Systems Analysis and Control

Matthew M. Peet
Illinois Institute of Technology

Lecture 11: Different Types of Control

Overview

In this Lecture, you will learn:

Limits of Proportional Feedback

• Performance Specifications.

Derivative Feedback
• Pros and Cons
• PD Control
• Pole Placement

More on Steady-State Error

• Response to ramps and parabolae
• Limits of PD control

Integral Feedback
• Elimination of steady-state error
• Pole-Placement

M. Peet Lecture 11: Control Systems 2 / 32

Recall the Inverted Pendulum Problem
Transfer Function 1
Ĝ(s) = M gl
Js2 − 2
For a simple proportional gain: K̂(s) = k
Closed Loop Transfer Function:
k
M gl
Js2 − 2 +k
Impulse Response Impulse Response
There are two cases:
2.5 18
x 10
6

2 16

1.5
14
1
12
Amplitude

Amplitude
0.5
10
0
8
−0.5
6
−1
4
−1.5

−2 2

−2.5 0
0 2 4 6 8 10 12 0 5 10 15 20 25
Time (sec) Time (sec)

M gl M gl
Figure: Case 1: k > 2
Figure: Case 2: k < 2

Both cases are unstable!

M. Peet Lecture 11: Control Systems 3 / 32
Differential Control
Now suppose we furthermore have a performance specification:
• Overshoot
• Rise Time
• Settling Time -
TDs G(s)
u(s) + y(s)

Problem: There is no solution using proportional gain: K̂(s) = k.

Now we must consider a New Kind of Controller:

Derivative Control: Choose K̂(s) = TD s
The controller is of the form

u(t) = TD ė(t)

The controller is called Differential/Derivative Control because it is

proportional to the rate of change of the error.
M. Peet Lecture 11: Control Systems 4 / 32
Differential Control
Differential control improves performance by reacting quickly.

Prediction:
• To measure ẏ(t), recall the definition of derivative:

e(t + ∆t) − e(t)

ẏ(t) ∼
=
∆t
• The ẏ(t) term depends on both the current position and predicted position.

I A way to speed up the response.

M. Peet Lecture 11: Control Systems 5 / 32

Differential Control: Using the Delay is Dangerous!

Problem: Differential control is implemented using delay.

• y(t) is the measurement.
• ẏ(t) cannot be measured directly
I Approximate using the delayed response:

e(t) − e(t − ∆t)

ẏ(t) ∼
=
∆t

I Delay can cause instabilities.

M. Peet Lecture 11: Control Systems 6 / 32

Differential Control: Noise is Dangerous!

Noise Amplification:
• Measurement of ẏ(t) is heavily influenced by noise.
e(t) − e(t − ∆t)
ẏ(t) ∼
=
∆t
• Sensor measurements have error
• As ∆t → 0, the effect of noise, σ is amplified:
e(t) − e(t − ∆t) 2σ
ẏ(t) ∼
= + →∞
∆t ∆t
M. Peet Lecture 11: Control Systems 7 / 32
Failure of Derivative Control
Inverted Pendulum

Controller: K̂(s) = TD s
Closed Loop Transfer Function:

TD /Js
M gl
s2 + TD /Js − 2J

2nd-Order System As we learned last lecture, stable iff both

• TD /J > 0
• −M gl
2J > 0

Derivative Feedback Alone cannot stabilize a system.

M. Peet Lecture 11: Control Systems 8 / 32

PD Control

Differential Control is usually combined with proportional control.

• To improve stability
• To reduce steady-state error.
• To reduce the effect of noise.
Controller: The form of control is
u(t) = K [e(t) + TD ė(t)]
or
û(s) = K [1 + TD s] ê(s)
M. Peet Lecture 11: Control Systems 9 / 32
PD Control
2nd-order system

Lets look at the effect of PD control on a 2nd-order system:

1
Ĝ(s) =
s2 + bs + c

Controller: K̂(s) = K [1 + TD s]
Closed Loop Transfer Function:

K̂(s)Ĝ(s) K [1 + TD s]
=
1 + K̂(s)Ĝ(s) s2 + bs + c + K [1 + TD s]
K [1 + TD s]
=
s2 + (b + KTD )s + (c + K)

The poles of the system are freely assignable for a 2nd order system.
• TD and K allow us to construct any denominator we desire.

M. Peet Lecture 11: Control Systems 10 / 32

PD Control
2nd-order system

Suppose we want poles at s = p1 , p2 .

Im(s)

Re(s)

• We want the closed loop of the form:

1 1
= 2
(s − p1 )(s − p2 ) (s − (p1 + p2 )s + p1 p2 )
Thus we want
• c + K = p1 p2 which means K = p1 p2 − c.
• b + KTD = −(p1 + p2 ) which means TD = − p1 +p
K
2 +b
= − pp11+p2 +b
p2 −c
PD feedback gives Total Control over a 2nd-order system.
M. Peet Lecture 11: Control Systems 11 / 32
PD Control
Example:

Suppose we have the 2nd-order system Im(s)

1
Ĝ(s) =
s2 + s + 1
and performance specifications:
• Overshoot: Mp,desired = .05 Re(s)
• Rise Time: Tr,desired = 1s
• Settling Time: Ts,desired = 3.5s.

As we found in Lecture 9, these specifications mean that the poles satisfy:

σ < −.9535ω, σ < −1.333, ωn > 1.8

We chose the pole locations:

s = −1.5 ± 1.4ı
M. Peet Lecture 11: Control Systems 12 / 32
PD Control
Example:

The desired system is

1
(s2 − (p1 + p2 )s + p1 p2 )
The closed loop is
K [1 + TD s]
s2 + (b + KTD )s + (c + K)
To get the pole locations:
p1,2 = −1.5 ± 1.4ı
we choose
• The gain
K = p1 p2 − c = (−1.5 + 1.4ı)(−1.5 − 1.4ı) + 1 = 1.52 + 1.42 − 1 = 3.21
• The derivative gain
p1 + p2 + b −3 + 1 2
TD = − =− = = .623
K 3.21 3.21
This gives the controller:
K̂(s) = K(1 + TD s) = 3.21 + 2s
M. Peet Lecture 11: Control Systems 13 / 32
PD Control
Problem with Steady-State Error

6 Step Response Step Response

x 10
3 1

2.5
0.8

2
Amplitude

Amplitude
0.6
1.5
0.4
1

0.2
0.5

0 0
0 2 4 6 8 10 0 0.5 1 1.5 2 2.5 3 3.5
Time (sec) Time (sec)

Figure: Open Loop Figure: Closed Loop

Although the PD controller gives us control of the pole locations, the
steady-state value is
K 3.21
yss = = = .7625
c+K 4.21

M. Peet Lecture 11: Control Systems 14 / 32

PD Control
Inverted Pendulum

Lets look at the effect of PD control on

the inverted Pendulum:
1/J
Ĝ(s) =
s2 − M2Jgl

Controller: K [1 + TD s]

Closed Loop Transfer Function:

K̂(s)Ĝ(s) K/J [1 + TD s]
= M gl
1 + K̂(s)Ĝ(s) −s2 + K/J [1 + TD s]
2J
K/J [1 + TD s]
=
s2 + K/JTD s + (K/J − M2Jgl )

M. Peet Lecture 11: Control Systems 15 / 32

PD Control
Inverted Pendulum

To achieve the performance

specifications: Im(s)
• Overshoot: Mp,desired = .05
• Rise Time: Tr,desired = 1s
• Settling Time: Ts,desired = 3.5s.
We want poles at Re(s)

s = −1.5 ± 1.4ı

Thus we want
• c + K = p1 p2 which means K = p1 p2 − c.
• b + KTD = −(p1 + p2 ) which means

p1 + p2 + b p1 + p2 + b
TD = − =−
K p1 p2 − c

M. Peet Lecture 11: Control Systems 16 / 32

PD Control
Inverted Pendulum

The closed loop is

K/J [1 + TD s]
M gl
s2 + K/JTD s + (K/J − 2J )

To get the pole locations p1,2 = −1.5 ± 1.4ı

we choose
• The gain
M gl
K/J = p1 p2 − c = 4.21 +
2J
• The derivative gain
p1 + p2 + b 3
TD = − = M gl
p1 p2 − c 4.21 + 2J
This gives the controller: !
1 3
K̂(s) = K(1 + TD s) = 4.21J + M gl 1 + M gl
s
2 4.21 + 2J

M. Peet Lecture 11: Control Systems 17 / 32

PD Control
Inverted Pendulum: Problem with Steady-State Error

Step Response
1.4

1.2

1
Amplitude
0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3 3.5
Time (sec)

The steady-state error with this controller is (K = J = M = g = l = 1)

K/J 4.21
yss = M gl
= = 1.135
(K/J − 2J ) 4.21 − .5

Derivative Control has No Effect on the steady-state error!

M. Peet Lecture 11: Control Systems 18 / 32
Recall: Steady-State Error

Lets take another look at steady-state error

yHtL
Recall:
0.7
• We measured steady-state error
0.6 using the step response.
0.5
I ess = 1 − limt→∞ y(t)
0.4
Sometimes this doesn’t work.
5 10 15 20 25 30
t
• Assumes objective doesn’t move.
Figure: Suspension Response for k = 1
Problems:
• If target is moving, we may never
catch up.
• Even if we can catch a moving
target, we may not catch an
accelerating target.
For these problems, the step response is
not appropriate.

M. Peet Lecture 11: Control Systems 19 / 32

Ramp and Parabolic Inputs
There are other types of response we can consider.

• Ramp response tracks error for a target with constant velocity.

• Parabolic response tracks error for a target with a constant acceleration.
M. Peet Lecture 11: Control Systems 20 / 32
Ramp and Parabolic Inputs
We can use the final value theorem to find the response to ramp and parabolic
inputs:

Ramp Response:
Recall the ramp input:
1
u(t) = t û(s) =
s2
The steady-state error to a ramp input is

1 − Ĝ(s)
ess = lim sê(s) = lim s(1 − Ĝ(s))û(s) = lim
s→0 s→0 s→0 s

M. Peet Lecture 11: Control Systems 21 / 32

Ramp and Parabolic Inputs
We can use the final value theorem to find the response to parabolic inputs:

Parabolic Response:
Recall the parabolic input:
1
u(t) = t2 û(s) =
s3
The steady-state response to a parabolic input is

Ĝ(s)
lim sŷ(s) = sĜ(s)û(s) =
s→0 s2

Note: The steady-state error to a parabolic input is usually infinite.

M. Peet Lecture 11: Control Systems 22 / 32
Ramp and Parabolic Inputs
The effect of the numerator

For steady-state error, the numerator of the transfer function becomes

important: for
n(s)
Ĝ(s) =
d(s)
Steady state error is
 
d(s) n(s)
lim (1 − Ĝ(s))sû(s) = lim − sû(s)
s→0 s→0 d(s) d(s)
d(s) − n(s)
= lim sû(s)
s→0 d(s)

û(s) is the test signal

• Step Input: sû(s) = 1
• Ramp Input: sû(s) = 1
s
• Parabolic Input: sû(s) = 1
s2

M. Peet Lecture 11: Control Systems 23 / 32

Ramp and Parabolic Inputs
Systems in Feedback

ĜK̂
When in feedback, the closed loop has the form
1 + ĜK̂
Hence steady-state error has the form
!  
ĜK̂ 1
ê(s) = 1 − sû(s) = sû(s)
1 + ĜK̂ 1 + ĜK̂
Step Response:
1
ess,step = lim
s→0 1 + Ĝ(s)K̂(s)
Ramp Response:
1 1 1
ess,ramp = lim = lim
s→0 1 + Ĝ(s)K̂(s) s s→0 sĜ(s)K̂(s)
Parabolic Response:
1 1 1
ess,parabola = lim 2
=
s→0 1 + Ĝ(s)K̂(s) s 2
s Ĝ(s)K̂(s)
M. Peet Lecture 11: Control Systems 24 / 32
Example of Ramp Response
Consider the Suspension Example: Open Loop:
s2 + s + 1
Ĝ(s) =
s4 + 2s3 + 3s2 + s + 1
s + 2s + 3s + s + 1 − s2 − s − 1
4 3 2
s4 + 2s3 + 2s2
1 − Ĝ(s) = =
s4 + 2s3 + 3s2 + s + 1 s4 + 2s3 + 3s2 + s + 1
Ramp Response:
1 − Ĝ(s) s3 + 2s2 + 2s
lim = lim 4 =0
s→0 s s→0 s + 2s3 + 3s2 + s + 1

What happens when we close the loop?

Closed Loop Transfer Function:
k(s2 + s + 1)
s4 + 2s3 + (3 + k)s2 + (1 + k)s + (1 + k)
Ramp Response:
1 s4 + 2s3 + 3s2 + s + 1 1
ess,ramp = lim = lim =∞
s→0 sĜ(s)K̂(s) s→0 k(s2 + s + 1) s
Proportional response isn’t capable of controlling a ramp input
M. Peet Lecture 11: Control Systems 25 / 32
Example of Ramp Response

The only way to control a ramp input using feedback is to put a pole at the
origin:
Controller: K̂(s) = T1I s
Ramp Response:

1 s4 + 2s3 + 3s2 + s + 1 TI s
ess,ramp = lim = lim = TI
s→0 sĜ(s)K̂(s) s→0 s2 + s + 1 s

By including 1/s in the controller, the steady-state error becomes finite.

M. Peet Lecture 11: Control Systems 26 / 32

Integral Control

The purpose of integral control is primarily to eliminate steady-state error.

Controller: The form of control is
Z t
1
u(t) = e(θ)dθ
TI 0

or, in the Laplace transform

1
û(s) = ê(s)
TI s
One must be careful when using integral feedback
• By itself, an integrator is unstable.
I A pole at the origin.

M. Peet Lecture 11: Control Systems 27 / 32

Integral Control
Suspension Problem Again

Now lets re-examine the suspension problem

Controller: K̂(s) = T1I s
Closed Loop Transfer Function:

Ĝ(s)K̂(s) s2 + s + 1
=
1 + Ĝ(s)K̂(s) TI s5 + 2TI s4 + 3TI s3 + (TI + 1)s2 + (TI + 1)s + 1

If we set TI = .1, then the transfer function has poles at

• p1,2 = −.55 ± .89ı, p3 = −2.26, p4,5 = .6384 ± 1.877ı

Integral feedback can Destabilize the system where proportional feedback

couldn’t!

M. Peet Lecture 11: Control Systems 28 / 32

Integral Control

Integral Feedback is destabilizing!

Integral feedback is always combined with proportional or differential feedback:
PI Feedback: Proportional-Integral
 Z t 
1
u(t) = K e(t) + e(θ)dθ
TI 0
 
1
K̂(s) = K 1 +
TI s
PID Feedback:
Proportional-Integral-Differential
 Z t 
1
u(t) = K e(t) + e(θ)dθ + TD ė(t)
TI 0
 
1
K̂(s) = K 1 + + TD s
TI s

M. Peet Lecture 11: Control Systems 29 / 32

PID Control
Example

Finally, lets see the effect of PID control on a second-order system:

 
1 1
Ĝ(s) = 2 K̂(s) = K 1 + + TD s
s + bs + c TI s
Closed Loop:  
ĜK̂ K 1 + T1I s + TD s
=  
1 + ĜK̂ s2 + bs + c + K 1 + T1I s + TD s
 
K s + T1I + TD s2
=  
s3 + bs2 + cs + K s + T1I + TD s2
KTD s2 + Ks + K T1I
= K
s3 + (b + KTD )s2 + (c + K)s + TI
Steady-State Response:
K
TI
yss,step = K
=1 No Steady-State Error!
TI
M. Peet Lecture 11: Control Systems 30 / 32
PID Control

Pole Placement: The three pole locations can be determined exactly.

• Given three poles: p1 , p2 , p3 .
• Construct Desired denominator:
1 1
= 3 2
(s − p1 )(s − p2 )(s − p3 ) s + ad s + bd s + cd

Three equations:
• b + KTD = ad
• c + K = bd
• K
TI = cd
Which can be solved as
• K = bd − c
• TI = K
cd
• TD = adK−b

M. Peet Lecture 11: Control Systems 31 / 32

Summary
What have we learned today? In this Lecture, you learned:

Limits of Proportional Feedback

• Performance Specifications.

Derivative Feedback
• Pros and Cons
• PD Control
• Pole Placement

More on Steady-State Error

• Response to ramps and parabolae
• Limits of PD control

Integral Feedback
• Elimination of steady-state error
• Pole-Placement

Next Lecture: Midterm Review

M. Peet Lecture 11: Control Systems 32 / 32