Lecture 18 - Summary of the Course Preparations for Final Examination

K. J. Åström 1. Dynamics
• Most problems require knowledge about dynamical systems
1. Dynamics • Problem 6 - Physical modeling and observability
2. PID Control • Problem 5 - Linearization
3. The Simple Feedback Loop • Problem 3 - Stability
2. PID Control
4. Feedforward
• Problem 1, Problem 2, Problem 3
5. State Feedback and Observers
3. The Simple Feedback Loop
6. Summary
• Problem 4, Problem 1, Problem 2
Theme: Reviewing what we learned and filling some gaps. 4. Feedforward
• Problem 10, Problem 1, Problem 2
5. State Feedback and Observers
• Problem 7, Problem 8, Problem 9

Homework Problems and Midterm 1. Dynamics

1. Dynamics 1. State models
• Homework 1, Homework 2, Homework 3, Midterm • Linearization
2. PID Control • Linear time invariant systems
• Homework 2, Midterm 2. Input-output models
3. The Simple Feedback Loop
• Transient response, step response, impulse response
• Homework 2, Homework 4, Homework 5, Midterm • Transfer function, poles, zeros
4. Feedforward • Frequency response, Bode and Nyquist plots
• Homework 6, Midterm
3. Controllability, observability, system structure
5. State Feedback and Observers
4. Relations between representations
• Homework 6, Homework 7
• Coordinate changes, standard forms
5. Stability
6. Modeling (from physics and from data)


c K. J. Åström December, 2000 1
 State Models Linearization
• State a collection of variables that summarize the past for dx
the purpose of predicting the future, e.g. = f ( x, u), y = g( x, u)
dt
dx The stationary solution x 0 , u0 is obtained by solving the
= f ( x, u)
dt equation f ( x0 , u0 ) = 0. For small deviations x = x0 + δ x,
y = g( x, u)
u = u0 + δ u, y = y0 + δ x around the equilibrium we have
x state, u control signal, y measured signal dδ x f f
= Aδ x + B δ u A= ( x0 , u0 ) B= ( x0 , u0 )
• The states describe storage of mass energy and momen- dt x u
g g
tum. Constitutive equations are also needed to obtain the δ y = Cδ x + Dδ u C= ( x0 , u0 ) D= ( x0 , u0 )
x u
model.
• Key problem: How accurately do we need to describe the Change notation: δ x → x, δ u → u, δ y → y
storage? dx
= Ax + Bu
• Synonyms: internal descriptions, white boxes dt
y = Cx + Du

Linear Time Invariant Systems Input-Output Models

dx
= Ax + Bu
Input Output
dt System
y = Cx + Du

t
x( t) = e x(0) +
At
e A( t− s) Bu( s) ds
0

t • Table over all input-output pairs
y( t) = Ce x(0) + C
At
e A( t− s) Bu( s) ds + Du( t) • Superposition: if (u 1, y1) and (u2 , y2) are i-o pairs so is
0
(au1 + bu2, a y1 + b y2)
Think in scalars interpret as vectors and matrices! • Table can be simplified drastically if initial conditions
If the matrix A can be diagonalized A = T Λ T −1 we have neglected
t
eAt = TeΛt T −1 . The elements of e At is a sum of terms eλ i t . In – Impulse response y(t) = 0 g(t − s)u(s)ds
general when the
 matrix has multiple eigenvalues the elements – Transfer function Y (s) = G (s) U (s)
λit
have the form ki Pi (t) e where λ i is an eigenvalue and P(t)
– Frequency response
is a polynomial


c K. J. Åström December, 2000 2
 Laplace Transforms Some Properties of the Laplace Transform
Consider a function f defined on 0 ≤ t < ∞ and a real number
∞
σ > 0. Assume that f grows slower than eσ t for large t. The L f = F ( s) = e− st f ( t) dt
Laplace transform F = L f of f is defined as 0

∞ Transform of the derivative

L f = F ( s) = e− st f ( t) dt
∞ ∞
∞
0 df 
L = e− st f ( t) dt = e− st f ( t) + s e− st f ( t) dt = − f (0) + sL f
dt 0 0 0
Example 1:

∞ ∞ 1 Linearity: L (a f + bg) = aL f + bL g
1 
f ( t) = 1 , F ( s) = e− stdt = − e− st =
0 s 0 s Differentiation: L df
dt
= sL f − f (0)
t
Example 2: Integration: L 0 f (τ )dτ = 1s L f
t

∞
1 − st∞ 1
Convolution: L 0 f (t − τ )g(τ )dτ = F (s) G (s)
f ( t) = e− at, F ( s) = e−( s+a) tdt = − e  =
0 s+a 0 s+a Final value Theorem: lim s→0 sF (s) = lim t→∞ f (t)
Initial value Theorem: lim s→∞ sF (s) = lim t→0 f (t)

Poles and Zeros Shape of eλ t

The transfer function of an LTI system is a rational function Real λ
λ <0 λ >0
b1 sn−1 + b2 sn−2 + . . . + bn
10

−1 B ( s) 1
G ( s) = C( sI − A) B= = n 8
A( s ) s + a1 sn−1 + a2 sn−2 + . . . + an 0.8
6
0.6

y
The denominator polynomial A(s) = det(sI − A) is the
4
0.4

0.2 2

characteristic polynomial of the matrix A. (Notice a slight 0 0

λt λt
0 1 2 3 4 0 0.5 1 1.5 2
notational conflict!)
Poles are the zeros of the polynomial A(s), they are also the Complex λ = σ + iω
eigenvalues of the matrix A. A pole λ means that the response σ = −0.25ω σ = 0.25ω
of the system has the component e λ t .
40
0.6
30

0.4 20

Zeros are the zeros of the polynomial B (s), they depend on

y

y
0.2 10

all matrices in the state description and reflect how the states 0 0

are coupled to the inputs and outputs. A zero λ means that

−0.2 −10

−0.4 −20

transmission of the signal e λ t is blocked. ωt ωt

0 5 10 15 0 5 10 15

c K. J. Åström December, 2000 3
 Frequency Response Frequency Response ...
0.1 The steady state transmission of sinusoids is given by G (iω )
0 •  G (iω ) gain, amplitude ratio
0 5 10 15
• arg G (iω ), phase
1

Graphical representations of G (iω )

−1

0 5 10 15 • Nyquist plot
An LTI system is completely characterized by its steady state • Bode plots
response to sinusoidal signals.
u( t) = u0 sin ω t
 
y( t) = u0  G ( iω ) sin ω t + arg G ( iω )

The Nyquist Curve Bode Plot

Represent the complex number G (iω ) as a point in the com- Represent the function G (iω ) as two curves, one for the
plex plane. The Nyquist curve is the locus of the point as fre- magnitude  G (iω ) (the gain curve or amplitude ratio) and one
quency ω changes. for the phase arg G (iω ) (the phase curve). Use logarithmic
Im G(iω ) scales for frequency and gain and linear scales for the phase.
• Interpretation Ultimate point • Interpretation 10
0
Gain

• Critical point
−1

• Asymptotes
10

−1 ϕ Re G(iω ) −2

• Positive feedback
10

a
• Minimum phase 10
−3
−1 0 1 2 3

• Negative feedback
10 10 10 10 10
ω
• Relation between curves
w

Phase
0

• Stability −50

−100

π d log  G ( iω )
arg G ( iω )  −150

2 d log ω 10
−1 0
10 10
w
1 2
10
3
10

c K. J. Åström December, 2000 4
 Non-minimum Phase Controllability and Observability
Right half plane zero
a−s dx
= Ax + Bu
Transfer function G (s) = ,  G (iω ) = 1 and dt
a+s y = C x + Du
ω
arg G (iω ) = −2 arctan
a The system can be steered from one point to another if the control-
lability matrix Wc has full rank
Time delay  
Transfer function G (s) = e−sT ,  G (iω ) = 1 and Wc =  B AB ... An−1 B 

arg G (iω ) = −ω T The state can be computed from the output if the observability
matrix Wo has full rank
Right half plane pole 
C 


s+b 

 CA 


Transfer function G (s) = ,  G (iω ) = 1 and 





s−b Wo = 


 .
.





 . 


 

b ω n −1
arg G (iω ) = −2 arctan = −π + 2 arctan CA
ω b

System Structure Relations

An LTI system can be split into four subsystems. The transfer dx
function is uniquely given by the subsystem S co = Ax + Bu
dt
y = Cx + Du
• Sco controllable and u y
observable Soc Σ Input-output relation
• Sc ō controllable but not
t
observable y( t) = Ce x(0) + Du( t) + C
At
e A( t−τ ) Bu(τ ) dτ
-
Soc Soc- 0
• Sc̄o not controllable but
t
observable = Ce x(0) +
At
g( t − τ )u(τ ) dτ
0
• Sc̄ō neither controllable --
Soc Impulse response
nor observable
g( t) = Dδ ( t) + CeAt B

c K. J. Åström December, 2000 5
 Relations ... Coordinate Changes

dx dx
= Ax + Bu = Ax + Bu
dt dt
y = Cx + Du y = Cx + Du

Transfer function Change coordinates by the transformation z = T x, x = T −1 z

b0 sn + b1 sn−1 + b2 sn−2 + . . . + bn
G ( s) = D + C[sI − A]−1 B = n dz
= T AT −1 z + T Bu = Ãz + B̃u
s + a1 sn−1 + a2 sn−2 + . . . + an

∞ dt
= e− stg( t) dt y = CT −1 z + Du = C̃z + Du
0

Same form but different matrices. Some relations

or equivalently
g( t) = Dδ ( t) + CeAt B = Dδ ( t) + C̃e Ãt B̃
dn y d n− 1 d nu d n− 1 u
n
+ a1 n−1 + . . . + an y = b0 n + b1 n−1 + . . . + bn u G ( s) = D + C( sI − A)−1 B = D + C̃( sI − Ã)−1 B̃
dt dt dt dt
W̃c = T Wc , W̃o = Wo T −1 X S

The Diagonal Form Controllable Canonical Form

If the matrix A can be diagonalized we have If the controllability matrix Wc has full rank we have
       
− a1 − a2 . . . − an−1 − an  1
 λ1 0   β1  

 



 


 
 
 
 

 1 0 0 0 




 0

 λ2   β2     
dz  









 dz 
=

 0 1 0 0







z+ 0




= 
 z+ . 
 u 
 
 
 

u
dt 
 . . 
 
 . 

dt 

 . . 




 .
. 


 . 
 
 . 
 
 . 
 
 .
    
 
 
  
0 λn βn 0 0 1 0 0
   
y =  γ 1 γ 2 . . . γ n  z + Du y =  b1 b2 . . . bn−1 bn  z + Du

Transfer function
The transfer function is
b1 sn−1 + b2 sn−2 + . . . + bn
G ( s) = D +
β 1γ 1 β 2γ 2 β nγ n sn + a1 sn−1 + a2 sn−2 + . . . + an
G ( s) = D + + + ... +
s − λ1 s − λ2 s − λn
The characteristic polynomial is
The mode zi is observable if γ i = 0 and controllable if β i = 0 sn + a1 sn−1 + a2 sn−2 + . . . + an

c K. J. Åström December, 2000 6
 Observable Canonical Form From Transfer Functions to State Equations
   
 −a1 1 0 ... 0  b1  Y ( s) 1 1

 
 
 
 = = 4
   b2 
 −a2
 0 1 0 
 
 U ( s) ( s + 1) 4 s + 4s3 + 6s2 + 4s + 1
dz 
 .





 .



=

 .. 

 z + 

 .
.


 u
dt 
 
 
 
 Hence

   

 − a n− 1 0 0 1


 
 

   
−an 0 0 0 bn ( s4 + 4s3 + 6s2 + 4s + 1) Y ( s) = U ( s)
 

y = 1 0 0... 0  z + Du
Switch to time domain
Transfer function d4 y d3 y d2 y dy
4
+ 4 3
+ 6 2
+4 +y=u
dt dt dt dt
b1 sn−1 + b2 sn−2 + . . . + bn
G ( s) = D +
sn + a1 sn−1 + a2 sn−2 + . . . + an

Characteristic polynomial

sn + a1 sn−1 + a2 sn−2 + . . . + an

From Transfer Functions to State Equations ... From Transfer Functions to State Equations

d4 y d3 y d2 y dy Y ( s) b1 s2 + b2 s + b3
+ 4 + 6 +4 + y=u = 3
dt4 dt3 dt2 dt U ( s) s + a1 s2 + a2 s + a3
s2
Introduce X1 = U
s3 + a1 s2 + a2 s + a3
d3 y d2 y dy s 1
x1 = , x2 = , x3 = , x4 = y X2 = 3 U = X1
dt3 dt2 dt s + a1 s + a2 s + a3
2 s
1 1 1
X3 = 3 U = X2 = 2 X1
s + a1 s2 + a2 s + a3 s s
dx1 1 1
= −4x1 − 6x2 − 4x3 − x4 + u sX 1 = −a1 X 1 − a2 X 1 − a3 2 X 1 + U
dt s s
dx2
= x1 = −a1 X 1 − a2 X 2 − a3 X 3 + U
dt
dx3 sX 2 = X 1
= x2 sX 3 = X 2
dt
dx4 Y = b1 X 1 + b2 X 2 + b3 X 3
= x3
dt


c K. J. Åström December, 2000 7
 From Transfer Functions to State Equations ... From Transfer Functions to State Equations

1 1 Y ( s) b1 s2 + b2 s + b3
sX 1 = −a1 X 1 − a2 X 1 − a3 2 X 1 + U = 3
s s U ( s) s + a1 s2 + a2 s + a3
= −a1 X 1 − a2 X 2 − a3 X 3 + U s3 X 1 + a1 s2 X 1 + a2 sX 1 + a3 X 1 = b1 s2 U + b2 sU + b3 U
sX 2 = X 1 1 1 1 
sX 3 = X 2 sX 1 = −a1 X 1 + b1 U + −a2 X 1 − a3 X 1 + b2 U + b3 U
s s s
Y = b1 X 1 + b2 X 2 + b3 X 3 1 1 1 
X 2 = 2 −a2 X 1 − a3 X 1 + b2 U + b3 U
s s s
Back to time domain sX 1 = −a1 X 1 + b1 U + X 2 = −a1 X 1 + X 2 + b1 U
1 
dx1 sX 2 = −a2 X 1 + b2 U + −a3 X 1 + b3 U
= −a1 x1 − a2 x − a3 x3 + u s
dt 1 
dx2 X3 = −a2 X 1 − a3 X 1 + b3 U
= x1 s
dt sX 2 = −a2 X 1 + b2 U + X 3 = −a2 X 1 + X 3 + b2 U
dx3
= x2 sX 3 = −a3 X 1 + b3 U
dt
y = b1 x1 + b2 x2 + b3 x3

From Transfer Functions to State Equations ... Stability

Stable Unstable
For the nonlinear system 1 1
sX 1 = −a1 X 1 + b1 U + X 2 = −a1 X 1 + X 2 + b1 U
sX 2 = −a2 X 1 + b2 U + X 3 = −a2 X 1 + X 3 + b2 U dx
= f ( x) 0.5 0.5
dt
sX 3 = −a3 X 1 + b3 U
only stability of a solution. 0
0 10 20 30 40 50
0
0 10 20 30 40 50
In the time domain For LTI systems stability Stable
1 Unstable
1
of one solution guarantees
dx1
= −a1 x1 + x2 + b1 u stability of all 0 0
dt
dx2 dx
= −a2 x1 + x3 + b2 u dt
= Ax −1 −1
dt 0 10 20 30 40 50 0 10 20 30 40 50
dx3
= −a3 x1 + b3 u An LTI system is stable if all eigenvalues of A are in the left half
dt
y = x1
plande or, equivalently, if the characteristic equation
det (sI − A) = sn + a1 sn−1 + a2 sn−2 + . . . + an = 0

has all roots in the left half plane.

c K. J. Åström December, 2000 8
 The Nyquist Criterion Stability Margins
The margins express the distance from the Nyquist curve to
5

3
the critical point in different ways. Two classical measures: gain
2 margin gm and phase margin ϕ m. An alternative is the shortest
1
distance to the critical point −1 which is 1/ M s , where Ms is the
0

−1
maximum sensitivity.
−2

−3

−4

−5
−5 −4 −3 −2 −1 0 1 2 3 4 5

Investigate propagation of sinusoids through the closed loop.

An oscillation can be maintained if L(iω 0 ) = −1. Notice that
the feedback is positive for the frequency ω 0 .
The point −1 is the critical point!
Stability based on encirclements of critical point.

Sensitivity to Process Variations Modeling

Closed loop transfer function with error feedback Modeling from physics
PC • Use knowledge of physics and engineering to derive
T=
1 + PC differential equations
Small variations in the process • Try to understand what factors are relevant for the problem
• Useful to think of frequencies and magnitudes
d T  1 d P d P
= =S
T  1 + PC  P  P • States describe storage of mass momentum and energy

Large variations stable if Modeling from data

• Introduce perturbations (steps, pulses, sinusoids) to the
∆ P 1
< process and observe its response
 P T 
• Fit LTI models (system identification)
Maximum sensitivities Ms = maxω  S (iω ), Mt = maxω T (iω )
Often useful to combine the approaches

c K. J. Åström December, 2000 9
 2. PID Control PI Control of First Order System
b
Process: P(s) =
s+a
ki
Controller: C(s) = k +
s
Closed loop transfer function from reference r to output y
Y ( s) PC b( ks + ki )
= = 2
1   sTd  R( s ) 1 + PC s + ( a + bk) s + bki
U ( s) = k β R( s) − Y ( s)) + R( s) − Y ( s) + γ R( s) − Y ( s)
sTi 1 + sTd/ N
All poles can be assigned. A good characteristic polynomial is

s2 + 2ζ ω 0 s + ω 20
Notice
• Set point weighting β and γ (Cheap 2 DOF) Identification of coefficients of equal powers of s gives
• Filtering of derivative k = (2ζ ω 0 − a)/ b, ki = ω 20 / b

PD Control of Second Order System PI Control of Second Order System

b b
Process: P(s) = Process: P(s) =
s2 + a1 s + a 2 s2 + a1 s + a 2
Controller: C(s) = k + k d s ki
Controller: C(s) = k +
Closed loop transfer function from reference r to output y s
Closed loop transfer function from reference r to output y with
Y ( s) PC b( kd s + k) b( kd s + k) error feedback
= = 2 = 2
R( s) 1 + PC s + a1 s + a2 + b( kd s + k) s + ( a1 + bkd ) s + a2 + bk
Y ( s) PC b( ks + ki )
= =
R( s ) 1 + PC s( s2 + a1 s + a2 ) + b( ks + ki )
All poles can be assigned. A good characteristic polynomial is b( ks + ki )
= 3
s + a1 s + ( a2 + bk) s + bki
2
s2 + 2ζ ω 0 s + ω 20
Not enough parameters to assign all poles.
Identification of coefficients of equal powers of s gives

k = (ω 20 − a2 )/ b, kd = (2ζ ω 0 − a1 )/ b


c K. J. Åström December, 2000 10
 PI Control of Second Order System PID Control of Second Order System
A compromise b
Process: P(s) =
s2 + a1 s + a 2
Y ( s) b( ks + ki )
= 3 ki
R( s ) s + a1 s + ( a2 + bk) s + bki
2
Controller: C(s) = k + + kds
s
Make characteristic polynomial equal to
Closed loop transfer function from reference r to output y
s3 + 2ω 0 s2 + 2ω 20 s + ω 30
Y ( s) PC b( kd s2 + ks + ki )
= =
R( s) 1 + PC s( s2 + a1 s + a2 ) + b( kd s2 + ks + ki )
and also choose ω 0 (response speed). Hence
b( kd s2 + ks + ki )
=
a1 2ω 20 − a2 a2 /2 − a2 ω 30 a31 s3 + ( a1 + bkd ) s2 + ( a2 + bk) s + bki
ω0 = , k= = 1 , ki = =
2 b b b 8b
All poles can be assigned.
Pole pattern can be chosen but the response speed is given by
process parameter a 1 .

PID Control of Second Order System ... Controllers with Error Feedback
Error feedback
Y ( s) PC b( kd s2 + ks + ki )
= = 3 ki
R( s) 1 + PC s + ( a1 + bkd ) s2 + ( a2 + bk) s + bki U ( s) = C( s)( R( s) − Y ( s)) = ( k + + kd s)( R( s) − Y ( s))
s
A good characteristic polynomial is
gives the closed loop transfer function from reference r to
s + 2s ω 0 +
3 2
2ω 20 s + ω 30 output y

Identification of coefficients of equal powers of s gives Y ( s) PC b( kd s2 + ks + ki )

= = 3
R( s ) 1 + PC s + ( a1 + bkd ) s2 + ( a2 + bk) s + bki
2ω 20 − a2 ω 30 2ω 0 − a1
k= , ki = , kd = All poles can be placed arbitrarily. Can something be done
b b b
about the zeros?

c K. J. Åström December, 2000 11
 Controllers with Set Point Weighting Saturation Windup and Windup Protection
Controller with set point weighting (a kind of feedforward) The feedback loop is broken if signals saturates. Problem
with unstable modes in the process and the controller. The
ki
U ( s) = k( β R( s) − C( s)) + ( R( s) − Y ( s)) + kd s(γ R( s) − Y ( s)) integrator in a PI is a typical example.
s
ki ki –y
= (β k + + γ kd s) R( s) − ( k + + kd s) Y ( s) KT d s
s s Actuator
= Cr ( s) R( s) − C( s) Y ( s) model Actuator
e = r− y
K Σ
gives the closed loop transfer function
K 1 – +
Σ Σ
Y ( s) Cr ( s) P( s) b(γ kd s + β ks + ki )
2 Ti s
= = 3
R( s ) 1 + P( s) C( s) s + ( a1 + bkd ) s2 + ( a2 + bk) s + bki es
1
Tt
The numerator can be modified by choosing the parameters β
and γ .
Look for the error. Assume system works.

Controller Tuning Ziegler-Nichols’ Step Response Method

• Model based: Make a model and apply some control • Switch controller to manual.
design method e.g. the methods just discussed • Make a step in the control variable.
• Make simple experiments on the process and adjust • Log process output. Normalize the curve so that it corre-
• Ziegler and Nichols methods sponds to a unit step.
– Good idea but systems are not well damped and too • Determine parameters a and L. The controller parameters
little process information is used are obtained from a table.
– Open loop experiments (bump test)
– Closed loop experiments K
Controller k Ti Td Tp
• Improved tuning rules A0 P 1/a 4L
PI 0.9/a 3L 5.7L
1.2/a L/ 2 3.4L
a

L
PID 2L

c K. J. Åström December, 2000 12
 Ziegler-Nichols’ Frequency Response Method Properties of Ziegler and Nichols Rules
• Switch the controller to pure proportional. Properties
• Adjust the gain so that the closed loop system is at the + Easy to explain and use
stability boundary.
+ Very common
• Determine the gain k u (the ultimate gain) and the period
- The closed loop system obtained too oscillatory ζ  0.2.
Tu (the ultimate period) of the oscillation.
Part of the criterion (quarter amplitude damping)
• Suitable controller parameters are obtained from a table.
- Too large overshoot
Im G(iω )

Ultimate point Reg. k Ti Td Tp - Sensitive to process variations

−1 ϕ Re G(iω ) P 0.5ku Tu - Can only be used on a restricted class of processes
a

ω PI 0.4ku 0.8Tu 1.4Tu A good idea to base tuning on simple process experiments.
PID 0.6ku 0.5Tu 0.125Tu 0.85Tu Large scope for improvements.
More process information needed.

A Ziegler Nichols Replacement Computer Implementation

• ZN is basically a sound method, BUT • Sampling, aliasing and anti-aliasing filters
Damping specified is too low  ζ = 0.2 • Approximate derivatives by differences
Too little process information (Two parameters k u, Tu ) • Use backward differences
• An additional parameter give a big improvement • Organization of the code
κ =  G (ω 180 )/ G (0)
K/Ku vs kappa Ti/Tu vs kappa
p( tk) = k ∗ b( y sp( tk ) − y( tk))
0 1
10 10
e( tk) = ysp( tk ) − y( tk)
Td  
d( tk ) = d( tk−1) − kN y( tk) − y( tk−1)
Td + Nh
−1 0
10 10
v= p( tk) + i( tk ) + d( tk )
u( tk )= sat(v)
kh kh  
−2
10 10
−1
i( tk+1) = i( tk ) + e( tk) + u−v
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 Ti Tr

c K. J. Åström December, 2000 13
 3. The Basic Feedback Loop Some Observations
d n
• A system based on error feedback is characterized by four
r u x y transfer functions (The Gang of Four)
F ∑ C ∑ P ∑
• The basic feedback system with two degrees of freedom is
characterized by six transfer function (The Gang of Six)
−1 • To fully understand a system it is necessary to look at all
systems
Six transfer functions are required to describe the system • It may be strongly misleading to only show properties of
a few systems for example the response of the output to
P PC PCF command signals. This is a common error in the literature.
X = D− N+ R
1 + PC 1 + PC 1 + PC
P 1 PCF • The properties of the different transfer functions can be
Y= D+ N+ R illustrated by transient or frequency responses.
1 + PC 1 + PC 1 + PC
PC C CF
U =− D− N+ R
1 + PC 1 + PC 1 + PC

A Possible Choice The Sensitivity Functions

Six transfer functions are required to show the properties of a
1 L
basic feedback loop. Four characterize the response to load S= , T= , L = PC, S+T =1
1+ L 1+ L
disturbances and measurement noise.
PC P Disturbance reduction
1 + PC 1 + PC Ycl ( s)
C 1 =S
Yol ( s)
1 + PC 1 + PC
Sensitivity to small process variations
Two more are required to describe the response to set point
changes. d T  d P
=  S
T   P
PCF CF
1 + PC 1 + PC Large variations do not give instability if
∆ P 1
<
 P T 


c K. J. Åström December, 2000 14
 Maximum Sensitivities The Sensitivity Functions

1 1 L PC 1 L Ycl ( s) ∆ P 1
S= = , T= = S= , T= , = S, <
1+ L 1 + PC 1+ L 1 + PC 1+ L 1+ L Yol ( s)  P T 
Maximum sensitivities A compromise: Good disturbance rejection |S| small! Good
Ms = max  S( iω ), Mt = max  T ( iω ) robustness T  small, but S + T = 1 and
Bode’s integrals: sL(s) goes to zero as s → ∞
The value 1/ Ms is the shortest distance from the Nyquist curve



of the loop transfer function L(iω ) to the critical point −1. ∞ ∞
1
log  S( iω )dω = log  dω = π pi
0 0 1 + L( iω )
The condition
∞
∞  1
L(1/ iω )
log  T (1/ iω )dω = log   dω = π
∆ P( iω ) 1 0 0 1 + L(1/ iω ) zi
<
 P( iω )  M t
The water bed effect.
guarantees that the perturbation ∆ P does not give instability.

4. Feedforward System with Two Degrees of Freedom

• Feedback • Feedforward l n

• Closed loop • Open loop r y

u x ∑
F ∑ C ∑ P
• Acts only when there • Acts before deviations
are deviations show up
• Market Driven • Planning −1
• Robust to model errors • Not robust to model
 S  < 1 for some ω errors  S  = 1 for all ω
• Risk for instability • No risk for instability The system is said to have two degrees of freedom (2DOF)
because there are separate signal paths from set-point to u
Feedforward is used to and from measured signal to u.
• Improve response to reference signals Set point weighting in PI control is a simple version of 2DOF
• Reduce the effect of measurable disturbances
t
u( t) = k( β r( t) − y( t)) + ( r(τ ) − y(τ )) dτ
0


c K. J. Åström December, 2000 15
 Many Versions of 2DOF Response to Reference Signals
l n

r x y Mu um
F ∑ C
u
∑ P ∑ r
ym uf b y
−1 My Σ C Σ P

Mu um
r −1
ym uf b y
My Σ C Σ P
Notice that Mu and M y can be viewed as generators of the
−1 desired output y m and the inputs u m which corresponds to y m.
Assume that process P, controller C and desired response M y
Horowitz: "Some structures have been presented as funda-
are given are given. Then M u = M y / P. Does not depend on C!
mentally different from the others ... all 2DOF configurations
have basically the same properties and potentials."

FC = Mu + Mu C

Reduction of Measurable Disturbance 5. State Feedback and Observers

Basic idea: PID controller predicts by linear extrapolation
de
e p = e + Td
dt

Can we predict better?

• The state captures all information about the past that is
relevant to predict the future.
Transfer function from disturbance to process output with • Find a way to compute the state from measurements of
combined feedback and feedforward input and output

Gd =
Y ( s)
=
P2 (1 − P1 G f f )
= P2 (1 − P1 G f f ) S
• A nice structuring of the control problem
D ( s) 1 + PC

Notice combined effect of feedback and feedforward. Ideal

feedforward compensator G f f = P1−1 = P y/ P yu.


c K. J. Åström December, 2000 16
 Determinants, Inverses, and Eigenvalues Determinants, Inverses, and Eigenvalues
Consider the following matrices and their transpose A vector e is an eigenvector if Ae = se and the number s is
  called an eigenvalue. We have
a11 a12
A=
a21 a22 Ae = se
( sI − A) e = 0
The determinant is
det ( sI − A) = 0
det A = a11 a22 − a12 a21
The eigenvalues are thus given by the characteristic equation
If the determinant is different from zero the matrix has an  
s − a11 −a12
inverse which is given by det ( sI − A) = det = s2 − ( a11 + a22 ) s + a11 a22 − a12 a21
−a21 s − a22
 
−1 1 a22 −a12
A =
det A −a21 a11
The roots of this equation are the eigenvalues of the matrix.

State Feedback The Closed Loop System

System and linear controller
dx
= Ax + Bu = ( A − B L) x + B lr r
dx dt
= Ax + Bu
dt y = Cx
u = − Lx + lr r u = − Lx + lr r

The closed loop system Transfer function from reference r to output y

dx Y ( s)
= Ax + Bu = Ax + B (− Lx + lr r) = ( A − B L) x + B lr r = C( sI − A + B L)−1 B lr
dt R( s )

If the system is controllable; i.e. if controllability matrix

Wc = ( B AB ... An−1 B )

is regular, then the matrix L can be chosen to give arbitrary

eigenvalues of A − B L.


c K. J. Åström December, 2000 17
 Computing the State Feedback Matrix L Observers
Start by determining controllability! Observer
System  
 C 
Simple problems, determine closed loop characteristic polyno- 
 

dx 
 CA  
mial, pick L to match desired polynomial = Ax + Bu 
 
dt  . 
Wo =  


 . 
. 
y = Cx 
 

Easy if system is in controllable canonical form  n−1

CA
   
 −a1 −a2 ... −an   1 The state can be computed from the output if system observ-

 
 
 


 0  
 0
 1 0    able; i.e. if the observability matrix W o is regular. Many different
dz 
 0









=


1 0 
 z + 


0 

 u ways, differentiation, simulation of model, A compromise:
dt 
 . 
 
 . 


 .. 
  .
. 


 
 

    
0 0 0 0 d x̂
= A x̂ + Bu + K ( y − C x̂) d x̃
dt = ( A − K C) x̃
dt
Transform to controllable canonical form W̃c = T Wc .
Find a matrix K so that A − K C has desired eigenvalues.
Use Matlab for numerical solutions.

Design of Observers State Feedback from Observed State

Basic compromise: Trust models or measurements? uc Model and u ff
Feedforward
Use results for feedback design A − K C = ( A − C K ) , T T T T
Generator
xm u fb y
∑ L ∑ Process
compare with state feedback A − B L
−
Simple problems, determine closed loop characteristic polyno-
xˆ
mial, pick L to match desired polynomial Observer

Easy if system is in observable canonical form W̃o = Wo T −1

   
 −a1 1 0 ... 0  b1  dx

 
 
 
 = Ax + Bu
   b2  dt
 −a2
 0 1 0 
 

dz  
 .





 .


 y = Cx
=
 .. 



z+
 .
.


 u
dt 
 
 
 
 u = u m + L( xm − x̂)

 
 
 


 − a n− 1 0 0 1 
 
     d x̂
= A x̂ + Bu + K ( y − ŷ)
−an 0 0 0 bn dt
 
y = 1 0 0... 0z y = C x̂


c K. J. Åström December, 2000 18
 The Closed Loop System The Closed Loop System ...

dx dx
= Ax + Bu = ( A − B L) x + B Lx̃ + B lr r
dt dt
y = Cx d x̃
= ( A − K C) x̃
u = − L x̂ + lr r dt
d x̂ Characteristic equation
= A x̂ + Bu + K ( y − C x̂)
dt
det ( sI − A + B L) det ( sI − A + K C) = 0
Introduce the state x̃ = x − x̂ instead of x̂ .
dx Transfer function from reference r to output y
= Ax + Bu = Ax − B Lx̂ + B lr r
dt
= Ax − B L( x − x̃) + B lr r = ( A − B L) x + B Lx̃ + B lr r Y ( s)
= C( sI − A + B L)−1 B lr
R( s )
d x̃
= ( A − K C) x̃
dt

Integral Action
Meaning of the model An alternative
dx dx
= Ax + Bu = Ax + B (u + v)
dt dt
dv dv
=0 =0
dt dt
Feedback from estimated states x̂ and v̂

        
d x̂ A B x̂ B K
= + u+ ε
dt v̂ 0 0 v̂ 0 Kv
ε = ( y − C x̂)
u = − L x̂ − v̂

c K. J. Åström December, 2000 19