You are on page 1of 13

Examination

Control Systems Theory and Design


21.09.2016

Problem 1 (25 Marks)

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

H4 (s)
H1 (s)
u − y
H3 (s)

H2 (s)


H9 (s) H8 (s) H7 (s) H5 (s)

H6 (s)
H(s)

Figure 1: Block diagram of a system

a) Figure 1 shows a block diagram of a SISO system with input u(t) and output y(t).
What is the transfer function H(s) from u(t) to y(t)? Simplify component fractions,
i.e. represent H(s) in the form

N(s)
H(s) = ,
D(s)

where N(s) and D(s) does not include fractions.

b) Consider a system with transfer function

5(s + 2)
H(s) = (1)
s + 2 + s(s + 2)(s + 2)

and a corresponding state space realization


   
0 1 0 0  
ẋ = 0
 0 1 x + 0 u
  y = 10 5 0 x (2)
−2 −5 −4 1
i) Is the system stable? Give reasons.
Hint: The next questions can be answered without calculations.
ii) Is the state space model controllable? Give reasons.
iii) Is this a minimal realization? Give reasons.
iv) Is the state space model observable? Give reasons.

c) Give MATLAB code for checking the minimality of the state space model in (2).
The workspace is empty.
Problem 2 (25 Marks)

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


Consider the loop in figure 2.

x̂ y
Observer F Plant

K(s)

n
Figure 2: Feedback loop

The plant is given by the state space model

ẋ = Ax + Bu (3)
y = Cx,

where A ∈ R3×3 , B ∈ R3×2 , C ∈ R1×3

a) An observer-based state feedback controller is to be designed.

i) First assume that the state x is measurable. Consider the problem of finding
a state feedback gain F that minimizes the cost function
Z ∞
J= x⊤ Qx + u⊤ Ru dt
0

What are the sizes of matrices Q and R? Explain your answer.


ii) Now consider the problem of designing an observer. Assume the output is
corrupted by a zero-mean, white noise signal n. Two designs are presented;
the weighting matrices for each design are given below

design 1 design 2
Qe = B ∗ B ⊤ Qe = 100 ∗ B ∗ B ⊤
Re = I Re = I

where I is the identity matrix of appropriate dimensions. Which observer


design corresponds to which response plot in figure 3, when a step disturbance
d = σ(t) is applied? Explain your answer.
3

A)
B)
2
y(t)

0 2 4 6 8 10

Figure 3: Output y for different observer designs

iii) Why does the loop not bring the mean value of the output back to zero when
a step disturbance is applied?
iv) How would you modify the controller to bring the steady state mean value to
zero when a step disturbance is applied?

b) i) Derive a state space model of the controller K.


ii) Write the equation governing the estimation error dynamics, i.e. the equation
˙ Recall that x̃ = x − x̂.
of x̃.
iii) Derive a state space model from disturbance input d to output y. Use the state
 ⊤
vector x⊤ x̃⊤ .

c) Write MATLAB code to compute a vector containing the values of x̃ as t → ∞


 ⊤
when d = 0 σ(t) . Assume matrices A cl and B cl of the closed-loop system are
already in the workspace.
Problem 3 (25 Marks)

Given the following 10th order model

ẋ = Ax + Bu (4)
y = Cx + Du,

where
 
−6 9.9 −1 8 −1.5 −0.7 −7 0 0 0
 0 −8.0 7 14 1.6 0 −5.5 57 0 0
 
0 0 −2.8 2.0 2.2 −5.4 5.6 6.3 0 0
 
0 0 0 −2.1 0 0 6.2 6.9 0 0
 
0 0 0 0 −5 6.1 80 −7.5 0 0
A= 0

 0 0 0 0 −4.2 4.9 2.6 0 0
0 0 0 0 0 0 −3.0 3.2 0 0
 
0 0 0 0 0 0 0 −3.8 0 0
 
0 0 0 0 0 0 0 0 .25 3 
0 0 0 0 0 0 0 0 0 .05
 ⊤
B = −0.17 0 0.08 0.026 0 0.008 0.11 −0.13 −0.19 −0.04
 
C = 1.14 1 1 0 0 1 0.17 0.32 0.12 0.5
D = 0.5

Bode Diagram
5 Frequency (ωd): 0.761
10th order
Damping (ζ): 0.873
Gred1
Magnitude (dB)

0 Gred2

-5

10 -3 10 -2 10 -1 10 0 10 1 10 2
Frequency (rad/s)
Figure 4: Bode magnitude plot of original and reduced open-loop models

a) i) The model is to be reduced by balanced truncation or residualization to a 4th


order model. Why can these reduction techniques not be applied to the system
as it is?
ii) Describe, step by step, what needs to be done to be able to perform balanced
reduction and indicate how this is done for the given system.
iii) Figure 4 shows the Bode magnitude plot for the full system and two possible
reduced systems. Which model reduction technique was used for which reduced
model? How can you recognize the different techniques?
b) An LQG controller is to be designed for the reduced order model. Where would the
poles of Ared2 + Bred2 F (i.e. the closed-loop poles of the reduced 4th order model
Gred2 under state feedback) approximately be if we let R → ∞, where R is a weight
in the cost function Z ∞
J= x⊤ Qx + u⊤ Ru dt. (5)
0

Hint1: Use the values from the Bode plot and remember that poles of a second order
system can be approximated by s1,2 = −σ ± jωd where ωd is the damped p frequency
and σ = ωn ζ. The value of ωn can also be approximated from ωd = ωn 1 − ζ 2
Hint2: Even though the eigenvalues of the full-order model are all real, the reduction
procedure might introduce complex poles to approximate the behavior of the system.

c) The LQG controller is now to be implemented on the real system. In simulation


using the full-order model, you notice that the response looks like the dashed line
in Figure 5. However, when implementing the controller in the real plant (subject
to physical limitations) the response looks like the solid line in figure 5.

i) What happens to the control input u(t) when the controller is implemented?
ii) What is the observed implication of the effect described in (c.i) on the closed-
loop response?
iii) How can you change your tuning, regarding the weights in the cost function
(5) in order to try to avoid the problem described in (c.ii)?

10
y(t)

-5

0
u(t)

-5

-10
0 5 10 15
time [s]
Figure 5: Response of simulation (dashed) and experiment (solid)
Problem 4 (25 Marks)

Hint: Tasks (a) and (b) can be solved independently.

−2

u(k) x1 (k) x2 (k) x3 (k)


z −1 z −1 z −1
− −

−0.5 0.1 −1

y(k)
2

Figure 6: Block diagram of a system


a) A discrete-time state-space model is given in Figure 6.
 T
i) Derive the state space model with states x1 x2 x3 .
ii) Is the system stable? Give reasons.
iii) Is the system controllable? Give reasons.
Hint: This question can be answered without calculation.
iv) Compute the state vectors for the time instants k = 1, 2 for the inputs given
 T
in Table 1 and initial states x(0) = 1 2 4 .

k 0 1

u(k) 3 −4

Table 1: Input u(k)


v) Compute the outputs y(0), y(1) and y(2) for the input and the initial states
given in a)-iii).
b) i) Consider the following MATLAB code for generating different transfer func-
tions.

1 G1 = zpk(-.2, .5, 5, 1);


2 G2 = tf([0, 0, .2],[1 0],1);
3 G3 = c2d(tf([1], [1 0]), 1, 'zoh');
4 G4 = tf([0, 0.2],[1 0 .5], 1);
5 G5 = tf(ss([0 1; -.5 1], [0;1], [-.2 1], 0, 1));
6 G6 = d2d(tf([.2 -.2], [1 2 1],.2), 1, 'zoh');
Which transfer function corresponds to which line in the MATLAB code? Give
reasons.
0.2 z−1 1
Ga = Gb = Gc =
z (z + 1)2 z−1
5z + 1 0.2 z − 0.2
Gd = Ge = Gf =
z − 0.5 z2 + 0.5 z2 − z + 0.5

ii) Consider the transfer functions Ga , . . . , Gf from b)-i). Find the corresponding
impulse responses (I) - (VI) in Figure 7. Give reasons.

(I) (II)
0.3
0.2
0.2 0.1
y(k)

y(k)
0.1 0
−0.1
0
0 1 2 3 4 5 0 1 2 3 4 5
k k

(III) (IV)
1.5 10

1 5
y(k)

y(k)

0
0.5
−5
0
0 1 2 3 4 5 0 1 2 3 4 5
k k

(V) (VI)
1 5

0.5 3
y(k)

y(k)

0 1

−1
0 1 2 3 4 5 0 1 2 3 4 5
k k
Figure 7: Impulse responses
Solution
Control Systems Theory and Design
27.02.2015

Ohne Gewähr No guarantee

Solution to Problem 1
a)

(H1 + H2 )(H3 − H4 )(1 + H8 H9 )(1 − H5 H6 )


H(s) =
(1 + H8 H9 )(1 − H5 H6 ) − (H1 + H2 )(H3 − H4 )H5 H7 H8 H9

b) i) The system is stable, all poles in the LHP (−2, −1, −1).
ii) The state space representation is in controller canonical form, all states can be
controlled by the input u.
iii) The realisation is not minimal, there is a pole-/zero-cancellation for s = −2.
iv) From ii) and iii) it follows that the system is not observable.

c)

1 A = [ 0 1 0 ; 0 0 1 ; −2 −5 −4];
2 B = [ 0 ; 0; 1 ] ; C = [10 5 0 ] ; D = 0;
3 sys = ss ( A , B , C , D ) ;
4 sysr = minreal ( sys ) ;
5 isequal ( sys , sysr )
6 % i f ans = 0 −−> s y s i s not minimal

1
Solution to Problem 2
a) i) Q ∈ R3×3 because the system has 3 states, and Q penalizes the states.
R ∈ R2×2 because the system has 2 inputs, and R penalizes each input.
ii) Design 1 corresponds to A) (dashed plot) and design 2 to the B) (solid plot).
In design 2, the measured output covariance is comparatively smaller than the
process noise covariance, leading to a controller with higher bandwidth (i.e.
faster poles) with little output noise attenuation.
iii) Input disturbance rejection requires integral action in the controller, which is
not present in this case.
iv) The controller has to be augmented with an integrator, whose
 input
 is the
tracking error. The augmented state feedback gain Faug = F FI is cal-
culated using the
 usual procedure,
  but considering the agumented matrices
A 0 B
Aaug = , Baug =
−C 0 0

b) i)
x̂˙ = Ax̂ + Bu + L(ŷ − y − n)
x̂˙ = Ax̂ + BF x̂ + L(C x̂ − y − n)

x̂˙ = (A + BF + LC)x̂ − Ly − Ln
u = F x̂

ii) Using x̃ = x − x̂ → x̃˙ = ẋ − x̂˙

ẋ = Ax + B(u + d)
ẋ = Ax + BF x̂ + Bd
− x̂˙ = (A + BF + LC)x̂ + L(−Cx − n)
x̃˙ = A(x − x̂) + LC(x − x̂) + Bd − Ln

x̃˙ = (A + LC)x̃ + Bd + Ln

iii) From equation of the plant, ẋ = Ax + BF x̂ + Bd, substitute x̂ = x − x̃ to get


ẋ = (A + BF )x − BF x̃ + Bd.

      
ẋ A + BF −BF x B
˙x̃ = 0 A + LC x̃
+
B
d
 
  x
y= C 0

c) C cl = [zeros(3) eye(3)]
sys = ss(A cl,B cl,C cl,0)
ssgain = dcgain(sys)
x tilde = dcg(:,2)

2
Solution to Problem 3
a) i) From the triangular structure one can see that the system is unstable, to do
balanced truncation/residualization a balanced realization is needed, however,
balanced realization is not defined for unstable systems.
ii) 1. The system must be decomposed in stable and antistable part, such that
G = Gstab + Gastab .
In this case:
 
−6 9.9 −1 8 −1.5 −0.7 −7 0
 0 −8.0 7 14 1.6 0 −5.5 57 
 
0
 0 −2.8 2.0 2.2 −5.4 5.6 6.3 
0 0 0 −2.1 0 0 6.2 6.9 
Astab =  
0
 0 0 0 −5 6.1 80 −7.5 

0
 0 0 0 0 −4.2 4.9 2.6 
0 0 0 0 0 0 −3.0 3.2 
0 0 0 0 0 0 0 −3.8
 >
Bstab = −0.17 0 0.08 0.026 0 0.008 0.11 −0.13
 
Cstab = 1.14 1 1 0 0 1 0.17 0.32
Dstab = 0.5
and
 
25 3
Aastab =
0 .05
 >
Bastab = −0.19 −0.04
 
Castab = 0.12 0.5
Dastab = 0
2. A balanced realization for Gstab is computed and reduced by truncation/resid-
ualization, resulting in reduced subsystem e.g. Gtrunc . In this case, as the
order of the desired reduced system is 4, and there is 2 unstable poles, the
stable subsystem is reduced to order 2.
3. The reduced system is given by Gred = Gtrunc + Gastab
iii) Gred1 corresponds to balanced truncation and Gred2 to balanced residualiza-
tion. Balanced truncation keeps the same feedthrough value D, which means
that high frequency behavior will be the same as the original system; in the
bode plot Gred1 has the same high frequency behavior as the original system.
Balanced residualization assumes that ”unimportant” states are constant (i.e.
not dynamic), this means that low frequency behavior will be the same as the
original system; in the bode plot, Gred2 has this characteristic.
b) From the bode plot, the stable pole pair of the reduced system has ωd = 0.761 and
ζ = 0.873, it follows then that ωn ≈ 1.5603 so the poles are s1,2 ≈ −1.36 ± 0.761j.
From (a.ii) we know that the reduced system contains these poles plus the unstable
poles of the original plant, therefore the other two poles are s3 = 0.25 and s4 = 0.05.
When R → ∞, control is expensive, stable poles remain where they are and unstable
poles are mirrored with respect to the imaginary axis; for this reason the poles of
Ared1 + Bred1 F are

3
s1,2 ≈ −0.63 ± 0.789j s3 = −0.25 s4 = −0.05

c) i) Actuators have physical limits, in this case the actuator saturates so the control
input u cannot take values beyond −3.5.
ii) The system becomes unstable as the input cannot reach the value commanded
by the controller.
iii) As the controller is LQG, increasing the value of the tuning parameter R would
bring the control effort down, avoiding reaching saturation. This will in turn
make the response slower.

4
Solution to Problem 4
a) i)
   
0.5 −2 0 1
x(k + 1) = 1 −0.1 0 x(k) + 0 u(k)
  
0 0 −1 0
 
y(k) = 2 0 1 x(k)

ii) The system is marginally stable.


iii) The system is not controllable, x3 is completely decoupled from the system.
iv)
   
−0.5 −5.85
x(1) =  0.8  x(2) =  0.58 
−4 4

v)

y(0) = 6 y(1) = −5 y(2) = −7.7

b) i) G1 ⇔ Gd , G2 ⇔ Ga , G3 ⇔ Gc , G4 ⇔ Ge , G5 ⇔ Gf , G6 ⇔ Gb
ii) Ga ⇔ (I), Gb ⇔ (IV), Gc ⇔ (III), Gd ⇔ (VI), Ge ⇔ (II), Gf ⇔ (V)
⇒ e.g. by calculating the first three output values for the impulse response

You might also like