You are on page 1of 31

ME 531 - Design of Control Systems

Basic Properties of Feedback


Control of Translational Motion of a UUV

bv
F
m
bv&

mv& + bv = F   ( ms + b)V ( s) = F ( s)
 ⇒ 
x& = v   sX ( s) = V ( s)

1
X ( s) 1 m
= = = G ( s)
F ( s) s( ms + b) s( s + b m)

Open-Loop Control:
w

r + x
+ u
K Σ G(s)

• Without feedback, the position of the vehicle is very difficult to control.

• If the vehicle is given a step input in r (force), what will happen to velocity? to position?

Step Response
UUV Open-Loop Control

1.8

1.6
(
Note: response is rounded due to s + b
m )
1.4
and integrates the step due to s.
1.2
Amplitude

0.8

0.6

0.4

0.2

0
0 10 20 30 40 50 60 70 80
Time (sec.)
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 2

Closed-Loop Control

+
r + e u + x
Σ D(s) Σ G(s)
-

For proportional control: D( s) = K p (the control effort, u, is linearly proportional to the


error, e)

Kp
m

s( s + )
Kp
X ( s) D( s)G ( s) b
m
= = =
m

R( s) 1 + D( s)G ( s) Kp
m s +
2 b
m s+
Kp
m
1+
s( s + b
m )

s2 + s+ =0
b Kp
Characteristic Equation: m m

b
Poles: s1,2 = − ± ( b 2 m) 2 − K p
m
2m

for ( b 2 m) 2 ≥ Kp
m , the roots are on the real axis

for ( b 2 m) 2 < Kp
m , the roots are complex.
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 3

Performance criteria: we want good transient response

⇒ M p < 10%, t r < 5 sec, t s < 10 sec

we want good disturbance rejection

⇒ ess = 0

Root Locus:

Locus of Roots for Proportional Control


0.2

0.15

0.1

0.05
Imag Axis

-0.05

-0.1

-0.15

-0.2
-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2
Real Axis
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 4

Closed-loop step response:

Closed-Loop UUV System Response

Kp = 0.2 Kp = 0.5
0.8 1

0.8
0.6
position (m)

position (m)
0.6
0.4
0.4

0.2
0.2

0 0
0 20 40 60 80 100 0 20 40 60 80 100
time (sec) time (sec)
Kp = 1 Kp = 2
1.5 1.5

1 1
position (m)

position (m)

0.5 0.5

0 0
0 20 40 60 80 100 0 20 40 60 80 100
time (sec) time (sec)
Kp = 5 Kp = 20
1.5 2

1.5
1
position (m)

position (m)

0.5
0.5

0 0
0 20 40 60 80 100 0 20 40 60 80 100
time (sec) time (sec)

• low gains result in slow response


• high gains result in high overshoot
• settling time remains the same for each case - this is because the real part of the complex
closed-loop poles is unchanging with increasing K p
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 5

Disturbance rejection:

Why?

1
X ( s) m 1
= , which has a DC gain of: .
W ( s) s2 + b
m s+
Kp
m Kp

• For the system to reject disturbances adequately, K p must be very large, but if K p is
large, then the transient response is bad.
• In this case we have competing objectives that we cannot satisfy with proportional control
alone.
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 6

Proportional/Derivative Control

• Previously, our closed-loop system had too much overshoot, too slow of a rise time, and
too long of a settling time.
• Basically, we need higher proportional gain (to speed up the response and decrease the
steady-state error) and more damping.

• Try a combination of proportional and derivative control:

de
D( s) = k p + kd s ⇔ u = k p e + kd
dt

The control effort in this case is the sum of a term that is proportional to the error
and a term that is proportional to the derivative of the error.

+
r + e u + x
Σ D(s) Σ G(s)
-

1
m
G ( s) =
s( s + b
m )
( k s+ k )
d p

( k s+ k )
m

s( s + ) d p
X DG b
m
= = =
m

R 1 + DG
1+
( k s+ k )
d p
m
s2 + ( )s +
b + kd
m
kp
m

s( s + b
m )
Characteristic Equation (PD Control):

b + kd kp
s2 + s+ =0
m m
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 7

Poles:
b + kd  b + kd 
2
kp
s1,2 =− ±   −
2m  2m  m

Zero:

kp
s= −
kd

Question: For a fixed value of proportional gain (say k p = 20 ), how does the value of k d affect
the location of the closed-loop poles?

Locus of Roots for PD Control, Kp = 20, Kd variable

0.5

0.4

0.3 0
25
0.2 50
75
0.1
Imag Axis

100 125
125
0
100
-0.1
75
-0.2 50
25 0
-0.3

-0.4

-0.5

-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
Real Axis
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 8

• Notice that higher values of Kd cause the roots to move toward the negative real axis.
–> increases damping, decreases M p
• Notice that both σ and ω n are increased.
–> faster t r and faster t s

Characteristic equation:

b k kp
s2 + s+ d s+ =0
m m m
kd  b kp 
s +  s2 + s +  = 0
m  m m
 1
s m 
1 + kd  2  =0
s + b
ms+
kb
m

Note: the form of this last equation will help us in the future when we learn how to
construct the root-locus diagram (it is called Evans’ form).

Closed-Loop step response:

• Note how the behavior changes with increasing k d :


< decreased overshoot
< faster rise time
< faster settling time
• This corresponds with the pole locations for changing k d .
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 9

Kp = 20, Kd = 0 Kp = 20, Kd = 50
2 1.4

1.2
1.5
1
position (m)

position (m)
0.8
1
0.6

0.4
0.5
0.2

0 0
0 20 40 60 0 20 40 60
time (sec) time (sec)
Kp = 20, Kd = 75 Kp = 20, Kd = 100
1.4 1.4

1.2 1.2

1 1
position (m)

position (m)

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 20 40 60 0 20 40 60
time (sec) time (sec)
Kp = 20, Kd = 125 Kp = 20, Kd = 150
1.4 1.4

1.2 1.2

1 1
position (m)

position (m)

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 20 40 60 0 20 40 60
time (sec) time (sec)
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 10

Response to a step disturbance:

Note: the zero steady-state error criteria is still not satisfied!

1
X ( s) m
= b + kd kp
W ( s) s2 + m s+ m

1
The DC gain is , which has been unchanged by the addition of derivative feedback.
kp

* PD control has greatly improved the transient response, but disturbance rejection is still
unacceptable.
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 11

Proportional/Integral/Derivative Control (PID)

Integral feedback has the following form:

u(t ) = k I ∫ e dτ
t

t0

U ( s) k
or: = D( s) = I
E ( s) s

Integral feedback can provide a finite value for the control even when the error is zero. This is
advantageous because it allows disturbances to be canceled with zero error.

• the error, e, no longer needs to be finite to produce a control u to counter a constant


disturbance w.
• control is a function of all past values of e, rather than the current value only.
• past errors charge up the integrator to the point that error are driven to zero (control
remains non-zero even though the error goes to zero).

For PID control:

de
+ k I ∫ e dτ
t
u = k pe + kd
dt t0

k
D( s) = k p + kd s + I
s
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 12

Root-Locus for PID control:

Locus of Roots for PID Control, Kp = 30, Kd = 200, Ki variable


1.5

0.5
Imag Axis

-0.5

-1

-1.5
-1.5 -1 -0.5 0 0.5 1 1.5
Real Axis
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 13

Closed-Loop step response for PID control:

Kp = 30, Kd = 200, Ki = 0.5 Kp = 30, Kd = 200, Ki = 1


100 200

80
150
position (m)

position (m)
60
100
40

50
20

0 0
0 10 20 30 40 0 10 20 30 40
time (sec) time (sec)
Kp = 30, Kd = 200, Ki = 2 Kp = 30, Kd = 200, Ki = 5
250 300

250
200
200
position (m)

position (m)

150
150
100
100
50
50

0 0
0 10 20 30 40 0 10 20 30 40
time (sec) time (sec)
Kp = 30, Kd = 200, Ki = 15 Kp = 30, Kd = 200, Ki = 50
400 800

600
300
400
position (m)

position (m)

200 200

0
100
-200

0 -400
0 10 20 30 40 0 10 20 30 40
time (sec) time (sec)
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 14

Disturbance response:

Equations of Motion:

mx&& + bx& = u + w
d
( r − x ) + k I ∫ ( r − x ) dτ + w
t
mx&& + bx& = k p (r − x ) + kd
dt t0

Clearing the integral:

&&& + bx&& + k d &&


mx x + k p x& + k I x = k d r&& + k p r& + k I r + w&

• The dynamics are only dependent on w& (not w).


• & = 0 and there will be no induced error.
For a constant w, w
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 15

kd s2 + k p s + k I
D( s) =
s
kd s2 + k p s + k I
X ( s)
=
( )
m

R( s) s2 ( s + b
m) +
kd s2 + k p s + k I
m

ms + m s+
kd 2 p kI k
X ( s) m
= 3 b kd 2 k p
R( s) s + ( m + m) s + m s + kI
m

Characteristic Equation:

s3 + ( b m + )s + s+ =0
kd 2 kp kI
m m m

k I  3  b kd  2 k p 
+ s + + s + s =0
m   m m m 
 1 
1+ kI  m
=0
 [
 s s 2 + ( b m + k d m) s + kp
m ] 

X
DC gain of = 1.
R

What about disturbance rejection?

s
X ( s) m
= = H ( s)
W ( s) s3 + ( b m + kd
m )s 2
+
kp
m s+
kI
m

1
xss = lim sX ( s) = lim sH ( s)W ( s) = lim s H ( s) = 0
s→ 0 s→ 0 s→ 0 s

therefore we have zero steady-state error!

Conclusion: we can satisfy all transient response and disturbance rejection criteria by using a PID
controller.
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 16

Integrator Advantages:

• Zero steady state error in response to constant reference inputs


• Zero steady state error in response to constant disturbances

Integrator Disadvantages:

• De-stabilizing influence on dynamic behavior


< increases overshoot
< increases settling time
• Integrator windup

Integrator Windup and Anti-Windup Techniques

What is windup and how is it caused?

• Windup is a condition where the integrator becomes over-charged inadvertently -


or wound-up.
• Windup occurs when an integrator integrates when it shouldn’t.
< Typically this happens when the actuator saturates.
< The actuator is trying as hard as it can, but is unable to track the reference
command. The integrator keeps adding up the error.
• The effect is best illustrated by example. Consider the UUV translation control:
• Assume that the thrusters saturate at 50 N.
• Give the vehicle a step input of 10 m.
• Tether snags causing the integrator to windup.
• Saturation occurs.
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 17
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 18
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 19

Integrator Wind-up Prevention:

• Turn off integrator when saturation of actuator occurs

Note: When uc exceeds umax (saturation), the feedback loop around the integrator is
activated which quickly drives e1 to zero - effectively turning off the integrator.

• Limit the output of the integrator

When the integrator output is bounded, its damaging effects on the dynamics are limited,
but its ability to overcome disturbances is also limited.

• Turn on the integrator only after transients have died out.

• When giving smooth trajectory commands, activate integrator wen desired


velocity = 0.
4.6
• Turn on when t > t s >
σ
• Turn on when the error is less than some ε , where the appropriate value of ε is
chosen by taking into account the DC gain.
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 20
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 21
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 22
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 23
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 24

Steady-State Tracking and System Type


Inputs:

Input Polynomial System Type


Step 1(t) 0th order
Ramp t 1st order
Parabolic t2 2nd

System type describes the order of the input polynomial going into a system for whihc the
tracking error will be constant.

Consider the following system:

r + x
+ e u +
Σ D(s) Σ G(s)
-
H(s)

Y ( s) DG
= = T ( s)
R( s) 1 + DGH
E ( s) = R( s) − Y ( s) = R( s) − T ( s) R ( s) = R ( s)[1 − T ( s)]
E ( s)
= 1 − T ( s)
R( s)
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 25

Consider the general input:

tk
r (t ) = 1(t )
k!

If: k = 0 ⇒ unit step


k = 1 ⇒ ramp with unit slope
k = 2 ⇒ parabola with unit 2nd derivative

Laplace transform of r (t ) :

1
R ( s) = k +1
s

Applying the Final Value Theorem to this general case:

ess = lim sE ( s)
s→ 0

ess = lim sR( s)[1 − T ( s)]


s→ 0

[1 − T ( s)]
1
ess = lim s k +1
s→ 0 s
ess = lim k [1 − T ( s)]
1
s→ 0 s

The result of evaluating this limit can be either zero, a non-zero constant, or infinite.

• The system type is the value of k for which the above ess is a non-zero constant.
• For example: if k = 0 and ess is a constant, then the system is type zero - which means
that it has a constant steady-state error to a step input.

If k = 1 , then the system is type I, and it has a constant ess for a ramp and zero
ess for a step input.
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 26

Example: Determine the system type for the following system

r + e A 1 x
Σ
(τs + 1) s
-

K
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 27

Example: Determine the system type for the following system

r + e  1  1 x
Σ K p  1 + TD s + 
 TI s  Js 2
-
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 28

Stability
Bounded Input - Bounded Output (BIBO) Stability:

A system is said to be BIBO stable if every bounded input results in a bounded output -
independent of what might be happening to some internal states of the system!

Definition: A system with impulse response h(t) is BIBO stable if and only if the integral

∫−∞
h(τ ) dτ < ∞

Asymptotic Internal Stability:

A system is asymptotically internally stable if the output and all internal variables never become
unbounded and go to zero as time goes to infinity for sufficiently small initial conditions.

What does this imply?

• Consider the LCC system with the characteristic equation:


sn + a1sn −1 + a2 sn − 2 + K+ an = 0
With characteristic roots pi such that:

sn + a1sn −1 + a2 sn − 2 + K+ an = ( s − p1 )( s − p2 )K( s − pn ) = 0

The solution to the differential equation whose characteristic equation is given above is:
n
y (t ) = ∑Ke
i =1
i
pi t
where pi are the roots, and Ki values depend on IC’s

The system is stable if and only if every term goes to zero as t → ∞ .


e pi t → 0 for all pi

This will happen only when all the poles are strictly in the left half plane (LHP) where
Re{ pi } < 0

In other words, a system is stable if the roots of its characteristic equation lie in the LHP.

This can easily be determined with MATLAB or with any polynomial root solver.
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 29

Routh’s Stability Criterion

• This was especially useful before computer-aided analysis was available.


• It is still useful for determining ranges of CE coefficients for stability.
• It allows us to make statements about a system’s stability without solving for the roots.

Consider the CE for a system of order n:

a ( s) = sn + a1s n −1 + a2 sn − 2 + K+ an −1s + an = 0

• { }
A necessary condition for stability is that all of the coefficients ai must be positive (if

{a } is negative or zero, CE will have roots in the RHP)


i

Q: If all of the coefficients are positive, is the system stable? - not necessarily!

• A stronger test is needed. The following test was proposed by Routh in 1879:

{ }
Calculate a triangular array based on ai . A necessary and sufficient condition for
stability is that all of the elements in the first column of the Routh array be positive.

How to construct the Routh array: (see pp 215-216 in text)

The first row contains all the even numbered coefficients and the second row contains all
the odd numbered coefficients of the CE.

Routh Array:
n sn : 1 a2 a4 K
n − 1 sn −1: a1 a3 a5 K
n− 2
n− 2 s : b1 b2 b3 K
n − 3 sn − 3: c1 c2 c3 K
M M M M M M
2 s2 ∗ ∗
1 s ∗ 0
0 s0 ∗
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 30

where:
1 a2 1 a4
a1 a3 a1 a5
b1 = b2 =
a1 a1
a1 a3 a1 a5
b1 b2 b1 b3
c1 = c2 =
b1 b1
For the Routh array, the number of sign changes in the first column is equal to the number of
RHP poles.

Example:

a ( s) = s5 + 2 s4 + 4 s3 + 3s 2 + 2 s + 2 = 0
ME531 - Design of Control Systems
Basic Properties of Feedback
Page 31

• See pp 217-219 in text for a discussion of special cases for Routh’s test:

Special Case I: zero in 1st column


Special Case II: row of zeros

• Look over example 4.20 and 4.21: how to use Routh’s test to determine stability versus
parameter range.

You might also like