Lecture 6 - SISO Loop Analysis

SISO = Single Input Single Output

Analysis:
Stability
Performance
Robustness

EE392m - Winter 2003

Control Engineering

6-1

ODE stability

x& = f ( x, t )
x2

Lyapunovs stability theory - nonlinear

systems

stability definition
first (direct) method

x1

exponential convergence

second method: Lyapunov function

generalization of energy dissipation

Lyapunovs exponent
dominant exponent of the
convergence
for a nonlinear system
for a linear system defined
by the poles
EE392m - Winter 2003

Control Engineering

x
Ae-ax
-Ae-ax
6-2

Stability: poles
x&
y

= Ax + Bu
= Cx + Du

y = H (s) u

1 Imag

H ( s ) = C (Is A ) B + D

Characteristic values = transfer

function poles
l.h.p. for continuous time
unit circle for sampled time

1
Real

Imag

I/O model vs. internal dynamics

N (s)
g1
gn
H (s) =
=
+K+
+ g0
D ( s ) s p1
s pn

EE392m - Winter 2003

Control Engineering

Real

6-3

Stability: closed loop

Plant : P ( s )

PID controller : C ( s )
1
s
e
kP + kI + kD
s
Ds + 1

y = P ( s )u
u = C ( s )e;

e = y yd

1 + P ( s )C ( s )] yd
e e = [1
442443
1

S (s)

yd
The transfer function poles are the zeros of
1 + P ( s )C ( s )
Watch for pole-zero cancellations!
Poles define the closed-loop dynamics (including stability)
Algebraic problem, easier than state space sim
EE392m - Winter 2003

Control Engineering

6-4

Stability
For linear system poles describe stability
almost, except the critical stability
For nonlinear systems
linearize around the equilibrium
might have to look at the stability theory - Lyapunov

Orbital stability:
trajectory converges to the desired
the state does not - the timing is off
spacecraft
FMS, aircraft arrival

EE392m - Winter 2003

Control Engineering

6-5

Performance
Need to describe and analyze
performance so that we can
design systems and tune
controllers
There are usually many
conflicting requirements
Engineers look for a
reasonable trade-off

Optimizer

kD , kP , kI

Performance

s
kI
u = k D
+ kP +
s
Ds + 1

Plant model
sim

EE392m - Winter 2003

Control Engineering

6-6

Performance: Example
Optimizer

Selecting optimal b in the

Watts governor - HW
Assignment 1

Performance

Plant model, given b

sim

damping b

2.15
2.1

Performance index
in a transient vs b

2.05
2
1.95
1.9
1.85
1.8
1.75

EE392m - Winter 2003

Control Engineering

2.5

3.5
DAMP ING b

4.5

6-7

Performance - poles
Steady state error: study transfer functions at s=0.
Step/pulse response convergence, dominant pole
a = min{Re p j }j =1
n

C + Ae at dominant exponent
0

Caution! Fast response (poles far to the left) leads to peaking

fast response
slow response
0
EE392m - Winter 2003

Control Engineering

6-8

Performance - step response

Step response shape characterization:
overshoot
undershoot
settling time
steady state error
rise time

0
EE392m - Winter 2003

Control Engineering

6-9

Performance - quadratic index

Quadratic performance

J=

response, in frequency domain

y (t ) yd (t ) dt =
2

t =0

1
2

1
2

~(i ) 2 d =
e

1
2
S (i ) ~
yd (i ) d =
2

S (i )

d
2

S ( s ) = [1 + P( s )C ( s )]

STEP

If yd(t) is a zero mean random process with the

spectral power Q (i)

1
2
J = E y (t ) yd (t ) dt =
t =0
2
EE392m - Winter 2003

S (i ) Q (i )d

Control Engineering

6-10

Transfer functions in control loop

disturbance
feedforward
noise

d
v

Plant
P( s)

n
Controller
C ( s)

reference

output

control

error

yd

e = S ( s )d S ( s ) yd + T ( s )n + S y ( s ) v
y = S ( s )d + T ( s ) yd + T ( s ) n + S y ( s )v
u = Su ( s ) d + Su ( s ) yd + Su ( s )n + T ( s )v
EE392m - Winter 2003

Control Engineering

6-11

Transfer functions in control loop

e = y yd + n

e = S ( s )d S ( s ) yd + T ( s ) n + S y ( s )v

y = P( s )( u + v ) + d
u = C ( s ) e

y = S ( s ) d + T ( s ) yd + T ( s )n + S y ( s )v
u = Su ( s ) d + Su ( s ) y d + Su ( s ) n + T ( s )v

Sensitivity S ( s ) = [1 + P( s )C ( s )]1

Complementary sensitivity T ( s ) = [1 + P( s )C ( s )]1 P( s )C ( s )

Noise sensitivity Su ( s ) = [1 + P( s )C ( s )]1 C ( s )

Load sensitivity S y ( s ) = [1 + P( s )C ( s )]1 P( s )

EE392m - Winter 2003

Control Engineering

6-12

Sensitivities
d
disturbance

disturbance d

Plant
P(s)

Controller
C ( s)
reference
yd

y
output

Feedforwa rd
F ( s)

e
error

yd

Plant
P(s)

y
output
e
error

reference

y = d + F ( s ) P ( s ) yd

y = S ( s )d + T ( s ) yd
1
,
L( s ) = P( s )C ( s )
1 + L(i )
Feedback sensitivity

S (i ) =

S FF (i ) = 1

Feedforward sensitivity

good for any frequency

|S (i)|<<1 for |L (i) |>>1
never unstable
|S (i)|1 for |L (i) |<<1
can be bad for |L (i) | 1 - ringing, instability
EE392m - Winter 2003

Control Engineering

6-13

Sensitivity requirements
e = S ( s ) d S ( s ) yd + T ( s ) n + S y ( s ) v
y = S ( s )d + T ( s ) yd + T ( s ) n + S y ( s ) v
u = S u ( s ) d + Su ( s ) y d + S u ( s ) n + T ( s ) v

S ( i ) =

1 + P(i )C (i )
P(i )
S y (i ) =
1 + P(i )C (i )
C (i )
Su (i ) =
1 + P(i )C (i )

Disturbance rejection and reference tracking

|S (i)|<<1 for the disturbance d ; |Sy(i)|<<1 for the input noise v

Limited control effort

|Su(i)|<<1 conflicts with disturbance rejection where |P(i)|<1

Noise rejection
|T (i)|<<1 for the noise n, conflicts with disturbance rejection

EE392m - Winter 2003

Control Engineering

6-14

Robustness
Ok, we have a controller that works for a nominal model.
Why would it ever would work for real system?
Will know for sure only when we try - V&V - similar to debugging
process in software

Can check that controller works for a range of different

models and hope that the real system is covered by this range

This is called robustness analysis, robust design

Was an implicit part of the classical control design - Nyquist, Bode
Multivariable robust control - Honeywell: G.Stein, G.Hartmann, 81
Doyle, Zames, Glover - robust control theory

EE392m - Winter 2003

Control Engineering

6-15

Control loop analysis

u(t)

Plant

y(t)

Feedback
controller

y = gu
u& = k I ( y yd )

y& + gk I ( y yd ) = 0

Why control might work if the process differs from the model?
Key factors
Uncertainty

modeling error (uncertainty) characterization

time scale (bandwidth) of the control loop
Step response for
the design model:

y(t)=gu(t)
Modeling error
EE392m - Winter 2003

Actual step response

u(t)

Plant

y(t)

Feedback
controller

0
Control Engineering

6-16

Robustness - Small gain theorem

Nonlinear uncertainty!

u(t)

Operator gain
Gu G u

y(t)

The loop is guaranteed stable if

G can be a nonlinear operator

G <1

L2 norm
1
2
2
u = u (t )dt =
2

G.Zames

~(i ) 2 d
u

(Open-loop stability assumed)

Desoer and Vidyasagar, Feedback

Systems: Input-Output Properties, 1975

L2 gain of a linear operator

1
1 ~
2
2
2
2
~
Gu =
G (i )u (i ) d sup G (i )
u (i ) d

142
4 43
4 1
2
2442443
2

EE392m - Winter 2003

Control Engineering

6-17

Robustness
Additive uncertainty

Multiplicative uncertainty

u(t)

P(s)

y(t)

P(s)

u(t)

C(s)

Su(s)
Condition of robust stability
C (i )
(i ) < 1
1 + P(i )C (i ) 123
1442443

Su

EE392m - Winter 2003

y(t)

C(s)
T (s)

Condition of robust stability

P(i )C (i )
(i ) < 1
1 + P(i )C (i ) 123
1442443

Control Engineering

6-18

Nyquist stability criterion

=1

u(t)

G(s)

y(t)

Homotopy Proof
G(s) is stable, hence the loop is stable
for =0. Increase to 1. The instability
cannot occur unless G(iw)+1=0 for
some 0 1.
|G(i180)|<1 is a sufficient condition

Subtleties: r.h.p. poles and zeros

Formulation and real proof using the
agrument principle, encirclements of -1
stable unstable stable as 01
EE392m - Winter 2003

Control Engineering

Compare against
Small Gain Theorem:

6-19

Gain and phase margins

v

P (s )

C (s )

=
y

1
C (s )

P (s )

Im L(s)

1/gm

Loop gain
L( s ) = P( s )C ( s )
S ( s ) = [1 + L( s )]
Nyquist plot for L
at high frequency L(i ) 1
1

EE392m - Winter 2003

gain
-1
margin

Control Engineering

180
m
gc

Re L(s)

phase
margin

6-20

Gain and phase margins

Bode plots

phase
crossover

gain
crossover

gain
margin

Im L(s)

1/gm
-1

180
m
gc

gc

Re L(s)

phase
margin

180
EE392m - Winter 2003

Control Engineering

6-21

Advanced Control
Observable and controllable system
Can put poles anywhere
Can drive state anywhere

Why cannot we just do this?

Large control
Error peaking
Poor robustness, margins
Observability and controllability = matrix rank
Accuracy of solution is defined by condition number

Analysis of this lecture is valid for any LTI control,

including advanced
EE392m - Winter 2003

Control Engineering

6-22