Professional Documents
Culture Documents
Sachin C. Patwardhan
Dept. of Chemical Engineering,
I.I.T. Bombay
Automation Lab
IIT Bombay
Layer 1 P
PID & Operating constraint Controll
MV PV
Load
Pl t
Plant Disturbances
Aggregate Production
Layer 4
Rate Models
Steady
S d SState / D Dynamic
i
Layer 3
First Principles Models
Mathematical Models
Qualitative
¾Qualitative Differential Equation
¾Qualitative signed and directed graphs
¾Expert Systems
Quantitative
¾Differential Algebraic systems
¾Mixed Logical and Dynamical Systems
¾Linear and Nonlinear time series models
¾Statistical correlation based (PCA/PLS)
Mixed
¾Fuzzy Logic based models
TT-1
Heater Coil
L
T
Thyrister
4-20 mA Control
Input Unit
Signal
TT-2 TT-3
E
Experimental
i t l Setup:
S t S
Schematic
h ti Di
Diagram
Example:
p Stirred Tank Heater-Mixer
dT1 F1 Q ( I1 ) ⎫
= (Ti1 − T1 ) + ⎪
dt V1 V1 ρC p ⎪
dh2 ⎪⎪
=
1
[F1 + F2 ( I 2 ) − F ] ⎬ From Physics
dt A2 ⎪
dT2 1 ⎡ UA(T2 − Tatm ) ⎤ ⎪
= ⎢ F1 (T1 − T2 ) + F2 (Ti 2 − T2 ) − ⎥ ⎪
dt h2 A2 ⎢⎣ ρ C p ⎥⎦ ⎪⎭
Q( I1 ) = 7.979 I1 + 0.989 I12 − 0.0073I13 ⎫
3⎬
Correlations from expt. data
F2 ( I 2 ) = 3.9 + 27 I 2 − 0.71I 2 + 0.0093I 2 ⎭
2
0
U = 139.5 J / m 2 Ks ; F (h) = k h2 − h
Local Linearization of
Mechanistic /Grey-box
y
Model
Local Linearization
IIT Bombay
dh1 a1 a3 γ 1k1
=− 2 gh1 + 2 gh3 + v1
dt A1 A1 A1
Tank3 Tank 4
dh2 a a γk
=− 2 2ggh2 + 4 2 gh g 4 + 2 2 v2
dt A2 A2 A2
dh3 a3 (1 − γ 2 )k2
=− 2 gh3 + v2
dt A3 A3
Tank 1
Tank
dh4 a (1 − γ 1 )k1
2 =− 4 2 gh4 + v1
dt A4 A4
Pump 2
Pump1
V2
V1
Manipulated Inputs : v1 and v2
Measured Outputs : h1 and h2
Model Parameters
IIT Bombay
A1, A3 [cm2] 28
A2, A4 [cm2] 32
Model as ODE-IVP
IIT Bombay
⎡ a a3 γ 1k1 ⎤
⎡h1 ⎤ ⎢ − 1 2 g gh1 + 2ggh3 + v1 ⎥ ⎡f1 ( X, U) ⎤
⎢ ⎥ ⎢ A1 A1 A1
⎥ ⎢ ⎥
⎢ ⎥ ⎢ a2 a4 γ 2k2 ⎥ ⎢ ⎥
⎢h2 ⎥ − 2 gh2 + 2 gh4 + v 2 ⎢f2 ( X, U)⎥
d ⎢ ⎥ ⎢ A2 A2 A2 ⎥ ⎢ ⎥
=⎢ ⎥ ≡
dt ⎢ ⎥ ⎢ − a3 2 gh + (1 − γ 2 )k2 v ⎢ ⎥
h
⎢ 3⎥ ⎢
⎥ f ( X, U )
A3 A3 ⎥ ⎢ ⎥
3 2 3
⎢ ⎥ ⎢ a (1 − γ )k ⎥ ⎢ ⎥
⎢h ⎥ ⎢ − 4 2 gh4 + 1 1
v ⎥ ⎢f ( X, U)⎥
⎣44⎦ ⎣4 ⎦
⎣144A A4
1
1
42 3
444424444443
4 ⎦
dX F( X, U)
dt
X = [h1 h2 h3 h4 ] ; U = [v1 v 2 ]
T T
I i i l Condition
Initial C di i : X(0)
Local Linearization
IIT Bombay
Local Linearization
IIT Bombay
Measurement Model
⎡ ∂G ⎤ ⎡ ∂G ⎤
Y (t ) ≅ G (X ) + ⎢
∂X ⎥ [X (t ) − X ] = Y + ⎢ ∂X ⎥ [X(t ) − X]
⎣ ⎦ (X) ⎣ ⎦ (X)
⎡ ∂G ⎤
y (t ) = Y (t ) − Y = ⎢
∂X ⎥ [X(t ) − X ]
⎣ ⎦ (X)
1/16/2014 System Identification 18
Automation Lab
Local Linearization
IIT Bombay
⎢ ∂D ⎥ = ⎢ ∂d1 ∂d 2 ∂d d ⎥ ; ⎢
∂X ⎥ = ⎢ ∂x1 ∂x2 ∂xn ⎥
⎣ ⎦ ⎢ ⎣ ⎦ ⎢
.... .... .... .... ⎥ .... .... .... .... ⎥
⎢ ∂f ∂f n ∂f n ⎥ ⎢ ∂g ∂g r ∂g r ⎥
⎢ n
.... ⎥ ⎢ r
.... ⎥
⎣⎢ ∂4
1
d1 ∂d 2
44424444
∂d d ⎦⎥
3
⎢⎣ ∂x1 ∂x2
1 444 424444
∂xn ⎦⎥
3
(n × d ) ( r × n)
1/16/2014 System Identification 19
Automation Lab
Local Linearization
IIT Bombay
Define matrices
⎡ ∂F ⎤ ⎡ ∂F ⎤ ⎡ ∂F ⎤ ⎡ ∂G ⎤
A = ⎢ ⎥ ; B = ⎢ ⎥ ; H = ⎢ ∂D ⎥ ; C = ⎢ ⎥
⎣ ∂X ⎦ ⎣ ∂U ⎦ ⎣ ⎦ ⎣ ∂X ⎦
computed at ( X ,U ,D )
⎡ 1 A3 ⎤ ⎡ γ 1k1 ⎤
⎢− T 0
A1T3
0 ⎥ ⎢
A1
0 ⎥
⎢ 1 ⎥ ⎢ ⎥
⎢ 0 1 A4 ⎥ ⎢ γ 2 k2 ⎥
− 0 0
dx ⎢ T2 A2T4 ⎥ ⎢ A2 ⎥
=⎢ ⎥x + ⎢ u
dt ⎢ 0 1 (1 − γ 2 )k 2 ⎥
0 − 0 ⎥ ⎢ 0 ⎥
⎢ T3 ⎥ ⎢ A3 ⎥
⎢ 1 ⎥ ⎢ (1 − γ 1 )k1 ⎥
⎢ 0 0 0 − ⎥ ⎢ 0 ⎥
⎣ T4 ⎦ ⎣ A4 ⎦
⎡kc 0 0 0⎤ Ai 2hi
y=⎢ ⎥ x Ti = for i = 1,2,3,4
⎣0 kc 0 0⎦ ai g
P- P+
(T1,T
T2) (62 90)
(62,90) (63 91)
(63,91)
(T3,T4) (23,30) (39,56)
1/16/2014 System Identification 21
Automation Lab
IIT Bombay
Quadruple
Q p Tank System
y
P-
Continuous Time State Space h1, h2 [cm] (12.4,12.7)
Model Matrices h3, h4 [cm] (1 8 1 4)
(1.8,1.4)
v1,,v2 [V] (3,3)
⎡- 0.01595 0 0.04186 0 ⎤ k1,k2, [cm3/V] (3.33,3.35)
⎢ 0 - 0
0.01107
01107 0 0
0.03334
03334 ⎥
A=⎢ ⎥ γ1,γ2 (0.7,0.6)
⎢ 0 0 - 0.04186 0 ⎥
⎢ ⎥
⎣ 0 0 0 - 0.03334 ⎦ Developed
p at ( X, U )
⎡0.08325 0 ⎤ Steady State Operating
⎢ 0 ⎥
0.06281 ⎡0.5 0 0 0⎤ Point
B=⎢ ⎥ C=⎢ ⎥
⎢ 0 0.04786⎥ ⎣ 0 0.5 0 0 ⎦
⎢ ⎥
⎣0.03122 0 ⎦
Rearranging
y ( s ) = G p ( s )u ( s ) + G d ( s )d( s )
G u ( s ) = C[sI − A ] B and G d ( s ) = C[sI − A ] H
−1 −1
⎡ γ 1c1 (1 − γ 2 )c1 ⎤
⎢ 1 + T1s (1 + T3 s )(1 + T1s ) ⎥
G(s) = ⎢ ⎥
⎢ (1 − γ )
1 2c γ c
2 2 ⎥
⎢⎣ (1 + T4 s )(1 + T2 s ) 1 + T2 s ⎥⎦
where c1 = T1k1kc / A1 and c2 = T2 k 2 kc / A2
Minimum Phase Case
⎡ 2.6 1.5 ⎤
⎢ 1 + 62 s (1 + 23s )(1 + 62 s ) ⎥
G− ( s ) = ⎢ ⎥
⎢ 1.4 2 .8 ⎥
⎢⎣ (1 + 30 s )(1 + 90 s ) 1 + 90 s ⎥⎦
1/16/2014 System Identification 24
Automation Lab
IIT Bombay
Manipulated
Inputs From Measured Outputs
p
Control Computer To Control
C
Computer
t
Control Computer
/DCS
3
3
2.8
Measurred Output
2.8
ed Output
2.6 2.6
ADC
D
Measure
2.4 2.4
2.2 2.2
2 2
1.8 1.8
0 5 10 15 20 0 5 10 15 20
Sampling Instant Sampling Instant
2.8 2.8
2.7 2.7
Manipulated Input Sequence
2.6 2.6
D
DAC
put
Manipulated Inp
2.5 2.5
2.4 2.4
2.3 2.3
2.2
2.2
2.1
2.1
2
2 0 2 4 6 8 10 12 14 16 18 20
0 2 4 6 8 10 12 14 16 18 20
Sampling Instant
Sampling Instant
dx
Solve = Ax(t ) + Bu(t )
dt
with
i h initial di i x(t k ) = x(kT ) ≡ x(k )
i i i l condition
and u (t ) = u (k ) for kT ≤ t < (k + 1)T
I an (ideal)
In (id l) digital
di it l control
t l system,
t d t arrives
data i or is
i sentt outt
at discrete time instants (0, T, 2T, 3T,...., kT,....)
Thus we need model that relates
Thus,
x(kT) and / or y (kT) with u(kT) for k = 0,1....
This is achieved by converting differential eqn.
eqn in
continuous time to difference eqn. in discrete time
Notation
x(k) ≡ x(t = kT), y (k) ≡ y (t = kT) and u(k) ≡ u(t = kT)
1/16/2014 System Identification 29
Automation Lab
Integrating Factor
IIT Bombay
dx
Note : To integrate scalar ODE = ax(t ) + bu (t )
dt
with
ith initial diti x(kT ) ≡ x(k ) for
i iti l condition f kT ≤ t < (k + 1)T
we need integrating factor e -at where
a 2 t 2 a 3t 3 a nt n
e = 1 + at +
at
+ + .... + + ....
2! 3! n!
Defining exponential exp( At ) as follows
t 2
t 3
e At = I + At + A 2 + A 3 + ....
2! 3!
d At 2t 2 3t 2 3
e = A+ A + A + ....
dt 2! 3!
⎡ t2 2 t3 3 ⎤
= ⎢I + At + A + A + ....⎥ A = e At A
⎣ 2! 3! ⎦
1/16/2014 System Identification 30
Automation Lab
dx
Using e − At
as integrating factor e − At
= e − At [Ax(t ) + Bu(t )]
dt
e − At ⎡ dx
⎢⎣ dt − Ax (t )
⎤ d − At
=
⎥⎦ dt e x [
(t ) = e − At
]
Bu(t )
( k +1)T t k +T
⎡ ( k +1)T
⎤
[− At
]
∫kT d e x(t ) = t∫ e Bu(τ )dτ = ⎢⎢ kT∫ e Bdτ ⎥⎥u(k )
− Aτ − Aτ
k ⎣ ⎦
⎡ ( k +1)T
⎤
e − A ( k +1)T
x(k + 1) = e − A ( kT )
x(k ) + ⎢ ∫ e Bdτ ⎥u(k )
− Aτ
⎢⎣ kT ⎥⎦
⎡( k +1)T A [( k +1)T −τ ] ⎤
x(k + 1) = e AT [ ] x( k ) + ⎢ ∫ e Bdτ ⎥u(k )
⎢⎣ kT ⎥⎦
1/16/2014 System Identification 31
Automation Lab
⎣0 ⎦
C
Computer controll relevant
l discrete
di models
d l
x(k + 1) = Φx(k ) + Γu(k )
y (k ) = Cx(k )
T
Φ = exp( AT ) ; Γ = ∫ exp( Aτ )B dτ
0
t2 2 t3 3
e At = I + At + A + A + ....
2! 3!
−1
= ΨΨ + ΨΛΨ [ −1
] t2
[
t + ΨΛ2 Ψ −1 + ....
2!
]
⎡ t2 2 t3 3 ⎤ −1
= Ψ ⎢I + Λt + Λ + Λ + .....⎥ Ψ
⎣ 2! 3! ⎦
1/16/2014 System Identification 34
Automation Lab
IIT Bombay
Computation of Model Matrices
Λt ⎡ t2 2 t3 3 ⎤
e = ⎢I + Λt + Λ + Λ + .....⎥
⎣ 2! 3! ⎦
⎡ 2 t
2
⎤
⎢1 + λ1t + λ1 + ... 0 ... 0 ⎥
⎢ 2! ⎥
⎢ t 2
⎥
=⎢ 0 1 + λ2t + λ2 + ...
2
... ...
2! ⎥
⎢ ... ... ... 0 ⎥
⎢ 2 t
2 ⎥
⎢ 0 ... ... 1 + λnt + λn + ...⎥
⎣ 2! ⎦
⎡e λ1t 0 ... 0 ⎤
⎢ ⎥
0 e λ2 t ... 0 ⎥
=⎢
⎢ ... ... ... ... ⎥
⎢ λn t ⎥
⎢⎣ 0 ... ... e ⎥⎦
1/16/2014 System Identification 35
Automation Lab
IIT Bombay
Computation of Model Matrices
⎡ ⎤ −1
e At
t2 2 t3 3
= Ψ ⎢I + Λt + Λ + Λ + .....⎥ Ψ = Ψ e Λt Ψ −1 [ ]
⎣ 2! 3! ⎦
T
⎡T ⎤ −1
Γ = ∫ exp( Aτ )B dτ = Ψ ⎢ ∫ exp(Λτ ) dτ ⎥ Ψ B
0 ⎣0 ⎦
⎡T λ1τ ⎤
⎢ ∫ e dτ 0 ... 0 ⎥
⎢0 ⎥
⎤ ⎢ 0 ⎥
T
⎡T
∫ dτ
λ2τ
e
⎢ ∫ exp((Λτ ) dτ ⎥ = ⎢
... 0 ⎥
⎣0 ⎦ ⎢ ... 0 ⎥
⎢ ... ...
T
... ⎥
⎢ ⎥
⎢ 0 ... ... ∫ e λnτ dτ ⎥
⎣ 0 ⎦
Alternate Method
IIT Bombay
dx
= Ax + Bu
dt
Initial Condition : x(kT ) = x(k ) and u(t ) = u(k ) over kT ≤ t < (k + 1)T
Taking Laplace Transform
sx( s ) − x( k ) = Ax
A ( s ) + Bu
B ( s)
1 − e −Ts
[sI − A ]x( s) = x(k ) + Bu (k )
s
⎡ 1 − e −Ts ⎤
x( s ) = [sI − A ] ⎢x(k ) +
−1
Bu(k )⎥
⎣ s ⎦
⎧ −1 ⎡ 1 − e −Ts ⎤⎫
x(k + 1) = L ⎨[sI − A ] ⎢x(k ) +
−1
B (k )⎥ ⎬
Bu
⎩ ⎣ s ⎦⎭
= Φ = L [sI − A ]
T
AT −1 −1
e Γ = ∫ e Aτ Bdτ
t =T
0
1/16/2014 System Identification 37
Automation Lab
Alternate Method
IIT Bombay
⎧ 1 ⎡s 0 ⎤⎫ T ⎡ e −τ 0⎤ ⎡1⎤
−1
Φ=L ⎨ ⎢ ⎥ ⎬ Γ=∫ ⎢ ⎥ ⎢ ⎥ dτ
⎩ s ( s + 1) ⎣1 s + 1⎦ ⎭
−τ
0
⎣1 − e 1 ⎦ ⎣0 ⎦
⎡ e −T 0⎤ ⎡ 1 − e −T ⎤
=⎢ ⎥ =⎢
−T −T ⎥
⎣1 − e 1⎦ ⎣ − +
T 1 e ⎦
1/16/2014 System Identification 38
Automation Lab
IIT Bombay
Computation of Model Matrices
⎡T ⎤
p( Aτ )Bdτ ⎥ = [Φ − I ]A −1B
When matrix A is invertible ⎢ ∫ exp(
⎣0 ⎦
Quadruple Tank System
Discrete Time State Space
p Model Matrices
Sampling Time T = 5 sec
⎡0.9233 0 0.1813 0 ⎤
⎢ 0 0.9462
0 9462 0 0
0.1493
1493⎥
Φ=⎢ ⎥
⎢ 0 0 0.8112 0 ⎥
⎢ ⎥
⎣ 0 0 0 0.8465⎦
⎡ 0.4001 0.02276⎤
⎢0.01209 0.3055 ⎥
⎢ ⎥ ⎡0.5 0 0 0⎤
Γ= C=⎢
⎢ 0 0 2159 ⎥
0.2159 ⎣ 0 0.50 5 0 0⎥⎦
⎢ ⎥
⎣ 0.1438 0 ⎦
1/16/2014 System Identification 39
Automation Lab
y (k ) = G p (q )u(k )
G p (q ) = C[qI − Φ] Γ
−1
[1 - 1.498 q -1
] [ ]
+ 0.551q -2 h1 (k ) = 0.04329 q -1 + 0.03549q -2 v2 (k )
1 498 h1 (k − 1) + 0.551
h1 (k ) - 1.498 0 551h1 (k − 2) = 0.04329
0 04329 v2 (k − 1) + 0.03549
0 03549 v2 (k − 2)
z Transforms
IIT Bombay
Note : z = eTs
T
and T ≡ sampling interval, s ≡ Laplace operator
1/16/2014 System Identification 43
Automation Lab
z Transforms : Properties
IIT Bombay
Linearity
Ζ[αf(k) + β g (k )] = αΖ[f(k)] + β Ζ[g(k)]
Time Shift
Ζ{ f (k - n))} = z − n f ( z )
⎡ n −1 ⎤
Ζ{ f (k + n)} = z ⎢ f ( z ) − ∑ f ( j ) z ⎥
n −j
⎣ j =0 ⎦
Final Value Theorem
Initial Value Theorem
If (1 - z -1 ) f ( z ) does
d nott have
h any poles l
lim
f (0) = f ( z) on or outside the unit circle, then
z→∞ lim lim
f (k ) = (1 - z -1 ) f ( z )
k→∞ z →1
1/16/2014 System Identification 44
Automation Lab
z Transforms : Examples
IIT Bombay
Example 1
y(kT) = α with k ≥ 0 (step function)
Ζ{y(kT)} = α + αz −1 + αz − 2 + ......
α
=
1 − z −1
Example 2
y(kT) = kT with k ≥ 0 (ramp)
Ζ{y(
y(kT))} = 0 + Tz −1 + 2Tz − 2 + ......
= T ( z −1 + z − 2 + ....)
Tz
=
( z − 1) 2
we have
{ }
y ( z ) = C[zI - Φ] Γ u( z ) = G ( z )u( z )
−1
⎡ 0.2 ⎤ ⎡ v1 ( z ) ⎤
⎡ h1 ( z ) ⎤ 0 01034 ⎥ ⎢
0 01138 z + 0.01034
0.01138 ⎥
⎢ ⎥ ⎢
⎢
z - 0.9233 2
⎥ ⎢ ⎥
⎢ ⎥= z - 1.734 z + 0.749
⎢ ⎥
⎢ ⎥ ⎢ ⎥
⎢ 0
0.006045
006045 z + 0
0.005614
005614 0
0.1528
1528 ⎥ ⎢v ( z ) ⎥
⎢h ( z )⎥ ⎢ 2
⎥
⎣1223⎦ ⎢⎣ z 2 - 1.793 z + 0.8009 ⎥
14444444442444z4 - 0.9462
44444 3⎦ 1⎢⎣
2 3 ⎦⎥
y( z) G( z) u(z)
()
d = dead time
[1 + a1 z -1
+ a 2 z ]
-2
y ( z ) =[b1 z -d-1
+ b2 z ]
-d- 2
u( z )
Practical Difficulty
A reliable mechanistic model may not be available
for the system under consideration.
Instead
I d of
f starting
i from
f well
ll d
developed
l d nonlinear
li
mechanistic model, can we estimate parameters of
linear p
perturbation directly
y from operating
p g data?
Remedy
1.. Inject known input
nput perturbat
perturbations
ons in
n the system
2. Record measured outputs generated in response
to the perturbations
3 Estimate dynamic model parameters using optimization
3.
Tank 3 Tank 4
Tank 1 Tank 2
Control Control
V l 1
Valve Valve 2
on 4 Tank Setup
Input 1 Input 2
Output 2
Output 1
1
ut 1 (mA)
0
Inpu
-1
1
Input 2 (mA)
-1
0
L
-5
0 200 400 600 800 1000 1200
4
el 2 (mA)
2
0
-2
2
Leve
-4
-6
0 200 400 600 800 1000 1200
Time (sec)
0
y1
-5
0 500 1000
1
0.5
u1
0
-0.5
0 500 1000
Samples
Id tifi ti D
Identification Data
t V lid i d
Validation data
1/16/2014 System Identification 56
Automation Lab
IIT Bombay
Quadruple Tank: Grey Box Linear Model
⎡α1 0 α 2 0 ⎤ ⎡ β1 β2 ⎤
⎢0 α α ⎥ ⎢β β 4 ⎥⎥
0 ⎡γ 1 0 0 0⎤
Φ=⎢ 3 4 ⎥ Γ=⎢ 3 C=⎢
⎢ 0 0 α5 0 ⎥ ⎢0 β5 ⎥ ⎣0 γ2 0 0⎥⎦
⎢ ⎥ ⎢ ⎥
⎣ 0 0 0 α 6⎦ ⎣β 6 0⎦
( ) Min N
Θ
ˆ , xˆ (0) = ∑ [ε ( k ) ]T
Wε(k )
Θ, xˆ (0) k =0
Subject to
xˆ (k + 1) = Φ (Θ)xˆ (k) + Γ(Θ)u(k)
yˆ (k) = C(Θ)xˆ (k)
h k = 0,1,2,........N - 1
where
Alternate Approach
IIT Bombay
⎡ y1(k)⎤ ⎡ b1 b2 q + b3 ⎤⎡u1(k)⎤
⎢ ⎥ ⎢ q+a q 2
+ a q + a ⎥⎢ ⎥
⎢ ⎥ ⎢ 1 2 3
⎥⎢ ⎥
⎢ ⎥=⎢ ⎥⎢ ⎥
⎢ y (k)⎥ ⎢ b4 q + b5 b ⎥⎢u (k)⎥
⎢ 2 ⎥ ⎢ 2 ⎥⎢ 2 ⎥
6
⎣424
1 ⎣q4+4
3⎦ 1
a 4 q + a5
444 42444444
q + a 6 ⎦⎣
31 424 3⎦
y(k) G(q) u(k)
Define parameter vectors
Θ1 = [a1 b3 ]
T
a2 a3 b1 b2
Θ 2 = [a4 b6 ]
T
a5 a6 b4 b5
1/16/2014 System Identification 60
Automation Lab
Estimate Θ1 (
ˆ , yˆ (0), yˆ (1), yˆ (2) by
1 1 1 y solving
gf)
following
g
optimization problem
Min
( )
N
Θ1 , yˆ1 (0),
ˆ ) yˆ1 (1),
) yˆ1 (2) = [
∑ 1
ε (k ) ]2
Summary
IIT Bombay
y (k ) = Cx(k )
n( z ) (C adj[zI - Φ]Γ )
x ( 0) = 0 =
d ( z) det[zI - Φ]
Stability:
y Discrete Time Systems
y
Consider scenario x(0) is non - zero and u(k) = 0
x(1) = Φx(0)
x(2) = Φx(1) = Φ 2 x(0)
..............
x(k ) = Φ k x(0)
Consider
C id special
i l case
where Φ is diagonalizable ⎡(λ1 )k 0 ... 0 ⎤
⎢ ⎥
Φ = ΨΛΨ −1
k
Λ = ⎢ 0 (λ2 )k ... 0 ⎥
⎢ ... ... ⎥
⇒ Φ 2 = ΨΛ2 Ψ −1 ⎢
... ...
k⎥
⇒ Φ k = ΨΛk Ψ −1 ⎢⎣ 0 ... ... (λn ) ⎥⎦
A = ΨΛΨ -1 ⇒ Φ = e AT = Ψ e ΛT Ψ -1 [ ]
Thi implies
This i li that
h
Matrix A and Matrix Φ have identical eigenvectors
Also, if λ is eigenvalue of A then e λT is eigenvalue of Φ
If α > 0 then
th eαT > 1
⇒ All poles in left half of s - plane
map outsidesid
id ide the
h unit
i circle
i l ini z - plane
l
1/16/2014 System Identification 73
Automation Lab