ELEC2146

Electrical Engineering Modelling and Simulation

Nonlinear Dynamic Systems

Dr Julien Epps
S2, 2011

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 1

 Overview

Nonlinear systems in state space
– Equilibrium points
– Linearization
– Phase portraits

Motivation
– Many practical systems are nonlinear
– Need to model/simulate them is even greater:

Because they do not neatly fall within linear theory

In order to find suitable techniques for analysis

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 2

 Equilibrium

Point or points at which no change
occurs

For a dynamic system:
– All derivatives are zero
– Linear systems:

System stable ⇒ single equilibrium point
– Nonlinear systems:

May have multiple equilibrium points

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 3

 Equilibrium

Nonlinear system example (Klee, 2007):
dx1
= x1 (a − bx1 − cx2 )
dt
dx2  x2 
= x2  d − λ 
dt  x1 

 x2  dx1
x2  d − λ  = 0 ⇒ x2 = ,0
 x1  λ
aλ
x1 (a − bx1 − cx2 ) = 0 ⇒ x1 = ,0
bλ + cd
 aλ ad 

Equilibrium: (0,0) and  , 
trivial  b λ + cd b λ + cd 
Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 4
 Linearisation

Points of equilibrium
– System will revert to equilibrium if perturbed

By ‘small’ amount
– Helpful practically as ‘operating points’
– Choose them as operating points for linearisation

Recall:
– Taylor series approximation to a function f(x)
df x − x0 d 2 f ( x − x0 ) 2
f ( x) = f ( x0 ) + + 2 + ...
dx x = x0 1! dx x = x0
2!
– Locally about an operating point x = x0

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 5

 Linearisation

Taylor series example
– Approximate f ( x) = tan −1 x about x = 0
– Find derivatives:
df 1 d2 f − 2x d 3 f 6x4 + 4x2 − 2
= = =
dx 1 + x 2 dx 2
(1 + x 2 ) 2 dx 3
(1 + x 2 ) 4
– Evaluate at x = 0
df d2 f d3 f
=1 =0 = −2
dx x = 0 dx 2 x =0
dx 3

– Substitute into series (to accuracy required)

x−0 ( x − 0) 2 ( x − 0) 3 1 3
f ( x) = 0 + 1 +0 −2 ... ≈ x − x
1! 2! 3! 3
Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 6
 Linearisation

Taylor series example
– This was for a function
– We are interested in linearising systems
– Principle is same

Linearising DEs
dx
= f ( x, t )
dt
– Define variable ∆x, representing variation about the
operating point x0:
x = x0 + ∆x
– Apply 1st order Taylor expansion to nonlinear terms
in f ( x, t )
– DE is now linear in ∆x (and hence x also)
Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 7
 Linearisation Example

Single-input, single-output example
– Linearise the following equation about the operating
point x = 1:
Note: Need to
use numerical
dx techniques to
= f ( x, u , t ) = −3 + 3 x − 2 ln x − u (t ) estimate
dt equilibrium points
(come out as x =
– Define x = 1 + ∆x , ∆x = x − 1 0.417, x = 1)

d∆x
= −3 + 3(1 + ∆x) − 2 ln(1 + ∆x) − u (t )
dt
– Represent nonlinear terms by their Taylor expansion

1 ( x − 1) 1 ( x − 1) 2
ln x = ln x x =1 + − 2 + ...
x x =1 1! x x =1 2!

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 8

 Linearisation Example

Single-input, single-output example
– Choose first-order approximation
1 ( x − 1)
ln x ≈ 0 + = x − 1 = ∆x
x x =1 1!
– Replace in DE:
d∆x
= −3 + 3(1 + ∆x) − 2(∆x) − u (t )
dt
– Equation is now linear in ∆x
– Once we have solution, substitute ∆x = x − 1 to get
it back in terms of x

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 9

 Linearisation

Typically
– There is more than one input or output or
– There is more than one state variable

In that case
– Equilibrium point is multi-dimensional
– Nonlinearities may exist in more than one state
variable
– Procedure must be generalised using partial
derivatives

Also
– May need more than one linearisation point

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 10

 Linearisation

What about more than one state eqn ?
– Equations may be nonlinear in more than one
(state) variable
– Linearise entire system of equations:

∆x& = A∆x + B∆u

∆ y = C∆ x + D∆ u

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 11

 Linearisation

Rewrite original state-space equations in
form: (note: can now be nonlinear)
dx1
= f1 ( x1 , x2 ,..., xn , u1 , u2 ,..., u m )
dt
dx2
= f 2 ( x1 , x2 ,..., xn , u1 , u 2 ,..., um ) state
dt equations

M
dxn
= f n ( x1 , x2 ,..., xn , u1 , u 2 ,..., u m )
dt
y1 = g1 ( x1 , x2 ,..., xn , u1 , u 2 ,..., um )
output
M equations
y p = g p ( x1 , x2 ,..., xn , u1 , u 2 ,..., u m )
Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 12
 Linearisation

Requires:
 ∂∂xf1 ∂f1
∂x2 L ∂f1
∂xn
  ∂∂uf1 ∂f1
∂u 2 L ∂f1
∂u m
 ∂f 21 ∂f 2 ∂f 2   ∂f 21 ∂f 2 
∂f 2
L L
A =  ∂x1 ∂x2 ∂xn  B =  ∂u1 ∂u 2 
∂u m
M M O M M M O M 
 ∂f n ∂f n ∂f n
  ∂f n ∂f n ∂f n

 ∂x1 ∂x2 L ∂xn  x = x 0 ,u =u 0  ∂u1 ∂u 2 L ∂um 
x = x 0 ,u = u 0

 ∂∂gx1 ∂g1
∂x2 L ∂g1
∂xn
  ∂∂gu1 ∂g1
∂u 2 L ∂g1
∂u m
 ∂g12 ∂g 2 ∂g 2   ∂g12 ∂g 2 ∂g 2
L L
C=  ∂x1 ∂x2 ∂xn  D =  ∂u1 ∂u 2 ∂u m
M M O M  M M O M 
 ∂g n ∂g n ∂g n
  ∂g n ∂g n ∂g n

 ∂x1 ∂x2 L ∂xn  x = x 0 ,u =u 0  ∂u1 ∂u 2 L ∂um 
x = x 0 ,u = u 0
Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 13
 Linearisation

More complex example (Klee, p613)
– Inverted pendulum

– Express DE in state-space form

– Find equilibrium point(s)
– Linearise system

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 14

 Phase Portraits

How do state variables co-vary ?

Phase portrait
– Plot of one state variable vs. another

Produces a contour when plotted at many points in time
– Could plot three, beyond this hard to visualise

Features of interest
– Extreme points ⇒ maximum overshoot
– Circular-style trajectories (orbits)

Periodicity

Oscillation
– Points of convergence (equilibrium points)

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 15

 Phase Portraits

Perfect periodicity
– Trajectory writes over itself

x2

x1

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 16

 Phase Portraits

Sinusoidal periodicity
– Elliptical
– Axes of ellipse depend on amplitudes of state
variables
x2

x1
Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 17
 Phase Portraits

Decaying periodicity
– Spiral
– Heads toward steady-state equilibrium point

x2

x1
Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 18
 Phase Portraits

Nonlinear periodicity
– Other shapes
– Note: definition of periodicity is x(t) = x(t+T)

Not necessarily sinusoidal

x2

x1

– Sustained oscillations on closed path: limit cycles

Dr Julien Epps ELEC2146 Electrical Engineering Modelling and Simulation 19