Examination

Control Systems Theory and Design

14.07.2014

Problem 1 (25 Marks)

a) A state space model of a dynamic system is given as

    
−1 1 0 0 x1 0
−1 −1 0 0 x2  1
ẋ = A x + B u =    +  u
0 0 −2 0 x3  0
0 0 0 2 x4 1
  (1)
x1
  x2 
y =Cx= 1 0 1 1  x3  .


x4

i) What are the eigenvalues λi (i = 1, 2, 3, 4) of the matrix A and what can be

inferred about the stability of the system governed by Eq. (1)?
ii) Is the state space realization (1) stabilizable? Give reasons for your answers.

b) A second order system is given by the transfer function

Y (s) ω2
G(s) = = 2 . (2)
U (s) s + 2 δ ω s + ω2

Assume that feedback control is to be used to increase the damping ratio δ to δ + ∆,

where ∆ > 0, without altering the natural frequency ω. The desired behavior is
then described by the the transfer function

ω2
Gdesired (s) = . (3)
s2 + 2 (δ + ∆) ω s + ω 2

i) Write down a state space realization (A2 , B2 , C2 , D2 ) of the system (2) and
(Adesired , Bdesired , Cdesired , Ddesired ) of the system (3) both in controller canonical
form. Are these representations controllable? Give reasons.
ii) Calculate the state feedback gain F such that Adesired = A2 + B2 F .
iii) Now assume that the states of the system G cannot be measured, but only the
output y. Is it possible to represent the state feedback control law u = F x that
you derived in ii) as an ideal PD-controller u = KP y + KD ẏ? If so, determine
the gains KP and KD . If not, explain why.

1
Problem 2 (25 Marks)

a) Figure 1 shows a simplified version of the A320 aircraft’s fly-by-wire system used to
control climbing and descent. Write down a state-space realization of the controller
using the labeled signals.

kδ

δ ẋ R x η
jδ
−

jq jn kq kn
q
n

Figure 1: Simplified fly-by-wire controller of the A320 aircraft.

b) Consider the feedback interconnection of two multivariable systems depicted in

Fig. 2. The respective state space realizations of the two subsystems are

ẋ1 = A1 x1 + B1 u1 ẋ2 = A2 x2 + B2 u2
y 1 = C1 x 1 y 2 = C 2 x2

r u1 y1 y
(A1 , B1 , C1 )
−

(A2 , B2 , C2 )
y2 u2

Figure 2: Feedback interconnection of two systems

i) Write downthestate space realization for the closed-loop system with the state
x
vector x = 1 .
x2
ii) How many inputs and outputs has the closed-loop system if y1 (t) ∈ R3 and
y2 (t) ∈ R2 ?
iii) Given the matrices A1, B1, C1 and A2, B2, C2 in the workspace, complete
the following Matlab code on your answer sheet. It generates the closed-loop
system in Matlab and simulates the system response for each input-output
pairing.

2
 1 % a s s e m b l e t h e c l o s e d −l o o p system
2 sys1 = ss ( . . . , . . . , . . . , 0 ) ;
3 sys2 = ss ( . . . , . . . , . . . , 0 ) ;
4 cl_sys = feedback ( . . . , . . . ) ;
5

6 % p l o t channel step responses

7 [ ny , nu ] = s i z e ( . . . ) ;
8 for i = 1 : ny
9 for j = 1 : nu
10 [ y , t ] = step ( cl_sys ( ny , nu ) ) ;
11 figure (1)
12 plot ( . . . , . . . )
13 hold a l l
14 end
15 end

c) An LQR controller augmented with integrators to achieve integral action is to be

designed for a strictly proper multivariable system with 2 inputs, 2 outputs and 3
states. Note that this controller does not include an observer since all states are
available for feedback.
The model matrices A, B, C are present in the workspace.

i) How many states has the closed-looped system?

ii) Write down Matlab code to construct the state feedback gain F and integral
feedback gain FI. Use Q = C T C, QI = I and R = I as weights for the design.
iii) Draw a block diagram of the controller (not of the closed-loop system) including
the augmented integral action.

3
Problem 3 (25 Marks)

a) The impulse response g(t) of a continuous time system G(s) is given by

g(t) = (4e−2t + 15e−5t )σ(t).

i) Determine the z-transform of g(t) for T = 1s.

ii) Use your results from (i) to determine a discretization Gi (z) of G(s).
iii) Determine an exact (zero order hold) discretization Ge (z) of G(s) for T = 1s.
iv) Is Gi (z) stable? Give reasons.
v) Is Ge (z) stable? Give reasons.
vi) If the Tustin approximation is used to discretize G(s) as Gt (z), how many zeros
do you expect in Gt (z)? Give reasons.

b) Assume the Tustin approximation Gt (z) of G(s) for T = 1s is given in time domain
by the difference equation

y(k) = −a1 y(k − 1) − a2 y(k − 2) + b0 u(k) + b1 u(k − 1) + b2 u(k − 2).

and the coefficients a1 , a2 , b0 , b1 , b1 as well as an input sequence are available in the

Matlab workspace as a1, a2,b0,b1,b2 and u.
Complete the following Matlab code on your answer sheet to simulate the model
response y(k) to the input sequence u(k) and to plot the model response.

1 N = length ( u ) ;
2 y = zeros ( . . . ) ;
3 a = ...;
4

5 for k = a : N
6

7 phi = [ −y ( k −1) , . . . , . . . , u ( k −1) , . . . ] ’ ;

8

9 y(k) = [ . . . , . . . , b0 , . . . , . . . ] ∗ phi ;
10

11 end
12

13 plot ( y ) ;

4
Problem 4 (25 Marks)

a) Assume that the structure of a discrete-time LTI system to be identified is given by

b1 z + b2
G(z) = .
z2 + a1 z + a2
i) Write down the linear regression model of G(z).
ii) The measurements
T
YmT = y(0) y(1) . . .
   
Um = u(0) u(1) . . . u(N ) , y(N )

of input and output data are available. Construct the vectors Y ∈ RN ,

p ∈ Rnp and the matrix M such that the vector of model errors is given by
E(p) = Y − M p, where p contains the parameters to be identified.
iii) The solution p? minimizing kE(p)k22 is given by

−1
p? = M T M M T Y.

The condition number κ = σσmax min

is defined as the ratio of the largest and
smallest singular value of a matrix. Determine the condition number of the
matrix M T M as a function of κM = σσn1 , where κM is the condition number of
p
M and an SVD of M is given by
 
σ1 0
..
.
 
 
 
 0 σnp   
 T Σs
M =U V = U V T.

 .
..  0
 
 ··· 0 ··· 
 
..
.

If the condition number of M T M is poor, how could we change the identifica-

tion experiment to improve the condition number κ?

b) Assume that measured impulse response data of a SISO LTI system are given in the
form of a matrix
···
 
g(1) g(2) g(k)
...
g(2) g(k + 1) 
 
Hk =  . .. .. .
 .. . . 
g(k) g(k + 1) · · · g(2k − 1)

i) Assume H_k is in the Matlab workspace. Write a Matlab script which deter-
mines the dynamic order n of a minimal state space representation assuming
the data is noise free.
ii) Now assume the measured impulse response data is corrupted by noise. De-
scribe in words how the dynamic order can be estimated in this case.