Examination

Control Systems Theory and Design

02.03.2018

Problem 1 (25 Marks)

Consider a person, who is driving an electric unicycle as depicted in Figure 1. The person
might be thinking about the way the unicyle works and how a state space model could
help to explain it.

state space model?

α

l mp

mw

τ r

s s

Figure 1: Electric unicycle

The linearised differential equations of the electric unicycle are given as

(2mw + mp )s̈(t) + mp lα̈(t) − 1r τ (t) = 0,
(1)
mp ls̈(t) + mp l2 α̈(t) − mp glα(t) + τ (t) = 0,
where s is the position of the unicycle, α the angle, and τ the motor torque that is driving
the unicycle.

a) Assume that
mp = 100, mw = 10, l = 1, r = 0.5, g = 10
and bring the differential equations into following form
     
s̈(t) ṡ(t) s(t)
M0 + D0 + K0 = F0 τ (t).
α̈(t) α̇(t) α(t)

b) Determine the state space model

ẋ(t) = Ax(t) + Bu(t), y(t) = Cx(t) + Du(t),
 
where x(t) = s(t) α(t) ṡ(t) α̇(t) is the state vector. Assume that the angle
α(t) is measured and the torque τ (t) is the input. Hint: Use the facts
   
0 I 0
A= , B= .
−M0−1 K0 −M0−1 D0 −M0−1 F0

1
c) Is the linearised state space model stable? Give reasons.

Hint: The characteristic polynomial p(λ) of A can be determined as

p(λ) = det(λM0−1 D0 + λ2 I + M0−1 K0 ).

d) Is the model observable? Give reasons.

Hint: No calculation needed!

e) Is the model in (b) a minimal realisation? Give reasons.

f) A Matlab function named check model.m is to be written. The function header

is given below. Write the function body, such that it performs as described in the
comments.

1 function [ stab , con, ob, min ] = check_model( sys )

2 % input: state space model sys
3 % output:
4 % * stab: 1 if stable,
5 % 0 if not stable
6 % * con: 1 if controllable
7 % 0 if uncontrollable
8 % * ob: 1 if observable
9 % 0 if unobservable
10 % * min: 1 if minimal realisation
11 % 0 if not minimal
12
13 % write function body
14
15 end

2
Problem 2 (25 Marks)

Tasks (a), (b) and (c) can be solved independently.

a) Given is the block diagram in Figure 2

d1

G1 (s) G2 (s)

u y
G4 (s)

G3 (s)
d2

Figure 2: Block diagram

i) If u ∈ R2 , d1 ∈ R3 , d2 ∈ R, y ∈ R2 , what are the dimensions of Gi (s),

i = 1, ..., 4?
ii) Write down the transfer function Gyd1 (s) from d1 to y in terms of Gi (s), i =
1, ..., 4.
iii) Assume now the loop is closed with unity positive feedback, i.e. u = y. Write
down the transfer function from d1 to y.

b) i) Write down a state space model of the dynamics from u1 to y2 for the intercon-
nection in Figure 3 in terms of Ai , Bi , Ci , Di . Use the state vector [xT1 xT2 ]T .

x1 x2
u1 y1 = u2 y2
(A1 , B1 , C1 , D1 ) (A2 , B2 , C2 , D2 )

Figure 3: Series interconnection

ii) Assume the model (A1 , B1 , C1 , D1 ) is unstable and (A2 , B2 , C2 , D2 ) is stable.

What can be said about stability of the resulting interconnection, assuming
[xT1 xT2 ]T is the state vector of a minimal realization? Explain.
iii) Write down a state space model of the dynamics from r to y3 for the intercon-
nection in Figure 4 (note that Di = 0) in terms of Ai , Bi , Ci . Use the state
vector [xT3 xT4 ]T .
iv) Assume the model (A3 , B3 , C3 ) is unstable and (A4 , B4 , C4 ) is stable. What can
be said about stability of the resulting interconnection, assuming [xT3 xT4 ]T is
the state vector of a minimal realization? Explain.

3
 x3
r y3
(A3 , B3 , C3 )
−

(A4 , B4 , C4 )
y4 u4 = y3
x4

Figure 4: Feedback interconnection

c) Given the state space model

   
0 1 0 0 0  
u
ẋ = −4 −3 0  x + 1 0 1
u2
0 0 −3 0 1 (2)
      
y1 1 3 0 0 1 u1
= x+
y2 0 0 1 0 0 u2

i) Which of the following block diagrams is a representation of the system? Ex-

plain.

u1 y1 u1 y1 u1 y1 u1 y1

u2 y2 u2 y2 u2 y2 u2 y2

a b c d

ii) Is the state space model (2) controllable? Explain. Hint: No calculation is
required, consider the structure of the system (from (c.i)) and each subsystem

4
Problem 3 (25 Marks)

Tasks (a) and (b) can be solved independently

Given is a balanced state space model

ẋ = Ax + Bu
(3)
y = Cx + Du

where A ∈ R348×348 , B ∈ R348×1 , C ∈ R1×348 , D ∈ R1×1 , for which an output feedback

controller is to be designed. Due to the high order of the model, model order reduction
techniques are to be used to obtain an approximate, low-order model more suitable for
controller design. The Hankel singular values of the full-order model are depicted in
Figure 5.
Hankel Singular Values
2500

2000

1500
abs

1000

500

0
2 4 6 8 348
Order

Figure 5: Hankel singular values of full-order model

a) i) What number of states would be a reasonable choice for a reduced order model?
Explain. Hint: There is more than one possibility.
ii) Introducing the partition according to significant/negligible Hankel singular
values    
A11 A12 B1  
A= B= C = C1 C2 (4)
A21 A22 B2
what are the dimensions of each block-partition (Aij , Bi , Cj ) in (4)? Relate
this answer to your choice in (a.i).
iii) Give expressions for Atrunc , Btrunc , Ctrunc , Dtrunc in terms of the block-partitions
in (4), for the model matrices of a reduced order model using balanced trun-
cation.
iv) Give expressions for Ares , Bres , Cres , Dres in terms of the block-partitions in (4),
for the model matrices of a reduced order model using balanced residualization.
v) Figure 6 shows the Bode magnitude plot of the full-order model and the
reduced-order models using balanced truncation and residualization. Which
curve corresponds to which method? Explain.

5
 Bode Diagram
Full-order
Reduced 1
Reduced 2

40

Magnitude (dB)
0

-40
10 -2 10 -1 10 0 10 1 10 2
Frequency (rad/s)

Figure 6: Bode plot of full- and reduced order models

b) An observer-based state feedback controller is to be designed for the reduced order

model using the structure in Figure 7, where Gf ull (s) is the full-order model. A
schematic of the observer is shown in Figure 8, where the reduced-order model
matrices (Ared , Bred , Cred , Dred ) are used.

u − y+n
BP y l
× Gf ull (s)
u
Bred 1 Cred
f s

x̂ Ared
Observer n

x̂
Figure 7: Closed-loop system
Figure 8: Observer schematic

i) Assume the loop in Figure 7 is cut open at point BP (plant’s input). Give
MATLAB code to compute a state space model from n to u. Assume Ared ,
Bred , Cred , f , l are already in the workspace.
ii) Given are two possible sets of values for the observer tuning parameters Qe
and Re . Match each design (1 and 2) with its corresponding
– controller Bode magnitude plot (a and b) in Figure 9
– initial value response (c and d) in Figure 10
Give reasons for your answer.

design 1 Qe = I, Re = 100
design 2 Qe = I, Re = 0.1

6
 0 a 20
c
b
d

|K(s)| (dB)
-40
10

y(t)
-80
0

10-4 100 104 0 20 40 60 80

Frequency (rad/s) Time (s)

Figure 9: Bode plot of K(s) Figure 10: Closed-loop response

iii) How should the Bode magnitude plot of K(s) look like in the low-frequency
range if good tracking is desired?
iv) The full-order model in (3) corresponds to a flexible beam. The goal of the
controller developed in part (b) is to damp vibrations. Give an interpretation
of the low frequency behavior of K(s) seen in Figure 9; relate your answer to
(b.iii).

7
Problem 4 (25 Marks)

r(k) e(k) u(k) y(k)

D(z) zoh G(s)
− T
H(z)

Figure 11: Feedback loop

A continuous-time state space realisation of a plant with transfer function

s−1
G(s) = (5)
s2 + 3s + 2
is given as

ẋ(t) = Ax(t) + Bu(t), y(t) = Cx(t) + Du(t), (6)

where
   
−1 −2 0  
A= , B= , C= 1 1 , D = 0. (7)
0 −2 1

Consider the feedback loop given in Figure 11.

a) Find a discrete-time state space representation of the plant that describes the be-
haviour of the plant in the sampling instants exactly. Assume T = 1.

Hint: The eigenpairs of A are

     
1 2
(λ1 , v1 ) = −1, , (λ2 , v2 ) = −2, ,
0 1

where Avi = λi vi holds for i = {1, 2}.

Do not develop eAT into a power series.

b) How many zeros does the discretised system in part (a) have? Give reasons.

c) Give an expression for the zeros of the discretised system if G(s) is sampled at high
sampling rates (T → 0) with a zero order hold? Give reasons.

d) Consider the unit step responses (I) - (IV) given in Fig. 12. Which step response
corresponds to the response of the discretised system in (a)?

e) Give Matlab code to plot the step response of the discretised system in (a). Assume
that the workspace is empty.

8
 (I) (II)

1 1

y(k)

y(k)
0.5 0.5

0 0
0 1 2 3 4 5 0 1 2 3 4 5
k k

(III) (IV)
0.6 0.2

0.4 0
y(k)

y(k)
−0.2
0.2
−0.4
0
0 1 2 3 4 5 0 1 2 3 4 5
k k

Figure 12: Unit step response

9
Formulas

∞ ZT
AT
X (AT )n 1
Φ=e = = I + AT + A2 T 2 + . . . , Γ= eAτ B dτ
n=0
n! 2
0

F = −pTa−1 C −1 with
 
p = ān−1 − an−1 ān−2 − an−2 . . . ā0 − a0 and
 
1 an−1 an−2 . . . a1
0 1 an−1 . . . a2 
Ta =  .. , where ai are the open-loop transfer function coefficients
 
.. . . . .. 
. . .
0 0 0 ... 1

4.6
s1,2 = −σ ± jωd , ts = , ζ = sin(θ), θ = tan−1 (σ/ωd )
σ
 −1  
−1 a b 1 d −b
A = =
c d det(A) −c a
Controller canonical form
bm sm + bm−1 sm−1 + · · · + b1 s + b0
G(s) =
sn + an−1 sn−1 + · · · + a1 s + a0
      
ẋ1 0 1 0 ··· 0 x1 0
 ẋ2   0 0 1 0   x2  0
      
 ..   .. . . ..   ..  +  ..  u
 .   .= . . 
  .  .
   
  
ẋn−1   0 0 0 ··· 1   xn−1  0
ẋn −a0 −a1 −a2 · · · −an−1 xn 1
 
x1
   x2 

y = b0 b1 · · · bn−1  .. 

.
xn

−1
eA = eXΛX = XeΛ X −1
Kirchhoff’s loop rule R

v0 vR L
v0 = vR + vL vL

Figure 13: Kirchhoff’s loop rule

Current-voltage relation

L iL R iR
diL vR = R iR
vL = L
dt
vL vR

10
Solution to Problem 2
a) i) G1 ∈ R3×2 , G2 ∈ R1×3 , G3 ∈ R1×2 , G4 ∈ R2×1
ii) From the block diagram y = G4 G2 d1 , therefore

Gd1 y = G4 G2

iii) In closed loop u = y, so

y = G4 (G2 (d1 + G1 y) + G3 y) = G4 G2 d1 + G4 G2 G1 y + G4 G3 y

Solving for y

y = (I − G4 G2 G1 − G4 G3 )−1 G4 G2 d1
therefore

Gyd1,cl = (I − G4 G2 G1 − G4 G3 )−1 G4 G2

b) i) State space models are given by

ẋ1 = A1 x1 + B1 u1 (1) ẋ2 = A2 x2 + B2 u2 (3)

y1 = C1 x1 + D1 u1 (2) y2 = C2 x2 + D2 u2 (4)
As u2 = y1 , substitute (2) into (3) and (4)

ẋ2 = A2 x2 + B2 (C1 x1 + D1 u1 )
y2 = C2 x2 + D2 (C1 x1 + D1 u1 )

Finally, write state space model

      
ẋ1 A1 0 x1 B1
= + u
ẋ2 B2 C1 A2 x2 B2 D1 1
 
  x1
y2 = D2 C1 C2 + D2 D1 u1
x2

ii) From the triangular structure, the eigenvalues of the interconnection are the
eigenvalues of A1 and A2 . If any of the 2 has unstable eigenvalues, the inter-
connection is unstable, unless there is an unstable pole-zero cancellation. As
the model is minimal, there is no pole-zero cancellation, meaning the intercon-
nection is unstable.
iii) In this case u3 = r − y4 and u4 = y3 proceed similarly to task (b.iii) to get
      
ẋ3 A3 −B3 C4 x3 B
= + 3 r
ẋ4 B4 C3 A4 x4 0
 
  x3
y 3 = C3 0
x4

1
 iv) Nothing can be concluded about this interconnection by knowing only stability
properties of each subsystem.

c) i) Note that the state space model can be partitioned as

   
0 1 0 0 0  
u 1
ẋ =  −4 −3 0  x+ 1
 0 
u2
0 0 −3 0 1 (5)
      
y1 1 3 0 0 1 u1
= x+
y2 0 0 1 0 0 u2

where the dynamics are completely decoupled and the only coupling is between
input u2 and output y1 via a feedthrough, therefore the correct choice is (c).
ii) Considering the fact that we have 2 dynamically decoupled systems we can
look at each independently. The system corresponding to x3 is a first-order
system with input u2 so it is trivial to see that it is controllable. Subsystem
[x1 x2 ] is in controllable canonical form and it is therefore controllable.

2
Solution to Problem 3
a) i) Either 2, 4 or 6 would be a reasonable choice as there is a large gap between
the Hankel singular values. Given that the original system is of order 348, one
could (correctly) argue that a system of order 2 would likely be too simple
to properly represent the original system and looking, e.g. at Bode plots or
exploratory experiments would give a better idea.
ii) Assume we chose to reduce the system to order 6. Then

A11 ∈ R6×6 , A12 ∈ R6×342 , A21 ∈ R342×6 , A22 ∈ R342×342

B1 ∈ R6×1 , B2 ∈ R342×1

C1 ∈ R1×6 , C2 ∈ R1×342
iii) Atrunc = A11 , Btrunc = B1 , Ctrunc = C1 , Dtrunc = D
iv) Recall the assumption made in balanced residualization, ẋ2 ≈ 0, then

0 = A21 x1 + A22 x2 + B2 u
Solve for x2

x2 = −A−1
22 (A21 x1 + B2 u)

and substitute into original state space model

ẋ1 = A11 x1 + A12 (−A−1

22 (A21 x1 + B2 u)) + B1 u
−1
y1 = C1 x1 + C2 (−A22 (A21 x1 + B2 u)) + Du

or

ẋ1 = (A11 − A12 A−1 −1

22 A21 )x1 + (B1 − A12 A22 B2 )u
y1 = (C1 − C2 A−1 −1
22 A21 )x1 + (D − C2 A22 B2 )u

therefore

Ares = A11 − A12 A−1

22 A21 , Bres = B1 − A12 A−1
22 B2 ,

Cres = C1 − C2 A−1
22 A21 , Dres = D − C2 A−1
22 B2

v) ”Reduced 1” is the reduced model resulting from balanced residualization: In

the original model D = 0 (as can be seen by the roll-off at high frequencies), the
fact that the reduced model has D 6= 0 results from the reduction procedure
as can be seen from the expression Dres where C2 A−1 22 B2 6= 0. Moreover, the
low-frequency behavior matches the original because ẋ2 ≈ 0.
”Reduced 2” is the reduced model resulting from balanced truncation: High
frequency behavior matches as Dtrunc = D, and there is a (small but nonzero)
error in low frequencies.

3
b) i) The transfer function from n to u corresponds to the controller K(s), a state
space model for it can be constructued in MATLAB with the command
ss(Ared + Bred ∗ f + l ∗ Cred), −l, f, 0)
ii) Design 2 assumes the noise covariance is small (which means it is assumed
that there is little measurement noise) for this reason the observer will be fast
(larger bandwidth, i.e. plot b) and for this reason will also let more noise
through (plot d).

design 1 a c
design 2 b d

iii) For good tracking, K(s) should have a very large gain, ideally infinite (i.e.
integral action) at low frequencies.
iv) As can be seen in the Bode plot, the gain at low frequencies is very small, this
contrasts with the objective for tracking (answer from (b.iii) ). This can be
explained by observing that for a flexible beam it makes no sense to track a
reference, what is desired in that vibrations in a band of frequencies are rejected
which is why the controller has the highest gain in the mid-frequency range.
As always, noise should be rejected so the gain at high frequencies should be
0.