Control Systems

Ina Taralova

Ecole Centrale de Nantes

E-mail : [email protected]

1
Dynamical system

2
gaz

liquid

Physical
process
 Modeling, identification,
dynamical system and control

3
gaz
inputs outputs
liquid

Physical
process
The process is the device or the plant or the system under control.
The input and output relationships are defined by the cause and effect
correlation of the process
 First european temperature regulator of
Cornelis Drebbel (1572-1633) from Holland

Drebbel invented the first regulator - thermostat, which

used a column of mercury and a system of floats and
levers to maintain a steady temperature within a
furnace. He later invented an incubator for hatching
eggs which used the same principle for temperature
regulation

Read more:
http://www.madehow.com/inventorbios/16/Cornelius-
Van-Drebbel.html#ixzz4KWqsUIRe
Watt’s Flyball Governor
(18th century)

5
Examples of Modern Control Systems

Manual control for fluid level regulation using the valve : the
human operator observes the fluid level and adjusts the debit
accordingly
Design Example: blood glucose and insulin evolution for a healthy person

7
For all these, we need modeling!
• A model is
– a set of relations (equations), allowing to
describe the behaviour of a (dynamical) system
• All models are imperfect; according to the
problem to be solved, certain features are
taken into account, whereas others are
neglected

8
 Why the modeling?

Natural sciences, physics,

• To simulate the behaviour of a system, in order
to be able to:
– Analyze and understand its behaviour

engineering, ...
– Predict its future evolution (tsunamis, population
of sea lions …)
– Estimate the parameters - or the states - which are
not directly observable or mesurable
– Control the system to reach a desirable behaviour:

Engineering
(feedback) control systems
– Optimize the system’s behaviour (by changing its
parameters - electronic circuits; control input, etc)
– Determine the faults in case of bad functioning, etc 9
 Classes of models
(based on the system’s knowledge)

Known Inknown

Structure; Parameters Black box

Structure; Parameters

Structure; Parameters Grey box

Structure; Parameters

Structure; Parameters White10box

 Classes of models
(based on the system’s knowledge)

• Statical - dynamical
• Deterministic - stochastic
• Continuous - discrete
• Distributed parameters - fixed parameters
• Stationnary - non-stationnary
• Linear - non-linear
• …..
11
 Identification

• Process of obtaining the model

(mathematical) of the system using data
from observation, and - if possible - the a
priori knowledge about the system

12
Control theory studies also
dynamical systems.
What is different?

13
 Control system
• Regulation : the system has to follow a
constant set point - ex. the temperature of
the shower, whatever the “perturbations”
(leaking cold water tap, shower head, etc)
• Tracking : the system follows a trajectory in
time (the trajectory of a car, plane, shuttle,
...)
• The dynamical model has inputs (e.g. the
set point, or the reference trajectories to be
followed), and outputs (e.g. the measurable
variables) 14
 Control systems are input - output systems

uU Dynamical yY

System

f
y=f(u)
y

u
In a steady-state regime, for each input u there exists a unique output
15 y
 Modeling (example)
• Model of an impedance

U linear U non-linear Z As a function of

the frequency

I I f
DC AC

16
 Linear system =>
superposition principle
 f(au1+bu2)=a f(u1)+b f(u2)

• Necessary and sufficient condition:

 f(u1+u2)= f(u1) + f(u2) f is additive
 f(au)=a f(u) f is homogeneous

17
 Example

u1 Linear y1
System

u2 Linear y2
System

u1  u 2 Linear
y1  y 2
System
18
 What if the system is inherently
not linear?
• 1. The nonlinear system may be locally
approximated (modelled) by a linear one,
and then apply the linear system theory (see
next slides)
• 2. Otherwise, analyze it using the nonlinear
dynamical systems theory (e.g. Master
CORO or EMARO courses of Ecole
Centrale de Nantes)
19
 How to represent a nonlinear input-
output system by a linear one
• Suppose a linear input-output relation for one
portion of the characteristics y=f(u) , and then
restrict the system functioning to that region

f
y=f(u)
y

20
u
 Linearizing
• By a “local” approximation of the nonlinear
system by a linear one

x0
Approximation by the tangent line at the point x0 (tangent model):
as many models as values x0 21
In the following slides, we shall
consider linear (or linearized)
systems only

22
In control theory : Input-output dynamical model

uU System yY

(process)

dn y di y d y du dmu
a n n  ...  a i i  ...  a1  a 0 y  b 0u  b1  ...  b m m
dt dt dt dt dt

23
 Input-Output Dynamical model

uU yY
System
(process)

dn y di y dy du dmu
a n n  ...  a i i  ...  a1  a 0 y  b 0u  b1  ...  b m m
dt dt dt dt dt

• Physical realisation requires m<= n, where

n is the system order
24
 To simplify the calculations, let transform
the differential equation to an algebraic one:

dn y di y dy du dmu
a n n  ...  a i i  ...  a1  a 0 y  b 0u  b1  ...  b m m
dt dt dt dt dt

Laplace transform

ansny(s)  ...  aisi y(s)  ...  a1sy(s)  a0y(s)  b0u(s)  b1su(s)  ...  bmsmu(s)

• Supposing y(0)=dy(0)/dt=…=dn-1y(0)/dtn-1 =0 et
25
u(0)=du(0)/dt=…=dm-1u(0)/dtm-1 =0
 Definition of the Laplace
transform
• Let f(t) be defined for t>0 and zero for t<=0;
let s be a complex variable; F(s) is then the
Laplace transform of f(t) defined by

st
F(s)  L[f ( t )]   e f ( t )dt
0
• This transform is bijective, i.e. the inverse
transform exists and it is given by
1
[f (t )]  L [F(s )] 26
 Nice properties of the Laplace
transform:
• It is a linear transform:
L[f1 (t )  f 2 (t )]  F1 (s )  F2 (s )
L[a.f (t )]  a.F(s )
• The Laplace transform of a derivative is
given by:
df (t )
L[ ]  s.F(s )  f (0)
dt
27
Laplace transform: some theorems
• Theorem of the initial value
lim f (t )  lim s.F(s )

t0 s
• Theorem of the final value
lim f (t )  lim s.F(s )
t s0
• Theorem of the delay
L[f ( t  T)]  e  sT F(s )
L1[F(s  a )]  e at f (t ) 28
 Laplace Transform Properties
Addition/Scaling L[af 1 (t )  bf 2 (t )]  aF1 ( s )  bF2 ( s )

d 
Differentiation L  f (t )  sF ( s )  f (0)
 dt 
Integration 
L  f (t )dt  
F ( s) 1
s
  f (t )dt
s
 t 0

t
Convolution  f (t  τ)f (τ )dτ  F (s) F (s)
0
1 2 1 2

Initial-value theorem f (0 )  lim sF ( s )

s 

Final-value theorem lim f (t )  lim sF ( s )

t  s 0
29 29
 Laplace Transforms of Common
Functions
Name f(t) F(s)
1 t0
Impulse f (t )   1
0 t0
1
Step f (t )  1
s
1
Ramp f (t )  t
s2
1
Exponential f (t )  e at
sa
1
Sine f (t )  sin(t )
 2  s2
30 30
 Transfer functions

U (s ) Y(s )
G(s)

ansny(s)  ...  aisi y(s)  ...  a1sy(s)  a0y(s)  b0u(s)  b1su(s)  ...  bmsmu(s)

b1 bm m
m m1 1  s  ...  s
Y(s) bms  bm1s  ...  b1s  b0 b0 b0 b0
G(s)   n1
 .
U(s) n
ans  an1s  .... a1s  a0 a0 1  a1 s  ...  an sn
a0 a0

K = G(0) = b0/a0 is the static gain of the system

31
 Examples of transfer functions W(s)
• RC Circuit
i
R

VI C VO

Laplace
vI ( t )  R * i( t )  vo ( t )  VI ( s )  R * I ( s )  Vo ( s )
1 Laplace 1 I( s )
vo ( t )   i( t )dt Vo ( s )  *  I ( s )  s * C * Vo ( s )
C C s
VI ( s )  R* s * C * Vo ( s )  Vo ( s )  ( 1  s * R * C )* Vo ( s )
Vo ( s ) 1
 W( s )  , where T  R * C is the time constant
VI ( s ) 1  s* T 32
Transfer function of a RC Circuit

i
R
VI VO
C

1 Vo output
W (s)   
1  s * T Vi input 33
 Examples of transfer functions W(s)
• RLC Circuit L
i
R

VI C VO

di(t ) Laplace
vI (t )  R *i(t )  L *  vo (t )  .......................
dt
1 Laplace 1 I (s)
vo (t )   i(t )dt  Vo (s)  *  .................
C C s
VI (s)  (R  L * s) * s *C *Vo (s) Vo (s)  .......................
Vo (s)
 W (s)  ..............
VI (s)
34
where T1  R *C , T2  L *C are the time constants
 Examples of transfer functions W(s)
• RLC Circuit L
i
R

VI C VO

di( t ) Laplace
vI ( t )  R* i( t )  L*  vo ( t ) VI ( s )  ( R  L* s )* I ( s )  Vo ( s )
dt
1 Laplace 1 I( s )
vo ( t )   i( t )dt Vo ( s )  *  I( s )  s* C * Vo ( s )
C C s
VI ( s )  ( R  L* s )* s* C* Vo ( s )  Vo ( s )  ( 1  s* R* C  s 2 * L* C )* Vo ( s )
Vo ( s ) 1
 W( s )  where T1  R* C , T2  L* C are
VI ( s ) 2
1  s* T1  s * T2
35
the time constants
 Study of a DC motor
Constant inductor flux

i
f
v(t) J
(t)

R L

i f
v(t) + + J
(t)
- e(t)

36
Example of a train

37
 R L

i f
+
v(t) J
- e(t) (t)

di (t ) 
v(t )  e(t )  Ri (t )  L  electrical equation
dt 
d (t ) 
J  cm (t )  f . (t ) 
dt 
2  mechanical equation
d  (t ) d (t ) 
J  cm (t )  f .
dt 2 dt 
e(t )  kem (t ) 
 equations of the motor
cm (t )  kemi (t ) 38
 Find the transfer functions
(we assume zero IC)

V (s)  L[ v( t )] ? (s)  L[( t )]

G1(s)

V(s)  L[ v( t )] (s)  L[( t )]

?
G2(s)

IC : initial conditions 39
 Transfer function of a DC motor G1(s)=(s)/V(s)
• when applying the Laplace transform to the input-output differential equation

di( t ) 
v( t )  k em ( t )  Ri ( t )  L
dt  v( t )  k ( t )  R  f .( t )  J d( t )   L  f d( t )  J d ( t ) 
2

1  d( t )   em    dt 2 
i( t )   f .( t )  J   k em  dt  k em  dt 
k em  dt  

2
k em  Rf RJ  Lf d( t ) LJ d 2 ( t )
v( t )  ( t )  
k em k em dt k em dt 2

 d( t )   d 2 ( t )  2
• transfer function, zero IC => L    s et L  2 
s 
 dt   dt 

k em
2
 (s ) k em  Rf
G 1 (s )  
V (s ) 1  RJ  Lf s  LJ
s 2
2 2
k em  Rf k em  Rf 40
Transfer function of a DC motor G1(s)=(s)/V(s)
• from the bloc diagram

di( t ) 1
(s)  C m (s)
v( t )  e( t )  Ri( t )  L V (s)  E(s)  R  Ls  I(s) f  Js 
dt
e( t )  k em ( t ) E (s)  k em (s) C m (s)  k em I(s)
c m ( t )  k em i( t ) C m (s)  k em I(s) 1
I(s)  V(s)  E(s) 
J
d( t )
 c m ( t )  f .( t ) C m (s)  f  Js  (s) R  Ls 
dt E (s)  k em (s)

V(s) 1 I(s) Cm(s) 1 (s)

k em
R  Ls f  Js 
E(s)
k em

k em k em

G 1 (s ) 
 (s )

R  Ls f  Js 

2
k em  Rf
V (s) 1  kem k RJ  Lf LJ
em
1 2 s 2 s2
R  Ls f  Js  k em  Rf k em  41
Rf
 Transfer function of a DC motor G2(s)=(s)/V(s)
• when applying the Laplace transform to the input-output differential equation

d( t ) di( t ) 
v( t )  k em  Ri( t )  L
dt dt 
1  d( t ) d 2 ( t )   d( t ) R  d( t ) d 2 ( t )  L  d 2 ( t ) d 3( t ) 
i( t )  f. J   v( t )  k em  f. J  f 

k em  dt 2 
dt   dt 
k em  dt 2 
dt  k em  dt 2  J dt 3 
 
di( t ) 1  d 2 ( t ) d 3( t ) 
 f. J 3 

dt 
k em  dt 2
dt 
2
k em  Rf d( t ) RJ  Lf d 2 ( t ) LJ d 3( t )
v( t )   2

k em dt k em dt k em dt 3
 d(t)   d2(t)  2  d3(t)  3
• Transfer function zero IC => L   s , L   s  et L 3   s 
2
 dt   dt   dt 

k em
2
 (s ) k em  Rf
G 2 (s )  
V (s )  RJ  Lf LJ 2

s 1  2 s 2 s  42
 k em  Rf k em  Rf 
 Transfer function of a DC motor G2(s)=(s)/V(s)
• supposing L neglectable (very small)

d( t ) 
v( t )  k em  Ri( t ) 
dt  d( t ) R  d( t ) d 2 ( t ) 
v( t )  k em  f. J 
1  d( t ) d 2 ( t )  dt 
k em  dt 2 
dt 
i( t )  f.  J  
k em  dt dt 2 

RJ d 2 ( t ) Rf  k 2 em d( t )
v( t )  2

k em dt k em dt

 d(t)   d2(t)  2
• transfer function, zero IC => L   s et L 2   s 
 dt   dt 
k em
2
 (s ) k em  Rf
G 2 (s )  
V (s )  RJ 

s1  2 s  43
 k em  Rf 
 Transfer Functions
Any initial conditions (not zero)
At t=0+ the input and the output and their derivatives are  0

(a nsn  .... a1s  a 0 )Y(s) - J(s)  (bmsm  ... b1s  b0 )U(s) - I(s)

• J(s) depends on ai and the initial conditions on y

• I(s) depends on bj and the initial conditions on u

J (s)  I (s) N(s)

 Y (s)  G(s) U (s)  with G(s) 
D(s) D(s)
• If u(t), G(s) and the initial conditions on u et y are known:
one calculates Y(s) and after y(t)=L-1[Y(s)]
44
 Transfer Functions
Any initial conditions (not zero)
At t=0+ the input and the output and their derivatives are  0
J (s)  I (s) N(s)
N.B. Y (s)  G(s) U (s)  avec G(s)  (1)
D(s) D(s)
From (1) we can deduce
 J (s)  I (s) 
Y (s)  G(s) U (s)   (2)
 N(s) 
• It is equivalent to consider that the initial conditions are
zero and the input is
1  J(s)  I(s) 
 u(t)  L    u(t)  u0 (t)
 N(s) 
45
Fake input
 Stability of linear systems
• Definition
– A linear system with input u and output y is stable

u (t )  0;  initial conditions / y
y (t )  0
when t  
U (s) Y (s)
u(t) G(s) y(t) y

y0

t 46 t
0 0
 Stability of linear systems
• Stability criterion
– A linear system with input u and output y is stable

– All poles (roots of the denominator) of the transfer

function W(s) are negative (or have a negative real part,
if complex)
– N.B. Poles can be tuned via (feedback control) pole
placement, by means of a regulator K(s):
T(s) = W(s)/(1+K(s)*W(s))
47
 u y

time time

 The system reaction to the step function u(t) is called

a step response y(t).
 Can you predict qualitatively the shape of the step response,
if you know the transfer function W(s)?

48
 Time (step) response of dynamical systems

Peak time tp -time to reach the to reach the first peak of the overshoot ymax
y(t) Settling time ts - time to reach 2% or 5% of the final value y∞

ymax
y∞ ± 2% (or 5%) of the final value y∞

tp ts time

The overshoot could be defined in percentage as

X1 = (ymax - y∞)/ y∞
NB The maximum required overshoot is generally around 25%
49
 Time response of dynamical systems – static error
er -static error defined by : er = (u(t) - y(t))/u(t) when t => ∞ (in practice after the settling time) and u(t) step function
Settling time ts - time to reach 2% or 5% of the final value y∞ ( Static error in % may also be used )

y(t)

step er
y∞

ts time

The static, or steady-state error is the error after the transient

response has decayed leaving only the continuous response
50
 How to find time response from the
transfer function ?

Yr ( s )  1 / s Y (s )

t 
51
http://web.iku.edu.tr/courses/ee/ee670/ee670/lecture%20notes/ch04.pdf
 Unit step response of first order systems
Yr ( s )  1 / s Y (s )

Expand the response Y(s) into partial fractions : Y(s) = …

52
53
 Transient Response

• Estimates the shape of the curve based on

the foregoing points on the x and y axis
• Typically applied to the following inputs
– Impulse
– Step
– Ramp
– Quadratic (Parabola)
First Order System
Impulse K Exponential
response
1  sT
Step response Step,
K K
- exponential
s s  1/ T
Ramp response K KT KT Ramp,
- - step,
s 2
s s  1 / T exponential
No oscillations (as seen by poles) 55
55
 First Order Feedback System
Y (s) K K
 
R ( s ) 1  K  sT 1  sT

Reference
R (s ) E (s ) U (s ) 1 Y (s )
S K
1  sT

B (s ) 1
 Unit step response of second order systems, real poles

Yr ( s )  1 / s Y (s )

G(s)
Expand the response Y(s) into partial fractions : Y(s) = …

Take the inverse Laplace transform of each fraction : y(t) = …..

Calculate the static (steady-state) error after the transient has decayed

e(t) = step(t) – y(t) => e(∞) = …….

Therefore, the percentage of static error is …… %

57
Show that the static error is defined by G(0) = ….
 1
Unit step response of second order systems W ( s) 
2 1
1 s s2
 n
 2
n

58
 Second Order System: Parameters

Interpretation of damping ratio

  0 : Undamped oscillation (Re  0, Im  0)
0    1 : Underdamped (Re  0  Im)
1   : Overdamped (Re  0, Im  0) Critically damped for   1

Interpretation of undamped natural frequency

n gives the frequency of the oscillation
60
61
 Second Order System

Y (s) K  N2
Impulse response :  2  2
R( s) Js  Bs  K s  2N s   N2
Oscillates if poles have non - zero imaginary part (ie, B 2  4 JK  0)
B
Damping ratio :   where Bc  2 JK
Bc
K
Undamped natural frequency :  N 
J

62
Second order system canonical form
1
1  2 /  N * s  1 /  N2 * s 2

where
tp   /( N * sqrt (1   ))
2

X 1  exp( / sqrt (1   ) 2

63
Second order system time specifications

64
Yes, you can! The poles location completely defines the step response :
 Stable (LHP) or unstable (RHP)
 Oscillatory (complex poles), aperiodic (real poles) or constant (poles
at the origin)

65
 Discrete transfer functions
When the signal values are defined in discrete time

yk

Sampling time tk-tk-1

0 1 2 3 4 5 6 7 8 k

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

tk=t [s]
Sampling time 0.1s 66
U( z ) Y( z )
G(z)

67
Discretization is achieved via Analog to
Digital (A/D) converters with sampling

68
Z transform discrete time systems
Definition of Z- transform :

Some Z-transform nice properties :

69
 Example
Calculate the z-transform for the unit step function
Solution.

70
 Step invariance method for discretisation
(MatLab ‘zoh’ – zero order hold option)
Principle : Same step (time) response at sampling instants

Apply z transform on both sides :

And accordingly ([1])

So finally

71
[1] Ogata, K. Discrete-Time Control Systems , Englewood Cliffs, NJ: Prentice-Hall, 1987
Calculate the z-transform for the following transfer function

1. Recall that

2. Calculate by partial fraction expansion :

3. Substitute t=kT and take Z- transform (you may use Z – transform tables)

4. Multiply by

72
Calculate the z-transform for the following transfer function

1. Recall that

2. Calculate by partial fraction expansion :

3. Substitute t=kT and take Z- transform (you may use Z – transform tables)

4. Multiply by

73
74
 Tutorial exercise

For the following second order system, identify the

canonical form (on the white board)

75