system identificationkarimpor.profcms.um.ac.ir/imagesm/354/stories/iden/system_iden4.pdf ·...
TRANSCRIPT
-
SYSTEMSYSTEMIDENTIFICATIONIDENTIFICATION
Ali KarimpourAssociate Professor
Ferdowsi University of Mashhad
کراريشناسائی سیستم تالیف دکتر مهدي : مراجعAdvanced PID Control Technologies by Karl J. Astrom and Tore Hagglund
-
lecture 4
Ali Karimpour Jan 2014
2
Lecture 4Classical System Identification Classical System Identification
MethodsMethodsTopics to be covered include:
Introduction.
Classical Identification Methods for Discrete Systems.
Classical Identification Methods for Continuous Systems.
-
lecture 4
Ali Karimpour Jan 2014
3
Classical System Identification MethodsClassical System Identification Methods
Topics to be covered include:
Introduction.
Classical Identification Methods for Discrete Systems.
Classical Identification Methods for Continuous Systems.
-
lecture 4
Ali Karimpour Jan 2014
4
Introduction
. است LTIمنظور از روشهاي کالسیک روشهاي مبتنی بر نظریه سیستمهاي
مبتنی بر پاسخ پله •مبتنی بر پاسخ ضربه •
مبتنی بر پاسخ فرکانسی •مبتنی بر پاسخ ورودي هاي ساده •
:موارد کاربرد این روشها
. باشد LTIسیستم واقعی که در حال شناسائی آن هستیم، در واقع •
.نویز در سیستم تقریبا قابل اغماض باشد •
.چنانچه سیستم با این روشها شناسائی نشد روشهاي پیچیده تر خطی و غیر خطی استفاده می شود
.مناسب جهت طراحی کنترلر •
-
lecture 4
Ali Karimpour Jan 2014
5
Classical System Identification MethodsClassical System Identification Methods
Topics to be covered include:
Introduction.
Classical Identification Methods for Discrete Systems.
Classical Identification Methods for Continuous Systems.
-
lecture 4
Ali Karimpour Jan 2014
6
Classical Identification Methods for Discrete Systems
Classical Identification Methods for discrete systems are:
1. Time domain identification.
2. Frequency domain identification.(It is a parametric procedure such as linear parametric models)
-
lecture 4
Ali Karimpour Jan 2014
7
Classical Identification Methods for Discrete Systems
In causal discrete LTI systems relation between input and output is:
+u(t) y(t)
v(t))()()()(
0tvktukgty
k
g(k): impulse response
In stable causal discrete LTI systems relation between input and output can be approximated by:
)()()()(0
tvktukgtym
k
System Identification: Determination of impulse response (g(k)) of system.
Classical Identification Methods (Time domain identification)
-
lecture 4
Ali Karimpour Jan 2014
8
Classical Identification Methods for Discrete Systems
Classical Identification Methods (Time domain identification)
1. Impulse response test.
2. Step response test.
3. Arbitrary input test.
-
lecture 4
Ali Karimpour Jan 2014
9
Classical Identification Methods for Discrete Systems
I) Impulse response test:
)()()(0
ktukgtym
k
0....)2()1(,1)0( uuu
Classical Identification Methods (Time domain identification)
)0()0( gy )1()1( gy
.........)()( mgmy
II) Step response test:
)()()(0
ktukgtym
k
1....)2()1()0( uuu
)0()0( gy )1()0()1( ggy
.........)(....)1()0()( mgggmy
)1()()( kykykg
-
lecture 4
Ali Karimpour Jan 2014
10
Classical Identification Methods for Discrete Systems
III) Arbitrary applied signal test:
)()()(0
ktukgtym
k
muuuu ,...,,, 210
Classical Identification Methods (Time domain identification)
0)0()0( ugy
.........
01 )(....)1()0()( umgugugmy mm
0/)0()0( uyg
01 )1()0()1( ugugy
01 /))0()1(()1( uugyg .........
-
lecture 4
Ali Karimpour Jan 2014
11
Classical Identification Methods for Discrete Systems
Example 5-1: Let y(t)=12u(0)+8u(-1)+4u(-2)+2u(-3)+e(t) e(t) is a zero mean white noise with σ2=0.25.
Classical Identification Methods (Time domain identification)
I) Impulse response test: y=[11.5283 8.4696 3.8795 1.5283]
II) Step response test: y=[11.5283 20.4696 23.8795 25.3708 ]
III) Arbitrary applied signal test: u= [1;0.5 ;-0.5;-1] , y=[11.5283 14.4696 1.8795 -12.4717]
,5283.1)3(ˆ,8795.3)2(ˆ,4696.8)1(ˆ,5283.11)0(ˆ gggg 8247.0ˆ2 gg
,4914.1)3(ˆ,4098.3)2(ˆ,8.9413)1(ˆ,5283.11)0(ˆ gggg 3098.1ˆ 2 gg
,2503.1)3(ˆ,3491.4)2(ˆ,.99807)1(ˆ,5283.11)0(ˆ gggg 9521.0ˆ 2 gg
-
lecture 4
Ali Karimpour Jan 2014
12
Classical System Identification MethodsClassical System Identification Methods
Topics to be covered include:
Introduction.
Classical Identification Methods for Discrete Systems.
Classical Identification Methods for Continuous Systems.
-
lecture 4
Ali Karimpour Jan 2014
13
Classical Identification Methods for Continuous Systems
Classical Identification Methods for continuous systems are:
1. Time domain identification.
2. Frequency domain identification.
-
lecture 4
Ali Karimpour Jan 2014
14
Classical Identification Methods for Continuous SystemsTime domain identification.
In causal discrete LTI systems relation between input and output is:
+u(t) y(t)
v(t))()()()( 0 tvdtuhtyt
h(t): impulse response
System Identification: Determination of impulse response (h(t)) of system.
Classical Identification Methods (Time domain identification)
-
lecture 4
Ali Karimpour Jan 2014
15
Classical Identification Methods (Time domain identification)
1. Impulse response test.
2. Step response test.
3. Doublet-Pulse Method
• First order systems.
• Second order systems.
• n-order systems.
• Prony analysis
Classical Identification Methods for Continuous SystemsTime domain identification.
-
lecture 4
Ali Karimpour Jan 2014
16
I) Impulse response test:
Classical Identification Methods (Time domain identification)
What is the draw backs of impulse response test?
dtuhty t )()()( 0
dthty t )()()( 0 )(th
II) Step response test:
dtuhty t )()()( 0 dhtyt
0 )()( dtdyth )(
What is the draw backs of step response test?
Classical Identification Methods for Continuous SystemsTime domain identification.
-
lecture 4
Ali Karimpour Jan 2014
17
Dynamic models should give the relation between the input and the output signal during transients.
Step response of the system is used to describe the dynamical behavior.
Step response of process control.
Temperature control, level control, ….
Step response that is uncommon in process control.
Oscillation process, spring constructions, ….
Classical Identification Methods for Continuous SystemsTime domain identification.
-
lecture 4
Ali Karimpour Jan 2014
18
Step response of the integration process.
Temperature control in a well isolated chamber, level control, ….
Step response of a system with long time delay.In transportation systems, ….
Step response of a non-minimum phase system.Water level in boilers, generation of hydro generators, ….
Classical Identification Methods for Continuous SystemsTime domain identification.
-
lecture 4
Ali Karimpour Jan 2014
dsTp eTsK
susysg
1)()()(
19
1- FO systems: First order systems.
2- FOTD systems: First order system with time delay.
3- SOTD systems: Second order system with time delay.
dsTp esTsT
Ksusysg
)1)(1()()()(
21
1)()()(
TsK
susysg p
5- FOTDI: First order system with time delay with integration. sLv e
sTsK
susysg
)1()()()(
4- Integral and time delay processes:sLv e
sK
susysg )()()(
Classical Identification Methods for Continuous SystemsTime domain identification.
-
lecture 4
Ali Karimpour Jan 2014
20
6- Non-interacting tanks or multiple lags
n
p
TsK
susysg
)1()()()(
7- Oscillatory Systems
12)()()()(
2
sTsTK
susysg p
8- System with inverse responses.
14.11
)()()(
2
sssT
susysg
Classical Identification Methods for Continuous SystemsTime domain identification.
-
lecture 4
Ali Karimpour Jan 2014
21
Step response test (The Bump Test)
The bump test is a simple procedure that is commonly used in process control.
FOTD systems can be find by bump test. sLp eTsK
susysg
1)()()(
Classical Identification Methods for Continuous SystemsTime domain identification.
-
lecture 4
Ali Karimpour Jan 2014
22
Integral and time delay process sLv esK
susysg )()()(
LaK v
The bump test is a simple procedure that is commonly used in process control.
Step response test (The Bump Test)
Classical Identification Methods for Continuous SystemsTime domain identification.
-
lecture 4
Ali Karimpour Jan 2014
23
The bump test is a simple procedure that is commonly used in process control.
Oscillatory Systems
21/2 ed21/2 npT
2)ln/2(11
d
p
n T212
Step response test (The Bump Test)
Classical Identification Methods for Continuous SystemsTime domain identification.
22
2
2)()()(
nn
np
ssK
susysg
-
lecture 4
Ali Karimpour Jan 2014
24
Double step response test (The Doublet-Pulse Method)
Tp is chosen a little longer than the time delay of the process.
The pulse amplitude a is chosen so that the response is well above the noise level.
FOTD systems can be find by bump test. sLp eTsK
susysg
1)()()(
pp
p
p
TtLorTtLyyT
T
ayyK
2
)1ln(
minmax
max
min
min
2
max
Classical Identification Methods for Continuous SystemsTime domain identification.
-
lecture 4
Ali Karimpour Jan 2014
25
Prony analysis
Classical Identification Methods for Continuous SystemsTime domain identification.
systemorder High )( sg
n
iii
t
i tfeAty i1
)2cos()(
n
i
jtfjjtfjt
i
iiii
ieeeeeAty
1
22
22)(
n
i
kTfjji iiieeAkTy1
)2(
2)(
....................
-
lecture 4
Ali Karimpour Jan 2014
26
Bode Diagram Method
Classical Identification Methods for Continuous SystemsFrequency domain identification.
1)()()(
sk
susysh
Ak log20
?1
-
lecture 4
Ali Karimpour Jan 2014
27
Bode Diagram Method
Classical Identification Methods for Continuous SystemsFrequency domain identification.
22
2
2)()()(
nn
n
ssk
susysg
220 I
2121log20
rM
221 nr
122
II
2log20rd
-
lecture 4
Ali Karimpour Jan 2014
28
Bode Diagram Method
Classical Identification Methods for Continuous SystemsFrequency domain identification.
Minimum phase and non-minimum phase systems?
Systems with time delay?
-
lecture 4
Ali Karimpour Jan 2014
29
Bode Diagram Method
Classical Identification Methods for Continuous SystemsFrequency domain identification.
Noise reduction in frequency domain.
)())(cos()( tejHtjHA
))(sin()(2
jHjHAys
))(cos()(2
jHjHAyc
222)( cs yyAjH
c
s
yyjH 1tan)(
Integration reduce the noise effect.
-
lecture 4
Ali Karimpour Jan 2014
30
Relay Feedback
frequency cross Phase 180 daK
4180
d-dRelay method is used to derive phase
crossover frequency and gain of systemat that frequency.
Classical Identification Methods for Continuous SystemsFrequency domain identification.
-
lecture 4
Ali Karimpour Jan 2014
31
Relay Feedback
frequency cross Phase 180 daK
4180
Classical Identification Methods for Continuous SystemsFrequency domain identification.