0% found this document useful (0 votes)
11 views7 pages

Week 2

Uploaded by

gosaliowen
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
11 views7 pages

Week 2

Uploaded by

gosaliowen
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

Models in Control system Design

·
Modelling
-model :

abstraction of system useful for analyas/design

to
-resequl mentally distinguish the
Following
:

* real
physical system
based deal
Ideal
*

physical model on interconnected blocks

eq resistors , masses , beams , Isotropic media ,

Newtonran Fluids , electrons , etc .

model
A Ideal mathematical From applying laws
of physics
(nonlinear partiall differential equanious (

A amplified math . Model From approximating idea model

cynear constant -
coefficient differential equations)
·
uncertainty in models

-models are never a


perfect representation of
system
a .

-uncertainty
: cannot predict exact system response to inputs
* unknown unmodeled Inputs :

eg . disturbance imples

* unknown / unmodeled dynamics


models
Despite this are still useful
-

-models allow us to

* generalize from observations of system behaviour to


predict
system behaviour in new situations .

exploit mathematical A
computational tools to
develop prototype
*

designs withou breaking our


physical system .

· Two modelling Approaches


-

Analytical modelling :

models From fundamental physics principle


* Explo centures
of work
modelling physical phenomena
* Idealised models based on differential equations ,
with
parameters
that depend on measured quanturies .

-Models From data :

useful it .
* cannot early) measure model parameters

* do not have good Fundamental science medels (eg complex


brological systems)
·
simple Models

-linear constant cofferent ODE

* Lil systems (With O Initial conditions (

* Transfer Functions
-

why :

* Approximate dynamics of wide


range
of
physical systems ,
over

wide range of Inputs


and
operating conditions .

of analysis
* Balance expressiveness and ease

A Engineered systems often designed to Follow such models

leg controllers) .

* In control often the arm is a system in


operating
, to
keep
region where dynamics are
approximately linear

* Avid Over Fitting" Cit modelling From data)

Analytical Modelling
·
Recall
:

Linear constant coefficient ordinary differential equations (LCCDEs)

Input u
output y related by
-

daray-itadNote dt
M-1
a ene

together with initial conditions on derivatives of ufy at 0


U
y
Alte

Transfer FUnCLOU :
-

bism+bm-sm-+
#
Giss= =
..
+
bist bo

s"+an- s + .. + ais+ do

UCS)- - Yes
* take LT
of both sides (with initial conditions (
=
0

* Mearrange to get YCs/UCS) =

transfer Function

· Basic Principles

Dynamics often anse From transfer/loss/storage of energy


-

Approach Identify low of mass /energy and conservation law


:

apply
-

A conservation of mass

*
conservation o energy

* conservation o charge

* newton's law of motion

- In control : also identity/choose inputs & outputs

Example 1 Simple RIC archit


:

L R
-

VCt

Lricts -T
ict)

-choice of inputs A outplets


:

VIt) Input , Vacts ourpret

-which physical law to KVL


apply
:

conservation
:

of energy

-
gives
relationship

rCt) =

V(t) +
reCt) + Ucht)

cat : iCt) , VRL= RICt) , UCt=


LA
Lc d
URLtS Red ViCtS =

=>
,
=

At

Lc d Red
: rct) =

+ Vet

vi =

est
=
V

+(Y)
·
Example

ma nameAtte cit
2
:

Ev

input applied Forsef output position X


:
-

,
:

~which physical law to apply


:

Newton 2nd law E =

me
-

gives relationship :

Hooke's law
- x(t)
f(t-kXCE) Max
tra
-

xxc
=

I At
2
-> Get
I

tut FCS) -

KXCS) -
strXcs) =
ms XCs)

=frs+k
System Identification
· system Identication

-
Arm
:

End a model of a dynamical system (ie .


differential

)
equation relating Input 4 and output

-
Given :

Input
*
signal(s) :
4 .,
H .. ... UN and

A (Noisy) measured output signals :

40 ...... YN

⑮ - >

I monel-
noisy measurements -

Trade-off between Fitting


-

the data but not over Ittng


·
Three main steps in system (D

Determine model structure


-

* From physics (based on cstem knowledge) , or


eq mass-spung-damper system

physics-> 2nd order system is


appropriate
* Generic Cample model 2"order LCCDE based
,
eg .

on measured response .

-choose test inputs :

* LT) systems :

Often choose unit step or sinusoids at

different Frequencies -
frequency response

Generally : want inputs that excite dynamics of system

-
It parameters of model :

* Graphically (simple cases , no noise , etel

*
solving an optimisation problem (In generall
· General Strategy
experimental

->
- l/
y data

I
-

L
W stem

F-
-S want this to be small

ene
could be
multiple ,
mputs
predicted output
depends on parameters

Search over
parameters & unty predicted response 48 closely
matches measured response y .

Need to be able to
compute response predicted by model Ye

Need a
way to measure how close aquals y and Ye are

Conly have sampled data)

-usually signals at sampleromes to it to


only compare .....

· what are the parameters ?

Example 1 :
coefficients bo ,
b ..... bm and no , a , .... an-1

bism+ bm-1Sm+ bist bo

=
.. +

+ s"+an- s
+ ais+ do ..
assume m<n
-
->
strictly Proper
nth order

-Example 2 :
Poles P , , p .
A
, P2 , P2 *... and coefficients A ,
A ,*, Az Ac*...
,

= I" .
(C a
Pic K
Ar EK
(constrains poles to have multiplically one)

-choice of parameterization can affect performance of system


ID methods ,

Computing model analytically


·

response

Example first-order model response to unit Input


:

step
-

ECs) = 0 ,
S + 02

* parameters :
f :
(A , , 2)

A Input : UCt = Act)

* output
:

Y6(t)

ucs) =
I -
Yols= o

-0
.[ 5 -

+2)
Yolt) = 6, (1 - - 2t) 1(t)
* get Formula For yAct as a
Function of 1 and t

Computhy model
response numerically
·

-
For Fixed model parameters , and a Fixed sample times to ,
to ,

... tw , enough to compute youti) numerically .

Eg .
do this using numerical differential equation solver

numerator benominator
- -

->
sample time

-> input vector

zero initial
-> Default
conditions

· Least squares Identification


-

given :

*
Input signal u (can extend to case of multiple inputs /
A measured output at sampled times y (t) , .., y (te)

* model with parameter vector t

End parameter vector t so that for given Input(s)


:

-
:

* error between measured output y and output predicted by the

model yo is minimized in
least-squares sense

Hemore e minz
N
I
(YCtr)-Y(tr))
·
L
W
-S

t .

Example First order model


- :

* ECs) =
E, - -
2

S +02

*
Step response :
Predicted by model

20 (t) = &, [1-e- tet] 1(t)

* measured step response :

values y (t0) , y(t) . . .


, y (tw)

constants
* objective Funchon : data
- I

cost(8 , 182) =
[Yctil-t
I
,
(l-efeti)]
variable
Aim
*
:

minimize with respect to A 2

·
solving the optimisation problem

-
Assume we have

* MATLAB Function 'cost' that takes argument o <model parameter)

and computes :

cost (8) = EcyCtr) -


YoC5K))2
k =
0
simulate the model

up to time t k
Initial guess For parameters
*
theta0

-
Ophmize : darvative free optimization method
=>
theraopt = frisearch (cost , the a 0)

* minimum
requirement
:
can evaluate yo (tr) by simulation

* Better opt method


.
available it can
compute dervative of YoLtk)

with respect to t .

* Typically only do Local Optimisation , limited guarantee

You might also like