Lecture 11 - Feedforward Design Introduction

K. J. Åström Recall systems with two degrees of freedom

d n
1. Introduction
r F ∑ C u ∑ P x ∑
y
2. Attenuation of measured disturbances
3. Improved response to reference signals
4. System inverses −1
5. Summary

Theme: A simple and useful idea. System inverses. • Recall systems with two degrees of freedom

Feedback and Feedforward Reduction of Load Disturbances by Feedforward

• Feedback • Feedforward Process
• Closed loop • Open loop Disturbance v
Gv
• Acts only when there • Acts before deviations
are deviations show up
• Market Driven • Planning u y
G ff Gu Σ
• Robust to model errors • Not robust to model
 S  < 1 for some ω errors  S  = 1 for all ω Feedforward
• Risk for instability • No risk for instability Control signal

Feedforward is a nice complement to feedback. Properly used + Reduce effects of disturbances that can be measured
it can improve a control system substantially. Its use is
+ Improve response to reference signals
increasing. It requires good models and should always be used
together with feedback. + No risk for instability
- Requires good models

c K. J. Åström October, 2000 1
 Reduction of Disturbance that Can be Measured Combine with Feedback

Transfer function from disturbance to process output with

combined feedback and feedforward
Transfer function from disturbance to process output with pure Y ( s) P2 (1 − P1 G f f )
feedforward Gd = = = P2 (1 − P1 G f f ) S
D ( s) 1 + PC
Y ( s)
Gd = = P2 (1 − P1 G f f ) Notice combined effect of feedback and feedforward.
D ( s)

Sensitivity to Modeling Errors The Ideal Feedforward Compensator

Pure feedforward Transfer function from disturbance to process output
dGs dG f f Y ( s) P2 (1 − P1 G f f )
Gd = P2 (1 − P1 G f f ), = = = P2 (1 − P1 G f f ) S
Gs Gf f D ( s) 1 + PC

Combined feedback and feedforward What is a good feedforward compensator?

dGs dG f f The choice
Gd = P2 (1 − P1 G f f ) S, =S
Gs Gf f
Pyd
G f f = P1−1 = .
Pyu

gives perfect compensation. Sensitivity?


c K. J. Åström October, 2000 2
 Improved Response to Reference Signals by Response to Reference Signals
Feedforward
d n Mu um
r
r y ym u y
F ∑ C u ∑ P x ∑
My Σ Σ
C P

−1
−1

An alternative representation Notice that Mu and M y can be viewed as generators of the

desired output y m and the inputs u m which corresponds to y m.
Mu um
r Assume that process P, controller C and desired response M y
ym u y
My Σ C Σ P are given are given. Then M u = M y / P. Does not depend on C!

−1

Relation: F C = M y C + Mu

Design of Feedforward Example of Feedforward Design

Let that process P, controller C and desired response M y are 1 1
Assume P = and M y = then
given be given. Since Mu = M y / P and the transfer function s+1 sT + 1
should be stable, causal and not include derivatives we find
s+1 1
1−T 
that. Mu = = 1−
sT + 1 T sT + 1
• Unstable process zeros must be zeros of M y
The direct term (high frequency gain of M u ) is 1/T which
• Time delays of the process must be time delays of M y implies that fast response requires high gain of M u . Notice
• The pole excess of M y must be greater than the pole
Mu (∞) 1
excess of P =
Mu(0) T
Take process limitations into account!
Large control signals if T << 1!


c K. J. Åström October, 2000 3
 Example of Feedforward Design Discussion
1 1 Reduction of measurable disturbance
Assume P = and My = then
( s + 1) 4 (sT + 1)4
Y ( s) P2 (1 − P1 G f f )
Gd = = = P2 (1 − P1 G f f ) S
Mu =
(s + 1)4 D ( s) 1 + PC
(sT + 1)4
1
4(1 − T )(sT )3 + (6(1 − T 2 )(sT )2 + (4(1 − T 3 )sT + 1 − T 4  Response to reference signals
= 4 1−
T (sT + 1)4
Y ( s) P( CM y + Mu)
Direct term is 1/T 4 which implies that fast response requires high = = M y + ( PMu − M y ) S
R( s ) 1 + PC
gain of Mu . Notice
Mu (∞) 1 Feedback will deal nicely with frequencies where the sensitivity
= 4
Mu (0) T is small. Feedforward only needs to act for frequencies where
the sensitivity is close to one. Approximate models can be
The high frequency gain of Mu depends strongly on the pole excess used.
and the speed up that we require. Bounds on the control signal limit
how fast response we can obtain.

System Inverses Approximate Inverses

A feedforward compensator requires inversion of a system or Since it is difficult to obtain an exact inverse we have to
part of the system dynamics. Formally easy to invert transfer approximate. One possibility to find an approximate inverse
functions because G −1 (s) = G1(s) . But! G † of the transfer function G is to find the transfer function X(s)
which minimizes
G ( s) =
1
, G −1 ( s) = s + 1  ∞
s+1 J= (u2 ( t) − v2 ( t)) dt
0
requires differentiation!
where V (s) = G (s) X (s) U (s) and u is a given input signal.
e− s −1 Notice that the approximate inverse is matched to a particular
G ( s) = , G ( s) = ( s + 1) e s
s+1 input u.
requires differentiation and prediction
s−1 s+1
G ( s) = , G −1 ( s) =
s+1 s−1

The inverse is unstable


c K. J. Åström October, 2000 4
 Approximate Inverses Use of Feedforward
Solving an optimization problem for step signals give the • Temperature control in buildings. Measure outside temper-
following approximate inverses ature and adjust heating
1−s • Level control in steam generators. To deal with shrink and
G ( s) = , G † ( s) = 1 swell.
1+s

Steam valve
G ( s) = e− s, G † ( s) = 1 F F
Feed
water
L Drum

Oil Turbine
Air
Raiser Down comer

Summary
• Feedforward is a useful technique.
• It can be used to improve response to reference signals
and reduction of the effect of measurable disturbances
• Feedforward has poor sensitivity S = 1!
• Useful to combine with feedback
• Use of feedforward is increasing


c K. J. Åström October, 2000 5