Lecture 16 - Controller Structures Introduction

K. J. Åström Many common issues in design of machines, electronics,

computer software, mechatronics
1. Introduction
• How to deal with complexity
2. Feedback and Feedforward
• Modularization
3. Linear Schemes
• Standardization
4. Nonlinear Schemes
• Structures
5. Gain Scheduling and Adaptation
• Paradigms, Design principles
6. Summary
• Top Down and Bottom Up
Theme: Building complex control systems.

Bottom Up Design Bottom Up Design of Control Systems

• A way to view systems System principles
Components
• A number of building blocks
• Controllers • Feedback
• Ideas to combine them
• Observers • Feedforward
• What are the building blocks of control?
• Estimators • Model following
• What principles can be used to select and combine them?
• Filters • Cascade
• The danger: Can it be done better?
• Limiters • Split range
• Commissioning: Close loops one by one.
• Dead zones • Gain scheduling
• Selectors • Adaptation

&
c K. J. Åström August, 2000 1
 Top Down Design of Control Systems Feedback
• Model complete system A very powerful idea with dramatic impact
• Design an integrated system Controller Process
y sp e u y
• System concepts Σ K G(s)

– State feedback
– Observers −1
– Model predictive control
• 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 Feedback and Feedforward

Process • Feedback • Feedforward
Disturbance v • Closed loop • Open loop
Gv
• Acts only when there • Acts before deviations
are deviations show up

u y • Market Driven • Planning

G ff Gu Σ • Robust to model errors • Not robust to model
e S e < 1 for some some errors e S e = 1 for all
Feedforward frequencies frequencies
Control signal • Risk for instability • No risk for instability

+ Reduce effects of disturbances that can be measured

+ Improve response to reference signals
+ No risk for instability
- Requires good models &
c K. J. Åström August, 2000 2
Combination of Feedback and Feedforward Linear Schemes
Feedforward
• Model following - Systems with two degrees of freedom
v
(2DOF)
G ff1 Σ G ff2
• Filters
yc y sp e u
Process
y • Cascade control
Gm Σ G fb Σ G p2 Σ G p1
• State feedback
• Observers
−1
• Attenuation of disturbances with specific character
• The Smith Predictor
• Model Predictive Control

Model Following - 2DOF Applications of Model Following

yc y sp e u y
Model Σ Controller Process • Coordination in multi-axis motion control
• Robotics
-1 • Path following
• Mixing in chemical processes
Feedforward
u ff
• Coordinated production changes
yc y sp e u y
Model Σ Controller Σ Process

−1

&
c K. J. Åström August, 2000 3
 Filters Cascade Control
Typical filters How to use several sensors. State feedback is the ultimate
case!
• Low pass
Process
• High pass y sp
Cp Cs
u
P1
ys
P2
y

• Band pass
Inner loop
• Notch
• Body bending filters Outer loop

Typical applications
0.4
• Reduce disturbances 0
0 10 20 30
• Improve robustness (high frequency roll-off) 0

• Smooth reference signals −0.5

−1
0 10 20 30

When is Cascade Control Useful? When is Cascade Control Useful?

A v D v
Key idea: make tight feedback around essential places where
there are essential perturbations (disturbances or uncertain- y
u ys y
ties) T=1 T = 10 u
Guidelines: B v ys

• Well defined relation between primary and secondary u ys y

variables T = 10 T=1
E y
• Essential disturbances and process variations in inner loop
C v u v
• Inner loop faster
u ys y ys
• Tight feedback (high gain and high bandwidth) in inner
loop

&
c K. J. Åström August, 2000 4
 Attenuation of Disturbances with Specific Systems with Time Delays
Character
• The derivative of the output gives poor prediction for
e
∑
υ systems with time delay
• Better predictions are possible by using past control
Idea: Exploit model of distur-
signals u(t − τ ), 0 < τ < T )d
1
1 + sT
bances (internal model principle)
• Constant disturbances (Inte- Replace the regular PID controller
e υ
gral Action) ∑ Z t
1 d yf
• Sinusoidal disturbances 2 ζ as u = ke + e( s) ds − kd
ki dt
s2 + 2ζ as + a2
• Periodic disturbances by the PPI (Predictive PI) controller
A disturbance observer is an Z Z
e υ t t
alternative. ∑ 1
u = ke + e( s) ds − kp u( s) ds
ki t− L
e − sT
A simple form of the Smith predictor

The Smith Predictor Model Predictive Control

Controller
d
Process
y sp u y
Σ Gf Gm
†
Gp Σ

Gm

−
e
−1 Σ

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

&
c K. J. Åström August, 2000 5
 State Feedback and Observers Nonlinear Schemes
uc Model and u ff • Limiters
Feedforward
xm u fb
Generator
∑ L ∑ Process
y
• Split range
−
• Ratio control
xˆ
Observer • Selectors
• Fuzzy control
• Use model to estimate variables that are not directly • Gain scheduling
measurable
• Neural networks
• States are the variables required to account for storage of
mass, momentum and energy • Adaptation
• Estimate the state
• Feedback from full state deviation
• Feedforward to generate um and ym

Limiters Rate Limiter

Limiters are often used
• To avoid saturation vlim=2, k=5, 100

• An element in circuits for windup protection

1

y
Σ
u e 1
• To protect equipment to rapid changes s
0

A simple amplitude limiter

−1
−1
u uh y
0 1 2 3 4
ul
A rate limiter causes delays (JAS)

&
c K. J. Åström August, 2000 6
 Jump and Rate Limiter Split Range
1
vlim=2, alim=0.5
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 Cooling valve
boilers.
Closed
0 0.5 1.0

Ratio Control Selector Control

Arrangement to obtain two flows that are proportional to each Scheme used to achieve several control objectives, e.g. control
other, e.g. oil and air in boilers temperature unless pressure is too high. A way to constrain
A B
process variables during operation.
y b a
a
yk a(yk +b) PV uh
SP u Σ Π zmax
SP
Cmax
PI SP
Div PV u
y PI M
PV y sp
yk y un I
u z y
yk C M N G1 G2
A
X
The scheme B is preferable! Why? zmin Process
SP ul
Cmin
PV

&
c K. J. Åström August, 2000 7
 Control of Fuel and Air in a Boiler Fuzzy Control
An elegant solution • Rule based control
Oil • Linguistic variables high, low, medium
• Membership functions
MV
• If temperature high then increase flow a little
M PI
I SP
N
Power Linguistic
1

demand variables 0.5

de
M dt
0

e u
A SP Linear
e Fuzzyfier Fuzzy
Defuzzifier
−0.5

X PI Filter Logic
MV −1
1
0.5 1
0 0.5
0
−0.5
−0.5
−1 −1

Air

Fuzzy Control Neural Networks

• A nonlinear state feedback Representation of functions of many variables
X

– How do we get the states? y( t) = f ai u i ( t)
– What does the nonlinearity look like?
– Rules and interpolation Real and artificial neurons Feedforward neural network
u1 w1
– Why so few rules
u1
– When is it useful u2
w2
y
• Excellent to automate successful manual operations un
u2 y1
wn
• Intuitive u3
1

• A lot of controversy: The No Model Myth 0

u4 y2

• Fuzzy control is more useful than its detractors claim but −1

−4 −2 0 2 4 u5
less useful than the propagandists claim A nonlinear function with a learning mechanism!
• Neuro-fuzzy
&
c K. J. Åström August, 2000 8
 Gain Scheduling Uses of Gain Scheduling
Controller
parameters Gain • Many uses
schedule

Operating
– Linearization of actuators
condition
– Surge tank control
Command
Control
signal
signal – Control over wide operating regions
Controller Process Output
• Important issues
– Choice of scheduling variables
– Granularity of scheduling table
Example of scheduling variables
– Interpolation schemes
• Production rate – Bump-less parameter changes
• Machine speed – Man machine interfaces
• Mach number and dynamic pressure • Importance of auto-tuning
• Room occupancy

Adaptation Uses of Adaptation

Self-tuning regulator
Specification Process parameters Process dynamics

• Tuning Tools
Controller
• Automatic Tuning
Estimation Varying Constant
design

Controller
parameters • Gain Scheduling Use a controller with Use a controller with
varying parameters constant parameters
Reference
Controller Process • Adaptive feedback
Input Output
• Adaptive feedforward Unpredictable Predictable
variations variations

• Integrated systems
• Certainty Equivalence Use an adaptive
controller
Use gain scheduling

• Many control and estimation schemes

• Dual control
– Control should be directing as well as investigating!
&
c K. J. Åström August, 2000 9