You are on page 1of 5

EEEN60114

Three hours

Mathematical formulae tables supplied by the Examinations Office

UNIVERSITY OF MANCHESTER

Faculty of Engineering and Physical Sciences


School of Electrical and Electronic Engineering

Optimal and Robust Control

Date: 4th June 2014


Time: 09:45 – 12:45

Answer all questions.

Electronic calculators may be used, provided that they cannot store text.

1 of 5
EEEN60114

Question 1
This question is about the Linear Quadratic Regulator (LQR).
(a) Consider the dynamic system
x  Ax  Bu ,
where A =1 and B = 2 . The cost function to be minimised by the optimal control
strategy u is given by

J   xT  Q
uT 
0   x
R  u 
dt ,
0 0
where Q  1 and R    0 .

(i) Find P > 0 such that AT P  PA  Q  PBR 1 BT P  0 .


(ii) Find the optimal gain K as a function of  .
(iii) Find the pole of the closed-loop system as a function of  . In particular,
discuss the cases when  approaches 0 and  , and provide an
interpretation using the cost function.
(iv) Using the Nyquist plot of K ( sI  A) 1 B , show that the optimal gain K must
be greater than 1 .
[3, 2, 5, 5 marks]

(b) An extension of the LQR problem can be developed by introducing cross terms
in the cost function. Consider the controllable dynamic system
x  Ax  Bu
and the cost function

J   xT  Q
uT  T
S   x
R  u 
dt
0 S
where Q  SR 1S T  0 and R  0 . Using the new input uˆ  u  R 1 S T x , the system
can be rewritten as
x  ( A  BR 1S T ) x  Buˆ .

(i) Show that



Q  SR 1 S T 0   x
J   xT 
uˆ T     dt .
0  0 R  uˆ 

(ii) Using the optimal control strategy for û show that the optimal gain is given
by K  R 1 ( B T P  S T ) , where P is the positive-definite solution of the
Riccati equation AT P  PA  Q  ( PB  S ) R 1 ( PB  S )T  0.
[3, 7 marks]
Total [25 marks]

2 of 5
EEEN60114

Question 2

This question is about the Kalman filter and Linear Quadratic Gaussian control.

(a) A stochastic dynamical system is given by


x  Ax  Bu  1 ,
y  Cx  2 ;

where  1 and  2 are white noise processes with

E[ 1 1T ]  Q  0, E[ 2 2T ]  R  0, E[ 1 2T ]  0 .


(i) State the optimal gain L and the Riccati equation to be solved.
Hint: You could use the data provided in Question 1 and the duality
between the Kalman filter and the LQR problem, but it is not necessary.
(ii) In practice, the weighting matrices are chosen without the information
about the noise processes. If a control engineer has good confidence in
the model of the plant, but is using a low-cost sensor, provide a sensible
selection of R when Q  BB T . Explain your answer.

[4, 3 marks]

(b) The combination of LQR control and a Kalman filter is referred to as Linear
Quadratic Guassian (LQG) control.
(i) Sketch a diagram that describes the implementation of a LQG controller
with integral action.
(ii) State the guaranteed phase margin of a LQG controller.
[4, 3 marks]

(c) The transfer function between the elevator  e (s ) and the altitude h(s ) of the
Boeing 747 aircraft can be described by
h( s ) 30( s  6)
 .
 e ( s) s( s 2  4s  13)
An LQG controller must be designed using Loop Transfer Recovery.
(i) List the steps necessary to design the controller. You may use MATLAB
commands, but this is not necessary.
(ii) The desired phase margin for the system is 50 degrees. Is this phase
margin guaranteed? Explain your answer.
(iii) Discuss whether integral action would be required to track step references.
[5, 4, 2 marks]
Total [25 marks]

3 of 5
EEEN60114

Question 3
Consider a positive feedback interconnection of an uncertain plant
P(s)  P0 (s)  (s) ,
1
where P0 ( s)  and (s)  RH  , and a controller C (s)  1 .
s  0.1s  1
2

(a) Give a precise mathematical definition of the H  norm of a MIMO system.


[4 marks]
(b) Compute the generalised plant M (s) in P(s)  Fu (M (s), (s)) , where Fu ( . , . ) is
the upper linear fractional transformation.
[4 marks]
(c) Compute the lower linear fractional transformation Fl (M (s), C (s)) .
[4 marks]
(d) Assuming  is restricted to a real scalar quantity, find all values of  such that
the feedback interconnection is internally stable.
[HINT: The equation ax 2  bx  c  0 has solutions in the open left-half plane if
and only if the real scalars a, b, c are either all strictly positive or all strictly
negative.]
[6 marks]

(e) Assuming any (s)  RH  , compute the largest  such that the feedback
interconnection is internally stable for all     .

[7 marks]

Total [25 marks]

4 of 5
EEEN60114

Question 4

K
Consider a positive feedback interconnection of an uncertain plant P( s)  ,
Ts  1
where K  [9,11] and T  [0.4,0.6] , and a controller C (s) .
(a) Compute the nominal plant P0 ( s) .
[4 marks]
3
(b) Show that W  is the least conservative constant weight W such that
8

P(s)  P0 (s)1  W(s) : (s)  RH  ,    1
for all allowable values of K and T .
[6 marks]
(c) State the “Small-Gain Theorem”.
[4 marks]
(d) Given any P0 ( s)  RH  , W (s)  RH  that also satisfies W (s)1  RH  , and a
controller C (s) , derive a test that guarantees robust stability for the all plants
within the family
 
 : P0 (s)1  W (s)(s) : (s)  RH  ,    1 .

[5 marks]
(e) Choose a controller C (s) that robustly stabilises the family of plants  .
[6 marks]

Total [25 marks]

END OF EXAMINATION PAPER

5 of 5

You might also like