Вы находитесь на странице: 1из 47

Control Engineering

Dr. Viswanath Talasila


Department of Telecommunication Engineering
Ramaiah Institute of Technology, Bengaluru
Module 12
Lecture 2
Linearization of state space dynamics

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Contents

1) What is linearity?

2) Taylor series and the Jacobian

3) From the Jacobian to the linearized model

4) 2 fun examples !
Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila
Are physical systems linear or nonlinear?

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Ok, but what do we mean by “non-linear” ?
Consider a simple illustrative example below.

y=3x
x y x y

1 3 1 1

2 6 2 4
y=x2
3 9 3 9

4 12 4 16
y=x
5 15 5 25

6 18 6 36

Question: which of the above two functions is non-linear?


To answer this, we use the property of superposition

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Ok, but what do we mean by “non-linear” ?
Superposition:
Adding the signals, x1 and x2 ,we have x = x1 + x2 . And we assume x1=x2

y=3x
x1 , x2 x y= y=
x1 , x y=(x1 + x2 )2 y=(x12 + x22 )
3(x1+x2) 3x1+3x2
x2
1 2 4 2
1 2 6 6
2 4 16 8
2 4 12 12
3 6 36 18
3 6 18 18
4 8 64 32
4 8 24 24
5 10 100 50
5 10 30 30
6 12 144 144
6 16 48 48

Observations: 1) In the second case adding x1


1) In the first case adding x1 and x2 give: 1+1=2. and x2 gives: 2+2=4.
1) Then 3(x1)+3(x2)=6 which is the same as 1) Then (x1 + x2 )2 = 16
2) 3(x1 + x2)=3x = 3*2=6 2) But: (x12 + x22 ) = 8

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Ok, but what do we mean by “non-linear” ?

y=3x

Observations:
1) In the first case adding x1 and x2 give: 1+1=2. 1) In the second case adding x1
1) Then 3(x1)+3(x2)=6 which is the same as and x2 gives: 2+2=4.
2) 3(x1 + x2)=3x = 3*2=6 1) Then (x1 + x2 )2 = 16
2) But: (x12 + x22 ) = 8

In the linear case, giving inputs


directly to a system and computing the In the nonlinear case, this is simply not
final output is the same as adding up true !
the inputs and then giving that to the
system
Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila
Examples of nonlinear systems

Nonlinear elements in circuits Nonlinear mechanical element:


no load curve of a DC machine
(diodes, transistors, transformer spring
etc)

in the beginning there is a linear relationship


between winding mmf and the air gap flux current I through a diode is a non-linear
but after this, saturation has come which function of the voltage V across its
shows the non linear behavior of the curve terminals:

There are thousands of such practical examples of nonlinear systems

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Linearization
• So far we have studied linear systems

• In practice almost all systems are non-linear


• i.e. they do not satisfy the property of superposition

• Analysis and Control of nonlinear systems is painful

• There are many tools, simulation packages etc developed for linear
systems
• Very useful !

• Linearize the non-linear system !

For linearizing a nonlinear system, we need three concepts:


equilibrium points, Taylor series and the Jacobian

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


General form of a (simple) nonlinear system and
equilibrium points

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


General form of a (simple) nonlinear system and
equilibrium points

What happens when we start at any


possible initial condition?

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Equilibrium Points

There are various kinds of equilibria – we will encounter a couple of them in todays lecture

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Why we linearize around an equilibrium point

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Linearizing around an equilibrium point

Let us consider the nonlinear


function shown in the
plot here. How can we
approximate the function?

In general, a function is
represented by a Taylor series.
In other words, we can
approximate the function if we
consider a finite number of
elements of its Taylor series

source: Wikipedia

Linearization is basically an approximation of a nonlinear function


Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila
The Taylor series

where x=a is an equilibrium


point of the differential equation

The derivatives above are w.r.t. x, thus we are looking are partial derivatives !!

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Jacobian - 1

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Jacobian - 2

We observe that the first order


derivatives have the following
matrix structure

How to then handle an nth order state space system?


1. Approximating a two-dimensional function requires two partial derivatives !
2. For an N-dimensional function we will require N partial derivatives

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Jacobian - 2

Given an nth order state space system, the state vector has n states, assume
we have ‘n’ functions as well - then the general Jacobian matrix will be of the
form

Ok, so how do we go from the Jacobian to the linearized state space system?
Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila
From the Jacobian to the Linear State Space system

To go from the computed Jacobian to the state space system is really simple:

1) First compute the equilibrium points of the nonlinear state space system
1) An equilibrium point is simply a solution of our equations when the dynamics are set to 0

2) Substitute the equilibrium point values into the Jacobian

3) The result is the linear A matrix

4) That’s it !!!

Write down Compute the


Compute At any equilibrium
nonlinear Jacobian J
(first order equilibrium point, substitute
state space
derivatives of the points those values into J
equations Taylor series)

The computed linear system A is valid only in a neighbourhood around the


specific equilibrium point – and not anywhere else !

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


From the Jacobian to the Linear State Space system

Write down Compute the


Compute At any equilibrium
nonlinear Jacobian J
(first order equilibrium point, substitute
state space
derivatives of the points those values into J
equations Taylor series)

Compute the Jacobian by using the first


order derivatives of the Taylor series

Tejas

Evaluate the Jacobian at the equilibrium


points and we obtain A

aircraft multiple
linear models

The computed linear system A is valid only in a neighbourhood around


the specific equilibrium point – and not anywhere else !
Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila
Predator-Prey Model

Tiger in the Bandipur forest Male sambar deer in Bandipur forest

Tiger is a symbol of wilderness and well-being


of the ecosystem. By conserving and saving
tigers the entire wilderness ecosystem is
conserved

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Prey Model

Let H(t) represent the number of deers (prey)

In absence of predators, the deer


population keeps increasing

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Prey Model

Let H(t) represent the number of deers (prey)

In absence of predators, the deer


dH
= rh H population keeps increasing
dt

http://faculty.ucr.edu/~hanneman/spatial/pred_prey/hunting_gathering.html
http://www.math.psu.edu/tseng/class/Math251/Notes-Predator-Prey.pdf
http://www.cds.caltech.edu/~murray/amwiki/index.php/Predator_prey

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Prey Model

dH In absence of predators, the deer


= rh H population keeps increasing
dt

Obviously population cannot increase forever !

Let K represent the maximum


population of deers (in the
absence of tigers

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Prey Model

dH In absence of predators, the deer


= rh H population keeps increasing
dt

Obviously population cannot increase forever !

K represents the maximum


dH æ Hö population of deers (in the
= rh H ç1- ÷ absence of tigers and
dt è Kø absence of natural death

For the remainde r of this talk we assume that K ∞ i.e. the forest has infinite resource capacity

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Prey Model
Enter the predators !!

In the presence of predators, the prey population can change dramatically

dH
= rh H - aLH
dt
a is an interaction parameter between the tigers and deers
Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila
The Predator Model

Let L(t) represent the number of tigers (predator)

Predators die without any prey


(exponentially starve to death)

http://www.math.psu.edu/tseng/class/Math2
51/Notes-Predator-Prey.pdf

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Predator Model
Let L(t) represent the number of tigers (predator)

dL
= -rl L Predators die without any prey
dt (exponentially starve to death)

http://www.math.psu.edu/tseng/class/Math2
51/Notes-Predator-Prey.pdf

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Predator Model

Let L(t) represent the number of tigers (predator)

dL Predators die without any prey


= -rl L (exponentially starve to death)
dt

In the presence of prey, the predator population can recover

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Predator Model
Let L(t) represent the number of tigers (predator)

dL Predators die without any prey


= -rl L (exponentially starve to death)
dt

In the presence of prey, the predator population can recover

dL
= -rl L + aLH
dt

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Predator-Prey Model

Two coupled differential equations

dH dL
= rh H - aLH = -rl L + aLH
dt dt
In matrix form we have

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Predator-Prey Model

Two coupled differential equations

dH dL
= rh H - aLH = -rl L + aLH
dt dt
In matrix form we have

é dH ù
ê ú é ùé ù
ê dt ú=ê r -aH H
h
úê ú
ê dL ú ê aL -rl úë L û
ê ú ë û
ë dt û

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Simulation and Analysis of the Predator-Prey equations

Two coupled differential equations

dH dL
= rh H - aLH = -rl L + aLH
dt dt

• In general, every time an “interaction” takes place between


tigers and deers, multiple things can happen
• Deers usually escape (the hunt success rate is usually low)
• When a “proper” interaction happens
• the deer population definitely reduces immediately
• But the tiger population will not increase immediately
• Thus, the interaction term cannot be the same for both
• Further, deers reproduce much faster than tigers !

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Simulation and Analysis of the Predator-Prey equations
Two coupled differential equations

dH dL
= rh H - aLH = -rl L + aLH
dt dt
• In general, every time an “interaction” takes place between tigers and deers, multiple
things can happen
• Deers usually escape (the hunt success is usually low)
• When a “proper” interaction happens
• the deer population definitely reduces
• But the tiger population will not increase
• Thus, the interaction term cannot be the same for both
• Further, deers reproduce much faster than tigers !

Thus the new coupled differential equations are

dH dL
= rh H - a1LH = -rl L + a2 LH
dt dt
with a1 > a2 , rh > rl
Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila
The Jacobian comes back…

dH dL
= rh H - a1LH = -rl L + a2 LH
dt dt

There are two nonlinear terms in this coupled state space system
1) We can compute the equilibrium points
2) Linearize around the equilibrium point

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Jacobian comes back…
Recall the predator-prey model
dH dL
= rh H - a1LH = -rl L + a2 LH
dt dt

What are the equilibrium points of this system?


æ rl rh ö
(H, L) = (0, 0) and (H, L) = ç , ÷
è a2 a1 ø

é r -a L -a1H ù
J =ê ú
h 1
The Jacobian matrix is
êë a2 L -rl + a2 H úû

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


The Jacobian comes back…

é r -a L -a1H ù
The Jacobian matrix is J= ê h 1
ú
êë a2 L -rl + a2 H úû

é a1rl ù
ê 0 - ú
At the equilibrium point æ rl rh ö we have J = ê a2 ú stable
ç , ÷ ê ar ú with eigenvalues center
è a2 a1 ø ê 2h 0 ú
êë a1 úû

Compute the Jacobian at the other equilibrium point: saddle point !

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Saddle point and Center

The center equilibrium occurs when a system has only two


eigenvalues on the imaginary axis

Source: wikipedia
saddle

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Model of the Van der Pol Oscillator
The Van der Pol oscillator equation is an example of an oscillator with nonlinear damping

Physics, engineering and biology:


Action potential of neurons, model
geological faults in seismology, tunnel
diodes and so on.

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Model of the Van der Pol Oscillator
Given the Van der Pol oscillator equation

Let us assume that μ = 0

we get the harmonic oscillator !

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Model of the Van der Pol Oscillator
The Van der Pol oscillator equation is an example of an oscillator with nonlinear damping

Physics, engineering and biology:


Action potential of neurons, model
geological faults in seismology, tunnel
diodes and so on.

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Model of the Van der Pol Oscillator

The Jacobian is

The origin (0,0) is an equilibrium point


and around the equilibrium, the
linearized state space equation is

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Model of the Van der Pol Oscillator
The original non-linear equation was

Around the equilibrium, the linearized state space equation is

The eigenvalues are we have an unstable equilibrium point !!

For μ>0 the equilibrium is always unstable ! In fact, close to the origin the system
is always unstable, but away from the origin, it is always stable – and it goes to a
limit cycle
Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila
Model of the Van der Pol Oscillator
The original non-linear equation was

BUT: we have a limit cycle !!


Around the equilibrium, the linearized state space equation is A limit cycle is a closed trajectory in phase
space such that at least one other
trajectory merges into this as t 

The eigenvalues are

Thus we have an unstable equilibrium point !!


In fact we have a stable limit cycle !

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


State space to transfer function –
the Van der Pol example

The original non-linear equation was

Around the equilibrium, the linearized


state space equation was

The corresponding transfer function is

Step response of the transfer function – really no


correlation with the complete nonlinear system

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


References
• https://www.electrical4u.com/types-of-systems-linear-and-non-linear-system/ (no load curve of DC
machine)
• https://www.allaboutcircuits.com/technical-articles/understanding-i-v-curves-of-non-linear-devices/ Diode
I/V curves
• https://en.wikipedia.org/wiki/Hooke%27s_law Wikipedia, nonlinear spring
• Van der Pol https://sites.oxy.edu/ron/math/341/10/ws/22.pdf

Control Engineering Module 12 – Lecture 2 Dr. Viswanath Talasila


Overview

Summary: Lecture 2
Contents: Lecture 3
 Difference between
linear and nonlinear
 Controllability and
 Taylor series and the
Jacobian Observability
 Simple pole placement
 2 detailed examples of
problem
linearization (predator-
prey and Van der Pol
oscillator

Control Engineering Module 12 – Lecture 1 Dr. Viswanath Talasila

Оценить