Nonlinear Control and Servo systems

Lecture 1

Giacomo Como, 2015

Dept. of Automatic Control

LTH, Lund University

November 2, 2015
Overview Lecture 1

• Practical information
• Course contents
• Nonlinear control systems phenomena
• Nonlinear differential equations
Course Goal

To provide students with solid theoretical foundations of nonlinear

control systems combined with good engineering ability

You should after the course be able to

◮ recognize common nonlinear control problems,
◮ use some powerful analysis methods, and
◮ use some practical design methods.
Today’s Goal

◮ Recognize some common nonlinear phenomena

◮ Transform differential equations to autonomous form,
first-order form, and feedback form
◮ Describe saturation, dead-zone, relay with hysteresis, backlash
◮ Calculate equilibrium points
Course Material

◮ Textbook
◮ Glad and Ljung, Reglerteori, flervariabla och olinjra metoder,
2003, Studentlitteratur,ISBN 9-14-403003-7 or the English
translation Control Theory, 2000, Taylor & Francis Ltd, ISBN
0-74-840878-9. The course covers Chapters 11-16,18. (MPC
and optimal control not covered in the other alternative
textbooks.)
◮ H. Khalil, Nonlinear Systems (3rd ed.), 2002, Prentice Hall,
ISBN 0-13-122740-8. A good, a bit more advanced text.
◮ ALTERNATIVE: Slotine and Li, Applied Nonlinear Control,
Prentice Hall, 1991. The course covers chapters 1-3 and 5,
and sections 4.7-4.8, 6.2, 7.1-7.3.
Course Material, cont.

◮ Handouts (Lecture notes + extra material)

◮ Exercises (can be downloaded from the course home page)
◮ Lab PMs 1, 2 and 3
◮ Home page
http://www.control.lth.se/course/FRTN05/
◮ Matlab/Simulink other simulation software
see home page
Lectures and labs
The lectures (28 hours) are given as follows:

Mon 13–15, E:C Nov 2 – Dec 7

Wed 8–10, E:C Nov 4 – Dec 9
Fri 8-10 E:C Nov 6
Thu 8-10 M:2112B Dec 10
Lectures are given in English.

———————
The three laboratory experiments are mandatory.
Sign-up lists are posted on the web at least one week before the
first laboratory experiment. The lists close one day before the first
session.
The Laboratory PMs are available at the course homepage.
Before the lab sessions some home assignments have to be done.
No reports after the labs.
Exercise sessions and TAs
The exercises (28 hours) are offered twice a week
Tue 15:15-17:00 M:2112B Wed 15:15-17:00 M:2112B
NOTE: The exercises are held in the seminar room of the Automatic
Control Department, M-building, second floor see schedule on home
page.
EXCEPTIONS: (i) for the first two weeks only Wednesday exercise
sessions are scheduled at 13:15–15:00 instead of 15:15-17:00.
(ii) on November 24 and 25 the exercise sessions will be held in a
different room to be announced in due time

Christian Grussler Olof Troeng Mahdi Ghazei

The Course

◮ 14 lectures

◮ 14 exercises

◮ 3 laboratories

◮ 5 hour exam: January 13, 2015, 14:00-19:00, MA10 I-J.

Open-book exam: Lecture notes but no old exams or exercises
allowed.
Course Outline

Lecture 1-3 Modelling and basic phenomena

(linearization, phase plane, limit cycles)

Lecture 2-6 Analysis methods

(Lyapunov, circle criterion, describing functions))

Lecture 7-8 Common nonlinearities

(Saturation, friction, backlash, quantization))

Lecture 9-13 Design methods

(Lyapunov methods, Backstepping, Optimal control)

Lecture 14 Summary
Todays lecture

Common nonlinear phenomena

◮ Input-dependent stability
◮ Stable periodic solutions
◮ Jump resonances and subresonances

Nonlinear model structures

◮ Common nonlinear components
◮ State equations
◮ Feedback representation
Linear Systems

u y = S(u)
S

Definitions: The system S is linear if

S(αu) = αS(u), scaling

S(u1 + u2 ) = S(u1 ) + S(u2 ), superposition

A system is time-invariant if delaying the input results in a delayed

output:
y(t − τ ) = S(u(t − τ ))
Linear time-invariant systems are easy to analyze

Different representations of same system/behavior

ẋ(t) = Ax(t) + Bu(t), y(t) = Cx(t), x(0) = 0

Z
y(t) = g(t) ⋆ u(t) = g(r)u(t − r)dr
Y (s) = G(s)U (s)

Local stability = global stability:

Eigenvalues of A (= poles of G(s)) in left half plane
Superposition:
Enough to know step (or impulse) response
Frequency analysis possible:
Sinusoidal inputs give sinusoidal outputs
Linear models are not always enough

Example: Ball and beam

mg sin(φ)

φ
mg

Linear model (acceleration along beam) :

2
Combine F = m · a = m ddt2x with F = mg sin(φ):

ẍ(t) = g sin(φ(t))
Linear models are not enough

x = position (m) φ = angle (rad) g = 9.81 (m/s2 )

Can the ball move 0.1 meter in 0.1 seconds with constant φ?
Linearization: sin φ ∼ φ for φ ∼ 0
(
ẍ(t) = gφ
x(0) = 0

t2
Solving the above gives x(t) = 2 gφ
2∗0.1
For x(0.1) = 0.1, one needs φ = 0.1 2 ∗g ≥ 2 rad
Clearly outside linear region!
Contact problem, friction, centripetal force, saturation

How fast can it be done? (Optimal control)

Warm-Up Exercise: 1-D Nonlinear Control System

ẋ = x2 − x + u

◮ stability for u = 0?
◮ stability for constant u = b?
◮ stability with linear feedback u = ax + b?
◮ stability with non-linear feedback u(x) =?
Stability Can Depend on Amplitude

Motor Valve Process

r 1 1 y
+ ? 2
s (s+1)

−1

Valve characteristic f (x) =???

Step changes of amplitude, r = 0.2, r = 1.68, and r = 1.72
Stability Can Depend on Amplitude

Motor Valve Process

r 1 1 y
+
s (s+1)2

−1

Valve characteristic f (x) = x2

Step changes of amplitude, r = 0.2, r = 1.68, and r = 1.72
Step Responses
0.4
Output y

0.2

0
0 5 10 15 20 25 30
4 Time t
Output y

0
0 5 10 15 20 25 30
10 Time t
Output y

0
0 5 10 15 20 25 30
Time t
r = 0.2

r = 1.68
Stable Periodic Solutions

Example: Motor with back-lash

0 1
5 y
Constant 5s 2+s
Sum P−controller Backlash
Motor

−1

1
Motor: G(s) = s(1+5s)
Controller: K = 5
Stable Periodic Solutions

Output for different initial conditions:

0.5

Output y
0

−0.5
0 10 20 30 40 50
Output y 0.5 Time t

−0.5
0 10 20 30 40 50
0.5 Time t
Output y

−0.5
0 10 20 30 40 50
Time t

Frequency and amplitude independent of initial conditions!

Several systems use the existence of such a phenomenon
Relay Feedback Example
Period and amplitude of limit cycle are used for autotuning

PID

A y
Σ u
Process
T
Relay

−1

1
u
y
0

−1

0 5 10
Time

[ patent: T Hgglund and K J strm]

Jump Resonances

20
y
Sine Wave 5s 2+s
Sum Saturation Motor

−1

Response for sinusoidal depends on initial condition

Problem when doing frequency response measurement
Jump Resonances

u = 0.5 sin(1.3t), saturation level =1.0

Two different initial conditions
6

2
Output y

−2

−4

−6
0 10 20 30 40 50
Time t

give two different amplifications for same sinusoid!

Jump Resonances

Measured frequency response (many-valued)

1
10

lin
ea
0

r
10

sa
sa

tu
tu

ra
ra

te
te

d
Magnitude

d
−1
10

−2
10

−3
10 −1 0 1
10 10 10
Frequency [rad/s]
New Frequencies
Example: Sinusoidal input, saturation level 1

a sin t y
Saturation

0
10 a=1 1
Amplitude y

−2
10 −1

0
−2
10 0 1 2 3 4 5
Frequency (Hz) Time t

0
10 a=2 1
Amplitude y

−2
10 −1

0
−2
10 0 1 2 3 4 5
Frequency (Hz) Time t
New Frequencies
Example: Electrical power distributionP
∞
energy in tone k
THD = Total Harmonic Distortion = k=2 energy in tone 1

Nonlinear loads: Rectifiers, switched electronics, transformers

Important, increasing problem
Guarantee electrical quality
Standards, such as T HD < 5%
New Frequencies

Example: Mobile telephone

Effective amplifiers work in nonlinear region

Introduces spectrum leakage

Channels close to each other

Trade-off between effectivity and linearity

Subresonances

Example: Duffing’s equation ÿ + ẏ + y − y 3 = a sin(ωt)

0.5

y
0

−0.5
0 5 10 15 20 25 30
Time t
1
a sin ωt

0.5

−0.5

−1
0 5 10 15 20 25 30
Time t
When is Nonlinear Theory Needed?

◮ Hard to know when - Try simple things first!

◮ Regulator problem versus servo problem
◮ Change of working conditions (production on demand, short
batches, many startups)
◮ Mode switches
◮ Nonlinear components

How to detect? Make step responses, Bode plots

◮ Step up/step down
◮ Vary amplitude
◮ Sweep frequency up/frequency down
Some Nonlinearities
Static – dynamic

u
|u| e

Abs Math
Saturation
Function

Look−Up
Sign Dead Zone
Table

Relay Backlash Coulomb &

Viscous Friction
Nonlinear Differential Equations

Problems
◮ No analytic solutions
◮ Existence?
◮ Uniqueness?
◮ etc
Finite escape time

Example: The differential equation

dx
= x2 , x(0) = x0
dt
has solution
x0 1
x(t) = , 0≤t<
1 − x0 t x0
Finite escape time
1
tf =
x0
Finite Escape Time

Finite escape time of dx/dt = x2

5

4.5

3.5

2.5
x(t)

1.5

0.5

0
0 1 2 3 4 5
Time t
Uniqueness Problems
√
Example: The equation ẋ = x, x(0) = 0 has many solutions:
(t − C)2 /4 t > C

x(t) =
0 t≤C
2

1.5

1
x(t)

0.5

−0.5

−1
0 1 2 3 4 5
Time t

Compare with water tank:

√
dh/dt = −a h, h : height (water level)
Change to backward-time: “If I see it empty, when was it full?”)
Local Existence and Uniqueness
For R > 0, let ΩR denote the ball ΩR = {z : kz − ak ≤ R}.

Theorem
If, f is Lipschitz-continuous in ΩR , i.e.,

kf (z) − f (y)k ≤ Kkz − yk, for all z, y ∈ ΩR ,

then (
ẋ(t) = f (x(t))
x(0) = a
has a unique solution

x(t) , 0 ≤ t < R/CR ,

where CR = maxx∈ΩR kf (x)k

Global Existence and Uniqueness

Theorem
If f is Lipschitz-continuous in Rn , i.e.,

kf (z) − f (y)k ≤ Kkz − yk, for all z, y ∈ Rn ,

then
ẋ(t) = f (x(t)), x(0) = a
has a unique solution

x(t) , t ≥ 0.
State-Space Models

◮ State vector x
◮ Input vector u
◮ Output vector y

general: f (x, u, y, ẋ, u̇, ẏ, . . .) = 0

explicit: ẋ = f (x, u), y = h(x)
affine in u: ẋ = f (x) + g(x)u, y = h(x)
linear time-invariant: ẋ = Ax + Bu, y = Cx
Transformation to Autonomous System

Nonautonomous:
ẋ = f (x, t)
Always possible to transform to autonomous system
Introduce xn+1 = time

ẋ = f (x, xn+1 )
ẋn+1 = 1
Transformation to First-Order System

dk y
Assume highest derivative of y
dtk h iT
Introduce x = y dy dk−1 y
dt . . . dtk−1

Example: Pendulum

M Rθ̈ + kθ̇ + M gR sin θ = 0

 T
x= θ θ̇ gives

ẋ1 = x2
k g
ẋ2 = − x2 − sin x1
MR R
A Standard Form for Analysis

Transform to the following form

Nonlinearities

G(s)
Example, Closed Loop with Friction

F
Friction

_
0 u v
C G
_

⇐⇒

Friction

−G
1+CG
Equilibria (=singular points)

Put all derivatives to zero!

General: f (x0 , u0 , y0 , 0, 0, 0, . . .) = 0
Explicit: f (x0 , u0 ) = 0
Linear: Ax0 + Bu0 = 0 (has analytical solution(s)!)
Multiple Equilibria

Example: Pendulum

M Rθ̈ + kθ̇ + M gR sin θ = 0

Equilibria given by θ̈ = θ̇ = 0 =⇒ sin θ = 0 =⇒ θ = nπ

Alternatively,

ẋ1 = x2
k g
ẋ2 = − x2 − sin x1
MR R
gives x2 = 0, sin(x1 ) = 0, etc
Next Lecture

◮ Linearization
◮ Stability definitions
◮ Simulation in Matlab