Linear Systems and Control

MCEN3006 / MCEN6019

Jonathan Paxman
j.paxman@curtin.edu.au
Room 204.413
Real
World A
Approximations bundle
Non of
Linear really
Linear good
Systems linear
Sometimes! tools
The Dynamic Model

A model is the abstract

representation of a system
or process that serves to
link causes and effects.
All real-world systems have a myriad of
physical, biological, economic,
aesthetic, etc, attributes. It is the
essence of the art of modelling to
assemble only that small subset of all
of the characteristics or features of the
system that is sufficient to serve the
specific objectives of the model as set
by the modeller.
Every model represents a
compromise between the
accuracy of the model itself,
and the effort that was
necessary to obtain that
model.
 Modelling is
the selective loss
of information.
The Stages of a Dynamic Investigation - 1

1. Specify the system to be studied and imagine a

simple PHYSICAL MODEL whose behaviour will
match sufficiently closely the behaviour of the
actual system.
2. Derive a mathematical model to represent the
physical model; that is, write the differential
EQUATIONS OF MOTION of the physical model
("motion" here means "change").
The Stages of a Dynamic Investigation - 2

3. Study the DYNAMIC BEHAVIOUR of the

mathematical model, by solving the differential
equations of motion.
4. Make DESIGN DECISIONS: i.e. choose the
physical parameters of the system and/or augment
the system so that it will behave as desired.
State space equations
x = Ax + Bu
y = Cx + Du

 These standard equations describe a system in state-

space form.
 x is a column vector contain all system states.
 u is a column vector containing all system inputs (or a scalar if
the system is single input).
 y is a column vector containing all system outputs (or a scalar
if the system is single output).
 The system dynamics are described completely by the state-
space matrices A , B , C , and D .
State space equations
x = Ax + Bu
y = Cx + Du
 State space equations can be directly obtained from the equations
of motion of a system.
 In a rigid mechanical system, the states may be positions (linear or
angular) and velocities of independently moving parts of the system.
 In an RLC circuit, the states may be the “states” of the energy
storing devices – capacitor voltages and inductor currents.
 In a tank system, the states may be the levels of the tanks
 The choice of state is not unique. Two different choices of x may
result in different state space matrices, but the input-output
behaviour will be the same.
Nonlinear state space equations

𝒙𝒙̇ = 𝑓𝑓 𝒙𝒙, 𝒖𝒖
𝒚𝒚 = 𝑔𝑔(𝒙𝒙, 𝒖𝒖)
 We can write the equations of a nonlinear system in a similar way,
replacing the linear 𝐴𝐴𝐴𝐴 + 𝐵𝐵𝐵𝐵 by the more general 𝑓𝑓 𝒙𝒙, 𝒖𝒖 .
 That is, the derivative of the state is still a function of the state and
the input, but no longer necessarily a linear function.
 Similarly, the output is a function (no longer necessarily linear) of the
input and the state.
 It is more difficult to obtain analytical solutions for nonlinear state
equations like these, however it is often possible to obtain numerical
solutions using a package like MATLAB, as we will see in computer
laboratories 1 and 2, and in the first assignment.
 If the system stays near a particular state (e.g through a control
system), then the dynamics can often be linearised to obtain linear
equations.
 Ways to simplify the modelling task

Linear Systems & Control (MCEN 3006)

Curtin University is a trademark of Curtin University of Technology
CRICOS Provider Code 00301J

12
Dynamic modelling: rigid body mechanics

 Draw free body diagrams for each of the

independent bodies
 Apply Newton’s 2nd law: ∑ 𝐹𝐹 = 𝑚𝑚𝑚𝑚 to each of the
translating free bodies, in each direction of
motion.
 Apply: ∑ 𝑇𝑇 = 𝐼𝐼𝛼𝛼 to each of the rotating free
bodies, in each direction of rotation.
 (Can also use Hamiltonian or Lagrangian
mechanics)
Rigid body mechanics example:
mass-spring-damper (review)

� 𝐹𝐹𝑧𝑧 = 𝑚𝑚𝑧𝑧̈

𝐹𝐹 𝑡𝑡 − 𝑞𝑞 𝑧𝑧̇ − 𝑘𝑘𝑘𝑘 = 𝑚𝑚𝑧𝑧̈

𝑞𝑞 𝑘𝑘 𝐹𝐹(𝑡𝑡)
𝑧𝑧̈ + 𝑧𝑧̇ + 𝑧𝑧 =
𝑚𝑚 𝑚𝑚 𝑚𝑚
Dynamic modelling practice example:
mass-spring-damper (review)

 Apply Laplace Transforms to obtain an equation relating

�
𝑍𝑍(𝑠𝑠) to 𝐹𝐹(𝑠𝑠), and hence obtain a Transfer Function
𝑞𝑞 𝑘𝑘 𝐹𝐹(𝑡𝑡)
𝑧𝑧̈ + 𝑧𝑧̇ + 𝑧𝑧 =
𝑚𝑚 𝑚𝑚 𝑚𝑚
𝑞𝑞 𝑘𝑘 �
𝐹𝐹(𝑠𝑠)
𝑠𝑠 2 𝑍𝑍 𝑠𝑠 − 𝑠𝑠𝑠𝑠 0− − ̇ −)
𝑧𝑧(0 + −
𝑠𝑠𝑠𝑠 𝑠𝑠 − 𝑧𝑧(0 ) + 𝑍𝑍 𝑠𝑠 =
𝑚𝑚 𝑚𝑚 𝑚𝑚
𝑞𝑞 𝑘𝑘 �
𝐹𝐹(𝑠𝑠)
𝑠𝑠 2 + 𝑠𝑠 + 𝑍𝑍 𝑠𝑠 = �
+ terms independent of 𝐹𝐹(𝑠𝑠)
𝑚𝑚 𝑚𝑚 𝑚𝑚
1
𝑍𝑍 𝑠𝑠 = 𝑚𝑚 �
𝐹𝐹(𝑠𝑠) �
+ terms independent of 𝐹𝐹(𝑠𝑠)
𝑞𝑞 𝑘𝑘
𝑠𝑠 2 + 𝑚𝑚 𝑠𝑠 + 𝑚𝑚

Transfer Function
Dynamic modelling practice example:
mass-spring-damper (review)

 To obtain state-space equations, we need to define state

variables which are sufficient to describe future
behaviour of the system (along with the input):
𝑞𝑞 𝑘𝑘 𝐹𝐹(𝑡𝑡)
𝑧𝑧̈ + 𝑧𝑧̇ + 𝑧𝑧 =
𝑚𝑚 𝑚𝑚 𝑚𝑚
𝑥𝑥1 𝑧𝑧 𝑥𝑥̇ 1 = 𝑥𝑥2
𝒙𝒙 𝑡𝑡 = 𝑥𝑥 =
2 𝑧𝑧̇ 𝑞𝑞 𝑘𝑘 1
𝑥𝑥̇ 2 = − 𝑥𝑥2 − 𝑥𝑥1 + 𝑢𝑢
𝑚𝑚 𝑚𝑚 𝑚𝑚
𝑢𝑢 𝑡𝑡 = 𝐹𝐹 𝑡𝑡
𝑥𝑥̇ 1 0 1 𝑥𝑥1 0
𝑦𝑦 𝑡𝑡 = 𝑧𝑧(𝑡𝑡) = − 𝑘𝑘 𝑥𝑥2 +
𝑞𝑞 1 𝑢𝑢
𝑥𝑥̇ 2 − 𝒙𝒙̇ = 𝑨𝑨𝑨𝑨 + 𝑩𝑩𝑢𝑢
𝑚𝑚 𝑚𝑚 𝑚𝑚
𝑥𝑥1 𝑦𝑦 = 𝑪𝑪𝑪𝑪 + 𝑫𝑫𝑢𝑢
𝑦𝑦 = 1 0 𝑥𝑥
2
State space
Linear Systems & Control (MCEN 3006)
Linear Systems & Control (MCEN 3006)
Linear Systems & Control (MCEN 3006)
 T

T Satellite θ
dynamics

Linear Systems & Control (MCEN 3006)

Dynamic modelling: passive electrical systems

 Define states for each of the energy storage

devices (capacitor voltages, inductor currents)
 Apply Kirchoff’s Current Law, Kirchoff’s Voltage
Law, nodal analysis or mesh analysis to obtain
solvable system of independent equations

� 𝐼𝐼 = 0 � 𝑉𝑉 = 0
node loop
Electrical system example

 Consider the pictured system, with input vi and

output vo .
 Is knowledge of the future input sufficient to predict the
output at any time? What else do we need to know?
 We could call the capacitor voltages vC1 and vC 2 the
states of the system, since they are sufficient to predict
the future outputs of the system along with the future
inputs.
 Electrical system example
Define: 𝑥𝑥1 = 𝑣𝑣𝐶𝐶1
𝑥𝑥2 = 𝑣𝑣𝐶𝐶2
𝑢𝑢 = 𝑣𝑣𝑖𝑖
𝑦𝑦 = 𝑣𝑣𝑜𝑜

 We can use KCL to write a set of ODEs describing this system in

terms of the states 𝑣𝑣𝐶𝐶1 and 𝑣𝑣𝐶𝐶2 .
1 1 1 1 1
𝑣𝑣𝐶𝐶1 − 𝑣𝑣𝑖𝑖 𝑣𝑣𝐶𝐶1 − (𝑣𝑣𝑖𝑖 − 𝑣𝑣𝐶𝐶2 ) 𝑑𝑑𝑣𝑣𝐶𝐶1 𝑥𝑥̇ 1 = − + 𝑥𝑥1 − 𝑥𝑥2 + + 𝑢𝑢
node 2: + + 𝐶𝐶1 =0 𝐶𝐶1 𝑅𝑅1 𝐶𝐶1 𝑅𝑅2 𝐶𝐶1 𝑅𝑅2 𝐶𝐶1 𝑅𝑅1 𝐶𝐶1 𝑅𝑅2
𝑅𝑅1 𝑅𝑅2 𝑑𝑑𝑑𝑑
1 1 1
𝑣𝑣𝑖𝑖 − 𝑣𝑣𝐶𝐶2 − 𝑣𝑣𝐶𝐶1 𝑑𝑑𝑣𝑣𝐶𝐶2 𝑥𝑥̇ 2 = − 𝑥𝑥1 − 𝑥𝑥2 + 𝑢𝑢
node 3: − 𝐶𝐶2 =0 𝐶𝐶2 𝑅𝑅2 𝐶𝐶2 𝑅𝑅2 𝐶𝐶2 𝑅𝑅2
𝑅𝑅2 𝑑𝑑𝑑𝑑
𝑦𝑦 = −𝑥𝑥2 + 𝑢𝑢
1 1 1
𝑥𝑥̇ 1 − + − 𝑥𝑥1 1
+
1
𝐶𝐶1 𝑅𝑅1 𝐶𝐶1 𝑅𝑅2 𝐶𝐶1 𝑅𝑅2
= 𝑥𝑥2 + 𝑢𝑢
𝐶𝐶1 𝑅𝑅1 𝐶𝐶1 𝑅𝑅2
𝒙𝒙̇ = 𝑨𝑨𝑨𝑨 + 𝑩𝑩𝑢𝑢 𝑥𝑥̇ 2 − 𝑅𝑅
1
− 𝑅𝑅
1
0
𝑦𝑦 = 𝑪𝑪𝑪𝑪 + 𝑫𝑫𝑢𝑢 𝐶𝐶2 2 𝐶𝐶2 2
𝑥𝑥1
𝑦𝑦 = 0 −1 𝑥𝑥 + 𝑢𝑢
2 State space
Electromechanical example: DC motor
𝐽𝐽 – moment of inertia
𝐵𝐵 – Damping coefficient
𝐾𝐾 – Motor torque coefficient
𝑅𝑅 – Stator resistance
𝐿𝐿 – Stator inductance

𝑑𝑑𝑑𝑑(𝑡𝑡) � 𝑉𝑉 = 0
𝑒𝑒 𝑡𝑡 = 𝑅𝑅𝑅𝑅 𝑡𝑡 + 𝐿𝐿
𝑑𝑑𝑑𝑑 loop
𝜏𝜏 𝑡𝑡 = 𝐾𝐾𝐾𝐾 𝑡𝑡
𝑑𝑑𝜔𝜔(𝑡𝑡)
𝐽𝐽 = 𝜏𝜏 𝑡𝑡 − 𝐵𝐵𝐵𝐵(𝑡𝑡) � 𝑇𝑇 = 𝐼𝐼𝛼𝛼
𝑑𝑑𝑑𝑑

 Moment of inertia 𝐽𝐽, rotational damping coefficient 𝐵𝐵

 Calculate the transfer function from e(t) to ω(t)
 Hence calculate the impulse and step responses of the system
Dynamic modelling: incompressible fluid tank systems

 Define states for the height of fluid in each tank

 Obtain equations for the flow out of each
valve/aperture as a function of the fluid height in
the corresponding tank
 Write an equation for each tank such that the
rate of change of volume of the tank equals the
net volumetric flow into the tank:
𝑑𝑑𝑉𝑉
= � 𝑞𝑞
𝑑𝑑𝑡𝑡
Fluid tank example
 Aperture flow rates (obtained
from Bernouli’s principle and
some simplifying
assumptions):
𝑚𝑚̇ 2 = 𝛽𝛽1 ℎ1
𝑚𝑚̇ 3 = 𝛽𝛽2 ℎ2
 Find equations relating the
flow into the first tank (input)
to the flow out of the second
tank (output)
 Place the equations into state
space form with respect to ℎ1
and ℎ2 .
 Assume fluid density is 𝜌𝜌
Fluid tank example
Transfer functions and stability
 An LTI system is asymptotically stable if the
impulse response ℎ(𝑡𝑡) satisfies the condition:
∞

∫ h(t ) dt < ∞
0

 An asymptotically stable system guarantees that

the output will be bounded for any bounded
input.
 Conversely, if a system is not asymptotically
stable, there exists a bounded input for which
the output is unbounded.
Transfer functions and stability
 An LTI system is marginally stable if there exists a
constant a such that the impulse response ℎ(𝑡𝑡) satisfies
the condition:
T

∫ h(t ) dt < aT
0

 That is, the integral diverges no faster than linearly.

 An system with an impulse response which settles to a
constant or a sinusoidal steady state is marginally stable.
 A system is unstable if it is neither asymptotically stable
nor marginally stable.
Transfer functions and stability
 A system is unstable if it is neither asymptotically
stable nor marginally stable.

 Caution:
 There are many definitions stability, not all of which
are equivalent. Be careful that you know which one
you are using.
 In this unit when we say a system is stable, we will
usually be referring to asymptotic stability.
Transfer functions and stability
 Consider a system with a rational transfer
function
n( s )
G ( s) =
d (s)
 Further assume that G(s) is proper. This means
deg[n( s )] ≤ deg[d ( s )]

 Note that all physically realizable transfer functions

are proper.
 We can write G(s) in terms of its poles pi by
factorizing the denominator:
Transfer functions and stability
 Suppose we factorize the denominator:
n( s )
G ( s) =
( s − p1 )( s − p2 ) ( s − pn )
 The system will be asymptotically stable if and only if all
of the poles pi have a negative real part.
 Or equivalently if all of the poles lie to the left of the imaginary
axis in the complex plane.
 The system will be marginally stable if it has distinct
poles on the imaginary axis and all other poles lie to the
left of the imaginary axis..
 Otherwise, the system will be unstable
Stability and pole locations
Stability and pole locations
Block diagrams of linear systems
If component systems are linear and represented by transfer functions, we can
rearrange block diagrams in some useful ways:

Series interconnection:

H G G.H

Parallel interconnection:

G
+
G+H
+
H
Block diagrams of linear systems
Feedback interconnection:

+ G
G
1+G.H
-

+ G
G
1-G.H
+

H
Block diagrams of linear systems
 These three simple rules enable very complex
block diagrams to be simplified to a few blocks
or a single block.
Preview: feedback control
 Suppose we have an integrating plant, and wish to
control it with simple proportional feedback?

+ 1
k
s
-

 What is the closed loop transfer function?

k
s+k
 Comment on the stability of the feedback system
 Asymptotically stable for all k >0
Preview: feedback control
 What if the plant is unstable?

+ 1
k
s −1
-

 What is the closed loop transfer function?

k
s + k −1

 Comment on the stability of the feedback system

 Asymptotically stable for all k >1
 More inverted pendulums!

https://youtu.be/-7xvqQeoA8c
More inverted pendulums!

https://www.youtube.com/watch?v=l5I8jaMsHYk
More inverted pendulums!

https://www.youtube.com/watch?v=bvim4rsNHkQ