digital control system is
).
\ .
\ *
\)/
\ &
\; UGV
\
;
\ (4.55)
*
\)/
\ &
\
The closed-loop stability of this system depends on the location of the roots of the
characteristic equation:
*
\)/
\ &
\ (4.56)
If all the roots lie inside the unit circle of Figure 4.8, the system is closed-loop
stable.
frequency of the damped oscillations at reasonable controller gains is usually 0.7 to 0.9
times the critical frequency, the critical frequency is also important as a measure of the
speed of response.
Ti
Process fluid flow X
T
X
us ( t)
A
X
Radians per time
2Q
T X=
T
Consider the well-insulated, perfectly mixed tank heat exchanger with constant
flow in and out (level control) as shown in Figure 4.9. Assume that the inlet temperature
varies in a sinusoidal fashion between a minimum and a maximum values as shown in
Figure 4.10.
The system of Figure 4.9 is modelled by a first-order transfer function. Suppose
that the input to the process is a sine wave us(t) of amplitude A and frequency Z as
shown in Figure 4.10. It is given by
WU
V # UKP
X V (4.57)
The period of one complete cycle of sine wave, as shown in Figure 4.10, is T units of
time. The frequency is expressed in variety of units. For example, the unit of frequency
is hertz (cycles per second) in electrical engineering and is given as
X (4.58)
6
Frequency Response 291
However, radians per time (radians/time) is the common unit of frequency used for
frequency response analysis in process control (chemical engineering) and is given as
Q
X (4.59)
6
or the period T (pick-to-pick time) is given as
Q
6 (4.60)
X
In a linear system, if the input is a sine wave with frequency Z , the output is also
a sine wave with the same frequency. The output has, however, a different amplitude
and lags (falls behind) or leads (rises ahead of) the input. In our example of heat
exchanger of Figure 4.9, at low frequencies the inlet temperature changes on a much
slower timescale than the characteristic time of the process, so the outlet temperature
is virtually identical to the inlet temperature. At high frequencies, the inlet temperature
changes so rapidly that the thermal mass of the process does not allow the outlet
temperature to change significantly. This type of system is known as a low-pass filter,
Input us ( t) Output ys ( t)
B
A
X
Radians per time
%T
(a) Output lags the input
Input us ( t)
Output ys ( t)
B
A
X
Radians per time
%T
(b) Output leads the input
Figure 4.11 Sinusoidal input-output.
that is, low-frequency inlet forcing is directly felt at the output. High input frequencies
are effectively filtered by the process and not felt at the output.
292 Response Analysis of Control System and Stability Criterion
§ '6 ·
G Q ¨ ¸ radians (4.61)
© 6 ¹
§ '6 ·
or G Q ¨ ¸ degree (4.62)
© 6 ¹
where
T = period
DT= lag between an input peak and an output peak
If the output leads the input, the phase angle f is positive, as shown in Figure 4.10(b).
The second measure is the amplitude ratio (AR), also called magnitude ratio (MR),
which is defined as the ratio of the maximum amplitude of the output sine wave B
over the maximum amplitude of the input sine wave A. The amplitude ratio (AR) may
be given as
$
#4 (4.63)
#
For a given process, both phase angle f and amplitude ratio AR will change if frequency
wÿ is changed. Different processes have different dependence of MR and fÿ on w. Since
each process is unique, the frequency-response curves are like finger-prints. The order
of system and the values of parameters (such as time constants, steady-state gain, and
damping coefficient) can be known by looking at the curves of AR and f.
The advantage of using the frequency response to analyze or design control
systems is that the system response is obtained easily from the response of the
individual elements, no matter how many elements are included. In contrast,
calculations of the transient response are quite tedious with only three components in
the system and are too difficult to be worthwhile for four or more components.
-R
)
U (4.64)
U RU
Now, the output can be found using Laplace transform method and substitution of
s = jw method.
Frequency Response 293
- R #XU R - R #XU R
V
-R#
UKP X V EQU X V
UR
[
V G
U X
R
UX
R
U X
R
V
U
At V o f and G R o , the exponential terms decay with time, so the first term can
be neglected at long times. Thus, after a long time, the response of a first-order process
to a sinusoidal input is given by
-R# - R #XU R
[UU
V
UKP X V
EQU X V (4.67)
U RX U RX
-R#
[UU
V UKP X V G (4.68a)
U RX
294 Response Analysis of Control System and Stability Criterion
where
G VCP XU R (4.68b)
The ss subscript is used to indicate the long-term behaviour of the output, after the
initial transients have decayed. From Eqs. (4.68) and (4.69), following observations can
be made:
(a) The output approaches zero at high frequencies.
(b) The ultimate response (steady-state) of a first-order process to a sinusoidal
input is also a sinusoidal wave with the same frequency w.
(c) The amplitude ratio AR is a function of the frequency:
-R#
[ U R X -R
#4 (4.69)
W #
U R X
(d) The output wave lags behind (phase lag) the input wave by an angle|f|, which
is also a function of the frequency w as can be seen from Eq. (4.68).
Substitution of s = jw method
Substitution of s = jw method is an easier way to generate frequency response plots.
Substituting s = jw in the transfer function of Eq. (4.64) results
-R -R LXU R - R LXU R - R - R LXU R - R
) LX
LXU R LXU R LXU R U R L
X U RX
or
-R - RXU R
) LX
L (4.70)
U R X U R X
Notice that the result G(jw) is a complex number (real and imaginary portion) that
changes as a function of frequency. Since the magnitude of a complex number Re + jIm,
can be represented by 4G +O then the amplitude (or magnitude) of the transfer
function is
#4 )
LX
- R U RX -R
(4.71a)
U RX
U R X
§ +O ·
Also, the phase angle of a complex number can be represented by VCP ¨ ¸ , so
© 4G ¹
G VCP XU R (4.71b)
It can be noticed that these equations are exactly the relationships shown with
Laplace transform method of Eqs. (4.68) and (4.69). The last two relationships also
Frequency Response 295
indicate that the amplitude ration AR and the phase lag, for the ultimate response of
a first-order process are equal to the modulus and argument, respectively, of its transfer
function when s = jw.
Xmd
X
Re |G ( jX ) |
G
AR
Increasing
XX
XX
G ( jX )
XX
As shown in Figure 4.12, with continuously varied frequency from zero to infinity,
a curve is formed in the G-plane. The real and imaginary parts of G(jw) at a specific
value of the frequency wÿ define a point in this coordinate system. Thus, at point 3 in
Figure 4.12, the frequency has a value w3. The shape and location of the curve are
unique characteristics of the system. We observe the following from the curve:
· The distance at point 3 from the origin (0, 0) is the amplitude ratio AR at the
frequency w3 and given as
&KUVCPEG ª¬ 4G ^) LX `º¼ ª¬+O ^) LX `º¼ ) LX #4
· The phase angle f with the real axis is the phase shift at the frequency w3 and
given as
+O ^) LX `
G VCP CTI ) LX = Phase shift
4G ^) LX `
296 Response Analysis of Control System and Stability Criterion
Thus, as the frequency varies from 0 to ¥, we trace the whole length of the Nyquist
plot and we find the corresponding values for the amplitude ratio AR and phase shift.
ª § · º § · Q
CTI ) LX CTI VCP « ¨ ¸ U R » VCP ¨ ¸ U R q TCFKCPU
¬« © U R ¹ ¼» ©UR¹
Im( G ) Im( G )
U nit circle
Xmd
,p ,p G ( jX)
–1 Xmd 2 +1
X
Re( G )
XU 1
p G ( jX)
X
, p
Re ( G )
+1
2
( a) F irst- order lag ( b) F irst- order lead
Im( G )
+j
Im( G )
U nit circle
–1 +1 Kp
Re ( G )
X Q X X Re ( G )
2D Xmd
G ( jX) G ( jX )
X Q –j
D
(c) Dead-time (d) Dead-time with first-order lead
When w = 0, arg G = 0 and |G| = 1. As w goes to infinity, |G| becomes infinite and
arg G goes to +90° or p/2 radians. The Nyquist plot for a first-order lead system is shown
in Figure 4.13(b).
Therefore,
-R
) LX arg G(jw) = arc tan (wtp) Dw (4.75)
X U R
It can be noted that the magnitude is exactly the same as for the first-order lag alone.
Phase angle is decreased by the dead-time contribution. Figure 4.13(d) shows the
Nyquist plot, which is a spiral that wraps around the origin as it shrinks in magnitude.
Xmd
Re ( G )
G ( jX)
Xm
) LX
§ X · Q TCFKCPU
CTI ) LX CTE VCP ¨ Q (4.76)
X © ¸¹
-R - RU RX - R L
) LX
X U R LX
X X U R
-R § · Q
) LX
X X UR
CTI ) LX CTE VCP ¨ ¸
© XU R ¹
CTE VCP XU R (4.77)
The Nyquist curve for integrator with first-order lag process is shown in Figure 4.15.
It can be noted that the results given in Eq. (4.77) could have been derived by combining
the magnitudes and arguments of an integrator as shown in Eq. (4.76) and a first-order
lag system shown in Eq. (4.72).
Im( G )
–K pUp
X md
Re ( G )
G ( jX)
X m
Figure 4.15 Nyquist plot for an integrator and first-order lag process.
300 Response Analysis of Control System and Stability Criterion
) LX
-R
- R U RX L- R [U RX
U RX L [U RX
U RX [U RX
-R § [U RX ·
) LX CTI ) LX CTE VCP ¨ ¸
U RX [U RX © U RX ¹ (4.78)
Figure 4.16 shows the Nyquist plot for a second-order under-damped process.
Following observations are made:
· When frequency is zero (i.e., w = 0), |G| is equal to Kp (i.e., AR = 1) and arg
G is equal to zero (i.e., f = 0). Therefore, the Nyquist plot starts (w = 0) on the
positive real axis at Re[G] = Kp and at a distance from the origin (0, 0) equal to
1, as shown in Figure 4.16.
· It intersects the imaginary axis (arg G = p/2) when frequency is equal to the
reciprocal of the time constant (w = 1/tp). At this point, |G| = Kp/2z. Therefore,
the smaller the damping coefficient, the farther out on the negative imaginary
axis the curve will cross. This shape is unique to an under-damped process.
Im( G )
Unit circle
Xmd
Kp
–1 +1
Re( G )
[r X
[
Kp
[ X
1
Up
Figure 4.16 Nyquist plot for second-order system.
Frequency Response 301
–1 Xmd +1
X0
Re( G )
transfer functions can be handled much more easily using Bode plots. The two curves
show how magnitude ratio and phase angle (argument) vary with frequency. Phase angle
is usually plotted against the log of frequency, using semi-log graph paper as illustrated
in Figure 4.18. The magnitude ration is sometimes plotted against the log frequency on
a log-to-log plot. However, usually it is more convenient to convert magnitude to log
modulus, defined by the equation
–10
–90 –20
–30
–180
0.01 0.1 1.0 0.01 0.1 1.0
Frequency X, radians per time Frequency X , radians per time
2JCUGCPING
G CTI ) LX
CTE VCP XU R
VCP XU R (4.81)
Figure 4.19 shows a Bode plot curve for first-order lag system in which the first
part is a plot of Eq. (4.81). The true curve is shown as the solid line on the upper part,
as shown in Figure 4.19. The L-curve can be easily sketched by considering the low- and
high-frequency asymptotes, which is one of the most convenient features of Bode plots.
As (wtp) ® 0 , Eq. (4.81) shows that AR ® 1. This is indicated by the low-frequency
asymptote. As (wtp) ® ¥, Eq. (4.82) becomes asymptotic to
log AR = log (wtp)
which is a line of slope 1, passing through the point
wtp = 1 and AR = 1.
This line is indicated as the high-frequency asymptote. The high-frequency
asymptote intersects L = 0 line at the frequency wc = 1/tp. This line where the two
asymptotes intersect is known as corner frequency or breakpoint frequency. The log
modulus is flat (horizontal) out to this point and then begins to drop off. A lag
contributes negative phase angle.
10 20
L ow-frequency
asymptote
1 0
High-frequency
asymptote
–40
0.01 XUp
1/Up
Corner or break-
0
Phase angle, degree
point frequency
– 45
–90 XUp
1/Up
It may be noticed in Figure 4.19 that the deviation of the true AR curve from the
asymptotes is a maximum at the corner (or breakpoint) frequency. Using wc = 1/tp in
Eq. (4.80) gives
#4
as the true value, whereas the intersection of the asymptotes occurs at AR = 1. Since
this is the maximum deviation and is an error of less than 30 percent, for engineering
purposes, it is often sufficient to represent the curve entirely by the asymptotes.
Alternatively, the asymptotes and the value of 0.707 may be used to sketch the curve
if more accuracy is required. In the lower half of Figure 4.19, we have the phase curve
as given by Eq. (4.81). Since
G VCP
XU R VCP
XU R
it is evident that f approaches 0o at low frequencies and 90o at high frequencies. This
verifies the low- and high-frequency portions of the phase curve. At the corner
frequency, wc = 1/tp,
GE VCP
X EU R VCP
Q
There are asymptotic approximations available for the phase curve, but they are
not so accurate or as widely used as those for the AR. Instead, it is convenient to note
that the curve is symmetric about 45°.
. NQI X U R (4.84)
The Bode plot for a first-order lead system is shown in Figure 4.20. The high-
frequency asymptote has a slope of +20 dB/decade. The breakpoint frequency is 1/tp. The
phase angle goes from zero to +90o and is +45o at w = 1/tp. Thus, a lead contributes
positive phase angle.
)
U G &U ) LX G LX &
) LX CTI ) LX X &
Frequency Response 305
10 20
0 0
0.1 –20
–40
0.01 XUp
1/Up
90
45
0 XUp
1/Up
Figure 4.20 Bode diagram for first-order lead system.
. NQI ) LX NQI
(4.85)
0 0 0
–90
–180
XUp XU p
X Q
D
X Q
D
Figure 4.21 Bode diagram for dead-time system.
) LX
-R
- R U RX L- R [U RX
U RX L [U RX
U RX [U RX
-R
) LX
U RX [U RX
§ [U RX · § [U X ·
¨ R
2JCUG CPING CTI ) LX CTE VCP ¨ ¸
VCP ¸
© U RX ¹ ©
¨ U X ¸
R ¹
ª º
« »
#4 . NQI « » (4.87)
«
¬« U RX [U RX »
¼»
Figure 4.22 shows the Bode plot for a second-order under-damped process. If wtp is used
as the abscissa for the general Bode plot, z will be a parameter. In other words, there
is different curve for each value of z. The corner (or breakpoint) frequency is the
reciprocal of the time constant.
The high-frequency asymptote has a slope of 40 dB/decade.
§ · ª º
»
#4 ¨ ¸ NKO . NQI « NQI XU R
¨ U RX ¸ X of « U X »
© ¹ ¬« R ¼»
This asymptote (as shown in Figure 4.22) has slope 2 and intersects the line
AR = 1 at wtp = 1. As shown in Figure 4.22, the log modulus curve has unique shape.
With lower damping coefficient, the peak in the L curve increases. A damping coefficient
Frequency Response 307
10
[
5
[
[
2
A mplitude ratio AR
[
1
0.5
[1.0
[
0.2
0.1
0.05
A symptote
0.02
[
[
[
–45
[
Phase angle, degree
[
–90
–135
–180
0.1 0.2 0.5 1.0 2.0 5.0 10.0
0.1 1
Up Up
XUp
of about 0.4 gives a peak of about +2 dB. This property is used extensively in tuning
feedback controllers.
For zÿ ³ 1, the second-order system is equivalent to two first-order systems in
series. In fact, the AR for zÿ ³ 1 as well as zÿ < 1 attains slope of 2 and phase of 180o.
Therefore, AR is consistent. Also, for zÿ < 0.707, the AR curves attain maxima in the
vicinity of wtp = 1. This can be checked by differentiating the expression for the AR with
respect to wtp and setting the derivative to zero. This results into
XU R OCZ [ [ (4.88)
for the value of wtp at which the maximum AR occurs. The value of the maximum AR,
obtained by substituting (wtp )max into Eq. (4.87) is:
#4OCZ [
[ [
0.8
[ 0.6
0.4
0.2
0
2.0 4.0 6.0 8.0 10.0
( AR) max
Figure 4.23 shows a plot of the maximum AR against z. The frequency attained at
maximum AR is called the resonant frequency, and is obtained from Eq. (4.88),
XU R [ (4.89)
UR
It may be seen that AR values exceeding unity are attained by systems for which
zÿ< 0.707. This is in sharp contrast to the first-order system, for which the AR is always
less than unity.
Illustrative Examples 309
ILLUSTRATIVE EXAMPLES
Example 4.1 Find the roots of a first-order process with a zero-order hold and
proportional sampled-data controller
-R
)/
U
U QU
*
\)/
\ <«
ª G6U U -R º
» - R D \ D
« U U Q U »
¬ ¼
6U
where
D { G UQ
The open-loop characteristic equation is
Z b = 0
The root of the open-loop characteristic equation is b. Since b is less than 1, this
root lies inside the unit circle and the system is open-loop stable.
- E - R D
*
\)/
\ &
\
\D
since D(z) = Kc,
Solving for the closed-loop root gives
\ D - E - R D
There is a single root. It lies on the real axis in the z-plane, and its location depends
on the values of the feedback controller gain Kc. When the feedback controller gain is
zero (the open-loop system), the root lies at z = b. As Kc is increased, the closed-loop
root moves to the left hand along the real axis in the z-plane.
Example 4.2 Stabilize an unstable process with proportional (P) control. The block
diagram is shown in Figure 4.24. The process response may be
considered as
§ · § ·
;
U ¨ ¸ /
U ¨ ¸ .
U
© U ¹ © U ¹
Solution: Since the transfer function of this process possess a pole at s = 1 > 0, this
310 Response Analysis of Control System and Stability Criterion
L ( s)
5
Gd =
s– 1
+
set
Y ( s) + E ( s) M ( s) +
G c = Kc Gf = 1 10 Y( s)
Gp =
s– 1
–
Gm = 1
process is unstable. Figure 4.25 shows the response of the process with P control. Curve
A shows the response of the uncontrolled system to a unit step change in the load L
which verifies its unstable character. Let us now introduce a feedback control system
with proportional (P) control only. Assume that for measuring sensor and the final
control element
Gm = Gf = 1
t
Figure 4.25 Response of process with P control.
We know that the closed-loop response of the system is given by Eq. (4.4), which
for the present system becomes
- E
;
U ; UGV
U .
U
U - E U - E
Illustrative Examples 311
have negative common pole if - E ! .
Therefore, the original system can be stabilized with simple proportional control. Curve
B of Figure 4.25 shows the dynamic response of the controlled system to a unit step
change in the load for Kc = 1. By comparing it to the behaviour of the uncontrolled
system, the stabilizing effect of the controller can be realized.
Therefore, according to the criterion, the system is stable. For a unit step change in the
input, the response of the system will be as shown in Figure 4.26(a). Now, let us
introduce a PI controller in which the measuring element and the final control element
have the transfer function given as
Gm = Gf = 1
y y
t t
(a) Open-loop stable response (b) Destabilized response with PI control
Figure 4.26 System response for Example 4.3.
) R)E
;
U ; UGV
U )UGV ; UGV
U
) R)E
312 Response Analysis of Control System and Stability Criterion
U U - E U U
-E
) R)E U U UU U
)UGV
) R)E U U -E
-E U U - E U
U U UU U
Let Kc = 0 and U . Then the poles of Gset are determined by the roots of the
polynomial
U U U
and are found to be
p1 = 7.185 p2 = 2.59 + j(11.5) p3 = 2.59 j(11.5)
It can be noticed that p1 and p2 have positive real parts. Therefore, according to the
criterion, the closed-loop response is unstable. Figure 4.26(b) shows the response of the
system to a unit step change of the set-point. Now, this can be compared with the
response of the uncontrolled system for noticing the destabilizing effect of the PI
controller. For different values of Kc and t1, the response becomes stable. By lowering
the gain to Kc = 10 and increasing t1 = 0.5, we find that all the poles of Gset have
negative real parts. That is, the closed-loop system is stable.
Example 4.4 Examine the stability of two feedback loops of Figure 4.24 of Example 4.2.
)R Gf = 1, Gm = 1, Gc = Kc
U
Therefore, the corresponding characteristic equation is
) R)H )E)O u u - Eu
U
which has the root given as
p = 1 10Kc
and the system is stable if p < 0 (i.e., - E ! ).
Example 4.5 Examine the stability of two feedback loops of Example 4.3.
§ ·
)R)H )E)O
u u -E ¨ ¸ u
U U © U U ¹
Example 4.6 Determine the stability by the Routh criterion for the given characteristic
equation.
s4 + 3s2 + 5s2 + 4s + 2 = 0
Solution: In the given characteristics equation, all the coefficients are positive.
Therefore, the system may be stable. To test this stability, let us form the Routh array
shown in Table 4.2.
The elements in the array are found by applying the formulas presented in the
rules: for example, b1, which is the element in the first column, third row, is obtained
by
C C C C
D
C
D
Since there is no change in sign in the first column, there are no roots having positive
real parts, and the system is stable.
314 Response Analysis of Control System and Stability Criterion
Example 4.7 Determine the stability with the Routh-Hurwitz criterion for feedback
control system of Example 4.3. The characteristic equation is given as
-E
U U - E U
U+
Solution: The corresponding Routh array for the given problem can be formed as
shown in Table 4.3.
Example 4.8 (a) Determine the values of Kp using U U U for which the
control system in Figure 4.24 is stable.
(b) Determine the roots of the characteristic equation using Theorem 3
of Routh test for the value of Kp for which the system is on the
threshold of instability.
Illustrative Examples 315
6s2 + 66 = 0
U r L
Therefore, two of the roots on the imaginary axis are located at and .
The third root can be found by Eq. (4.90) in factored form:
(s s1)(s s2)(s s3) = 0 (4.91)
where s1, s2 and s3 are the roots. Introducing the two imaginary roots U L and
U L into Eq. (4.91) and multiplying out the terms give
s3 s3s2 + 11s 11s3 = 0
316 Response Analysis of Control System and Stability Criterion
Comparing this equation with Eq. (4.90), we see that s3 = 6. The roots of the
characteristic equation are therefore U L , U L and s3 = 6.
Example 4.9 Determine the stability of the system shown in Figure 4.24 using
U U U , Kp = 5 and U + for which a proportional-plus-
integral (PI) controller is used.
-R
U + U
U U U
§ ·§ ·§ ·
U + U¨ U ¸¨ U ¸ ¨ U ¸
© U ¹© U ¹© U ¹
Because there are two sign changes in the first column, we know from Theorem
2 of the Routh test that two roots have positive real parts. From the previous example
we know that for Kp = 5, the system is stable with proportional control. With integral
action present, however, the system is unstable for Kp = 5.
Example 4.10 Determine the frequency response of a second-order process with the
transfer function given as
-R
)
U
U U [U U
Solution: Substitute s = jw.
) LX
-R -R U X L[UX
U X L[UX U X L[UX U X L[UX
- R U X - R [UX
) LX
L
U X [UX U X [UX
Illustrative Examples 317
§ [UX ·
Phase angle G CTI ) LX VCP ¨ ¸
© U X ¹
which is a phase lag since G .
Example 4.11 Determine the frequency response of (a) P, (b) PI, (c) PD, and (d) PID
feedback controllers.
Solution: (a) The transfer function of proportional (P) feedback controller is given as
Gp(s) = Kp
Therefore,
Amplitude ration AR = Kp
Phase angle I 0
Therefore,
Amplitude ration #4 ) R LX -R
XU +
§ ·
Phase angle G CTI ) R LX VCP ¨ ¸
© XU + ¹
(c) The transfer function of proportional-derivative (PD) feedback controller is
given as
) R
U - R U & U
Therefore,
§ ·
Amplitude ration #4 ) R LX - R ¨ U &X ¸
© U +X ¹
§ ·
Phase angle G CTI ) R LX
VCP ¨ U &X ¸
© U +X ¹
It can be noticed that f can take negative (phase lag) or positive (phase lead) values
depending on the values of tD, tI and w.
Example 4.12 Determine the frequency response of a pure capacitive process with the
transfer function given as
-R
)
U
U
Solution: Substitute U LX .
-R - R LX -R
) LX L
LX LX LX X
Consequently, for the ultimate response,
-R
Amplitude ratio #4 ) LX
X
Phase angle VCP f
G
In this case, the ultimate sinusoidal response of the system lags behind the input
wave by 90o.
Example 4.13 Determine the frequency response of a pure dead-time process with the
transfer function given as
)
U G U F U
Solution: Substitute U LX
)
LX GU F U
Consequently, for the ultimate response,
Amplitude ratio #4 ) LX
Phase angle G CTI ) LX U FX
In this case, the sinusoidal response of the process lags behind the input wave,
since f < 0.
Example 4.14 Draw Bode plot for a system with the transfer function given as
)
U
U U
U U
Review Questions 319
G2
0
G1
– 90
– 180
Total G = G 1 + G2
1/U1 1/U2
0
– 20
– 10
L , dB
– 20
– 20
– 30
Total L = L 1 + L 2
– 40
1/U1 1/U2
Figure 4.27 Bode plots for G1 and G2.
REVIEW QUESTIONS
4.1 What do you mean by stability of a control system?
4.2 With neat sketch explain the general conditions for stability of a process control
system.
4.3 What are stability criterions for a control system?
4.4 The control system of Figure 4.28 has the following transfer functions:
For PI controller,
U
)
U
320 Response Analysis of Control System and Stability Criterion
+
+ +
G1 G2 C
R
–
B
H
Find the characteristics equation and its roots, and determine whether the system
is stable.
4.5 Write the characteristics equation and construct the Routh array for the
control system shown in Figure 4.29. Verify whether system is stable for (a) Kp = 9.5,
(b) Kp = 11, (c) Kp = 12.
+
R Kp 3 C
( s + 3)(0.5s + 1)
–
3
s+ 3
-
)
U U U
4.14 What are the characteristics of the response of a linear system with a transfer
function G(s) to a sustained sinusoidal input?
4.15 Define the frequency response of a control system.
4.16 Discuss the means that are used to represent the results of the frequency response
analysis for a dynamic system.
4.17 Explain the two important measures (a) amplitude ratio (AR), and (b) the phase
shift, that are obtained from time-domain plot of frequency response.
4.18 Derive amplitude ratio and phase shift for the frequency response of first-order
system using (a) Laplace transform method and (b) substitution of s = LX method.
4.19 With an example and neat sketches define and discuss Bode plots and Nyquist
plots.
4.20 For the following first-order system construct its Bode plot:
)
U UU
4.21 Construct the Bode diagram and Nyquist plot of a first-order system with dead time
with the following transfer function:
- R GU VF U
)
U
U RU
4.22 Construct the Bode diagrams for two control systems in series with transfer
functions
)
U )
U
U U
4.23 With proper derivation draw a Nyquist plot for the following systems:
(a) First-order lag system
(b) First-order lead system
(c) Dead-time system
(d) Dead-time and first-order lag system
(e) Integrator system
(f) Integrator and first-order system
(g) Second-order under-damped system
(h) Third-order system
322 Response Analysis of Control System and Stability Criterion
4.24 With proper derivation draw a Bode plot for the following systems:
(a) First-order lag system
(b) First-order lead system
(c) Dead-time system
(d) Second-order system
SELF-CHECK QUIZ
A State True/False:
4.1 A system is considered as unstable if, after it has been disturbed by an input
change, its output took off and did not return to the initial state of rest.
4.2 A system exhibiting unbounded response to abounded input is unstable.
4.3 Bounded is an input that always remains constant.
4.4 Unbounded output exists only in practice because all physical quantities are
limited.
4.5 The system to be stable, the poles of the transfer function must lie in the left half
of the s-plan (LHP).
4.6 The Routh-Hurwitz criterion for closed-loop systems requires calculation of the
actual values of the roots of the characteristic polynomial.
4.7 The Routh-Hurwitz criterion is a necessary and sufficient criterion for the stability
of linear systems.
4.8 The Routh-Hurwitz criterion states that the number of roots of q(s) with positive
real parts is equal to the number of changes in sign of the first column of the Routh
array.
4.9 Routh test theorem is applied to determine the roots.
4.10 As per Routh test theorem, the necessary and sufficient condition for all the roots
of the characteristic Eq. (4.2) to have negative real parts (stable system) is that all
elements of the first column of the Routh array (a0, aa, b1, c1, etc.) be negative.
4.11 The algebraic method for determining stability is limited in its usefulness whether
a system is stable. It does not give us any idea of the degree of stability or the roots
of the characteristic equation.
4.12 The steady-state performance specification is usually steady-state error.
4.13 The location of a complex root can be converted directly to a damping coefficient
and a time constant.
4.14 The damping coefficient is equal to the reciprocal of the radial distance from the
origin to the root.
4.15 The time constant is equal to the cosine of the angle between the negative real axis
and a radial line from the origin to the root.
Self-Check Quiz 323
4.16 A root-locus plot shows how the roots of the closed-loop characteristic equation
vary as the gain of the feedback controller changes from zero to infinity.
4.17 Frequency response may be defined as the steady-state behaviour of the system
when forced by a sinusoidal input.
4.9 As per Routh test theorem, the necessary and sufficient condition for all the roots
of the characteristic equation to have negative real parts (stable system) is that all
elements of the first column of the Routh array (a0, aa, b1, c1, etc.) be
________________.
4.10 The algebraic method for determining stability is limited in its usefulness whether
a system is _____. It does not give us any idea of the _________ of stability or the
______ of the characteristic equation.
4.11 The steady-state performance depends on both the system ________ and
______________ and the type of ______________.
4.12 The dynamic performance of a system may be obtained by observing the location
of the ________ of the system characteristic equation in the __________.
4.13 The location of a complex root can be converted directly to a ____________ and a
___________.
4.14 The damping coefficient is equal to the ________ of the angle between the negative
real axis and a __________ from the origin to the root.
4.15 The time constant is equal to the reciprocal of the _________ from the origin to
the root.
4.16 The root-locus method is a powerful tool for designing and analyzing __________
control systems.
4.17 In the complex plane, the root loci are the plots of the roots of ____________ as
the controller gain Kc is varied form zero to ______.
4.18 A Nyquist plot is generated by plotting the complex number G(jw) in a two-
dimensional diagram whose ordinate is the _________ of G(jw) and whose abscissa
is the ________ of G(jw).
5
5.1 INTRODUCTION
A major difficulty in control-system design is to reconcile the large-scale, fuzzy, real
problems with simple, well-defined problems that the control theory can handle. A good
design is often a compromise based on reasonable trade-offs between cost and
performance. Another aspect is that design is often arrived at by interaction between
customer and vendor. Many subjective factors such as pride, tradition and ambition
enter this interaction.
It is also useful to remember that control problems can be widely different in
nature. They can range from design of a simple loop in a given system to design of an
integrated control system for a complete process. The approach to design can also be
widely different for mass-produced systems, and one-of-a-kind systems.
The relationship between process design and control design is also important.
Control systems have traditionally been introduced into given processes to simplify or
improve their operation. It has, however, become clear that much can be gained by
considering process design and control design in one context. The availability of control
system always gives the designer an extra degree of freedom, which can frequently be
used to improve the performance or economy. Similarly, there are many situations
where difficult control problems arise because of improper process design.
In this chapter, various aspects of design of process control systems are discussed.
327
328 Design of Process Control Systems
y(t) Maximum
Unit-step overshoot
input
1.00
0.90
Steady-state
Delay time error bound
(td)
0.50
0.10
The peak time (tp) is the time required for the response to reach the first (or
maximum) peak.
The settling time (ts) is defined as the time required for the response to settle to
within a certain per cent of its final value. Typical percentage values used are 2% and
5%. The settling time is related to the largest time constant of the controlled system.
The delay time (td) is the time required for the response to reach half of its final
value for the very first time.
The per cent overshoot (PO) represents the amount that the response overshoots
its steady-state (or final) value at the peak time, expressed as a percentage of the steady-
state value. Figure 5.1 shows a typical unit-step response of a second-order system
)
U
X P
(5.1)
U
[X P U X P
È [Q Ø
21 GZR
É Ù (5.2)
Ê [ Ú
Q
VR
(5.3)
XP [
VU (5.4)
[X P
where the 2% criterion is used for the settling time (ts). If 5% is used, then ts can often
be approximated by ts = 3/(zwn). A precise formula for rise time (tr) and delay time (td)
in terms of damping ratio (wn) and undamped natural frequency cannot be found. But
useful approximations are
[ [
VF # (5.5)
XP
and
[ [
VT # (5.6)
XP
Note that the above expressions are only accurate for a second-order system. Many
systems are more complicated than the pure second-order system. Thus, when using
these expressions, the designer should be aware that they are only rough
approximations. The time-domain specifications are quite important because most
control systems must exhibit acceptable time responses. If the values of ts, td, tr, tp and
PO are specified, then the shape of the response curve is virtually determined. The
time-domain system specifications, such as PO, tr, z, etc., can be applied to discrete-time
systems with minor modifications.
330 Design of Process Control Systems
Steady-state accuracy
If the output of a system at a steady state does not exactly agree with the input, the
system is said to have steady-state error. This error is one measure of the accuracy of
the system. Since actual system inputs can frequently be considered combinations of
step, ramp and parabolic types of signals, control systems may be classified according to
their ability to follow step, ramp, parabolic inputs, etc. In general, the steady-state error
depends not only on the inputs but also on the type of control system. Open-loop transfer
function of a stable unity feedback control system can generally be expressed as
M
U \
U \ !
U \O
R
U R !
U RP
)
U
U 0
U
(5.7)
where G0(s) represents the open-loop transfer function of a stable unity feedback control
system, zi(¹0, i = 1, 2,
, m) are zeros and pj(¹0, j = 1, 2,
, n) and 0 (a pole at the
origin with multiplicity N) are poles, such that m < n + N.
The classification (or type) of feedback system refers to the order of the pole of
the open-loop transfer function G0(s) at s = 0 (i.e. the value of the exponent N of s in
G0(s)). In other words, the classification is based on the number of pure integrators in
G0(s). A system is called type 0, type 1, type 2,
, if N = 1, 2, 3,
, respectively.
Static error constants describe the ability of a system to reduce or eliminate
steady-state errors. Therefore, they can be used to specify the steady-state performance
of control systems. For a static unity feedback system with open-loop transfer function
G0(s), the position error constants Kp, velocity error constant Kv and acceleration error
constant Ka are defined as
Kp = NKO G0(s) (5.8)
U
Kv = NKO sG0(s) (5.9)
U
Ka = NKO s2G0(s) (5.10)
U
In terms of Kp, Kv and Ka, the systems steady-state error for the three commonly
used test signals i.e. a unit-step u(t) a ramp input tu(t) and a parabolic input
(1/2)t2u(t), can be expressed, respectively, as
G
(5.11)
- R
G
(5.12)
- XR
G
(5.13)
- CR
where the error e(t) is the difference between the input and output, and e(¥) = limt®¥ e(t).
Therefore, the value of the steady-state error decreases as the error constants
increase.
Specification of Control Systems 331
Just as damping ratio (z ), settling time (ts), delay time (td), peak time (tp), and
per cent overshoot (PO) are used as specifications for a control systems transient
response, so Kp, Kv and Ka, can be used as specifications for a control systems steady-
state errors.
Mp
1
wp w
0.707
wb
/R
for z £ 0.707 (5.15)
[ [
X D X P ËÍ
[ [ [ ÛÝ (5.16)
Sensitivity to parameters
During the design process, the design engineer may want to consider the extent to
which changes in system parameters affect the behaviour of a system. One of the main
advantages of a feedback system is that it can be used to make the response of a system
relatively independent of certain types of changes or inaccuracies in the plant model.
Ideally, the parameter changes due to heat, humidity, age, and other causes should not
appreciably affect a systems performance. The degree to which changes in system
parameters affect system transfer functions, and hence performance, is called
sensitivity. The greater the sensitivity, the worse is the effect of a parameter change.
Specification of Control Systems 333
Figure 5.3 shows the modelling of a typical closed-loop control system, where Gp(s)
represents the plant or process to be controlled, Gc(s) is the controller, and H(s) may
represent the feedback sensor dynamics. The model Gp(s) is usually an approximation
Control
Controller input Process
R(s) Gc(s) Gp(s) C(s)
Sensor
H(s)
to the actual plant dynamic behaviour, with parameters at nominal values and high-
frequency dynamics neglected. The parameter values in the model are often not
precisely known and may also vary widely with operating conditions. For the system
shown in Figure 5.3, the closed-loop transfer function Gcl(s) is
%
U )E
U) R
U
)EN
U (5.17)
4
U )E
U) R
U *
U
The sensitivity of the closed-loop performance to the elements in the forward path
reduces as the loop-gain increase. This is a major reason for using a feedback. In
summary, to reduce the sensitivity we need to increase the loop-gain. But, in general,
increasing the loop-gain degrades the stability margins. Hence, we usually have a
trade-off between low sensitivity and adequate stability margins.
)E
U) R
U )E
U) R
U
%
U 4
U &
U 0
U (5.18)
)E
U) R
U )E
U) R
U )E
U) R
U
The tracking error e(t), defined as e(t) = r(t) c(t), with its corresponding Laplace
transform E(s), becomes
334 Design of Process Control Systems
D(s)
+
+ E(s) U(s)
R(s) Gc(s) Gp(s) C(s)
+
+
+
N(s)
Figure 5.4 A unity feedback control system showing sources of noise and disturbance.
)E
U) R
U
'
U 4
U &
U 0
U (5.19)
)E
U) R
U )E
U) R
U )E
U) R
U
In terms of the sensitivity function (S) and closed-loop transfer function (G), the
output C(s) and tracking error E(s) becomes
C(s) = Gcl(s)R(s) + S(s)D(s) Gcl(s)N(s) (5.20)
and
E(s) = S(s)R(s) S(s)D(s) + Gcl(s)N(s) (5.21)
It is clear that S(s) must be kept as small as possible to minimize the effects of
disturbances. Tracking and disturbance rejection require small S, while noise
suppression requires small Gcl.
form of the plant uncertainty can be parametric, non-parametric or both. Typical sources
of uncertainty include unmodelled high-frequency dynamics, neglected nonlinearities,
plant parameter variations (due to change of environmental factors), etc.
Therefore, writing specifications for control system is not easy. The different
aspects of controller performance and robustness are inter-related and, in many cases,
competing. While a good controller can compensate for some deficiencies in the plant
(system to be controlled), the plant implies significant limits on controller performance.
For these reasons, it is important to have the plant designer, the person responsible for
the controller specifications, and the control designer work together from the beginning
to the end of any project with demanding control requirements.
necessary. Hence, selection of the state variables to be controlled has to be made. The
following guidelines may assist in the selection of controlled variables:
(a) Guideline 1: Always select state variables representing inventories which are not
self-regulatory. A self-regulatory state is one which, after a step change in one
of the input variables (manipulated or disturbance variable), will achieve a new
steady-state value. Conversely, a non-regulatory state variable does not achieve
a new steady-state value, but continues changing.
(b) Guideline 2: Select state variables which, although self-regulatory, may exceed
equipment or process constraints. The long-term integrity of the process
demands that certain equipment constraints are not violated. Obvious examples
include minimum and maximum flow rates in pumps and compressors, and
minimum and maximum allowable pressures and temperatures in vessels.
Process constraints include quality specifications and material handling limits.
Thus, the application of this guideline is again intuitive.
(c) Guideline 3: Select state variables which, although self-regulatory, may
seriously interact with other inventories. Areas of particular importance are
those where the process interactions are strong and where the number of
interactions are large. An example where many interactions are present is a
steam boiler supplying a large plant. The steam pressure is self-regulatory, but
the fluctuations would affect all steam users.
(d) Guideline 4: Select state variables that are direct measures of product quality or
that strongly affect it. One should remember that ultimately the process must
provide to the satisfaction of the market. Controlling variables that directly or
strongly relate to product quality, which ensures we achieve this goal.
When the number of controlled variables exceed the available number of
manipulated variables, only those selected under Guidelines 24 should be reconsidered.
Thus, the guidelines are really a hierarchy of importance and should be treated as
guidelines only, not as hard and fast rules.
(b) Develop the cause and effect matrix (CEM) for each of the unit operations in
the process, testing for independent accessibility. The CEM indicates the
structural relationship between manipulated variables and control objectives.
Control objectives can be arbitrarily complex, involving more than one
measurement variable.
(c) Apply the combining CEM to generate the CEM for the whole process from the
individual CEMs. The design should be performed on the whole plant, not
separately on individual unit operations. This produces an integrated design
which avoids inconsistent control objectives or multiple use of manipulated
variables.
(d) Achieve structural controllability of the CEM of the complete process, by
deleting manipulated variables or control objectives, as required.
(e) Generate a feasible pairing of manipulated variables and control objectives by
applying the recording algorithm described below to the whole plant CEM.
m1 m2
F
x
Figure 5.5 Mixing system.
Control System Design Using Models 339
stream of flow rate (F) and composition x. It is desired to control both F and x by
adjusting m1 and m2. A total mass balance yields:
F = m1 + m2 (5.22)
And a component balance yields:
O
Z (5.23)
O O
Taking Eq. (5.23) and re-arranging to solve for m2 and substituting the resultant
expression into Eq. (5.22) yield:
O
( (5.24)
Z
Now
(
M
(5.25)
O O
which can be derived from Eq. (5.22). Also,
(
M
(5.26)
O Z
Z
M
M Z (5.27)
M
Hence
Ë O O Û
4)# Ì
( Z ZÜ (5.28)
Ì Ü
ÌÍ Z Z ÜÝ
Here, k11 and k¢11 are open-loop gains and l11 is relative gain.
The pairing of the variables will be dependent on the choice of the desired
concentration of the outlet stream. If, for example, x is chosen to be 0.6, then
Ë Û
4)# Ì Ü (5.29)
Í Ý
and the most desirable pairings are F/m2 and x/m2. Note that if the desired operating
point were to change, say to x = 0.4, the most desirable pairings between the variables
would alter. While the method may suggest that the pairings should change, this would
rarely occur in practice.
The first limitation of RGA method is that it does not eliminate the interactions,
it merely tries to minimize the effect. The second major limitation is that it only relies
upon steady-state information. If dynamic interactions are more important than those
that occur at the steady state, then clearly RGA is not a good method for such systems.
340 Design of Process Control Systems
where
G(s) = Resulting open-loop transfer function
Gc(s) = Transfer function of the cascade series regulator
Gp(s) = Transfer function of the uncompensated process.
In parallel compensation, as shown in Figure 5.6(b), the regulator is situated in a
special feedback loop. The transfer function of this system is
)
U)
U
)
U (5.31)
)
U)E
U )
U)
U
where
G(s) = Resulting open-loop transfer function
G1(s) = Transfer function of the outside compensating feedback loop part of the
system
G2(s) = Transfer function of the inside compensating feedback loop part
Gc(s) = Transfer function of the parallel regulator.
R(s) C(s)
G1(s) G2(s)
+
U(s)
Gc(s)
U(s)
Gp(s)
R(s) C(s)
G1(s) G2(s)
+
Parallel compensation can always be selected in such a way as to obtain the same
result as in cascade compensation, and vice versa. In practice, the choice between these
compensating depends on the operational characteristics of the component functional
element blocks. Both of these methods use feedback. Their advantage is that sensitivity
to disturbances and to system parameter variations is reduced. Feedback is most
effective when the process dynamics need a high bandwidth. For this reason most
valuable design methods have been developed using these systems.
Figure 5.6(c) shows a special block diagram and is mostly used for measurable
disturbance compensation. The regulator is part of a supplementary feed-forward loop,
along which the signal to be compensated is introduced into the system. The
compensation method discussed in Figure 5.6(c) substantially differs from that of
Figures 5.6(a) and 5.6(b) in its working principle. The block diagram shows that this
compensation cannot affect the stability of the closed-loop system, but improves the
systems performance by reducing the dynamic error related to the disturbing action
that is to be compensated. The design of a feed-forward compensator is a calculation of
the inverse of a dynamic system.
The advantage of feed-forward, compared with feedback is that corrective action
may be taken before the disturbance has influenced the variables. Since, feed forward
is an open-loop compensation method, it requires a good process model. The use of feed
forward control is better applied to digital control, where this is widely permitted.
Analytical methods
Analytical methods for control design of SISO, LTI systems can be divided as:
· Transfer function method
· Time-domain method
· Rule-of-thumb (empirical) method.
A very suitable way to study linear control systems is based on the concept of the
transfer function. It is then natural to apply methods that allow to change this transfer
function. One of the most commonly used design methods is that of pole placement. The
main idea is to determine the parameters of a general form regulator, so that the closed-
loop system has the desired properties. This method can be applied to both continuous
and sampled linear systems. Pole placement is a general approach in SISO process
control design. Many other design methods, such as root-locus, dipole compensation,
Smith predictor, Dahlin control algorithm, dead-beat control, model algorithm control,
etc., can also be considered as particular forms of the pole placement design.
Usually the performance of a controlled system is expressed in the time-domain by
the specification of its step or impulse response. These types of specification (as
explained in Section 5.2), include rise-time, overshoot, settling-time, steady-state error,
etc. The performance criteria that satisfy time specifications, is usually expressed as a
time integral optimization criteria. This is a very modern design technique but serious
difficulties may appear in the analytical solution process.
The Smith predictor (to be explained in Chapter 6 in detail), is a special design
method for dead-time process control. This method can be useful in the compensation
of large dead-time processes, where the part of the process with the dead-time (time
delay) cannot be separated but the dead-time itself may be precisely determined. Using
this method, the regulator does not compensate the dead-time delay, but improves the
performance of the part of the process without dead-time.
Empirical (rule-of-thumb) methods have been developed for tuning standard
controllers. For a PID controller, it is used for primary parameter setting, namely the
proportional factor (or proportional band), the integral time constant, and the derivative
time constant. Difficulties in precise estimation of process parameters, the nonlinear
behaviour of real processes, the stochastic character of certain disturbances and the
difficulties in finding analytical solutions, all lead to limited usefulness of analytical or
graphicalanalytical methods. Another important reason for introducing empirical
methods of controller design is the fact that standard controller tuning and performance
checks must frequently be done locally in the working process. Even if analytical
controller design is available, final fine-tuning is necessary almost every time.
safety systems do not substitute for the regulatory controllers and, therefore, do not
allow continued operation when these controllers fail. Some of the design considerations
are discussed in details that are necessary for a reliable process control system.
Reliability
Reliability is a statistical parameter used to evaluate equipment on the basis of failure
probabilities. Reliability is defined as the probability of an equipment performing its
required function for a specific time interval under stated conditions. If malfunctions are
random, rates of failure will be uniform during this period and the probability will be
an exponential function of time as:
R(t) = el t (5.32)
where
R(t) = Reliability
l = Failure rate (failure/hour)
t = Time (hour).
An equipment typically goes through three distinct failure rate regions during its
lifetime, as shown in Figure 5.7. Improper handling or use can also cause failures, but
these are not usually considered in reliability.
1
M=
MTBF
T1 T2 M
(Mean life)
Time
The infant mortality is commonly an early failure period, when failure rates are
high. This may be caused due to inadequate quality control or poor manufacturing
Controller Design 345
practices. Most manufacturer conduct burn-in tests, during which devices are operated
for specific periods under simulated conditions to eliminate early failures before
shipment. Burn-in is the operation of the component, module, or system under some
increased stress, e.g. increased temperature. The increased stress is intended to cause
most of the early failures to occur in the manufacturers shop rather than in the users
plant.
The second stage is a normal operating interval in which random operating
stresses cause failures. Although these stresses cannot be predicted in a deterministic
fashion, data for large numbers of similar products typically show constant failure rates
during this period. This gives the frequency at which the devices can be expected to fail
(failure rate). The reciprocal of the failure rate is the mean time between failure
(MTBF), usually given in hours.
The final stage is the wear-out period, during which aging is the dominant failure
mechanism. Wear-out is a relatively minor problem with solid-state devices, since
normal operating lives are extremely long.
The reliability of any module is, therefore, a function of time and the module
failure rate. Reliability can be increased by:
· The use of high quality components
· Low stress on components
· A minimum number of components
· Burn-in at board level
· Environmental protection
· Heat soak of the final assembly, including cabinet and cables.
Availability
Reliability does not take into consideration that the equipment can be repaired and put
back on line. Another reliability parameter, availability, considers both MTBF and the
mean dead-time (MDT), which includes the mean time to repair (MTTR).
Availability is the probability that the system is working throughout the total
mission time. Systems availability (A) can be calculated as
7RVKOG
# (5.33)
6QVCNVKOG
If a system is always working, the availability is 100%. The availability of series string of
modules is simply the product of the availability of the individual modules, and given as
As = Aa ´ Ab ´ Ac ´ L´A m (5.34)
where
As = Overall series system availability
Aa, Am = Availability of individual module.
Similarly, availability of parallel combination of two modules, can be given as
Ap = Aa + Ab Aa Ab (5.35)
where
Ap = Overall parallel system availability
Aa, Ab = Availability of module a and b respectively.
346 Design of Process Control Systems
Availability can also be defined in terms of the mean time between failures (MTBF)
and the mean downtime (MDT)
/6$(
A = (5.36)
/&6
MDT = MTDF + MTTR (5.37)
MTTR = MTDL + MTRF + MTRO (5.38)
where
MTDF = Mean time to diagnose the system fault
MTTR = Mean time to repair
MTDL = Mean time to determine a fault location
MTRF = Mean time to replace a faulty component
MTRO = Mean time to return the system to operable condition
Backup systems
Backup systems are used to allow the system to safely shut down, allow for limited
production when the control system fails, and allow for continued plant operation. Each
system must be evaluated to determine what level of back-up is justified.
Continuous
sine wave
Sampled
sine wave
Sample-and-hold delay
The delay from a sample-and-hold results from using stored data. Consider a digitized
sine wave, as in Figure 5.8. At the beginning of the sample (just after the vertical
segment), the data is new. But during the sample interval, the data ages. At the end
of the sample interval, the data is approximately one sample old. The average age of the
data over the interval is half the sample interval. The phase, assuming this half-sample-
interval delay, is given as
( 65#/2.'
2JCUG5* ( 65#/2.'
where
F = Frequency in hertz (Hz)
TSAMPLE = Sample interval in second
Phase = 360 ´ F ´ t°DELTA.
The phase describes the time shift between input and output. This time shift (lag)
can be expressed in unit of time (tDELTA) but more often is expressed in degree, where
360° is equivalent to one period of the input sine wave.
The sample-and-hold does little to change the amplitude of the sampled signal. In
other words, the gain is approximately 0 dB. So, the transfer function of a sample-and-
hold is given as
65*
U F$²
( 65#/2.' (5.39)
Delay in Digital Controllers 349
The approximation of Eq. (5.39) is nearly exact. In fact, the phase is exact, but the
gain is a bit inaccurate at high frequencies. However, Eq. (5.39) is still commonly used
because the gain inaccuracy is usually insignificant. Below one-fourth the sample
frequency the impact on gain is less than 1 dB and so is small enough to ignore. Above
one-fourth the sample frequency the impact is still limited to just a few dB and the
system rarely needs to be analyzed to accuracy better than that at such high
frequencies. So, although the exact gain for the sample-and-hold is available, the simpler
form of Eq. (5.39) is almost always adequate.
Calculation delay
The second form of delay caused by digital controllers is called calculation delay. This
delay is generated because of the time required to execute the control laws. Unlike
analog systems, which process control laws continuously, digital systems require a finite
amount of time after sampling to calculate the control law output. The transfer function
of calculation time is simply
TCALC(s) = 0 dB Ð(360 ´ F ´ TCALC )° (5.40)
where TCALC is in second. Calculation delay does not cause attenuation. It can vary from
a small portion of the sample interval to nearly the entire interval. It is a function of
the complexity of the control laws and the skill with which they are coded. Calculation
delay is often much larger than the sample-and-hold delay and usually a portion comes
from careless construction of the control algorithms.
Velocity delay
The third form of delay is caused by estimating velocity from position. Only motion
control systems that rely on position sensors are subject to this delay. Most controllers
are designed for single-integrating plants such as electrical, translational mechanics,
rotational mechanics, fluid mechanics, heat flow, etc. Motion controllers control a
double-integrating plant because they apply torque, but they usually measure position
rather than velocity. (Note that motion controllers relying on techometer feedback do
not suffer from velocity-estimation delay.) The controller usually forms velocity as the
difference of the two most recent positions, given as
VN » (PN PN1)T (5.41)
where
VN = Current velocity
PN = Position
PN1 = Position from the previous sample.
Imperfections in this estimation generate additional phase lag equivalent to a
sample-and-hold. Consider that the difference is formed by a combination of new data
(PN) and data one sample old (PN1) so that the average age of the data is half of the
sample interval and is given as
T VEL-EST(s) » 0 dB Ð(180 ´ F ´ TSAMPLE)° (5.42)
This delay is identical to that generated by the sample-and-hold, as in Eq. (5.39), and can
be reduced by the inverse trapezoidal method.
350 Design of Process Control Systems
The total delay caused by digital controls (as shown in Table 5.1) is the sum of the
sample-and-hold delay, calculation delay and, for position-based motion systems, velocity
estimation delay.
Position-based motion
Source General system (in T) system (in T)
Sample-and-hold 0.5 0.5
Calculation delay 0.10.9 0.10.9
Velocity estimation 0.5
Total 0.61.4 1.11.9
Digital control systems rarely perform with the responsiveness of their analog
counterparts. Still, the advantages of digital systems often outweigh the performance
loss, especially if the performance differences are small. A small performance-difference
implies that the sample rate is high enough that the resulting phase lags do not
dominate the total phase lag in the loop.
and treble on a stereo system. Two parameters, the gain and the reset time, must be
adjusted, so the tuning procedure is significantly more complicated than when only one
parameter is involved. Finally, the tuning of three-mode or proportional-integral-
derivative (PID) controllers represents the next higher degree of difficulty. Here three
parameters, the gain, the reset time, and the derivative time must be adjusted.
Because controller tuning is usually done by trial and error, it can be quite tedious
and time-consuming. Consequently, it is desirable to have good preliminary estimates
of satisfactory controller settings. The values of the tuning parameters depend on the
desired closed-loop response and on the dynamic characteristics, or personality, of the
other elements of the control loop, particularly the process. A number of procedures and
formulas have been introduced to help enhance tuning effectiveness and give insight
into tuning itself.
In this section, a single tuning procedure will be applied to multiple controllers. As
with analog controllers, digital controllers are tuned in zones. Each of the six
controllers, i.e., proportional (P), proportional-integral (PI), proportional-integral-plus
(PI+), proportional-derivative (PD), proportional-integral-derivative (PID) and
proportional-integral-derivative-plus (PID+) has either one or two zones. The
proportional and differential gains combine to determine behaviour in the higher zone
and thus will be set first. So, the P and D gains must be tuned simultaneously. The
integral gain and a command filter determine behaviour in the lower zone.
The higher zone is limited by the control loop outside the control law: the plant,
the power converter and the feedback filter. The lower zone is limited primarily by the
higher zone. Note that sampling delays can be thought of as parts of these processes;
calculation delay and sample-and-hold delay (as explained in Section 5.4) can be thought
of as part of the plant and feedback delay as part of the feedback filter.
The tuning in this section will set the loop gains by optimizing the response to the
command. Higher loop gains will improve command response and they will also improve
the disturbance response. Depending on the application, command or disturbance
response may be more important. However, command response is usually preferred for
determining stability for a practical reason: commands are easier to generate in most
control systems. Disturbance response is also an important measure. When tuning, the
command should be as large as possible to maximize the signal-to-noise ratio. This
supports accurate measurements. However, the power converter must remain out of
saturation during these tests.
In this section, consistent sets of stability requirements are placed on all of the
controllers. Of course, in industry, requirements for controllers vary from one
application to another. The requirements used here are representative of industrial
controllers, but designers will need to modify these requirements for different
applications. The specific criteria for tuning will be as follows: in a response to a square-
wave command, the high-frequency zone (P and D) can overshoot very little (less than
20%), and the low-frequency zone can overshoot upto 15%. These tuning methods will
be based on time-domain measures of stability, chiefly overshoot in response to a square
wave.
flow, pressure and temperature. The type of controller and its setting used for any one
type of process control loop are sometimes much the same from one application to
another. For example, most flow control loops use PI controllers with wide proportional
band and fast integral action. Some of the examples discussed hereinafter are the
common practice, which work in most applications.
L evel transmitter
L evel sensor PV
LT
Fluid inlet F0
L evel controller
Surge tank
Fluid outlet, F
Control
valve
Pump
(a) Single liquid tank
Controller Tuning 353
Fluid inlet
F0
LC LC LC
h1 h2 h3
F
F1 F2 F3
(b) Tanks in series
Figure 5.9 P versus PI control for liquid level.
increased to the point that it is equal to F0, the level will stop changing since the tank
is just an integrator. Now, if we use a proportional (P) level controller, no change will
take place. The level will remain at the higher level, and the entering and exiting
flows will be equal. If, however, we use a proportional-plus-integral (PI) level controller,
the controller will continue to increase the outflow beyond the value of the inflow to
drive the level back down to its set-point. Thus, there is an inherent problem with PI
level controllers, that is, they amplify flow rate changes. The change in the flow rate
out of the tank is actually larger (for a period of time) than the change in the flow rate
into the tank. When the liquid tanks are working in series, this amplification gets
worse as it works its way down through the series units. Thus, a small disturbance
started out at the beginning can result in large fluctuations by the time it reaches the
last unit in the train. There are, of course, many applications where it is desirable to
control level tightly, for example, in a reactor where control of residence time is
important.
Vapour
Flow
FC transmitter
Control
valve Flow
FC controller
L iquid
condensate
Figure 5.10 P versus PI control for liquid level.
Vapour
PT PC PT
Cooling
water
L iquid
condensate
(a) Fast pressure loop (b) Slow pressure loop
Figure 5.11 Pressure control loop.
Controller Tuning 355
Figure 5.11(b) shows an example of a slower pressure loop in which the pressure
is held by throttling the water flow to a condenser. The water changes the DT, driving
force for condensation in the condenser. Therefore, the heat transfer dynamics and the
lag of the water flowing through the shell side of the condenser are introduced into the
pressure control loop.
Disadvantages of Trial and Error Method of Tuning. The trial and error tuning method
has the following disadvantages:
· It is quite time-consuming if a large number of trials is required to optimize Kp,
Ti and Td or if the process dynamics are quite slow. It may result in loss of
production and poor product quality.
356 Design of Process Control Systems
c( t) c( t)
c set c set
Time Time
(a) K p < K pu (b) K p < K pu
c( t) c( t)
c set c set
Time Time
(c) K p < K pu ( without saturation) (d) K p < K pu ( with saturation)
Figure 5.12 Results of trial and error tuning.
Controller type KP KI KD
P controller 0.50 KMAX 0 0
PI controller 0.45 KMAX 1.2 f0 0
PID controller 0.60 KMAX 2.0 f0 0.125/f0
Controller Tuning 357
Select controller
Done
The Ziegler-Nichols method is too aggressive for many industrial control systems.
For example, for a proportional controller, the method specifies a GM of just 6 dB,
compared with the 12 dB in the P controller. In general, the gains from Ziegler-Nichols
will be higher than from the methods presented here. Table 5.5 shows a comparison of
tuning the P, PI and PID controllers according to the method presented in Section 5.5
and Ziegler-Nichols method. The terms KMAX = 4.7 and f0 = 357 Hz were found
experimentally. Both sets of gains are stable, but the Ziegler-Nichols method provides
smaller stability margins.
Controller Kp Ti Td
- RW
Proportional-only P = 0.50Kpu
Proportional-integral PI
- RW
= 0.45Kpu
6W
- RW 6W 6W
Proportional-integral-derivative PID = 0.60Kpu
Controller Tuning 359
The quarter decay ratio response is very desirable for disturbance inputs because
it prevents a large initial deviation from the set-point without being too oscillatory.
However, it is not so desirable for step changes in set-point because it causes a 50%
overshoot. This is because the maximum deviation from the new set-point in each
direction is one-half the preceding maximum deviation from the opposite direction. This
difficulty can easily be corrected by reducing the proportional gain from the value
predicted by the formulas of Table 5.4. In fact, decay ratio is a direct function of the
controller gain and can be adjusted at any time by simply changing the gain. In other
words, if for a given process the quarter decay ratio response is too oscillatory, then a
reduction of the gain will smooth the response.
Although widely applied, the Ziegler-Nichols continuous cycling method has some
of the disadvantages as the trial-and-error method listed in section 5.7.2. However, the
continuous cycling method is less time-consuming than the trial-and-error method
because it requires only one trial and error search.
5.7.5 Cohen and Coon (CC) Rules or Process Reaction Curve Method
Cohen and Coon (CC) method of controller tuning was proposed by Cohen and Coon
(1953) and is often used as an alternative to the Ziegler and Nichols (Z-N) method. CC
rule is an open-loop method in which the control action is removed from the controller
by placing it in manual mode and an open-loop transient is induced by a step change in
the signal to the valve. Thus, this method is based on a single experimental test that
is made with the controller in the manual mode. After inducing a small step change in
the controller output, the process response is measured and recorded. This step
response is also referred to as the process reaction curve.
Figure 5.14 shows a block diagram of a typical control loop for measurement of the
process reaction curve. In this case, the control action is removed and the loop opened
for the purpose of introducing a step change to the valve. The step change is recorded
at the output of the measuring element. The step change to the valve is conveniently
provided by the output from the controller, which is in manual mode. The response of
the system (including the valve, process and measuring element) is called the process
reaction curve.
M /s U 0
R0
Gc Gv Gp C
L oop opened
C
H
To recorder
Figure 5.14 Block diagram of control loop for the process reaction curve.
360 Design of Process Control Systems
Figure 5.15 shows a typical S-shape process reaction curve showing graphical
construction to determine first-order with transport lag model. The S-shaped
process reaction curve can be represented by a first-order with transport lag model and
is given as
- U G6F U
) R
U (5.43)
6U
Using expression in Eq. (5.43), Cohen and Coon obtained the controller settings
shown in Table 5.5. The computations was made by CC considering the response having
¼ decay ratio, minimum offset, minimum area under the load-response curve, and other
favourable properties. The graphical construction of S-shaped process reaction curve
shown in Figure 5.15 reduces the process reaction curve to the first-order with the
transport lag model given by Eq. (5.43).
Bu
Bu
Tangent line, slope = — = S
T
0 t
0 Td
Input
0 t
0
Figure 5.15 S-shape process reaction curve.
Steps for Cohen and Coon (CC) tuning method are summarized as follows:
Step 1 After the process reaches steady state at the normal level of operation,
switch the controller to manual model. In a modern controller, the
controller output will remain at the same value after switching as it had
before switching. This is called bumpless transfer.
Step 2 With the controller in manual mode, introduce a small step change in the
controller output that goes to the valve and record the transient, which
is the process reaction curve as shown in Figure 5.15.
Controller Tuning 361
Step 3 Draw a straight line tangent to the curve at the point of inflection, as
shown in Figure 5.15. The intersection of the tangent line with the time
axis is the apparent transport lag Td. The apparent first-order time
constant T is obtained from the equation given as
$W
6 (5.44)
5
where Bu is the ultimate value of B at large t and S is the slope of the
tangent line. The steady-state process gain that relates B to M in Figure
5.14 is given by the equation
$W
-U (5.45)
/
Step 4 Using the values of Ks, T, and Td from Step 3, the controller settings are
found from the relations given in Table 5.5.
-R 6 § 6F ·
Proportional (P) - U 6F ¨© 6 ¸¹
6 § 6F ·
Proportional-plus-integral (PI) -R
- U 6F ¨© 6 ¸¹
6F
U+ 6F 6
6F
6
-R 6 § 6F ·
Proportional-plus-derivative (PD) - U 6F ¨© 6 ¸¹
6F
U& 6F 6
6F
6
6 § 6F ·
Proportional-plus-integral-plus-derivative (PID) -R
- U 6F ¨© 6 ¸¹
6F
U+ 6F 6
6F
6
U& 6F
6
F
6
362 Design of Process Control Systems
It can be noticed in Table 5.5 that all of the controller settings are a function of
6F
the dimensionless group , the ratio of the apparent transport lag to the apparent
6
time constant. Also, proportional gain Kp is inversely proportional to steady-state
process gain Ks.
The modelling method of process reaction curve shown in Figure 5.15 is crude and
does not give a very good fit. Finding the point of inflection and drawing a tangent line
at this point is quite difficult, especially if the data for the process reaction curve are not
accurate and if they scatter. A better method of fitting the process reaction curve to a first-
order with transport lag model is to perform a least-square fit of the data. However, the
disadvantage to this fitting procedure is that more time and effort required.
Advantages of CC (or process reaction curve method): The process reaction curve
method offers several significant advantages, as follows:
· Only a single experimental test is required.
· It does not require trial-and-error.
· The controller settings are easily calculated.
Disadvantages of CC (or process reaction curve method): The process reaction curve
also has disadvantages as follows:
· The experimental test is performed under open-loop conditions. Thus, if a
significant load change occurs during the test, no corrective action is taken and
the test results may be significantly distorted.
· It may be difficult to determine the slope at the inflection point accurately,
especially if the measurement is noisy and a small recorder chart is used.
· The method tends to be sensitive to controller calibration errors. By contrast,
the Z-N method is less sensitive to controller errors in Kp since the controller
gain is adjusted during the experimental test.
· The method is not recommended for processes that have oscillatory open-loop
responses.
5.7.6 Auto-tuning
Astrom and Hagglund described an automatic tuning, called auto-tuning method, an
alternative to Ziegler-Nichols continuous cycling method. This method has the following
features:
· The system is forced by a relay controller which causes the system to oscillate
with small amplitude. The amplitude of the oscillation can be constrained by
adjusting the amplitude of the input variations.
· Usually, a single closed-loop experiment is sufficient to find the dynamic model,
and the experiment does not require a priori information about the process
dynamic model.
The auto-tuner uses a relay with dead-zone to generate the process oscillation. The
period Tu is found simply by measuring the period of the process oscillation. The
ultimate gain is given by
4d
K pu (5.46)
Sa
Controller Tuning 363
where
d = Relay amplitude set by the operator
a = Measured amplitude of the process oscillation
Controller Machine
Feedback filter
Set KP low
No
Done
system. Higher gain amplifies noise, so setting the gain low will relieve the noise
problem but at the expense of degrading the control system performance. In case of
substantial noise, setting the proportional gain requires balancing the need for
performance and the elimination of noise.
Integral gain also causes instability. In the open loop, the integral, with a 90o
phase-lag, reduces the phase margin. In the time domain, the common result of adding
integral gain is overshoot and ringing. As a result, larger integral gains usually reduce
bandwidth, especially when these gains are large enough to benefit from command filter
(as discussed in Section 5.7.9).
With PI control, the P gain provides a similar operation as it did in the P controller
(Section 5.7.7), and the I gain provides DC stiffness. Larger I gain provides more
stiffness and, unfortunately, more overshoot. The PI controller is shown in Figure 5.18.
Note that KI is in series with KP; this is common, but it is also common to place the
two gains in parallel.
Controller Machine
PI control
Feedback filter
PI controllers have two zones: high and low. The high zone is served by KP and
the low by KI. Figure 5.19 shows a flowchart for tuning of PI controller. The process
for setting the proportional gain is the same as it was in the P controller. After the
higher zone is complete, KI can be tuned. Here it is raised for 15% overshoot to a square
wave. Again, a square wave is an unreasonably harsh command to follow perfectly; a
modest amount of overshoot to a square wave is tolerable in most applications.
No
Done
As can also be seen in Figure 5.20, when KFR is zero, command is filtered by, KI/(s + KI)
which is a single-pole low-pass filter at the frequency KI (in rad/sec). This case will allow
the highest integral gain but also will most severely limit the controller command
response. Typically, KFR = 0 will allow an increase (of almost three times) in the integral
gain but will reduce the bandwidth by one-half when compared with KFR = 1 (PI control).
Finding the optimal value of KFR depends on the application, but a value of 0.65 has been
found to work in many applications. This value typically allows the integral gain to more
than double while reducing the bandwidth by only 1520%.
One question about PI+ control that naturally arises is, why select the frequency
of the command low-pass filter to be KI? Why not set that frequency either higher or
lower? The reason is that this frequency is excellent at cancelling the peaking caused
by the integral gain. One way to look at PI+ control is that it uses the command filter
to attenuate the peaking caused by PI. The peaking caused by KI can be cancelled by
the attenuation of a low-pass filter with a break of KI.
PI+ is often referred to as pseudo-derivative feedback with a feed-forward (PDFF).
This method is shown in Figure 5.21. Although the equivalence between Figures 5.20
Controller Machine
Command
Error + Controller
+ Power G
G Response
Power
KFR KP s
converter s
+
+
KKII Sat Plant
K
KII
1 KFR ss
s +
s KKII
PI control
Command Feedback filter
and 5.21 is not obvious upon inspection, construction of the control law for
Figure 5.20 is
Î Ë -+ Û ÞÈ -+ Ø
%QPVTQN - 2 Ï%QOOCPF Ì - (4
- (4 Ü (GGFDCEM ß ÊÉ U ÚÙ (5.47)
Ð Í U - + Ý à
and of the control law for Figure 5.21 is
Ë - Û
- 2 Ì
%QOOCPF (GGFDCEM + - (4 %QOOCPF (GGFDCEM Ü (5.48)
%QPVTQN
Í U Ý
With some algebraic processing, Eq. (5.47) reduces to Eq. (5.48).
Controller
Controller Machine
Machine
KFR
+
+ Controller Response
++ K
KII ++ Contro Power Power
Powe G Respons
K
KPP
s converter s
Command
Comma
- -
Sat Plant
Plant
Sa
Feedback filter
Zero
Apply K I. Setwave
square KP low. Set atKabout
command FR to 10%
1.
of the desired loop bandwidth.
Use large amplitude, but avoid saturation.
No
High DC
stiffness General Fastest response
Done
Controller Machine
Error
Erro K
KDDss
+ Controller
Control
+ + Power Power G Response
KPP
K converter s
Command - +
Plant
KI Sat
s
Use of D gain advances the phase of the loop by virtue of the 90° phase lead of a
derivative. Using D gain will usually allow the system responsiveness to increase, for
example allowing the bandwidth to nearly double in some cases. Differential gain has
shortcomings, too. Derivatives have a high gain at high frequencies. So, while some D
do help the phase margin (PM), too much D hurts the gain margin (GM) by adding gain
at the phase crossover, typically a high frequency. This makes the D gain difficult to
tune. The designer sees overshoot improvement because of increased PM, but a high-
frequency oscillation, which comes from reduced GM, becomes apparent. The high-
frequency problem is often hard to see in the time-domain because a high-frequency
ringing can be hard to distinguish from normal system noise. So, a control system may
be accepted at installation but have marginal stability and thus lack the robust
performance expected for factory equipment.
Another problem with derivative gain is that derivatives are very sensitive to
noise. Even small amount of noise from wiring or resolution limitations may render the
D gain useless. In most cases, the D gain needs to be followed by a low-pass filter to
reduce the noise content. However, the lower the break frequency of the filter, the less
benefit can be gained from the D gain.
Figure 5.24 shows the flowchart for tuning of a PID controller. A PID controller
is a two-zone controller. The P and D gains jointly form the higher-frequency zone.
I gain forms the low-frequency zone. The benefit of the D gain is that it allows the P
gain to be set higher than it could be otherwise. The first step is to tune the controller
as if it were a P controller, but to allow more overshoot than normal (perhaps 10%),
understanding that the D gain will cure the problem. Typically, the P gain can be raised
2550% over the value from the P and PI controllers. The next step is to add a little
D gain to cure the overshoot induced by the higher than normal P gain. The P and D
gains together form the high-frequency zone. Next, the integral gain is tuned, much as
it was in the PI controller. The expectation is that the P and I gains will be about 30%
higher than they were in the PI controller.
Controller Tuning 371
Done
Note: BW = bandwidth; PM = phase margin; GM = gain margin; KVP = velocity loop proportional
gain in PI velocity loop (amps/rad/sec); KVI = velocity loop integral gain in PI velocity loop
(sec1); KVD = velocity loop derivative gain (sec) in PI velocity loop; KVFR = PI+ gain.
*PI+ and PID+ are a continuum, which can be evaluated with KVFR ranging from 0 to 1. The
evaluation of PI+ and PID+ here set KVFR to 0.65. When KVFR is 1, the methods are equivalent
to PI and PID, respectively.
Controller Machine
Error K
KDDss
Command +
Controller
Control Power Response
KFR + + K Power GG
KPp
converter ss
+ -
K
KI + Plant
KKII SatSat
11–K
KFR
FR
KI ss
ss K I
High DC
stiffness General Fastest response
Done
Controller Machine
Error
KDs
+ Controller Response
+ + Power Power G
KP
converter s
Command -
Plant
Sat
PD control
Feedback filter
Done
many applications. The introduction of the I term provides DC stiffness but reduces PM.
The command filter in PI+ and PID+ allows even higher DC stiffness but reduces
bandwidth. The D term provides a higher responsiveness but reduces GM and adds a
phase shift, which is a disadvantage if this loop is to be enclosed in an outer loop.
The flowchart in Figure 5.29 provides a procedure for selecting a controller. First
determine whether the application needs a D gain; if not, avoid it as it adds complexity,
increases noise susceptibility, and steals gain margin. Next, make sure the application
can support D gains; systems that are noisy configured as PI controllers may not work
well with a differential gain. After that, examine the application for the needed DC
Yes
Yes
Does the Digital Does the
No application Analog or application No
require a DC digital signal? require a DC
stiffness? stiffness?
Analog
Yes Yes
Yes Yes
Will the No No Will the
overshoot be overshoot be
a problem? a problem?
Yes Yes
stiffness. If none is required, avoid the integral gain. If some is needed, use the standard
form (PI or PID); if maximum DC stiffness is required, add the input filter by using PI+
or PID+ control.
Output ITA E
IA E
ISE
Time
Figure 5.30 Closed-loop responses using various time interval criteria.
378 Design of Process Control Systems
closed-loop responses. When we tune the controller performance using IAE, ASE or
ATAE performance criteria, we should remember the following two points:
· Different criteria lead to different controller designs.
· For the same time integral criterion, different input changes lead to different
designs.
The above procedure, although mathematically rigorous, has following serious
practical problems:
· It is very tedious.
· It relies on models (transfer functions) for the process, sensor and final control
element which may not be known exactly.
· It incorporates certain ambiguities as to which is the most appropriate criterion
and what input changes to consider.
Proportional control
· Accelerates the response of controlled processes.
· Produces an offset (i.e., non-zero steady-state error) for all processes except
those with terms 1/s (integrators) in their transfer function, such as the liquid
level in a tank or the gas pressure in a vessel.
Integral control
· Eliminates any offset.
· The elimination of the offset usually comes at the expense of higher maximum
deviations.
· Produces sluggish, long oscillating responses.
· If we increase the gain Kc to produce faster response, the system becomes more
oscillatory and may be led to instability.
Derivative control
· Anticipates future errors and introduces appropriate action.
· Introduces a stabilizing effect on the closed-loop response of a process.
Figure 5.31 illustrates in a very simple way the characteristics of the above
feedback controllers.
It is clear that a three-mode PID controller should be the best. This is true in the
sense that it offers the highest flexibility to achieve the desired controlled response by
having three adjustable parameters. At the same time, it introduces a more complex
tuning problem because we have to adjust three parameters.
Choosing the Controller 379
Uncontrolled
y( t ) response
PI
PID
t
Figure 5.31 Closed-loop responses to unit step change in load for various types of feedback controller.
To balance the quality of the desired response against the tuning difficulty we can
adopt the following rules in selecting the most appropriate controller:
(a) As far as possible, use simple proportional controller. Simple proportional
controller can be used if (i) we can achieve acceptable offset with moderate
values of Kc or (ii) the process has an integrating action (i.e., a term 1/s in its
transfer function) for which the P control does not exhibit offset. Therefore, for
gas-pressure or liquid-level control we can use only P controller.
(b) If a simple P controller is unacceptable, use PI. A PI controller should be used
when proportional control alone cannot provide sufficiently small steady-state
errors (offsets). Therefore, PI will seldom be used in liquid-level or gas-pressure
control systems, but very often (almost always) for flow control. The response
of a flow system is rather fast. Consequently, the speed of the closed-loop
system remains satisfactory despite the slowdown caused by the integral
control mode.
(c) Use a PID controller to increase the speed of the closed-loop response and
remain robustness. The PI eliminates the offset, but reduces the speed of the
closed-loop response. For a multi-capacity process whose response is very
sluggish, the addition of a PI controller makes it even more sluggish. In such
cases, the addition of the derivative control action with its stabilizing effect
allows the use of higher gains which produce faster responses without
excessive oscillations. Therefore, derivative action is recommended for
temperature and composition control where we have sluggish, multi-capacity
processes.
380 Design of Process Control Systems
ILLUSTRATIVE EXAMPLES
Example 5.1 The differential equation describing the dynamics of a second-order
system, is as follows:
F [ F[
U
[U
[ Z
V (5.49)
FV FV
Develop a computer program in C++ to solve this problem by using the
RungaKutta computer simulation method, for the following conditions:
t = 1, z = 0.4, x(t) = u(t) = 1, y(0) = 0.
We must first express Eq. (5.49) as two first-order differential equations by letting
y1 = y (5.50)
and
F[ F[
[ (5.51)
FV FV
Using these expressions in Eq. (5.49), we obtain
F[
[ (5.52)
FV
and
F[ [
[ [ Z
V (5.53)
FV U U U
Equations (5.52) and (5.53) have been written in terms of state variables, y1 and y2. We
are now ready to apply the RungaKutta method to these equations. The RungaKutta
method for solving differential equation is often called a marching solution because the
calculation starts at an initial value of the independent variable t and moves forward one
integration step at a time. It divides the independent variables t into increments of equal
length Dt, as shown in Figure 5.32.
The fourth-order RungaKutta method uses the following equations:
k1 = f(y0, t0)Dt (5.54)
Dt Dt Dt Dt
y
t
t0 t1 t2 t3 t4
Figure 5.32 Dividing the independent variable t into equal increments Dt.
Illustrative Examples 381
È M 'V Ø
k2 = f É y0 + , t0 + Dt (5.55)
Ê ÙÚ
È M 'V Ø
k3 = f É y0 + , t0 + Dt (5.56)
Ê ÙÚ
È 'V Ø
k4 = f É y0 + k3, t0 + Dt (5.57)
Ê ÙÚ
y1 = y0 + (k1 + 2k2 + 2k3 + k4) (5.58)
t1 = t0 + Dt (5.59)
The above equations are applied during the first increment Dt from t0 to t1. The
values obtained at the end of the first increment (y1, t1) are then used as a new set of
initial conditions in these equations to obtain a set of values of y and t at the end of
the second interval. This procedure of computing y and t at the end of successive
intervals generates the solution to the differential equation.
The set of equations Eqs. (5.54) to (5.59), used to solve a single first-order
differential equations:
F[
H
[ Y V (5.60)
FV
and
FY
H
[ Y V (5.61)
FV
with the initial conditions y0, w0, t0. The RungaKutta equations used to solve for y(t)
and w(t) are given as follows:
k1 = f1(y0, w0, t0)Dt (5.62)
l1 = f2(y0, w0, t0)Dt (5.63)
È M N 'V Ø
k2 = f1 É [ Y V Ù 'V (5.64)
Ê Ú
l2 = f2 ÈÉ [ M Y N V 'V ØÙ 'V (5.65)
Ê Ú
M N 'V
k3 = f1 ÈÉ [ Y V ØÙ 'V (5.66)
Ê Ú
È M N 'V Ø
l3 = f2 É [ Y V Ù 'V (5.67)
Ê Ú
k4 = f1(y0 + k3, w0 + l3, t0 + Dt)Dt (5.68)
l4 = f2(y0 + k3, w0 + l3, t0 + Dt)Dt (5.69)
y1 = y0 + (k1 + 2k2 + 2k3 + k4) (5.70)
w1 = w0 + (l1 + 2l2 + 2l3 + l4) (5.71)
t1 = t0 + Dt (5.72)
382 Design of Process Control Systems
A computer program written in C++ is shown in Figure 5.33. In the program, the
#include<stdio.h>
#include<math.h>
void main()
{
//response of 2nd order system by runga-kutta
float fndy1(float,float);
float fndy2(float,float);
float zeta=.4,dt=.05,k,l1,o,k1,k2,l2,l3,i,k3,l4,k4,tau=1,y1=0,y2=0,t=0;
printf(" T Y");
k=0;o=1;
for(i=0;i<o;i++)
{
k=k+1;
if(k==11)
{
printf("\n%f %f",t,y1);
k=1;
}
k1=fndy1(y1,y2)*dt;
l1=fndy2(y1,y2)*dt;
k2=fndy1(y1+.5*k1,y2+.5*l1)*dt;
l2=fndy2(y1+.5*k1,y2+.5*l1)*dt;
k3=fndy1(y1+.5*k2,y2+.5*l2)*dt;
l3=fndy2(y1+.5*k2,y2+.5*l2)*dt;
k4=fndy1(y1+k3,y2+l3)*dt;
l4=fndy2(y1+k3,y2+l3)*dt;
y1=y1+(k1+2*k2+2*k3+k4)/6;
y2=y2+(l1+2*l2+2*l3+l4)/6;
t=t+dt;
if(t<10.05)
{
o++;
}}printf("\n");
}
float fndy1(float y1,float y2)
{
return (y2);
}
float fndy2(float y1,float y2)
{
float zeta=.4;
int tau=1;
return(1/pow(tau,2)-y1/pow(tau,2)-2*zeta*y2/tau);
}
Figure 5.33 A C++ program for a step-response system of a second-order system forÿt = 1, z = 0.4.
functions DY1 and DY2 correspond to Eqs. (5.52) and (5.53). After defining the
parameters (t,z and Dt), the RungaKutta procedure is started.
The output results from running the program in Figure 5.33 are shown in
Figure 5.34.
Illustrative Examples 383
T Y
0.500000 0.107667
1.000000 0.359915
1.500000 0.658174
1.999999 0.927084
2.499999 1.122055
2.999998 1.228117
3.499998 1.253177
3.999997 1.218883
4.499999 1.151724
5.000001 1.076087
5.500003 1.010032
6.000005 0.963729
6.500007 0.940011
7.000009 0.936218
7.500010 0.946547
8.000012 0.964258
8.500014 0.983343
9.000016 0.999497
9.500018 1.010427
10.000020 1.015650
Press any key to continue
Figure 5.34 Output results of the program in Figure 5.33.
Example 5.2 Obtain the transient response of the liquid-level control system shown in
Figure 5.35 using a digital computer. The values of the parameters of the
block diagram are as follows:
Kp = Proportional gain, psi/ft tank level
Ti = Integral time, min
Kv = Valve constant = 0.070 (ft3/min)/psi
T1 = Time constant of tank 1 = 2.0 min
T2 = Time constant of tank 2 = 1.0 min
T3 = Time constant of tank 3 = 1.0 min
For convenience in simulating this system, the diagram of Figure 5.35(b) has been
reduced to that of Figure 5.36 in which Kv has been combined with the PI control
transfer function in one block and the transfer functions for the three tanks have been
combined in one block.
To obtain the differential equations for use in the RungaKutta method, we
proceed as follows. From the controller block, we may write
/
U È Ø 6K U
- R - X É - R - X (5.73)
'
U Ê 6K U ÙÚ 6K U
Cross-multiplying gives
TisM(s) = KpKvTisE(s) + KpKvE(s)
384 Design of Process Control Systems
PI controller
Water
supply
Water
Water
c R1
1
R
s § 1 · M CC
+ K p ¨¨1 ¸¸ M Kv 1 1 R1
© Ti s ¹ 2s 1 s 1 s 1
-
+ E ¥ 1µ M 0.55 C
R = u( t) K p K v ¦1
§ Ti s·¶ (2s 1)( s 1) 2
Figure 5.36 Reduced block diagram of control system for Example 5.2.
- R -X
O
- R - X G G (5.74)
6K
From the comparator of Figure 5.35, we have
e = 1 c (5.75)
and
G E (5.76)
Replacing e and G in Eq. (5.74) by the expression in Eqs. (5.75) and (5.76) gives
- R -X
O
- R - X E
E (5.77)
6K
The three tanks are represented by
%
U
(5.78)
/
U
U
U
The differential equation represented by Eq. (5.78) can be formed by cross-
multiplying. The result is
(2s + 1)(s + 1)2C(s) = 0.55M(s)
or,
(2s3 + 5s2 + 4s + 1)C(s) = 0.55M(s) (5.79)
n
Reorganizing s C(s) to be nth derivative of c in the time domain, Eq. (5.79) can be
written as
E
E E E O
or
E
E E
E O (5.80)
In order to apply the RungaKutta method, we must express Eq. (5.80) as three first-
order differential equations. The procedure will now be shown.
Let
x = c (5.81)
y = E (5.82)
z = E (5.83)
386 Design of Process Control Systems
Z = y (5.84)
[ = z (5.85)
#include<stdio.h>
#include<math.h>
void main()
{
float fnm(float,float,float,float);
float fnx(float,float,float,float);
float fny(float,float,float,float);
float fnz(float,float,float,float);
float
kp,m,m1,x1,x2,x3,x4,m2,m3,m4,y1,y2,y3,y4,z1,z2,z3,z4,kv,taui,x,y,z,t,dt,k;
Illustrative Examples 387
int i,o;
kv=.07;kp=107.2;taui=3.7;o=1;
m=kp*kv;
x=0;y=0;z=0;t=0;dt=.1;
printf(Time, min Level, ft);
k=0;
for(i=0;i<o;i++)
{
k=k+1;
if(k==6)
{
printf(\n%f %f,t,x);
k=1;
}
m1=fnm(m,x,y,z)*dt;
x1=fnx(m,x,y,z)*dt;
y1=fny(m,x,y,z)*dt;
z1=fnz(m,x,y,z)*dt;
m2=fnm(m+(m1*.5),x+(x1*.5),y+(y1*.5),z+(z1*.5))*dt;
x2=fnx(m+(m1*.5),x+(x1*.5),y+(y1*.5),z+(z1*.5))*dt;
y2=fny(m+(m1*.5),x+(x1*.5),y+(y1*.5),z+(z1*.5))*dt;
z2=fnz(m+(m1*.5),x+(x1*.5),y+(y1*.5),z+(z1*.5))*dt;
m3=fnm(m+(m2*.5),x+(x2*.5),y+(y2*.5),z+(z2*.5))*dt;
x3=fnx(m+(m2*.5),x+(x2*.5),y+(y2*.5),z+(z2*.5))*dt;
y3=fny(m+(m2*.5),x+(x2*.5),y+(y2*.5),z+(z2*.5))*dt;
z3=fnz(m+(m2*.5),x+(x2*.5),y+(y2*.5),z+(z2*.5))*dt;
m4=fnm(m+m3,x+x3,y+y3,z+z3)*dt;
x4=fnx(m+m3,x+x3,y+y3,z+z3)*dt;
y4=fny(m+m3,x+x3,y+y3,z+z3)*dt;
z4=fnz(m+m3,x+x3,y+y3,z+z3)*dt;
m=m+(1/6)*(m1+(m2*2)+(m3*2)+m4);
x=x+(1/6)*(x1+(x2*2)+(x3*2)+x4);
y=y+(1/6)*(y1+(y2*2)+(y3*2)+y4);
z=z+(1/6)*(z1+(z2*2)+(z3*2)+z4);
t=t+dt;
if(t<(10.1))
o++;
}printf(\n);}
float fnm(float m,float x,float y,float z)
{
float kv,kp,taui;
kp=107.2;kv=.07;taui=3.7;
return((-kv*kp*y)+(kp*kv*(1-x))/taui);
}
float fnx(float m,float x,float y,float z)
{
return(y);
}
float fny(float m,float x,float y,float z)
{
return(z);
}
float fnz(float m,float x,float y,float z)
{
return((-2.5*z)-2*y-.5*x+.275*m);
}Figure 5.37 A C++ program for a liquid-level control system for a step change in the set-point of
1.0 ft, Kp = 107 psi/ft and Ti = 3.7 min.
388 Design of Process Control Systems
The output results from running the program in Figure 5.37 are shown in
Figure 5.38.
m = Kpe
e= r x
x = y(t Td)
R = u(t) + E K M 1
Kpp Y
Ws 1
-
X W s
e t dds)
exp (
Figure 5.39 Block diagram of a proportional control system with transport lag.
Illustrative Examples 389
The difference between this example and Example 5.2 is the presence of a transport lag.
In the previous digital simulations, only the current value of y was needed and hence
stored. In this example, we must store values of y over the time interval
(t Td) to t (i.e. over the interval Td). Since we compute y only at discrete times, we
store values of y in an array of computer storage locations, called a stack. The diagram
in Figure 5.40 will help clarify this storage.
Present value
Present of of
value YY
S(1)
S(1)
S(2)
S(2)
S(3)
S(3)
S(4)
S(4)
Delay = =
Delay TdTd
x
S(N)
S(N)
S(N+1)
S(N+1)
Delayed value
Delayed of Yof(i.e.
value X) X)
Y (i.e.,
Figure 5.40 Array used to obtain a transport lag.
The array will be used to store past values of y that were computed at the end of
each computation interval. At the end of each computation interval, the values of y will
be moved one position toward the end of the stack and the value of y just computed
will be placed in the first storage location of the stack. This means, a current
value of y will not appear at the end of the stack until it has moved through each
storage location. The amount of time the current value of y is delayed will depend on
the number of storage locations and Dt. The number of storage locations N is
determined by:
6
0 F
'V
Let the values stored in the array be S(i), where I, which represents the array position,
will vary from 1 to (N + 1). The following terms are now defined for the computer
program to be developed:
390 Design of Process Control Systems
Y = y, present value of y
S(i) = Stored past values of y
S(1) = Current value of y, obtained at end of most recent computation interval
S(N + 1) = X, the delayed value of y, i.e. X = y(t Td).
An outline of the procedure for computing y at discrete values of t is as follows:
(i)
Let the array for storing values of y be of length N + 1, where, N = Td/Dt.
(ii)
Initialize the elements of the array to zero.
(iii)
Initialize the time variable, T = 0.
(iv)Set X = S(N + 1).
(v)Print T, Y and X.
(vi)Start the RungaKutta routine to integrate the differential equation over the
first computation interval DT.
(vii) Rearrange the contents of the array, as shown in Figure 5.40, by shifting the
contents of each storage location by one position. Start shifting from the
bottom. In this shifting, the oldest value of Y will be discarded and the value
of Y just computed will enter the first cell to become S(1).
(viii) Store the value of Y just computed into S(1), i.e. S(1) = Y.
(ix) Increment T by DT and return to step (iv) to repeat another cycle of
calculation.
Using the above steps, the C++ computer program shown in Figure 5.41 has been
written for the conditions: t = 1.0, Td = 0.2, R = u(t), Kp = 8.4 and DT = 0.02. For these
conditions:
6
0 F
'V
#include<stdio.h>
#include<math.h>
void main()
{
//first order system with transport lag
float s[11],kp,k1,k2,k3,k4,y,t,dt,x;
float fndy(float,float);
int i,o,j,k;
kp=8.399999;o=1;y=0;t=0;dt=.02;
for(i=1;i<12;i++)
{
s[i]=0;
}
printf("T Y X");
for(j=0;j<o;j++)
{
x=s[11];
k1=fndy(y,x)*dt;
k2=fndy(y+.5*k1,x)*dt;
k3=fndy(y+.5*k2,x)*dt;
k4=fndy(y+k3,x)*dt;
Illustrative Examples 391
y=y+(k1+2*k2+2*k3+k4)/6;
t=t+dt;
printf("\n%f %f %f",t,y,x);
k=10;
for(i=1;i<11;i++)
{s[k+1]=s[k];
k=k-1;}
s[1]=y;
if(t<(1.001))
o++;
}printf("\n");
}
float fndy(float y,float x)
{
float kp;kp=8.399999;
return(-y+kp-(kp*x));
}
Figure 5.41 A C++ program for the control of a first-order system with a transport lag.
The results from running the program in Figure 5.41 are shown in Figure 56.42.
T Y X
0.020000 0.166331 0.000000
0.040000 0.329369 0.000000
0.060000 0.489178 0.000000
0.080000 0.645823 0.000000
0.100000 0.799366 0.000000
0.120000 0.949868 0.000000
0.140000 1.097391 0.000000
0.160000 1.241992 0.000000
0.180000 1.383730 0.000000
0.200000 1.522661 0.000000
0.220000 1.658842 0.000000
0.240000 1.764660 0.166331
0.260000 1.841264 0.329369
0.280000 1.889770 0.489178
0.300000 1.911261 0.645823
0.320000 1.906787 0.799366
0.340000 1.877369 0.949868
0.360000 1.823995 1.097391
0.380000 1.747627 1.241992
0.400000 1.649195 1.383730
0.420000 1.529604 1.522661
0.440000 1.389730 1.658842
0.460000 1.235025 1.764660
0.480000 1.070641 1.841264
0.500000 0.901445 1.889770
0.520000 0.732024 1.911261
0.540000 0.566702 1.906787
0.560000 0.409547 1.877369
0.580000 0.264381 1.823995
0.600000 0.134793 1.747627
392 Design of Process Control Systems
Example 5.4 Obtain the simulation of a PID control of a first-order system shown in
Figure 5.43.
1
U
s
+
+ E ª§ 11 º· M + 1
RR=0
= 0 KKpp«1¨1 Tddss»¸ Y
T
T ss Ws 1
¬© i i ¼¹
-
Cross-multiplying this expression, solving for sM, and writing the result in the time
domain give
Illustrative Examples 393
-R
O
- R G G - R6F
G (5.89)
6K
This expression is not in the form in which the right side is free of derivatives of the
variables. To obtain the correct form, we proceed as follows. Since R = 0 for this
problem, we may write
e = y (5.90)
G = [ (5.91)
[ =W
V O [? (5.93)
U
where U(s) = 1/s has been written as u(t) (a unit-step) in the time domain. Taking the
derivative of both the sides of Eq. (5.93) gives
[
=
V O
[
? (5.94)
U
where the use has been made of the fact that the derivative of a unit-step function is
a unit-impulse function. Combining Eqs. (5.90)(5.92), with Eqs. (5.93) and (5.94) gives
G [ =W
V O [? (5.95)
U
Î È Ø Þ
G
[ É Ù >W
V O [@ß
Ï
V O (5.96)
UÐ ÊU Ú à
Substituting the expressions for e, G CPF
G from Eqs. (5.90), (5.95), and (5.96) into
Eq. (5.89) gives,
O
%= 6FU
V # $[ #O? (5.97)
where
-R
U
A = Td t , $ U 6F %
6K
U - R6F U
The right side of Eq. (5.97) contains the forcing term C Tdt (t). If Eq. (5.97) were
integrated, this term would contribute a constant value of CTdt. The reason for this
is that the integration of a unit is a unit-step, thus
V
Ô
V FV W
V
394 Design of Process Control Systems
Solving Eqs. (5.98) and (5.99) with the given initial conditions will produce a
response for the control system of Figure 5.43. The procedure for programming
Eqs. (5.98) and (5.99) by use of the RungaKutta method is straightforward.
Example 5.5 Determine controller settings for a PI controller using both Cohen and
Coon (CC) method for a control system shown in Figure 5.44.
+ § 1 · 1
Kc ¨1 ¸ e s C
R = u( t)
© U1s ¹ s1
–
As depicted in Figure 5.44, the transfer function of the model is already in the form
of first-order with transport lag in the form of Eq. (5.43). This is form used by CC
method to derive their tuning rules. Therefore, we obtain T and Td immediately without
having to draw a tangent line through the point of inflection, i.e., T = 1 and Td = 1. From
the block diagram of Figure 5.44 we also observe that the process gain Kproc = 1.
Substituting these values into the appropriate equations of Table 5.7 gives
6 § 6 · § ·
-E ¨ F ¸ ¨ ¸
- RTQE 6F © 6 ¹ © ¹
Illustrative Examples 395
6F
6
and U+ 6F
6F
6
Using these values for Kc and tI, the step response can be obtained by simulation
as shown in Figure 5.45.
c
1.5 CC
1.0
0.5 ZN
0.0 t
0.0 5.0 10.0
Figure 5.45 Unit step response of Example 5.5 (Figure 5.44)
Example 5.6 Determine controller settings for a PI controller using CC and Z-N methods
for a control system shown in Figure 5.46. In this problem, the process
reaction curve must be moulded by the method shown in Figure 5.15.
1
U= s
+ § 1 · 1
R=0 Kc ¨1 ¸ C
© U 1s ¹ ( s 1) 4
CC method
Since the transfer function of the plant is given as , we can obtain the value of
U
Td and T for use in the CC method analytically. A unit-step response for the plant
396 Design of Process Control Systems
transfer function is
§ ·
E
V ¨ V V V ¸ G V
© ¹
First and second derivatives can be readily obtained from this result, and thus
V
E
V V G
V
E
V
G V V
The location of the inflection point on the transient c(t) is obtained by setting the
second derivative to zero:
V
G V V
Solving for t gives as the root of interest in this problem t = 3. Knowing that the
point of inflection occurs at t = 3, we compute the slope of the tangent line through this
point to be
5 E
G
We can now determine Td as shown in Figure 5.47. From the expression for c(t),
we obtain the value of c at the inflection point to be c(3) = 0.353. The value of t where
the tangent line intersects the t-axis is obtained from the slope S; thus
5
6F
C
1.0
Slope = 0.224
0.9
0.8
0.7
0.6
0.5
0.4
0.3
Point of inflection
0.2
0.1
0.0 t
0 1 2 3 4 5 6 7 8 9 10
Figure 5.47 Process reaction curve for Example 5.6.
Illustrative Examples 397
1.0
Z-N
0.8 (CC) 1
(CC) 2
0.6
0.4
0.2
0.0
– 0.2
– 0.4
– 0.6
– 0.8
–1.0 t
0 2 4 6 8 10 12 14 16 18 20
Figure 5.48 Transients produced by Z-N and CC methods.
Z-N method
When we apply the Z-N method for a PI controller, we obtain the following results:
Kcu = 4
2W Q
Kc = 1.8
U+
The transient for this set of controller parameters is also shown in Figure 5.48. We
see that the response is stable and well damped.
398 Design of Process Control Systems
Applying the least square fit procedure out to t = 5, following results will be
produced:
Td = 1.5 and T = 3.0
Applying the CC method for these values of Td and T gives
Notice that the value of Kc is now considerably less than the value obtained from
the fitting procedure shown in Figure 5.47. This leads to expectation that the response
will now be stable. This expectation is fulfilled by the transient labelled (CC)2 as shown
in Figure 5.48.
REVIEW QUESTIONS
5.1 Explain the control systems specifications.
5.2 What is a MIMO process, and in what sense is the design of a control system for
a MIMO process different from that for a SISO process?
5.3 Discuss the performance specification for a single-input, single-output (SISO)
linear time-invariant (LTI) system. Derive the equation for rise time (tr) and delay
time (td) for a transient response specification.
5.4 Explain the various quantities that are required to specify the transient response
characteristics of SISO LTI system.
5.5 What are the robustness specification for SISO LTI systems?
5.6 What are the performance and robustness specifications for a multiple-input
multiple-output (MIMO) LTI system?
5.7 What are the control problems for standard industrial controllers in process
control applications? How a proper design can solve these problems?
5.8 Draw a block diagram of a popular controlled system. Explain why a compensator
is incorporated into the control loop. Define the transfer function of the
compensated system.
5.9 What is regulation performance? How regulation problems are solved?
5.10 Discuss the control design methods used for a SISO LTI system. Explain the
difference between these methods.
5.11 Discuss the various design considerations while designing a controller for process
control application.
5.12 Define the term delay in digital controllers. What is the impact of sampling delay
while employing digital controllers?
5.13 What is a controller tuning? Why tuning a controller is necessary?
5.14 Discuss with neat sketches and flowcharts the tunings of P, PI, PI+, PID, PID+
and PD controllers. Compare the tuning results of these controllers.
Self-Check Quiz 399
SELF-CHECK QUIZ
A State True/False:
5.1 The peak time tp is the time required for the response to reach half of its final
value for the very first time.
5.2 The rise time tr is the time required for the response to rise from x% to y% of
its final value.
5.3 The settling time ts is defined as the time required for the response to settle to
within a certain percent of its final value.
5.4 The delay time td is the time required for the response to reach the first (or
maximum) peak.
5.5 The bandwidth wb is defined as the frequency at which the magnitude of the
closed loop frequency response drops to 0.070 of its zero-frequency value.
5.6 The Smith predictor is a special design method for dead-time process control.
5.7 Reliability is defined as the probability of equipment performing its required
function for a specific time interval under stated conditions.
5.8 MTD = MTDL + MTRF + MTRO
5.9 MTTR = MTDF + MTRF
5.10 Calculation delay is generated because of the time required to execute the control
laws.
5.11 Control field tuning is often performed using trial and error methods suggested
by controller manufacturers.
5.12 Trial and error tuning procedure is applicable to processes that are open-loop
unstable.
5.1 _____________ is the time the system takes to attain a nearly constant value,
usually ±5 percent of its final value.
5.2 The __________ is the time required for the response to reach the first
(or maximum) peak.
5.3 The ___________is the time required for the response to rise from x% to y% of
its final value.
5.4 The __________ is defined as the time required for the response to settle to
within a certain percent of its final value.
5.5 The __________ is the time required for the response to reach half of its final
value for the very first time.
5.6 The ___________ represents the amount that the response overshoots its
steady-state (or final) value at the peak time, expressed as a percentage of
the_____________.
5.7 The _________________is defined as the maximum magnitude of the closed-loop
frequency response.
5.8 The frequency at which Mp occurs is called the ______________.
5.9 The __________ is defined as the frequency at which the magnitude of the closed
loop frequency response drops to 0.070 of its zero-frequency value.
5.10 ________ is the adjusting of the controller parameters to obtain a specified
closed-loop response.
5.11 Control field tuning is often performed using _____________.
Part III
6.1 INTRODUCTION
On the one hand, the implementation of control loops requires the availability of
adequate technology for the sensors, actuators and devices that make up the control
system. On the other hand, in order to achieve a satisfactory performance in its
operation, the selected control methodology or system should be able to solve the
problems inherent in the process dynamics and its environment. This satisfactory
performance would demand an ideal control methodology or strategy, able to satisfy the
following requirements:
· To guarantee the stability and, particularly, robustness
· To be as efficient as possible, as required by the desired performance criteria
· To be easy to implement and operate in real-time through the use of digital
computers.
Therefore, control systems are required to have a high dynamical performance and
robust behaviours, yet are expected to cope with more complex, uncertain and highly
nonlinear dynamic processes. Along with this increased process complexity is increased
abstraction and uncertainty in the models and their mathematical representation. One
significant approach in dealing with major changes and uncertainty in nonlinear
dynamical processes is through advanced or intelligent modelling and control. Intelligent
controllers are generally self-organizing or adaptive and are naturally able to cope with
significant changes in the plant and its environment, while satisfying the control
requirements. As with any advanced control theory, a central issue is the representation
and development of appropriate process models with known approximation errors. As
processes increase in complexity, they become less amenable to direct mathematical
modelling based on physical laws, since they may be:
· distributed, stochastic, nonlinear and time varying
· subject to large unpredictable environmental disturbances
· have variables that are difficult to measure, have unknown causal relationships,
or are too difficult or expensive to evaluate in real-time.
405
406 Advanced Process Control Strategies
While a human operator can cope with a large set of observations and frequently
conflicting constraints subject to multiple subjective and objective performance criteria,
the operator can be error prone, suffers from fatigue and non-repeatability, and can
generate potentially dangerous situations in safety-critical systems.
Therefore, one aspect of advanced control strategy or of intelligent control, is to
develop systems that incorporate the creative, abstract and adaptive attributes of a
human, while minimizing the undesirable aspects such as unpredictability,
inconsistency, fatigue, subjectivity and temporal instability. Advanced or intelligent
controllers are able to perform under significant process uncertainties and
incompleteness in the system and its environment, being both sufficiently robust or
reconfigurable to cope automatically with system failures and sufficiently adaptive to
cope with new goals or unanticipated situations.
Advanced control strategies attempt to compensate for process deviations in the
shortest possible time by accounting for process dynamics, dead times, time delays and
loop interactions. The benefits of advanced control are:
· Increased throughput
· Increased product recovery
· Energy conservation
· Reduced disturbances to other processing units
· Minimum rework or recycle of off-spec products
· Reduced operating manpower
· Increased plant flexibility.
In this chapter, following advanced control strategies are discussed in details:
· Cascade control
· Predictive control
· Adaptive control
· Inferential control
· Statistical process control
· Intelligent control (artificial intelligence).
utilizes multiple feedback loops and called cascade control. Cascade control is built up
by nesting the control loops, as shown in Figure 6.1.
R Primary Secondary
Process, Process, c
control loop, control loop,
Cp P1 P2
Cs
Inner loop
Outer loop
The system has two loops: The inner loop is called the secondary loop, and the
outer loop is called the primary loop. The outer loop controls the signals we are
primarily interested in. It is also possible to have a cascade control with more nested
loops.
An example of a feedback control system for temperature control of water in the
tank, is shown in Figure 6.2. In this application, the temperature of the water in the
Set-point
Stirrer
TT TC Temperature
controller
Steam
supply
Control valves
Cold
water
On/off
valve
Drain
Figure 6.2 Feedback control system for water temperature control.
408 Advanced Process Control Strategies
tank is maintained constant by adjusting the flow rate of steam. If a disturbance in the
steam-supply pressure occurs, the flow rate of steam changes, which in turn upsets the
temperature (controlled variable). Of course, once the temperature measuring device
senses the upset, it feeds the information back to the controller, which takes corrective
action so as to bring the temperature back to the set-point. Meanwhile, the disturbance
enters the process and upsets the controlled variable. One can visualize that, if the
variation in the steam-supply pressure is frequent, the controlled variable may not
remain at the set-point for very long.
A cascade control system can be implemented here to correct this problem, in
which, a second control loop can be added, as shown in Figure 6.3. In the presence of
steam-supply pressure fluctuations, the pressure sensor senses the changes, and the
pressure controller manipulates the steam valve so as to hold the down-stream pressure
constant. Thus, the effect of supply pressure fluctuations on the steam flow rate can be
eliminated. The performance of feedback control loop is greatly improved. Therefore, in
cascade control system, the output of one controller manipulates the set-point of another
controller.
Control valve
PT PC
Set-point
Stirrer
TT TC
Temperature
controller
Steam
supply
Control valves
Cold
water
On/off
valve
Drain
Figure 6.3 Cascade control system for water temperature control.
Cascade Control 409
A block diagram representation for the above water temperature control using
cascade control system, is shown in Figure 6.4. It may be noted from this figure that,
each controller is served by its own measurement device, but only one controller, the
primary or master controller, has an independent set-point, and only one controller, the
secondary or slave controller, has an output to the process. The secondary controller,
the manipulated variable, and its measurement device constitute the elements of the
inner or secondary loop. The outer loop consists of all the elements of the cascade
control system, including those of the inner loop.
Inner loop
Primary Secondary
controller controller
Temperature
+ + Pressure +
+ Temperature Pressure
control Process
controller controller
Set- + valve +
point
Pressure sensor
Temperature sensor
Temperature Sensor
Figure 6.4 Block diagram representation of a cascade control system for water temperature control.
To ensure that the cascade control systems function properly, it is necessary that
the dynamics of the inner loop be at least as fast as those of the outer loop, and
preferably faster. If the dynamics of the inner loop are much faster than those of the
outer loop, the inner controller will correct the effect of disturbances in that loop before
they have a chance to upset the controller variable. If this condition is not met, it is
generally impossible to tune the master controller satisfactorily. The commonly
encountered control loops, in order of decreasing speed, are flow, liquid-level, and
pressure, temperature composition.
There are two distinct advantages gained with a cascade control:
· The secondary controller can correct the disturbances affecting the secondary
variable, before a pronounced influence is felt by the secondary variable.
· Closing the control loop around the secondary part of the process reduces the
phase lag seen by the primary controller, resulting in increased speed of
response.
410 Advanced Process Control Strategies
Cascade control is particularly useful when disturbances are associated with the
manipulated variable or when the final control element exhibits nonlinear behaviour.
Cascade control systems are specially recommended in cases, where a valve (actuator)
manipulates flow. The gain of the valve is nonlinear, as it depends among others on the
pressure drop across the valve, which can change significantly during operation. A
secondary control loop with PI-controller can compensate for these gain changes
completely. Cascade control systems are applied more frequently in digital control
systems compared with analog control systems, as the additional effort of the secondary
controller is small.
%8
U )F
U=
)E
U)X
U) R
U)U
U?
&
U )E
U)X
U) R
U)U
U
)E
U)E
U)X
U) R
U) R
U)U
U (6.2)
%8
U )E
U)E
U)X
U) R
U) R
U
52
U )E
U)X
U) R
U)U
U
)E
U)E
U)X
U) R
U) R
U)U
U (6.3)
DD22 D11
D
Secondary loop
Gd2(s) Gd1(s)
MV(s) CV
E1(s) SP2(s) E2(s) CV2(s)
2(s) CV1(s)
+ +
Gc1(s) Gc2(s) Gv(s) Gp2(s) ¦ Gp1(s) ¦
SP1(s) –
– CVm2(s)
Gs2(s)
CVm1(s)
Gs1(s)
A key factor in cascade control is the relative dynamic responses of the secondary
and primary processes. Since the main reason for cascade is secondary disturbance, in
this section we are trying to evaluate the responses to secondary disturbances such as
step, sine and stochastic. For these simulation studies, the models for the sensors G1(s)
and valve Gv(s) were taken to the unity, and the dynamics of the plant models and
disturbance model are given as follows, with all times scaled so that the process models
have a common value of the fraction dead time. The relative dynamics between the
secondary and primary are defined by a variable h, which will be allowed to vary in the
following models (Tables 6.1 and 6.2).
G
IU
Secondary ) R
U PI-controller tuned accordingly
IU
G
U
Primary ) R
U PI-controller tuned accordingly
U
Process Control
G
I U
)R PI-controller tuned accordingly
=
IU?=
U ?
0.35
0.30
Relative cascade-to-single-loop
IAE of primary loop
0.25
0.20
0.15
0.10
0.05
0.00
0 5 10 15 20 25 30 35 40
Relative primary-to-secondary dynamics, h
Figure 6.6 Relative performance (IAEcase/IAEs1) of cascade and single-loop control for a step
disturbance in the secondary loop.
controller, which is often the case. However, many cascade control systems cannot
achieve such a remarkable improvement because the secondary loop is not so fast, and
some potential secondary loop dynamics are so slow as to prohibit cascade control.
Simple dynamic responses from cascade control are shown in Figure 6.7, for a step
disturbance in the secondary loop, D2(s) = 1/s at time = 10.
Primary
(CV1)
Secondary
(CV2)
Manipulated
(MV)
Time
(a) With h = 10.
Cascade Control 413
Primary
(CV1)
Secondary
(CV2)
Manipulated
(MV)
Time
(b) With h = 1.0
Figure 6.7 Performance of cascade control for a disturbance in the secondary loop.
The case with a very fast secondary demonstrates how quickly the secondary
controller attenuates the effect of the disturbance. The case of a much slower secondary
oscillations, which are more troublesome with the continuous disturbances experienced
in industrial plants, usually prohibit the use of PID controllers when h is less than about
3, although Figure 6.6 shows that some improvement in performance may be possible.
P
52K %8K
T52 Ç P
(6.4)
K
The standard deviation of the primary variable is plotted as a function of the relative
secondary/primary dynamics (h). Again, the faster the secondary, the better the
performance of the cascade. Dynamic responses for this system are given for h = 10 in
Figures 6.8(a), (b) and (c) for open-loop, single-loop and cascade control, respectively. It
is important to recognize that the results in Figure 6.6 are limited to the specific process
and disturbance studied; other disturbances, with different frequency components, would
give different results, although the general trend would be unchanged.
414 Advanced Process Control Strategies
CV1
CV2
0 Time 140
(a) Open-loop process
CV1
CV2
0 Time 140
CV1
CV2
0 Time 140
(c) Cascade control
Figure 6.8 Dynamic response for stochastic secondary disturbance with h = 10.
100
101
102 Single-loop
Amplitude ratio
^%8
LX ^
^ &
LX ^
103 Cascade
104
105
106
102 101 100 101 102
Frequency (rad/time)
Figure 6.9 Closed-loop frequency responses for single-loop and cascade control with hÿ = 10.
M Y1 Y2
G1(s) G2(s)
M Y1 Y2
G1(s) G2(s)
1
G1(s) S Y2SP
M Y1 Y2
G1 G2
1 1
Y1SP G1GC1 Y1 Y2
G2
1 + G1GC1
1
GC2 S Y2SP
Figure 6.10(b) shows the conventional feedback control system, where a single
controller senses the controlled variable Y2 and changes the manipulated variable M.
The closed-loop characteristic equation for this system is given as
1 + G1(s)Gc(s) G2(s) = 0 (6.5)
Figure 6.10(c) shows a series cascade system. There are now two controllers. The
secondary controller GC1 adjusts M to control the secondary variable Y1. The set-point
signal Y1SP to the GC1 controller comes from the primary; i.e. the output of the primary
controller GC2 is the set-point for the GC1 controller. The GC2 controller set-point is Y2SP.
The closed-loop characteristic equation for this system is not the same as that
given in Eq. (6.5). To derive it, let us first look at the secondary loop by itself. The
equation that describes this closed-loop system is given as
)%
; ;52 (6.6)
)%
1 + G1GC1 = 0 (6.7)
Next, we look at the controlled output Y2. Figure 6.10(d) shows the reduced block
diagram of the system in the conventional form. We can deduce the closed-loop
characteristic equation of this system by inspection.
)%
))% = 0 (6.8)
)%
However, let us derive it rigorously.
Y2 = G2Y1 (6.9)
Substituting Y1 from Eq. (6.6) gives
)%
; ) ;52 (6.10)
)%
M Y1
G1
Y2
G2
Y1SP
GC1 S
1
M Y1
G1
Y2
G2
1
YSP S GC2
Y1SP GC1 M Y2
G2
1 G1GC1
1
Y2SP
GC2 S
CV1(s)/SP2(s). During the first identification experiment (e.g. process reaction curve),
the primary controller is not in operation (i.e. the primary controller is in manual
or the cascade is open), which breaks the connection between the primary and
secondary controllers. The secondary is tuned in the conventional manner. This involves
a plant experiment, initial tuning calculation, and fine-tuning based on a closed-loop
dynamic response.
When the secondary has been satisfactorily tuned, the primary can be tuned. The
initial plant experiment perturbs the variable that the primary controller adjusts. In this
case, the secondary set-point is perturbed in a step for the process reaction curve. The
calculation of the initial tuning constants and the fine-tuning follows the conventional
procedures. Naturally, the secondary must be tuned satisfactorily before the primary
can be tuned.
The cascade control system uses more control equipmenttwo sensors and two
controllersthan the equivalent single-loop control system. Since the cascade control
requires all of this equipment to function properly, its reliability can be expected to be
lower than the equivalent single-loop system, although the slightly lower reliability is
not usually a deterrent to the use of cascade control. Since the cascade involves more
equipment, it costs slightly more than the single-loop control system. The increased
costs include a field sensor and transmission to the control room, a controller (whose
cost may be essentially zero if a digital computer system is used), and costs for
installation and documentation. These costs are not usually significant compared to the
benefits achieved through a properly designed cascade control strategy.
PI
controller
Control
reagent
Control
valve
1 1
L ( s)
5s + 1 ( s + 1) 3
+
+ § 1 · 1 +
M ( s) K c ¨1 ¸ Y ( s)
© U I s¹
( s + 1) 3
–
4
Curve I Zeigler-N ichols settings:
K c = 3.65, UI = 3.0
3
After obtaining satisfactory settings for controller with Kc = 2.84 and tI = 5.0, the
response of system to a step change in L(s) of 10 units has been obtained as shown in
Figure 6.14. It can be noted that the response is oscillatory and has a long tail. This
response illustrates the fact that the feedback control system does not begin to respond
until the load disturbance has worked its way through the forward loop and reaches the
measuring element, with the result that the composition can move far from the set-
point during the transient.
3
Curve I PI control with K c = 2.84, UI = 5.0
2 Curve II FF control with K c = 2.84,
–1
UI = 5.0, G F =
(5s + 1)
1
0 t
5 10 15 20 25
–1
Thus, if the change in the load disturbance L(s) can be detected as soon as it occurs
in the inlet stream, this information can be fed forward to a second controller. The
second controller adjusts the control valve in such a way as to prevent any change in
the outlet composition from the set-point. The controller that uses information fed
forward from the source of the load disturbance is called feedforward controller. After
modifying Fig. 6.12(b), a block diagram of combination of feedforward controller GF (s)
and the feedback controller GC (s) is created as shown in Fig. 6.15.
The transfer function of a feedforward controller is a relationship between a
manipulated variable and a disturbance variable (usually a load change) and is given as
§ /CPKRWNCVGF 8CTKCDNG · /
U
) (
U ¨© ¸¹ (6.20)
&KUVWTDCPEG ; EQPUVCPV .
U
To design a feedforward controller and known GF(s), we must know both GL(s) and GM(s).
As we have discussed, the objective of most feedforward controller is to hold the
controlled variable constant at its steady-state value. Therefore, the change in Y(s)
should be zero. The output response Y(s) can be written as
Y(s) = GL(s)GM(s)L(s) + GF(s)GM(s)L(s) + GC(s)GM(s)E(s) (6.21)
Feedforward Control 425
L ( s) GL 1 GM 1
5s + 1 (5s + 1)
3
G C (s)
Feedforward controller
+ +
M ( s)
+ E § 1 · +
GM 1
+ Y( s)
GC Kc ¨1 ¸ 3
UIs¹ ( s + 1)
– ©
Feedforward controller
In order to determine the transfer function of GF(s) that will prevent any change in the
control variable Y(s) from its set-point R(s), which is zero (i.e. setting Y(s) = 0) and
solving for the relationship between M(s) and L(s) give the feedforward controller
transfer function, which is given as
/
U § ).
U ·
{ )(
U ¨ ¸ (6.23)
.
U © ) /
U ¹
with Y(s) = 0 and R(s) = 0, the result will be
GF (s)= GL(s) (6.24)
For the example under consideration in Fig. 6.15,
).
U
U
Therefore, )(
U (6.25)
U
For the load response of the control system in Fig. 6.15, with GF(s) given by Eq. (6.25),
were obtained for a step change in L(s), there would be no deviation of Y(s) from the
set-point (i.e., perfect control). This response is shown as Curve II in Figure 6.14, which
of course is a horizontal line at Y(s) = 0. Rather than using GF(s) of Eq. (6.25) in
the feedforward controller, one can try using only the constant term of GF(s), that is
GF(s) = 1. This response gives Curve III in Figure 6.14, which has a very large
undershoot before the feedback controller returns Y(s) to the set-point. We try using
GF(s) = 0.5, to obtain Curve IV of Figure 6.14 in which case the undershoot is less, but
response is still unsatisfactory. As shown by Curves III and IV, omitting the dynamic
part of GF (s) can give very poor results.
426 Advanced Process Control Strategies
Perturbation (disturbance)
Predictive control
have PID components, and, in general, there is only one tuning parameter, the choice
of how fast the controlled variable (c) should move to the set-point. The symbol CI
represents the model inverse. This is because, in normal modelling approach, one sets
the input and evaluates the response of the output. By contrast, MBC determines what
input causes a set, predetermined output response. This is called the model inverse.
This open-loop model structure will work satisfactory if the model is perfect and there
are no constraints. However, nearly all controller models are not perfect, and therefore,
they require some form of feedback approach, as illustrated in Figure 6.17(b).
Set-point MBC m
controller, CI Process, P c
(model inverse)
Set-point
MBC m c
controller, CI Process, P
+
MBC
controller, CM
f
MBC
controller, CA
As shown in Figure 6.17(b), three functions of MBC, model (CM), inverse (CI) and
adjustment (CA), has been introduced. The difference between the model CM and the
process output (c) is monitored and used to adjust a controller feedback ( f ), which
usually is either a bias to the set-point or a model coefficient. While the common aim
of this strategy is to make intelligent control decisions, the choices of the model
structure, control law, adjusting mechanism, and adjustable parameters have led to
many variations in model-based controllers.
Model structure, therefore, is of primary importance. Choices for control law,
adjustment mechanism, and adjustable parameters, and for other features such as
428 Advanced Process Control Strategies
Modelling approaches
Following modelling approaches are dominant in a model-based control system:
· Transfer function models
· Time series models
· Nonlinear phenomenological models.
Transfer function models are based on open-loop Laplace transform descriptions of
the process response to step input, and have been the traditional control modelling
approach. Their familiarity and simplicity of the resulting MBC are advantages which
offset their limitations of linear and simplistic dynamic modelling.
Time series models represent the open-loop response of the process with a vector
of impulses, which are empirically determined. The precision of the modelled dynamic
process is an advantage, which offsets the limiting assumption that the process is linear,
as well as the need for using matrix/vector algebra. This is the most common modelling
approach in the industrial use of MBC.
Nonlinear phenomenological models are design-type simulators. For markedly
nonlinear or non-stationary process applications, their control intelligence can offset the
disadvantage of their modelling and computational complexity.
In all the above cases, the control is as good as the modelled representation of the
process. Initializing the controller with a model, which has been validated by process
testing, is the first and most critical implementation step. Based on the above
approaches, following model-based controllers are popular:
· Internal mode control (IMC)
· Model predictive control (MPC)
· Process-model-based control (PMBC)
In this case, CI is the realizable and stable part of the model inverse. For the first-
order plus dead-time (FOPDT), model inverse can be calculated as
UR
%+ U (6.28)
-R -
If the model is in error, then steady-state offset can be removed by biasing the set-point
by that error, as shown in Figure 6.18. To eliminate spikes in m due to both
measurement noise and set-point changes, the biased set-point is filtered (CF) to obtain
a reference trajectory (r). The inverse CI calculates m, which makes the idealized
process output (c) track the reference trajectory (r).
Controller
Set-
point + m c
CF CI Process
CM
Figure 6.19 illustrates the IMC function blocks for a process, which can be
modelled as an FOPDT. The functions are leads, lags, delays, summations and gains,
and can be configured in any distributed control system (DCS), single-loop controller
(SLC) or programmable logic controller (PLC). The single tuning parameter is tp, the
filter time constant. The lower values of tf make the controller more aggressive, and
the higher values make it gentler.
The major technical drawback of IMC is that the model is linear and stationary.
If either the process gain or the time constants change, the model-based calculations
become either too aggressive or sluggish and therefore, returning is required. This is
similar to PID control in which changes in process dynamics also require returning. If
the process dynamics change substantially, the model should be re-parameterized.
Since an exact inverse is not possible, the IMC approach segregates and eliminates
the aspects of the model transfer function that makes calculation of a realizable inverse
impossible. The first step is to separate the model into the product of the two factors
(6.29)
)O
U )O
U)O
U
430 Advanced Process Control Strategies
Controller
1 Up 1
s Process
Ufs 1 Kp K c
+
K pR R p s
U ps 1
where
)O
U = The non-invertible part has an inverse that is not casual or is unstable.
The inverse of this term includes predictions (eq s) and unstable poles
, with t < 0
U U
appearing in Gcp(s). The steady-state gain of this term must be 1.0.
)O
U = The invertible part has an inverse that is casual and stable, leading to a
realizable, stable controller. The steady-state gain of this term is the gain
of the process model Km.
Thus, the IMC controller eliminates all the elements in the process model Gm(s)
that lead to an unrealizable controller by taking the inverse of only the invertible factor
to give
)ER
U ËÍ)O
U ÛÝ (6.30)
This design equation ensures that the controller is realizable and that the system is
internally stable (at least with a perfect model), bit it does not explicitly guarantee that
the behaviour of the control system is acceptable. However, the performance of such
controllers is acceptable in many cases.
As discussed earlier, all realistic processes are modelled by transfer functions
having a denominator order greater than the numerator order. Thus, from Eq. (6.30),
the controller, which is the inverse of the process model, will have a numerator order
greater than the denominator order. This results in first or higher-order derivative in
the controller, which generally leads to unacceptable manipulated-variable behaviour
and, thus, poor performance and poor robustness when model errors occur.
Achieving a good control performance requires modifications that modulate the
manipulated-variable behaviour and increase the robustness of the system. The IMC
design method provides one feature to account for both of these concerns: filtering the
feedback signal. The filter can be placed before the controller, as shown in Figure 6.20,
Predictive Control Systems 431
D(s) Gd (s)
Gm(s)
MV(s)
Em(s)
so that the closed-loop transfer functions for the controlled and manipulated variables
become
%8
U )H
U)ER
U) R
U
(6.31)
52
U ) H
U)ER
U=) R
U )O
U?
/8
U ) H
U)ER
U
(6.32)
52
U ) H
U)ER
U=) R
U )O
U?
%8
U )F
U= ) H
U)ER
U) R
U?
(6.33)
&
U ) H
U)ER
U=) R
U )O
U?
/8
U )F
U)H
U)ER
U
(6.34)
&
U ) H
U)ER
U=) R
U )O
U?
Now, four desirable properties of the filter are determined as a basis for selecting
the filter algorithm:
(a) The steady-state value of the filter needs to be determined. Application of the
final value theorem to the closed-loop transfer function in Eq. (6.31) with the
requirement of zero steady-state offset yields
1442443
Ë '52 ) H
)ER
) R
Û
NKO %8
V NKO Ì Ü
V U
Í ) H
)ER
=) R
)O
? Ý
U (6.35)
0
È Ø
)H
U (6.36)
ÉÊ U H U ÚÙ
In this equation, the exponent N is selected to be large enough so that the product
Gf (s)Gcp(s) has a denominator polynomial in s of order at least as high as its numerator
polynomial. The filter time constant can be adjusted to satisfy the performance
specifications. Increasing the filter time constant modulates the manipulated-variable
fluctuations and increases robustness at the expense of larger deviations of the
controlled variable from its set-point during the transient response.
The predictive control system is difficult to implement in analog computer because
of the dead time in the model Gm(s), but it is straightforward with digital computers,
regardless of the model structure. The simple models here can be expressed in discrete
form. The IMC system in Figure 6.20 with a single filter will be considered, and the
dynamic model will be assumed to be of first-order with dead time. Thus, the predictive
control system equations in continuous form are:
%8O
U - O GZR
R O U
)O
U (6.37)
/8
U U O U
-O
)O
U (6.38)
U O U
/8
U U O U
) R
U)ER
U (6.39)
6R
U - O U H U
Predictive Control Systems 433
with CVm(s), the predictive value of the controlled variable, that is, the output from the
model Gm(s). The dynamic model can be simulated in discrete form, as
È 'V Ø Ë È 'V Ø Û
%8O P GZR É Ù %8O P - O Ì GZR É Ü /8P * (6.40)
Ê UO Ú Í Ê U O ÙÚ Ý
with Dt, the digital controller execution period and the dead time modelled as
G = qm /Dt, an integer value.
Note that the product of Gf (s)Gcp(s) can be implemented as one algorithm in this
case: a lead-lag transfer function for discrete form
U H 'V
U O 'V U O 'V
/8P /8P 6RP 6R (6.41)
U H 'V - O
U H 'V - O
U H 'V P
with Tp, the target (that is, the set-point as corrected by the feedback signal); the
difference between the measured and predicted values of the controlled variable.
In summary, the predictive control system execution at step n involves the
following:
· Calculate the predicted controlled variable, Eq. (6.40).
· Calculate the difference between the measured and model-predicted controlled
variables, (Em)n = CVn (CVm)n.
· Correct the set-point with the feedback signal, (Tp)n = SPn (Em)n.
· Calculate the manipulated-variable value, Eq. (6.41).
The results for a first-order with dead-time process model are given in Figure 6.21
for a good performance for a step disturbance. The filter tuning constant has a large
0.55
0.50
0.45
UH 0.40
R
U
0.35
0.30
0.25
0.20
0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70
R
Fraction dead time,
R U
Figure 6.21 Tuning correlation for a single-loop IMC disturbance response on a first-order with
dead-time process.
434 Advanced Process Control Strategies
value for small fraction dead times, although one might initially expect the opposite
correlation because systems that are easier to control require more filtering. The reason
for these results is the need to moderate the high-frequency variation in the
manipulated variable. Thus, the ratio of process-time-constant to filter-time-constant in
the lead-lag element in the controller should not be too large; these results indicate that
a reasonable ratio is around 2. A smaller filter performance, but the variability in the
manipulated variable would be unacceptably large for many applications.
In general, all parameters appearing in the IMC model and the control algorithm
could be tuned, but it is a common practice to use the best estimates for the dynamic
model. Thus, only the filter time constant (tf) is considered available for tuning.
Process output
Process
Model output
Model
Adaptation
mechanism
After a given time instant, the adaptation mechanism ensures that the absolute
value of the error considered in Figure 6.22 will be bounded by the smallest possible
limit, according to the level of noise and perturbations acting on the process. If no noise
or perturbations are involved, the error tends asymptotically towards zero.
In adaptive control system, the control itself includes a controller (including its
supporting hardware and instrumentation), a performance index, and an adjustment
algorithm, as shown in Figure 6.23. The overall system receives a command input r(t),
and controller sends a control input u(t) to the uncertain plant or process in order to
Controller
Environment
Performance
index Feedback
Adjustment
rule
effect an output signal c(t). The process output signal is fed back to both the controller,
for control input adjustment purposes, and the performance index, which measures how
well the controller is performing its tasks. The adjustment algorithm, using the control
input, process output, and plant performance measure, then adjusts controller
parameter values in order to improve system performance.
Some application scenarios also allow the measurement of parameters that
characterize the system operational environment. Adjustment rules in such cases may
incorporate such information if a need arises for a controller to discriminate between
the effects of plant dynamics and the effects of environmental changes on plant output
measurements. An alternate approach is for the designer to use the information to
incorporate a model of the environment as part of an uncertain, augmented plant during
the system design process.
Adaptive control system uses both a priori knowledge about the controlled process,
as well as the automatic incorporation of an acquired outline knowledge based on
observations of the process. Adaptive control is an extension and generalization of
classical off-line feedback control synthesis, by which all or some of the controller
parameters are adjusted automatically in response to online process observations or
output residuals (hence the name self-tuners for a subset of adaptive controllers).
Controller
parameters
Environment
Redesign
Feedback
Estimator
Plant parameter estimates
Command
input r (t)
Control Plant
input u(t) output c(t)
Controller Plant or
process
Controller Environment
parameters
Update law
e(t)
Reference Output error cm(t) +
model
Model output
The update law, in this case, is an error-driven recursive algorithm that calculates
incremental changes in controller parameter values directly from command input,
control input, plant output and desired model output data. These stepwise changes are
made so that the controller will minimize, in gradual fashion, the magnitude of an
output error regardless of whether the causes for error are plant dynamics or
environmentally based disturbances.
A directly-adapted closed-loop control system consequently may drift dangerously
close to instability before the output error increases to a size sufficient to effect any
corrective adjustments. This is not necessarily the case with indirect adaptation, since
a change in identified parameters may or may not be accompanied by corresponding
output error.
Model
cm(t)
Controller parameter
Tuning
mechanism
r(t)
u (t) c(t)
Controller Plant
c(t)
The overall control system can be conceived as if it consists of two loops: an inner
loop, the ordinary control system, composed of the plant and the controller; and an outer
loop that comprises the parameter-adjustment, or tuning, mechanism. The controller
parameters are adjusted by the outer loop so as to make plant output c(t) close to the
model output, cm(t).
440 Advanced Process Control Strategies
Figure 6.27 illustrates another example and a different way to adjust the
parameters of the controller. We postulate a reference model, which tells us how the
controlled process output ideally should respond to the command signal (set-point). The
model output is compared to the actual process output. The difference (error em)
between the two outputs is used through a computer to adjust the parameters of the
controller in such a way as to minimize the integral square error:
V
minimize +5' ³ =G O
V? FV (6.42)
The model chosen by the control designer for reference purposes is to a certain
extent arbitrary. Most often a rather simple linear model is used.
Reference
model
Outer loop
em
New values for
controller parameters Adaptation
mechanism
e Process
Controller (plant)
+
Controlled
output
Inner loop
We notice that the model-reference adaptive control is composed of two loops. The
inner loop is an ordinary feedback control loop. The outer loop includes the adaptation
mechanism and also looks like a feedback loop. The model output plays the role of the
set-point while the process output is the actual measurement. There is a comparator
whose output (error em) is the input to the adjustment mechanism. The key problem is
to design the adaptation mechanism in such a way as to provide a stable system (i.e.
bring the error em to zero). This is not a trivial problem, but an extensive discussion on
how it can be solved, which requires mathematical analysis.
parameters using some estimation techniques and the second step is to online design
the controller using the estimated unknown parameters. The inner loop consists of
ordinary control system and is composed of the plant and the controller. The outer loop
consists of the parameter-adjustment mechanism. The later, in turn, is made of a
recursive identifier and a design block. The identifier updates an estimate of the
unknown plant parameters according to which the controller parameters are tuned
online by the design block. A block diagram of STAC is shown in Figure 6.28.
Tuning mechanism
Recursive
Design identifier
Plant
parameters
Controller
parameter
r(t)
u(t) c(t)
Controller Plant
c(t)
The most prevalent adaptive or self-tuning controllers (STAC) in active use today
are probably those derived from the ubiquitous fixed-parameter PID controllers
commonly found in process control applications. PID controllers, along with their
relatives, the PI and PD controllers, are classic examples of very simple low-order
controllers that have the capability of providing satisfactory performance when tuned for
any given operating point. This is well tailored for process control, where the local
operating point of the controlled sub-process varies only a small amount as part of a
much larger, multi-faceted process that runs for long periods of time.
The need of self-tuning is purely economic in nature and is usually performed on
a periodic basis (as opposed to the continual searching of the direct adaptive controllers
described above). Automated calibration of process control greatly reduces the
manpower requirements necessary to maintain all of the local controllers at their
respective optimum settings. Automation also allows for faster calibration to optimal
performance after any set-point verifications that may arise from commanded changes
in the process.
Another example of a self-tuning regulator (STR) is shown in a block diagram form
in Figure 6.29. It has again two loops. The inner loop consists of the process and an
ordinary linear feedback controller. The outer loop is used to adjust the parameters
of the feedback controller and is composed of (a) a recursive parameter estimator and
(b) an adjustment mechanism for the controller parameters.
442 Advanced Process Control Strategies
Parameter
adjustment
mechanism
m y
Parameter
New values Outer loop estimator
of controller
parameter
Manipulated y
Set-point variable Process
Controller
m (plant) Controlled
+
output
Inner loop
The parameter estimator assumes a simple linear model for the process:
[
U - R GZR
VF U
(6.43)
O
U UU
Thus, using measured values for the manipulated variable (m) and the controlled output
(y), it estimates the values of the parameters Kp, t and td, employing a least-squares
estimation technique. Once the values of the process parameters Kp, t and td are
known, the adjustment mechanism can find the best values for the controller
parameters using various design criteria, such as: phase or gain margins and integral
of the squared error, etc.
Both the parameter estimator and adjustment mechanism require involved
computations. For this reason, the STR can be implemented only through the use of a
digital computers.
The first type of a self-tuning adaptive controller, called the self-tuning regulator
(STR), was proposed in 1971 by Astrom and Wittenmark and has since been
implemented in several industrial applications. Subsequent modifications, the self-tuning
controller (STC) and the generalized predictive controller (GPC), have also been used
with industrial processes. These models are used on a differential equation (digital)
model of the process.
Gc(s) SPi(s)
Gpi(s) S CVi(s)
Gdi(s)
MV(s)
D(s)
Gdt(s)
Gpt(s) S CVt(s)
Figure 6.31 Block diagram of feedback control system with a true controlled variable, CVt(s) and
an inferential controlled variable, CVi(s).
cases, the inferential variable is not as accurate as an on-stream sensor of the true
variable. Also, the approximate relationship used for the inferential variable has a
limited range, beyond which the inferential variable might not be satisfactory. It is
important to remember that zero steady-state offset for the true variable is possible only
when it is measured, perhaps infrequently, and used in the control system to adjust the
set-point of the inferential controller, SPi(s).
Figure 6.31 can be used to determine the relationship necessary for a good
inferential control. First, the response of the true controlled variable to a disturbance
can be evaluated.
CVi(s) = Gdt(s)D(s) + Gpt(s)MV(s)
14243
)RV
U)E
U)FK
U (6.44)
)FV
U &
U &
U
)RK
U)E
U
A key goal of the control system is to maintain zero steady-state deviation in
the controlled variable. This can be evaluated by applying the final value theorem to
Eq. (6.32) with a step disturbance and PI feedback controller as
- RV - FK - E 6+
NKO %8
V NKO %8
U - FV '& '& (6.45)
V U - RK - E 6+
Thus, the criterion for a perfect steady-state inferential control in response to a
disturbance is that
- FV - FK
- RV - RK
446 Advanced Process Control Strategies
As the process relationships deviate from this criterion, the performance of the
inferential controller degrades. Thus, an important engineering decision is the selection
of a proper inferential measured or calculated variable.
Another example of inferential control is illustrated in Figure 6.32. Figure 6.32(a)
shows a block diagram of the process with one unmeasured controlled output (y) and one
secondary measured output (z). The manipulated variable m and the disturbance (d)
affect both the outputs. The disturbance is considered to be unmeasured. The transfer
functions in the block diagram indicate the relationships between the various inputs and
outputs, and they are assumed to be perfectly known.
Unmeasured
d
Process disturbance
Gd1 Gd2
Secondary
z measurement
Gp2 S
(a) Process
Controller y
Set-point m Process of
Gc
ySP + Figure 6.32(a) z
Gd 1
G p1 G p2
Gd 2
Estimate of
unmeasured
output, y
Gd1 z
S
Gd 2
Estimator
(b) Corresponding inferential control system
Figure 6.32 Example of inferential control system.
Intelligent Control 447
From Figure 6.32(a), we can easily derive the following inputoutput relationships:
[ ) R O )F F (6.46)
\ ) R O )F F (6.47)
From Eq. (6.46), we can solve with respect to F and find the following estimates of the
unmeasured disturbance:
) R
F \ O (6.48)
)F )F
È ) Ø )
[ É ) R F ) R Ù O F \ (6.49)
Ê )F Ú )F
Equation (6.49) provides the needed estimator which relates the unmeasured controlled
output to measured quantities like m and z. Figure 6.32(b) shows the structure of the
resulting inferential control system. Notice that the estimated value of the unmeasured
output plays the same role as a regular measured output. That is, it is compared to the
desired set-point and the difference is the actuating signal for the controller.
It is important to emphasize that the success of an inferential control scheme
depends heavily on the availability of a good estimator, which in turn depends on how
well we know the process. Thus, if the process transfer functions Gp1, Gp2, Gd1 and Gd2
are perfectly known, a perfect estimator can be constructed and consequently we will
have perfect control. When the process transfer functions are only approximately known
(which is usually the case), the inferential scheme provides control of varying quality,
depending on how well the process is known.
In chemical process control, the variable that is most commonly inferred from
secondary measurements is composition. This is due to the lack of reliable, rapid, and
economical measuring devices for a wide spectrum of chemical systems. Thus,
inferential control may be used for the control of chemical reactors, distillation columns,
and other mass transfer operations such as driers and absorbers. Temperature is the
most common secondary measurement used to infer the unmeasured composition.
Expert systems
Expert systems are computer software systems that mimic the tasks routinely carried
out by experts. Expert systems mimic human judgement using computer programs by
applying substantial knowledge in specific areas or expertise to solve finite, well-defined
problems. These computer programs contain human expertise (called heuristic
knowledge) obtained either directly from human experts or indirectly from books,
publications, codes, standards, or databases, as well as general and specialized
knowledge that pertains to the specific situation. Expert systems have the ability to
reason using formal logic, to seek information from variety of sources including
Intelligent Control 449
databases and the user, plant operator, and to interact with conventional programs to
carry out a variety of tasks including sophisticated computations.
Usually, the domain of expertise of these systems is restricted to a well-defined
area. Unlike humans, these systems lack common sense and hence cannot extrapolate
their knowledge much beyond their narrow domain of expertise. They encode the
relevant knowledge about the problem domain in a form, which is suitable for
manipulation. The structure of the expert system separates the knowledge from the
mechanism for manipulating the knowledge. This has significant advantages because it
makes the system easier to maintain and update. It is this separation of the knowledge
base from the reasoning process that makes expert systems different from other
software packages. Figure 6.33 shows the typical structure of an expert system.
The principal components of an expert system are:
· The inference engine
· The knowledge base
· The interface between the human and the expert system.
Reasoning mechanism
(inference engine)
Knowledge-base rules,
facts, frames, objects
The inference engine gathers the information needed from the knowledge base,
from associated databases, or from the user. It guides the search process in accordance
with a programmed strategy; uses rules of logic to draw inferences or conclusions for
the process involved; and presents the conclusions (where warranted) with explanations
or bases for the conclusions.
The knowledge base consists of information stored in retrievable form in the
computer, usually in the form of rules or frames. The correctness and completeness of
the information within the knowledge base is the key to obtaining correct results or
solutions using expert systems.
The interface between the human and the expert system must translate user input
into the computer language, and it should present conclusions and explanations to the
450 Advanced Process Control Strategies
user in a written or graphical form. It should also include an editor to assist in adding
to or changing the knowledge base.
Figure 6.34 illustrates the expert control system (ECS). Real-time expert systems
are being used for process control applications. It requires additional features such as
the ability to include the time element in the reasoning process.
Expert controller
Disturbance
Inference
engine
Output, c
Plant
Reference Command
input, r input, u
Knowledge
base
Expert system shells are currently available in the market, tailored to meet the
requirements of process control applications. It requires close integration with the
underlying distributed control system (DCS) for exchange of sensor data and
communication with the plant operator. Figure 6.35 shows the structure of a real-time
expert system. Examples of real-time expert system shells include G2 (from Gensym
Corporation), and Expert 90 (from Bailey Controls).
Inference
Real-time expert system
User engine
interface
Knowledge base Database
Sensor Control
data actions
Operator
Process
Figure 6.35 Structure of a real-time expert system.
Intelligent Control 451
Expert system applications to process control can be divided into two areas: control
system design and real-time process control. In an expert system for design, the goal is
to capture the expertise of the design engineer to reduce the design time and to produce
an optional design. The user or operator will interact with the program in an interview
mode, supplying relevant data on the specific problem and the expert system will
suggest an appropriate control configuration. In the real-time process control, the expert
systems are being applied for batch process control, distributed control, fault diagnosis,
and statistical process control.
A number of characteristics of expert systems are unique and generally
advantageous. Some of the advantages of implementing an expert system in computer-
aided process control are:
· Experts need not be present for a consultation, it may be delivered to remote
locations where expertise may not be available.
· Expert systems do not suffer from some of the shortcomings of human beings,
e.g. tiredness or carelessness as the workload increases.
· The techniques inherent in the methodology of expert systems minimize the
recollection of information by requesting only relevant data from the user or
appropriate databases.
· Expert knowledge is saved and readily available because the expert system can
become a repository for undocumented knowledge that might otherwise be lost.
Expert systems have following disadvantages also that affect their use:
· They must be kept up to date as the condition changes.
· Results are very dependent on the correctness of the knowledge incorporated
into expert systems.
· They do not benefit from experience except through updating of the knowledge
base (based on human experience).
· Expert systems are unable to solve problems outside their domain of expertise
and in many cases are unable to detect the limitations of their domain.
Neural networks
Neural networks, like expert systems, robotics, natural language processing, machine
vision and fuzzy logic, are parts of an overall technology umbrella of artificial
intelligence (AI). In process control applications, the difficulties that arise in the control
of complex systems, are mainly due to system complexity, and the presence of
nonlinearities and uncertainties. Such systems are characterized by poor models, high
dimensionality of the decision space, distributed sensors and decision makers, high noise
levels, multiple sub-systems, levels, time-scales and performance criteria, complex
information patterns, overwhelming amounts of data, and stringent performance
requirements. Even modern control theory is not in a position to cope with these
situations. To address these problems in a systematic way, a number of methods have
been proposed that are known as an intelligent control theories or methodologies.
Amongst these methods, perhaps the most attractive approach is that of artificial neural
networks (ANNs).
452 Advanced Process Control Strategies
x2 Sum Transfer
w2 Output
path y3
yn
Nucleus wn
xn
Axon
Synapses
(a) Neuron (b) Processing element
Figure 6.36 Neural net analogy with human brain.
The learning rules used to train the networks can generally be classified as
supervised or unsupervised. Supervised learning rules require the desired network
output to be available and they adapt the weights so that the output error is reduced.
Unsupervised learning is used to organize the networks structure based only on the
training inputs presented to the network. ANNs learn from experience, generalize from
previous examples to new ones and abstract essential characteristics from input
containing irrelevant data. ANNs have other features such as high dimensionality,
adaptability, self-organization, and variety of connections among neutrals. Therefore,
ANNs can use nonlinearity, learning and generalization capabilities for applications to
advanced process control. Figure 6.37 shows a three-layer fully connected neural
network.
Advantages of neural network include:
· Good fit for nonlinear models
· Ability to adapt, generalize and extrapolate results
· Speed of execution in recall mode
· Ease of maintenance.
Intelligent Control 453
Manipulated variables
Steam flow Reflux flow
Output layer
(2 nodes)
Hidden layer
(4 nodes)
cm(t)
S
Neural network
training
+
r(t)
Plant
c(t)
Figure 6.38 Plant modelling using neural network.
Critic
r(t) e u c(t)
Neural network
Plant
controller
+
of small changes as the controller is operating normally. Gradually, the NN will take
over the control action because it makes better choices of the control signal than the
controller does.
Another configuration suggested is shown in Figure 6.40(b), in which the NN is
used to make the performance of the system appear linear to the controller. For
example, if the plant is a robot, which is represented by a highly nonlinear model, then
the controller would implement the computed torque algorithm to provide the desired
acceleration signal to the NN. This will, in turn, perform the inverse dynamic evaluation
Intelligent Control 455
Neural
network
+
+ +
Controller Plant
+
Neural Plant
Controller network
and determine the force needed to yield the desired acceleration. This approach is
different than feed-forward architecture, and requires the NN to be pre-trained to learn
the true dynamics of the plant off-line. Because the NN makes the closed-loop system
appear to be linear to the controller, this configuration is called linearization
architecture.
Fuzzy logic
Many techniques have been developed and used for the purpose of process control, such
as feedback control, transfer functions, state-space time-domain, optimal control,
adaptive control, robust control etc. Different important concepts such as root locus,
Bode plot, phase-margin, gain-margin, eigenvalues, eigenvectors, etc., have been
imported from different areas or developed in the control field. Most of these techniques
rely more on system mathematical models in their design process. These mathematical
model based techniques are difficult to represent accurately the system dynamics.
Therefore, the techniques such as set-point control, sliding mode control, neural
network control, fuzzy logic control, that rely less on mathematical model of the system,
but more on the designers knowledge of the actual system are more often resorted to.
In 1985, Professor L.A. Zadeh of the University of California, Berkeley, presented
his paper Outlining fuzzy set theory. Fuzzy control provides a formal methodology for
456 Advanced Process Control Strategies
Logical sum
Rule-1
Antecedent Consequent
block block
logical product logical product
Rule-2
Antecedent Consequent
block block
logical product logical product
...
Rule-N
Antecedent Consequent
block block
logical product logical product
(the antecedent block) and a conclusion (the consequent block). The conclusion
is satisfied when the conditions are satisfied. The inference process proceeds
from the conditions to the conclusion, and then to the logical sum.
(b) The defuzzifier operation, which unifies the results of the rule processes and
calculates a final value of the output. Defuzzification allows fuzzy-based controls
to be interfaced to all types of output devices commonly available today.
The main characteristics of fuzzy control are, time varying, nonlinear behaviour
and the low frequency of its measurements. Fuzzy logic is very human concept,
potentially applicable to a wide range of processes and tasks that require human
intuition and experience. Fuzzy logic control can be applied by means of software,
dedicated controllers, or fuzzy microprocessors embedded in digital products. A block
diagram of fuzzy logic control system is shown in Figure 6.42.
Disturbance
Reference Command
input, r input, u Output, c
Fuzzy Plant or
e controller process
Fuzzy logic control is basically an intelligent control technique and has been shown
to yield promising results for many applications that are difficult to handle by
conventional techniques. Fuzzy logic control has today found application in computer-
aided process control.
Fuzzy logic development tools. As more industries are introducing this technology,
there is a rising demand for software and hardware tools to facilitate fuzzy logic
application development. In accordance with this rising demand, a number of fuzzy logic
development tools have been created, some of which are described now.
1. CubiCalc (from HyperLogic Corp.) runs under Microsoft Windows. Working
with CubiCalc evolves around an expert system paradigm (i.e. the user builds
a rule base, which can then be queried) rather than a control system paradigm
(i.e. the user builds a fuzzy processing box with input and output data paths).
This expert system paradigm and the highly intuitive user interface make the
software easily comprehensible to almost every user, regardless of technical
background. This is a tool designed for diverse users and equally diverse
applications. Because of its low cost and ease of use, CubiCalc is widely used
fuzzy logic software.
458 Advanced Process Control Strategies
Figure 6.43 illustrates an example of SPC control using computer for a heat
exchanger. In this example, an SPC supervisor operates with a conventional PID
controller in a control loop. The supervisor uses SPC techniques to analyze the process
outputs and to decide whether any deviations from the set-point can be attributed to
noise. Without the supervisor, the PID controller responds to both random variations
(noise) and to assignable deviations. With the supervisor, it acts when there is a
deviation of the process mean from the set-point and reduces the effect of controller
action to random process noise.
Cold water
Cold water supply Heat exchanger Heated water
Condensate
Control
valve TT
Electro-pneumatic
valve positioner
Steam
supply
DAC ADC
ADC
(a) Physical heat exchanger system
SPC
supervisor
PID Heat
+ control exchanger
As shown in Figure 6.43, the heat exchanger is placed under computer control and
a statistical supervisor, which performs a series of statistical tests to oversee the control
action. Some tests are designed to filter out the random noise variation and some to
detect slow process drift. In the absence of process drift or set-point changes, the
Algorithms for Processes with Dead Time 461
controller should ideally take no corrective action. When all tests are passed then this
implies that the process is under satisfactory control and no change of controller action
is required. In this case, the SPC supervisor maintains a constant control action. When
the test indicating a large offset fail, PID control is implemented. When the test
indicating a slow drift in the process mean fail, then integral-only action is used to bring
the process to set-point.
The implementation of real-time SPC places increased demands on instrumenta-
tion, communication networks and computer technology. Developments in microchip
technology offer improved accuracy and stability for primary sensors. The use of digital
transmitters eliminates the error contribution of analog-to-digital conversion (ADC),
Microprocessor-based smart transmitters contribute to better accuracy and higher
reliability by their improved rangeability. Automatic pressure and temperature
compensation, remote calibration and self-diagnostics also contribute to better data
quality.
The ability to interface a personal computer (PC) with the control system also
increases the feasibility of adding real-time SPC to existing plants. Intelligence at the
I/O interface levels enables routine operating functions to be handled locally, while
conserving a higher-level processing capability for SPC-type applications.
L Process
r (t) e
G (s) m K
K pp bb T s c
Gcc(s) e (dqds)
exp
11
UWss
r (t) e m Kp b c
Gc(s) K p
Us 1
e (
exp T dqs s)
d
Ws 1
of qd. Since there is no delay in the feedback signal (b), the response of the system would
be greatly improved. The scheme, of course, cannot be implemented, because b is an un-
measurable (fictitious) signal.
Now, a model of the process is developed and a manipulated variable (m) is applied
to the model, as shown in Figure 6.45. If the model were perfect and disturbance,
L = 0 (i.e. no load disturbances are present), then the controlled variable c will become
L Process
e1 m c
r (t) el m Kp bb c
GGcc(s) K
e T(d qs ds)
exp
m
p
Us 1
Ws 1
em
bm cm
Kp bm cm em
K Tdq
p
Us 1 e (
exp s s)
d
Ws 1
Model em = c c m
equal to the error cm and em = c cm = 0. The arrangement shown in Figure 6.45 reveals
that although the fictitious process variable b is unavailable, the value of bm can be
derived which will be equal to b unless modelling errors or load upsets are present. It
is used as feedback signal, as shown in Figure 6.46. The difference (c cm) is the error
(em), which arises because of modelling errors or load upsets.
The arrangement shown in Figure 6.46, controls the model well but perhaps not
the process, if load occurs or if the model is not accurate. To compensate for these
Algorithms for Processes with Dead Time 463
L Process
r (t) e1 m Kp b c
G (s) K
Gcc(s) p
Us 1
e (
exp T dqs s)
d
Ws 1
Kp
K bm cm em
Tdq
Us 1
p
e(
exp s s)
d
Ws 1
Model em = c cm
errors, a second feedback loop is implemented using em, as shown in Figure 6.47. This
is called the Smith predictor control strategy. The Gc(s) is a conventional PI or PID
controller, which can be tuned much more tightly because of the elimination of dead
time from the loop.
L Process
r (t) e1 e2 m
G (s)
Gcc(s) KK pp b T s c
Us 1
e (d qds)
exp
Ws 1
Kp bm cm em
K
Us 1
p
e (
exp T d qs s)
d
Ws 1
Model em = c cm
Thus, Smith reasoned that eliminating dead time from the control loop would be
beneficial, which is certainly true but not possible via a feedback controller, only
physical changes in the process can affect the feedback dead time. Therefore, Smith
suggested that controlling a model of the process, without the dead time (or other non-
invertible element), would provide a better calculation of the manipulated variable to be
implemented in the true process. Smith had retained the conventional PI control
algorithm and relied on the general predictive structure in which the controller was
calculated by the elements in the dashed box, as shown in Figure 6.48. These elements
perform the function of the predictive control algorithm, Gcp(s).
The system, as shown in Figure 6.48, consists of a feedback PI algorithm (Gc) that
controls a simulated process, Gm ( s) , which is easier to control than the real process.
464 Advanced Process Control Strategies
D(s) Gd(s)
Gcp(s)
+
SP(s) + CV(s)
Gc(s) Gp(s) S
MV(s) +
Gmm(s)
(s) Gm(s)
Em(s)
Gm ( s ) has the same meaning here as for IMC predictive control in Eq. (6.29), and the
absence of dead time or inverse response in the model Gm ( s) allows much more
aggressive control of the model than that of the true plant.
The calculated manipulated variable resulting from controlling the model is
implemented in the true process, which could yield good control as long as the model
were perfect. Naturally, the model will not be perfect, and some form of feedback is
required to achieve a zero steady-state offset. Smith had recognized the value of the
predictive structure and, as shown in Figure 6.48, proposed correcting the model with
the difference between the measured and the predicted controlled variables. Note that
the prediction is determined using the complete linear dynamic model, Gm(s), including
any non-invertible dynamics. The feedback signal, Gm(s), can be interpreted as a
correction to the model Gm ( s) .
The closed-loop transfer function of the system in Figure 6.47 is
%8
U )E
U) R
U
52
U
)E)O
U )E
U=) R
U )O
U? (6.50)
If the models were perfect, the characteristic equation would not contain a dead
time, because Gm(s) and Gp(s) would cancel. Thus, for the case with a perfect model, the
characteristic equation involves only the expression [1 + Gc(s) Gm ( s) ], which is easier to
control and allows a more aggressive adjustment of the manipulated variable. Naturally,
the true process is never known exactly, and the actual behaviour and stability depend
on all terms without cancellation. Application of the final value theorem to Eq. (6.50),
for a step change in the set-point and a PI algorithm for the controller, gives
'52 ÎÑ )E
U) R
U ÞÑ
NKO %8
V NKO Ï ß (6.51)
V U U ÐÑ )E
U)O
U )E
U=) R
U )O
U?àÑ
Gp(0) = Kp =
and Gm(0) = Km = )O
-O
Algorithms for Processes with Dead Time 465
and
È Ø
- R - E É
Ê 6+ U ÙÚ
NKO %8
V '52 '52 (6.52)
V È Ø È Ø
- E É
6 U Ù - O - E É
6 Ù - R - O
Ê + Ú Ê + UÚ
Thus, zero steady-state offset for a step input with Smith predictor control does not
require a perfect model. It requires only that the steady-state gains for the two models
) and that the controller algorithm, G (s), have an integral mode.
be identical (Km = - O c
Again, the performance and robustness of the Smith predictor control system
depend on the controller tuning. It is cautioned that the PI controller in the Smith
predictor should not be tuned using correlations, which are developed for the feedback
dynamics. The purpose of the PI controller is to calculate an approximate inverse
rapidly, as demonstrated by the following equation:
/8
U )E
U
)ER
for large Gc(s) (6.53)
52
U 'O
U )E
U)O
U )O
U
Ë Û
- E Ì
/8
U )E
U Í 6
U ÜÝ
(6.54)
6R
U )O
U)E
U Ë Û -O
- E Ì
Í 6
U ÜÝ U O U
IMC controller:
/8
U U O U
(6.55)
6R
U - O U H U
Thus, the tuning correlations in Figure 6.21 along with Eqs. (6.56) can be used to
estimate initial tuning for the Smith predictor with a first-order with dead-time process
model.
The Smith predictor is easily programmed in a digital computer system. The digital
form of a first-order with dead-time model, the digital models are programmed using
Eq. (6.40) for G(s) and the same equation with no dead time, G = 0, for )O
U.
466 Advanced Process Control Strategies
Set-point + +
Digital + Gp(s) c
Gh0(s)
controller exp (qds)
t t
Analytical
predictor t
mathematically rigorous manner. The key thing to remember is that optimal control is
primarily about transitions between steady-state operations and is only applicable to
certain classes of process operations.
Batch processing operations by nature dynamic or transient activities, where most
of the states given by temperatures, compositions or pressures, are changing
continuously with time. How one moves the system from some initial state to its final
state is the subject of optimal control. There are processes which produce a range of
products with varying specifications because of market demand, such as paper
production, glass plates, etc. Here optical control plays important role in changing the
process in an optimal way to move from one product to the next minimizing wastage
or recycled material. Some of the examples of an optimal control are:
(a) Tubular reactor temperature profile control, where the temperature profile
maintained along a reactor directly affects the concentration of products out of
the reactor,
(b) Optimal transitions for an evaporator system wherein it is desired to produce
number of concentrates from a dilute feed of a sugar industry or a fruit juice
production etc.
The key elements to optimal control are as follows:
(a) An objective in terms of performance measure or objective function (J) to be
achieved which sets out the cost for the desired transition.
(b) A set of dynamic constraints which must be satisfied as the system moves
between states. These are typically the differential equations and algebraic
constraints arising from dynamic mass and energy balances plus any
constitutive equations.
(c) A set of control u which may be manipulated in order to drive the system along
the optimal trajectory. This could include flow rates, temperatures or pressures
in process systems.
(d) A set of operating constraints which may include constraints on the state
variables plus the manipulated variables.
(e) A means of computing the optimal pathway and the control systems.
Mathematically, we can look at each element above and describe the problem as
optimizing (maximizing or minimizing) the performance index as follows:
V
J = f1(x, u, tf) + Ô
V
f2(x, u, tf ) dt (6.57)
and we want to compute the controls u(t), which minimizes J subject to the state
equations and other constraints.
This is quite a common type of problem in process industries. It often occurs when
productivity is the key issue and batches of material need to be pushed through
the system at maximum rate. This type of optimal control problems takes no account
of the off-specification material produced or the energy which is expended in moving
between t0 and tf.
Ô Ç B W FV
VH
, K K (6.63)
V
K
where ui(t) is the ith control and ai is some cost associated with that control, e.g. the
cost of steam in dollars per tonne. We can sum overall relevant controls, as shown
Optimal Control 469
above. Typical of this type of application could be to minimize energy consumption whilst
moving the product quality from one level to another.
VH
, Ô
V
eTWe dt (6.64)
where
e(t) ^ Z
V Z
V^
Z
V = Desired trajectory
x(t) = Actual trajectory
W = Weighting matrix.
This type of control could apply to a reactor, which has a specified concentration
or temperature profile over the cycle.
ui
0
t tf
Figure 6.50 Discretized control variable.
The optimization procedure seeks to find the values of uki for all k and I which
gives an optimal value for the performance index J. To do this, following are required:
(a) An expression of the performance index (J)
(b) Initial values for all uki
(c) An optimization algorithm for nonlinear constrained problems
(d) An integration routine to solve the state equations.
Now the problem for the unknown values uki can be solved as shown in
Figure 6.51. In this solution technique, it is important to employ an efficient differential
equation solver and to ensure that the equations are solved accurately. Otherwise,
sloppy solutions will affect the optimizers ability to find an optimal solution. METLAB
provides routines for both integration and optimization in the respective toolboxes.
An alternative approach to discretizing the controls is called control para-
metrization, where we assume that the control u(t) is in the form of a polynomial.
That is,
Optimal Control 471
Define J
Initialize u ki
Integrate state
equations
x f (u , x )
Evaluate J
index
Optimizer
No
Update u ki
values
W
V C C V C V ! C V
M
M (6.67)
We can then select an order k (usually < 4 or 5) and solve for the unknown
coefficients ak for each control u. It uses the same approach shown in Figure 6.51. This
nonlinear programming technique can be used to solve most optimal control problems
including those with state constraints.
repeatedly until the trajectories stabilize. The final iteration leads to the optimal
trajectories for x and u.
ILLUSTRATIVE EXAMPLES
Example 6.1 Design the secondary (slave) and primary (master) loops for a series
cascade control system shown in Figure 6.52.
1 11
M 1 Y1 Y2
§ 1 §¨ 1 s + 1··¸ (s 1)
¨ s© 2 1¸¹ s 1 5s5s 11
©2 ¹
1 1
K1 S K2 S
Y1SP Y2SP
-E (6.68)
=
U ? U U
or
U U U - E (6.69)
Illustrative Examples 473
To solve for the ultimate gain (Ku) and ultimate frequency (wu), we substitute iw for s:
K X X K X - E
or
U K - (6.72)
To have a damping coefficient of (0.707), the roots must lie on a radial line whose
angle with the real axis is cos1 (0.707) = 45°. On this line the real and imaginary parts
of the roots are equal. So for a closed-loop damping coefficient of 0.707
- or - (6.73)
Now, the closed-loop relationship between Y1 and Y1SP is
) )% >
U @
U
; ;52 ;
) )% 52 (6.74)
>
U @ U
or
; ;52 (6.75)
U U
Designing the primary (master) loop. The closed-loop characteristic equation for
the master loop is
))% -
) )% (6.76)
))% U U U
474 Advanced Process Control Strategies
or
U U U - (6.77)
Solving for the ultimate gain (Ku) and ultimate frequency (wu) by substituting iw for
s, gives
Ku = 30.8 wn = = 2.26
It is useful to compare these values with those found for a single conventional
control loop, Ku = 19.8 and wu = 1.61. We can see that cascade control results in a higher
controller gain and a smaller closed-loop time constant (the reciprocal of the frequency).
Therefore, the system will show a faster response with cascade control than with a
single-loop.
0.9e 2 s 0.2e 2 s
Gd 1 Gd 2
70s 1 60s 1
Reflux
m ratio 1.2e 1s y Overhead
G p1 6 composition
30s 1
1 z Temperature
Gp2 6
20s 1 of top tray
1.2e s 4.5(60s 1)
30s 1 (70s 1)(20s 1)
60 s 1 z
6 4 .5
70 s 1
Estimate of
unmeasured Estimator
output, [
(b) Corresponding inferential control system
Figure 6.53 Example of inferential control system.
GU GU
[
U F
U O
U
U U
and
GU GU
\
U
F
U O
U
U U
Having developed the four process-transfer functions it is easy to design the
inferential control system, as shown in Figure 6.53(b).
Example 6.3 Design a controller for the three-tank mixing process shown in
Figure 6.54 by applying the IMC procedure.
The IMC controller design requires a transfer function model of the process. The
linearized third-order model will be used here. In this case,
xA0
B
xxA1A1
xA2
A
xA3
AC
-O
)O
U
)O
U , )O
U
U U
U
Thus, the model can be inverted directly to give
)ER
U
ËÍ)O
U ÛÝ
U O U
U
-O
This controller in the predictive structure could theoretically provide good control
of the controlled variable. However, there are several drawbacks with this design. First,
the controller involves first, second and third derivatives of the feedback signal. These
derivatives cannot be calculated exactly, although they can be estimated numerically.
Second, the appearance of high-order derivatives of a noisy signal could lead to
unacceptably high variation and large overshoot in the manipulated variable. Finally,
these high derivatives could lead to extreme sensitivity to model errors. Therefore, this
controller would not be used without modification.
Example 6.4 Design an IMC controller for the three-tank mixing process shown in
Figure 6.54 (Example 6.3), using the alternative first-order with dead-
time approximate model for the process using process reaction curve as
- O GZR
R O U GU
)O
U
U O U U
This model must be factored into invertible and non-invertible parts:
-O
)O
U
)O
U GZR
R O U G U
U O U U
The invertible part is then employed in deriving the controller:
U O U U
)ER
U ËÍ)O
U ÛÝ
-O
This controller is a proportional-derivative algorithm, which still might be too
aggressive but can be modified to give acceptable performance.
Example 6.5 Simulate the dynamic response of the linearized three-tank mixing
process in Example 6.4, operating at the base-case inlet flow rate, under
IMC feedback control.
The true process, Gp(s), is taken as the linear, third-order system, and the
controller and dynamic model, Gm(s), will be based on the approximate first-order with
dead-time model. This structure mismatch, which is typical of realistic applications,
precludes perfect control. Thus, the results of this exercise give a realistic evaluation
of the performance of IMC controllers.
The controllers with filter and model transfer functions are:
Review Questions 477
%8O P G
%8O P
G
/8P
U H
/8P /8P 6RP 6RP
U H
U H
U H
In this example, the closed-loop simulation is performed using the equations for the
controller (based on an approximate model) and the linearized third-order model for the
plant. Thus, a significant model mismatch exists between the process and the model.
REVIEW QUESTIONS
6.10 For the mixing system shown in Figure 6.55 and a disturbance in the feed
concentration, discuss how would you add one sensor to improve control
performance through cascade control.
DD
AC
6.11 Figure 6.9 represents the frequency responses for a single-loop and cascade
control with a disturbance in the secondary loop.
(a) Sketch the general shapes and discuss the frequency responses for cascade and
single-loop control for (i) a set-point change and (ii) a disturbance in the
primary loop.
(b) Calculate the frequency responses for (a), with Gd1(s) = Gp1(s) and h = 10.
6.12 Prepare a digital computer program to perform the control calculations for the
cascade control system of water temperature control shown in Figure 6.3. Include
initialization, reset wind-up and other factors required for a good implementation.
6.13 What are the two types of a cascade control system? Illustrating with a block
diagram, derive the closed-loop characteristics equations for both the systems.
6.14 Explain how a cascade controller is tuned.
6.15 Two open-loop transfer functions GM1(s) and GM2(s) are connected in parallel. They
have the same input M(s) but each has its own output, Y1(s) and Y2(s), respectively.
In the closed-loop system, a proportional controller K1 is installed to control Y1
by changing M.
However, a cascade control system is used, where another proportional controller
K2 is used to control Y2 by changing the set-point of the K1 controller. Thus we
have a parallel cascade system.
(a) Draw a block diagram of the system.
(b) Derive the closed-loop transfer function between Y2 and Y2set.
(c) What is the closed-loop characteristic equation?
6.16 What is predictive control? Discuss with the block diagrams different types of
popular predictive control systems used in industrial applications.
6.17 What is a model-based control? Explain with block diagram.
6.18 What are the modelling approaches that are dominant in the model-based control
systems?
Review Questions 479
6.19 Discuss various types of model-based controllers that are popular for process
control applications.
6.20 Explaining with a block diagram, derive model inverse C1 for the first-order-plus
dead-time (FOPDT) internal mode control (IMC).
6.21 Derive the closed-loop transfer functions for the controlled and manipulated
variables of IMC system.
6.22 What are the desirable properties of the filter as the basis for selecting the filter
algorithm of IMC system?
6.23 What is a multiple-loop (multivariable) control system? What are the
multivariable predictive controller techniques that are popular in process control
applications?
6.24 What is adaptive control? Why is it needed in process control applications?
6.25 Discuss with general topology the functioning and applications of adaptive control
system.
6.26 Give two examples of adaptive control for processing units. Describe the functions
of the adaptive control schemes you proposed.
6.27 What are adaptive control adjustments? With block diagrams explain the
functions of each of them.
6.28 Discuss with block diagram the two types of adaptive control schemes that are
used for modelling of plants or processes.
6.29 Explain with block diagrams the uses of self-tuning controllers and regulators.
6.30 Discuss about the different adaptive control techniques that are applied in
industrial processes.
6.31 Discuss the logic of an inferential control scheme. Why is this control scheme
needed? Describe two examples of inferential control systems.
6.32 Discuss the inferential design criteria. Why case changes in disturbance and
manipulated variables included when selecting an inferential variable?
6.33 Develop an inferential control scheme that can be used to control the drying of
solids with warm air. Discuss how you would develop the estimator of the
inferential structure.
6.34 What is an intelligent control used for the process control and monitoring? How
the control algorithm is developed in intelligent control?
6.35 What is an artificial intelligence (AI)? Why is this technique used in process
control systems?
6.36 What are different types of artificial intelligent systems? Explain each of them.
6.37 What is an expert system? What is the structure of an expert system? With
illustrative example, explain the use of expert control system for process control
applications.
480 Advanced Process Control Strategies
-
)O
U U
7.50 Design an IMC controller for a process which is of first-order with a transport lag:
GZR
U F U
) -
U U
SELF-CHECK QUIZ
A State True/False:
6.1 Advanced control strategies attempt to compensate for process deviations in the
shortest time possible by accounting for process dynamics, dead times, time delays
and loop interactions.
Self-Check Quiz 481
6.2 In cascade control, the nonlinear gain of the valve can be compensated with a
secondary control loop with PI-controller.
6.3 If the manipulated variable affects one variable, which in turn affects a second
controlled variable, the structure leads to parallel cascade control.
6.4 If the manipulated variable affects both variables directly, the structure leads to
series cascade control.
6.5 The methodology of predictive control was introduced in 1964 in a doctoral thesis.
6.6 In model-based control (MBC), computers use a process model to make control
decisions.
6.7 Transfer function models of model-based predictive control are based on open-loop
Laplace transform descriptions of the process response to step input.
6.8 Time series models of model-based predictive control are based on design-type
simulators in which their control intelligence can offset the disadvantage of their
modelling and computational complexity.
6.9 Nonlinear phenomenological models of model-based predictive control are based on
open-loop response of the process with a vector of impulses, which are empirically
determined.
6.10 The major technical drawback of IMC is that the model is linear and stationary.
6.11 The Identification and Command (IDCOM) method is a type of multivariable model
algorithm control, which is based on a process impulse response that utilizes a
predictive heuristic scenario technique to calculate the manipulated variable.
6.12 Feedforward control detects the disturbances as they exit the process by using the
measured inputs and begins to correct for it before the disturbance upsets the process.
6.13 Adaptive control system uses both a priori knowledge about the controlled process,
as well as the automatic incorporation of acquired outline knowledge based on
observations of the process.
6.14 Artificial intelligence attempts to emulate certain mental processes of humans by
using computer models.
6.15 Expert systems are computer software systems that mimic the tasks routinely
carried out by experts.
6.16 Smith predictor (SP) algorithm is a model-based adaptive approach to control the
systems with long dead times.
(c) Design-type simulators in which their control intelligence can offset the
disadvantage of their modelling and computational complexity.
(d) All of these
6.10 Multivariable control (MVC) is a technique that services
(a) Single-input/single-output (SISO) algorithms simultaneously as opposed to the
multiple-input/multiple-output (MIMO)
(b) Single-input/multiple-output (SIMO) algorithms simultaneously as opposed to
the single-input/single-output (SISO)
(c) Multiple-input/multiple-output (MIMO) algorithms simultaneously as opposed
to the single-input/single-output (SISO)
(d) None of these
6.11 Dynamic Matrix Control (DMC) is a predictive control technique based on
(a) A process step response and calculates manipulated variable moves via an
inverse model
(b) A process impulse response which utilizes a predictive heuristic scenario
technique to calculate the manipulated variable
(c) Both (a) and (b)
(d) None of these
6.12 The Identification and Command (IDCOM) method is a type of multivariable model
algorithm control based on
(a) A process step response and calculates manipulated variable moves via an
inverse model
(b) A process impulse response which utilizes a predictive heuristic scenario
technique to calculate the manipulated variable
(c) Both (a) and (b)
(d) None of these
6.13 In adaptive control system, the control includes
(a) A controller (including its supporting hardware and instrumentation)
(b) A performance index
(c) An adjustment algorithm
(d) All of these
6.14 Adaptive control system uses
(a) A priori knowledge about the controlled process
(b) The automatic incorporation of acquired outline knowledge based on
observations of the process
(c) Both (a) and (b)
(d) All of these
6.15 In self-tuning adaptive controller (STAC)
(a) The specifications are given in terms of a performance index
(b) The specifications are given in terms of a reference model that indicates how
the plant output should respond ideally to the command signal
(c) Both (a) and (b)
(d) None of these
484 Advanced Process Control Strategies
Computer-Based Control
7
7.1 INTRODUCTION
The earliest proposal to use computer for process control (real-time) application was
made by Brown and Campbell in their paper, in 1950. The paper contained a diagram,
as shown in Figure 7.1, wherein a computer was used in both the feedback and
feedforward loops. Brown and Campbell assumed that analog computing elements will
be most likely to be used but they did not rule out the use of digital computing elements.
The first digital computer developed specifically for control (real-time) was for airborne
operation, and in 1954, a Digitrac digital computer was successfully used to provide an
automatic flight and weapons control system.
Raw material
Product
Plant Process reaction
Means ofof
Means
Computer quality
quality
measurement
Appropriate measurement
Appropria
parameter
te
Reference
Referen
Figure 7.1 Computer in a plant control.
489
490 Computer-Aided Process Control
Data acquisition
computer
Measurement
interface
Sensors /
transducers
Process
processing by the process control computer. The smart instruments ensure that the
actuator, transmitter or sensor function according to design.
The major active application of digital computers is in process control and plant
optimization. Computer control systems, once prohibitively expensive, can now be
tailored to fit most industrial applications on a competitive economic basis. The advances
in the use of computer control have motivated many changed in the concepts of the
operations of industrial processes. Video display terminals now provide the focus for
operators to supervise the whole plant from a control room. A few keyboards and
screens replace large panels of instruments, knobs and switches. Control rooms are now
much smaller and fewer people are required to supervise the plant.
Process control computers now have the capability to implement sophisticated
mathematical models. Plant managers and engineers can be provided with
comprehensive information concerning the status of plant operations to aid effective
operation. With the use of microprocessor-based instruments and new emerging
techniques, it is possible for the automatic tuning of controller parameters for the best
operating performance. The expert systems and advanced control techniques, such as
model-based predictive control, are being applied with the help of computers for
optimization of the process operation.
Computer system
v Final
Set-point Controller control y(t)
y(t )
equation Digital-to- element Process
e (digital u analog u(t)
u(t) u(t)
u(t) Proces
+ (control
– algorithm) converter valve)
Measured
Measured Controlled
Analog-to- Sensor
digital
variable
Variable variable
(measuring element)
converter
Digital computer
Digital computer
v(t
v(t))
Set-point Digital
Digital Hold Process
Process y(t)
y(t)
algorithm
e algorithm u(t) device u(t)
+ u(t) u(t)
–
Direction On/off
Temperature sensor
Air Flow
Air flow Reversible motor
control
Amplifier
Heater control
Temperature Blower
measurement
(0–10 V) Computer
Variable
air inlet
Computer
Amplifier Heater
control Direction On/off
Sensor
Reversible
motor
Air flow
control
Variable
Fully air inlet
closed
Air inlet Fully
position open
(b) Computer control
494 Computer-Aided Process Control
Computer
ADC Temperature
measurement
some preset process operating and then discharging the product. Figure 7.5 illustrates
a chemical reactor, a typical example of a batch process.
In this batch reactor, batch cycle consists of the charging sequence of each of the
two chemicals, a heat-up operation, cool-down sequence, and discharge of the final
product-mix for separation and subsequent processing. A chemical is produced by the
On/off valve
Chemical 1
On/off valve PT PC
Set-point
Stirrer
Chemical 2
TT TC
Temperature
controller
Hot
water
Cold
water
On/off valve
Product-mix exit
Figure 7.5 Chemical reactor (batch reactor).
reaction of two other chemicals at a specified temperature. The two chemicals are mixed
together in a reactor (sealed vessel) and feeding hot or cold water through the water
jacket, which surrounds the vessel, controls the temperature of the reaction. The hot
or cold-water flow, the flow of chemicals into the reactor and the flow of product mix
out of the reactor are controlled by adjusting the respective valves. The pressure inside
496 Computer-Aided Process Control
the reactor and the temperature of the product mix are continuously monitored. For
implementing computer control, these actions /sequences are put as logical steps in the
form of a software (program).
Batch process can be classified, on the basis of the number of products (which
could be typically the number of individually identifiable chemicals, substances or items
produced), as
(i) Single-product batch process: A single-product batch process produces the
same product in each batch and same operations are performed in each batch
using the same amount of raw materials.
(ii) Multi-grade batch process: A multi-grade batch process produces products
that are similar but differ only in formula quantities. The same operations are
performed on each batch, but the quantities of raw materials and processing
conditions (e.g. temperature) are varied, i.e. the procedure remains the same,
but the formula is changed.
(iii) Multi-product batch process: A multi-product batch process produces products
utilizing different methods of production or control. Both the procedure
and formula are changed. With each batch, the operations performed, the
amounts of raw materials and processing conditions, may vary in a multi-
product process.
Implementing an automatic batch process is more difficult for a multi-grade process
than that for a single-product process. Most difficult process to be automated is the
multi-product process amongst the three types.
Batch process can also be classified, on the basis of the structure of the process
facility, as
(i) Series (single-stream) structure batch process: A series-structure batch
process is a group of units through which a batch passes sequentially, as
shown in Figure 7.6(a). It could be a single unit such as a reactor, or several
process units in sequence. Two different recipes could be executing at the
same time in this type of process.
(ii) Parallel (multi-stream) batch process: A parallel structure batch process is
that in which several batches may be executing at the same time, as shown
in Figure 7.6(b), and they could also be performing the same operation. Many
batch processes combine single-stream and multi-stream structures to give a
series-parallel structure, as shown in Figure 7.6(c).
The batch process classifications by product and by structure can be combined as
a matrix, as shown in Figure 7.7, to show the degree of complexity in automating these
various combinations. As illustrated in the figure, the multi-product, series/parallel
batch processes require the highest degree of sophistication in control equipment to
achieve effective batch control.
Classification of Computer-Aided Process Control System 497
Reactor 1 Reactor 2
(Product A) (Product B)
(Product C) (Product D)
(b) Parallel (multi-stream) structure
(Product A ) (Product B)
Reactor 1 Reactor 3
Series
Simplest Complex
Parallel
(f )A batch processes generally have recipes associated with each batch of product
that is made, and the same equipment might be used to produce many
different products or product grades. These recipes direct the processing of the
raw materials into final product. Continuous processes usually operate under
the same recipe for large period of time.
(g) Batch processes generally require more sequential logic which is a part of the
normal operating mode of the batch process. Whereas, in continuous
processes, the sequential logic is used only during the start-up or shut-down
of the system.
(h) A batch system uses number of regulatory control loops that operate within
the phases of the process, e.g. temperature control during the heating phases.
The discontinuous nature of batch process causes many unique control
problems with these loops.
(i) Batch processes often include the processing of phases and steps that can
fail, so special failure routines to be executed in that event are also
included. In continuous process, the process is generally shut-down when
failure occurs.
From the above description, it is obvious that programmable logic controllers
(PLCs) are the most suitable device to control the batch processes. Both the logical and
continuous variables can be handled within the same unit and both binary combinational
logic and sequencing operations can be easily carried out using PLCs.
Supervisory computer
M easurements
Controller Process-1
+
–
Controller Process-2
+
–
Controller Process-N
+
–
(b) Allocation, i.e. optimal distribution of a limited resource among several parallel
(alternative) process units (such as fuel use, feedstock selection, etc.)
(c) Scheduling i.e. batch processing, cleaning, maintenance, relining.
Classification of Computer-Aided Process Control System 501
where
e = Error (measurement reference)
Kp = Proportional controller gain
Ti = Integral time of the controller
Td = Derivative time of the controller.
Almost all industrial control problems are solved by applications of the above
algorithm or its close variation. This is implemented in DDC systems by using difference
equation equivalent to Eq. (7.1).
FG GM GM
(7.2)
FV 6
and
e dt = ekT, k = 0, 1, 2,
The control equations can be written as
È 6 6F Ø
OM - P É GM (7.3)
Ê 6K 5M GM GM ÙÚ
PT
Chemical-1
Stirrer
TT
Chemical-2
Hot
water
Cold
water
Jacket-water exit
On/off valve
Product-mix exit
Input
Input Multiplexer
Multiplexer Output
Output
multiplexer
Multiplexer control
Control multiplexer
Multiplexer
Computer DAC
ADC Computer DAC
ADC
Operators’
Operators
Peripheral
Peripheral Console
Console programming
Programming
devices
Devices console
Console
Figure 7.9 Direct digital control system for chemical reactor.
Computer-Aided Process Control Architecture 503
Early DDC system application had drawbacks, e.g. if only one processor was used,
a single failure could affect a large number of controlled variables and possibly disable
an entire process. To avoid such failure, redundant processors were used to provide
back-up for hardware and also additional processors used in parallel to take care of
complex control functions.
Level-2
Level-2
comDrive
computer
Data storage
Data storage MMI
MMI
Computer
gateway
(b) A colour-CRT (cathode ray tube display unit) based largely programmed
operator interface system is easy to adapt to the individual plant situation.
(c) A programmed menu type of instruction system for the microcomputers of the
controller box permits the final programming (configuration) of the total
system to be done by using a keyboard.
(d) A very wide selection of control algorithms or computational schemes within
the programmed menu permits easy selection and testing of alternate control
schemes for a process.
(e) Data highway data transmission and communication capabilities, between
separate processing units of the system, provides very wide band
communications and the possibility of redundancy for extra safety.
(f) Relatively easy communication with the host computer systems for
supervisory control or hierarchy control or other higher-level process control
functions. However, these new distributed computer control systems
themselves are generally restricted to supplying the needs of first level or
dynamic control. Supervisory control is externally supplied by an associated
host computer.
(g) Extensive diagnostic schemes and devices for easy and rapid maintenance
through replacement of entire circuit boards.
(h) Redundancy and other fail-safe techniques help promote high system reliability.
(i) Relay and logic functions normally associated with programmable logic
controllers (PLCs) have been incorporated into the set of control algorithms of
each control box.
(j) A global database to simplify the task of data collection for use in upper-level
hierarchical control functions.
(k) It incorporates personal computers (PC) into the system as a special controller,
as a special operator console, as a programming device and as a special
configuration device, to select control loop parameters and properties of any
control loop in the system.
Level-4 Corporate-level
MIS management information
Production
Plant control level
Level-3 level
(production planning and scheduling)
Level-3
Level-2
Computer
gateway
Industrial process
An industrial plant is faced with the continual problem of adjusting its production
schedule to match its customers need, as expressed by the order stream being
continually received. It requires to maintain a high plant-productivity at lowest practical
production costs. Another important role of computer-aided control system is to work as
control system enforcer in which the lower-level computer continually ensures that the
control system equipment is actually carrying out its designed job, and thus, keeping the
units of the plant production system operating at some optimal level.
REVIEW QUESTIONS
7.1 What is the role of digital computers in industrial process control?
7.2 With a block diagram describe the important elements of a computer-aided process
control system.
7.3 For a tank-level control shown in Figure 7.13, draw a block diagram representation
for a computer-aided control:
LC
Heater
7.4 Explain the classification of computer-aided process control system. Which category
of the classification is used most for process control functions?
7.5 In which applications, batch processes are widely used? What are the different
types of a batch process? Explain with an example, the working of a computer-aided
process control for a batch process.
7.6 What are the differences between batch and continuous processes?
7.7 Give examples of supervisory computer control processes. Draw a block diagram
for the supervisory computer control of a furnace temperature control.
7.8 Explain with sketch and block diagram, a direct digital computer control system for
chemical reactor.
7.9 Consider the two-stirred-tank heaters shown in Figure 7.14. The flow rates of the
effluent streams are assumed to be proportional to the liquid static pressure that
causes the flow of the liquid. Cross-sectional areas of the two tanks are A1 and A2
(ft2) and the flow rates are volumetric.
(a) Develop a suitable type of computer control system.
(b) Draw a block-diagram representation indicating important elements of the
computer control system.
7.10 Develop a computer-aided process control system using suitable category of
classification for an industrial process, as shown in Figure 7.15.
512 Computer-Aided Process Control
F1, Ti
A1
F2, T2
F3, T3
F0
F3
F1
L1
F2
L2
Superheated
Flue steam
gas
TC-2
TC-1
Treated
Treated
water
water
Treated
water
Radiation
Fuel
Figure 7.16 Cascade control for boiler superheated steam temperature control.
SELF-CHECK QUIZ
A State True/False:
7.1 The first digital computer developed specifically for control (real-time) was for
airborne operation.
7.2 Ferranti Argus 200 system was a large system with a provision of 120 control loops
and 256 measurements.
514 Computer-Aided Process Control
7.3 Process control computers do not have any capability to implement sophisticated
mathematical models.
7.4 The device that converts continuous electrical signal into a discrete-time signal is
called digital-to-analog converter.
7.5 In batch process, there is continuous feed and product stream flows require
unusual amount of logic and sequencing in the control.
7.6 A batch process is transient under normal conditions.
7.7 A multi-grade batch process produces products utilizing different methods of
production or control.
7.8 A single-product batch process produces the same product in each batch and same
operations are performed in each batch using the same amount of raw materials.
7.9 In a continuous control process, the production is maintained for long duration
without interruption, typically over several months or even years.
7.10 Continuous process may produce batches in that the product composition may be
changed from time to time.
7.11 In 1969, Honeywell Company, USA, designed TDC 3000 system, a distributed
computer control system as an alternative to the unwieldy and unreliable
centralized computer control system.
7.12 In hierarchical computer control system there is hierarchy of computers connected
on a network each performing distinct functions.
7.1 The earliest proposal to use computer for process control (real-time) application
was made by
(a) Brown and Campbell (b) Routh-Hurwitz
(c) Ramo-Wooldridge (d) Ferranti Argus
7.2 The first industrial computer control system for closed-loop control, called RW-300
(Ramo-Wooldridge Company) system, was installed at
(a) Port Arthur refinery (b) Texaco company
(c) Texas (d) All of these
7.3 The device that converts continuous electrical signal into a discrete-time signal is
called
(a) Digital-to-analog converter (b) Analog-to-digital converter
(c) Both (a) and (b) (d) None of these
7.4 The device that converts discrete signal into a continuous electrical signal is called
(a) Digital-to-analog converter (b) Analog-to-digital converter
(c) Both (a) and (b) (d) None of these
7.5 A process in which sequence of operation is carried out to produce a quantity of
a product and the sequence is then repeated to produce further batches is called
(a) Batch or sequential process (b) Continuous process
(c) Direct digital control (DDC) (d) Supervisory control
Self-Check Quiz 515
7.6 A process in which the production is maintained for long duration without
interruption, typically over several months or even years is called
(a) Batch or sequential process (b) Continuous process
(c) Direct digital control (DDC) (d) Supervisory control
7.7 A widely used industry for batch processes where the operations carried out
frequently involve mixing raw materials, carrying out some preset process
operating and then discharging the product, is
(a) Food-processing industries (b) Chemical industries
(c) Both (a) and (b) (d) None of these
7.8 A batch process which produces the same product in each batch and same
operations are performed in each batch using the same amount of raw materials,
is known as
(a) Multi-grade batch process (b) Single-product batch process
(c) Multi-product batch process (d) None of these
7.9 A batch process which produces products that are similar, but differ only in formula
quantities, is known as
(a) Multi-grade batch process (b) Single-product batch process
(c) Multi-product batch process (d) None of these
7.10 A batch process which produces products utilizing different methods of production
or control, is known as
(a) Multi-grade batch process (b) Single-product batch process
(c) Multi-product batch process (d) None of these
7.1 The earliest proposal to use computer for process control (real-time) application
was made by ___________ in their paper in the year ________.
7.2 In 1954, a Digitrac digital computer was successfully used to provide an
_____________ and weapons control system.
7.3 The first direct-digital control (DDC) computer system was the _________________
system installed in ___________ at the ICI _______________ at Fleetwood, Lancashire.
7.4 The device that converts continuous electrical signal (analog) into a discrete-time
signal is called __________________.
7.5 The device that converts discrete signal into a continuous electrical signal is called
_____________________.
7.6 In _____ process, there is discontinuous feed and product stream flows require
unusual amount of logic and sequencing in the control.
7.7 Discrete loads of raw materials are usually fed into the system for a_______,
whereas ________feed of materials in case of continuous process.
7.8 In _________ process, raw materials are usually processed in an identical fashion.
7.9 Intermittent operations are primarily involved rather than continuous operations
as in the case of _________process.
7.10 In 1969, Honeywell Company, USA, designed ______system, a distributed computer
control system as an alternative to the unwieldy and unreliable centralized
computer control system.
8
8.1 INTRODUCTION
Almost every modern (general-purpose) digital computer can be used for industrial
process (real-time) computer control. To make it useful for process control, the digital
computer should be equipped with additional features such as ability of efficient and
effective communication both with plant and operating personnel, capability of rapid
execution of tasks (algorithms) to provide for a real-time control action. With the
advancement in technology and development of microcomputers and minicomputers, it
is possible to use them even for very large industrial process control applications with
several thousand input/output (I/O) points. Therefore, any general-purpose digital
computer with a stored program, sufficient input/output capability, interrupt handling
ability and faster communication features, can be used for industrial process control
applications.
In Chapter 7, we have discussed how computer can be (or are being) used for the
control of industrial processes. In this chapter, various hardware and software of
computer required for process control applications will be discussed in detail.
Memory
Arithmetic logic
unit (ALU)
General-purpose
General purpose Arithmetic and logic Control
register
register unit (ALU) unit
Data bus
Data bus
Address
bus
Control
bus
Main memory Input/output
interface
Peripherals
Computer fetches data from primary (main) memory under command of control
unit and places them in one of the registers. It performs arithmetic operations using
ALU based on application program requirements, saves them temporarily in
accumulator (a type of register), and transfers it finally to primary (main) memory
storage after the calculations or operations are over. The processed data are further
transferred to the input/output (I/O) devices as per the requirements of the application
program.
Important features of the CPU which determine the processing power available and
hence influence the choice of computer for process control applications include:
Word length
Instruction set
Addressing method
Number of registers
Information-transfer rates
Interrupt structure.
The word length used by the computer is important both in ensuring adequate
precision in calculations and also, in allowing direct access to a large area of main
storage within one instruction word. It is always possible to compensate for short word-
lengths, both for arithmetic precision and for memory access, by using multiple-word
operations. But, the disadvantage is that it increases the time for the operations. The
amount of main storage directly accessible is also influenced by the number of address
lines provided on the input/output interface.
The basic instruction set of the CPU is also important in determining its overall
performance and therefore following features are desirable to have:
Flexible addressing modes for direct and immediate addressing
Relative addressing modes
Address modification by use of index registers
Instruction to transfer variable length blocks of data between storage units or
locations within memory
Single commands to carry out multiple operations.
These features reduce the number of instructions required to perform housekeeping
operations and hence both reduce storage requirements. It also improves an overall
speed of the operation by reducing the number of access to main memory required to
carry out the operations. A consequence of extensive and powerful instructions set is,
however, that an efficient assembly language programming becomes more difficult in
that the language can become complex. Thus, it is desirable to be able to program the
system using a high-level language which requires a compiler designed to make
optimum use of the special features of the instruction set.
Information transfer is another important area, which must be considered
carefully, both within the CPU, between backup storage and the CPU, and with the
input/output devices while selecting a computer for process control applications. The
rate at which such transfers can take place, the ability to carry out operations in parallel
with processing of data and the ability to communicate with a large range of devices can
be crucial to the process control applications. Flexible and multi-level interrupt structure
is also a vital requirement for the computer to be used for process control applications.
Organisation of a General-Purpose Computer 519
8.2.2 Storage
The main function of the storage device is to store data and instructions (programs). The
most elementary unit of stored information is the bit (a 0 or 1 value) and a byte
(consisting of 8 bits) representing a character of the information or data, is considered
as an elementary storage unit. The storage capacity of a computer is expressed in terms
of certain multiples of a byte:
1 byte = 8 bits
1 Kilobyte (KB) = 103 bytes = 1,024 bytes
1 Megabyte (MB) = 106 bytes = 1,024 KB
1 Gigabyte (GB) = 109 bytes = 1,024 MB
1 Terabyte (TB) = 1012 bytes = 1,024 GB
For economical and technological reasons, storage devices come in many sizes,
speeds and costs. A storage device ranges from an inexpensive, low-capacity, slow-speed
device (e.g. floppy disks) to a larger, more expensive, faster ones (e.g. non-removable or
hard disks and tapes) suitable for permanent storage of the information of large
commercial, governmental or educational enterprises. The storage devices store two
distinct types of informationdata and instructions (programs). Storage devices are of
two types:
(a) Main storage or immediate access storage (IAS)
(b) Auxiliary or secondary memory storage and
(c) Cache memory.
The main memory (also called primary memory) storage is a high-speed random
access memory (RAM), which is volatile in nature, which means, it keeps the data as
long as the power is available to the computer. As soon as the power is off, the content
of the memory is erased (removed). It stores information or program, which is required
for execution. The operation of main memory is very fast. RAM permits both read and
write operations by the CPU. Another kind of memory, which CPU can read but cannot
write or change, is called read-only memory (ROM). Variations that can be
reprogrammed by off-line devices are called PROM programmable ROM (PROM) or
erasable programmable ROM (EPROM). The use of ROM has eased the problem of
memory protection to prevent loss of programs through power failure or corruption by
malfunctioning of the software. The alternative to using ROM is the use of memory
mapping technique that traps instructions, which is an attempt to store in a protected
area. Computers require the main (or internal or primary) memory not only for storing
instructions and data awaiting execution of a program, but also for storing the results,
which can thereafter be transferred to the output unit.
The auxiliary memory storage (such as tape or disk) is non-volatile in nature that
means, it keeps data, information, program, or instructions even after power of
computer is off. These devices provide bulk storage for programs or data, which are
required infrequently at a much lower cost than the high-speed main memory. But, it
has disadvantages of taking much longer access time and the need for interface boards
and software to connect to the CPU. Auxiliary storage devices operate asynchronously
to the CPU and care has to be taken in deciding on the appropriate transfer technique
for data between the CPU, fast access (main) memory and the auxiliary storage. On a
520 Computer Hardware for Process Control
real-time system, it is not desirable to make use of the CPU to carry out the transfer
as not only is this a slow method but no other computation can take place during
transfer.
An additional storage, called cache memory, is also used in the process control
computers for improvements in performance and utilization. Cache memory is an
auxiliary memory that provides a buffering capability by which the relatively slow and
ever-increasingly large main memory can interface to the CPU at the processor cycle
time in order to optimize performance. It is a high-speed storage that is much faster
than the main storage, but extremely expensive compared with the main storage.
Therefore, only relatively small caches are used.
Memory
Memory Arithmetic and logic Control unit
Control unit
unit (ALU)
Input/output
Input/output (I/O) interface
(IO) interface
Computer-related
RTC devices
interrupt
Disk
Plant-related
devices Magnetic
tape
Digital
signal Line
printer
process
Plant or process
Pulse
signal Console
keyboard
ADC/
Plant
DAC
Operator devices
Telemetry
channels Keyboard
data entry
Display
Alarm
handler
Log printer
Log printer
Figure 8.1(b). A bus contains one wire for each bit needed to specify the address of a
device or location in memory, plus additional wires that distinguish among the various
data-transfer operations to be performed. A bus carries electrical signal and the wires
may be tracks on a printed circuit board or in a ribbon cable form. This printed circuit
board (called cards) or ribbon can be plugged into the slots provided in the computer.
A bus can transmit data in either direction between any two components of the
computing system. Without a bus, a computer would need separate wires for all possible
connections between peripherals, which will be a very difficult situation. A bus is divided
into the following three functional groups:
Data line
Address line
Control and status lines.
522 Computer Hardware for Process Control
The data lines indicate what the information is, the address lines provide
information on where the data or information is to be sent or where it is to be obtained
from, and the control and status lines indicate when it is to be sent.
On a microcomputer, the bus is usually called an expansion bus because its design
determines the degree to which the minimum configuration of the system can be
expanded with regard to memory, processing speed, graphics capability, and peripheral
support. The expansion bus is the collection of wires, paths, connectors, and controllers
responsible for distributing the data and instructions from the microprocessor to the
peripheral expansion cards. Slots connected to the bus provide places to plug those cards
in, and the bus then provides a mechanism for communicating with them.
When microcomputers were first introduced during the 1970s, there was no
standard mechanism for expanding a computers capabilities. Every manufacturer had a
different scheme, and manufacturers had not even standardized across different models
in a single product line. With increased use of computers for process and other control
applications, various standards were introduced. Following generations of expansion
buses were introduced with IBM-compatible PCs:
S-100 bus
ISA (Industry Standard Architecture) bus
ISA AT(Advanced Technology) bus
MCA (Micro-channel Architecture) bus
EISA (Extended Industry Standard Architecture) bus
NU-bus
Now, they are explained.
S-100 bus
With the introduction of 8080 computer by Altair, it came with an expansion bus with
a published specification that used common parts and connectors. This open architecture
soon became a standard. One-hundred pins were provided for various signals on what
became known as S-100 bus. This meant that new video cards, more memory, and serial
and parallel ports could be added to the computer as needed. Thus, concept of add-on
board concept was born and number of companies started manufacturing S-100 bus
cards. The S-100 bus was capable of a 16-bit addressing, but it had some shortcomings.
It was subject to interference and cross-talks. It did not perform reliably with high-
frequency signals, and there were ambiguities in the standard that led some
manufactures to define specific pins in multiple ways.
ISA bus
In 1981, IBM announced its new personal computer, the IBM-PC, with the concept of
open architecture and flexibility of its expansion bus. This new IBM bus was named ISA
(Industry Standard Bus). In some ways it was inferior to S-100 bus. It strictly used an
8-bit architecture with parity protection and had only 62 pins consisting of 3 grounds,
5 voltage-supply lines, 20 address lines, 8 data lines, 10 lines devoted to interrupts, and
a variety of special-purpose signal lines. The ISA bus was processor-specific and its edge-
triggered interrupts meant that each expansion card could have only one interrupt.
Organisation of a General-Purpose Computer 523
ISA AT bus
In 1984, IBM announced its new advanced-technology (AT) computer built around the
16-bit Intel-80286 chip, which could be run as fast as 12 MHz. An additional connector
was added next to the 8-bit ISA connector, which allowed additional address and control
signals while maintaining downward compatibility with ISA expansion boards.
Information could be transferred at up to 2 Mbytes per second on this bus, but speed
was too slow due to its rating of 8 MHz frequency.
MCA bus
To increase the market share, IBM announced a new computer line, the Personal
System/2 (PS/2) in 1987. It introduced a new expansion bus based on a concept of micro-
channel architecture (MCA). The MCA increased data throughput to 10 Mbytes per
second, more than ten times the speed of the ISA AT bus. MCA expansion cards were
smaller in size than those used with ISA bus and were designed to take advantage of
assembly advances using surface-mount components. It was a full 32-bit bus with every
fourth pin as a ground, reducing interference and allowing much faster cycle times.
Expansion cards could now communicate directly with the Video Graphics Array (VGA)
card and the expansion cards could be configured by the microprocessor. Interrupts were
level-sensitive and remained active during the entire cycle, making it easier to address
multiple cards with the same interrupt. Most significantly, no longer was the main
processor the sole repository of computing power and expansion cards could now have
their own processors and memory. The expansion cards could now be more intelligent
and powerful, freeing the main processor for additional tasks. But, the disadvantage of
MCA card was that it was not downward compatible with the old ISA cards.
EISA bus
A group of computer manufacturers got together and formed the Gang of Nine; Wyse,
AST, Tandy, Compaq, Hewlett-Packard, Zenith, Olivetti, NEC, and Epson and developed
a joint standard named EISA (Extended Industry Standard Architecture). EISA remained
compatible with the old ISA boards. It was 32-bit standard, even faster than MCA, with
a maximum transfer rate of 33 Mbytes per second. The expansion cards were almost
twice as large, allowing for more components.
The EISA bus is synchronous and can be used only in the 80386 and 80486 chips.
Interrupts are no longer edge-triggered and can be shared. The EISA standard is
producing high-standard cards for all microcomputer applications. New disk controllers
reduce hard-disk access times below 1 millisecond threshold. On-board intelligence and
memory hold the promise of extremely fast response, increased performance, and widely
enhanced video resolutions.
NU-bus
Apple Computer Corporation introduced NU-bus. Until the introduction of this bus in
the late 1980s, the Macintosh had been a closed machine. This was widely cited, along
with higher cost, as the reason for Apples inability to match the market share of IBM
compatibles. The 32-bit NU-bus was introduced with Mac-II. It operates on a 10-MHz
synchronous clock, providing access to all main logic board resources through six
524 Computer Hardware for Process Control
Euro-DIN connectors. This has led to many enhancements for the Mac that might not
otherwise have evolved.
Analog-to-digital converters
In a digital computer control system, the plant output (or controlled variable) is usually
an analog signal, which is measured by sensors (or instruments). This analog signal has
to be converted into a digital signal before it is fed to the computer. The conversion of
analog measurements to a digital data involves three operations namely sampling,
quantization and encoding, as shown in Figure 8.3.
Sampling
Sampling Quantization
Quantization Encoding
Encoding
Analog
Analog
00 ------22nn
signal Digital
Digital
signal output
L
output
fs = 1/T
0 T 2T 3T
110 qq
101
Binary output code
Coding point
100
011
Decision points
010
001
000 r(t)
1 2 3 4 5 6 7
FS FS FS FS FS FS FS FS
8 8 8 8 8 8 8
Normalized analog input voltage
(a) Quantizing intervals
+½ LSB
q/2VV
e == ±q/2
0
r
ö
-½ LSB
LSB Quantizing error
(b) Quantizing error
Hardware unit
LimPos
r(t) c(t)
KADC Digital
KADC computer
LimNeg
spacing 2n (1 LSB-list significant bit) for a converter having n-bit binary output word-
length. Note that the maximum output code does not correspond to a full-scale input
value, instead to (1 2n). Full-scale (FS) because there exist only (2n 1) coding points,
as shown in Figure 8.4 for a 3-bit ADC.
Quantization of a sampled analog waveform involves the assignment of a finite
number of amplitude levels corresponding to discrete values of input signal Vs between
0 and VFS. The uniformly spaced quantization intervals 2n represent the resolution limit
for an n-bit converter, which may also be expressed as the quantizing interval q equal
to VFS/(2n 1)V.
To illustrate this concept consider a 3-bit ADC, with an input voltage range of
±10 volt, that is, a full-scale (FS) range of (10 + 10) = 20 volt. With a word-length of
3 bits, the number of levels available for quantizing the analog signal is 23 = 8.
Thus, the interval between successive levels, called quantization level or quantization
step-size, is given by
S P (5 (5 (8.1)
or
S
= 2.5 volt
When we view the output (binary output code) c(t), as the sum of the actual input
(normalized analog input) r(t), and a random error (noise) e, called the quantization
error, whose value is bounded between ± q/2. That is,
c(t) = r(t) + e (8.2)
and
S S
G (8.3)
A block diagram is shown in Figure 8.4(c) to model the function of an ADC so that
we can properly use it when simulating digital filters or digital control systems. As
shown, the limiter is a part of the hardware, but is a useful when simulating the ADC
function on a digital computer. For a 2s complement, n-bit ADC, the positive and
negative limit settings (LimPos and LimNeg) are given by
LimPos = 2(n1) 1 (8.4)
and
LimNeg = 2(n1) (8.5)
The parameter KADC represents the gain or scale of the ADC device and is given
by
P
- #&% (8.6)
(5
where FS is a full-scale range of the ADC, such that FS = 10 + 10 = 20 for a 3-bit ADC.
For 3-bit ADC, we also have
528 Computer Hardware for Process Control
- #&%
S
The output c(t) of ADC for an analog input signal of r(t) volts is rounded to the
nearest level and given by
c(t) = round [r(t)*KADC] (8.7)
ADC outputs c(t), for different analog input r(t) volts, in both decimal as well as in
binary, for a 3-bit word-length are given as follows:
For r(t) = 3 volt, c(t) = round (3*0.4) = 110 = 0012
For r(t) = 5 volt, c(t) = round (5*0.4) = 210 = 0102
For r(t) = 7 volt, c(t) = round (7*0.4) = 310 = 0112
For r(t) = 9 volt, c(t) = round (9*0.4) = 410 = 1002
For r(t) = 12 volt, c(t) = round (12*0.4) = 510 = 1012
For r(t) = 15 volt, c(t) = round (15*0.4) = 610 = 1102
For r(t) = 17 volt, c(t) = round (17*0.4) = 710 = 1112
The relationship between the analog input and digital output for a 3-bit, ±10 volt,
2s complement ADC is shown in Figure 8.4(b).
While a variety of circuit designs are available for executing analog-to-digital
conversion, most popular of these employ the following techniques:
Successive approximation
Single and dual-loop integration
Voltage-to-frequency conversion
Synchro-to-digital conversion
Simultaneous conversion
Tracking conversion.
The decision as to which type of ADC conversion is best suited for a particular
application will have to be made based on various influencing factors such as speed,
accuracy, cost, size and the inherent noise-reduction capability. However, the successive
approximation technique is the most widely utilized ADC for computer interfacing
primarily because its constant conversion period (T) is independent of input signal
amplitude and its timing requirements are therefore conveniently uniform.
Upper
Upper limit
limit
7/8
3/4 Lowerlimit
Lower limit
Normalized analog output
Normalized analog output
5/8
Differential
Differential
1/2 nonlinearity
non-linearity
3/8
1/4
1/8
0
000 001 010 011 100 101 110 111
Binary
Binaryinput
input code
code
(a) 3-bit D/A converter transfer relationship
+Vref
LSB MSB
+
RL –
RR11 RRf
f
LimPos
Digital ZOH
KDAC
computer
LimNeg
Analog
Analog
ADC
ADC
sensor
sensor
Analog
Analog ADC
ADC
sensor
sensor Digital
Digital computer
Plant or process
Plant or process computer
to be controlled
to be controlled Analog
Analog
sensor
sensor
DAC
DAC
Analog
Analog
DAC
DAC
sensor
sensor
Sample-and-hold devices
An ideal sample and (zero-order) hold device takes in an analog signal r(t) and outputs
a signal c(t), that is equal to the analog signal r(t) at each sampling instant and is
constant between sampling instants, as shown in Figure 8.7(a). Figure 8.7(b) shows the
primitive hardware outline of an actual sample-and-hold device.
In Figure 8.7(c), the sampling switch closes for a finite time (q), to allow the
capacitor to charge. When the sampling switch opens, a transient occurs in the signal,
r(t) Ideal
Ideal sample and
sample and
c(t)
zero-order
Zero-order hold device
hold device
r(t)
c(t)
t
(b) The time behaviour
Clock-driven sampling
switch
Amplifier with
r(t) high-input impedance to c(t)
prevent loading capacitor
Capacitor for
holding gain
(c) Hardware outline of the operation
Figure 8.7 Ideal sample-and-hold device.
c(t). Finally, despite the presence of the amplifier the signal c(t) does not remain
perfectly constant between sampling instants. Figure 8.8 illustrates these effects.
For rapidly changing signals, or for rapid sampling (small values of T), the value
q and the length of the switching transient may be critical. For slow-sampling (long
values of T), the non-ideal holding due to capacitor-discharge is important. Here a
rapidly changing signal is to be A/D converted, the sample-and-hold device is inserted
before the A/D converter to obtain an improvement.
532 Computer Hardware for Process Control
r(t)
Ideal holding
Ideal holding
c(t)
Switching transient
Switching transient Non-ideal holding
Non-ideal holding
Capacitor
Capacitor charge
charge
Sampling Sampling
switch closes switch opens
Figure 8.8 The time behaviour of non-ideal sample-and-hold device.
Multiplexing (MUX)
Multiplexing is a technique that allows a number of lower bandwidth communication
channels to be combined and transmitted over a higher bandwidth channel. At the
receiving end, de-multiplexing recovers the original lower bandwidth channels.
Multiplexing scans across a number of analog signals and time-sharing them
sequentially into a single analog output channel. The main reason for multiplexing is
to make efficient use of the full bandwidth of the communication channel and to achieve
a lower transmission cost. Following three basic multiplexing methods are used:
(a) Space division multiplexing (SDM): It refers to the physical grouping of many
individual channels or transmission patterns to form a channel with a high
aggregate bandwidth. Hundreds of twisted wire pairs, coaxial cables, or optical
fibres can be grouped to form a larger diameter cable. Each wire pair, coaxial
cable, or fibre in the main cable is an individual communication channel
capable of being frequency-division, or time-division multiplexed.
(b) Frequency division multiplexing (FDM): Frequency division multiplexing
divides a higher bandwidth channel into many individual smaller bandwidth
channels. Signals (data, voice or video) on these channels are transmitted at
the same time but at different carrier frequencies. Guard bands are needed
between the channels to help reduce inter-channel interference. FDM
Process-Related Interfaces 533
techniques are mostly used to combine low-speed circuits onto single voice-
grade lines for transmission to remote places e.g. television broadcasting in
which the stations broadcast programs continuously, each at a different
frequency through the atmosphere which acts as the transmission medium.
(c) Time division multiplexing (TDM): In time division multiplexing, the entire
bandwidth of the channel is dedicated to a low-speed channel for a short period
of time, and then to the other low-speed channels in round-robin fashion or
some other predetermined sequence. A familiar example of TDM is the input
output bus of a computer servicing many peripherals, one at a time for short
periods of time. The time slots may be allocated on a fixed, predetermined basis
or on a demand basis. TDM is usually subdivided into the following categories:
Synchronous time division multiplexing (STDM)
Asynchronous time division multiplexing (ASTDM)
Message switching multiplexing (MSM).
Figure 8.9 illustrates the relationship between frequency or bandwidth and time in
FDM and TDM.
Channel 1 C C C C C
h h h h h
a a a a a
Channel 2 n n n n n
n n n n n
Frequency
Frequency
Frequency
e e e e e
Channel 3 l l l l l
1 2 3 4 5
Channel
Cannel 44
Time
Time
frequencyTime
or Time Time
freqency or or
frequency FDM TDM
bandwidth
bandwidth space
space
bandwith space
Figure 8.9 Relationship between frequency and time in FDM and TDM.
As networks grow and get more complex, hierarchies of multiplexing are required
in which low bandwidth channels are multiplexed onto higher bandwidth channels,
which in turn are multiplexed on even higher bandwidth channels, etc. In the FDM
hierarchy, multiplex levels correspond to increasingly higher frequency bands. In TDM
hierarchy, they correspond to increasingly higher pulse rates.
Modem
A modem is device used to transmit data between computers, workstations, and other
peripheral devices interconnected by means of conventional communication lines
supporting analog transmission. Modems transform (modulate) data from a digital device
to analog form suitable for transmission over such lines. Since, in general, data flow in
534 Computer Hardware for Process Control
both the directions, modems are also able to receive an analog signal from some
remote device and restore (demodulate) it back to its original digital form, as shown in
Figure 8.10.
Modem
Modem Modem
Modem
Communication
link
Digital Digital
Digital
Digital Modulator
Modulator Demodulator
Demodulator
Central
Central signal
Signal signal
signal
Terminal
Terminal
processor
processor
Mod.
Modulator
Demodulator
Demodulator
signal Modulator
Modulator
signal
Figure 8.10 Modem.
keeps time. For greater precision, a hardware timer is used. The interrupts activate the
RTC software. RTC is usually provided with battery back-up so that even in the absence
of main power the clock function is not lost. RTCs are also used in batch processing and
online computer systems.
RS-232C interface
RS-232C is a communication interface standard initially used in a telephone data
communication. It has now become the interface standard of choice for most personal
computers (PCs) and has found increasing popularity in sensor systems. RS-232C
employs a single-ended voltage interface, as illustrated in Figure 8.11(a), where a logic
1 is a voltage between 15 to 3 volt and a logic 0 is a voltage between +3 to +15 volt.
Sensor data are sent to the computer using a bit serial digital transmission protocol.
Since RS-232C is a single-ended interface, the distance between the transmitter and
receiver must be short to minimize susceptibility to the interference.
Logic11==–15
Logic 15to to
– 33volt
volt
Logic 00==++3
Logic 3 toto+15
+15volt
volt
RS-422A interface
RS-422A is a differential voltage interface illustrated in Figure 8.11(b), which extends the
maximum cable length between transmitter and receiver as well as maximum data rate.
The polarity of the transmission line determines the transmitted logic level and allows
for reliable detection at the receiver at signal levels as low as 200 milli-volts. The
difference interface can tolerate a significant attenuation of the signal. This allows for
transmission over long distances at higher data rates. The differential interface is also
suited to transmission over twisted-pair cabling which will reduce the susceptibility to
interference.
RS-485 interface
The EIA standard RS-485 extends the RS-422A standard by allowing up to 32 receivers
to be connected in a multi-drop configuration on a single cable. Higher-order protocols
are used to provide an organized utilization of the cable. The RS-485 transceivers are
designed to tolerate simultaneous contention on the cable.
Comparison of RS-232C, RS-422A and RS-485 types of interfaces, is shown in
Table 8.1.
Table 8.1 Comparison of RS-232C, RS-422 and RS-485 Interfaces
Company management
level (level-4)
Computer
Production management gateway
level (level-3)
Field
Field level
level control
control (level-
(level-1)
0)
1 1 1 1 1
SENDER 0 0 0 0 RECEIVER
RECIEVER
Character
1 1 0 0 1 0 1 1 1 1 0 0 1 0 1 1
Strobe
Parity bit
Sender P O N M H Receiver
1 1 1 1 1 1
2 0 0 0 0 0
3 1 0 0 0 0
4 0 1 1 1 1
Process computer
Field terminal
5 0 1 1 1 0
66 0 1 1 0 0
7
7 0 1 0 1 0
8 (Parity bit) 1 1 1 1 1
A parallel transmission system is shown in Figure 8.14 in which seven wires have
been provided for a 7-bit ASCII code and two wires for parity bit and strobe signal. All
the ASCII 7-bit of a character is transmitted in the time taken to transmit a single bit.
As seen in Figure 8.14, all the bits related with one character are transmitted at a
time, but the individual characters such as M, N, O and P are transmitted serially, i.e.
one after the other. Therefore, a parallel transmission technique is also called as
parallel-by-bit but serial-by-character transmission system. It is most suitable for short
distance communications. It is a very efficient and an extremely high-speed transmission
system.
Most of the parallel interfacing requiring simple control applications is achieved
through the use of peripheral interface adapter (PIA) chips. PIA is the natural
interfacing device to interpose between an A/D or D/A converter and a process
control computer (or microprocessor) system. A typical configuration of PIA is shown in
Figure 8.15.
542 Computer Hardware for Process Control
Process computer
PIA
ADC DAC
As the distance between the source (plant devices) and the receiver (control
computer) increases it becomes more difficult, when using analog techniques, to obtain
a high signal-to-noise ratio, particularly in an industrial environment where there are
numerous sources of interference. Use of analog systems is, therefore, generally limited
to short distances. The use of a parallel digital transmission provides high data transfer
rates but is expensive in terms of cabling and interface circuitry and again is normally
used only over short distances (or when high rates of transfers are required).
Station-1 Station-2
Station- 2 Station-3
Medium bus
Station-4 Station-5
Station-6
Figure 8.16 Bus topology.
544 Computer Hardware for Process Control
In a bus topology, the simplest of the LAN topology, the bus is normally passive
and all the devices are simply plugged into the common node of the transmitting
medium. The bus is inherently reliable because of its passive nature but there may be
limitation on the length of a bus. In a ring topology, all the nodes are connected to the
nearest adjacent nodes in a notional ring.
Station-3
Repeater
Station -2
Station-4
Repeater
Repeater
Repeater
Station-1
Figure 8.17 Ring topology.
The demands on a production level, are quite different from those on a processing level
with sensors and actuators. Therefore, no single network can efficiently operate at all
levels. The requirements at each level, in terms of data transmission rate, protocol, and
interfaces, govern the type of communications system, which can be installed. It is
inevitable that several networks will be found within the same process industry
(factory). In order to ensure transparency between various levels within the
communications hierarchy as well as between the devices on a particular network,
standards have to be adopted.
Field-bus user
r7
ye
La
Application layer 6
y er
La
Presentation layer 5
y er
La
Session layer 4
y er
La
Transport layer 3
y er
La
r2
Network layer
ye
La
Data-link layer 1
y er
La
Physical layer
Physical layer
The physical layer defines how data are to be physically transported from device to
device. It includes specifications of electrical signals, cabling, connectors, network
topology and transfer speed. The cabling or transmissions media is the backbone of all
networks. Three types of cables are used.
Twisted pairs, that provide most economical solution for data transmission at
the field level. It allows transmission rates up to 375 KBPS (kilobits per second)
over distance up to 300 m, and 31.25 KBPS up to distance of 1500 m. Shielded
cables provide more immunity against interference, as well as higher
transmission rates.
Coaxial cable, which is more expensive than twisted pairs, and is used for carrier
or broadband transmission in higher-level networks. It allows high transmission
rates of 5 MBPS over 1000 m over carrier or 10 MBPS (megabits per second)
over several kilometres over broadband.
Glass-fibre (fibre optic) cable, that is simple to lay and connect. It has a greater
transmission capacity (more than five times that of the coaxial cable), more
immunity to electromagnetic interference, and has high transmission rates (in
the range of GBPS, gigabits per second).
The network topology describes the way in which the devices in the network are
connected together. The basic topologies are star, ring and bus arrangements. All other
network topologies are built from these basic elements. Bus, ring and star topologies are
shown in Figures 8.16, 8.17 and 8.19, respectively.
The star structure has a central node, which handles all data exchange with its
peripheral devices, each served by a point-to-point connection. In a ring structure, the
Communication and Networking 547
devices are connected in a physical ring. There is no central control, rather, each device
assumes the role of controller at strictly predefined intervals. In a bus architecture, all
the devices are connected to a single data line, the so-called bus, along which
information is passed. Since mutual communication between any of the connected
devices is allowed, transmission must be strictly regulated.
Data transmission lines should be cheap and reliable and for this reason, the
standard topology at field level is the serial bus on which data are sent sequentially. This
saves on cabling, is more reliable, and allows transmission over long distances. The
serial interface standards such as RS-232C and RS-485, are described in Section 8.3.5.
Data-link layer
The data-link layer ensures that data can be exchanged between devices. This layer
governs not only network access and data format, but also mechanism to ensure data
security. When several communication devices are present on a single line, clear rules
must be given following which data are allowed to be transmitted. In principle, there are
two methods of regulating access: central and de-central. In central bus control, a fixed
master assigns the right to one station at a time. If the master fails, then the entire
communication network breaks down. For this reason, de-centralized bus control with
flying masters has been developed. In this case, the right to transmit can be assigned
to several stations. The de-centralized bus control models such as, the CSMA/CD model,
ethernet (as per IEEE 802.3) and the Token passing model (as per IEEE 802.2), have
found a wide acceptance and are found all over the world. An intelligent interface, a
so-called gateway, is required for the linkage of these networks.
1. CSMA/CD bus access method: In CSMA/CD (carrier sense multiple access with
collision detection) method, all the stations on the bus have the right to
transmit. Each station continuously listens to (or senses) the bus. If the bus is
free, then any of the stations can transmit its data. If several stations want to
transmit simultaneously, a collision is detected and all stations withdraw. A
random generator in each of the stations then determines the time interval,
which must elapse before the station can renew its attempt to transmit. The
CSMA/CD bus access is very popular in office and higher automation systems.
It is not suitable for field-buses, since short response times, as required for
alarms, cannot be guaranteed.
2. Token passing method: In this method, the token, i.e. the right to transmit, is
passed from device to device. The passing sequence is defined during the
planning of the system. This method allows each device a fair share of the bus,
since each is allowed to transmit within a preset period of time. The time taken
to pass the token around the system determines the frequency of polling of the
individual members. This bus access method is used in control and field
applications, where reaction to events in the process periphery, must take place
within specified time.
3. The master-slave method: In this method, one bus device is assigned as the
master. Only master is allowed to transmit. The master polls (or addresses) each
participant in turn, supplies it with data, and asks it to transmit its data. This
method is generally used in sensor/actuator applications.
548 Computer Hardware for Process Control
Application layer
The application layer defines the services which are supported by the network, for
instance, read and write commands, program management, up-and-downloading of data,
and virtual device images. User interfaces with this layer for programming the network.
MAP protocol
Manufacturing automation protocol (MAP) was defined in 1983 by General Motors to
allow open data exchange between all the automation equipment the company used.
Since then, several manufacturers of automation equipment have come to offer MAP.
MAP is not very fast, and is suitable for transmission of medium to large data volumes.
The cost of MAP per node is relatively high. MAP supports only token passing bus-access
method (IEEE 802.2). MAP facilitates communication between computer, robot
controllers, programmable logic controllers (PLC), computerized numerical control
(CNC), and direct numerical control (DNC). The application where time is important, the
MAP Enhanced Performance Architecture (MAP/EPA) is used.
TOP protocol
The TOP protocol, defined by Boeing for use in technical/scientific and office automation
environment, lays emphasis on the exchange of documentation. TOP is used for both
CAD and CAM (computer-aided design and computer-aided manufacturing) documenta-
tion and office documentation such as text and graphics. TOP supports a range of
transmission media and access methods including CSMA/CD, token passing ring, etc.
8.5.2 Field-buses
Field-buses are used to link sensors and actuators at the process level with the system
intelligence, i.e. with DCS systems, PLCs and controllers, so that the information they
supply can flow into the plant information system. Field-buses offer several advantages
over hard-wired installation such as:
Bidirectional digital transmission
Simple cabling and cross-wiring
Easy expandability and retrofitting
Industrial Communication Systems 549
Smart transmitters
In smart transmitters, the intelligence is located in the field instruments itself.
Todays smart transmitters operate as standard 4- to 20-mA devices with a superimposed
digital signal, which are able to transport measurement and configure data. The
transmitters are configured and interrogated on site by plugging a hand-held terminal
into the supply line.
Various protocols exist for the smart transmitters such as, DE (closed), INTENSOR
(open), and HART (open). HART has got wide acceptance in the industries. Smart
transmitters are often seen as the first step toward obtaining an intrinsically safe field-
bus as the basis for an international standard.
Rack-bus
The rack-bus is a serial, character-oriented bus developed to link intelligent transmitters
to external networks via gateway. By means of the bus, the data provided by the
measuring points are made available to the supervisory system, and, in addition, the
transmitters may also be remotely configured. The rack-bus operates on the fixed
masterslave principle, where the master is the gateway, a special module that provides
the link to external networks, and transmitters are the slaves. Each transmitter has a
unique address. The gateway is interfaced to the field-bus via special modules for
RS-232C, RS-422 and 4- to 20-mA current loop.
The rack-bus is intended for environments where, despite a high interference
potential, reliable data transport must be assured. Rack-bus realizes layers 1, 7 and 2
of the OSI model. It uses twisted pairs as transmission medium.
550 Computer Hardware for Process Control
MODBUS
The MODBUS is a transmission protocol for process control systems. In contrast to the
other buses, no transmission medium is defined. The user can therefore, choose
between RS-232C, RS-422, or 4- to 20-mA current loop, all of which are suitable for the
transmission rates which the protocol defines. The MODBUS is accessed on master
slave principle, the protocol providing for one master and up to more than 250 slaves.
Only the master initiates a transaction, which is either a query/response type, where
only a single slave is addressed, or a broadcast/no-response type, where all slaves are
addressed. Certain characteristics of the MODBUS protocol are fixed, such as frame
format, frame sequences, handling of communications errors, and exception conditions
and the functions performed. Other characteristics are user-selectable such as,
transmission medium, transmission characteristics, and transmission mode (RTU or
ASCII). All functions supported by the MODBUS protocol are identified by an index
number and are designed as control commands for both field instrumentation and
actuators.
Although the MODBUS is relatively slow in comparison to some other buses such
as PROFI-BUS or FIP-BUS, it has advantages of wide acceptance among instrument
manufacturers and users. It has been regarded as defacto industrial standard with
proven capabilities.
PROFI-BUS
The PROFI-BUS (process field-bus), is a German standard. Unlike the MODBUS, the
PROFI-BUS specifies layers 1, 2 and 7 in accordance with the OSI model. Its aim is to
provide a high-performance communications system, for simple, inexpensive equipment.
Its functionality and data transfer capabilities ensure transparency to higher network
levels such as MAP. PROFI-BUS is designed to meet the following specifications:
Communication with simple field instrumentation with immediate response,
masterslave access with central polling and broadcast messages
Simple and economic transmission system, also for international standard
applications
Application layer with functions corresponding to the MMS (manufacturing
message) standard
The interface between layers 2 and 7 designed such that field-bus functions are
imaged as MMS functions.
The PROFI-BUS uses the MAP/EPA three-layer protocol comprising of the
physical, data link, and application layers. It uses a hybrid access method combining a
centralized masterslave system with decentralized token passing. The PROFI-BUS has
already been published as a German standard.
FIP-BUS
The FIP-BUS (factory information protocol bus) is the result of the combined efforts of
a number of French companies. The FIP-BUS specifies both protocol and interface.
It aims at very fast transmission rates and strictly defines scanning intervals. The FIP-
BUS protocol has hybrid central/de-central control based on the broadcasting principle.
Industrial Communication Systems 551
A central unit, called the bus arbitrator, organizes broadcasting. Most of the data
transmitted on the bus are variables, each of which is identified by a name. A variable
is processed by one transmitter only but can be read by any number of receivers.
The FIP-BUS offers the user freedom in the design of the system architecture. It
eliminates all connections between the functional organization of the automation
process and its physical implementation. Process control functions can be distributed
between the various devices on the bus.
The use of a field-bus implies a transition from analog to digital data transmission.
In comparison to the standard 4- to 40-mA transmitter signals, digital instruments are
capable of supplying additional measurement information including technical units,
events and status. Digital instruments can monitor their own performance and provide
rapid diagnosis, and can be remotely configured and interrogated.
Comparison between MOD-BUS, PROFI-BUS and FIP-BUS, is given in Table 8.2.
facility is very common in most operating systems and real-time applications. It not only
enables a computer to communicate with a rich variety of external devices, but is also
helpful to the system in managing its own device and program resources. Although
basically implemented by hardware, the logical power of interrupts is also provided in
a convenient form to users of some modern programming languages.
REVIEW QUESTIONS
8.1 With neat sketch explain the organization of a general-purpose digital computer.
What are its important components? Discuss the functions of each of these
components.
8.2 Define the structure of a conventional digital computer, and identify the
characteristics of each hardware component in this structure.
8.3 What are the basic and optional features of a central processing unit? Are all of
them needed for a process control computer?
8.4 What are the important features of the CPU which determine the processing
power?
8.5 What is the basic instruction set of the CPU in determining overall performance?
8.6 What is the main function of a storage device of a digital computer? How is the
storage capacity of a computer expressed? What are the types of storage device?
8.7 Define bit, byte and word. Why are they needed to encode information in a digital
system?
8.8 What is the real-time clock needed for? How does it function? Why is it very
difficult to use a computer for process control without a real-time clock? Do you
have any suggestions on how you can count time elapsed without a real-time clock?
8.9 Draw a neat sketch and explain the typical I/O interfaces of a process control
computer.
8.10 What is a bus? What are the types of generation of expansion bus interfaces used
in process control computer system?
8.11 Discuss the types of process related interfaces used in process control computer for
connecting wide variety of instruments and actuators (sensors/transducers) to the
process or plant for measurement of process parameters.
8.12 What are the types of analog interfaces? Discuss each of them with a neat sketch.
8.13 The conversion of analog measurements to digital data involves three important
operations, namely sampling, quantization and encoding. Explain these operations
with neat illustrations.
8.14 What is quantization? Illustrate the concept of quantization by considering 3-bit
analog-to-digital converter. Derive equation of the quantization level.
8.15 Derive an equation to calculate the ADC outputs c(t) for different analog inputs r(t)
volts, in both decimal as well as in binary for a 3-bit word-length. Illustrate with
a diagram the relationship between the analog input and digital output for a 3-bit,
±10 volts, twos complement ADC.
554 Computer Hardware for Process Control
8.16 Explain in a simple physical terms how you can convert a continuous signal to a
discrete-time signal and vice versa. Why are these two operations necessary in a
computer-process I/O interface?
8.17 What is easier and less costly to do?
(a) multiplex N analog signals first and then use a common A/D converter
(b) convert the N analog signals to digital first and then multiplex them to enter
the computer through a single, common word storage?
8.18 Explain with neat sketches the working of a three-bit digital-to-analog converter.
8.19 With neat illustrations discuss the different types of circuit-designs available for
executing DAC.
8.20 What is multiplexing? What are the methods of multiplexing? Explain each of them
with neat sketches.
8.21 What is modem? What are the types of modems available depending on their
transmission speed?
8.22 What is the use of digital interfaces? Discuss the different types of digital interfaces
used for digital data transfer between devices. Provide comparison between RS-
232C, RS-422 and RS-485 interfaces.
8.23 Draw a communication network hierarchy for a process industry showing different
process control levels. Explain the function of each communication level.
8.24 Discuss different types of transmission techniques used for data transfer between
two devices.
8.25 What is local area network? Explain its functioning in a process control
environment.
8.26 What is ISO network model? With neat sketch discuss the structure of ISO model.
Differentiate between the various layers of an ISO model.
8.27 What is CSMA/CD? For what purpose is this used? Explain with an example.
8.28 What is protocol? Differentiate between manufacturing automation protocol (MAP)
and technical office protocol (TOP).
8.29 What is the use of field buses in industrial process control system? What are
different types of field buses? Discuss the advantages and disadvantages.
8.30 Compare between PROFI-BUS and FIP-BUS.
SELF-CHECK QUIZ
A State True/False:
8.3 A bus can transmit data in either direction between any two components of the
computing system.
8.4 The main memory of a computer system is volatile.
8.5 The EISA bus is synchronous and can be used only in the 80386 and 80486 chips.
8.6 Multiplexing is a technique that allows number of lower band-width communication
channels to be combined and transmitted over higher bandwidth channels.
8.7 Cache memory provides a buffering capability in order to optimize the performance
of the CPU.
8.8 In a bus interface, the address lines indicate what the information is, whereas the
data lines provide information on where the data or information is to be sent.
8.9 The function of ZOH (zero-order-hold) in a DAC circuit is to hold a numerical value
constant until the next one arrives.
8.10 In time division multiplexing, the higher bandwidth channel is divided into many
individual smaller bandwidth channels.
8.11 Real time clock (RTC) is a mechanism whereby elapsed time can be monitored by
a computer system.
8.12 A serial transmission is most suitable for short distance communication, whereas
a parallel transmission is suitable for long distance communication.
8.13 Twisted pair cables are more expensive than coaxial cables for transmission of
data.
8.14 Both asynchronous communication interface adapter (ACIA) and universal
asynchronous receive and transmit (UART) are same device used for the serial data
transfer.
8.15 Glass-fibre cables have lesser transmission capability than coaxial or twisted pair
cables.
9.1 INTRODUCTION
Digital computers have become the primary means of implementing process control
algorithms. Real-time software is the medium in which these solutions are expressed.
No computer system can operate without a software, which is a program that helps
execute the tasks performed by the system. Computers are sequential devices, and as
such, can act only as sampled-data controllers, with time discretized. Fundamental
characteristics of the real-time software, as distinct form the regular (general) software,
are that the control algorithms must be executed at their scheduled sample intervals
(with some specified tolerance) and that associated software components, which interact
with the sensors and actuators, can have critical time-window constraints.
The nature and difficulty of producing a real-time software depends on the
complexity and timing constraints of the problem. The tighter the time constraints with
respect to the computers basic computing speed limitations and the more things that
need to be serviced simultaneously (at least simultaneously from the viewpoint of the
control object), the more difficult will it be to complete the software design and
implementation successfully.
In this chapter, details about computer software used for real-time applications,
have been discussed.
to be developed from old programs used previously for similar functions. There is also
a major trend to make all the programs as general as possible in order to enhance their
use for related applications, to minimize their individuality, and thus to reduce the cost
of developing the next system.
The desire for minimum individuality in process control computer programs has
resulted in a very common program organization, consisting of the following three major
parts:
1. System software or executive software
2. Application software or process control
3. System support software.
and business applications. Since the priority interrupt capability vitally affects the
overall management of the computer system, this function is usually included within the
executive (system software) program of the process control computer. The other time-
based and externally simulated functions are included in the application program of the
system.
A second major difference between process control programs and typical scientific
or business programs is the great dependence upon multi-programming or parallel
execution of functions. The time-based operation of the control computer, the large
amount of time required to complete many functions (such as printing-plant reports, and
reading process variables), and the necessity to check the status of many tasks, mean
that the computer must shift its attention between many different functions while
completing only small parts of any one function at any time. To keep track of the
ongoing status of each of the several tasks under way at any time requires a very
sophisticated executive program. All successful process control executive programs must
have this capability.
Figure 9.1 illustrates the operations that are executed by programming in the
process computer system. The diagram shows the overall system as carried out by a
single computer containing all functions. The modular system shown in the diagram
allows any particular module to be modified without affecting the other modules. This
greatly simplifies both the initial programming effort and later required program
modifications. This is made possible by the use of the data tables indicated in the
diagram. A further advantage of such a program system is that programs developed by
others for any of the modules can be readily integrated into the overall program.
Executive
program
Timing and priority information
Data transfer
interface Pertinent plant steady-state To actuator
model
To other computer
system
Figure 9.1 Process control computer programming system.
Application programs
U
til
rt iti
File manager es
po
sup
ge
m
ua
Sch
ste
ng
La
OS kernel
-sy
edu
sub
ler
Interrupt Software
Dispatcher
I/O
handler
Hardware
CPU
I/O devices
Operating system
OS
OS kernel
Kernel
In the 60s, 70s and 80s, a series of well-known real-time operating systems have
been launched for process control computer systems, based on IBM, VAX, PDP, GE and
other computers. Some of the popular OS are MS-DOS, UNIX, VMS, etc. Today new OS
have been developed for better manmachine interface (MMI) such as Windows NT,
Windows 95 and Windows 98, etc. Following types of operating systems are generally
used in process control environments:
Single-task or single-job
Foreground/background
Multi-tasking.
User
Ap
pl
CP d
(C man
ica
)
Basic disk operating tio
so m
n
es co
oc le system (BDOS) so
ftw
r
pr nso
ar
BIOS e
Co
Basic input/output
system Software
CPU Hardware
I/O devices
Disk
User
Application
CCP software
Software
BDOS
BIOS
Hardware
CPU Disk I/O devices
User
Common data
Computer hardware
Interrupt handling
Memory management
Code sharing
Device sharing
Inter-task communication and data sharing.
Computer hardware
ate
cre Executing
Executing In
ter
a sk ru
T Sc pt
hed
e ule
let
de or
int
sk
Event, awake
Ta err
up
Task dele t
te
Dormant Ready
Suspend
Tas
k cre
ate ke
Cre
ate awa
t,
Ta en
s k Ev
de d
let Suspended
Suspended p en
e S us
Figure 9.6 Task state transition diagram in the task-control block for RTOS.
running. A task enters the ready state if it was executing and its time slice runs out,
or if it was pre-empted. If it was in the suspended state, then it can enter the ready state
if an event initiated to occur. If the task is in the dominant state, then it enters the
ready state upon creation (if another task is executing). Tasks that are waiting on a
particular resource, and are not ready, are said to be in the suspended or blocked state.
The dominant state is used only in systems where the number of task-control blocks is
fixed. This state is best described as a task that exists but is not available to the
operating system.
When the task management is evoked, the operating system checks the ready list
to see if the next task is eligible for the execution. If it is eligible, then the TCB of the
currently executing task is moved to the end of the ready list, and the eligible task is
removed from the ready list and made the executing state.
In addition to rescheduling, the operating system checks the status of all resources
in the suspended list. If a task is suspended on a resource, that task can enter the ready
state. The list structure is used to arbitrate two tasks that are pending on the same
resource. If a resource becomes available to a pending task, then the resource tables are
updated and the eligible task is moved from the suspended list to the ready list.
is held in ROM, the need has arisen for modular operating system in which only the
parts which are required for a particular application are stored in the ROM.
In concurrent programming, it is assumed that there is a processor available for
each task; no assumption is made as to whether the processor will be an independent
unit (i.e. a separate CPU) or simply a single CPU, which is time-shared between the
tasks. Furthermore, no assumption is made either about the timing or the priority of
any task. Tasks, which require to exchange information or synchronize for other
purposes, do so by means of signals. Another assumption, which is made, is that a
number of primitive instructions exit. These are instructions, which are parts of a
programming language or the operating system and their implementation, and
correctness is guaranteed by the system.
programming languages for embedded real-time (process control) systems. Today, C++,
Visual C++ and Visual Basic have become very popular programming languages for
real-time application software developments. These languages provide better man
machine interface (MMI). Higher-level languages for process control applications may be
divided into two major branches: general-purpose languages and problem-oriented
languages.
General-purpose languages
Figure 9.8 shows these two branches and their further subdivision into several different
categories:
Higher-level languages have the following two major advantages over an assembler
or any other so-called lower-level language:
The same program can be run on any type of computer for which the
corresponding translator (also called compiler or interpreter) program has been
written. That is, higher-level languages are not usually machine architecture
dependent as are assemblers.
The average user interprets usually much more rapidly their symbology than
those of the assembler or machine languages they replace; that is, they are
more like normal English or mathematical statements.
A compiler takes the code (program) as written by the programmer and converts
it completely to the machine language of the particular computer. This translation may
include converting the program into an assembler language as an intermediate step.
An interpreter, on the other hand, stores the program in the computer in
approximately the form developed by the programmer. The computer then executes the
program one line at a time by interpreting, or converting, each line to machine language
and carrying out the instructions included therein before going to the next line.
Real-Time Application Software 573
When compiling a program, the computer can sense the complexity of the entire
program and incorporate some means to handle it. This look-ahead capability is, by
definition, not available with the interpreter. Compiler languages differ among
themselves in terms of their capabilities. Those that permit complex programs to be
readily written in them are called systems programming languages. Some of these can
be used in the preparation of very difficult programs, such as large computer operating
systems or executive programs. Otherwise these programs must be written in the
assembler language of the particular computer.
Problem-oriented languages
Problem-oriented languages are those that have been especially developed for a
particular type of application, such as process control applications. In fact, many
problem-oriented languages have been especially developed for relatively narrow types
of applications within process control, such as control of batch reactors.
As indicated in Figure 9.8, problem-oriented languages can be further divided into
compiler-type languages and fill-in-the-blanks languages. One of the major types of fill-
in-the-blanks programs originated with the general-purpose control programming
(GPCP), developed by Humble Oil and Refining Company. The system was later picked
up by the IBM as PROSPRO and by General Electric as BICEPS.
Compiler-type languages are usually modified or extended versions of the general-
purpose compiler languages. FORTRAN and BASIC have been especially popular as the
basics for such efforts. A well-known example of this kind of program is AUTRAN
(Automatic Utility TRANslator), which was originally developed by the Merck, Sharp and
Dohme Automation Department and perfected and marketed by Control Data
Corporation.
The fill-in-the-blank system is a largely prewritten, interpreter-type program that
offers a large number of choices in its execution in addition to a well-developed skeleton
data base. By means of responding to a series of questions or a menu displayed on CRT
(VDU) screen, the system developer or programmer is able to make a proper choice of
algorithms for each of the control loops from the available options. The programmer can
also supply all the necessary system parameters to the computers database. The result
is a completely operative control program for the process, developed from the CRT-
displayed options. Program development may also be handled by means of a set of
preprinted forms, one for each control loop, one for each analog or digital input, one for
each process function, etc. These forms are then read into the computers memory.
The prewritten fill-in-the-blanks program must be interpretively executed, since it
is developed with no knowledge of how the final programmer will organize the execution
of the final program or how many and what type of functions will be involved.
All these real-time system application development languages must take into
consideration the following users requirements:
Security, a measure to the extent to which a language is able to detect errors
automatically either at compile time or through run-time support system.
Economically, it is essential to detect errors at the compilation stage rather
than at run-time, since the earlier the error is detected the less it costs to
correct it.
574 Computer Software for Process Control
Reliability, a measure of the ease with which the operation of a program can
be understood without resort to supplementary documentation such as
flowcharts or natural language descriptions. It results in reduction in
documentation cost, easy error-detection and easy maintenance.
Flexibility, a measure of less number of uses of assembly level (or hard-wired)
coding.
Portability, a measure of being able to transfer a program from one computer
to another and able to compile and run on the target computer. This features
speeds up developments, reduces cost, and increases security. It is often difficult
for real-time process control systems in that they often make use of specific
features of the computer hardware, the field devices (sensors and instruments),
and the real-time operating systems.
Simplicity, contributing to the security and thereby reducing the cost of
training, reducing the portability of programming errors arising from
misinterpretation of the language features.
reliability of complex systems, and techniques are available for making such systems
fault tolerant. In hardware, the emphasis is on improved components and construction
techniques and the use of redundancy to maintain critical systems functions in the event
of single failures or multiple failures.
In software, the emphasis has been on improved software construction techniques
using software engineering tools. In some applications, redundancy is introduced into
the software systems. The use of redundancy gives the system resilience. It can be used
to limit the scope of the faults or mask errors so that they do not lead to failure, to
protect inter-process communications in distributed systems, and to ensure the integrity
of decision processes in decentralized process control systems. In addition, in time-
critical systems, it can be used to ensure that results are computed in a timely manner.
Following software fault tolerance techniques are used:
Fault masking, in which the massive redundancy of software is used to mask
errors. The most common technique is N-modular redundancy, proposed by
Avienzis, which involves the n-fold replication of processes. The replicated
processes are operated either in serial or in parallel and the results computed
by the variant paths are compared. If an error is generated by one path, it will
be detected and the erroneous result can be suppressed, thus masking the fault.
The usual replication factor is three and diverse techniques are used in the
programming of the triplicated paths.
Error detection and recovery, in which the method detects the error, assess the
damage caused by the error, and initiate an error-recovery mechanism, in order
to compute an error-free result with the minimum delay. Two recovery
techniques are used: the forward recovery technique and backward recovery
technique. The forward-error-recovery technique minimizes the delay by
transforming the erroneous state directly into a correct or acceptable state. In
backward-error-recovery technique, the system moves backwards through
previous states until it can restore a previously saved correct state and restart
processing in the forward direction.
Watchdog mechanisms, in which the performance of time-critical application
processes is monitored. This is done using a real-time time-lapse counter, which
is preset to trip after a predetermined time. The predetermined time is set
somewhat less than the time-critical period so that the fault recovery can take
place and the system can still provide a timely and satisfactory response.
Protection of communications, in which the communication medium of a
computer-aided process control system may not be immune to faults and must
be regarded as a potential source of errors. Typical errors are erroneous
messages or messages which fail to arrive. In loosely coupled systems, such
errors may have a serious effect because inter-process messages are used to
ensure the proper synchronization and operation of the distributed systems. In
synchronous systems, prompt recognition of a communication fault is essential
if recovery actions are to be initiated and deadlock is avoided. In real-time
systems, these actions must be carried out within a critical time period.
Software fault-tolerance techniques can be used in concurrent real-time systems
to protect the inter-process communications. Communication transactions can
be protected using watchdog timers.
576 Computer Software for Process Control
Single task
The simplest design strategy is to treat the whole of the software system as a single
Design of Software for Process Control (Real-Time) Systems 577
program written using the techniques applicable to a sequential program. This approach
is recommended for simple systems with a small number of inputs where the response
time for each input is similar, and the time taken for the various paths through the
procedures forming the control loop does not vary greatly from cycle to cycle. As the
number of inputs, actions to be performed, and outputs increases, the range of
environment response time, usually increases. It thus becomes more difficult to satisfy
the minimum time requirement, if all the actions are carried out at every control cycle.
One strategy is to partition the inputs and actions according to the required
environment response time.
Two-tasks (foregroundbackground)
A commonly used design approach is to partition the system into two sections, usually
referred to as the foreground and background partitions. The typical division is to place
the time-dependent actions in the foreground and the time-independent actions in the
background. The general rule for forming the partition is that the number of actions
placed in the foreground partition should be minimized. An implicit assumption is that
in the system there will be a single task in each partition, thus limiting problems of
resource sharing and synchronization to resolving conflicts across the partition
boundary.
Multiple task
A natural extension of the foregroundbackground division is to partition the system
into many subsets and treat each subset as a separate task. That is, the software is
divided into number of separate programs. The multi-tasking approach developed
during a period when it was assumed that the implementation would be on a single
processor with a real-time operating system being used to share resources between the
tasks. But today, the tasks are distributed to more than one processor for faster and
better operation.
MASCOT system
The MASCOT is the earliest formal methodology for assisting the development of real-
time software, the first version of which was developed by Jackson and Simpson during
the period 197175. Extensive changes were made to the technique and its version.
The official standard for MASCOT-3 was published in 1987.
The name MASCOT is an acronym for Modular Approach to Software
Construction, Operation and Test. It is, first and foremost, a consistent methodology to
be applied to the complete life cycle of a software project from the initial understanding
and the clarification of the requirement through the maintenance and extension of the
operational system. MASCOT incorporates the following features:
· Design representation
· Deriving the design
· Constructing software so that it is consistent with the design
· Executing the constructed software so that the design structure remains visible
at run time
· Facilities for testing the software in terms of design structure.
Figure 9.9 illustrates the interactions involved in MASCOT. The operators and the
software exchange information collaborate with each other in order to achieve, via
sensors and actuators, a more effective interaction with the process or plant.
Process
Software Operators
Structured system analysis. Structured system analysis uses the following tools to
build structured specification of a software:
· Data flow diagram
· Data dictionary
· Structured English
· Decision tables
· Decision trees
Following are various steps involved in structured analysis:
Step 1-Draw the data flow diagram (DFD): The DFD is a pictorial representation
of all aspects of the logical data flow. It uses four basic symbols (as per Gane and
Sarsen), as shown in Figure 9.10.
With the help of the above symbols, data flow diagram of a software problem
is drawn and further refinement (breakdown) is done till a logical flow of data is
achieved.
Step 2-Put in the details of data flow: Data items are identified that are required
to go into various data flows. In case of large system, a data dictionary is created to keep
track of the various data elements involved.
580 Computer Software for Process Control
Arrow
Flow of data
Step 3-Define the logic of processes: The logical steps (and algorithm) within each
process is determined. For developing the logic within the process, decision tree and
decision table techniques are used.
Step 4-Define data store: Exact content of each data store and its format are
defined. These help database design and building database.
Step 5-Define the physical resources: Now that the designer (developer) knows
what is required online and the format of each element, blocking factors are decided.
In addition, for each file, the file name, organization, storage medium, and records,
down to the field level, are specified.
Step 6-Determine the input/output specifications: The input and output forms are
specified. Input screens, display screens, printed output format, are decided.
Step 7-Perform sizing: The volume of input (daily, hourly, monthly, etc.), the
frequency of each printed reports, the size and number of records of each type that are
to pass between the CPU and mass storage, and the size of each file are estimated.
Step 8-Determine the hardware requirements: Based on the information estimated
in Step 7, the hardware configuration such as, storage capacity, processor speed, CPU
size, etc., is decided.
Determining the hardware configuration is the final step of Gane-and-Sarsens-
specification method. The resulting specification document, after approval by the client,
is handed over to the design team, and the software process continues. Figure 9.11
Design of Software for Process Control (Real-Time) Systems 581
Reports/trends
Heat start Equipment data
Operator Laboratory
Equipment data
update
Process data
Sample analysis data
Data
Heat data
handling Heat information
Production
Process events
schedule Tracking
Production data
Heat data
Interface Raw material schedule
with field weights Process
data bus tracking Heat Model
Production schedule data result
Production status
Confirm
activity Process
modelling
Trigger event
Figure 9.11 A typical example of DFD for modelling of steel making process.
illustrates an example of data flow diagram (DFD) for processing modelling of a steel
making process.
Structured design. Structured design is a specific approach to the design process
that results in small, independent, black-box modules, arranged in a hierarchy in a top-
down fashion. Structured design uses following tools to build the systems specifications
document:
· Cohesion
· Coupling
· Data flow analysis.
Cohesion of a component is a measure of how well it fits together. A cohesive
module performs a single task within a software procedure, requiring little interaction
with procedures being performed in other parts of a program. If the component includes
part which are not directly related to its logical function, it has a low degree of cohesion.
Therefore, cohesion is the degree of interaction between two software modules.
Constantine and Yourdon defined seven categories or levels of cohesion in order to
increase the strength of cohesion from the lowest to highest:
· Coincidental cohesion (Bad)
· Logical cohesion
· Temporal cohesion
· Procedural cohesion
· Communicational cohesion
· Sequential cohesion
· Functional cohesion. (Good)
582 Computer Software for Process Control
9.7.5 Algorithms
An algorithm, by definition, is the description of the mechanization of a particular
equation by the digital computer. Algorithms are set of several equations that are used
in the process control field, especially for the implementation of PID control.
In DCS control, the digital computer computes the actual movement of the valve
or other final actuator. In supervisory control, on the other hand, the computer merely
computes the required change in the variables set-point, and the final control actuation
is left to an analog electronic or pneumatic controller.
There are two major approaches for digitally representing the action of
conventional three-mode pneumatic or electronic controllers, position and velocity
algorithms. For position algorithm, the computer output is the corrected valve (or other
final control element) position and integration is done on the computer. For the velocity
algorithm, the computer output is the change that the valve should undergo between
sampling periods. Therefore, integration must be done by the final control element via
a stepping motor, integrating amplifier, or other similar devices.
584 Computer Software for Process Control
Position algorithm
The three-mode controller can be represented by
P
'G
2P - R G - &
'V
- K Ç G'V 2/ (9.1)
with
e = S V (9.2)
where
Pn = Valve position at time n
PM = A median valve position
Kp = Proportional-mode gain
Ki = Integral-mode gain
KD = Derivative-mode gain
e = Error
S = Set-point
V = Variable.
The position algorithm requires that the computer recalculates the full value of the
valve setting increment at each time. In addition, this value must be transmitted to the
valve positioner as an analog signal, requiring a digital-to-analog converter (DAC) or the
equivalent for each output.
Velocity algorithm
If the computer output goes to a stepping motor or integrating amplifier, the computer
simply calculates the required change in valve position. The output is a digital pulse
train, where a stepping motor is used. The stepping motor in turn drives a slide wire
and the slide wire output is proportional to the correct valve position. Therefore, the
combination of stepping motor (digital correction) and slide wire (analog signal to the
final control element) acts as a digital-to-analog converter (DAC). No special DAC is
required, and computation word length in the computer may possibly be reduced.
The velocity algorithm output Yn is found by subtracting the outputs of two
successive position algorithm calculations, and is given as
Yn = Pn Pn1 (9.3)
-
- R
GP GP - K GP 'V &
GP GP GP GP (9.4)
'V
r + e f u y
Controller Actuator Plant
z
Sensor
The block diagram representation simplifies the control system analysis, and
thereby, indirectly, also the control system design for control engineers. Furthermore,
it is the mechanism that helps with decoupling the reaction of the control system to
different control inputs. Most simulation software vendors offer today a block-diagram
editor as a graphical modelling tool. Unfortunately, block-diagram editors are not a
panacea for all the graphical modelling needs. Block diagram editors are not the right
tool to model a physical plant, e.g. electrical circuits or multi-body systems.
However, the controllers, after having been designed in an abstract fashion, need
to be implemented using physical components. Although the control engineer can choose
these components, they may still have some non-ideal characteristics, the effect of which
ought to be analyzed before the system is actually built. In this case, even the controller
blocks become physical systems, and the same restrictions that were previously
mentioned with respect to the physical plant to be controlled apply to them as well.
In summary, block diagrams are only useful to describe the higher levels of the
control system architecture, but are rarely a good choice for describing the physical
layer at the boot of the hierarchy.
Hierarchical modelling
Control systems are frequently built like an onion. One control loop encompasses
another. For example, it is quite common that a local nonlinear control loop is built
around a nonlinear robot arm, with the purpose of linearizing the behaviour of the robot
arm, such that, from the outside, the arm with its local controller looks like a linear
system. This system then becomes a block in another control system at a hierarchically
higher level. The purpose of that control layer may be to decouple the control inputs
from each other, such that each control input drives precisely one link (the physical
586 Computer Software for Process Control
configuration may be different). This control system then turns into a series of blocks
in a yet higher-level control configuration, in which each individual control input is
controlled for performance.
Evidently, control engineers would like their block-diagram editors to behave in
exactly the same fashion. One entire block diagram may become a single block at the
next higher level. Most of the block-diagram editors currently on the market offer such
a hierarchical decomposition facility.
Plant modelling
One part of control system to be simulated is the plant to be controlled. As was
mentioned earlier, this plant can represent anything that is controllable. It can be a
thermal power plant, or a chemical processing plant, or an aircraft, any other process
industry. In order to test the control design, the control engineer should be able to
simulate the plant with the control system around it.
Linearization
One way that control engineers deal with control systems is to linearize the plant (or
process) or at least part thereof to be controlled. This then enables them to perform the
controller design in a simplified fashion, since there exist analytical controller design
strategies for linear systems, whereas the nonlinear system control design would have
to be done by trial and error.
Control engineers want the linearization of the original model to be done in an
automated fashion. Moreover, this has to happen inside the modelling environment,
since the original nonlinear model needs to be interpreted in this process.
This feature is very important for control engineer. They want to be able to
compare the behaviour of the linearized model with that of the original nonlinear model
before they go about designing the controller. Then, after the controller has been
synthesized, they would like to simulate the control behaviour of the controller when
applied to the original nonlinear process. Finally, they may want to use the linear
control system design only as a first step on the way to determining an appropriate
controller for the original nonlinear process. The so-synthesized controller can be
interpreted as an approximation of the final parameter values. The parameters are then
fine-tuned using the nonlinear plant model together with some parameter optimization
algorithm.
Some of the currently available environments, such as ACSL and SIMULINK, offer
a limited model linearization capability. A linear model of the type:
Z #Z $W (9.5)
is obtained from the original nonlinear model:
Z H
Z WV (9.6)
by approximating the two Jacobians:
H H
# and $ (9.7)
Z W
Software for Modelling and Simulating Control Systems 587
through numerical differences. The facility is limited in the following three ways:
(a) There is no control over the quality of the numerical difference approximation,
and thereby the linearization. The problem can be arbitrarily poorly
conditioned. A symbolic differentiation of the model to generate the Jacobians
may be more suitable and is entirely feasible.
(b) The approximation is necessarily local, i.e. limited to an operating point
< x0, u0 >. If, during simulation, the solution starts to deviate a lot from this
operating point, the approximation may be meaningless.
(c) The approximation makes the assumption that the state variables must be
preserved. This assumption may be too strong. If a sub-system is represented
by the state-space model:
Z H
ZWV and y = g (x, u, t) (9.8)
one may wish to preserve is the input-output behaviour, but this behaviour should be
preserved over an entire trajectory or even a set of trajectories. This can often be
accomplished by a model of the type:
\ #\ $\ (9.9)
y = Cz + Du (9.10)
if only the length of the linear state vector z is chosen sufficiently larger than that of
the original state vector x.
Parameter identification
Contrary to the process parameters that can be determined (at least in an approximation
fashion) form physical considerations, controller parameters are technological
parameters that can be freely chosen by the designer. Hence, a tool is needed to
determine optimal controller parameter values in the sense of minimizing (or
maximizing) a performance index.
Although some simulation environments offer special tools for parameter
identification, they all proceed in a purely numerical fashion. Let p be the vector of
unknown parameters, and PI the performance index to be optimized. It is fairly
straightforward to augment the model at compile time by a sensitivity model that
computes the sensitivity factor 2+
R
. If there are k parameters and n equations in the
original model, the augmented model will have n(k + 1) equations.
The control engineer can then look at the magnitude of the sensitivity parameters
as a function of time, and pick a subset of those (those with large magnitudes) for
optimization. Let us assume the reduced set of parameters pr is of length kr (< k).
Optimizing PI(pr) implies making 2+ R
= 0. The latter problem can be solved by Newton
iteration:
2+ N 2+ N
EN (9.11)
RTN RTN
and
RTN RTNE N (9.12)
588 Computer Software for Process Control
Each equation implies solving the augmented set of the original equations and the
equations partially differentiated with respect to design parameters. Even equations for
the Hessian matrix (the second partial derivative) can be symbolically generated at
compile time, if the code is simultaneously generated that prevents these additional
equations from being executed during each function evaluation.
Frequency domain
Control engineers like to switch back and forth between the time domain and the
frequency domain when they are dealing with linear (or linearized) systems. Most
simulation systems offer the capability to enter blocks as transfer functions.
The polynomial coefficients are used in a set of differential equations (using the
controller-canonical form), thereby converting the transfer function back into the time
domain.
Although this feature is useful, it does not provide the control engineer with true
frequency analysis capabilities. Control engineers like to be able to find the bandwidth
of a process or plant, or determine the loop gain of a feedback loop. Such operations are
much more naturally performed in the frequency domain, and it seems useful,
therefore, to have a tool that would transform a linear (or linearized) model into the
frequency domain, together with frequency domain analysis tools operating on the so-
transformed model.
Real-time applications
Control systems are often not fully automated, but represent a collaborative effort of
human and automated control. Human operators alone, because of the time-critical
nature of the decisions that must be reached, cannot control complex systems (such as
power-generation or an aircraft system or other real-time industrial process). Humans
are not fast and not systematic enough for this purpose. Yet, safety considerations
usually mandate at least some human override capability, and often, humans are in
charge of the higher echelons of the control architecture. That is, they are in control
of those tasks that require more intelligent and insight, yet are less time-critical.
Simulation of such complex control systems should allow human operators to drive
the simulation in just the same manner as they would drive the real system. This is
useful for both the system debugging as well as the operator training. However, since
humans cannot be time-scaled, it is then important to perform the entire simulation in
real time.
Another real-time aspect of simulation is the need to download controller design
into the digital controllers that are used to control the actual process once the design
has been completed. It does not make sense to ask the control engineer to re-implement
the final design manually in the actual controller, since this invariably leads to new bugs
in the code. It is much better if the modelling environment offers a fully automated real-
time code generation facility, generating real-time code in a particular computer
programming language (C, C++, FORTRAN, Ada, etc.).
Finally, some simulators contain hardware in the loop. For example, flight
simulators for pilot training are elaborate electromechanical devices. The purpose of
these simulators are to make the hardware components behave as closely as possible
Software for Modelling and Simulating Control Systems 589
to those that would be encountered in the real system. This entails simulated scenery,
simulated force feedback, possibly simulated vibrations, etc. Evidently, these simulations
need to be performed in real time as well.
MATLAB
MATLAB (MATrix LABoratory) is an interactive software program for scientific and
engineering calculations. It is used for computer-aided analysis and design of control
systems. The MATLAB consists of the base program plus a variety of toolboxes (called
control system toolbox), and collection of special files called M-files that extend the
functionary of the base program. Most of the statements, functions and commands are
computer-platform-independent. Regardless of what particular system you use, your
interaction with MATLAB is basically the same. MATLAB interprets and acts on input
in the form of one or more of the objects such as statements and variables, matrices,
graphics and scripts.
Originally designed as a teaching programme, giving interactive access to the
linear algebra routines, EISPACK and LINPACK, MATLAB was released into the public
domain in around 1980. Powerful though the basic matrix calculator capabilities of
MATLAB are, its real flexibility is due to its support of macro files (or M-files). A macro
file, in its simplest form, is just a collection of ordinary MATLAB commands which are
stored in a file. When called, such a script of commands is executed just as if it had
been typed by the user. MATLABs real strength lies in its ability to use M-files to create
new functions.
MATLAB underwent further developments. It was rewritten in C programming
language for efficiency and enhanced portability and released as a commercial product
in 1985. Like its competitors, the main market was initially the CASD (discussed latter
in this chapter) market, where, supported by two sets of toolbox extensions called the
control and signal processing toolboxes, MATLAB made rapid inroads into academia
and industry. A recent development has been the provision of add-on graphical input of
system models, in the form of block diagrams, support for point-and-pick nonlinear
simulation, and enhanced graphical functionality. At least one package, MATRIXx has
evolved further by the addition of data structures and more sophisticated support for
macro development.
Application of the many classical and modern control system design and analysis
tools is based on mathematical models. MATLAB can be used with systems given in the
form of transfer function descriptions. MATLAB can be used in manipulating
polynomials, computing poles and zeros of transfer functions, computing closed-loop
590 Computer Software for Process Control
transfer functions, computing block diagram reductions, and computing the response of
a system to different frequency responses.
these programs often remains clearly visible through the new interface. Some of the
general-purpose simulation languages are briefly discussed here:
· ACSL (from Mitchell & Gauthier Association): Advanced Continuous
Simulation Language (ACSL) available since 1975 has long been the
unchallenged leader in the market of simulation languages. This situation
changed in recent years due to success of SIMULINK and SystemBuild. ACSL
is a language based on the Continuous System Simulation Language (CSSL)-
standard. An ACSL program is preprocessed to FORTRAN for platform
independence. The resulting FORTRAN program is then further compiled to
machine code. As a consequence, ACSL simulations always run efficiently, which
is in contrast to the simulation code generated by most block-diagram
simulators. User-defined FORTRAN, C and Ada functions can be called from an
ACSL model. ACSL can handle ODEs and DAEs, but no over-determined DAEs.
For a long time already, ACSL has supported state-event handling in a
numerically reliable way (by means of schedule statement), such that
discontinuous elements can be handled. Recently, ACSL has been enhanced by
a block-diagram front-end, a post-processing package for visualization and
animation, and a MATLAB-like numerical computation engine.
A block in ACSLs block-diagram modeller can take any shape and the input/
output points can be placed everywhere, contrary, e.g. to the much more
restricted graphical appearance of SIMULINK models. Consequently, with
ACSL, it is easier to get a closer correspondence between reality and its
graphical image. Unfortunately, ACSL is not (yet) truly modular. All variables
stored in a block have global space. This means that one has to be careful not
to use the same variable name in different blocks. Furthermore, it is not
possible to define a block once, and to use several copies of this block. As a
result, it is not convenient to build up user-defined block libraries. ACSL is
running from PCs to supercomputers. With more than 25 years of experience,
ACSL is fairly robust, contains comparatively decent integration algorithms, and
many small details that may help the simulation specialist in problematic
situations. Although the ACSL vendors have lost a large percentage of their
academic users to SIMULINK, ACSL is still fairly popular in industry.
· Simnon (from SSPA Systems): Simnon was the first direct-executing, fully
digital simulation system on the market. Designed originally as a university
product, Simnon is a fairly small and easily manageable software system for the
simulation of continuous-time and discrete-time systems; Simnon offered, from
its conception, a mixture between a statement-oriented and a block-oriented
user interface. Meanwhile, a graphical front-end has been added as well. Simnon
has been for years a low-cost alternative to ACSL, and enjoyed widespread
acceptance, especially in academia. Due to its orientation, it suffered more than
ACSL from the SIMULINK competition.
· Desire (from G.A. and T.M. Korn): Desire is another direct-executing simulation
language, designed to run on small computers at impressively high speed. It
contains a built-in micro-compiler that generates machine code for Intel
processors directly from the model specification. Since no detour is done through
a high-level computer language, as is the case in most other compiled simulation
Software for Modelling and Simulating Control Systems 593
the analysis and design of linear plant and controllers. Many modern packages also
provide support for the modelling, simulation and linearization of nonlinear systems and
some have the capability of implementing a control law in software.
Modern CACSD has been made possible by the synthesis of several key
developments in computing. The development and continued dominance of high-level
procedural languages such as FORTRAN, C, C++, etc., enabled the development and
distribution of standard mathematical software. The emergence of fully interactive
digital computer operating systems such as UNIX and its user Shells influenced the
development of CACSD packages, which have been constructed along similar lines. The
ready availability and cheapness of raster-graphic displays has provided the on-screen
display of data from control systems analysis, and the creation of tools for modelling
control systems using familiar block diagrams. CACSD has potential to make order-of-
magnitude improvements in the ease-of-use, ease-of-manipulation and efficiency of the
interaction between the control designer, his model, analysis tools and end-product-
software for embedded controllers. The driving force of all these developments is the
seemingly continual increase in computing power year-on-year and the result has been
to make computers accessible to large numbers of people while at the same time making
them easier to use.
A control engineer often describes systems through the use of block diagrams. This
is not only the traditional graphical representation of a control system but also an
almost discipline-independent, and thus universally understandable, representation for
dynamic systems. The diagrams may also constitute a complete documentation of the
designed system. Block diagrams are self-documenting and, when appropriately
annotated, may form complete and consistent specifications of control systems.
The main features of CACSD are analysis tools built around a matrix environment
and modelling, simulation, and code generation tools constructed around the block
diagram representation. For the most part, control systems analysis and design is done
from textual interface and modelling, simulation, and code generation rely on a
graphical user interface. There are links between the two environments, usually
provided by some form of linearization. In addition to serving as a documentation aid,
the overall cost and cycle time for developing complex controllers is radically reduced
if analysis/simulation code and/or real-time code is automatically generated from the
block diagrams. This eliminates time-consuming manual coding, and avoids the
introduction of coding bugs.
traditionally generated code, primarily due to error reduction of not hand coding. This
reduction of costs provided the evidence that automatic translation of block diagrams is
both feasible and desirable. However, due to advances in both computer graphics and
code-generation techniques, the first tool was obsolete by the late 1980s. In recent years,
several commercial block-diagram-based tools are available. These tools include
SystemBuild from Integrated Systems Incorporated, Model-C from Systems Control
Technology, the PC-based Xanalog from Xanalog, Simulab/Simulink from the Math
Works, and BEACON from General Electric. Some of these tools primarily serve as
interface to analysis packages such as MATRIXx (SystemBuild), CTRK-C (Model-C) and
MATLAB (Simulink). In some cases, they can also be used to directly generate a
computer language such as C, Ada or FORTRAN.
many of todays block diagram tools have been developed as companion packages by the
same commercial vendors that also sell matrix environment. Through linearization, it
thus becomes possible to transform a nonlinear block diagram to a linear
representation, which can then be analyzed and used for design in the matrix
environment.
Postscript Paper
documentation
FORTRAN
Code Simulation
Netlist
generation
Ada
Optimize memory Controller
Graphical block diagram editor Optimize run-time Assembler hardware
and user interface
Code usage
Figure 9.13 BEACON architecture.
Computer (PC) Based Data Acquisition System 599
except graphical location contained in the block diagram is written to the ASCII
netlist file.
· An automatic code generator, which translates the block diagrams into
simulation and/or real-time computer code. For each block defined in the
graphical editor, the code generator excepts to find a block description language
(BDL) definition as described later.
The BEACON architecture is one of the most open and extendible in the industry,
allowing for straightforward extensions to the capability of the system and easy
interfacing to other systems. Therefore, the architecture of other block diagram
environments are often variants of that of BEACON. Some of the most common
differences found in other systems are:
· Built-in simulation capabilities: Many of todays commercial systems have a
nonlinear simulation engine directly built into the system, avoiding BEACONs
explicit translation step. Simulation results may then also be directly displayed
on or accessed from the original diagram (e.g. in the form of time histories). This
allows the user to immediately see the effects of any changes made to the
diagram. One drawback of this approach is that these non-compiled approaches
all have some kind of threaded-code or interpretative model execution, leading
to much slower simulations than explicitly compiled simulation models such as
those coming out of BEACONS. Some systems allow for either of the two
approaches.
· The avoidance of an explicit netlist: Many systems have a monolithic
architecture with no direct access to the information in a modelled system. This
prevents users from directly interfacing the block-diagram editor to other tools
or filters.
· No code generation: Some older systems have built-in simulation capabilities
only, with no generation of real-time or explicit simulation code.
Transducer-1
Transducer -1 Signal
conditioner-1
Physical plant parameters
Signal
Transducer-2
Transducer -2
conditioner-2 Multiplexer A-D
(telemetry) converter
Signal
Transducer-3
Transducer -3
conditioner-3
Computer
Computer
" (for
(fordata
data processing)
processing
Transducer-N
Transducer -N Signal
conditioner-4 Recording and display
The output of the signal conditioner is fed to the multiplexing (telemetry) device.
With the help of multiplexing, all individual signal data (called lower bandwidth
communication channels) are combined and transmitted over a higher bandwidth
channel as explained in Section 9.3.1. At the receiving end, de-multiplexing recovers the
original lower bandwidth channels. It scans across a number of analog signals and time-
sharing them sequentially into a single analog output channel. The multiplexed data is
converted into digital signal with the help of analog-to-digital converter (ADC), as
discussed in detail in Section 9.3.1. The converted digital signals are fed to the computer
for further processing, mathematical computation, storage, etc. The final and processed
data is either displayed on electronic digital display panel or recorded on magnetic media
and chart recorders.
Data acquisition system is generally, a process of collecting input signal in analog
or digital form, as rapidly, as accurately, completely and economically as necessary and
possible. Data acquisition system can be of two types: an analog data acquisition system
and a digital data acquisition system. Analog data acquisition system deals with
information in analog form, whereas digital data acquisition system handles information
in digital forms.
REVIEW QUESTIONS
9.1 What is a software? What are its types? What are features of software for process
control applications?
9.2 What are the major differences between programming in the process control
applications and programming in general scientific and business applications.
9.3 Illustrate with diagram the operations that are executed by programming in the
process computer system.
9.4 What is an operating system (OS)? Explain the functions of OS with a block
diagram indicating its various components.
9.5 What is a real-time operating system (RTOS)? Draw the structure of RTOS.
Review Questions 601
9.6 Explain with its structure the difference between single-task single-user operating
system and multi-tasking operating system.
9.7 What is concurrent programming? Explain its importance.
9.8 Discuss the different types of real-time application languages for process control
application. What are features?
9.9 What is software fault tolerance? What are software fault tolerance techniques
used?
9.10 Explain with neat sketch the working of a PC based data acquisition system (DAS).
9.11 What should be the design strategies for a real-time software system for process
control application?
9.12 What are the design techniques and tools for designing a real-time process control
software?
9.13 Discuss in detail the structured system analysis and design (SSAD) method for
designing software for real-time applications.
9.14 What are real-time programming languages? Discuss the features of these
languages.
9.15 Discuss about the software for simulation of continuous-time system. What are
special requirements for a simulation tool?
9.16 Discuss about standard modelling and simulation software available for process
control applications.
9.17 Discuss the use of digital computers for computer-aided control system design
(CACSD).
9.18 The step response of the following differential equation is to be obtained
numerically with the aid of a digital computer.
F [ F[
[
FV FV
F[
where and y = 0 at t = 0
FV
Integration step sizes (Dt) of 0.1, 0.5 and 1.0 are to be used.
(a) Which of the step sizes will give a numerical solution closest to the analytical
solution?
(b) Which step size will require the least computation time?
(c) If it is possible to get an impulse response for the above differential equation,
show how you would provide for it in solving the differential equation by the
computer.
602 Computer Software for Process Control
SELF-CHECK QUIZ
A State True/False:
9.1 Computers are sequential devices, and as such, can act only as sampled-data
controllers, with time discretized.
9.2 The system, application and system support software, are common within the
computer control systems used for industrial process control.
9.3 Real time operating system does not have any control on the resources of a
computer system such as processors, memories, mass storage devices, input and
output facilities, etc.
9.4 A real-time operating system (RTOS) is used for the process control computer
application.
9.5 In foreground-background systems, also called two-task system, the system is
partitioned into two sections, usually referred to as the foreground and background
partitions.
9.6 Typically, the time-dependent actions (tasks) are placed in the background section
and time-independent actions are placed in the foreground section.
9.7 In round robin system, several processes are executed parallel to completion, often
in conjunction with a cyclic executive.
9.8 Systems that use pre-emption schemes instead of round robin or first-come-first-
serve scheduling are called round-robin system.
9.9 In a multi-tasking or multi-programming system, the partition of system is done
into many subsets and each subset is treated as a separate task.
9.10 In multi-programming the operating system ensures that each user can run multiple
programs as if they had the whole of the computer system for their program.
9.11 Data acquisition system is generally, a process of collecting input signal in analog
or digital form, as rapidly, as accurately, completely and economically as necessary
and possible.
9.12 The MASCOT is the earliest formal methodology for assisting with development of
real-time software, the first version of which was developed by Jackson and
Simpson during the period 197175.
9.13 SSAD is a software engineering approach to the specification, design, construction,
testing and maintenance of software for maximizing the reliability and main-
tainability of the real-time system as well as for reducing software life-cycle costs.
9.14 Cohesion of a component is a measure of interconnections among modules in the
software.
9.15 Coupling is a measure of how well it fits together.
9.1 The fundamental characteristics of the real-time software, as distinct form regular
(general) software, are that the
(a) Control algorithms must be run at their scheduled sample intervals (with some
specified tolerance)
Self-Check Quiz 603
(b) Associated software components, which interact with the sensors and
actuators, can have critical time-window constraints
(c) Both (a) and (b)
(d) None of these
9.2 The nature and difficulty of producing real-time software depends on the
(a) Complexity
(b) Timing constraints of the problem
(c) Both (a) and (b)
(d) None of these
9.3 System software or executive program
(a) Consists of all the programs that supervise the overall operations of the
computer system
(b) Aids the user in the development of application programs
(c) Is a set of tools for development, generation, test, run, and maintenance of
programs, disregarding their application
(d) All of these
9.4 The main components of systems software are
(a) Real-time operating system
(b) Process-oriented programming language and language processors
(c) Utility programs and programming tools
(d) All of these
9.5 System software or executive program consists of
(a) Programs for tasks directly related to the primary functions of the system.
(b) Programs that help the user to prepare the application programs
(c) Consists of all the programs that supervise the overall operations of the
computer system
(d) None of these
9.6 Application software consists of
(a) Programs for tasks directly related to the primary functions of the system.
(b) Programs that help the user to prepare the application programs
(c) Consists of all the programs that supervise the overall operations of the
computer system
(d) None of these
9.7 System support software consists of
(a) Programs for tasks directly related to the primary functions of the system
(b) Programs that help the user to prepare the application programs
(c) Consists of all the programs that supervise the overall operations of the
computer system
(d) None of these
9.8 The important functions performed by application programs are
(a) Conversion of plant input data to engineering units
(b) Scheduling, optimization, and control correction computations
(c) Operators console, logging, and other operator and management presentations
(d) All of these
604 Computer Software for Process Control
9.1 No computer system can operate without _______ which is the program necessary
to execute the tasks performed by the system.
9.2 The fundamental characteristics of the_____________, as distinct form regular
(general) software, are that the control algorithms must be run at their scheduled
sample intervals (with some specified tolerance).
9.3 Access to the hardware of the process control computer system and to the I/O
devices is through the __________.
9.4 In a single-user, single-task, single-job operating system, the whole of the software
system is treated as a single_____________.
9.5 Example of single-user, single-task, OS is _________ operating systems.
9.6 The foreground-background systems is also called _____ system.
9.7 Two-task system is partitioned into two sections, usually referred to as (a)
______________ and (b) ___________ partitions.
9.8 Foreground-background systems are an improvement over the _____________
systems in that the polled loop is replaced by code that performs useful processing.
9.9 The foreground tasks run in______________, or combination fashion.
9.10 In round robin system, several processes are executed _________to completion,
often in conjunction with a cyclic executive.
9.11 Systems that use pre-emption schemes instead of round robin or first-come-first-
serve scheduling are called ______________systems.
9.12 In a ___________system, the partition of system is done into many subsets and
each subset is treated as a separate task.
9.13 __________systems are simple examples of multi-tasking operations, where each
user of the system is allocated a portion of CPU time in a round-robin fashion.
9.14 A ___________takes the code (program) as written by the programmer and
converts it completely to the machine language of the particular computer.
9.15 The MASCOT was the earliest formal methodology for assisting with development
of real-time software, the first version of which was developed by _____________
during the period__________.
9.16 The official standard for MASCOT 3 was published in ______.
9.17 The name MASCOT is an acronym for ______________________________________.
10
Data bus
Input/output
Main memory
interface
As already explained in Chapter 8, Section 8.2.1, the CPU receives data via input
device (such as keyboard, mouse) in the form of binary digits (0s and 1s) and stores the
data for later processing. It performs arithmetic and logical operations on the data in
accordance with previously stored instructions (programs), and delivers the results to
the user through an output devices (such as printer, visual display, plotter). Figure 10.1
shows block diagram of a microcomputer.
equipment for the automotive industry, the PLCs are now used in virtually every type
of industry imaginable. A PLC produces on/off voltage outputs and can actuate elements
such as electric motors, solenoids (and thus pneumatic and hydraulic valves), fans,
heaters and light switches. They are vital parts of industrial automation equipment
found in all kinds of industries.
Power supply
The power supply unit provides the isolation necessary to protect solid-state components
from most high-voltage line spikes. The power supply unit converts power line voltages
Programmable Logic Controllers 609
Programming device
I/O Bus
I/O system modules
Output Output
device device
to those required by the solid-state components. All PLC manufacturers provide the
option to specify line voltage conditions. In addition, the power supply is rated for heat
dissipation requirements for plant floor operation. This dissipation capability allows
PLCs to have high-ambient-temperature specifications and represents an important
difference between programmable logic controllers (PLCs) and personnel computers
(PCs) for industrial applications. The power supply unit drives the I/O logic signals, the
central processing unit, the memory unit, and some peripheral devices.
Memory unit
The memory unit of the PLC serves several functions. It is the library, where the
application program is stored. It is also where the PLCs executive program is stored.
An executive program functions as the operating system of the PLC. It is the program
that interprets, manages and executes the users application program. Finally, the
memory unit is the part of the programmable controller, where the process data from
the input modules and control data for the output modules are temporarily stored as
data tables. Typically, an image of these data tables is used by the CPU and, when
appropriate, sent to the output modules.
Memory can be volatile or non-volatile. The content of the volatile memory is
erased if power is removed. Obviously, this is undesirable, and the units with volatile
memory provide battery backup to ensure that there will be no loss of program in the
event of a power failure. Non-volatile memory does not change state on loss of power
and is used in cases in which extended power failures or long transportation times to
job site (after program entry) are anticipated.
Programmer units
The programmer unit provides an interface between the PLC and the user during
program development, start-up and troubleshooting. The instructions to be performed
during each scan are coded and inserted into memory with the programmer unit. The
programmer unit vary from small hand-held units (size of a large calculator) to desktop
stand-alone intelligent CRT-based units. PLC manufacturers are now providing
controller models that use personal computer (PC) which allows the computer to
interface with a serial input module installed in the programmable controller.
Programming units are the liaison between what the PLC understands and what
the engineer desires to occur during the control sequence. Some programmer units
store programs on other media such as cassettes, tapes and floppy disks. It provides
Programmable Logic Controllers 611
automatic documentation of the existing program using a printer attached to it. With
off-line programming, the user can write a control program on the programming unit,
then take the unit to the PLC in the field and load the memory with the new program,
all without removing the PLC. Online programming allows cautious modification of the
program while the PLC is controlling the process or the machine.
Peripheral devices
Peripheral devices are grouped into several categories such as programming aids,
operational aids, I/O enhancements and computer interface devices. Programming aids
provide documentation and program recording capabilities. The definite trend in
programming aids is PC-compatible software that allows the PLC to be emulated by the
personal computer. The software is sold by the PLC manufacturer or a licensee and is
often model-specific.
Operational aids include a variety of resources that range from colour graphics
CRTs to equipment or support programs that can give the operator specific access to
processor parameters. In this situation the operator is usually allowed to read and
modify timer, counter and loop parameters but not have access to the program itself.
Some aids facilitate the interaction between the PLC and dumb terminals, such as
printers, to deliver process information in a desired format. Some devices have the
ability to set up an entire panel and plug into the PLC through RS 232C ports, thereby
saving enormous panel and wiring costs.
The I/O enhancement group is a large category of PLC peripheral equipment. It
includes all types of modules, from dry contact modules to intelligent I/O to remote
I/O capabilities. Some I/O simulators used to develop and debug programs that can be
categorized in the I/O enhancement group. These are hardware modules which can be
plugged into the PLC.
The computer interface device group is a rapidly expanding section of PLC
peripheral devices. These devices allow peer-to-peer communications (i.e. one PLC
connected directly to another), as well as network interaction with various computer
systems.
Program
Program
memory
m emory
00
00 00
01
01 30 30
02
02 31
03
03 32
04
04
Processor
33
05
05 34
06
06 35
07 36
07
08 37
08 37
09
09 09
Accumulator
1PB 22CR
CR 3 CR
1 PB 11CR
CR 3CR SOLENOID A
11LS
LS
115 V AC
personal computers (PCs). The personal computer (PC) as programming unit has become
very popular with a graphical display. The display typically shows several ladder diagram
lines at a time and also indicates the power flow within each line during the operation
to make debugging and testing simpler. Other units are programmed with logical gates
instead of a ladder diagram. The program is entered by moving a cursor along the
screen (using arrow keys or mouse). When the cursor reaches the location, where the
next element is to be added, confirmation is given via additional keys.
An increasing number of PLCs are programmed in the personal computers (PCs)
in English-statement type languages and higher-level languages (such as C, BASIC, etc.)
using Windows-based operating systems. The programming software is installed (from
many PLCs under MS-Windows) into a PC programming terminal. This software enables
the user to generate PLC programs and to download the resulting program into the PLC
via a standard RS-232 PC port or via a supplier-specific data highway communication
link. With the terminal connected during run the user can observe the current state of
variables, may capture the time history of contacts, or modify the program.
wiring diagram of the physical arrangement of the various components (switches, relays,
motors, valves, etc.) and their interconnections. It is used by electrician (field
maintenance engineer) to do the actual wiring of a control panel.
(e) Starter, relay or a solenoid coil (f) Heating element of an overload device
XOO
OOX
(T.O.)
(g) Time-delay relay-contact (h) Selector switch
Figure 10.5 Ladder diagram symbols.
of Figure 10.5(g) signifies the contacts of a time delay relay. Symbol of Figure 10.5(h)
represents a selector switch with the relative switch positions being indicated by the X
(closed) and the O (open) near the actuator.
Line, L1
Line, Line, LL2
Line,
S1 S2 C1
2
S3 S4
C2
S1
C3
S5
As shown in the figure, all the conditions have to be satisfied simultaneously. The
series connection is a logical AND, whereas the parallel connection is a logical OR. The
letters S1, S2, S3, S4 and S5 denote switches and letters C1, C2, and C3 denote the coil.
The relay contacts usually have negligible resistance, whether they are limit switches,
pressure or temperature switches. The output element could be any resistive load (relay
coil) or a lamp, motor or any other electrical device that can be actuated. Each rung of
the ladder diagram must contain at least one output element, otherwise a short circuit
would occur.
Figure 10.7(b) illustrates a ladder diagram for automatic control circuit of a water
tank process. Figure 10.7(a) shows a water storage tank, a pressure tank, a pump and
an assortment of pilot devices to provide control for the system. The ladder diagram for
the system indicates that the control circuit has an automatic as well as a manual mode.
The manual mode in this case is a start-and-stop pushbutton (P.B.) together with safety
overloads for the pump. These elements are shown in the lower horizontal line (ABC)
of the ladder diagram. They have been arranged so that the electrical connection from
line L1 to L2 required the stop P.B. to be inactive, the start P.B. to be pressed, and the
mode selection switch to be in AUTO position. The normally-open contacts, that are
activated when the pump relay is energized, allow the pump relay to remain engaged
after the start P.B. has been released. The normally-open contacts levelled M are
activated when the pump relay is energized and allow the pump relay to remain engaged
after the start P.B. has been pushed and released.
The automatic portion of this water control circuit is presented in the top-half
(PQRST) of the ladder diagram. The water pump is energized only when each of the
control constraints has been satisfied and the connection from L1 through the pump
616 Microcomputer-Based Process ControlA Programmable Logic Controller (PLC)
PSL
103 Air input
SOL
LSH
101
Stock input Pressure
tank LSH
102
Storage
LSL
tank
103
Discharge
Pump
(a) Schematic diagram
Line, L1 Stop P.B. Line, L2
1 2 Q LSH 101 PSL 103 Air solenoid
P T
SOL
OFF
AUTO MAN
1 2 3 4
A
B
M
starter relay to L2 has been completed. The requirement for automatic control of this
water system is more complex. Figure 10.7(a) shows that there is a high-level and a low-
level monitoring as well as pressure control for the pressure tank and a low-level
monitoring of the storage tank. The high-level sensor for the pressure tank (LSJ 101)
has two sets of contacts: (a) normally-open contact to control the operation of the air
solenoid and (b) a normally-closed contact to provide a condition for pump operation. The
operation of the pump is also governed by the low-level sensors on the pressure and
storage tanks (LSL 102 and LSL 103). Pump damage because of low-water levels in the
storage tank and guaranteeing of pressure stability in the pressure tank, is achieved by
placement of LSL 103 and LSL 102 in series with the starter and the normally closed
LSH 101 contacts. To provide continual pump operation after LSL 102 is momentarily
activated, an interlock with a set of pup relay contacts must be provided.
PLC Communications and Networking 617
Therefore, the method of developing a ladder diagram for a control scheme is to:
· review the control functions required in the circuit
· select the control elements that accomplish this task and
· then arrange the symbols for concurrent control constraints in a sequential
fashion between the supply (voltage) lines on the drawing.
computer
computer
their own control over the access method, access mechanisms, package lengths,
insertion and deletion of nodes on the rug (fly), network start-up and recovery
mechanism.
REVIEW QUESTIONS
10.1 Differentiate between a microprocessor and microcomputer.
10.2 With a neat illustration explain the working of different parts of a microcomputer.
10.3 What is a programmable controller? Discuss the principles on which control
equipment are based.
10.4 What are the parameters on which the overall comparison between different
control equipment types can be performed?
10.5 What is a programmable logic controller (PLC)? Why is it used in industrial
process controls?
10.6 With neat illustration discuss the architecture of a PLC. Explain the functioning
of its various parts.
10.7 Explain with a block diagram basic structure of a PLC.
10.8 How PLC is programmed? Discuss with a neat sketch the ladder diagram
approach of a PLC programming.
10.9 What is a ladder diagram? What are its symbols? Draw and discuss a ladder
diagram for an automatic control circuit for a water tank process.
10.10 With neat sketch discuss the PLC network that is used for communication of
plant parameters in a process control environment.
10.11 Discuss the various selection criteria that must be considered for selection of a
suitable PLC for a specific process control.
10.12 What are the general rules for installation of a PLC?
10.13 What are the advantages of using a PLC?
10.14 Draw a ladder diagram logic for automatic control of a three-tank mixing process
shown in Figure 10.9.
Review Questions 621
xA0
B
xA1
xA1
xA2
A
xA3
AC
10.15 Draw a ladder diagram logic for an automatic control of a fired heater system
shown in Figure 10.10.
Flue gas
Feed
PC
...........
FC
TC
SP
MV CV
FC
DP X Ö
FI FY FY
r
AI
SELF-CHECK QUIZ
A State True/False:
10.1 The early PLC was designed only for logic-based sequencing operations.
10.2 In modern PLCs, plug-in modules are used to convert I/O signal levels to one that
is compatible with the bus architecture.
10.3 Hand-held programmer enables the operator to enter a program multiple contacts
at a time.
10.4 The input to a PLC is always analog and the output from PLC is always digital in
nature.
10.5 The primary objective of PLC communication is to exchange real-time control data
between various PLCs and computers and/or other networks.
11.1 INTRODUCTION
The instrumentation used to implement automatic process control has gone through an
evolutionary process. In the beginning, plants used local and large pneumatic
controllers. These controllers later became miniaturized and centralized onto control
panels and consoles. The first applications of process control computer resulted in a mix
of the traditional analog and the newer direct digital control (DDC) equipment in the
same control room. These systems were inflexible and also cumbersome because the
changing of control configurations necessitated changes in the routing of wires.
With the introduction of microcomputer-based distributed control system (DCS), in
the early 1970s, it became an accepted factory standard for process control applications.
623
624 Microcomputer-Based Process ControlA Distributed Control System (DCS)
Todays DCS has evolved into a flexible and powerful integrated control system that
supplies data acquisition, advanced process control and batch control capabilities.
In DCS, equipment is separated in functional areas and is installed in different
work areas of a process plant. The plant operator monitors and manipulates the
set-points of the process parameter from central control room. The operator views the
process information transmitted from the processing area and displayed on the computer
terminal (CRT or visual display unit) and also changes control conditions from a
keyboard.
The controlling portion of the DCS, distributed at various locations, performs the
following two functions at each location:
· Measurement of analog variable and discrete (digital) inputs
· Generation of output signals to actuators that can change process conditions.
The input and output signals can be both analog and discrete (digital). The plant
(process) information is communicated between central location and remotely located
controller locations over electrical transmission media. The communication path is
either a cable from each remote location to the central station, or a single cable data
highway interfacing all the remote stations. Figure 11.1 shows a distributed control
system in its basic form.
High-level operators console
Supervisory computer
Data highway
Process
As shown in Figure 11.1, the operators console in the control room (high-level
operators interface) is connected through a data highway to several distributed system
components. These components can be located either in rooms adjacent to the control
room or out in the field. A DCS consists of the following modules:
· Operator stations that use microprocessor-based CRT displays and keyboard
communication with control devices and displays
Overview of a Distributed Control System 625
11.2.3 Keyboard
A keyboard may be connected in the monitor housing, or it may be separate, connected
by a cable to the CRT. The key may be movable pushbuttons, or they may be printed
squares on a flexible membrane. The switches of membrane type keyboard have a
flexible, hermetically sealed covering. When the pattern of a key printed on the
membrane is pressed, a conductive elastomer sheet is pushed through an opening under
the key picture, making contact with another conductive sheet beneath it.
Some keyboards use special function keys, programmable by the user. Others use
a conventional typewriter keyboard combined with blocks of special function keys whose
purpose is predefined. Special keyboards are also used to construct graphic display, in
order to accommodate the requirement for colour selection and to facilitate the
construction of special symbols.
11.2.4 Displays
The operator of a distributed control system depends on the displays on the video screen
of his CRT for plant information. Following types of display are commonly used by the
DCS suppliers:
· Group displays
· Overview displays
· Detail displays
· Graphic displays
· Trend displays.
Group displays
The group display shows the operating parameters of group of control loops such as four,
eight, twelve, or sixteen control loops, arranged in rows so that they look like the faces
of instruments on an instrument panel. Figure 11.2 illustrates a four-unit group display.
80 – 80 – 80 – 80 –
60 – 60 – 60 – 60 –
40 – 40 – 40 – 40 –
20 – 20 – 20 – 20 –
0– 0– 0– 0–
PV SP OP PV SP OP PV SP OP PV SP OP
60% 40% 50% 82% 65% 68% 40% 20% 65% 70% 80% 40%
Each of the control loop is represented by a rectangle with bar graphs to indicate the
values of the process variable (PV) and the output (OP) signal.
A moving index positioned beside the process variable bar shows the set-point (SP)
value. Engineering values for process variable, set-point and output per cent are printed
in or below the rectangular area. The process variable range may be printed on the left
or right side of the bar graph. Several lines of text permit a tag number and a service
description to be shown. The rectangle may change the colour (usually to red) if an
alarm condition occurs.
The array of rectangles resembles the row of instruments on a panel board of
analog controllers. From the keyboard, an operator can select a loop, select an operating
mode, change set-point value, and change output value. Using configuration procedure,
an operator can also change the format of the loop arrangement and even change the
content of the display. This permits any combination of loops, arranged by process
association, alarm priority, sequential operation, or any other common parameter, to be
included in the same group display.
Overview displays
The overview display shows the bar essentials of a number of groups, each group in a
separate rectangle. A typical overview display is shown in Figure 11.3. The set-point is
shown as a straight line reference. Deviation of process variable from set-point appears
as a vertical bar. If the process variable for a particular loop is greater than the
set-point, a vertical line will rise up out of the reference line from the segment
corresponding to that particular loop. If the deviation is in the other direction, a vertical
line will drop down from the segment. If the process variable is at set-point value, there
will be no vertical line. The operator, looking at an overview display, can see at a glance
the condition of all the loops in a number of operating areas and can quickly spot a loop
that is out of control.
Digital conditions can also be displayed on an overview display. Discrete conditions
(e.g. an open or closed switch) can be shown as the presence or absence of a bar rising
from the reference line. Sequential events can be displayed by displaying messages that
change as the sequence advances.
If the overview display indicates an alarm condition in a particular group, and the
group parameters must be examined more closely, the operator frequently call up the
group display in which the alarm has occurred with a single keyboard stroke. If the
operator wants still more detailed information, another keystroke can often be used to
call up a detailed display of the loop in the group that has generated the alarm.
Detail displays
The detail display is specific to a single loop or control function. Figure 11.4 shows a
Configuration word
Algorithm type
Highway and station address
CONF 0100210003
HWY 1 STA 33
PID
FLC-508
900
FUEL FLOW
100
- 800
80 –
600
60 –
40 – 500
20 –
400
0–
PV SP OP
300
60% 40% 50%
100 80 60 4400 20 0
Minutes
ADDRESSES
typical detail display. This display has the same bar graph representation that is part
of the group display, but it includes additional information defining constants, limits, and
other characteristics of the function. The sources of the signals coming to it are listed
on the screen.
The list of detail displays varies from supplier to supplier. Each supplier of DCS has
its own complement of special lists, menus and libraries (e.g. units of measure and
messages etc.) that are displayed on the CRT.
Graphic displays
A graphic display capability allows a picture to be drawn on the screen so that the
operator can look at a portion of the process more realistically than by watching a row
of bar graphs. Figure 11.5 shows a graphic display representation of a process. Process
and control information is included in the picture, and it can be interactive, dynamically
Quality Condenser
Column
98.4%
94°C
Accumulator
Flow 251.7 GPM
Heater 151°C
210°C
Flow 337.2 GPM
Column Bottoms
feed level
51% Steam 13662 PPH
Re-boiler
changing as real-time information change. A pipeline, for example, can become filled
with colour when a valve is opened, the symbol of the valve can change colour, and its
condition can be identified by the label that indicates On or Off. Graphics are valuable
training tools and help the operator relate to plant conditions when a number of
variables are changing at one time.
630 Microcomputer-Based Process ControlA Distributed Control System (DCS)
Some graphic displays are also capable of showing movement, for example, when
liquid is flowing in a pipeline, when solids are travelling on a belt, when agitators are
turning, or when fuel is burning in a combustion process, the associated movements are
dynamically displayed.
Trend displays
Trend displays are the distributed control system equivalents of chart records. They are
a profile of values of a process variable showing changes that have taken place over a
period of time. Some detail displays, as shown in Figure 11.4, include a real-time trend
graph of the process variable values during a selectable period, e.g. 1 minute, 1 hour,
24 hour, etc. In some displays, several trend graphs can be displayed at once, allowing
comparison of the history of several variables. Figure 11.6 shows the display of two
variables.
Plate
Plate temperature
tmperature 250
TC – 42A
SP 600
PV 307.6
Re-boiler
Re-boilerlevel
lvel 120
LC – 207
SP 45
PV 106
The trend display information is valuable to the plant operator to observe the
recent pattern of operating history. It is valuable to the operator after an upset has
occurred; allowing him to determine which of several interrelated variables was the first
to be affected by the changing conditions. Trends over longer periods (over a week or
more) can be saved on a floppy disk storage and displayed when required.
Drops to an electrical cable can be made by splicing wires, but this is not possible
when a fibre glass is used. An optical electrical interface (OEI) units is required as a
node on the fibre-optics highway. This uses an LED to change electrical pulses coming
from a controller file or from an electrical highway into light pulses. The light pulses
are changed back into electrical signals by using a PIN diode, a light-sensitive device
that produces an electrical output when exposed to light. Table 11.1 summarizes the
relative features of twisted pairs, coaxial cables and fibre-optic cables.
Many DCS systems are designed to include limited online information functions,
perhaps providing, for example, a history module as part of a standard DCS function. In
most cases, such systems do not have sufficient capacity to meet the users needs, and
a computer-based data archiving system is necessary. The DCS does not have a
sufficient storage capacity nor processing speed for processing a large volume of data,
and the DCS does not normally support the desired database formats required by many
users. In addition, many computer systems are supported by huge quantities of standard
software packages, making the implementation of a real-time information system
affordable.
Velocity algorithm
Coordinated supervisory control with integrated feedback or feedforward combined in a
velocity control algorithm can be activated or deactivated without creating perturbation
to the process. The velocity algorithm implemented in the supervisory computer can be
converted to the position algorithm which is usually applied at the DCS level. The
control command (u), in the supervisory computer at the current sampling time (n) can
be expressed as follows:
u(n) = sat [u(n 1) + D] (11.1)
where
D = Incremental output from computer command
u(n) = Current computer position command at sampling instant (n)
u(n 1) = Past sampled position command
and
Upper limit if x ³ upper limit
34241
where
D f bi = Incremental value from the ith feedback loop
D ffj = Incremental value from the jth feed forward loop
ki = Assignable constant for ith loop
kj = Assignable constant for jth loop
638 Microcomputer-Based Process ControlA Distributed Control System (DCS)
The advantage of this algorithm for initialization of the control output command
from the supervisory computer is that initialization can be obtained simply by setting
D equal to 0 and making u(n 1) equal the value obtained from DCS (for example,
set-point value for a slave loop).
Anti-windup algorithm
For supervisory control, the command from the supervisory computer to the set-point
for the control loop at the DCS level is essentially a form of cascade control. Caution
is needed when control loops are in cascade. The upper or primary loop control
command (supervisory computer command) requires information from the lower or
secondary loop at the DCS level in order to command correctly. The information
includes valve saturation status and lower loop current set-point value at the DCS. This
information at the DCS level requires constant monitoring by the supervisory computer
of the real-time information communicated through interface module on the DCS
network. It is very important to prevent the computer command from causing the
set-point windup at the DCS level.
Therefore, an anti-windup algorithm is used to protect the upper loop command
from the windup. The supervisory control command output to the DCS set-point is
frozen at its last value if the DCS output to the valve is saturated and the direction of
the upper loop output increment is such that it would cause further valve saturation at
the DCS level. Otherwise, the upper loop within the supervisory computer will not be
frozen at its last value.
This anti-windup algorithm can be written by modifying Eq. (11.1) as follows:
u(n) = sat [u(n 1) + (1 sw) * D] (11.3)
where
321
N+ (D > 0) =
0, if £ 0
1, if £ 0
N (D £ 0) =
0, if > 0
1, if valve position reaches upper bound
321 321
Iu =
0, if valve position at upper bound is not reached
1, if valve position reaches lower bound
I1 =
0, if valve position at lower bound is not reached
instead of wiring program code for each special algorithm. Such an arrangement can be
extremely useful to facilitate the implementation of supervisory control by control
engineers.
safety device is not actuated, rather than just not starting the motor. A well-integrated
system might describe the nature of the problem to the operator by changing the
running light colour from red (stopped) to yellow, instead of to green (the colour
associated with running). If the systems are very well integrated, the operator may be
able to use the DCS MMI to query the PLC to determine which limit switches are
preventing the requested action from taking place.
The DCS can also be interfaced with the computer systems. While most computers
today have CRT-based interfaces, some are text-based and use cryptic (to the uninitiated)
series of commands or instructions. By having a simple function key or touch screen
button for starting a computer program for advanced control, or even a plant-wide
electronic mail system, operators can use a familiar and well-understood interface.
REVIEW QUESTIONS
11.1 Explain with a neat sketch the structure of a distributed control system (DCS).
Discuss the functioning of its various parts.
11.2 What is the difference between a PLC and DCS? How are PLC and DCS
integrated for industrial process control?
642 Microcomputer-Based Process ControlA Distributed Control System (DCS)
11.3 What are the different types of displays available on the video screen of the DCS
for plant information? Explain each one with a sample screen diagram.
11.4 How graphic display helps the operator of the plant? With a neat sketch show the
graphic display representation of a process.
11.5 What is a trend display? What is its significance in process control environment?
11.6 How is DCS programmed? How is the DCS configured to provide a desired
information?
11.7 Discuss the method of interfacing DCS with the following:
(a) Supervisory computers
(b) PLCs
(c) Smart transmitters
(d) Microprocessor-based analyzers.
11.8 What are the international fieldbus standards that are used in DCS
communication?
11.9 What is a data highway? Discuss the relative features of a data highway.
11.10 Discuss the advantages of using fibre optics as a data highway.
11.11 What is a network access protocol? Discuss the methods used for distributed
control highways.
11.12 List the advantages and drawbacks of various types of network access protocol.
11.13 How does a supervisory computer typically perform in a process control
application?
11.14 Why is supervisory computer control necessary for an industrial process control?
11.15 Discuss the different types of supervisory control algorithms used to command the
controllers.
11.16 How is DCS integrated with PLCs and other computers?
11.17 What is the manmachine interface (MMI)? Why is it important for computer-
aided process control applications?
11.18 List the advantages and features of using DCS for process control applications.
SELF-CHECK QUIZ
A State True/False:
11.1 The high-speed ladder logic of the PLC is usually performed independently of the
functions being performed in the rest of the DCS.
11.2 In DCS, equipment is separated in functional area and is installed in centralized
control room of the plant.
Self-Check Quiz 643
11.3 Because of the distributed nature of the DCS, a single failure typically shuts down
the entire plant.
11.4 The keyboards of DCS systems are programmable by the user.
11.5 CSMA/CD uses ring type of network for exchange of data.
11.1 Like many other computer systems, distributed control system (DCS) is also called
programmed _____________.
11.2 DCS is configured from the _________________ station.
11.3 One of the features that make distributed control systems so powerful is their
___________ library.
11.4 The most popular medium for the data highway for DCS communication is
_______.
11.5 The full form of algorithm SISO used in DCS is ____________________________.
Part V
Case Studies
12
12.1 INTRODUCTION
In previous chapters, the dynamic behaviour of several basic industrial processes was
examined. It was also discussed that the use of computers for controlling industrial
processes results in all-round process improvements. It maximizes plant yields,
increases plant capacity and productivity, ensures the repeatability in the product
quality, reduces (optimizes) raw material and energy costs, increases safety of plant
operation, and improved profitability.
With this background and to tie together the principles discussed so far, this
section presents a number of practical examples of implementing process control
systems. This will provide the readers with an opportunity to conceptualize, how to
design process control systems. It also gives an overall insight into as how a computer-
aided process control systems can be implemented.
647
648 Process ControlCase Study
Pneumatic
controller
Pure A PC
Pneumatic
transmitter
O
X #X
X, c 0 S#
PT Set-point
m Product stream
V,T 1,c1,k1 V ,T 2,c 2,k2
Sample
stream
Heater Heater
M easuring
Tank 1 Tank 2 element
Figure 12.1 Two stirred-tank chemical-reactor control system.
8
FE
X E §¨ X O·
E M 8E O
FV © S # ¸¹ (12.1)
Chemical-Reactor Control System 649
where
m = Molar flow rate of pure reactant A through the valve, lb moles/min
rA = Density of pure reactant A, lb moles/ ft3
It is assumed that the volumetric flow of reactant A through the valve m/rA is
much less than the inlet flow rate w with the result that Eq. (12.1) can be written as
FE
X M8 E
8 X E O (12.2)
FV
This equation may be written as
FE
U E E O (12.3)
FV 8 § 8·
M X ¨ M ¸
X © X¹
where
8
U (12.4)
X M8
FE
At steady state, , and Eq. (12.3) becomes
FV
E U E U OU (12.5)
8
M
X X §¨ M 8 ·¸
© X¹
where s refers to steady state.
Subtracting Eq. (12.5) from (12.3) and introducing the deviation variables
C1 = c1 c1s
C0 = c0 c0s
M = m ms
give
F%
U % / % (12.6)
FV § 8· 8
M ¸ X ¨ M
X © X¹
Taking the transformation of Eq. (12.6) yields the transfer function of the first reactor,
which is given as
§ 8· ª § 8 ·º
¨ M ¸ «X ¨ M X ¸»
%
U © X ¹ %
U ¬ © ¹¼
/
U (12.7)
UU UU
As with Tank 1, Eq. (12.8) can be written in terms of deviation variables and
arranged to give
F%
U % % (12.9)
FV 8
M
X
where
8
U (12.10)
X M 8
C2 = c2 c2s
Taking the transform of Eq. (12.9) gives the transfer function for the second
reactor, which is given as
§ 8 ·
¨ M ¸
%
U © X ¹ %
U
(12.11)
U U
8
U min
X M8
8
U min
X M8
Chemical-Reactor Control System 651
§ 8 · ª § 8 ·º § · ª § · º
¨ M ¸ «X ¨ M X ¸ » ¨ ¸ « ¨ ¸ »
%
U © X ¹ %
U ¬ © ¹¼
/
U © ¹
¬ © ¹¼
UU UU
or
C1s = 0.0733 lb mole A/ft3
§ 8· § ·
¨ M ¸ ¨ ¸
%
U © X ¹ %
U © ¹
U U
or
C2s = 0.0244 lb mole A/ft3
OU
cfm
S#
-X cfm/psi
OU
Since cfm, the normal operating pressure on the valve will be
S#
RU psi
Therefore, the valve equation may be derived as
where M = m 1.25rA
P = p 10.5
Taking the transform of Eq. (12.13) the valve transfer function will be
/
U
-X S # (12.14)
2
U
-O psi/(lb mole/ft3)
Since c2s is 0.0244 lb mole/ft3, the normal signal from the measuring element is
psig
The equation for the measuring device may be written as
b = 7.32 + Km(c2 0.0244) (12.15)
where b is the output pressure (psig) from the measuring device. In terms of deviation
variables, Eq. (12.15) becomes
B = KmC2 (12.16)
where B = b 7.32 and C2 = c2 c2s
2
U
-R (12.20)
U
Assuming the set-point and the signal from the measuring device to be the same
when the system is at steady state under normal conditions, we have the reference
value of the set-point
CRs = b = 7.32 psig
The corresponding deviation variable for the set-point is given as
CRs = cR cRs
8QNWOG
UF OKP
(NQY TCVG
GU F U GU
+ 1 1
CR + P A
N 1 + V C1 V
Kp K vSA
CR
Km
w M
1 k1
X
1 k2
X C2
psig
– X U 1s 1 U 1s 1
psig B Km e– U s
d
1
K1
C0 § V ·§ V·
¨ 1 k1 X ¸¨ 1 k2 X ¸
© ¹© ¹
+
+ + 1
CR KK p C2
( U s 1)( U2 s 1)
–
eU s
d
Numerical quantities for the parameters in the transfer functions are as follows:
t1 = 2
t2 = 1
-
-O -XS #
Open-loop gain = -- R -R - R
X §¨ M 8 ·¸ §¨ M 8 ·¸
© X ¹© X¹
X,T i
Product stream
X,T
Tank
H eater
At the steady state condition, the energy balance around the heating process of the
tank may be written as
3U X %
6U 6 KU
(12.21)
where
Qs = Heat input to the tank at steady state condition
Tis = Steady-state inlet temperature to the tank
Ts = Steady-state temperature of the effluent stream
C = Specific heat of fluid
For a satisfactory design, the steady-state temperature of the effluent stream
Ts must be equal to the desired temperature TR (set-point). Hence
3U X %
64 4KU (12.22)
Temperature Temperature
transmitter controller
X,Ti1 TT TC Set-point
Product stream
X,T
Tank
Q Temperature sensor
(Thermocouple)
Heater
Figure 12.6 shows a block diagram for stirred-tank heat exchanger control system.
The heat input is adjusted according to the error (TR T) to maintain the set-point. Tm
is the thermocouple reading. Let us assume that the controller has proportional (P)
control action. Thus, the relationship between controller temperature and the error is
given as
Q(t) = wC(TR Tis) + Kp(TR T) (12.25)
where
Kp = A constant of proportionality
Stirred-Tank-Temperature Control System 657
Disturbance Ti
Set-point +
TR + Q + T
Controller Heater Stirred-tank
Controlled
– variable
Tm Temperature T
sensor
U = Ti
+
+ Q 1 + 1
Kp
R
XC Uis + 1 C= T
B = Tm
e–Us
Thus, the overall transfer function is the product of the individual transfer functions. Its
magnitude will be product of the magnitudes and the argument will be the sum of the
arguments of the individual transfer functions. In general, if
)
U )
U)
U)P
U
Then
) LX ) LX ) LX )P LX
Now, the block diagram of Figure 12.6(b) can be redrawn as shown in Figure 12.7 after
substituting for the parameters as calculated earlier.
U = T ia
Q +
+ Kp Kp XC + 1
R C = Ta
XC =
0.202s + 1
–
B = T ma
e–0.0395s
Because of excessive phase lag, the control system of Figure 12.7 might offer stability
problems. The system represents proportional control of tank temperature with a delay
in the feedback loop. The factor is the process sensitivity , which gives the
X%
ultimate change in tank temperature per unit change in heat input Q. The proportional
sensitivity Kp, in Btu per hour per degree of temperature error is to be specified by the
designer. The open-loop transfer function for this system is given by
§ - R · U
¨ ¸G
)
U © ¹ (12.26)
U
-R
Figure 12.8 shows the plot of Bode diagram for G(s). The constant factor is included
in the definition of the ordinate for AR. At the frequency of 43 rad/min, the phase lag
is exactly 180o and
660 Process ControlCase Study
#4
- R
Therefore, if a proportional gain of 5000 Btu/(hr)(oF) is used,
#4
This is the amplitude ratio (AR) between the signals, which is dimensionless.
1.5 20
Corner or break-
point f requency
L ow-f requency
asymptote
A mplitude ratio A R
1.0 0
L , dB
H ig h-f requency
asymptote
0.5 – 20
– 40
0.1 X
0
e– 0.0395s
1
– 45 0.202s + 1
Phase angle, degree
– 90
e – 0.0396s
0.202s + 1
– 135
– 180
– 225
– 270 X
1 10 20 50 100
Figure 12.8 Bode diagram for open-loop transfer function of control system for stirred-tank
heat exchanger.
Stirred-Tank-Temperature Control System 661
The control system is redrawn for Kp = 500 with loop opened as shown in Figure
12.9(a). That means, the feedback signal B is disconnected from the comparator. It is
imagined that the set-point disturbance
R = sin 43t
is applied to the opened loop. Since the open-loop AR and phase lag are unity and 180o,
then
B = sin (43t 180o) = sin 43t
Now, let us assume that at some instant of time R is set to zero and simultaneously
the loop is closed. Figure 12.9(b) shows that closed loop continues to oscillate
indefinitely. This oscillation is theoretically sustained even though both R and U
are zero.
U= 0
+
+ = sin 43t 8.33
+ 1 C = sin (43t – 83°)
R = sin 43t 0.202s + 1
–
R=0
+ = sin 43t 8.33
+ 1 C = sin (43t – 83°)
0.202s + 1
–
A fter
closing loop
B = –sin 43t
e–0.0395s
Suppose Kp is set to slightly higher value and the same experiment is repeated.
This time the error signal Î is amplified slightly each time it passes around the loop.
Thus, if Kp is set to 5001, after the first time around the loop the signal Î becomes
§ · § ·
¨ ¸ UKP V . After the second time, it is ¨ ¸ UKP V and so on. The phase-angle
© ¹ © ¹
relations are not affected by changing the Kp. Thus, it can be concluded that, for Kp >
5000, the response is unbounded, since it oscillates with increasing amplitude. On the
basis of stability criterion discussed in Chapter 4, it is concluded that the control system
is unstable for Kp > 5000 because it exhibits an unbounded response to the bounded
input described earlier. The condition Kp > 5000 corresponds to AR > 1 for the open-loop
transfer function, at frequency 43 rad/min, where the open-loop phase lag is 180o.
662 Process ControlCase Study
Temperature Temperature
sensor signal
Computer-aided
control
Power
Electric Continuously control
oven variable Power
electrical input Continuously supply
variable power
unit
100°C
(0.84 V)
185°C
(1.93 V)
Time (seconds)
70°C
(0.73 V)
3600 7200 10800
T1 T2
20°C
(0 V)
3600 7200 10800
Time (seconds)
T1 T2
(c) Open-loop response of the oven cooling
Figure 12.11 Response of the oven temperature to step changes.
664 Process ControlCase Study
show the response of the oven temperature to step changes in the input voltage to the
heating element.
Table 12.1 summarizes the results of the open-loop response tests. The curve of
Figure 12.11(b) is chosen as a reasonable compromise among the curves obtained during
experimentation.
Discretization
For trapezoidal discretization,
T z 1
s (12.29)
2 z 1
Electric Oven Temperature Control 665
Î 6 \ Þ
&
\ Ï ß
Ð
6=
\
\ ? \ à
=
6 6 \
6 \
\ 6 6
+ (0.0037 T + 67.5T2)] (12.30)
Ë È Ø È Ø Û
W
\ \ W
\ Ì É 6 Ù É 6 Ù \ Ü
Í Ê 6 Ú Ê 6 Ú Ý
È Ø
É 6 Ù \ (12.31)
Ê 6 Ú
Ë È Ø È Ø Û
W
M W
M Ì É 6 Ù G
M É 6 Ù G
M Ü
Í Ê 6 Ú Ê 6 Ú Ý
È Ø
É 6 Ù G
M (12.32)
Ê 6 Ú
Oven heat losses compensation
Let us assume that the desired oven temperature is set at Td and that the actual
temperature Ta of the oven satisfies Ta = Td. Under this condition, the algorithm of
Eq. (12.32) is put into operation with u(0) = 0. Then for zero error the algorithm
produces zero output. However, to keep the temperature Ta equal to Td, an input is
needed to the oven, sufficient to offset the heat loss. Thus, an additional term a,
dependent on Td, is added to the right-hand side of Eq. (12.32), which therefore becomes
G U È Ø
)
U &
U ÉÊ Ù (12.34)
U U Ú
)
U &
U
(12.35)
)
U &
U
UU
G U
U
666 Process ControlCase Study
Since 150 s is a very long sampling interval, a lower value of T can be taken as
sampling interval to demonstrate the performance of the oven. Figure 12.12 shows the
Voltage applied to oven element
Oven temperature
100°C
Change of
desired value
20°C
3600 7200 Time (seconds)
performance of the oven under closed-loop control with sampling interval of 60s.
Table 12.12 shows an experimental result of the settling time for various sampling
periods.
Electric Oven Temperature Control 667
Temperature
sensor
Temperature signal
Continuously
variable Power
Electric electrical input supply
Regulated
oven power
Power
control
Amplifier
ADC DAC
Computer
Final
Reference
Analog-to-
digital Temperature sensor
converter (measuring element)
Computer
Flue products
Insulated walls
Thermocouple
Fuel
Burner
Air Flame
Slabs Slabs
Variable loads
Variable loads
Figure 12.14 Schematic diagram of reheat furnace.
The control objective is to alter the fuel flow rate by manipulating (controlling) a
valve so that the measured temperature at some point in the furnace follows a specified
temperature/time profile. An independent regulator is used to maintain the air/fuel ratio
which is required for efficient and safe combustion.
where
H0 = w0, * ÇY Z , K K
* ÇÇ Y Z Z K
KL K L (12.39)
K K L
H1 is the linear term, H2 is the second-order term of the function, and so on. This
expression may be written as:
F(x) = wTy (12.40)
where w is a vector of weights and y is an expanded vector of x. The vector w may be
found by adaption.
In this case, the controller model is trained on the existing data, and the model
learns about the system before the operation starts. This data can be obtained from
normal running data from an existing control system or by performing system
identification trials to determine the response of furnace temperature with respect to
gas (fuel) valve position. Adaption can then be used to track the system. In this way,
the controller is continually fed with the most appropriate weights in a similar manner
to a self-tuning control (STC) scheme.
7.5
6.5
6
0 200 400 600 800 1000 1200 1400
Time (samples)
(a) Furnace temperature
672 Process ControlCase Study
Control setting
5
4
Excitation
0
0 200 400 600 800 1000 1200 1400
Time (samples)
(b) Gas valve position demand signal
Figure 12.15 Predictive control for K = 0.
Response of control setting
8
Response v desired
7.5
6.5
6
0 200 400 600 800 1000 1200 1400
Time (samples)
(a) Furnace temperature
Control setting
5
4
Excitation
0
0 200 400 600 800 1000 1200 1400
Time (samples)
(b) Gas valve position demand signal
Figure 12.16 Predictive control for K = 0.1.
Reheat Furnace Temperature Control of Mill in an Integrated Steel Plant 673
7.5
6.5
6
0 200 400 600 800 1000 1200 1400
Time (samples)
(a) Furnace temperature
Control setting
5
4
Excitation
0
0 200 400 600 800 1000 1200 1400
Time (samples)
(b) Gas valve position demand signal
Figure 12.17 Predictive control for K = variable.
12.5.5 Adaption
As the nonlinear model explained above is a linear weight model, it has a fundamental
advantage over many other nonlinear modelling schemes. It has a unimodal error
surface. This enables the use of standard adaption schemes, which are normally used
for linear schemes in the nonlinear controller. To demonstrate this, a simple adaption
term may be added to the scheme already described in Section 12.5.2, as follows:
674 Process ControlCase Study
w¢ = w 2m ex (12.42)
where
w¢ = New weights vector
w = Old weights vector
e = Measured error between the predicted temperature and the actual
temperature
x = Extended input vector used to predict the temperature
m = Adaption coefficient.
Figure 12.18 shows the block diagram of a complete controller using a model and
adaption.
vd
Generalized
u predictive control v
(w, ut, m)
m
Model
+
Adaption
Furnace
Figures 12.19 and 12.20 show the furnace temperature and the gas valve position
demand signal for adaptive control with complete system for two iteration. The
controller is trained as discussed in Section 12.5.2 and then misadjusted by introducing
small errors in the weights vector. The controller is initialized with these misadjusted
weights. After completion, the furnace is reset and the controller is restarted using the
newly adapted model weights. It can be seen that both tracking and stability
performance are significantly improved on the number of iteration with the new
weights.
Reheat Furnace Temperature Control of Mill in an Integrated Steel Plant 675
7.5
6.5
6
0 200 400 600 800 1000 1200 1400
Time (samples)
(a) Furnace temperature
Control setting
5
4
Excitation
0
0 200 400 600 800 1000 1200 1400
Time (samples)
(b) Gas valve position demand signal
Figure 12.19 Adaptive control (iteration-1).
7.5
6.5
6
0 200 400 600 800 1000 1200 1400
Time (samples)
(a) Furnace temperature
676 Process ControlCase Study
Control setting
5
4
Excitation
0
0 200 400 600 800 1000 1200 1400
Time (samples)
(b) Gas valve position demand signal
Figure 12.20 Adaptive control (iteration-2).
T set-point Process
Furnace controller computer
T furnace
Fuel MODEL
w 2T 1 wT
Burner Flame
Air wx 2 x wU
T1 T2 T3
Control systems
In a steel plant, the computer control of the slab reheating furnace may be divided into
three major groups:
(a) Logic control function
(b) Process control function
(c) Manmachine interface.
Logic control functions cover sequence control, interlocking and basic closed-loop
control functions. Logic control functions are in charge of smooth and safe operation of
the plant as a complex combination of single devices.
Process control functions are responsible for yield of rolled products. Process
functions directly influence product quality and production costs.
The manmachine interface (MMI) is a link between operator and the automation
system and is an important component in computer-aided control concepts.
Following control functions are implemented in a slab reheating furnace of steel
plants:
· Temperature control of top and bottom zones
· Flow control and ratio control of zones
· Furnace pressure control
· Overall furnace control.
678 Process ControlCase Study
H H
q h
The slab-entry into the rolls generates a force F, so that the thickness of the
outgoing strip (h) is given as
F
h = q + (12.43)
M
where M is the stiffness coefficient. q and F are measured using synchro and load
cell respectively, to estimate the value of h using Eq. (12.43) for thickness control. The
value of h, estimated using Eq. (12.43) is further validated (calibrated) by measuring the
strip-thickness using nucleonic gage or laser-based system at downstream of the process
(rolling). The upper roll is moved vertically up and down using an electric motor as an
actuator to modify q and hence modify and control the strip thickness h.
680 Process ControlCase Study
S1
S0 S2
S1
S2
Tension
Tension
distributed
in strip
ed
pli
A p sion Strip width
ten
(a) Non-uniform elongation (b) Tension measured at various points
Figure 12.23 Flat measurement of rolled strip.
The strip tensions can be measured at various points such as s2, s1, s0, s1, s2, etc.,
sited systematically across the strip width, as shown in Figure 12.23(b). We fit a
polynomial using a least-square criterion, given as
s(w) = a0 + a1w + a2w2 (12.44)
Parameters a1, a2 are the measures of flatness with a1 = a2 = 0 implying perfect flatness.
The parameters a0 is not of interest since it is merely a measure of the mean tension
applied during rolling.
C
U M
14444244443
)
U
W
U U
U
C
U M
)
U
W
U U
U (12.45)
C
U M
)
U
W
U U
U
*
U
X P
(12.46)
X P YX P U U
*
LX
X P
(12.47)
X P X L YX PX
G U6 M
)
U (12.49)
U U
U
G U6 M
)
U (12.50)
U U
U
where k1, k2 and k3 are constants and have values 10, 2 and 0.5 respectively. Now, Z
transforming Eq. (12.48) yields
k1
G1c ( z ) (1 z 1 )Z 2
s (1 s )
k1[(T 1 e T )z 1 e T Te T ]
( z 1)( z e T )
682 Process ControlCase Study
=
\ 6 ?
)
\
\
\
\
(12.51)
\ \
Since the three processes of Eqs. (12.48)(12.49) are similar, the calculation done
for Eq. (12.48) will be similar for Eqs. (12.49) and (12.50).
Let D1(z) be the transfer function of the controller for loop-1; then we must have
)
\ &
\
*
\ (12.52)
)
\ &
\
or
*
\
&
\
)
\
*
\
Substituting the values leads to
\
\ \
&
\
\ È \ Ø
ÉÊ Ù
\ \ \ \ Ú
\
\ \
\
\ \ \
\ \ \
(12.53)
\ \ \
Similar equations can be derived to obtain D2(z) and D3(z) in analogous fashion.
W
\ 0
\
and u(z) = N(z)q(z)
G
\ &
\
where S
\ G
\ &
\ . Then
W
\
\ \ \
G
\ \ \ \
Figure 12.24 illustrates the realization of the above algorithm. The two other
controllers for D2(z) and D3(z) can also be realized in the same way.
+
u(k)
20.83
+
+
0.9048 1 0.9048
e(k)
T T T
+
+ +
1.16
2.7
0.59
Computer-aided
control system
Strip
Flatness Thickness
Rolling mill measurement measurement
Figure 12.25 Overview of computer-aided control system for rolling mill.
Coil yard
Reheat
furnace
Coiler
x x x x x x
x x
Slab
x x x x x x
Roughing mill Crop shear Five strands finishing mill Down coiler
Figure 12.26 Schematic of a hot-strip mill.
The heated slabs are discharged from the reheat furnace and taken to roughing
mill for rolling to achieve an initial reduction in dimension (thickness and width) and
thereby increasing the length of the product. The slab starts converting into strip of
thinner dimension. The strip is coiled in the coiler for mill synchronization and to retain
the heat for further rolling.
686 Process ControlCase Study
Uneven-shaped crops of head-end and tail-end of the hot rolled-coil is cut by the
shearing machine before it is fed to the finishing mills. Finishing mills are multi-strand
rolling mill in which the coil passes through four or five strands depending on the target
strip thickness. All strands of the roughing mill work in synchronization.
Finished strip passes through run-out table (ROT), where cooling of strip is done
using optimization model to achieve desired (target) mechanical properties of strip. The
finished strip is coiled in the down coiler and taken to the coil storage yard. These coils
are cut into different sizes as per customer requirement and shipped to the customer
location.
ADC DAC
Laser thickness
Laser thickness sensor (exit)
sensor (entry)
Strip x
Position control
Figure 12.27 Automatic gauge (thickness) control.
Both entry- and exit-strip thickness of a rolling strand is measured using a laser
thickness gauge meter. Based on these measurements, the roll gap and thus the
thickness of the rolled material (strip) is determined from the position of the roll
adjustment (position control system) and the measured rolling force. The effects of
strand and roll deformation are taken into account in a mathematical model. The strip
thickness measured by downstream thickness gauge is used for elimination of long-term
deviations in order to achieve maximum performance in terms of absolute thickness.
Thickness measurement obtained from an upstream thickness gauge is used for
feed-forward compensation of disturbances in strip thickness. This helps elimination of
any variations in the strip thickness.
required by the customer, and taking into consideration the variation in strip width
during rolling. This extra width causes a significant decrease in yield of the plant.
Therefore, it is important that the variations in the strip width is minimized or
eliminated to improve the plant yield. The automatic width control (AWC) system
ensures a minimum strip-width variation and thus contributes in improving the
performance of hot rolling mills. The automatic width control system is implemented in
the roughing mill strand. Figure 12.28 shows a schematic of automatic width control
system in the roughing mill strand.
Edger x
Roughing mill
Figure 12.28 Automatic width control system.
As shown in Figure 12.28, width of the slab is measured using non-contact image
based system before it enters the edger strand. The rolling material is virtually divided
into number of segments. The corresponding set-point for the roll gap of the edger is
calculated for each segment. The set-point calculation is done using a linearized
mathematical model with model parameters delivered by a process optimization
computer.
As the slab/strip passes through the edger the slab/strip tracking function
determines the position of each segment and sends the set-points to the edger roll gap
adjustment system. The automatic width control system ensures minimum width
variations after the strip passes through the roughing mill strand. The increased
Automation of Hot-strip Mill in an Integrated Steel Plant 689
longitudinal strip spread during rolling in the roughing strand, is taken into account and
compensated for. Thus, a uniform strip width is achieved from the head-end to tail-end.
This minimizes the cropping losses thereby further increasing yield of the mill.
ADC DAC
Maintenance
computer
HMD CCD-camera
x
Strip
the strip and improvement in strip quality. The level-2 automation system for a hot
strip mill comprises of the following models:
· Mill set-up model and rolling strategy
· Run-out table (ROT) cooling control model.
The rolling strategy takes care of the calculation of an optimal pass schedule for
the finishing mill rolling. Beside the consideration of conventional boundaries such as
thickness, temperature, etc., of the strip, the rolling strategy is mainly driven by strip
profile and flatness considerations which leads to a particular roll force requirement. To
produce a flat strip, the relative profile of the strip should be constant from pass to pass
(strand to strand) of the finishing mill. In the first strand, however, demand for relative
profile is decreased as because ratio between width and thickness of the strip is in a
range allowing material to flow in direction cross to running strip. Thus, in the first
strand, the profile can be varied without much effect to the strip flatness.
The restriction of constant relative strip profile, is therefore, not binding in the
passes, where material flow across the rolling direction is possible. This means that in
the case of a large thickness to width ratio, a change of the strip profile is possible
without influencing the flatness of the strip. According to Shohett and Townsend, the
difference between the actual relative profile and the target relative profile must be
within the following range to get a flat strip
È JØ È JØ
É Ù
%6 % # É Ù (12.57)
Ê YÚ Ê YÚ
where
h = Thickness of the strip
w = Width of the strip
CT = Relative target profile
CA = Relative actual profile.
Figure 12.30 shows the dependence of profile and flatness of strip. For each
thickness to width ratio, the allowed range of the deviation is to be found between actual
relative profile and target relative profile. These shape control curves of Figure 12.30,
are the foundation for the calculation of roll pass-schedule. The pass-schedule is
calculated such that the constant relative strip profile is achieved. The curves shown in
Figure 12.30 imply a region of admissible roll forces which guarantee that the flatness
of the strip is not disturbed due to the influences of parameters such as
· Thermal expansion of work roll
· Wear of work roll
· Thermal crown of work roll
· Roll bending
· Roll force.
The thermal expansion of the work roll is an essential determinant of the strip profile
and flatness. Thermal crown and change of work roll diameter are calculated
considering actual process conditions and cooling conditions. The evolution of thermal
crown is calculated from the temperature field within the work rolls as per the equation
given as
4Y
C
EVJ
\
4Y Ô T
6
T \ 6 FT
(12.58)
692 Process ControlCase Study
E = CA CT
Centreway strip
Flat strip
h/w
Edgeway strip
where
cth = Thermal expansion in width direction
z = Roll axis coordinate
b = Coefficient of thermal expansion
Rw = Work roll radius
r = Radius coordinate
T = Temperature
T0 = Initial roll temperature.
The temperature field within the work roll can be obtained by differential equation
of heat conduction in the radial direction and in direction of roll axis, given as
6 È 6 6 6 Ø
SE M É Ù (12.59)
V Ê T T T \ Ú
where
r = Density
c = Specific heat capacity
l = Thermal conductivity.
Figure 12.31 shows an example of the thermal crown. Together with the wear and
the roll contour, the total camber of the work roll is calculated. The wear of the work
roll concerning profile and flatness has a similar impact as the thermal crown. The roll
wear model is used to determine the reduction of the work roll diameter due to roll
wear between the strip and the work roll, and is given by the empirical formula:
Cwear(z) = LSt Kwear (pWB(z) + pWSt (z)) (12.60)
Automation of Hot-strip Mill in an Integrated Steel Plant 693
where
Cwear = Roll wear
LSt = Strip length
Kwear = Wear coefficient
pWB = Load between work roll and backup roll
pWSt = Load between work roll and strip.
mm
Thermal crown
120
80
40
-40
Wear of work rolls
-80
-120
-mm
Figure 12.31 Thermal work roll crown and wear of work roll.
The rolling force is adjusted to achieve the required roll gap profile. For each
strand, the required rolling force defines the pass schedule. Under consideration of roll
crown, thermal crown, shifting position and bending force, the shape control curves lead
to an allowed roll force range to fulfil the requirements of profile and flatness.
To achieve a constant profile in the first strand of finishing mill, influence of
varying rolling force is compensated by work roll bending in the first strand. The
shifting position for all strands is determined in such a way that the range of roll
bending for the strands is maximum. Control of flatness is done in exit-strands with the
help of work roll bending. Figure 12.32 shows a schematic overview of the calculation
of pass schedule done by rolling strategy. When strip enters the first strand, the
calculation of pass schedule for remaining strands of the finishing mill is calculated after
the first strand has been loaded and before loading of next strand.
Figure 12.33 shows the block diagram of set-up models for finishing mill of a hot-
strip mill.
Roll contour
Roll
Thermal roll crown crown
model
Wear of work rolls Unloaded total roll crown
Strip thickness
Profile and
flatness
model
Strip width
Strip parameter
Pass Pass schedule
schedule
Mill parameter model
target values for temperatures and cooling rates are achieved as accurately as possible.
The cooling model solves the following nonlinear heat equation:
6 È 6 Ø
S E
6 M
6 Ù (12.61)
V [ ÊÉ [ Ú
with the nonlinear boundary condition at the upper surface of the strip and with a
similar condition, and is given as
6
M
6 K W
9HNQY
2 6 (12.62)
[
where
r = Density
c = Specific heat capacity
l = Thermal conductivity
Automation of Hot-strip Mill in an Integrated Steel Plant 695
thickness, temperature
thickness, temperature
crown
model equation
profile, flatness
model
Roll force
Adaption of model
Pass schedule calculation
parameter
Roll Actual measured
set-up values
x x x
x x x x
x
x x x
x x x
T = Temperature
t = Time coordinate
y = Thickness coordinate
ju = Heat flow
Wflow = Water-flow density
P = Cooling pattern.
The heat flow (ju) within the cooling zones mainly depends on the water-flow
density, which is a function of the cooling pattern (P) (on/off information for each of the
valves) and the surface temperature of the strip. Furthermore, the heat-flow due to
696 Process ControlCase Study
convection and radiation for regions where the strip is exposed to the surrounding air,
must be considered. The heat transfer coefficient from the strip to the coolant is a
strongly nonlinear function of strip temperature and water-flow density (Figure 12.34).
Unstable film-boiling
a
Water flow density,
Stable film-boiling
T
Burn-out point Leidenfrost point
The main task of the cooling model is the pre-calculation of the cooling pattern (P),
which ideally meets the quantity and production requirements expressed by the cooling
strategy for the individual strip. Besides the cooling temperature, cooling strategy
contains information such as
· The cooling mode (water cooling at the upper, the lower or at both sides)
· When to start cooling (early or late cooling)
· An optimum cooling rate
· A warmer or even uncooled head or tail of the strip.
For equal cooling strategy, the cooling pattern may differ for different strips due
to the different mill-exit temperatures, different strip velocities, different thickness
(results of the mill set-up calculation) or different material behaviour.
In addition to the cooling pattern sensitivities of the down-coiler coiling
temperature with respect to mill-exit temperature, the strip velocity and sensitivities of
single valves are computed to allow the lower-levelled control system to react very fast
for varying set-point conditions. In case of the temperature measurement at the
beginning of the fine cooling section a feedforward control with respect to this
intermediate temperature is also realized, making use of the computed sensitivities.
After the strip has entered the cooling section, a cyclic set-up calculation computes
modified cooling patterns and actual sensitivities for the actual set-point. As soon as
measured down-coiler coiling temperatures or intermediate temperatures are available,
Automation of Hot-strip Mill in an Integrated Steel Plant 697
Speed
Speed
Figure 12.35 shows the main components and relationships of the components of
the cooling model. The closed-loop control is formed by the cooling strategy, the cooling
models and the heat transfer adaptation.
698 Process ControlCase Study
RM FM Cooling
Set-up Set-up Set-up
Level-2
Field instrumentation
ROT
x Coiler x x x x x cooling
x x
x x x x x x Down
Crop shear coiler
Roughing mill (RM) Finishing mill (FM)
RhS
TV-rh
WR
StR
TV-hp
Fuel
RhT
RhT
SG HPT
HPT &
& EG
Air Wr LPT
LPT
se
se Grid
se
se
Feed water
Condenser
ExP
Feed-water heaters
train m
m
FwP TU
hT = Corresponding enthalpy of wT
hTR = Steam enthalpy at the HPT discharge
wR = RhS mass flow-rate
hR = Corresponding enthalpy of wR
h0 = Fluid enthalpy at the LPT discharge
aT,aR = Suitable constants (£1) accounting for the steam extractions (se)
from the HPT and the RhT and LPT, respectively.
It may be observed that the enthalpy drops (hT hTR) and (hR h0) remain
approximately unchanged as the plant load varies, because turbines are designed to
work with constant pressure ratio across their stages, while the steam flow varies. This
means that the output power Pm consists of two contributions, PHP and PLP, which are
approximately proportional to the ShS-flow and to the RhS-flow, respectively. In turn,
the flow of wT and wR are determined by the state of the SG (i.e. pressures and
temperatures) and by the hydraulic resistances that the turbines (together with their
control valves) present at the SG boundaries.
Steam extraction (se), as shown in Figure 12.37, mutually influence sub-system SG
and TU. Any variations in the principal steam flow (wT) create variations in SE-flow and,
consequently, a change in the feed-water temperature at the inlet of the SG. Feed-water
mass flow-rate, on the contrary, is essentially imposed by the FwP, which is generally
equipped with a flow control system which makes the FwP act as a flow-generator. The
overall gain of the process loop due to the steam extractions is rather small, so that the
feed-water temperature variations may be considered a small disturbance for the SG,
which is, ultimately, the sub-process where the fundamental dynamics take place.
Therefore, power plant control is a function of steam generator dynamics with the
turbine flow characteristics acting as boundary conditions at the steam side, the feed-
water mass flow-rate and the feed-water temperature acting as exogenous variables, and
Eqs. (12.63)(12.65), determining the power output.
When assessing control objectives, one of the crucial problems is to define system
performance. For power plants, one needs to address the following:
· The kinds of services the unit is required to perform, usually defined in terms
of maximal rate for large ramp load variations, the participation band for the
power-frequency control of the power system, and the maximum amplitude and
response time for the primary speed regulation in case of contingencies
· The maximal amplitude of temperature fluctuations during load variations, to
limit equipment stress due to creep or fatigue
· Maximal transient deviation of throttle pressure and drum level, to avoid
potentially dangerous conditions, evaluated for the largest disturbances (e.g. in
case of load rejection).
In this section, the most common operating condition for a drum boiler (i.e. with
throttle pressure controlled at a constant value during load variation) has been
considered. The main objective here is to return the pressure to the nominal value
within a reasonable time after the disturbance (i.e. the load variation), while strictly
insuring that it remains within safety limits (which may also depend on the amplitude
of the disturbance).
702 Process ControlCase Study
Pe
Pf Y Y
Turbine
P/PI
valve servo
Boiler
and
turbine
Pe Q Q
Combustion Pf
PI/PID control
Pe
Pe Y Y
P/PI Turbine
valve servo
Boiler
and
turbine
Pf Q Q
Combustion Pf
PI/PID
control
Pe
Pe Y Turbine Y
P/PI
valve servo
Boiler
K and
g turbine
Pf Q Q
PI/PID Combustion Pf
control
the positioning of the throttle valve. As a result, the turbine-following control allows a
rapid regulation of throttles pressure and slow, but stable, regulation of generation.
Consequently, the turbine-following control is preferred for plants not used for the load-
following.
The boiler-following architecture, as shown in Figure 12.38(b), produces sub-
stantially more rapid responses to generation commands but they can be quite
oscillatory. Moreover, the pressure response is typically oscillatory.
Modern requirements of the load-following have led to the widespread use of two-
input, two-output pressure and generation control. There are number of approaches to
coordinated control. One configuration, commonly referred to as coordinated control or
integrated control, is shown in Figure 12.38(c). Properly designed coordinated-control
systems can provide excellent response to load demand changes.
Generators speed naturally synchronize because of their interconnection via the
electrical network. Ultimately, the (steady state) network synchronous speed is
704 Process ControlCase Study
Bf w
P
Pe
Pe P/PI Y
Pe
Figure 12.39 Frequency bias modification of boiler-following or coordinated control configuration.
Q Y
Position servo
wf
y
d Drum and
P/PI PID Valves circulation
loop
yd
(a) Single-element
Q Y
Position servo
wf
y Drum and
d
P/PI PID Valves circulation
loop
yd
(b) Two-element
Q Y
Position servo
wf
y Drum and
d
P/PI PID Valves circulation
loop
yd
(c) Three-element
Figure 12.40 A typical drum-level configuration.
Temperature control
An important goal of a plant control is to regulate steam temperature at every point of
the turbine, i.e. at the superheater and reheater outlets. For accomplishing this, there
are a number of control means. The most direct ones are attemporators which inject
water at the heat-exchanger inlet. By moderating the fluid temperature entering the
heat exchanger, it is possible to control the outlet temperature. Other possibilities are
associated with adjusting the heat transferred to the fluid as it passes through the
exchanger. This can be accomplished by changing the mass flow-rate of the gas past the
heat transfer surfaces with recirculated gas or excess air-flow, or the gas temperature
at the exchanger surfaces by altering the burner positions or tilt of the burners. Some
times, a combination of these methods is employed.
706 Process ControlCase Study
The dynamics of superheaters or reheaters is such that the response of the outlet
temperature to a change in inlet temperature is characterized by a series of first-order
lags with time constants that vary inversely with the steam-flow rate through the heat
exchanger. Because of the significant time delay of the outlet temperature response, a
cascade control arrangement, as shown in Figure 12.41, is typically required for a
temperature regulation. The attemporator outlet temperature is a convenient
intermediate feedback variable, although, depending on the heat-exchanger
construction, other intermediate steam temperatures may also be available for
measurement. Because of the strong dependence of the time lag on steam flow,
parameterization of the regulator parameters on steam flow is necessary for good
performance over a wide load range. Some control systems incorporate disturbance
feedforward.
Wx Qx
Disturbance feedforward
Computer
Gain scheduling
Txo
PI/PID PI/PID Attemporator Attemporator Txo
Sensor Iao
Sensor
Combustion control
The main purpose of the combustion control system is to regulate the fuel and air
inputs into the furnace to maintain the desired heat input into the steam generation
process while assuring appropriate combustion conditions (excess oxygen). In most
instances, regulating the furnace gas pressure is a secondary, but important, function
of the combustion controller. A typical combustion-control configuration is shown in
Figure 12.42. The heating-rate command-signal is generated by the pressure-generation
controller, as shown in Figure 12.38(a).
Pg Pressure Pg
regulator ID fans
Wa
Wa
Air FD fans Furnace
Q Master regulator
combustion Q
control Fuel Fuel Q
regulator delivery
The details of the combustion control system significantly depend on the type of
fuel. Oil and gas are typically regulated with flow-control valves. These controls are
usually fast. Also, oil- and gas-flow and the caloric content of the fuel can be reliably
measured. Pulverized coal presents a different situation. The fuel-flow is regulated by
adjusting the feeder speed (which directly changes the flow rate of coal into the
pulverizer) and the primary air flow (the air flow through the pulverizer that carries the
pulverized coal into the furnace). The pulverizing process is quite slow and adds delay
in fuel-delivery process. Moreover, the flow rate of coal is difficult to estimate accurately
and coals calorific content varies. The master combustion control proportions the fuel
and air requirements and establishes set-points for the lower-level controllers.
Plant limits
limits evaluator
limits Unit Boiler
level
Turbine Boiler
Astrom, Karl J. and Bjoorm Wittenmark, Computer Controlled Systems: Theory and
Design, Prentice Hall, Englewood Cliffs, New Jersey, 1990.
Bateson, Robert N., Introduction to Control System Technology, 5th ed., Prentice Hall,
Englewood Cliffs, New Jersey, 1993.
Bequette, Wayne B., Process Control: Modeling, Design, and Simulation, Prentice-Hall
of India, New Delhi, 2003.
Borrie, John A., Modern Control Systems, Prentice Hall, Englewood Cliffs, New Jersey,
1986.
Brown, Martin and Chris Harris, Neuro Fuzzy Adaptive Modeling and Control, Prentice
Hall, Englewood Cliffs, New Jersey, 1994.
Cheremisinoff, N.P., Process Level Instrumentation and Control, Marcel Dekker,
New York, 1981.
Considinc, Douglass M. (Editor-in-Chief), Process Instruments and Control Handbook,
4th ed., McGraw Hill, Singapore, 1993.
, Encyclopedia of Instrumentation and Control, McGraw Hill, New York, 1971.
Coughanowr, Donald R., Process Systems Analysis and Control, 2nd ed., McGraw Hill,
New York, 2000.
Deshpande, Pradeep B. and Raymond H. Ash, Elements of Computer Process Control
with Advanced Control Applications, 3rd printing, Instrument Society of America,
Prentice Hall, Englewood Cliffs, New Jersey, 1983.
Dorf, Richard C. and Bishop, Robert H., Modern Control System, 8th ed., Addison-
Wesley, 1999.
Eckman, Donald P., Automatic Process Control, Wiley Eastern, New Delhi, 1988.
Edward Bodinglano, C., Planning, Scheduling and Control Integration in the Process
Industries, McGraw Hill, New York, 1995.
709
710 Bibliography
Erickson, Kelvin T. and John L. Hedrick, Plantwide Process Control, John Wiley &
Sons, New York, 1999.
Fabian Garcia Nocetti, D. and Peter J. Flemming, Parallel Processing in Digital Control,
Springer-Verlag, Berlin, 1992.
Fisher, P.E. and G. Thomas, Batch Control Systems Design, Application and
Implementation, Instrument Society of America, New Jersey, 1990.
Frank Dsouza, A., Design of Control Systems, Prentice Hall, Englewood Cliffs,
New Jersey, 1988.
Franklin, Gene F., Powell, David J. and Workman, Michael, Digital Control of Dynamic
Systems, 3rd ed., Addison-Wesley, 1998.
Garrett, Patrick H., Advanced Instrumentation and Computer I/O Design: Real-time
System Computer Interface Design, IEEE Press, New York, 1994.
Gayakward, Ramakant and Leonard Sokoloff, Analog and Digital Control Systems,
Prentice Hall, Englewood Cliffs, New Jersey, 1989.
Gupta, Madan M. and Naresh K. Sinha, Intelligent Control System: Theory and Practice,
IEEE Press, New York, 1996.
Harriott, Peter, Process Control, 23rd Reprint, Tata McGraw-Hill, New Delhi, 1964.
Heath, Walter S., Real-time Software Techniques, Van Nostrand Reinhold, New York,
1991.
Hordeski, Michael F., Control Technology and Personal Computers, Van Nostrand
Reinhold, New York, 1992.
Isekmann, Rolf, Digital Control Systems, Vol. 1, Fundamental Deterministic Control,
2nd revised ed., Springer-Verlag, Berlin, 1989.
Isermann, Rolf, Digital Control Systems, Vol. 2, 2nd revised ed., Springer-Verlag, Berlin,
1991.
Johnson, Curtis D., Process Control Instrumentation Technology, 8th ed., Prentice-Hall
of India, New Delhi, 2006.
Johnson, Curtins D., Process Control Instrumentation Technology, John Wiley & Sons,
New York, 1982.
Jovie, Fran, Expert Systems in Process Control, Chapman & Hall, London, 1992.
Kane, Les (Ed.), Handbook of Advanced Process Control Systems and Instrumentation,
Gulf Publishing (Book Division), Houston, 1987.
Kuo, Benjamin C., Automatic Control Systems, 3rd ed., Prentice-Hall of India,
New Delhi, 1979.
Lamba, S.S. and Y.P. Singh, Distributed Computer Control Systems, Tata McGraw-Hill,
New Delhi 1992.
Bibliography 711
Landon, Yoan D., Adaptive Control: The Model Reference Approach, Vol. 8, Marcel
Dekker, New York, 1979.
Laplante, Phillip A., Real-time Systems Design and Analysis, An Engineers Handbook:
2nd ed., IEEE Press, New York, 1997.
Lavine, William S. (Ed.), The Control Handbook, IEEE Press, New York, 1996.
Leri, Shem-Tov. and Ashok K. Agrawala, Real-time System Design, McGraw Hill,
New York, 1990.
Liptak, Bela G. (Editor-in-Chief), Instrument Engineers Handbook: Process
Measurement and Analysis, Butterworth Heinemann, Oxford, 1995.
, Instrument Engineers Handbook, Vol. 2, Chilton Book, Philadelphia, 1970.
, Process Measurement: Instrument Engineers Handbook, revised ed., Chilton
Book, Randnor, Pennsylvania, 1982.
Lu, Yong-Zai, Industrial Intelligent Control Fundamentals and Applications, John Wiley
& Sons, New York, 1996.
Luyben, Michael L. and William L. Luyben, Essentials of Process Control, McGraw Hill,
New York, 1997.
Luyben, William L., Process Modeling, Simulation, and Control for Chemical Engineers,
2nd ed., McGraw-Hill, New York, 1990.
Marlin, Thomas E., Process Control: Designing Processes and Control Systems for
Dynamic Performance, McGraw Hill, New York, 1995.
Middleton, Richard H. and Graham C. Goodwin, Digital Control and Estimation: A
Unified Approach, Prentice Hall, Englewood Cliffs, New Jersey, 1990.
Mosca, Edoardo, Optimal Predictive and Adaptive Control, Prentice Hall, Englewood
Cliffs, New Jersey, 1995.
Perdikaris, George A., Computer Controlled Systems: Theory and Applications, Kluwer
Academic, The Netherlands, 1991.
Philips, Charles L. and H. Troy Nagle, Digital Control System: Analysis and Design,
3rd ed., Prentice Hall, Englewood Cliffs, New Jersey, 1995.
Rijnsdrop, John E., Integrated Process Control and Automation, Elsevier Science,
The Netherlands, 1991.
Roffel, Brian, Peter J. Vermeer and Patrick A. Chin, Simulation and Implementation of
Self-tuning Controllers, Prentice Hall, Englewood Cliffs, New Jersey, 1989.
Sanchez, Martin, Juan M. and Jose Rodellar, Adaptive Predictive Control for the
Concepts to Plant Optimization, Prentice Hall, Englewood Cliffs, New York, 1996.
Seborg, Dale E., Thomas F. Edgar and Duncan A. Mellichamp, Process Dynamics and
Control, John Wiley & Sons, New York, 1989.
712 Bibliography
Shinskey, F.G., Process Control Systems Application, Design and Adjustments, 3rd ed.,
McGraw Hill, New York, 1988.
Shunta, Joseph P., World Class Manufacturing through Process Control, Prentice Hall,
Englewood Cliffs, New Jersey, 1995.
Singh, S.K., Computer-Aided Process Control, Prentice-Hall of India, New Delhi, 2003.
Singh, S.K., Industrial Instrumentation and Control, 2nd ed., Tata McGraw-Hill,
New Delhi, 2003.
Smith, Carlos A., Principles and Practices of Automatic Process Control, 2nd ed., John
Wiley & Sons, New York, 1997.
Stephanopoulos, George, Chemical Process Control: An Introduction to Theory and
Practice, Prentice-Hall of India, New Delhi, 1984.
Stock, Michael, AI in Process Control, McGraw Hill, New York, 1989.
Van De Vegte, John, Feedback Control Systems, Prentice Hall, Englewood Cliffs,
New Jersey, 1994.
Warwick, Klvin and Ming T. Tham (Eds.), Fail Safe Control Systems, Chapman & Hall,
London, 1991.
Yan, John, Roza Longan and Lofh A. Zadeh, Industrial Aplications of Fuzzy Logic and
Intelligent System, IEEE Press, New York, 1995.
Answers to Self-Check Quizzes
Chapter 1
A State True/False
713
714 Answers to Self-Check Quizzes
1.11 disturbances
1.12 minimize
1.13 zero
1.14 feedforward
1.15 multivariable, manipulated
1.16 input, time integral
1.17 three-mode
Chapter 2
A State True/False
Chapter 3
A State True/False
3.1 controlled
3.2 controlled
3.3 step change
3.4 active
Answers to Self-Check Quizzes 715
Chapter 4
A State True/False
4.1 unstable
4.2 upper, lower
4.3 roots, characteristic
4.4 left-half
4.5 relative stability
4.6 right, imaginary
4.7 stability
4.8 stability
4.9 positive
4.10 stable, degree, roots
4.11 process, controller, disturbance
4.12 roots, s-plane
4.13 damping coefficient, time constant
4.14 cosine, radial line
4.15 radial distance
4.16 feedback
4.17 characteristic equation, infinity
4.18 imaginary part, real part
716 Answers to Self-Check Quizzes
Chapter 5
A State True/False
Chapter 6
A State True/False
Chapter 7
A State True/False
Chapter 8
A State True/False
Chapter 9
A State True/False
9.1 software
9.2 real-time software
9.3 operating system
9.4 sequential program
9.5 disk-based
9.6 two-task
9.7 (a) foreground (b) background
9.8 interrupt-only
9.9 round robin, pre-emptive priority
9.10 sequentially
9.11 pre-emptive priority
9.12 multi-tasking or multi-programming
9.13 Time-sharing
9.14 compiler
9.15 Jackson and Simpson, 197175
9.16 1987
9.17 Modular Approach to Software Construction, Operation and Test
Chapter 10
A State True/False
Chapter 11
A State True/False
11.1 configuration
11.2 operator
11.3 algorithm
11.4 coaxial cable
11.5 single-input, single-output
Index
721
722 Index