Examination

Control Systems Theory and Design

04.09.2012

Problem 1 (25 Marks)

Figure 1 shows the scheme of a heating system for a swimming pool. Water is heated
in the boiler and circulated by the pump. The temperatures in the boiler Tb and in the
swimming pool Ts are assumed to be homogeneous.

Tb(t) Ts(t)
Pump
Boiler Swimming Pool

Figure 1: Scheme of a heating system

The input u(t) is the thermal power that the boiler is producing. The system can be
described by the following equations

Ṫb (t) = −5Tb (t) + 5Ts (t) + u(t)

Ṫs (t) = 2Tb (t) − 3Ts (t) + Ta (t)
T̈a (t) = −0.1Ta (t),

where Ta denotes the ambient temperature and is roughly modeled by an oscillator.

 T
a) Find the state space representation with the state vector Tb (t) Ts (t) Ta (t) Ṫa (t)
of the system depicted in Figure 1. Assume that Ts is measured (i.e. y(t) = Ts (t)).

b) Is the system observable,

i) if the temperature in the pool Ts is measured (i.e. y(t) = Ts (t)).

ii) if the ambient temperature Ta is measured (i.e. y(t) = Ta (t)).

c) Is the system controllable? If not, which states belong to the uncontrollable sub-
space?

d) What is the physical meaning of the results in b) and c)?

e) What MATLAB command could be used to

i) check if the system is stable?

ii) compute the observability matrix?

1
Problem 2 (25 Marks)

Consider a state space realization of a plant model

   
2 0 0 0 2
ẋ =  3 −1 0  x + 0 2 u
−1 0 −1 1 0
 
1 0 0
y= x
0 1 1

a) Compute the transfer function G(s) of the system.

Hint: the inverse of a 3 × 3 triangular matrix can be computed as
 −1  
a 0 0 cf 0 0
 b c 0  = 1  −bf af 0 
acf
d e f be − cd −ae ac

b) Is the given state space realization minimal? Give reasons.

c) Consider the control loop in Fig 2. The state space realization given above describes
the dynamics of the plant.

x̂ u y
- Observer F Plant

Figure 2: Block diagram of an LQG control loop

i) Write down the state space model from disturbance d to output y, in terms of
the plant model (A, B, C), F and L, where F is the state feedback gain, L the
observer gain. Define the state vector x̄ as [xT , xT − x̂T ]T , where x is the state
of the plant.
ii) Given an unit step disturbance d, calculate the static output y in terms of the
plant model (A, B, C), F and L.
iii) The cost function is chosen as
Z ∞
V = [y T (t)y(t) + uT (t)Ru(t)]dt
0

Now assume that R → ∞. Where will the optimal closed-loop plant poles be
placed?

2
Problem 3 (25 Marks)

A discrete-time state space model is given in Figure 3

0.5

-0.5

z −1 x3 (k) 0.5 z −1 x2 (k) -1 z −1 x1 (k)

u(k)

0.5
-0.5
-1
y(k)
-2

Figure 3: Discrete-time system

 T
a) Derive the state space model with states x1 x2 x3 .
b) Compute the state vectors for the time instants k = 1, 2, for the inputs given in
 T  T
Table 1 and initial states x1 (0) x2 (0) x3 (0) = 16 8 0 .

k 0 1
u(k) −8 4

Table 1: Input sequence

c) Compute the output y(0), y(1), y(2) for the input and the initial states given in b).

Now assume a continuous-time plant is to be controlled using a discrete-time controller.

One can either synthesize a continuous-time controller, which is then discretized, or do
directly a discrete-time controller design.

d) Which method should be chosen if the sampling frequency is about ten times the
bandwidth of the system (ωs ≈ 10ωb )? Give reasons.
e) Complete the m-file below to generate a discrete-time optimal regulator with the
T T
cost function Σ∞
k=0 [x (k)x(k)+u (k)Ru(k)], where R = 0.1. Compare the controlled
 T
and uncontrolled free responses for the initial state 16 8 0 . Assume the model
depicted in Figure 3 is given in the MATLAB/SIMULINK file sys.mdl.
1 [ A , B , C , D]= dlinmod ( . . . , 1 ) ;
2 K = dlqr ( A , B , . . . , . . . )
3 syss=ss ( . . . , . . . , . . . , . . . ,1);
4 syscl=feedback ( syss , K ) ;
5 . . . ( ss ( A , B , C , D , 1 ) , . . . ) ; hold on
6 . . . ( C∗ syscl , . . . , ’ r ’ ) ;

3
Problem 4 (25 Marks)

a) [State Space Identification]

Consider the following discrete-time system

x(k + 1) = Φx(k) + Γu(k)

y(k) = Cx(k)
 
y1 (k)
with the state vector x(k), a scalar input signal u(k) and two output signals y(k) = .
y2 (k)
Assume that

   
1.4 0.5 0 2
Φ= 0
 0 −1 , Γ = 0
 (1)
0 0 0.8 0
 
1 1 1
C= (2)
0 0 1

i) Is the system stabilizable? Give reasons.

ii) Is the state space realization minimal? If not, what is the order of a minimal
realization?
iii) Let k = 100, and g(k) be the impulse response of system (1)-(2) at time k.
What is the dimension of the Hankel matrix
 
g(1) g(2) g(3) · · · g(k)
 g(2) g(3) · · · · · · g(k + 1) 
 .. .. .. 
Hk =  g(3)

. . .


 . . . .
 .. .. .. ..


g(k) g(k + 1) · · · · · · g(2k − 1)

iv) Under the assumption of ideal, noise-free measurements, what is the rank of
the Hankel matrix Hk ? Give reasons.
v) Now assume that the system matrices Φ, Γ and C are unknown. Write MAT-
LAB code to compute Γ and C, when the Hankel matrix Hk is in the workspace.

b) [Transfer Function Identification]

Assume that noisy input and output data

U T = [u(0) u(1) . . . u(N)], Y T = [y(0) y(1) . . . y(N)]

of an unknown system have been measured. The system is to be modeled by the

transfer function
b1 z −1
G(z) =
1 + a1 z −1 + a2 z −2
where a1 , a2 and b1 are unknown coefficients, to be estimated by computing the
solution to the normal equation p = (M T M)−1 M T Y .

4
 i) Give expressions for the measurement matrix M and the vector p, in terms of
the measurement data and parameters a1 , a2 and b1 .
ii) If the matrix M in Rn×m (n < m) has the rank n, can the estimation problem
be solved? Give reasons.

c) [Transfer Function Identification]

Consider a transfer function with two independent variables:

b1 z1−1
G(z1 , z2 ) =
1 + a1 z1−1 + a2 z1−2 + a3 z1−1 z2−1

Let u(k) and y(k) be the input and output signals, respectively. Write down the dif-
ference equation of the system in the form of a linear regression y(k1, k2 ) = mT (k1 , k2 )p
in terms of the measurement data and parameters a1 , a2 , a3 and b1 .