EMM514 Control Engineering II

Kenyatta University
BSc Mechanical Engineering
EMM514: CONTROL ENGINEERING II
March 12, 2020

EMM514: CONTROL ENGINEERING II
Prerequisites
EMM 507 Control Engineering I
Purpose
The aim of this course is to enable the student to;
1. understand design of controllers using PIDs
2. understand design of digital controllers
3. understand the various non linearities in control systems
4. understand the concept of optimal control systems
Learning Outcomes
At the end of this course, the student should be able to;
1. Design and implement PI,PD and PID controllers
2. Design and implement digital controllers
3. Analyse non linearities in control systems
4. Perform optimization of control systems
Course Outline
Controllers: Basic control actions, automatic controllers, actuators, sensors. Design using various
Design specifications, controller configurations, P, PI, PD and PID controllers, models, character-
istics and implementation, tuning of PID controllers
State space: state variable feedback controller design, controllability, observability, eigenvalue place-
ment, observer design for linear systems
Introduction to non-Linear control systems: sources of non-linearity, mathematical description of
non-linear systems
Optimal control: Concept of optimal control, cost function, minimum time and minimum energy
strategies, Lagrange multipliers,
i
Teaching Methodology
2 hour lectures and 1 hour tutorial per week, and at least five 3-hour laboratory sessions per
semester organized on a rotational basis.
Instructional Materials/Equipment
1. Control Engineering laboratories;
2. Computer laboratory;
3. Overhead projector;
Prescribed Text Books
1. Ogata K. (1996) Modern Control Engineering, Prentice Hall, 3rd Ed.
2. Shankar Sastry, Nonlinear Systems: Analysis, Stability, and Control, Springer, 1999 (SS)
3. Hassan K. Khalil, Nonlinear Systems, 3rd Ed., Prentice Hall, 2002 (HK)
4. Benjamin C. Kuo, Digital Control Systems, Second Edition, Oxford University Press
5. M. Sami Fadali and A. Visioli, Digital Control Engineering, Academic Press, 2009.
6. Donald E. Kirk, Optimal Control Theory, Dover publications, 2004
7. Naidu D. S., Optimal Control Systems, CRC press, 2003
References
1. C. L. Phillips and H. T. Nagle, Jr., Digital Control System Analysis and Design, Third edition,
Prentice-Hall, Englewood Cliffs
2. Raymond G. Jacquot, Modern Digital Control Systems, 1st edition, Marcel Dekker, Inc
3. Gene F. (2005) Feedback Control of Dynamic Systems, Prentice Hall, 5th Ed.
4. Hans P. Geering, Optimal Control with Engineering Applications, Springer, 2007
5. Journal of Dynamic Systems, Measurement, and Control
ii
Contents
1 CONTROLLERS 1
1.1 Definitions: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Proportional (P) Control Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Integral Control Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Derivative Control Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5 Proportional and Integral (PI) Control Action . . . . . . . . . . . . . . . . . . . . . . 15
1.6 Proportional and Derivative (PD) Control Action . . . . . . . . . . . . . . . . . . . . 18
1.7 Tuning of PID Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.8 Electronic P, PI, PD and PID Controllers . . . . . . . . . . . . . . . . . . . . . . . . 26
2 STATE VARIABLE METHODS 32

2.1 State space representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.1.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2 State Equations from Transfer Functions . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.1 Controllable canonical form . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2.2 Observable canonical form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2.3 Diagonal canonical form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2.4 Jordan Canonical Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3 Similarity Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.3.1 Eigen Values and Eigen Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3.2 Repeated Eigen values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.4 Laplace Transform Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.4.1 Transfer Function from state space equations . . . . . . . . . . . . . . . . . . 49
2.4.2 Time Solution of state equations . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.4.3 State Transition Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.4.4 Controllability and Observability . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.5 State Space Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.5.1 Pole Placement Design Technique . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.5.2 State Observer Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3 NON-LINEAR CONTROL SYSTEMS 76

3.1 Equilibrium Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.2 Linearization of non linear state space model and local stability . . . . . . . . . . . . 80
3.3 Describing functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
iii
4 OPTIMAL CONTROL 96
4.1 Performance Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.2 Solution to the Optimal Control problem . . . . . . . . . . . . . . . . . . . . . . . . 98
iv
Controllers Lecture Notes by A. M. Muhia
1 CONTROLLERS
1.1 Definitions:
Control system: An arrangement of physical components connected or related in such a manner

as to command, direct, or regulate itself or another system.
A control system must have
• An input or inputs
• An output or outputs
• An arrangement to achieve this input-output combination
Control action is that quantity responsible for actuating the system to produce the output. De-
pending on whether such a control action is dependent on the status of the output, control systems
are classified as
1. Open loop systems
2. Closed looped systems
Open Loop Systems
This is a control system in which the output has no effect on the control action i.e. the control
action is totally independent of the output of the system. The output is neither measured nor fed
back for comparison with the input. To each reference input, there corresponds a fixed operating
condition and as a result the accuracy of the system depends on the calibration. Open loop control
can be used, in practice, if the relationship between the input and output is known and if there are
neither internal nor external disturbances.
A general open loop control system is shown in Figure 1 below.
Figure 1: Open Loop Control System
The reference input r(t) is applied to the controller which generates the actuating signal u(t) to
give the controlled output c(t).
Closed Loop Control Systems
This is a control system in which the control action is somehow dependent on the output of the
system. There is a comparison of the state of the output and the reference state. This property is
known as feedback and is the main difference between open loop and closed loop systems.
1
Feedback is the property of the system which permits the output to be compared with the reference
input so that appropriate control action is formed
The generalized closed loop control system is shown in Figure 2 below
Figure 2: Closed Loop Control System
r (t) - reference input

u (t) - actuating signal
e (t) - error signal
c (t) - controlled output
b (t) - feedback signal
Part of the output is fed back to the input. The comparison between the reference input r (t) and
the feedback b (t) gives the error e (t) .
When feedback signal is positive, the system is called positive feedback system and e (t) = r (t) +
b (t).
When the feedback signal is negative, the system is called negative feedback system and e (t) =
r (t) + b (t).
The error is applied to the controller which gives the actuating signal u (t). The action of the
controller will be to drive the controlled output in such a manner so that the error is reduced to
zero i.e. the feedback signal is equal to the input signal. The manner in which the automatic
controller produces the control signal is called the control action
Industrial controllers are classified on the basis of the control action as;
Linear and Non-linear Control Systems
Linear systems are systems which obey the principle of superposition and proportionality.
If an input x1 produces output y1 by proportionality theorem an input αx1 produces an output
αy1 . By superposition theorem, an input αx1 + βx2 will produce an output αy1 + βy2 .
Linear systems are systems where mathematical tools like Laplace, Fourier etc. can be used. These
systems can be analysed mathematically or graphically.
• Practical systems are not fully linear but exhibit certain non-linearities. The common exam-
ples of non-linearity are saturation, friction, deadzone, backlash etc.
• Non-linear systems exhibit peculiar behaviour
• They do not obey the law of superposition.
• Standard test signals approach cannot be used here
2
• The stability of linear systems depends on the root location and is independent of the initial
state. In non-linear systems, stability depends on root location as well as the initial condition
and type of input.
• Non-linear systems exhibit self sustained oscillations of fixed frequency and amplitude called
Limit Cycles. Linear systems do not have this feature
• Non linear systems may exhibit hysteresis while linear systems do not show this behaviour
Table 1: Linear and Non Linear Systens

SR. NO Linear Systems Non Linear Systems
1 Obey superposition theorem Do not obey superposition theorem
2 Can be analysed by standard Cannot be analysed by standard
test signals test signals
3 Stability depends only on the Stability depends on the root location,
root location initial condition and type of input
4 Do not exhibit Limit Cycles Exhibit Limit Cycles
5 Do not exhibit Hysteresis / Exhibit Hysteresis /Jump Resonance
Jump Resonance
6 Can be analysed by Laplace, Cannot be analysed by these methods
Fourier and Z-transforms
An automatic controller compares the actual value of the plant with the desired value, determines
the deviation and produces a control signal which will reduce the deviation to zero or to a small
value. The manner in which the automatic controller produces the control signal is called the
control action. Industrial controllers are classified on the basis of the control action as
1.2 Proportional (P) Control Action
The control action or signal u (t) is proportional to the error signal e (t) i.e
u (t) = kp e (t) (1)
where kp is the proportional gain constant
Figure 3: General Proportional Controller
3
The controller is an amplifier with a gain kp (the gain maybe adjustable)

Proportional control is the basis of root-locus design
Taking the Laplace transform of the equation 1
U (s) = kp E (s) (2)
U (s)
⇒ kp =
E (s)
The closed loop transfer function is given by
Y (s) kp G (s)
= (3)
R (s) 1 + kp G (s)
The characteristic equation of the closed-loop system is
1 + kp G (s) = 0 (4)
The gain kp in equation 1.4 is varied to generate the root-locus.

Example 1.1
Construct the root-locus for the first order control system shown below
Figure 4: Example 1.1
open loop transfer function
k
G (s) =
Ts
characteristic equation
1 + G (s) H (s) = 0
k
1+ =0
Ts
4
Ts + k = 0
−k
s=
T
where k is varied from zero to ∞, the locus commences at the open-loop pole s = 0 and terminates
at minus infinity on the real axis.
Example 1.2
Construct the root locus for the 2nd order control system below
k
G (s) H (s) =
s (s + 4)
s = 0, −4
open loop zeros: none

1 + G (s) H (s) = 0
k
1+ =0
s (s + 4)
s2 + 4s + k = 0
for different values of k the roots can be obtained as
5
k characteristic equation roots

0 s2 + 4s = 0 s = 0, −4
4 s2 + 4s + 4 = 0 s = −2 ± j0
8 s2 + 4s + 8 = 0 s = −2 ± j2
16 s2 + 4s + 16 = 0 s = −2 ± j3.46
The loci commences at the open-loop s = 0, −4 when k = 0. At k = 4 they branch into the complex
space
Proportional Control of type 0, first order system
Consider the system below where k and c are constants
Figure 6: Proportional Control of type 0, first order system
The closed loop transfer function of the system is given by

Y (s) kp
=
R (s) ks + c + kp
kp
⇒ Y (s) = R (s)
ks + c + kp
1
If the input is a unit step, then R (s) = the steady state value of Y (s) is given by
s
k 1
Yss = limS→0 s
ks + c + kp s
kp
= <1
c + kp
kp
Yss =
c + kp
1
=
1 + c/kp
Yssk →∞ = 1
p
6
Figure 7: Response of Proportional Control of type 0, first order system
It can be seen that Yss → 1 as kp → ∞

The step response of the closed loop system is as shown below
The difference between Yss and the ideal value is known as the steady state error ess
The steady state error can be reduced by increasing the gain kp and can be eliminated completely
by increasing kp to infinity
In practical control systems, we cannot increase kp indefinitely as this
• could make the system unstable
• may be expensive (a high gain amplifier which is expensive to design and consumes more
power)
• may not be physically possible ; Amplifier have saturation limits and infinite gain is not
practically realizable
• amplifier noise
The time constant of a gain system gives an estimate of the speed of response of a system
The open loop time constant of the system is
1
ks + c
τ = kc
for the closed loop system
k
τ=
c + kp
since k , c , and kp are positive constants,
k k
<
c + kp c
therefore increasing kp reduces the time constant and this makes the system faster.
7
Proportional Control of a type 0, second order
Consider the system shown below
Figure 8: Proportional Control of type 0, second order system
kp
G (s) = 2
s + as + b
closed loop transfer function
Y (s) G (s)
=
R (s) 1 + G (s)
kp
= 2
s + as + b + kp
Steady state value of Y (s) for a unit step input is
kp 1
Yss = lim s 2
s→0 s + as + b + kp s
kp
= <1
b + kp
Proportional Control of a type 1, second order system

The open loop transfer function is given by
kp
G (s) =
ks2 + cs
The closed loop transfer function is given by
Y (s) kp
= 2
R (s) ks + cs + kp
8
Figure 9: Proportional Control of type 1, second order system
Steady state value of Y (s) for a unit step input
kp 1
Yss = lim s
s→0 ks2 + cs + kp s
kp
= =1
kp
Hence the set point is attained

SUMMARY
In general for proportional control
• The output has a non zero steady state error to step inputs if the plant is type 0.
The steady state error can be reduced by increasing the gain kp when possible
• Increasing the gain kp reduces the time constant making the system faster
Example 1.3
Consider the system below where k = 2 , c = 3. Determine
1. The value of kp to give a maximum steady state error of 0.2 for a unit step input
2. The time constant of the resulting system
9
Solution
kp
Yss =
c + kp
ess = 1 − Yss
kp
=1−
c + kp
c
=
c + kp
3
= 0.2
3 + kp
kp = 12
k
τ=
c + kp
2
=
15
Example 1.4
Consider the system below. Given that k = 10 , c = 5 ,determine kp such that the dumping factor
of the closed loop system ξ = 0.7071 and compute the time constant τ of the resulting closed loop
system
Y (s) kp
= 2
R (s) ks + cs + kp
kp/k
= 2 c (5)
s + k s + kp/k
The standard second order system is given by
10
Y (s) ωn2
= 2 (6)
R (s) s + 2ξωn s + ωn2
where
ωn - undumped natural frequency
ξ - damping factor
comparing the characteristic equation (i) and (ii)
r
2 kp kp
ωn = ωn =
k k
c
2ξωn =
k
1 c
ξ=
2ωn k
solving for kp
s
1 k c
0.7071 = ×
2 kp k
1 k c2
0.70712 =
4 kp k 2
c2
0.70712 = but c = 5 , k = 10
4kkp
25
kp =
40 × 0.70712
= 1.25
1
Time constant τ =
ξωn
r r
kp 1.25
ωn = =
k 10
= 0.3536
1
=
0.7071 ∗ 0.3536
τ =4
11
1.3 Integral Control Action

For integral control, the control signal u (t) is given by
ˆ t
u (t) = ki e (t) dt (1)
0
where ki is the integral gain constant

converting equation 1 to frequency domain
k
U (s) = i E (s)
s
Integral Control of type 0, first order system
Consider the system shown below. It is assumed that k and c are position constants
Figure 12: Integral Control of type 0, first order system
The open loop transfer function is
ki
s (ks + c)
The closed loop transfer function is obtained as
Y (s) k
= 2 i
R (s) ks + cs + ki
ki
= k (2)
c k
s + s+ i
2
k k
The integral control action increases the type of system by 1 and the order of the system by 1
If the input is unit step input, the steady state value of Y (s) becomes
ki
k 1
Yss = lim s
s→0 c k s
s2 + s + i
k k
12
=1
The steady state error is zero

Comparing (2) with the general equation of 2nd order system
Y (s) ωn2
= 2
R (s) s + 2ξωn s + ωn2
r
ki
⇒ ωn =
k
c
⇒ 2ξωn =
k
c
ξ=
2ωn k
c
= √
2 k1 k
This implies that it is not possible to vary ωn and ξ independently. Hence the integral control action
on its own does not give the designer the flexibility of setting the values of ωn and ξ independently
since it has only one design parameter ki
SUMMARY
Integral control adds a pole at the origin and this increases the type of the system by 1
It improves the steady state performance of a system.
NB: When used with a type 0 system, the steady state error is eliminated.
Adding a pole at the origin pulls the root-locus to the right, thus lowering the stability margin of
the system. In some cases integral control could even make a system unstable.
Example 1.5
Type 1 second order system
ki
ki
G (s) = 2 = k c
s (ks + c) 2
s s+
k
13
Figure 13: Integral Control of type 1, second order system
ki
= k
c k
s + s+ i
2
k k
open loop poles
c
s2 s + =0
k
−c
s1 = 0 s2 = 0 s3 =
k
Figure 14: Root Locus for Integral Control of type 1, second order system
For these reasons integral control is normally not used on its own, but is combined with other laws.
1.4 Derivative Control Action
The control signal u (t)is given by
d
u (t) = kd e (t)
dt
where kd is the derivative gain
14
when the slope of e (t) is large at the current time , the magnitude of u (t)will increase i.e the
derivative control law provides a large corrective action in anticipation before the error gets large.
This is the main advantage of derivative control.
However it has some shortcomings.
If the error is constant, there is no corrective action taken even if the error is large i.e it cannot
bring a constant or slowly varying error to zero by itself →, allows drift.
In frequency domain
U (s) = skd E (s)
U s = sk
⇒ E(s) d
The gain increases with frequency and this has the effect of amplifying high frequency noise
The gain curve has an infinite gain at infinite frequency for an ideal derivative controller. This is
physically unrealizable
For this reasons the derivative control action is not used by itself in control systems but usually it
is combined with P or PI controls.
1.5 Proportional and Integral (PI) Control Action
The control signal u (t) is given by
ˆ t
u (t) = kp e (t) + kI e (t) dt
0
ˆ
kp t
= kp e (t) + e (t) dt
TI 0
where
kp
= kI
TI
ˆ t
1
⇒ u (t) = kp e (t) + e (t) dt (1)
TI 0
kp - proportional gain
kI - Integral gain
TI - Integral time
Taking Laplace transform of equation (1)

1
U (s) = kp 1 + E (s)
sTI
PI control gives two design parameters kp and kI or (kp and TI ).
15
k
The controller adds a pole at the origin and a zero at s = − I to the open loop transfer function
kp
The type of the system is increased by one
PI Control of type 0, first order system
Figure 15: Proportional plus Integral Control of type 0, first order system
Open loop transfer function
kp s + ki
G (s) =
ks2 + cs
Closed loop transfer function
Y (s) kp s + ki
= 2
R (s) ks + c + kp s + ki
kp s/k
+ ki/k
= (2)
(c+k p )s k
s2 + k + ki
For a unit step input
kp s/k
+ ki/k 1
Yss = lim s
s→0 ( c+k p ) s k s
s2 + k + ki
=1
Hence the set point is attained

Comparing the characteristic equation of (2) and that of general second order system, then
r
ki
ωn =
k
c + kp
ξ= √
2 kki
16
This shows that it is possible to vary ωn and ξ independently

Example 1.6
Consider the system above with k = 1 , c = 1 and time constant τ = 0.2. Determine the PI
controller gains such that ξ = 0.7071
1
τ=
ξωn
1 1
ωn = =
ξτ 0.7071 × 0.2
= 7.071
q
ki
ωn = k
k
ωn2 = 2i
k
q
ki = ωn2 k 2
= 50
c + kp
ξ= √
2 kki
p
kp = 2ξ kki − c
p
kp = 2ξ kki − c
= 2 × 0.7071 × 7.071 − 1
=9
In general PI controller increases the type of system by one and therefore improves the steady state
performance of a system
However, it adds a pole at the origin which pulls the root locus to the right , thus degrading the
transient response of the system
17
Additional pole at the origin also lowers the stability margin of the system and for this reason , it
is not used with type 1 or higher systems as such systems could become unstable.
1.6 Proportional and Derivative (PD) Control Action
The control signal u (t)is given by
d
u (t) = kp e (t) + kD e (t) (1)
dt
Taking Laplace transform of (1)
U (s) = kp E (s) + skD E (s)

= kp + skD E (s)
= kp [1 + sTD ] E (s)
where
kp TD = kD
PD Control of type 0, first order system
Consider the system below where k and c are positive constants
Figure 16: Proportional plus Derivative Control of type 0, first order system
The closed loop transfer function of the system is given by
Y (s) skD + kp
=
R (s) (kD + k) s + c + kp
For a unit step input
kp
Yss = <1
c + kp
18
The output has a non-zero steady state error .

PD control does not improve the steady state performance of a system.
In general PD control increases the damping of the system and this reduces the overshoots. The
D component of the controller also makes the system faster. Reduce settling time and improve the
transient response
PD control adds a zero to the system
pulls root-locus to the left - improves the stability margin of the system
Proportional plus Integral plus Derivative (PID) Control Action
The control signal u (t) is given by
ˆ t
d
u (t) = kp e (t) + kI e (t) dt + kD e (t) (2)
0 dt
Taking Laplace transform of (1)
k
U (s) = kp E (s) + I E (s) + skD E (s)
s

k
= E (s) kp + I + skD
s

1
= kp E (s) 1 + + sTD
sTI
2
s TI TD + sTI + 1
= kp E (s)
sTI
PID adds two zeros and a pole to the system.
It increases the system type by 1
PID controller is used so as to utilize the best properties of PI and PID controller. In particular ,
PD.
Improves the transient response by increasing damping resulting in smaller overshoots, and improv-
ing speed of response
Improves the steady state performance
However, PID controller requires more components to implement and therefore is the most expensive
Example 1.7
A PID controller is inserted in series with a system having a transfer function
10
G (s) =
(s + 1) (s + 2)
19
The system has unity feedback. Find the gain constant of the PID controller required to locate the
closed loop poles at s = −50 , s = −4 ± j5
Solution
The system is shown below
10kp s2 TI TD + sTI + 1

Y (s)
= 3
R (s) s TI + 3TI + 10kp T1 TD s2 + 2TI + 10kp TI s + 10kp
10kp
s2 TI TD + sTI + 1

TI
= 10k
s3 + 3 + 10kp TD s2 + 2 + 10kp s + T p

I
Characteristic equation of closed loop system is
10kp
s3 + 3 + 10kp TD s2 + 2 + 10kp s +

(3)
TI
For closed loop poles s = −50 , s = −4 ± j5 the characteristic polynomial is
(s − 50) (s + 4 + j5) (s + 4 − j5)
= s3 + 58s2 + 441s + 2050 = 0 (4)
Equating coefficient of equal powers of polynomial (2) and (3)

kp = 43.9
TI = 0.214
TD = 0.125
Example
1
G (s) =
(s + 1) (s + 2)
20
√
Desired poles s = −2 ± j2 3
1.7 Tuning of PID Controllers
Ziegler Nicholas Rules
Tuning is the process of selecting the controller parameters to meet given performance specifica-
tions. Ziegler and Nicholas suggested rules for tuning PID controllers based on experimental step
response or based on the value of kp that results in marginal stability when only the proportional
control action is used
Ziegler Nicholas rules are used to select PID coefficient such that the step response of the resulting
system has a quarter decay ratio, which is satisfactory for many systems.
The concept of a quarter decay ratio is illustrated in the figure below
Figure 18: Quarter Decay Ratio
A quarter decay ratio means that

b c 1
= = etc
a b 4
Ziegler and Nicholas proposed two methods
The Ultimate Cycle method
The set-up below is used

The proportional controller has an adjustable gain.
The gain of the proportional controller is increased until the plant exhibits sustained oscillations
21
Figure 19: Ultimate Cycle method
NB: If the system does not exhibit sustained oscillations, whatever the value proportional gain
may have , then this method does not apply.
The period of oscillation pu (known as ultimate period) and the gain that causes the sustained
oscillations ku (known as ultimate gain) are determined
Figure 20: Ultimate Cycle sustained oscillations
Ziegler and Nicholas proposed the following coefficients for PID control
Types of controller Optimum gain
kp TI TD
P 0.5ku - -
PI 0.45ku 0.833pu -
PD 0.6ku - 0.125pu
PID 0.6ku 0.5pu 0.125pu
ku - ultimate gain
pu - ultimate period
22
The Process Reaction method
For the process reaction method, the open loop step response of the system is obtained
Figure 21: Process Reaction method
The open loop response is shown above. It is S-shaped.
NB: If an open loop step response is not S-shaped, this method does not apply
The parameters to be determined from the step response are the delay time L and the gradient at
the steepest point.
To obtain the delay time L a tangent is drawn at the steepest point of the step response and
extended to the time t axis as shown below
Figure 22: Process Reaction tuning
The gradient at the steepest point is given by

k
R=
T
For the approach, Ziegler and Nicholas proposed the following coefficients for PID control
23
Types of controllers Optimum gain

kp TI TD
1
P - -
LR
0.9 L
PI -
LR 0.3
1.2
PID 2L 0.5L
LR
Example 1.8
Use the ultimate cycle method to determine the gain for P, PI and PID controllers for the following
plant
1 1
G (s) = = 3
(s + 1) (s + 2) (s + 3) s + 6s2 + 11s + 6
Solution
A proportional controller is first used
The closed loop transfer function for the given system with a proportional controller is
kp
s3 + 6s2 + 11s + 6 + kp
The characteristic equation is
s3 + 6s2 + 11s + 6 + kp = 0
Setting s = jω and kp = ku
(jω)3 + 6 (jω)2 + 11 (jω) + 6 + ku = 0
−jω 3 − 6ω 2 + 11jω + 6 + ku = 0
Equating imaginary parts to 0

−jω 3 + 11jω = 0
ω 2 = 11
√
ω = ± 11
2π
ultimate period pu =
ω
2π
pu = √ = 1.89
11
Equating real parts
24
−6ω 2 + 6 + ku = 0
ku = 6ω 2 − 6
= 6 × 11 − 6
= 60
If P control is to be used with the given system then the gain kp should be set to
kp = 0.5ku
= 0.5 × 60
= 30
PI control
kp = 0.45ku
= 0.45 × 60
= 27
TI = 0.8133 × pu
= 0.8133 × 1.89
= 1.57
k
kI = Tp
I
= 17.18
PD control
kp = 0.6ku = 36
TI = 0.5pu = 0.5 × 1.89
= 0.947
25
TD = 0.125 × 1.89 = 0.947
k
kI = Tp = 38.1
I
kD = TD kp = 8.52
Example
The following table gives the measured open loop response of a system to unit step input.
Use Ziegler Nicholas rules to determine the controller gains for P, PI, and PID controller for the
system
t sec 0.00 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
Amplitude 0.00 0.01 0.06 0.15 0.28 0.43 0.57 0.70 0.80 0.88 0.92 0.95 0.96
1.8 Electronic P, PI, PD and PID Controllers

The Electronic P, PI, PD and PID Controllers are based on the op-amp circuit shown in the Figure
23 below where Z1 (s) and Z2 (S)are impedances
Figure 23: Basic Op-amp Circuit
V2 (s) Z2 (s)
=−
V1 (s) Z1 (s)
Proportional controller
V2 (s) R2
= = kp
V1 (s) R1
NB: R2 could be variable , making it possible to vary kp
Integral controller
V2 (s) 1 k
= = I
V1 (s) sCR1 s
26
Figure 24: Electronic Proportional Controller
Figure 25: Electronic Integral Controller
ˆ t
1
v2 (t) = V1 (t) dt
R1 C 0
Derivative controller
V2 (s)
= sCR1 = skD
V1 (s)
or
dv1 (t)
v2 (t) = CR1
dt
If the differentiator is implemented as shown in Figure 26above, the circuit amplifies noise and the
op-amp immediately gets saturated
A practical differentiator is shown in Figure 27 below
27
Figure 26: Ideal Electronic Derivative Controller
Figure 27: Practical Electronic Derivative Controller
V2 (s) sCR2
=
V1 (s) 1 + sCR1
This differentiator has a limited gain i.e the gain levels are at same frequency, so it does not get
saturated like the ideal one
Proportional and Derivative PD controller
The circuit is as shown in Figure 28 below
28
Figure 28: Electronic Proportional plus Derivative Controller
V2 (s) R2
= [1 + sCR1 ]
V1 (s) R1
R2
⇒ kp =
R1
TD = R1 C
kD = R2 C
Proportional and Integral PD controller
Figure 29: Electronic Proportional plus Integral Controller

V2 (s) R2 1
= 1+
V1 (s) R1 sCR2
29
R2
⇒ kp =
R1
TI = R2 C
1
KI =
R1 C
Proportional plus Integral plus Derivative (PID) Control
Figure 30: Electronic Proportional plus Integral plus Derivative Controller
V2 (s) (1 + sC1 R1 ) (1 + sC2 R2 )

=
V1 (s) sC2 R1

C1 R1 + C2 R2 1
= + + sC1 R2
C 2 R1 sC2 R1
C1 R1 + C2 R2
⇒ kp =
C2 R1
1
kI =
C2 R1
kD = C1 R2
Parallel form of the PID controller
The PID controller can also be implemented as shown in Figure 31 below

The P, I and D components are connected in parallel and the output of the respective parts summed
together.
This requires more components than the previous form but it gives the designer more flexibility as
the P, I and D gains can be set independently
30
Figure 31: Parallel Proportional plus Integral plus Derivative Controller
31
State Variable Methods Lecture Notes by A. M. Muhia
2 STATE VARIABLE METHODS
2.1 State space representation
2.1.1 Introduction
State space approach is a generalized time-domain method of modelling, analysing and designing
a wide range of control systems.
The approach can deal with
1. Multiple input, multiple output (MIMO) systems or multivariable systems
2. Nonlinear and time invariant systems
3. Alternative controller design approaches
The analysis require
1. Input variables
2. Output variables
3. State variables
2.1.2 Definitions
State: State of dynamic system is the smallest set of variables (state variables) such that knowl-
edge of the variables at time t = to plus knowledge of the inputs for t ≥ to , this information
determines the behaviour of the system for any time t ≥ to .
State variables: These are variables making up the smallest set of variables that determine the
state of the dynamic system.Let a dynamic system have n variables x1 , x2 , . . .., xn to describe the
behavior of the system while the input is given fort ≥ to and the initial state at t = to is specified.
If the future state of the system is determined, then such variables are a set of state variables.
State vector: The behavior of a given system is described by n state variables and can be
considered to be n components of a vector x. Such a vector is called a state vector. A state is thus
a vector that determines uniquely the system state x (t) for anyt ≥ to once the state at t = to is
given and the input for t ≥ to is specified.
State space: The n dimensional space whose coordinate axis consists of x1 , x2 , . . .., xn axis is
called state space. Any state can be represented by a point in the state space.
Consider the dynamic system shown below
The above dynamic system has MIMO.
32
The state of the system is described by a set of first order differential equations in terms of the
state variables x1 , x2 , . . .., xn and input variables u1 , u2 , . . .., ur as
ẋ1 (t) = f1 (x1 , x2 , . . . , xn ; u1 , u2 , . . . , ur ; t)
ẋ2 (t) = f2 (x1 , x2 , . . .., xn ; u1 , u2 , . . .., ur ; t) (1)
ẋn (t) = (fn x1 , x2 , . . . , xn ; u1 , u2 , . . . , ur ; t)
The output of the system y1 (t) , y2 (t) ,. . .,ym (t) may be a function of input variables, state
variables and time. This may be described as
y1 (t) = (g1 x1 , x2 , . . . , xn ; u1 , u2 , . . . , ur ; t)
y2 (t) = g2 (x1 , x2 , . . . ., xn ; u1 , u2 , . . . ., ur ; t) (2)
ym (t) = gm (x1 , x2 , . . .., xn ; u1 , u2 , . . .., ur ; t)
Definitions of u (t) , x (t) and y (t) would be

 
x1 (t)
 x2 (t) 
 
x (t) =  . 

 state vector
 .. 
xn (t)
 
u1 (t)
u2 (t)
 
 ..  input vector
y (t) =  
 . 
ur (t)
 
y1 (t)
 y2 (t) 
 
y (t) =  . 

 output vector
 .. 
ym (t)
Using the above vectors, equations (1) and (2) become
ẋt = f (x, u, t) state equation (3)
y (t) = g (x, u, t) output equation (4)
33
Time invariant systems If the vector functions of f and g do not involve time, then the system
is said to be a time invariant system
Then
ẋt (t) = f (x, u) (5)
y (t) = g (x, u) (6)
If equations 5 and 6 are linear then
ẋt = Ax (t) + Bu (t) (7)
y (t) = Cx (t) + Du (t) (8)
where A, B, C and D are constant matrices
Time varying systems If the system is time varying, then equations 3 and 4 result to
ẋ (t) = A (t) x (t) + B (t) u (t) (9)
y (t) = C (t) x (t) + D (t) u (t) (10)
where
A (t) – state matrix
B (t) – input matrix
C (t) – output matrix
D (t) – direct transition matrix
Example 1.1
Obtain the state space representation of the system described by
d3 y d2 y dy
3
+ 6 2
+ 11 + 6y = 6u
dt dt dt
where y is the output andu is the input to the system
Solution
Knowledge of y (0) , ẏ (0)and ÿ(0) together with the input u (t) for t ≥ 0 determines completely the
future of the system. Thus if y (t) , ẏ (t)and ÿ(t)are a set of the state variables then
Defining the state variables as
x1 = y
34
x2 = ẏ
x3 = ÿ
Then
ẋ1 = ẏ = x2
x2 = ẏ = x3
x3 = ÿ = −6x1 − 11x2 − 6x3 + 6u
By use of vector matrix notation, then the 3 first order differential equations can be combined into
one as       
ẋ1 0 1 x10 0
ẋ =
 2  0 0 1  x2  + 0 u
      
ẋ3 −6 −11 −6 x3 6
The output y is given by
 
h i x1
y= 1 0 0 x2 
 
x3
Example 1.2
Write the state variable formulation of the parallel RLC network shown below
Solution
Applying KCL at node A then
i = iR + iC + iL
ˆ
v (t) dv (t) 1
I sin ωt = +C + v (t) dt
R dt L
Differentiating with respect to t and rearranging yields
35
d2 v 1 dv 1 ω
+ + v = I cos ωt
dt2 RC dt LC C
Choosing
v (t) = x1 (t)
ẋ1 (t) = x2
1 1 ω
ẋ2 (t) = − x1 − x2 + I cos ωt
LC RC C
The vector-matrix differential form of the state equation can be written as
" # " #" # " #
ẋ1 0 1 x1 0
= 1 1
+ ω
ẋ2 − LC − RC x2 C I cos ωt
and the output " #

h i x
1
v= 1 0
x2
NB
Usually in the circuit problem, the current through the inductor and voltage across the capacitor
are chosen as the state variables
2.2 State Equations from Transfer Functions
Consider a system defined by
y (n) + a1 y (n−1) + · · · + a(n−1) ẏ + an y = b0 un + b1 un−1 + · · · + bn−1 u̇ + bn u (1)
where u is the input and y is the output

The equation can be written as
Y (s) b0 sn + b1 sn−1 + · · · + bn−1 s + bn

= n (2)
U (s) s + a1 sn−1 + · · · + an−1 s + an
The state space representation of the system defined by equation (11) and (12) can be presented
in controllable canonical form, observable canonical form and diagonal canonical form
36
2.2.1 Controllable canonical form
The controllable canonical form of the state space representation is given by

     
ẋ1 0 1 0 ··· ··· 0 0 x1
 ẋ2   0 0 1 ··· ··· 0   x2  0
      
 ..   .. .. .. ..   ..   .. 
      
 .   . . . .   .  .
 . =  .  + . u (1)
      
 ..   .. .. .. ..   ..   .. 
   . . . . 
   
ẋn−1   0 0 0 ··· ··· 1  xn−1  0
      
ẋn −an −an−1 −an−2 · · · · · · −a1 xn 1
and
 
x1
 x2 
 
. 
 
h  .. 
i
y = bn − an b0 bn−1 − an−1 b0 ··· · · · b1 − a1 b0 
 .  + b0 u
 (2)
 .. 
 
xn−1 
 
xn
Example 1.3
Consider the system given by
Y (s) s+3
= 2
U (s) s + 3s + 2
Obtain the controllable canonical form of the state space representation

Solution
n = 2 , b0 = 0, b1 = 1, b2 = 3 ,a1 = 1, a2 = 2
" # " #" # " #
ẋ1 0 1 x1 0
= + u
ẋ2 −a2 −a1 x2 1
" #" # " #
0 1 x1 0
= + u
−2 −3 x2 1
" #
h i x
1
y = b2 − a2 b0 b1 − a1 b0 + b0 u
x2
" #
h i x
1
= 3 1
x2
37
2.2.2 Observable canonical form
For the transfer function
Y (s) b0 sn + b1 sn−1 + · · · + bn−1 s + bn

= n
U (s) s + a1 sn−1 + · · · + an−1 s + an
The observable canonical form of the state space representation is given by
     
ẋ1 0 0 ··· ··· 0 −an x1 bn − an b0
 ẋ2  1 0 ··· ··· 0 −an−1   x2  bn−1 − an−1 b0 
      
 ..   .. .. ..   ..   ..
      

 .  . . .  .  
    . 
u
 .  = . + (1)
  
 ..   .. .. ..   ..  
    .. 
   . .  .   . 

ẋn−1  0 0 ··· ··· 0 −a2  xn−1   b2 − a2 b0 
      
ẋn 0 0 ··· ··· 1 −a1 xn b1 − a1 b0
and
 
x1
 x2 
 
. 
 
h  .. 
i
y= 0 0 ··· ··· 0  .  + b0 u
1   (2)
 .. 
 
xn−1 
 
xn
Example 1.4
Obtain the observable canonical form of the state space representation for the system with the
transfer function
Y (s) s+3
= 2
U (s) s + 3s + 2
Solution
n = 2 , b0 = 0 ,b1 = 1 ,b2 = 3 ,a1 = 1 , a2 = 2
" # " #" # " #
ẋ1 0 −a2 x1 b2 − a2 b0
= + u
ẋ2 1 −a1 x2 b1 − a1 b0
" #" # " #
0 −2 x1 3
= + u
1 −3 x2 1
" #
h i x
1
y= 0 1 + b0 u
x2
38
" #
h i x
1
= 0 1
x2
2.2.3 Diagonal canonical form
If the denominator polynomial has distinct roots then the transfer function can be written as
Y (s) b0 sn + b1 sn−1 + · · · + bn−1 s + bn

= (1)
U (s) (s + p1 ) (s + p2 ) · · · (s + pn−1 ) (s + pn )
c1 c2 cn−1 cn
= b0 + + + + (2)
(s + p1 ) (s + p2 ) (s + pn−1 ) (s + pn )
the diagonal canonical form of the state space representation of the system is given by
     

ẋ1 −p1 0 ··· ··· 0 0 x1 1
ẋ2  0 −p1 ··· ··· 0   x2  1
      

..  .. .. .. ..   ..   .. 
      

 .  . . . .   .  .
 .  + . u
 =     (3)
   .
..
  .. .. .. ..  . .

 .
  . . .  . 
 
 .
ẋn−1   0 0 0 ··· −pn−1 0  xn−1  1
      
ẋn 0 0 −an−2 ··· ··· −pn xn 1
and
 
x1
 x2 
 
. 
 
h  .. 
i
y = c1 c2 ··· · · · cn−1  .  + b0 u
cn   (4)
 .. 
 
xn−1 
 
xn
Example1.5
Obtain the diagonal canonical form of the state space representation for the system with the transfer
function
Y (s) s+3
= 2
U (s) s + 3s + 2
Solution
s+3 s+3 c1 c2
= = +
s2 + 3s + 2 (s + 1) (s + 2) (s + 1) (s + 2)
39
b0 = 0 , c1 = 2 , c2 = −1 , p1 = 1 , p2 = 2
" # " #" # " #
ẋ1 −p1 0 x1 1
= + u
ẋ2 0 −p2 x2 1
" #" # " #
−1 0 x1 1
= + u
0 −2 x2 1
" #
h i x
1
y = c1 c2 + b0 u
x2
" #
h i x
1
= 2 −1
x2
2.2.4 Jordan Canonical Form
If the denominator polynomial involves multiple roots then the state space representation can be
written in Jordan canonical form.
For example, if the pi ‘s are different from one another except that the first three are equal i.e.
p1 = p2 = p3 , then the factored form of the transfer function becomes
Y (s) b0 sn + b1 sn−1 + · · · + bn−1 s + bn

= (1)
U (s) (s + p1 )3 (s + p4 ) · · · (s + pn )
The partial fraction expansion of the equation becomes
Y (s) c1 c2 c3 c4 cn
= b0 + 3
+ 2
+ + + ··· + (2)
U (s) (s + p1 ) (s + p1 ) (s + p1 ) (s + p4 ) (s + pn )
The state space representation of the system in Jordan canonical form becomes
 
 
ẋ1 −p1 1 0 0 0    
 x1 0
−p1

   0 1 0
 ẋ2  
   
  x2  0
   ..    
 ẋ3   0 0 −p1 0 .   x  1
  3  
 =   +  u (3)
  
 ẋ4   0 ..   x4  1
 .   −p4 .    
 .   .    
 .   .. ··· 0
   
 
ẋn xn 1
0 ··· ··· ··· ··· −pn
and
40
 
x1
 
 x2 
h i 
 .. 
y = c1 c2 ··· · · · cn  . 

 + b0 u (4)
 . 
 .. 
 
xn
The characteristic equation of a given system remains invariant under different forms of state
variable representation. This can be true for the transfer function also. The choice of the states is
not unique
Example 1.6
Consider the system
...
y + 6ÿ + 11ẏ + 6y = 6u
where y is the output and u is the input. Obtain the state space representation of the system
Solution I
Let
x1 = y
x2 = ẏ
x3 = ÿ
ẋ1 = x2
ẋ2 = x3
ẋ3 = −6x1 − 11x2 − 6x3 + 6u
These can be written in the vector matrix differential equation form as
      
ẋ1 0 1 x10 0
ẋ2  =  0 0 1  x2  + 0 u
      
ẋ3 −6 −11 −6 x3 6
 
h i x1
y= 1 0 0 x2 
 
x3
The characteristic equation becomes
|SI − A| = 0
41
s3 + 6s2 + 11s + 6 = 0
Solution II
...
y + 6ÿ + 11ẏ + 6y = 6u
The transfer function is obtained as
Y (s) 6
= 3 2
U (s) s + 6s + 11s + 6
6
=
(s + 1) (s + 2) (s + 3)
By partial fraction expansion
Y (s) 3 −6 3
= + +
U (s) (s + 1) (s + 2) (s + 3)
3 −6 3
Y (s) = U (s) + U (s) + U (s)
(s + 1) (s + 2) (s + 3)
Defining
Y (s) = X1 (s) + X2 (s) + X3 (s)
where
3
X1 (s) = U (s) ẋ1 = −x1 + 3u
(s + 1)
−6
X2 (s) = U (s) ẋ2 = −2x2 − 6u
(s + 2)
3
X1 (s) = U (s) ẋ3 = −3x3 + 3u
(s + 3)
The state space representation becomes
      
ẋ1 1 0 0 x1 3
ẋ2  = 0 −2 0  x2  + −6 u
      
ẋ3 0 0 −3 x3 3
 
h i x1
y= 1 1 1 x2 
 
x3
42
|SI − A| = 0
s3 + 6s2 + 11s + 6 = 0
2.3 Similarity Transformation

It has been stated that the choice of states is not unique for a given system. Suppose that there
exists a set of state variables
h iT
X = x1 x2 ··· xn (1)
We may take another set of state variables
h iT
Z = z1 z ··· zn (2)
So that a linear or similarity transformation exists

Let X = P Z i.e
Z = P −1 X (3)
where P is a non-singular transformational matrix

Differentiating equation (27) yields
Ż = P −1 Ẋ (4)
Using the general state space equations
Ẋ = Ax + Bu
y = Cx
Then equation (28) becomes
Ż = P −1 Ax + P −1 Bu (5)
From equation (27) X = P Z

Equation (29) becomes
Ż = P −1 AP z + P −1 Bu (6)
and
y = Cx = CP z (7)
43
Equations (30) and (31) can be written as
Ż = Âz + B̂u (8)
y = Ĉx (9)
Where Â = P −1 AP, B̂ = P −1 B and Ĉ = CP

Hence similarity transformation the transformed system can be represented in the vector-matrix
differential form as
Ż = Âz + B̂u
and the output as

y = Ĉx
NB
1. The characteristic equations and hence the Eigen values of A and Âare invariant under simi-
larity transformation
2. The transfer function remains invariant under similarity transformation
2.3.1 Eigen Values and Eigen Vectors
Given the matrix equation
Ax = λx (1)
The values of the scalar λ for which non trivial solutions exist are called Eigen values and the
corresponding solutions x = 0 are called Eigen vectors
Equation (34) can be written in the form
(λI − Ax) = 0 whereI is the identity matrix
|λI − Ax| = 0 is the characteristic equation of A
The roots of the characteristic equation are called Eigen values of the matrix A
Corresponding to each Eigen value is a non-zero solution of x = expi This is called the Eigen vector
of A corresponding to λi
Example 1.7
" #
4 1
Determine the Eigen values and Eigen vectors of Ax = λx where A =
3 2
Solution
|λI − A| = 0
44
(λ − 4) (λ − 2) − 3 = 0
λ1 = 5, λ2 = 1
For λ1 = 5 , Ax = λx becomes " #" # " #

4 1 x1 x1
=5
3 2 x2 x2
4x1 + x2 = 5x1
3x1 + 2x2 = 5x2
x1 = x2
Eigen vector corresponding toλ1 = 5 is

" # " #
x1 1
exp1 = =
x2 1
simplest form
Forλ2 = 1 , Ax = λx becomes
" #" # " #
4 1 x1 x1
=1
3 2 x2 x2
4x1 + x2 = x1
3x1 + 2x2 = x2
x1 = −3x2
Eigen vector corresponding to λ2 = 1 is

" # " #
x1 1
exp2 = =
x2 −3
simplest form
45
We obtain the transformation matrix P from the Eigen vectors as follows

" #
h i 1 1
P = exp1 exp2 =
1 −3
The matrixÂ = P −1 AP is then obtained
" #−1 " #" # " #

1 1 4 1 1 1 5 0
Â = P −1 AP = =
1 −3 3 2 1 −3 0 1
Under similarity transformation the characteristic equation does not change i.e.

|λI − A| = λI − Â = λ2 + λs + 5

2.3.2 Repeated Eigen values
In some cases, the matrix A will have repeated Eigen values. The Eigen vectors are evaluated as
follows
Example 1.8
 
3 −3 2
Determine the Eigen values and Eigen vectors of Ax = λx where A = −1 5 −2
 
−1 3 0
Solution
|λI − A| = 0

λ − 3 3 −2

1 λ−5 2 =0

1 −3 λ
λ3 − 8λ2 + 20λ − 16 = 0
λ1 = 4 λ2 = λ3 = 2
For λ1 = 4
    
3 −3 2 x1 x1
−1 5 −2 x2  = 4 x2 
    
−1 3 0 x3 x3
This yield
46
x2 = −x1
and
x3 = −x1
and the simplest form of the corresponding Eigen vector as
 
1
exp1 = −1
 
−1
For λ2 = λ3 = 2
    
3 −3 2 x1 x1
−1 5 −2 x2  = 2 x2 
    
−1 3 0 x3 x3
The simultaneous equations obtained are
x1 − 3x2 + 2x3 = 0
We let x2 = α and x3 = βwhere are α and β constants

The above equation becomes
x1 − 3α + 2β = 0
x1 = 3α − 2β
and the resulting Eigen vector as

   
3 −2
exp = α 1 + β  0 
   
0 1
 
3
For β = 0and α = 1 , exp2 = 1
 
0
 
−2
Forβ = 1 and α = 0 ,exp3 =  0 
 
47
The transformation matrix P is obtained as

 
1 3 −2
P = −1 1 0
 
−1 0 1
The matrix Â = P −1 AP is then obtained as
 −1   
1 3 −2 3 −3 2 1 3 −2
Â = P −1 AP = −1 1 0 −1 5 −2 −1 1 0
    
−1 0 1 −1 3 0 −1 0 1
 
4 0 0
= 0 2 0
 
0 0 2
Checking the characteristic equation

|λI − A| = λI − Â = λ3 − 8λ2 + 20λ − 16 = 0

2.4 Laplace Transform Technique

Recall I
The Laplace transform of a function f (t) is defined as the integral
ˆ ∞
L {f (t)} = exp−st f (t) dt (1)
0
Example 1.9
Obtain the Laplace transform of f (t) = expat
Solution
ˆ ∞
L {f (t)} = exp−st expat dt
0
1
=
s−a
Recall II
n 0 o
L f (t) = sL {f (t)} − f (0) (2)
0
n o
L f ” (t) = s2 L {f (t)} − sf (0) − f (0) (3)
48
and so on for Laplace transform of higher derivatives(you should be able to prove this with a lot of
ease)
Example 1.10
Use the Laplace transform of second derivative to derive
s
L {cos (at)} = 2
s + a2
Solution
Let
f (t) = cos (at)
0
f (t) = −a sin at
and
f ” (t) = −a2 cos (at)
0
f (0) = 0
f (0) = 1
Using equation (37)
0
n o
L f ” (t) = s2 L {f (t)} − sf (0) − f (0)
n o
L −a2 cos (at) = s2 L {cos (at)} − s − 0

s2 + a2 L {cos (at)} = s
s
L {cos (at)} = 2
s + a2
2.4.1 Transfer Function from state space equations

Y (s)
The transfer function is given by G (s) = U (s)
The general state space representation of any given systems is
ẋ = Ax + Bu (1)
49
y = Cx + Du (2)
Taking the Laplace transforms of (38) and (39)
sX (s) − X (0) = AX (s) + BU (s) (3)
Y (s) = CX (s) + DU (s) (4)
Assuming that the initial conditions of the system are zero i.e. X (0) = 0 and rearranging (40) then
X (s) = (SI − A)−1 BU (s) (5)
Substituting (42) into (41) yields

Y (s) = C (SI − A)−1 B + D U (s) (6)
and the transfer function is obtained as
Y (s)
G (s) = = C (SI − A)−1 B + D (7)
U (s)
Example 1.11
Obtain the transfer function of the system whose state space representation is given as
" #" # " #
0 1 x1 0
ẋ = + u
−2 −3 x2 1
" #
h i x
1
y= 1 0
x2
Solution
Y (s)
G (s) = = C (SI − A)−1 B + D
U (s)
where
" # " #
0 1 0 h i
A= , B= , C= 1 0 and D = 0
−2 −3 1
" #
−1 1 s+3 1
(SI − A) = 3
s + 3s + 2 −2 s
50
" #" #
1 h i s+3 1 0
C (SI − A)−1 B + D = 3 1 0
s + 3s + 2 −2 s 1
1
= 3
s + 3s + 2
2.4.2 Time Solution of state equations
Consider the homogenous case

ẋ (t) = ax (t) (1)
Approach A
Assume a solution of the form
x (t) = b0 + b1 t + b2 t2 + · · · + bk tk (2)
Substituting (46) into (45) yields
h i
ẋ = b1 + 2b2 t + · · · + kbk tk−1 = ab0 + b1 t + b2 t2 + · · · + bk tk (3)
If the assumed solution is to be a true solution, equation (47) must hold true for any value of t
This implies that
b1 = ab0
1 2
b2 = a b0
2
1 k
bk = a b0
k!
The value of b0 is determined by substituting t = 0 into equation (46) i.e.
X (0) = b0
The solution of x (t) can be written as

1 2 2 1 k k
x (t) = 1 + at + a t + · · · + a t x (0)
2 k!
But
1 1
expat = 1 + at + a2 t2 + · · · + ak tk
2 k!
Therefore
51
x (t) = expat x (0) (4)
Approach B
Using Laplace transform of the homogeneous equation (45)
sX (s) − X (0) = aX (s)
1
X (s) = X (0)
s−a
Taking the inverse Laplace transform
x (t) = expat x (0)
2.4.3 State Transition Matrix
The approach to the solution of homogeneous scalar equation can be extended to the solution of
homogeneous state equation
ẋ (t) = Ax (t) (1)
Taking Laplace transform
sX (s) − X (0) = AX (s)
X (s) = (sI − A)−1 X (0)
Taking the inverse Laplace transform
x (t) = L−1 (sI − A)−1 x (0) (2)
x (t) = expAt x (0) (3)
Where
expAt = L−1 (sI − A)−1
expAt is called the state transition matrix and contains all information about the free motions of
the system described by (49)
Example 1.12
52
Obtain the state transition matrix of the following state equation

" # " #" #
ẋ1 0 1 x1
=
ẋ2 −2 −3 x2
Solution
expAt = L−1 (sI − A)−1
" #
1 s+3 1
(sI − A)−1 = 3
s + 3s + 2 −2 s
" #
1 s+3 1
=
(s + 2) (s + 1) −2 s
s+3 1
" # " 2 1 1 1
#
(s+2)(s+1) (s+2)(s+1) (s+1)
− s+2 (s+1)
− s+2
= −2 s
= −2 2 −1 2
(s+2)(s+1) (s+2)(s+1) (s+1)
+ s+2 (s+1)
− s+2
" 2 1 1 1
#
At −1 −1 −1 (s+1)
− s+2 (s+1)
− s+2
exp =L (sI − A) =L −2 2 −1 2
(s+1)
+ s+2 (s+1)
− s+2
" #
2 exp−t − exp−2t exp−t − exp−2t
=
−2 exp−t +2 exp−2t − exp−t +2 exp−2t
Non Homogeneous state equations
Consider the non-homogeneous state equation
ẋ (t) = ax (t) + Bu (t) (4)
Multiplying equation (52) by exp−at both sides
exp−at ẋ (t) = exp−at ax (t) + exp−at Bu (t)
exp−at [ẋ (t) − ax (t)] = exp−at Bu (t)
d
exp−at x (t) = exp−at Bu (t)

(5)
dt
Integrating equation (53) between0 and t results in
53
ˆ t
exp−at x (t) = exp−at Bu (t) dt + x (0)
0
ˆ t
at
x (t) = exp x (0) + exp at
exp−aτ Bu (τ ) dτ (6)
0
From equation (54)
• The first term on the right hand side is the response to the initial conditions
• The second term is the response to the input u (t)
Extending the same approach to the solution of homogeneous state equation yields
ˆ t
x (t) = expAt x (0) + expA(t−τ ) Bu (τ ) dτ (7)
0
The solution of x (t) is the sum of a term consisting of the transition of the initial state and a term
arising from the input vector
Example 1.13
Obtain the time response of the following system
" # " #" # " #
ẋ1 0 1 x1 0
= + u
ẋ2 −2 −3 x2 1
where u (t) is a unit step input

Solution
ˆ t
At
x (t) = exp x (0) + expA(t−τ ) Bu (τ ) dτ
0
From the previous example

" #
2 exp−t − exp−2t exp−t −2 exp−2t
expAt =
−2 exp−t +2 exp−2t − exp−t +2 exp−2t
" #
2 exp−(t−τ ) − exp−2(t−τ ) exp−(t−τ ) −2 exp−2(t−τ )
expA(t−τ ) =
−2 exp−(t−τ ) +2 exp−2(t−τ ) − exp−(t−τ ) +2 exp−2(t−τ )
" #
0 h i
expA(t−τ ) Bu (τ ) = expA(t−τ ) 1
1
" #
exp−(t−τ ) −2 exp−2(t−τ )
=
− exp−(t−τ ) +2 exp−2(t−τ )
54
ˆ t ˆ t
exp−(t−τ ) −2 exp−2(t−τ )
expA(t−τ ) Bu (τ ) dτ = dτ
0 0 − exp−(t−τ ) +2 exp−2(t−τ )
" #
1
2 − exp−t + 12 exp−2t
=
exp−t − exp−2t
" #
1 − exp−t + 12 exp−2t
x (t) = expAt x (0) + 2
exp−t − exp−2t
" #" # " #
2 exp−t − exp−2t exp−t −2 exp−2t x1 (0) 1
2 − exp−t + 12 exp−2t
= +
−2 exp−t +2 exp−2t − exp−t +2 exp−2t x2 (0) exp−t − exp−2t
2.4.4 Controllability and Observability
These tells us whether it is at all possible to control all the states of the system completely by
suitable choice of input and whether it is possible to reconstruct the states of a system from its
input and outputs
1. Controllability
For the linear time invariant system
ẋ (t) = Ax (t) + Bu (t)
y (t) = Cx (t) (1)
The system is said to be controllable if it is possible to find some input u (t) that will transfer
the initial state of the system x (0) to the origin of the state space, x (t0 ) = 0 with t0 finite.
The solution of the state equation yields
ˆ t
x (t) = Φ (t) x (0) + Φ (t − τ ) Bu (τ ) dτ (2)
0
where Φ (t) = expAt

For the system to be controllable
ˆ t
x (0) = Φ (t0 ) x (0) + Φ (t0 − τ ) Bu (τ ) dτ = 0 (3)
0
with finite t0
A linear time invariant continuous time system is completely controllable iff the RANK of
the controllability matrix M is equal ton
55
h i
M= B AB A2 B . . . An−1 B (4)
The rank of a matrix A is the maximum number of linearly independent columns of A; that
is, it is the order of the largest non singular matrix contained in A. This implies that the
controllability matrix M must be non singular for the system to be completely controllable.
If a system is not completely controllable, it implies that it has one or more natural modes
that cannot be affected by the input directly or indirectly.
Example 1.14
Determine whether the system represented by the given state space is controllable
" # " #" # " #

ẋ1 0.5 0 x1 0
= + u (t)
ẋ2 0 −2 x2 1
Solution
The controllability matrix is given by
h i
M= A AB
" #" # " #

0.5 0 0 0
AB = =
0 −2 1 −2
" #
0 0
M=
1 −2

0 0
M = =0

1 −2
The matrix is singular hence the system is uncontrollable

This is more obvious if we write the two differential equations separately as
ẋ1 = 0.5x1
ẋ2 = −2x2 + u (t)
It is evident that whereas x2 can be changed byu (t) the state x1 is unaffected by our choice
of the inputs since it is not coupled either directly to the input or to the state x2 hence this
state of x1 (0) exp−0.5t is uncontrollable
On the other hand if we had
56
ẋ1 = 0.5x1 + x2
ẋ2 = −2x2 + u (t)
The controllability matrix is obtained as
" #
0 1
M=
1 −2

0 1
M = = −1

1 −2
The matrix is nonsingular hence the system is controllable.

x1 can be controlled indirectly through x2
2. Observability
The linear time invariant system is said to be observable if the initial condition x (0)can be
determined from the output function y (t) for 0 < t < t1 where t1 is finite
y (t) = Cx (t)
ˆ t
= CΦ (t) x (0) + C Φ (t − τ ) Bu (τ ) dτ (5)
0
Thus given u (t) and y (t)for 0 < t < t1 with t1 being some finite value, the system is observable
if equation (60) can be solved forx (0)
The system is observable if the observability matrix N is nonsingular i.e. the rank of N is
equal ton

C

CA

N = .

..

CAn−1
Example 1.15
Consider the system represented by
" # " #" # " #

ẋ1 0.5 0 x1 0
= + u (t)
ẋ2 0 −2 x2 1
57
" #
h i x
1
y (t) = 0 1
x2
Determine whether the system is observable

Solution
The observability matrix is given by
" #
C
N=
CA
" # " #
h i 0.5 0 0 1
CA = 0 1 =
0 −2 0 −2

0 1
N = =0

0 −2
The matrix is singular and therefore the system is unobservable

The state x1 does not affect the output nor does it affect the state x2 which is coupled to the
output
58
2.5 State Space Design
2.5.1 Pole Placement Design Technique
State space design enables the design of a system having the desired closed loop poles or desired
It also enables inclusion of initial conditions if necessary
Pole placement design is based on the state model of the system. We assume that all the state
variables are measurable and are available for feedback
State model equations
ẋ (t) = Ax (t) + Bu (t) (1)
Figure 32: General control system
The plant input u (t) is made a function of the states of the form
u (t) = f (x (t)) (2)
Equation (62) is called the control rule or control law. In pole placement design, the control law is
specified as a linear function of the states of the form
u (t) = −kx (t) (3)
Figure 33: Pole placement design
59
This control law allows the poles of the closed system to be placed in any desirable location and is
expressed as
u (t) = −k1 x1 (t) − k2 x2 (t) − . . . − kn xn (t) (4)
The design problem is the specification of the desired root locations of the systems characteristic
equations and the calculations of the gains ki to yield these desired root locations.
A necessary and sufficient condition that the closed-loop poles can be placed at any arbitrary
location in the s-plane is that the system must be completely state controllable.
Determination of the MatrixK
1. Using Direct substitution method

If the system is of low order, direct substitution of matrixK into the desired characteristic
polynomial may be simpler.
e.g. if n = 3 and the desired poles are µ1 , µ2 and µ3 then
h i
K = k1 k2 k3 (5)
Desired characteristic polynomial
(s − µ1 ) (s − µ2 ) (s − µ3 )
This is also obtained as
|[sI − A + BK]|
We equate
(s − µ1 ) (s − µ2 ) (s − µ3 ) = |[sI − A + BK]|
to obtain the values of ki

Example 1.16
Consider the system
ẋ (t) = Ax (t) + Bu (t)
where  
0 1 0
A= 0 0 1
 
−1 −5 −6
60
 
0
B = 0
 
The system uses the state feedback control law u (t) = −kx (t). It is desired to have closed
loop poles at s = −2 ± j4 , s = −10.
Determine the state feedback gain matrix K
Solution
We first check for controllability of the system
h i
M= B AB A2 B

0 0 1

M = 0 1 −6 = −1

1 −6 31
The matrix is non singular hence the system is completely state controllable
Next we solve forK
Let h i
K = k1 k2 k3
|[sI − A + BK]|
     
s 0 0 0 1 0 0 h
i
= 0 s 0 −  0 0 1  + 0 k1 k2 k3
     

0 0 s −1 −5 −6 1
= s3 + (6 + k3 ) s2 + (5 + k2 ) s + (1 + k1 ) .........(i)
The desired characteristic equation is
(s + 2 − j4) (s + 2 + j4) (s + 10)
= s3 + 14s2 + 60s + 200.........(ii)
Comparing equations (i) and (ii)
61
k3 = 8 k2 = 55 k1 = 199
h i
K = 199 55 8
u = −199x1 − 55x2 − 8x3
2. Using Ackermann’s Formula

Cayley Hamilton Theorem
Every square matrix satisfies its own characteristic equation i.e. if the characteristic equation
of the nth order square matrix is
|λI − A| = 0
|λI − A| = λn + α1 λn−1 + α2 λn−2 + · · · + αn−1 λ + αn I = 0 (6)
Then
An + α1 An−1 + α2 An−2 + · · · + αn−1 A + αn I = 0 (7)
To obtain the inverse of the matrix A we divide equation byA on both sides
An−1 + α1 An−2 + α2 An−3 + · · · + αn−1 I + αn A−1 = 0
−1 h n−1 i
A−1 = A + α1 An−2 + α2 An−3 + · · · + αn−1 I (8)
αn
Example 1.17
Determine the inverse of the matrix
 
1 2 7
A = 4 2 3
 
1 2 1
using Cayley Hamilton Theorem
Solution
|λI − A| = 0
62

λ − 1 −2 −7

−4 λ−2 −3 = 0

−1 −2 λ − 1
λ3 − 4λ2 − 20λ − 35 = 0
−1 h 2 i
A−1 = A − 4A − 20I
35
 
−4 11 −5
−1 
= −1 −6 25 

35
6 1 −10
Consider the state equation

ẋ = Ax + Bu (9)
Where the state feedback control law
u (t) = −kx (t)
We assume that the system is completely state controllable and that the desired closed loop
poles are at
s = µ1, s = µ2 . . . . . . s = µn
Equation(69) becomes
ẋ = [A − Bk] x (10)
Defining Â = A − Bk
Then the desired characteristic equation is

sI − Â = 0

(s − µ1 ) (s − µ2 ) . . . . . . (s − µn ) = 0
sn + α1 sn−1 + α2 sn−2 + · · · + αn−1 s + αn = 0
Use of Cayley Hamilton theorem which states that Â satisfies its own characteristic equation
we obtain
63

Φ Â = Ân + α1 Ân−1 + α2 Ân−2 + · · · + αn−1 Â + αn I = 0 (11)
To derive Ackermann’s formula, we consider the case when n = 3

Considering that
I=I (12)
Â = A − Bk (13)
Â2 = [A − Bk]2
= A2 − 2ABK − B 2 K 2
= A2 − ABK − BK Â (14)
Â3 = [A − Bk]2 [A − BK]
h i
= A2 − ABK − BK Â [A − BK]
= A3 − A2 BK − ABK Â − BK Â2 (15)
Multiplying both sides of equations (72), (73), (74) and (75)with, α3 ,α2 α1 and α0 where
α0 = 0
α3 I = α3 I
α2 Â = α2 [A − Bk]
h i
α1 Â2 = α1 A2 − ABK − BK Â
Â3 = A3 − A2 BK − ABK Â − BK Â2
Adding the two sides of the equations
64
Â3 +α1 Â2 +α2 Â+α3 I = A3 +α1 A2 +α2 A+α3 I−α2 BK−α1 ABK−α1 BK Â−A2 BK−ABK Â−BK Â2
(16)
Since

Â3 + α1 Â2 + α3 I = Φ Â = 0
And
A3 + α1 A2 + α2 A + α3 I = Φ (A) 6= 0
then
Φ (A) − α2 BK − α1 ABK − α1 BK Â − A2 BK − ABK Â − BK Â2 = 0 (17)
h i h i
Φ (A) = B α2 K − α1 K Â − K Â2 + AB α1 K − K Â + A2 BK
2
 
h i α2 K − α1 K Â − K Â
Φ (A) = B AB A2 B  α1 K − K Â (18)
 

K
α2 K − α1 K Â − K Â2
 
h i−1
B AB A2 B Φ (A) =  α1 K − K Â (19)
 

K
h i
Premultiplying both sides of (79) by 0 0 1 and rearranging we obtain
h ih i−1
K= 0 0 1 B AB A2 B Φ (A) (20)
For an arbitrary positive integer n, then
h ih i−1
K= 0 0 ... 1 B AB ... An−1 B Φ (A) (21)
Example 1.18
Consider the system
ẋ (t) = Ax (t) + Bu (t)
65
where  
0 1 0
A= 0 0 1
 
−1 −5 −6
 
0
B = 0
 
loop poles at
s = −2 ± j4 , s = −10.
Determine the state feedback gain matrix K using Ackermann’s formular
Solution
The desired characteristic equation is given by
(s + 2 − j4) (s + 2 + j4) (s + 10) = s3 + 14s2 + 60s + 200
Φ (A) = A3 + 14A2 + 60A + 200I
where
 
0 1 0
A= 0 0 1
 
−1 −5 −6
 
199 55 8
Φ (A) =  −8 159 7 
 
−7 −43 117
 
h 0 i 0 1
B AB 2
A B = 0

1 −6

1 −6 31
h ih i−1
K= 0 0 1 B AB A2 B Φ (A)
  
h i 0 0 1 199 55 8
K= 0 0 1 0 1 −6  −8 159 7 
  
1 −6 31 −7 −43 117
66
h i
K = 199 55 8
3. Using Transformation Matrix T

Suppose that the system defined by
ẋ = Ax + Bu
and the control law is given by u (t) = −kx (t)

The matrix K can be obtained as follows
(a) Check the controllability condition for the system. If the system is completely control-
lable then
(b) Determine the characteristic polynomial of the matrix A
|sI − A| = sn + a1 sn−1 + a2 sn−2 + · · · + an−1 s + an
Determine the values ofa1 , a2 . . . an

(c) Determine the transformation matrix T that can transform the system state equation in
controllable canonical form. If the system is already in controllable canonical form then
T =I
Else
T = MW
Where M is the controllability matrix and

 
an−1 an−2 . . . a1 1
an−2 an−3 . . . 1 0
 
 
 . . . .
 
W = . . . .
 
 
 . . . .
 
 a1 1 . . . 0 0
 
1 −an−1 . . . 0 0
(d) Write the desired characteristic polynomial
(s − µ1 ) (s − µ2 ) . . . . . . (s − µn ) = sn + α1 sn−1 + α2 sn−2 + · · · + αn−1 s + αn
Determine the values of α1 , α2 . . . αn

(e) Obtain the matrix K as
h i
K = αn − an αn−1 − an−1 α1 − a1 T −1 (22)
67
Example 1.19
Consider the system
ẋ (t) = Ax (t) + Bu (t)
where  
0 1 0
A= 0 0 1
 
−1 −5 −6
 
0
B = 0
 
loop poles at
s = −2 ± j4 , s = −10.
Determine the state feedback gain matrix K using transformation matrix T
Solution
Check the system for controllability
h i
M= B AB A2 B

0 0 1

M = 0 1 −6 = −1

1 −6 31
The matrix is nonsingular hence the system is completely state controllable

Obtain the characteristic polynomial of the matrix A
|[sI − A]|
   
s 0 0 0 1 0

= 0 s 0 −  0 0 1 
   

0 0 s −1 −5 −6
= s3 + 6s2 + 5s + 1
Comparing with
68
s3 + a1 s2 + a2 s + a3
then
a1 = 6, a2 = 5, a3 = 1
The desired characteristic polynomial is given by
(s + 2 − j4) (s + 2 + j4) (s + 10) = s3 + 14s2 + 60s + 200
Comparing this with s3 + α1 s2 + α2 s + α3 then
α1 = 14, α2 = 60, α3 = 200
Since the system is already in controllable canonical form then T = I

The matrix Kis obtained as
h i
K = α3 − a3 α2 − a2 α1 − a1 T −1
h i
K = 199 55 8
2.5.2 State Observer Design
In the pole placement design approach, we assumed that all state variables are available for feedback.
In practice, however, not all the state variables are available for feedback. Then we need to estimate
unavailable state variables. A state observer estimates the state variables based on the measurement
of output and control variables
If a state observer estimates all state variables of the system, regardless of whether some state
variables are available for direct measurement, it is called a full order state observer. A necessary
and sufficient condition for observer design is that the system must be completely state observable.
Consider the system
ẋ = Ax + Bu
y = Cx (1)
The state variables can be estimated from the measured output and control variables
where
G-state observer gain matrix
69
Figure 34: General control system
Figure 35: State Observer
ỹ-estimated output
x̃-estimated state variables
For the observer
˜ = Ax̃ + G (y − ỹ) + Bu
ẋ
ỹ = C x̃ (2)
But y = Cx and ỹ = C x̃
Equation (84) becomes
˜ = Ax̃ + GC (x − x̃) + Bu
ẋ (3)
Subtracting (85) from (83) we obtain
˜ = A [x − x̃] − GC [x − x̃]
ẋ − ẋ
y − ỹ = C [x − x̃] (4)
˜ = ẋand
Taking x − x̃ = x̂ , ẋ − ẋ ˆ y − ỹ = ŷequation (86) becomes
70
ˆ = [A − GC] x̂
ẋ
ŷ = C x̂ (5)
We can choose appropriate Eigen values of [A − GC] to enable placement of poles of the closed
loop system at desired locations
The control design problem is to determine the matrix G;(n ∗ 1) matrix
where
 
g1
 g2 
 
G=
 .. 

.
gn
Determining the matrix G
1. Using Direct Substitution method

Similar to the case of pole placement, if the system is of low order, then direct substitution
of the matrix G into the desired characteristic polynomial may be simpler. e.g. if x is a 3
vector then G can be written as  
g1
G = g2 
 
g3
Substituting this G into the desired characteristic polynomial
|sI − (A − GC)| = (s − µ1 ) (s − µ2 ) (s − µ3 )
By equating the coefficients of the powers on both sides of this equation, we can obtain the values
of g1 , g2 and g3
Example 1.20
Consider the system
ẋ = Ax + Bu
y = Cx
Where
" #
0 20.6
A=
1 0
71
" #
0
B=
1
h i
C= 0 1
Design a full order state observer assuming that the desired Eigen values of the observer matrix
are µ1 = −10 µ2 = −10
Solution
Test the system for observability
Observability matrix " # " #
C 0 1
N= =
CA 1 0
|N | = −1
The system is completely state observable and determination of observer gain matrix is possible
Let " #
g1
G=
g2
|sI − (A − GC)| = 0

s −20.6 + g1
= s2 + g2 s − 20.6 + g1 = 0

−1 s + g2
(s − µ1 ) (s − µ2 ) = (s + 10) (s + 10) = 0
s2 + 20s + 100 = 0
Comparing the two equations

g2 = 20and g1 = 120.6 " #
120.6
G=
20
Using Ackermann’s Formula
72
The observer gain matrix is given by
−1  

0 C
 CA  0
   
G = Φ (A) 
 ..   .. 
  
 .  .
CAn−1 1
Where
Φ (A) = An + α1 An−1 + α2 An−2 + · · · + αn−1 A + αn I
Example 1.21
Determine the observer gain matrix for example 1.20 using Ackermann’s formula
Solution
" #−1 " #

C 0
G = Φ (A)
CA 1
" #
0 20.6
A=
1 0
h i
C= 0 1
(s − µ1 ) (s − µ2 ) = (s + 10) (s + 10)
s2 + 20s + 100
Φ (A) = A2 + 20A + 100I
" #2 " # " #

0 20.6 0 20.6 1 0
= + 20 + 100
1 0 1 0 0 1
" #
120.6 412
=
20 120.6
" #" #" # " #
120.6 412 1 0 0 120.6
G= =
20 120.6 0 1 1 20
73
Using Transformation Matrix Q

Following the same procedure as in deriving the state feedback matrix K then
1. Check the observability condition for the system. If the system is completely observability
then
2. Determine the characteristic polynomial of the matrix A
|sI − A| = sn + a1 sn−1 + a2 sn−2 + · · · + an−1 s + an
Determine the values ofa1 , a2 . . . an
3. Determine the transformation matrix Q that can transform the system state equation in
controllable canonical form. If the system is already in observable canonical form then Q=I
Else
Q = WN
WhereN is the observability matrix and
 
an−1 an−2 . . . a1 1
an−2 an−3 . . . 1 0
 
 
 . . . .
 
W = . . . .
 
 
 . . . .
 
 a1 1 . . . 0 0
 
1 −an−1 . . . 0 0
4. Write the desired characteristic polynomial
(s − µ1 ) (s − µ2 ) . . . . . . (s − µn ) = sn + α1 sn−1 + α2 sn−2 + · · · + αn−1 s + αn
Determine the values of α1 , α2 . . . αn
5. Obtain the observer gain matrix G as

 
αn − an
αn−1 − an−1 
 
G = Q
 .. 
 (6)
 . 
α1 − a1
Example 1.21
Determine the observer gain matrix for example 1.20 using the transformation matrix method
74
Solution
Observability has already been tested
" #
0 20.6
A=
1 0
h i
C= 0 1
" #
s −20.6
|sI − A| = = s2 − 20.6
−1 s
Comparing this with s2 + a1 s + a2

a1 = 0, a2 = −20.6
The system is already in state observable form, therefore
" #
1 0
Q=I=
0 1
The desired characteristic equation is
(s − µ1 ) (s − µ2 ) = (s + 10) (s + 10) = 0
s2 + 20s + 100 = 0
From this we obtain the values of α1 = 20 , α2 = 100

" #
α2 − a2
G=Q
α1 − a1
" #" # " #
1 0 120.6 120.6
= =
0 1 20 20
75
Non-linear Control Lecture Notes by A. M. Muhia
3 NON-LINEAR CONTROL SYSTEMS

Linear systems are systems which obey the principle of superposition and proportionality
If an input x1 produces an output y1 by proportionality theorem an input αx1 produces an output
αy1
Similarly if an input βx2 produces an output βy2 by superposition, an input αx1 + βx2 will produce
an output αy1 + βy2
Linear systems are systems where mathematical tools like Laplace, Fourier etc can be used. These
systems can be analyzed mathematically and graphically
Non-Linear systems exhibit peculiar behavior
1. They do not obey the law of superposition. As a result standard test signals lose their meaning
as the same signal gives different outputs at different operating points
2. The stability of linear systems depends only on the root location and is independent of the
initial state. In non-linear systems the stability depends on root location as well as initial
condition and type of input.
3. Non-linear systems exhibit self sustained oscillations of fixed frequency and amplitude called
Limit Cycles . Linear systems do not have the feature.
4. Linear systems are described by linear differential equations and it is usually possible to obtain
closed form solutions for linear systems. In general this is not possible for non-linear systems
5. Generally the analysis and design methods developed for linear systems e.g root locus, pole
placement, etc apply to all linear systems.
There is no analysis and design method that is universally applicable to all non-linear systems
Unique properties of non-linear systems

1. Multiple equilibrium point
An equilibrium point is a point where the system can stay forever without moving
2. Limit cycles
These are oscillations of fixed amplitudes and fixed period without external excitations
3. Bifurcation
The change in the qualitative behavior of non-linear systems e.g change in number of equi-
librium points, number of limit cycles, stability of equilibrium points etc as a result of quan-
titative change in system parameters
4. Chaos
Small changes in the initial condition result in large and often unpredictable changes in the
system output
76
5. Finite escape time

The situation where the state of an unstable non-linear system can go to infinity in finite time
Examples of non-linearities
Classified into two
1. Inherent (natural) non-linearities

those that naturally come with the systems hardware and motion e.g dead zone, saturation,
backlash, hysteresis etc
2. Intentional (artificial non-linearities)

those introduced into the system by the designer e.g relays, quantizers
3.1 Equilibrium Points
Consider an autonomous system without external inputs

ẋ = f (x t)
when x is the state vector with components x1 , x2 , . . . . . . xn and f (x t) is an n-state vector whose
elements are functions of x1 , x2 , . . . . . . xn andt
we assume that the system has a unique solution starting at a given initial state x (0) = x0
Suppose we have an autonomous system in which all states have settled down to constant values
(not necessarily zero). Such a system is said to be in equilibrium
The state xc is an equilibrium state of the system iff
ẋc = f (xc t) = 0
i.e since the states have settled to constant values
x˙c = 0
Example
Find the equilibrium state of the autonomous LTI system described by the state equations
" # " #" #
ẋ1 (t) −1 2 x1 (t)
=
ẋ2 (t) 1 −1 x2 (t)
Solution
At equilibrium ẋc = 0
0 = −x1 + 2x2
0 = x1 − 2x2
which gives a solution
77
Figure 36: Common Non-linearities
x1 = x2 = 0
hence the equilibrium state for this system is at the origin. i.e
(x1c , x2c ) = (0 , 0)
In general a LTI autonomous system
ẋ = Ax
has a single equilibrium point at the origin if the matrix A is non singular.
If A is singular , the system has infinity equilibrium points
78
Example
Find the equilibrium state of the non-linear system described by the state equations

ẋ1 = x2 − x1 x21 + x22

ẋ2 = −x1 − x2 x21 + x22
Solution
Let the equilibrium point be (x1e , x2e )
At equilibrium point
ẋ1e = ẋ2e = 0
Hence we can write
0 = x2e − x1e x21e − x22e .................(a)

0 = −x1e − x2e x21e − x22e ..............(b)

Multiplying (a) by x1e and (b) by x2e gives

0 = x1e x2e − x21e x21e − x22e .............(c)

0 = −x1e x2e − x22e x21e − x22e ..........(b)

Adding (c) and (d) gives

2
0 = x21e − x22e
whose only solution is
x1e = x2e = 0
Example
Find the equilibrium state of the non-linear system described by the state equations
ẋ1 = x2
ẋ2 = −x1 − x21 − x2
Solution
At equilibrium
0 = x2e
0 = −x1 − x21 − x2
x2e = 0 x1e = 0 or −1
79
Hence the equilibrium state for the system are
(x1e , x2e ) = (0 , 0) and (−1 , 0)

In general a non-linear system may have multiple equilibrium points
3.2 Linearization of non linear state space model and local stability
Based on Taylor series expansion about the operating (equilibrium) point consider a scalar function
f (h). For small x1 Taylor,s approximation theorem states that
∂f (h) 1 ∂ 2 f (h)
f (h + x) = f (h) + x+ + ······
∂h 2! ∂h2
∂f (h)
' f (h) + x + higher order terms
∂h
If h = 0 then

∂f (h)
f (x) = f (0) + x + higher order terms
∂h

h=0
Now consider the system
ẋ = f (x)
Using Taylor’s Theorem for small changes of state x around the origin we can approximate the
system as

∂f (x)
ẋ = f (x) ' f (0) + x + higher order terms
∂x

x=0
In practice linear approximation of the system is done by neglecting the higher order terms
Since it is assumed that the equilibrium state of the system is at the origin then f (0) = 0 hence

∂f (x)
ẋ = x
∂x

x=0
This equation is used to approximate nonlinear system with a linear one
Since Taylor’s Theorem is only valid for small variations around a point , the model obtained is
only useful in analyzing the stability in the neighborhood of the equilibrium point at the origin i.e
local stability
For nonlinear system
ẋ = f (x , u)
80
y = g (x , u)
with
n - state variables
r - inputs
m - outputs
   
ẋ1 f1 (x1 , x2 , . . . xn ; u1 u2 . . . ur )
ẋ2   f2 (x1 , x2 , . . . xn ; u1 u2 . . . ur )
   
 
 .. = .. 
   
 .   . 
ẋn fn (x1 , x2 , . . . xn ; u1 u2 . . . ur )
   
y1 g1 (x1 , x2 , . . . xn ; u1 u2 . . . ur )
 y2   g2 (x1 , x2 , . . . xn ; u1 u2 . . . ur )
   

 . = .. 
 .   
 .   . 
ym gm (x1 , x2 , . . . xn ; u1 u2 . . . ur )
The system is linearized to
∂ ẋ = A∂x + B∂u
∂y = C∂x + D∂u
Where A , B , C , and D are obtained as follows
 ∂f ∂f1 ∂f1 
1 ···
∂x ∂x2 ∂xn
 ∂f21 ∂f2 ∂f2 

 ∂x ∂x2 ··· ∂xn

A =  .1


 .
 .


∂fn ∂fn ∂fn
∂x1 ∂x2 ··· ∂xn
obtained at x = x0 , u = u0 equilibrium points
 ∂f ∂f2 ∂fn 
1 ···
∂u ∂u2 ∂ur
 ∂f21 ∂f2 ∂fn 

 ∂u1 ∂u2 ··· ∂ur

B= .


 .
 .


∂fn ∂fn ∂fn
∂u1 ∂u2 ··· ∂ur
 
∂g1 ∂g1 ∂g1
∂x1 ∂x2 ··· ∂xn
 
C=
 .. .. 
 . . 

∂gm ∂gm ∂gm
∂x1 ∂x2 ··· ∂xn
81
 
∂g1 ∂g1 ∂g1
∂u1 ∂u2 ··· ∂ur
 
D=
 .. .. 
 . . 

∂gm ∂gm ∂gm
∂u1 ∂u2 ··· ∂ur
Example
Consider a system described dy the differential equation
u2
ÿ = 1 − 2
y
Let the state variable of the state equation be x1 = y , x2 = ẏ. Obtain
1. Non linear state space model
2. Linearized modes about equilibrium point given by u = 1 , i.e (u0 = 1) , (x10 = 1) , (x20 = 0)
Solution
x1 = y x2 = ẏ
2
ẋ1 = ẏ = x2 ẋ2 = ÿ = 1 − u2
x1
Therefore the non linear state space model is
" #  
ẋ1 x2
= 2 
ẋ2 1 − u2
x1
which we want to linearize

u0 = 1 we obtain x10 and x20
NB: At equilibrium points derivatives are equal to zero i.e
ẏ = ÿ = 0 or ẋ1 = ẋ2 = 0
ẏ = x2 ⇒ 0 = x20
2
ÿ = 0 = 1 − u2
x1
x10 = 1
x20 = 0
Suppose that y = x1 → linear

" #
h i x1
y= 1 0
x2
82
to obtain A and B
" #  
ẋ1 x2 → f1 (x1 x2 u)
= 2 
ẋ2 1 − u2 → f2 (x1 x2 u)
x1
 
∂f1 ∂f1 " #
∂x1 ∂x2 0 1
A= ∂f2 ∂f2
 =
∂x1 ∂x2 x10 2 0
x20
" ∂f # " #
1 0
B= ∂u =
∂f2
∂u
−2
u0 =1
⇒ linearized model
      
ẋ
 1 =
0 1   x1   0 
+ u
ẋ2 2 0 x2 −2
 
h x1 i
y= 1 0 
x2
Stability of Linearized system
The Stability of Linearized system is analyzed as follows
1. If the Eigen values of the linearized system i.e strictly on the left half plane , the equilibrium
point of the non linear system is asymptotically stable
2. If at least one of the Eigenvalues of the linearized system lies in the right half plane , the
equilibrium point of the non linear system is unstable
3. If the Eigenvalues of the linearized system lies in the left half plane , but at least one of them
lies on the imaginary axis , no conclusion can be drawn about the stability of the equilibrium
plane of the non linear system
Example
Consider the nonlinear system
ẋ1 = x21 + x1 cos x2
ẋ2 = x2 + x21 + x1 sin x2
83
obtain the linearized approximation of the system and check for the stability of the equilibrium
point
Solution
" # " #
ẋ1 x21 + x1 cos x2 f1 (x1 x2 )
=
ẋ2 x2 + x21 + x1 sin x2 f2 (x1 x2 )
 
∂f1 ∂f1 " #
∂x1 ∂x2 cos x2 2x2 − x1 sin x2
A= ∂f2 ∂f2
=
2x1 + sin x2 1 + x1 cos x2 x10
∂x1 ∂x2
x20
ẋ1 = 0 = x220 + x10 cos x20
ẋ2 = 0 = x20 + x210 + x10 sin x20
x10 = 0 x20 = 0
" #
1 0
A=
1 1
Eigenvalues

λI − A = 0

λ − 1 0
=0

−1 λ − 1
(λ − 1)2 = 0
λ1 = λ2 = 1
Eigenvalues lie on the right half plane

Therefore the equilibrium point of the non linear system is unstable
3.3 Describing functions
Describing functions method is a frequency response method which can be used to approximately
analyse and predict non linear behaviour
84
Figure 37: Non-linear System
Consider the system below

The describing function is developed by applying a sinusoid to the non linearity i.e
m (t) = M sin ωt
Assumptions
1. There is only one non linear element in the system
2. The non linear component is time invariant
3. Corresponding to sinusoidal input m (t) = M sin ωt , only the fundamental component in the
output n (t)has to be considered
4. The non linearity has odd symmetry
The steady state n (t)is periodic and in general non sinusoidal. Thus n (t) can be represented by a
Fourier series as
∞ ∞
A0 X X
n (t) = + An cos nωt + Bn sin nωt
2
n=1 n=1
Where ˆ
2 T
An = n (t) cos nωt dt
T t0
ˆ
2 T
Bn = n (t) sin nωt dt
T t0
Where T is the period on the input sinusoid

Non linearity has odd symmetry → A0 = 0
Considering only the fundamental component of the Fourier series
n (t) = A1 cos ωt + B1 sin ωt
85
= N1 sin (ωt + φ)
Where
q
N1 = A21 + B12
N1 sin (ωt + φ) = N1 ∠φ.............(a)
From (a) it can be seen that n (t)can be approximated as a sinusoid of the same frequency as m (t)
but not of the same magnitude and phase
The non linearity gain is given by
N1 ∠φ
N (M , ω) =
M
The equivalent gain is called the describing function and can be represented as follows
Figure 38: Describing function
The describing function N (M , ω) in general is a function of both amplitude and frequency of the
input sinusoid
Derivation of Describing functions
1. Cubic non linearity

let n (t) = m3 (t)
The input must be assumed to be sinusoid i.e
m (t) = M sin ωt
n (t) = M 3 sin3 (ωt)
= M 3 sin ωt sin2 (ωt)

3 1 − cos 2ωt
= M sin ωt
2
86
M3
= [sin ωt − sin ωt cos 2ωt]
2
M3

1
= sin ωt − (sin 3ωt) − sin ωt
2 2
M3
= [3 sin ωt − sin 3ωt]
4
Ignoring third harmonic
3M 3
n (t) = sin ωt
4
3M 3
= ∠0
4
N1 ∠φ
N (M , ω) =
M
3M 3
=
4M
3M 2
=
4
2. Determine the describing function for the 2 position relay non linearity shown in Figure 39
below
Figure 39: 2-Position Relay
For a sinusoid input m (t) = M sin ωt , the output of the relay is as shown below
87
Figure 40: Output of 2-Position Relay
ˆ
2 T
A1 = n (t) cos ωt dωt
T 0
ˆ ˆ
1 π 1 2π
= v cos ωt dωt − v cos ωt dωt
π 0 π π
π 2π
v v
= sin ωt − sin ωt = 0

π 0 π π
ˆ
2 T
B1 = n (t) sin ωt dωt
T 0
ˆ ˆ
1 π 1 2π
= v sin ωt dωt − v sin ωt dωt
π 0 π π
88
π 2π
v v
= − cos ωt + cos ωt

π 0 π π
v v
− [−1 − 1] + [1 + 1]
π π
4v
=
π
4v 4v
n (t) = = ∠0
π π
N1 ∠φ
N (M , ω) =
M
4v
=
πm
Example
Determine the describing function for the saturation non linearity shown in Figure 41 below
Figure 41: Saturation
Solution
n (t) is given by



 −ks m (t) < −s

n (t) = km (t) −s ≤ m (t) ≤ s



ks m (t) > s
For a sinusoid input m (t) = M sin ωt , the output of the saturation non linearity is as shown in
Figure 42 For the sinusoid input m (t) = M sin ωt , the output of the saturation non linearity is
89
Figure 42: Output of Saturation
given by



 kM sin ωt 0 ≤ ωt ≤ θ1


θ1 ≤ ωt ≤ θ2



 ks

n (t) = kM sin ωt θ2 ≤ ωt ≤ θ3


−ks

 θ3 ≤ ωt ≤ θ4



θ4 ≤ ωt ≤ 2π

kM sin ωt
A1 = 0 Odd symmetry
ˆ
2 T
B1 = n (t) sin ωt dωt
T 0
90
ˆ 2π
2
= n (t) sin ωt dωt
2π 0
ˆ
1 2π
π 0
ˆ
4 π/2
π 0
ˆ ˆ
4 θ1 4 π/2
= kM sin2 ωt dωt + ks sin ωt dωt
π 0 π θ1
ˆ ˆ
4kM θ1 1 − cos 2ωt 4ks π/2
= dωt + sin ωt dωt
π 0 2 π θ1
θ π/2
2kM 1 1 4ks
= ωt − sin 2ωt −
cos ωt
π 2 0 π θ 1

2kM 1 4ks
= θ1 − sin 2θ1 + cos θ1
π 2 π

4 kM θ1 kM sin 2θ1
= − + ks cos θ1
π 2 4
q
2
s θ1 = sin−1 M
s 1− s2

but M sin θ1 = s sin θ1 = M cos θ1 =
M
" r r #
4 kM s 2kM s s2 s2
= sin−1 − 1 − 2 + ks 1− 2
π 2 M 4 M M M
" r #
4 kM s ks s2
= sin−1 + 1− 2
π 2 M 2 M
" r #
2kM −1
s s s2
= sin + 1− 2
π M M M
the describing function is given by
N ∠0
M
" r #
2kM −1
s s s2
= sin + 1− 2
πM M M M
91
" r #
2k −1
s s s2
= sin + 1− 2
π M M M
Example
Figure 43: Deadzone

0 0 < ωt < θ1
n (ωt) =
k [M sin ωt − A] θ1 < ωt < π2
ˆ
4 π/2
B1 = k [M sin ωt − A] sin ωt dωt
π θ1
ˆ
4k π/2 h i
= M sin2 ωt − A sin ωt dωt
π θ1
92
ˆ
4k π/2 M
= [1 − cos 2ωt] − A sin ωt dωt
π θ1 2
" π/2 π/2 #

4k M 1
= ωt − sin 2ωt + A cos ωt
π 2 2 θ θ1 1

4k M π
= − θ1 + A (0 − cos θ1 )
π 2 2

4k M π
= − θ1 + A cos θ1
π 2 2

4kM π A
= − θ1 + cos θ1
2π 2 M
q
2 2

A
But sin θ1 = M θ1 = sin−1 M
A cos θ1 = M −A 2 M
" r #
M 2 − A2

2kM π A A
= − sin−1 +
π 2 M M M2
" r #
M 2 − A2

2k π A A
N (M , ω) = − sin−1 +
π 2 M M M2
Example




−k [M − A] 0 < ωt < θ1


π
k [M sin ωt − A] θ1 < ωt <




 2
n (ωt) = k [M − n] π
2
< ωt < θ3


k [M sin ωt + A] θ3 < ωt < 3 π2







−k [M − A] 3 π2 < ωt < 2π

ˆ 2π
2
A1 = n (ωt) cos ωt dωt
2π 0
ˆ
1 2π
= n (ωt) cos ωt dωt
π 0
93
Figure 44: Backlash
ˆ ˆ
1 θ1 1 π/2
= −k [M − A] cos ωt dωt + k [M sin ωt − A] cos ωt dωt
π 0 π θ1
ˆ ˆ
1 θ3 1 3π/2
+ k [M − n] cos ωt dωt + k [M sin ωt + A] cos ωt dωt
π π/2 π θ3
ˆ
1 2π
+ −k [M − A] cos ωt dωt
π 3π/2
94
θ π/2
kM π/2 k [M − A]
θ
−k [M − A] 1 kA 3
= sin ωt − sin ωt + ωt + sin ωt
π 0 π θ1 2π θ1 π π/2
3π/2 3π/2 2π
kA kM k [M − A]
+ sin ωt + ωt − sin ωt
π θ 2π θ π 3π/2
3 3
−k [M − A] kA kA kM π k [M − A]
= sin θ1 − + sin θ1 + − θ1 + sin θ3
π π π 2 2 π
k [M − A] kA kA kM 3π k [M − A]
− − − sin θ3 + − θ3 +
π π π 2π 2 π
95
Optimal Control Lecture Notes by A. M. Muhia
4 OPTIMAL CONTROL
The objective of optimal control theory is to determine the control signal that will optimize (max-
imize or minimize) some performance criterion while at the same time satisfying the physical con-
straints of the system
Examples
To find a control strategy to transfer a system from an initial state to a final state using the mini-
mum energy
To find a control strategy to transfer a system from an initial state to a final state in minimum
time given the input to the system is limited to a certain value.
To send a rocket to space while minimizing fuel consumption
The formulation of the control problem requires
• A mathematical description (or model) of the process to be controlled
• specifications of the performance criterion
• a statement of physical constraints
4.1 Performance Index
1. Minimum time problem

We intend to transfer the system from an initial state x (t0 )to a specified final state x tf in
minimum time.
The performance index is given by
ˆ t
f
J = tf − t0 = dt
t0
2. Minimum control effort (energy) problem

The rate of expenditure of energy is proportional to u2 (t). So to minimize energy we minimize
ˆ t
f
J= u2 (t) dt
t0
For several control inputs , the equation above takes the form
ˆ t
f
J= uT (t) u (t) dt
t0
96
To allow for greater generality for weighting different control signals separately, we can write
ˆ t
f
J= uT (t) Ru (t) dt
t0
where R is a positive definite matrix
3. Tracking Problem
The objective is to maintain the state x (t) as close as possible to the desired state xd (t)
The performance index is
ˆ t
f
J= [x (t) − xd (t)]T ϕ (x (t) − xd (t)) dt
t0
where ϕ is a positive definite matrix
4. State regulator problem

This is a special case of the tracking problem above where xd (t) = 0
The performance index is ˆ t
f
J= xT (t) ϕx (t) dt
t0
5. Terminal control problem

In a terminal control problem, we would like to minimize the deviation of the final state

x tf from the desired state xd tf .
The performance measure is
T
J = x tf − xd tf H x tf − xd tf

If xd tf = 0
J = xT tf Hx tf

6. General optimal control problem

The general performance index is given by
ˆ t
T f
[x (t) − xd (t)]T ϕ (x (t) − xd (t))+uT (t) Ru (t) dt

J = x tf − xd tf H x tf − xd tf +
t0
If xd (t) = 0 then
ˆ t
T f
xT (t) ϕx (t) + uT (t) Ru (t)

J =x tf Hx tf + dt
t0
97
4.2 Solution to the Optimal Control problem
There are three methods of solving Optimal Control problems
1. Calculus of variations
2. The minimum principle
3. Dynamic programming
Calculus of variation
Deals with finding the optimum value of a function

Let x (t) be a scalar function with continuous first derivatives
The basic variational problem is to find the optimal x (t) usually denoted as x∗ (t) for which the
functional ˆ t
f
Jx (t) = v (x (t) , ẋ (t) , t) dt
t0
has an optimum value (minimum or maximum)

It is assumed that the integrand v has continuous first and second derivative w.r.t all its arguments
The necessary condition for optimality is that
∂v (x∗ (t) , ẋ∗ (t) , t) d ∂v (x∗ (t) , ẋ∗ (t) , t)

−
∂x dt ∂ ẋ
i.e
∂v d ∂v
− =0
∂x dt ∂ ẋ

for all t t0 , tf
This is known as the Euler-Lagrange equation
Compliance with the Euler-Lagrange equation is only a necessary condition for optimum. Op-
timality may sometimes not yield a max or min , just an inflection point where the derivative
vanishes.
However, if the Euler-Lagrange equation is not satisfied, for any function , this indicates that the
optimum does not exist for the functional
Example
Find ẋ (t) which minimize the cost function
ˆ 2h i
J= 2x2 (t) + 2x (t) ẋ (t) + ẋ2 (t) dt
0
that satisfies the boundary condition x (0) = 0 x (2) = 1

Solution
98
From the above function
v (x (t) , ẋ (t) , t) = 2x2 (t) + 2x (t) ẋ (t) + ẋ2 (t)
∂v
= 4x (t) + 2ẋ (t)
∂x
∂v
= 2x (t) + 2ẋ (t)
∂ ẋ

d ∂v
= 2ẍ (t) + 2ẋ (t)
dt ∂ ẋ

∂v d ∂v
− =0
∂x dt ∂ ẋ
⇒ 4x (t) + 2ẋ (t) − 2ẍ (t) − 2ẋ (t) = 0
4x (t) − 2ẍ (t) = 0
ẍ (t) − 2x (t) = 0
Solving the differential equation

m2 − 2 = 0
√ √ √
m=± 2 m1 = 2 m2 = − 2
√ √
x∗ (t) = c1 exp− 2 t +c2 exp 2 t
where c1 and c2 are constants

The boundary conditions are substituted to obtain the values of c1 and c2
x (0) = 0 ⇒ c1 + c2 = 0
√ √
x(2) = 1 ⇒ c1 exp− 2 2 +c2 exp 2 2 = 1
c1 = √ 1 √ c2 = √ −1 √
exp− 2 2 − exp 2 2 exp− 2 2 − exp 2 2
Example
99
Find the optimum of
ˆ 2h i
J= ẋ2 (t) + 2t x (t) dt
0
subject to the boundary conditions x (1) = 1 , x (2) = 5
Solution
v (x (t) , ẋ (t) , t) = ẋ2 (t) + 2t x (t)
∂v
= 2t
∂x
∂v
= 2ẋ (t)
∂ ẋ

d ∂v
= 2ẍ (t)
dt ∂ ẋ

∂v d ∂v
− = 2t − 2ẍ (t) = 0
∂ ẋ dt ∂ ẋ
⇒ ẍ (t) = t
t2
ẋ (t) = + c1
2
t3
x (t) = + c1 t + c2
6
Substituting the boundary conditions
x (0) = 1 c2 = 1
8
x (2) = 5 5= + 2c1 + 1
6
4
c1 =
3
t3 4
ẋ (t) = + t+1
6 3
Functional involving several Independent functions
100
Consider the functional

ˆ t
f
J (x1 , x2 , . . . xn ) = v [(x1 (t) , . . . xn (t)) , ẋ1 (t) . . . ẋn (t) , t] dt
t0
The above equation can be written in a more compact format as

ˆ t
f
J x (t) = v [x (t) , ẋ (t) , t] dt
t0
where  
x1 (t)
 x2 (t) 
 
x (t) =  . 

 .. 

xn (t)
  d 
ẋ1 (t) dt x1 (t)
 d
 ẋ2 (t)   dt x2 (t) 
 
ẋ (t) = 
 ..  
 =  .. 

 .   . 
ẋn (t) d
dt xn (t)
The corresponding Euler Lagrange equation is
∂v (x∗ (t) , ẋ∗ (t) , t) d ∂v (x∗ (t) , ẋ∗ (t) , t)

− =0
∂x dt ∂ ẋ
In expanded form
∂v (x∗ (t) , ẋ∗ (t) , t) d ∂v (x∗ (t) , ẋ∗ (t) , t)

− =0
∂x1 dt ∂ ẋ1
∂v (x∗ (t) , ẋ∗ (t) , t) d ∂v (x∗ (t) , ẋ∗ (t) , t)

− =0
∂x2 dt ∂ ẋ2
....
..
∂v (x∗ (t) , ẋ∗ (t) , t) d ∂v (x∗ (t) , ẋ∗ (t) , t)

− =0
∂xn dt ∂ ẋn
or simply
101

∂v d
− dt ∂v =0
∂x1 ∂ ẋ1

∂v d
− dt ∂v =0
∂x2 ∂ ẋ2
..
.

∂v d
− dt ∂v =0
∂xn ∂ ẋn
Example
Find the extrema for the functional
ˆ π/4 h i
J (x) = x21 (t) + ẋ22 (t) + ẋ1 (t) ẋ2 (t) dt
0
where the boundary conditions are
" #
" #" # " #
x1 (0) 0 x1 π4 1
= =
x2 (0) 0 x2 π4 −1
Solution
v (x (t) , ẋ (t) , t) = x21 (t) + ẋ22 (t) + ẋ1 (t) ẋ2 (t)

∂v ∂v d ∂v
= 2x1 (t) ; = ẋ2 (t) ; = ẍ2 (t)
∂x1 ∂ ẋ1 dt ∂ ẋ1

∂v ∂v d ∂v
= 0; = 2ẋ2 (t) + ẋ1 (t) ; = 2ẍ2 (t) + ẍ1 (t)
∂x2 ∂ ẋ2 dt ∂ ẋ2

∂v d ∂v
− =0 ⇒ 2x1 (t) − ẍ2 (t) = 0
∂x1 dt ∂ ẋ1

∂v d ∂v
− =0 ⇒ 2ẍ2 (t) + ẍ1 (t) = 0
∂x2 dt ∂ ẋ2
4x1 (t) − 2ẍ2 (t) = 0
ẍ1 (t) + 2ẍ2 (t) = 0
ẍ1 (t) + 4x1 (t) = 0
102
x1 (t) = c1 cos 2t + c2 sin 2t
ẍ2 (t) = 2x1 (t) = 2c1 cos 2t + 2c2 sin 2t
ẋ2 (t) = c1 sin 2t − c2 cos 2t + c3
1 1
x2 (t) = − c1 cos 2t − c2 sin 2t + c3 t + c4
2 2
Substituting the boundary conditions
x1 (0) = 0 c1 = 0 x2 (0) = 0 c4 = 0
π π −2
x1 =1 c2 = 1 x2 = −1 c3 =
4 4 π
x1 (t) = sin 2t
−1 2
x2 (t) = sin 2t − t
2 π
The Lagrange Multiplier
The elimination method (direct method) used previously gets tedious for higher order problems
The Lagrange multiplier method on the other hand adjoins the constraint to the original function
and the adjoined function is extremized in the usual way
Consider the extreme of the function f (x1 , x2 )subject to the condition g (x1 , x2 ) = 0
We form an augmented Lagrangian function
L (x1 , x2 , λ) = f (x1 , x2 ) + λg (x1 , x2 )
A necessary condition for extreme is that
df = dL = 0

∂f ∂g ∂f ∂g
+λ dx1 + +λ dx2 = 0
∂x1 ∂x1 ∂x2 ∂x2
Example
103
In a 2 dimension space , find the point on the line x1 + x2 = 5 that is nearest to the origin
Solution
Elimination method The square of the distance from the origin is given by
f (x1 , x2 ) = x21 + x22
This function is to be minimized subject to the condition
x1 + x2 = 5
x1 = 5 − x2
f (x2 ) = (5 − x2 )2 + x22 = 2x22 − 10x2 + 25
∂f (x2 )
= 4x2 − 10 = 0
∂x2
x∗2 = 2.5 x∗1 = 2.5
∂ 2 f (x2 )
=4>0
∂x22
√
this is a minimum point. It follows that the minimum distance from the origin is 5/ 2
The Lagrange Multiplier method
L (x1 , x2 , λ) = x21 + x22 + (x1 + x2 − 5)
For the values of x1 and x2 that satisfy the constraint , the function λ (x1 + x2 − 5) equals zero
and so we have simply added a zero to the function to be minimized
Taking partial derivative of the function w.r.t to all the variables and equating to zero
∂L
= 2x1 + λ = 0
∂x1
∂L
= 2x2 + λ = 0
∂x2
∂L
= x1 + x2 − 5 = 0
∂λ
104
solving simultaneously we obtain
x∗1 = 2.5 x∗2 = 2.5 λ∗ = 5
Example
A manufacture would like to package his products in cylindrical containers. The cost of each empty
container is proportional to the area of the material that is used to manufacture the can i.e the
cost of each container is proportional to its surface area. The manufacturer would like to maximize
the volume of the cylindrical container for a given surface area A0 . Assuming the radius of the
container is r and the height h ,determine the relationship between the radius r and the height h
that maximizes the volume of the container subject to the constraint that the surface area of the
constraint is a constant A0
Solution
Elimination method We maximize the volume
v (r, h) = πr2 h
subject to the surface area
A (r, h) = 2πr2 + 2πrh = A0
we eliminate one of the variables and maximize the resulting function
A − 2πr2 A
h= = −r
2πr 2πr
v = πr2 h
1
= Ar − πr3
2
hence
dv 1
= A − 3πr2
dr 2
r
A
r=
6π
r r
4A 2A
h= =
6π 3π
The second derivation is
105
d2 v
= −6πr < 0
dr2
Hence this is a maximum point
The Lagrange Multiplier method The constraint is written as
2πr2 + 2πrh − A0 = 0

L (r, h) = πr2 h + λ 2πr2 + 2πrh − A0
The function L is known as the Lagrangian which is a function of 3 variables: r , h , and λ

Taking partial derivative of L w.r.t each of these variables and equating to zero
∂L
= 2πrh + λ (4πr + 2πh) = 0
∂r
∂L
= πr2 + 2πrλ
∂h
∂L
= 2πr2 + 2πrh − A0 = 0
∂λ
solving the above equation simultaneously gives
r r
A 2A
r= h=
6π 3π
for a given surface area the volume is maximized if the height is twice the radius
Consider the problem of finding the extrema of a real valued function f (x) = f (x1 , x2 , . . . xn )
subject to the conditions
g1 (x) = g1 (x1 , x2 , . . . xn ) = 0
g2 (x) = g2 (x1 , x2 , . . . xn ) = 0
..
.
gm (x) = gm (x1 , x2 , . . . xn ) = 0
where f and g are continuous derivatives and m < n
106
Let λ1 , λ2 ,. . . . . .λn be the Lagrange multipliers corresponding to the m conditions

The Lagrange function is
L (x1 , x2 , . . . xn , λ1 , λ2 , . . . λm ) = f (x1 , x2 , . . . xn )+λ1 g1 (x1 , x2 , . . . xn )+· · ·+λm gm (x1 , x2 , . . . xn )
 
g1 (x1 , x2 , . . . xn )
 g2 (x1 , x2 , . . . xn ) 
 
= f (x1 , x2 , . . . xn ) + (λ1 , λ2 , . . . λm ) 
 .. 

 . 
gm (x1 , x2 , . . . xn )
L (x, λ) = f (x) + λT g (x)
The optimal values x∗ and λ∗ are solutions to the following (n + m) equations
∂L ∂f ∂g1 ∂g2 ∂gm

= + λ1 + λ2 + · · · + λm =0
∂x1 ∂x1 ∂x1 ∂x1 ∂x1
∂L ∂f ∂g1 ∂g2 ∂gm

= + λ1 + λ2 + · · · + λm =0
∂x2 ∂x2 ∂x2 ∂x2 ∂x2
..
.
∂L ∂f ∂g1 ∂g2 ∂gm

= + λ1 + λ2 + · · · + λm =0
∂xn ∂xn ∂xn ∂xn ∂xn
∂L
= g1 (x1 , x2 , . . . xn ) = 0
∂λ1
∂L
= g2 (x1 , x2 , . . . xn ) = 0
∂λ2
..
.
∂L
= gm (x1 , x2 , . . . xn ) = 0
∂λm
or more compactly
∂L ∂f ∂g
= + λT =0
∂x ∂x ∂x
∂L
= g (x) = 0
∂λ
107
NB:
The introduction of the Lagrange multiplier allows us to treat all variables in the augmented
function L (x1 , x2 , . . . xn , λ1 , λ2 , . . . λm ) as though they are independent variables
The Lagrange multiplier makes it easy to solve the problems of constrained extremization but its
value is not important
Extreme of functional with conditions
Consider the extremization of the performance index in the form of a functional
ˆ t
f
T (x1 , x2 , . . . xn , t) = v (x1 , x2 , . . . xn , ẋ1 , ẋ2 , . . . ẋn , t) dt
t0
subject to the boundary conditions (or plant system equations)
g1 (x1 , x2 , . . . xn , ẋ1 , ẋ2 , . . . ẋn , t) = 0
g2 (x1 , x2 , . . . xn , ẋ1 , ẋ2 , . . . ẋn , t) = 0
..
.
gm (x1 , x2 , . . . xn , ẋ1 , ẋ2 , . . . ẋn , t) = 0
we form the augmented functional
ˆ t
f
J= L (x1 , x2 , . . . xn , ẋ1 , ẋ2 , . . . ẋn , λ1 (t) , λ2 (t) , . . . λm (t) , t) dt
t0
where λs are the Lagrange multipliers and the Lagrangian is defined as
L (x1 , x2 , . . . xn , ẋ1 , ẋ2 , . . . ẋn , t) = v (x1 , x2 , . . . xn , ẋ1 , ẋ2 , . . . ẋn , t)

+λ1 (t) g1 (x1 , x2 , . . . xn , ẋ1 , ẋ2 , . . . ẋn , t) + · · · · · · + λm (t) gm (x1 , x2 , . . . xn , ẋ1 , ẋ2 , . . . ẋn , t)
The necessary condition for extremization of the functional subject to the conditions are

∂L d ∂L
− =0
∂x dt ∂ ẋ

∂L d ∂L
− =0
∂λ dt ∂ λ̇
In Control Systems
108
Applying the results in control systems

Consider the problem of extremizing the functional
ˆ t
f
J (x, u, t) = v (x, u, t)
t0
subject to the plant equation
ẋ (t) = f (x (t) , u (t) , t)
where    
x1 (t) u1 (t)
 x2 (t)   u2 (t) 
   
x (t) =  . 
 and u (t) =  . 

 ..   .. 
 
xn (t) um (t)
The optimization problem is to extremize the functional
ˆ t
f
J (x1 , x2 , . . . xn , u1 , u2 , . . . um , t) = v(x1 , x2 , . . . xn , u1 , u2 , . . . um ,
t0
ẋ1 , ẋ2 , . . . ẋn , u̇1 , u̇2 , . . . u̇m , t)dt
subject to the constraints
f1 (x1 , x2 , . . . xn , u1 , u2 , . . . um , t) − ẋ1 (t) = 0
f2 (x1 , x2 , . . . xn , u1 , u2 , . . . um , t) − ẋ2 (t) = 0
..
.
fn (x1 , x2 , . . . xn , u1 , u2 , . . . um , t) − ẋn (t) = 0
The Lagrangian is given by
L (x1 , x2 , . . . xn , u1 , u2 , . . . um , ẋ1 , ẋ2 , . . . ẋn , u̇1 , u̇2 , . . . u̇m , λ1 , λ2 , . . . λm , t)
109
= v (x1 , x2 , . . . xn , u1 , u2 , . . . um , ẋ1 , ẋ2 , . . . ẋn , u̇1 , u̇2 , . . . u̇m , t) +

λ1 (t) f1 (x1 , x2 , . . . xn , u1 , u2 , . . . um , t) − ẋ1 (t) + · · · +
λm (t) fm (x1 , x2 , . . . xn , u1 , u2 , . . . um , t) − ẋm (t)
The necessary conditions for extremization become

∂L d ∂L
− =0
∂x1 dt ∂ ẋ1
..
.

∂L d ∂L
− =0
∂xn dt ∂ ẋn

∂L d ∂L
− =0
∂u1 dt ∂ u̇1
..
.

∂L d ∂L
− =0
∂um dt ∂ u̇m

∂L d ∂L
− =0
∂λ1 dt ∂ λ̇1
..
.

∂L d ∂L
− =0
∂λm dt ∂ λ̇m
Example
Suppose that the system described by the state equations
ẋ1 (t) = x1 (t)
ẋ2 (t) = u (t)
is to be controlled to minimize the performance index
ˆ
1 2 2
J (x1 , x2 , u) = u (t) dt
2 0
110
with boundary conditions " # " #

x1 (0) 1
=
x2 (0) 2
and " # " #
x1 (2) 1
=
x2 (2) 0
find the optimal value of x1 (t) , x2 (t) and u (t)

Solution
ẋ1 (t) − x2 (t) = 0
ẋ2 (t) − u (t) = 0
The Lagrangian is given by
1 2
u (t) − λ1 (t) [ẋ1 (t) − x2 (t)] + λ2 (t) [ẋ2 (t) − u (t)]
2

∂L ∂L d ∂L
=0 = λ1 (t) = λ̇1 (t) ⇒ −λ̇1 (t) = 0
∂x1 ∂ ẋ1 dt ∂ ẋ1

∂L ∂L d ∂L
= −λ̇1 (t) = λ2 (t) = λ̇2 (t) ⇒ −λ̇1 (t) − λ̇2 (t) = 0
∂x2 ∂ ẋ2 dt ∂ ẋ2

∂L ∂L d ∂L
= u (t) − λ2 (t) =0 =0 ⇒ −u (t) − λ2 (t) = 0
∂u ∂ u̇ dt ∂ u̇

∂L ∂L d ∂L
= ẋ1 (t) − x2 (t) =0 =0 ⇒ ẋ1 (t) − x2 (t) = 0
∂λ1 ∂ λ̇1 dt ∂ λ̇1

∂L ∂L d ∂L
= ẋ2 (t) − u (t) =0 =0 ⇒ ẋ2 (t) − x2 (t) = 0
∂λ2 ∂ λ̇2 dt ∂ λ̇2
Solving the simultaneous equations
λ1 (t) = c1
111
λ2 (t) = −c1 t + c2
u (t) = −c1 t + c2
1
x2 (t) = − c1 t2 + c2 t + c3
2
1 1
x1 (t) = − c1 t3 + c2 t2 + c3 t + c4
6 2
substituting the boundary conditions
x1 (0) = 1 ⇒ c4 = 1
x2 (0) = 2 ⇒ c3 = 2
x2 (2) = 0 − 2c1 + 2c2 = −2

4
x1 (2) = 1 − c1 + 2c2 + 4 + 1 = 1
3
4
− c1 + 2c2 = −4
3
− 2c1 + 2c2 = −2
c1 = −3 c2 = −4
The optimal values of x1 (t) , x2 (t) and u (t) are
t3
x∗1 (t) = − 2t2 + 2t + 1
2
3t2
x∗2 (t) = − 4t + 2
2
u∗ (t) = 3t − 4
The Hamiltonian
The Hamiltonian H is defined as
H (x (t) , u (t) , λ (t) , t) = v (x (t) , u (t) , t) + λT [f x (t) , u (t) , t]
112
The above function is also known as the Pontryazin H function

Consider the problem of extremizing the functional
ˆ t
f
J (x, u, t) = v (x, u, t)
t0
subject to the plant equation
ẋ (t) = f (x (t) , u (t) , t)
The optimal control u∗ (t) and optimal state x∗ (t) can be obtained by solving simultaneously the
equations
∂
ẋ (t) = [H (x (t) , u (t) , λ (t) , t)] − state equation
∂x
∂
λ̇ (t) = [H (x (t) , u (t) , λ (t) , t)] − co − state equation
∂x
∂
0= [H (x (t) , u (t) , λ (t) , t)] − control equation
∂u
Example
Given a second order system
ẋ1 (t) = x2 (t)
ẋ2 (t) = u (t)
and performance index
ˆ
1 2 2
J= u (t) dt
2 0
find the optimal control u∗ (t) and optimal state x∗1 (t) and x∗2 (t) given the boundary condition
" # " # " # " #
x1 (0) 1 x1 (2) 1
= and =
x2 (0) 2 x2 (2) 0
Solution
The Hamiltonian is given by
1 2
H= u (t) + λ1 (t) x2 (t) + λ2 (t) u (t)
2
113
∂H
ẋ1 (t) = = x2 (t)
∂λ1
∂H
ẋ2 (t) = = u (t)
∂λ2
−∂H
λ̇1 (t) = =0
∂x1
−∂H
λ̇2 (t) = = −λ1 (t)
∂x2
∂H
0= = u (t) + λ2 (t)
∂u
solving the equations
λ1 (t) = c1
λ2 (t) = −c1 t + c2
u (t) = c1 t − c2
1 2
x2 (t) = c1 t − c2 t + c3
2
1 3 1 2
x1 (t) = c1 t − c2 t + c3 t + c4
6 2
substituting the boundary conditions we obtain
c1 = 3 c2 = 4 c3 = 2 c4 = 1
it implies
λ∗1 (t) = 3
λ∗2 (t) = −3t + 4
u∗ (t) = 3t − 4
114
3 2
x∗2 (t) = t − 4t + 2
2
t3
x∗1 (t) = − 2t2 + 2t + 1
2
115

EMM514 Control Engineering II

Uploaded by

Copyright:

Available Formats

EMM514 Control Engineering II

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

EMM514 Control Engineering II

Uploaded by

Copyright:

Available Formats

Kenyatta University

BSc Mechanical Engineering

EMM514: CONTROL ENGINEERING II

March 12, 2020

EMM 507 Control Engineering I

The aim of this course is to enable the student to;

1. understand design of controllers using PIDs

2. understand design of digital controllers

3. understand the various non linearities in control systems

4. understand the concept of optimal control systems

At the end of this course, the student should be able to;

1. Design and implement PI,PD and PID controllers

2. Design and implement digital controllers

3. Analyse non linearities in control systems

4. Perform optimization of control systems

1. Control Engineering laboratories;

Prescribed Text Books

1. Ogata K. (1996) Modern Control Engineering, Prentice Hall, 3rd Ed.

6. Donald E. Kirk, Optimal Control Theory, Dover publications, 2004

7. Naidu D. S., Optimal Control Systems, CRC press, 2003

4. Hans P. Geering, Optimal Control with Engineering Applications, Springer, 2007

5. Journal of Dynamic Systems, Measurement, and Control

2 STATE VARIABLE METHODS 32

3 NON-LINEAR CONTROL SYSTEMS 76

Control system: An arrangement of physical components connected or related in such a manner

• An arrangement to achieve this input-output combination

1. Open loop systems

2. Closed looped systems

Open Loop Systems

Figure 1: Open Loop Control System

Closed Loop Control Systems

Figure 2: Closed Loop Control System

r (t) - reference input

Linear and Non-linear Control Systems

• Non-linear systems exhibit peculiar behaviour

• They do not obey the law of superposition.

• Standard test signals approach cannot be used here

Table 1: Linear and Non Linear Systens

1.2 Proportional (P) Control Action

u (t) = kp e (t) (1)

where kp is the proportional gain constant

Figure 3: General Proportional Controller

The controller is an amplifier with a gain kp (the gain maybe adjustable)

U (s) = kp E (s) (2)

The closed loop transfer function is given by

The characteristic equation of the closed-loop system is

The gain kp in equation 1.4 is varied to generate the root-locus.

Figure 4: Example 1.1

open loop transfer function

Figure 5: Example 1.2

open loop transfer function

open loop zeros: none

for different values of k the roots can be obtained as

k characteristic equation roots

Proportional Control of type 0, first order system

Consider the system below where k and c are constants

Figure 6: Proportional Control of type 0, first order system

The closed loop transfer function of the system is given by

Figure 7: Response of Proportional Control of type 0, first order system

It can be seen that Yss → 1 as kp → ∞