Lecture 16 - Controller Structures

K. J. strm 1. Introduction 2. Feedback and Feedforward 3. Linear Schemes 4. Nonlinear Schemes 5. Gain Scheduling and Adaptation 6. Summary
Theme: Building complex control systems.

Introduction
Many common issues in design of machines, electronics, computer software, mechatronics How to deal with complexity Modularization Standardization Structures Paradigms, Design principles Top Down and Bottom Up

Bottom Up Design
A way to view systems A number of building blocks Ideas to combine them What are the building blocks of control? What principles can be used to select and combine them? The danger: Can it be done better? Commissioning: Close loops one by one.

Bottom Up Design of Control Systems

Components Controllers Observers Estimators Filters Limiters Dead zones Selectors System principles Feedback Feedforward Model following Cascade Split range Gain scheduling Adaptation

K. J. strm August, 2001

Top Down Design of Control Systems

Model complete system Design an integrated system System concepts
State feedback Observers Model predictive control

Feedback
A very powerful idea with dramatic impact
Controller y sp Process u G(s) y

Commissioning: Needs careful consideration.

+ Reduce effect of disturbances + Reduce effect of process variations + Linearize nonlinear systems + Does not require accurate process model - Measurement noise is injected into the system - Risk for instability

Feedforward
Process Disturbance v Gv

Feedback and Feedforward

Feedback Closed loop Acts only when there are deviations Feedforward Open loop Acts before deviations show up Planning Not robust to model errors S = 1 for all frequencies No risk for instability

G ff Feedforward

Gu

Market Driven Robust to model errors S < 1 for some some frequencies Risk for instability

Control signal

+ Reduce effects of disturbances that can be measured + Improve response to reference signals + No risk for instability - Requires good models
c K. J. strm August, 2001 2

Combination of Feedback and Feedforward

Feedforward v G ff1

Linear Schemes
Model following - Systems with two degrees of freedom (2DOF) Filters Cascade control State feedback Observers Attenuation of disturbances with specic character The Smith Predictor Model Predictive Control

G ff2 Process

yc

Gm

y sp

G fb

G p2

y G p1

Model Following - 2DOF

yc Model y sp

Applications of Model Following

y

e Controller

Process

Coordination in multi-axis motion control Robotics Path following Mixing in chemical processes Coordinated production changes

-1

Feedforward yc y sp

u ff

Model

Controller

Process

K. J. strm August, 2001

Filters
Typical lters Low pass High pass Band pass Notch Body bending lters Typical applications Reduce disturbances Improve robustness (high frequency roll-off) Smooth reference signals
0.4 0 0

Cascade Control
How to use several sensors. State feedback is the ultimate case!
Process y sp Cp Cs u P1 ys P2 y

Inner loop

Outer loop

10

20

30

0 0.5 1 0 10 20 30

When is Cascade Control Useful?

Key idea: make tight feedback around essential places where there are essential perturbations (disturbances or uncertainties) Guidelines: Well dened relation between primary and secondary variables Essential disturbances and process variations in inner loop Inner loop faster Tight feedback (high gain and high bandwidth) in inner loop
A u B u

When is Cascade Control Useful?

v T=1 v T = 10 ys T=1 v ys y y E u v ys y ys T = 10 y u ys D v y

C u

K. J. strm August, 2001

Attenuation of Disturbances with Specic Character

e 1 1 + sT

Systems with Time Delays

The derivative of the output gives poor prediction for systems with time delay Better predictions are possible by using past control signals u(t ), 0 < < T )d

Idea: Exploit model of disturbances (internal model principle) Constant disturbances (Integral Action) Sinusoidal disturbances Periodic disturbances A disturbance observer is an alternative.

2 as s2 + 2 as + a2

Replace the regular PID controller

1 u = ke + ki
t

e(s)ds kd

d yf dt

by the PPI (Predictive PI) controller

e

e sT

u = ke +

1 ki

e(s)ds k p

t L

u(s)ds

A simple form of the Smith predictor

The Smith Predictor

y sp

Model Predictive Control

Controller Process d Gf Gm

Gp

Gm 1 e

Beautifully simple Are there some snags? Design controller C as if there were no time delays in the process.
Cancellations may degrade performance Does not work if process is unstable Less general than state feedback

Widely used in process industries

c K. J. strm August, 2001 5

State Feedback and Observers

uc

Nonlinear Schemes
Limiters
y

Model and Feedforward xm Generator

u ff u fb L Process

Split range Ratio control Selectors Fuzzy control Gain scheduling Neural networks Adaptation

Observer

Use model to estimate variables that are not directly measurable States are the variables required to account for storage of mass, momentum and energy Estimate the state Feedback from full state deviation Feedforward to generate um and ym

Limiters
Limiters are often used To avoid saturation An element in circuits for windup protection To protect equipment to rapid changes A simple amplitude limiter

Rate Limiter

vlim=2, k=5, 100 1

1 s

y
0

1
u ul uh y
1 0 1 2 3 4

A rate limiter causes delays (JAS)

K. J. strm August, 2001

Jump and Rate Limiter

vlim=2, alim=0.5 1

Split Range
A simple way to use one controller to control two actuators. Commonly used for heating and cooling.
Heating valve Open

1 0 1 2 3 4

Commonly used in the power industry for load changes to save boilers.
Closed 0 0.5

Cooling valve

1.0

Ratio Control
Arrangement to obtain two ows that are proportional to each other, e.g. oil and air in boilers
A y a SP Div yk PI y yk PV y yk u B b a a(yk +b) SP PI PV u
y sp C un

Selector Control
Scheme used to achieve several control objectives, e.g. control temperature unless pressure is too high. A way to constrain process variables during operation.
PV SP M I u N uh

zmax

Cmax

M A X

G1

G2

The scheme B is preferable! Why?

zmin

SP PV

Cmin

ul

Process

K. J. strm August, 2001

Control of Fuel and Air in a Boiler

An elegant solution
Oil

Fuzzy Control
Rule based control Linguistic variables high, low, medium Membership functions If temperature high then increase ow a little

M I N Power demand M A X

MV SP

PI

Linguistic variables e de dt e u Defuzzifier

0.5

SP MV

PI

Linear Filter

Fuzzyfier

Fuzzy Logic

0.5

1 1 0.5 0 0.5 1 1 0 0.5 0.5 1

Air

Fuzzy Control
A nonlinear state feedback

Neural Networks
Representation of functions of many variables
y(t) = f ai ui(t)

How do we get the states? What does the nonlinearity look like? Rules and interpolation Why so few rules When is it useful

Real and articial neurons

u1 u2 un
1

Feedforward neural network

u1

w1 w2 wn
u3 u4 u5 y2

Excellent to automate successful manual operations Intuitive A lot of controversy: The No Model Myth Fuzzy control is more useful than its detractors claim but less useful than the propagandists claim Neuro-fuzzy
c

y
u2 y1

1 4 2 0 2 4

A nonlinear function with a learning mechanism!

8

K. J. strm August, 2001

Gain Scheduling
Controller parameters Gain schedule Operating condition Command signal Controller Control signal

Uses of Gain Scheduling

Many uses
Linearization of actuators Surge tank control Control over wide operating regions

Process

Output

Important issues

Example of scheduling variables Production rate Machine speed Mach number and dynamic pressure Room occupancy

Choice of scheduling variables Granularity of scheduling table Interpolation schemes Bump-less parameter changes Man machine interfaces

Importance of auto-tuning

Adaptation
Specification Self-tuning regulator Process parameters

Uses of Adaptation
Process dynamics

Tuning Tools
Controller design Controller parameters Reference Controller Process Input Output Estimation

Automatic Tuning Gain Scheduling Adaptive feedback Adaptive feedforward Integrated systems

Varying

Constant

Use a controller with varying parameters

Use a controller with constant parameters

Unpredictable variations

Predictable variations

Certainty Equivalence Many control and estimation schemes Dual control

Control should be directing as well as investigating!
c

Use an adaptive controller

Use gain scheduling

K. J. strm August, 2001