Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
054414 Process Control System Design LECTURE 11: SYSTEMATIC DISCRETE CONTROL DESIGN
Daniel R. Lewin Department of Chemical Engineering Technion, Haifa, Israel
11- 1 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
Objectives
On completing this section, you should: Be able to implement the IMC design procedure on discrete systems. Be able to compare common discrete control algorithms, namely:
The discrete PID algorithm (plus filter) Dead-beat Algorithm Dahlins Algorithm
using a systematic framework Be able to identify situations where adaptive controllers are needed, and be able to conceptually design such a system.
11- 2 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
LECTURE ELEVEN
d yyS S
+ +
q((z)) qq(s) z
p (z ) p(s)
+ +
yy
+
y = pq ys + (1 (z pq)d p -1 ) q=p f
q1 ( z ) =
(z
0.5 ) z +3
and q2 ( z ) =
(z
0.5 ) z 1.5
Non-causal
11- 4
Unstable
LECTURE ELEVEN
1 1 such that p pA = pM (stable, causal, and strictly proper) and the all-pass component, pA ( z ) , is defined as:
pA ( z ) = z
(1 ) ( z ) ( z ) (1 ) .
h i
1
i =1
Note that pA (e i T
) = 1
N is a time delay, selected so that pM ( z ) will be strictly proper, and h is the number of zeros in p ( z ) outside the unit circle.
11- 5 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
kz 3 z 0.5 kz q (z ) = pA1 ( z ) = z 4 , pM 1 ( z ) = z 0.5 kz z 0.5 z 1.5 pA 2 ( z ) = 0.667 z 1 2 (z 0.667 ) z 1.5 0.667 ( z 0.5 ) p2 ( z ) = 2 z ( z 0.667 ) q ( z ) = z ( z 0.667 ) (z 0.5 ) pM 2 ( z ) = 1.5 2 (z 0.5 ) p1 ( z ) =
11- 6 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
LECTURE ELEVEN
f (z ) =
dk (1 f (z ) ) z =1 = 0, 0 k < m dz k
input type (1 or 2)
1 =0 1 z 1 z =1
Systematic Discrete Design
z 1.5 z 3 2 ( z 0.5 )
= pq = pAf =
0.667 ( z 1.5 1 If ys = z/(z 1) (unit step): lim y (kT ) = lim (z 0.667 )(z)( ) ) z -3 =1 k z 1 i.e. design with q (1 ) = p 1 (1 ) guarantees offset-free response
11- 8
LECTURE ELEVEN
11- 9
As seen previously, any negative poles in q(z) lead to ringing (the controller output will exhibit a sign change in each sample). The closer these negative poles are to 1, the more severe will be the ringing.
11- 10
LECTURE ELEVEN
Example of Ringing
Consider the process p ( s ) =
(10 s +1)(25s +1)
1
sampled at T=3.
The equivalent discrete process as seen by a discrete controller (after ZOH) is (see Lecture 8):
p (z ) =
pM ( z ) =
pA ( z ) = z 1
11- 11
Example of Ringing
c (z ) =
1 p (z ) q (z )
q (z )
show ringing, the true response does, which is a result of the ringing exhibited by the input, due to the pole at z = -0.87.
PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
Note that while the output at the sample intervals does not
11- 12
LECTURE ELEVEN
qmod ( z ) = q ( z )
(z i ) i
=1
(1 i ) i
=1
z P =
q (z ) z P (1 i )
i =1 P
11- 13
qmod ( z ) = q ( z )
(z i ) i
=1
(1 i ) i
=1
z P =
q (z ) z P (1 i )
i =1 P
The modified controller, qmod(z), has the following properties: The negative poles zi = i , -1 i 0, have been substituted by poles at the origin, eliminating the ringing phenomenon. The P poles need to be added to keep the controller proper. The steady-state gain of the IMC controller has been conserved (why is this important?)
11- 14
LECTURE ELEVEN
(z + 0.87 )(z 0.5 ) 31.8 ( z 0.90 ) ( z 0.74 ) 17.0 qmod ( z ) = z (1 +z0.87 )( z 0.5 ) (z 0.5 )
The response with the modified controller now no longer satisfies the desired trajectory (note the overshoot!). However, we have eliminated the ringing. Thats life
11- 15
(z )
+
p (z )
Im(z)
p (e
iT
(eiT ) )
p (e iT )
(eiT ) =
Re(z)
11- 16
depends on T
Systematic Discrete Design
LECTURE ELEVEN
Computing
Compute,
m
(z) - Example
Solution:
pworst (z) = p(z) =
(1 + k ) (1 e T /(1) )
z
(1+ ) T
(1 e ) z (z e )
T T T
(z e
T /(1 )
iT z = e
(z) =
11- 17
Computing
= 0.1, = = k = 0.2
T = 0.5 T = 0.05
(z) - Example
= 1, = = k = 0.2
T = 0.5
T = 0.05
0 < < T
0 < < T
The results for T = 0.05 are close to those one would expect for a continuous system. For T = 0.5, the effect of uncertainty may be different, depending on the ratio of /.
11- 18 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
LECTURE ELEVEN
11- 19
(z )
10
LECTURE ELEVEN
This is the same equation used to translate an IMC controller to its classical equivalent! We shall use this interpretation to analyze several common discrete controllers: Discrete PID algorithm Dead-beat algorithm Dahlin algorithm
11- 21 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
11
LECTURE ELEVEN
Advantages: The velocity form is independent of initial condition, us. It is not subject to reset wind-up. Disadvantages: It responds aggressively to step changes in setpoint. The differential operator is sensitive to noise.
11- 23 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
ys(k)
11- 24
12
LECTURE ELEVEN
Dead-beat Algorithm
In dead-beat control, the desired response is to ensure zeroerror from the n+1 sample, where n = /T. The desired closed loop transfer function is: y ( z ) = ( z ) = z -(n+1) ys ( z ) (z ) z -(n+1) -1 = G -1 ( z ) Thus, CDB ( z ) = G ( z ) 1 (z ) 1 z -(n+1) Example. 1 e T 0.5 z 1 10 8.6z 1 T = 1, P(s) = , G(z) = HP(z) = 10 = 0.5s + 1 1 0.14z 1 1 e T 0.5z 1
Thus, CDB ( z ) = G -1 ( z )
11- 25
11- 26
13
LECTURE ELEVEN
Dahlins Algorithm
Recall that the IMC design for a delay process gives: y (s) 1 = (s) = e s s + 1 ys ( s ) This approach enables us to target not only the desired delay of the response, but also define how fast we would like it to settle (as determined by the value of ). Taking this idea into the discrete domain: (1 b)z-(n+1) (z ) = ,b = e T , = nT 1 bz-1 (z ) Thus, CDA ( z ) = G -1 ( z ) 1 (z )
= G -1 ( z )
11- 27
14
LECTURE ELEVEN
Hence, CDA ( z ) = G -1 ( z ) =
(1 b)z -3 ,b = e T 1 bz -1 (1 b)z -3
Q(z) =
Thus, DA and DB are both equivalent to IMC control with and without and a filter. In DA, selection of the filter parameter, b, allows robustness to be guaranteed.
11- 29 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
Here,
DA ( = 0.5 min)
DA ( = 2.0 min)
0.39z 3 1 0.61z 1 Note that the fastest response is obtained with DB, with the filter time constant provided a back-off on performance for DA. G(z) =
11- 30
15
LECTURE ELEVEN
Here,
DA ( = 0.5 min)
DA ( = 2.0 min)
0.39z 4 1 0.61z 1 With no filter, DB is unstable with delay uncertainty. DA with = 0.5 is close to the theoretical stability limit, and with = 2 gives a similar to nominal response. G(z) =
11- 31
Adaptive Control
In cases where parameter changes are expected to be large, robust control may lead to poor performance. Such cases are: Large, frequent disturbances (feed composition, feed quality) Batch operation (no steady-state) Inherent nonlinear behavior (pH control, highly nonlinear kinetics, catalyst decay, heat exchanger fouling, etc). A variety of adaptive control techniques are often used in these situations, where: The process changes are largely known and can be anticipated this calls for programmed adaptation, which is essentially a feedforward strategy. The process changes are largely unknown this calls for on-line adaptive control or self-tuning control, and involves feedback.
11- 32 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
16
LECTURE ELEVEN
Programmed Adaptation
Gain Scheduling based on knowledge of nonlinearity If the process gain is known to vary significantly, we can use a controller with a nonlinear gain: C(s) = f(e) x g(s): Options for f(e): a. Continuous gain (1 ) e , 0 1 f (e) = + 100 If 0, this leads to offset because the controller is insensitive to small e. b. 3-piece nonlinear controller
Kc , e eband f (e) = Kc,low , e < eband
11- 34 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
17
LECTURE ELEVEN
Self-tuning Controllers
When process changes can be neither measured nor anticipated, programmed adaptation cannot be used. Instead, a feedback strategy is employed, in which the controller is retuned on-line.
Controller computation
parameter estimates
output
11- 35
The figure shows a process in which a continuous flow of HCl, F2, at concentration CCl1 is neutralized using NaOH, F1. Both F2 and CCl1 are disturbances.
The pH of the exiting stream, F, must be kept at 7. The figure on the right shows the pH curve for this strong acidstrong base process.
11- 36 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
18
LECTURE ELEVEN
Assuming perfect mixing, reaction at equilibrium, constant volume and density, the system is described by: F (t ) = F1 (t ) + F2 (t )
dCCl 1 = (F (t ) CCl 1 (t ) F (t ) CCl (t ) ) dt V 1 60 dC Na 1 = (F (t ) CNa 2 (t ) F (t ) CNa (t ) ) dt V 2 60 pH = log C H , C H (t ) = KW COH (t ) , KW = 10 14 gmol2 /li2 COH (t ) = C Na (t ) + C H (t ) CCl (t )
Defining variables: P = F1 60 ,Q = F2 60 V V
Ch = CH
11- 37
KW , Cn 2 = C Na 2
KW , Cc 1 = CCl 1
KW
Combining equations and substituting dimensionless variables, we obtain the single ODE:
dCh 1 = (Ch + Cc 1C h2 C h3 ) P 1 + C h2 dt
+ C h + Cn 2C h2 C h3 Q , C h ( 0 ) = 1
11- 38
19
LECTURE ELEVEN
P: 1.5 1.6 at t = 5 s
11- 40
20
LECTURE ELEVEN
11- 41
Summary
On completing this section, you should: Be able to implement the IMC design procedure on discrete systems. Be able to compare common discrete control algorithms, using a systematic framework (IMC):
The discrete PID algorithm (plus filter) Dead-beat Algorithm Dahlins Algorithm
Much of the material covered in the continuous part of the course carries over with some exceptions.
Be able to identify situations where adaptive controllers are needed, and be able to conceptually design such a system.
11- 42 PROCESS CONTROL SYSTEM DESIGN - (c) Daniel R. Lewin Systematic Discrete Design
21