Lecture 8 - The Basic Feedback Loop Introduction

K. J. Åström • A nice collection of tools have been developed

• We have looked at a few examples
1. Introduction
• We will now investigate a typical control problem
2. A basic control problem
• A basis for control system design
3. Systems with two degrees of freedom
• How to judge a control system
4. Basic relations
• New concepts and insight
5. The sensitivity functions
– Sensitivity functions
6. Summary
– What is the effect of feedback
Theme: Understanding the basic feedback loop. Systems with • A basis for a serious look at the design problem
two degrees of freedom. The gangs of four and six. Sensitivity
• How to capture a complex reality in tractable mathematics
functions

A Basic Control System Key Issues

d n
Find a controller that
r F ∑ C u ∑ P x ∑
y A: Reduces effects of load disturbances
B: Does not inject to much measurement noise into the
system
−1
C: Makes the closed loop insensitive to variations in the
process
Ingredients: D: Makes output follow command signals well
• Controller: feedback C, feedforward F Convenient to use a controller with two degrees of freedom,
• Load disturbance d: Drives the system from desired state i.e. separate signal transmission from y to u and from r to u.
• Measurement noise n: Corrupts information about x A complete separation of the problem: Use feedback to deal
• Command signal or reference signal r gives desired with A, B, and C. Use feedforward to deal with D!
process variable x

c K. J. Åström October, 2000 1
 System with Two Degrees of Freedom Designing System with Two Degrees of Freedom
l n
Design procedure:
r
F ∑ C
u
∑ P
x ∑
y • Design the feedback C to achieve
– Small sensitivity to load disturbances l
– Low injection of measurement noise n
−1
– High robustness to process variations
• Then design the feedforward F to achieve good response
The system is said to have two degrees of freedom (2DOF) to command signals r
because there are separate signal paths from reference r to For many problems in process control the load disturbance re-
control u and from measured signal y to u. sponse is much more important than the set point response.
We have already encountered this in PI control The set point response is more important in motion con-

trol. Few textbooks and papers show more than set point re-
t
u( t) = k( br( t) − y( t)) + ( r(τ ) − y(τ )) dτ sponses.
0

Many Versions of 2DOF Relations between signals

yc y sp e u y d n
Model Σ Controller Process

r F ∑ C u ∑ P x ∑
y
-1

Feedforward
u ff −1
yc y sp e u y
Model Σ Controller Σ Process

P PC PCF
−1 X = D− N+ R
1 + PC 1 + PC 1 + PC
P 1 PCF
Horowitz: "Some structures have been presented as funda- Y= D+ N+ R
1 + PC 1 + PC 1 + PC
mentally different from the others ... all 2DOF configurations PC C CF
U =− D− N+ R
have basically the same properties and potentials." 1 + PC 1 + PC 1 + PC

FC = M y + Mu C = M + Mu C

c K. J. Åström October, 2000 2
 Some Observations A Possible Choice
• A system based on error feedback is characterized by four Six transfer functions are required to show the properties of a
transfer functions (The Gang of Four) basic feedback loop. Four characterize the response to load
• The basic feedback system with two degrees of freedom is disturbances and measurement noise.
characterized by six transfer function (The Gang of Six) PC P
• To fully understand a system it is necessary to look at all 1 + PC 1 + PC
C 1
systems
1 + PC 1 + PC
• It may be strongly misleading to only show properties of
a few systems for example the response of the output to Two more are required to describe the response to set point
command signals. This is a common error in the literature. changes.
• The properties of the different transfer functions can be PCF CF
illustrated by transient or frequency responses. 1 + PC 1 + PC

Amplitude Curves of Frequency Responses Step Responses

PI control k = 0.775, Ti = 2.05 of P(s) = (s + 1)−4 with PI control k = 0.775, Ti = 2.05 of P(s) = (s + 1)−4 with
M (s) = (0.5s + 1)−4 M (s) = (0.5s + 1)−4
PC F PC P PC F PC P
1+ PC 1+ PC 1+ PC 1.5
1+ PC 1.5
1+ PC 1.5
1+ PC

0 0 0
10 10 10 1 1 1

0.5 0.5 0.5

0 0 0
−1 −1 −1
10 10 10 0 10 20 30 0 10
−1
10
0
10C F
1
10
−1
10
0
10 C
1
10
−1
10
0
10 1
1
10 CF C20 30 0 10 20
1 30

1+ PC 1+ PC 1 1+ PC 1.5
1+ PC 1.5
1+ PC 1.5
1+ PC
10
1 1
10 10
1 1 1

0.5 0.5 0.5

0
0
10
0
10 10 0 0 0

−1 0 1 −1 0 1 −1 0 1 0 10 20 30 0 10 20 30 0 10 20 30
10 10 10 10 10 10 10 10 10


c K. J. Åström October, 2000 3
 An Alternative Remember
Show the responses in the output and the control signal to
a step change in the reference signal for system with pure
error feedback and with feedforward. Keep the reference signal Please remember to always look at all responses
constant and make a unit step in the process input. Show the when you are dealing with control systems.
response of the output and the control signal.
1.5
Response of y to step in r
1
1

0.5

0
0 10 20 30 40 50 60
0.5
2

1.5

0.5 0
0 0 1 2 3 4 5
−0.5
0 10 20 30 40 50 60

Four Responses The System

1
Response of y to step in r Response of y to step in d Process P(s) =
1 80 s−1
60 s−1
Controller C(s) =
0.5 40 s
20 Response of y to reference r
0 0
0 1 2 3 4 5 0 1 2 3 4 5 Y ( s) PC 1
Response of u to step in r Response of u to step in d = =
1 0 R( s) 1 + PC s+1

0.5
Response of y to step in disturbance d
0 −0.5
Y ( s) P s s
−0.5 = = 2 =
D ( s) 1 + PC s −1 ( s + 1)( s − 1)
−1 −1
0 1 2 3 4 5 0 1 2 3 4 5

What is going on?


c K. J. Åström October, 2000 4
 Focus on Feedback The Feedback Problem
d n d n
0 u x y 0 u x y
Σ C Σ P Σ Σ C Σ P Σ

−1 −1

• Neglect following of reference signals (the feedforward The signals have the following relations. Notice that there are
problem). only four transfer functions. The Gang of Four.
• Focus on on the feedback problem P PC
X = D− N
– Load disturbances 1 + PC 1 + PC
P 1
– Measurement noise Y= D+ N
1 + PC 1 + PC
– Model uncertainty PC C
U =− D− N
1 + PC 1 + PC

The Loop Transfer Function L(s) = P(s) C (s) The Sensitivity Functions
Tells a lot about the system. The transfer functions
1 1 1
10
• Sensitivity function S = =
1 + PC 1+ L
0
10
PC L
• Complementary sensitivity function T = =
10
−1 1 + PC 1+ L
are called sensitivity functions. They have interesting properties
−2
10
−1
10
0
10 and useful physical interpretations. We have
0 • The functions S and T only depend on the loop transfer
−100
function L

−200
• S+T =1
−300
• Typically S (0) small and S (∞) = 1 and consequently
T (0) = 1 and T (∞) small
−400
−1 0
10 10


c K. J. Åström October, 2000 5
 Quiz Disturbance Reduction
d n
Look at the block diagram
d n 0 u x y
Σ C Σ P Σ
r F ∑ C u ∑ P x ∑
y
−1

−1
Output without control Y = Y ol (s) = N (s) + P(s) D (s)

Find all relations where the signal transmissions are equal to Output with feedback control
either the sensitivity function or the complementary senstitivity 1   1
function Ycl = N + PL = Yol = SYol = SYol
1 + PC 1 + PC

The Audience is Thinking ... Disturbances with frequencies such that  S (iω ) < 1 are
reduced by feedback, disturbances with frequencies such that
 S (iω ) > 1 are amplifed by feedback.

Assessment of Disturbance Reduction Assessment of Disturbance Reduction

We have 1
Ycl 1
= =S
Ycl ( s) 1 Yol 1 + PC
= S( s) = 0.5
Ycl ( s) 1 + P( s) C( s)

Feedback attenuates disturbances of frequencies such that Geometric inter- 0

 S (iω ) < 1. It amplifies disturbances of frequencies such that pretation: Dis-

 S (iω ) > 1 turbances with

Im
−0.5

1
frequencies in-
10
side the circle −1

0
are amplified by
10
feedback. Dis- −1.5

−1
10
turbances with
frequencies out- −2
−2 −1.5 −1 −0.5 0 0.5 1
Re
−2
10
side are reduced.
−2 −1 0 1
10 10 10 10


c K. J. Åström October, 2000 6
 Limitations Robustness
Can the sensitivity be small for all frequencies? (No we have Small process changes T = PC/(1 + PC)
S (∞) = 1)! Can we get  S (iω ) ≤ 1?
dT dP CdP 1 dP dP
Yes if the Nyquist curve of the loop = − = =S
T P 1 + PC 1 + PC P P
transfer function L = PC is in the first
and third quadrant! Passive systems! 1
1

Otherwise we have 0.5

0.5
Large process changes

∞  0

log  S( iω )dω = π
0
Re pk  C ∆ P < 1 + PC
0

Im
−0.5

Im
−0.5

−1

where pk are poles of PC in the right or

∆ P
−1

half plane. The sensitivity can be

−1.5
1
<
decreased at one frequency at the −2
−2 −1.5 −1 −0.5
Re
0 0.5 1  P T  −1.5

cost of increase at another frequency. −2

−2 −1.5 −1 −0.5 0 0.5 1
Re

(The "water-bed effect"!)

Maximum Sensitivity Maximum Sensitivities

The number 1
Requirement on maximum sensitivities give constraints that tell
that the Nyquist curve should avoid certain circles
Ms = max  S( iω ) 0.5

Ms=2 and Mt=2 Ms=1.4 and Mt=1.4

0
is a measure of robustness. 3 3

The number 1/ Ms is the

Im

−0.5
2 2
smallest distance from the −1

Nyquist curve to the critical 1 1

point -1. −1.5

0 0
Reasonable values are in −2
−2 −1.5 −1 −0.5 0 0.5 1

the range of 1.2 to 2.

Re

−1 −1
A maximal sensitivity Ms guarantees a gain margin g m ≥
Ms /( Ms − 1) and a phase margin greater than ϕ m ≥ −2 −2

arcsin 1/ Ms . Constraints on both gain and phase margins can

−3 −3
be replaced by constraints on M s . −4 −2 0 2 −4 −2 0 2


c K. J. Åström October, 2000 7
 Summary of the Sensitivity Functions Summary

1 L
• Systems with two degrees of freedom allow a complete
S= , T = , Ms = max S (iω ), Mt = max T (iω ) separation of responses to reference signals and distur-
1+ L 1+L
bances.
The value 1/ Ms is the shortest distance from the Nyquist curve of
the loop transfer function L(iω ) to the critical point −1. • Design for disturbances and robustness first, then choose
feedforward F to give desired response to reference
 log T Ycl (s) ∆ P 1
S= = , < signals
 log P Yol (s)  P T 

Bode’s integral: sL(s) goes to zero as s → ∞ • A system with error feedback is characterized by four



transfer functions (Gang of Four)
∞ ∞
1 
log S (iω ) dω = log   dω = π pi • The basic feedback loop with two degrees of freedom is
0 0 1 + L ( iω )

∞
∞  1
L(1/ iω ) characterized by six transfer fnctions (Gang of Six)
log T (1/ iω ) dω = log  dω = π
0 0 1 + L(1/ iω ) zi
• The effect of feedback on disturbances is given by
The water bed effect.
Yclosedloop( s) = S( s) Yopenloop( s)

The Gang of Six

Responses of y and u to reference signal r is characterized by
PC F CF
,
1 + PC 1 + PC

Response of y to load disturbance d is characterized by

P
1 + PC

Response of u to measurement noise n is characterized by

C
1 + PC

Robustness to process variations is characterized by

1 PC
S= , T=
1 + PC 1 + PC


c K. J. Åström October, 2000 8