Modelado 1 UK

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 72

Process Modelling

Prof. Cesar de Prada


Dpt. Systems Engineering and Automatic Control
University of Valladolid, Spain
prada@autom.uva.es
http://www.isa.cie.uva.es/~prada/

Prof. Cesar de Prada, ISA, UVA 1


Outline
• Motivation
• Models
• Modelling methodology
• First principles models. Examples
• Linearized models
• States space models
• Impulse response models
Prof. Cesar de Prada, ISA, UVA 2
Motivation
• Models are required to analyse the
behaviour of a system and taking decisions
about its functioning, not by experimenting
all possible options, but in a systematic and
more rational way
Model

Prof. Cesar de Prada, ISA, UVA 3


Models
• They are approximate representations of the real
world
• Several types of models: Physical, qualitative,
mathematical…
• Different applications: design, training, what
happens if…?, decision making,…
• Abstraction: Only those aspects relevant to the
problem considered are included in the model.
• How can be generated, solved, used, validated…?

Prof. Cesar de Prada, ISA, UVA 4


What is a mathematical model?
• Set of equations that describe in a sound way
the behaviour of the system we are modelling.
• They are always approximations, not the real
system
• Different models for different types of
processes and aims
• Compromise between easy of use and fidelity
in the representation of the reality
Prof. Cesar de Prada, ISA, UVA 5
Adequate representation…

Proceso
u time

ym

time Model

time
Prof. Cesar de Prada, ISA, UVA 6
Continuous and discrete event
processes
q

Continuous processes: Discrete event systems:


Its variables evolve Its variables only change at
continuously over time certain time instants and can
and can have any value have only an integer number
within a given range of values
Prof. Cesar de Prada, ISA, UVA 7
Continuous / Discrete events
• Continuous processes
– They are described mainly by DAEs or PDEs.
– Main interest: the trajectory of some variables
• Discrete events processes
– They are describe mainly as sequences of
activities.
– Main interest: the statistical behaviour of some
variables

Prof. Cesar de Prada, ISA, UVA 8


Static and dynamic models
q = kρ h
q
Static model: It relates the
values of the process
h variables at an equilibrium
state

dh Dynamic model: It relates


Aρ = q − kρ h the value of the process
dt
variables over time

Prof. Cesar de Prada, ISA, UVA 9


Dynamic response

q
time

Prof. Cesar de Prada, ISA, UVA 10


Static and dynamic models
• Static models
– They describe processes at equilibrium
– They are described by algebraic equations (AE)
– Typical application: Process design
• Dynamic models (in continuous time)
– They describe the time evolution of the process
– They are described by DAE and PDE
– Wide range of applications: control, training,...
Prof. Cesar de Prada, ISA, UVA 11
Other kind of dynamic systems

u(kT)
y(t)
Computer D/A Process

y(kT)
A/D

 discrete time models


 They relate the process variables at the sampling
instants kT, k = 0, 1, 2, ,….
 Equations in differences: y((k+1)T) = f(y(kT), u(kT))
Prof. Cesar de Prada, ISA, UVA 12
How to obtain a process model?

Using reasoning Using experimentation and


and laws of nature data from the process
Prof. Cesar de Prada, ISA, UVA 13
First principles models
• They are obtained from reasoning and the
use of laws of nature: conservation of mass,
energy or momentum as well as others
from the application domain.
• They have a wide range of validity
• They require a deep knowledge of the
process and the applicable laws from
chemistry, physics, biology, etc.

Prof. Cesar de Prada, ISA, UVA 14


Experimental models: Identification
The model is obtained from the A model structure is
postulated and its
information provided by a set of
coefficients are
experiments. computed using the
data
U
Y
U Y
t Process
t

Model

Prof. Cesar de Prada, ISA, UVA 15


First principles models
Modelling methodology:

 Set the aims and range of application of the


model
Fix a set of hypothesis about its behaviour
Formulate the mathematical equations using the
appropriate laws of nature according to the
hypothesis made
Estimate the value of its parameters
Validate the model

Prof. Cesar de Prada, ISA, UVA 16


Many kinds of models
• Lumped parameters
• Distributed parameters
• Non-linear
• Linear
• In the time domain
• In the frequency domain
• ….
Prof. Cesar de Prada, ISA, UVA 17
Mass conservation law
Mass accumulation in the system per unit time =
Incoming mass per unit time -
Outgoing mass per unit time -
Mass generated inside the system per unit time -
Mass consumed inside the system per unit time -

dm Fi F0
= Fi − F0 + G − C
dt G m C

Prof. Cesar de Prada, ISA, UVA 18


Example: Gravity drained tank
q Mass conservation
p0 Accumulation per unit time =
Input flow q - outflow F
h
p1 dm
= qρ − Fρ
dt
F
m = Aρh F = Sv = Sk 1 p1 − p 0
m mass in the tank p1 = p 0 + ρgh F=k h
A tank cross section dh
ρ density, k constant Aρ = qρ − ρk h
dt

Prof. Cesar de Prada, ISA, UVA 19


Example: Gravity drained tank
q
Mass conservation

Accumulation per unit time =


h Input flow q - outflow F
dm
= qρ − Fρ
dt
u
F m = Aρh F = uk h
dh
A = q − uk h V = Ah
m mass in the tank dt
A tank cross section Algebraic
ρ density, k constant Non-linear differencial
equation
u valve opening equation
Prof. Cesar de Prada, ISA, UVA 20
States space models
d x( t )
= f ( x ( t ), u ( t ), v( t ), t )
dt
y( t ) = g( x , u ( t ), v( t ), t )

v Disturbances
u y Observed
x
Manipulated responses
variables
x states
Prof. Cesar de Prada, ISA, UVA 21
Model simulation
Integrating (numerically)
q
the model, one can obtain
the time evolution of the
liquid volume as a function
h of q

dh 1 uk
F = q− h V = Ah
dt A A
Euler integration 1 u ( t )k 
formula h ( t + ∆t ) = h ( t ) +  q ( t ) − h ( t )  ∆t
A A 

Prof. Cesar de Prada, ISA, UVA 22


Causality
dh Computational
q A = q−F
dt causality: the order in
F=k h which the variables
are computed
h q h F

F
Physical causality: The intended use of the
causes and effects model (what happens if…,
q h F control,.., may impose the
computational causality
Prof. Cesar de Prada, ISA, UVA 23
Hypothesis
ci
q q

c h h

F F
Perfect mix No mix
d Vc h Ah V
= qc i − Fc c( t ) = c i ( t − ) = c i ( t − ) = ci (t − )
dt v Av F

Prof. Cesar de Prada, ISA, UVA 24


Formulation
ci
Concentration q d (Vc) 
= qc i − Fc
ci dt  Vc
 c =
dV  V
= q−F
dt 
Volume V h
c dc dV
V +c = qc i − Fc
dt dt
F dc
V + c(q − F) = qc i − Fc
dt
Perfect mix dc
V = q (c i − c)
dt
ρ constant

Prof. Cesar de Prada, ISA, UVA 25


Numerical errors
q1 q2

h1 h2

F1
F2
dh dh Laws +
A1 1 = q 1 − F1 A 2 2 = q 2 + F1 − F2
dt dt constraints
F1 = k 1 h 1 − h 2 h 1 < h 2 ? F2 = k 2 h 2
F1 = k 1 sgn( h 1 − h 2 ) h 1 − h 2 0 ≤ h i ≤ h max qi ≥ 0

Prof. Cesar de Prada, ISA, UVA 26


Isothermal chemical reactor

Reaction: A B
FT

AT Row material
A

Reactor

Products
A, B

Prof. Cesar de Prada, ISA, UVA 27


Mathematical model
Hypothesis:
Product A
• perfect mix in the reactor CAi , Ti
F
•Temperature T constant
•Volume V constant
d cA − E RT A⇒B
V = Fc Ai − Fc A − Vke cA
dt CA CB T
d cB − E RT
V = − Fc B + Vke cA
dt
Mass balances of products A and B

Prof. Cesar de Prada, ISA, UVA 28


Pressure in a container
pf
a F
dm
= Fi − F = Fi − aC v p − pf
2 2
Fi
dt p
ρ
m = Vρ p = RT isothemal tank
M
VM d p
= Fi − aC v p2 − pf2 Low pressure
RT d t container
Constant
Prof. Cesar de Prada, ISA, UVA
composition 29
Energy conservation law
Ti d(mH) V2
q = qρH i − qρH +
dt R
si H = c e T m = Ahρ
dT V2
Ah = q (Ti − T ) +
V R T dt ρc e R

Non-linear differential equation


T temperature, V voltage
Hypothesis:
m mass of liquid in the tank
H enthalpy, ce specific heat T uniform distribution
A tank cross section Perfect thermal isolation
ρ density, R resistance Constant density
Prof. Cesar de Prada, ISA, UVA 30
Conservation of momentum
d (mv) d (Iω)
= ∑ Fi = ∑ Ti
dt dt
d2x d 2θ
m 2 = ∑ Fi I 2 = ∑ Ti
dt dt

F Reference
m system
x

Prof. Cesar de Prada, ISA, UVA 31


Suspended mass

d2x
m 2 = ∑ Fi
dt
d2x
m 2 = −kx + mg + F
-kx dt

0 m dx
=v
dt
x mg dv k 1
= − x+g+ F
F dt m m

Prof. Cesar de Prada, ISA, UVA 32


Coupled Vehicles
K spring coef. r
2
d y dy dx f viscous
M 2
= F − K ( y − x − L ) − f ( − ) friction coef.
dt dt dt
d2x dy dx No sliding
m 2 = K ( y − x − L) + f ( − ) friction
dt dt dt
No air
resistance
f F
m M
k
x y

Prof. Cesar de Prada, ISA, UVA 33


Single dimension crane
x F Location of mass M: x
M
Location of mass m: x + L sen θ
Two degrees of freedom: x, θ
θ L
Viscous friction
m No sliding friction
mg

Model the position of mass m and cart M with


respect to the reference system
Prof. Cesar de Prada, ISA, UVA 34
Single dimension crane
Conservation of momentum of
M x F the cart and mass m in the
horizontal direction x

θ L
d2x d 2 ( x + Lsenθ) dx
M 2 +m = F−f
dt dt 2 dt
m
d 2θ  d θ
2
d2x dx
(M + m) 2 + mL cos θ 2 − mLsenθ  = F − f
mg dt dt  dt  dt

Another equation is needed for θ

Prof. Cesar de Prada, ISA, UVA 35


Single dimension crane
Conservation of angular
M x F momentum with respect to a
set of axis associated to the cart
Movements described in this set
θ L of axis present an acceleration -
d2x d2x/dt2 in the horizontal direction
m 2 θ
m
dt x with respect to a set of fix axis
θ
d 2θ d2x
mg (I + mL ) 2 = − mgsenθL − m 2 cos θL
2

dt dt
d 2θ  dθ
2
d2x dx
(M + m) 2 + mL cos θ 2 − mLsenθ  = F − f
dt dt  dt  dt
Prof. Cesar de Prada, ISA, UVA 36
Liquid flow in a pipe
d mv
Momentum = A∆p 0 − A∆p v − AfLρv 2 − Ahρg
conservation dt
law 1
∆p v = 2 2 ρq 2 m = ALρ q = Av
a Cv
Non-linear L d q ∆p 0 1 fL
differential = − ( 2 2 + 2 )q 2 − gh
A dt ρ a Cv A
equation

a q h

∆pv
∆p0
Prof. Cesar de Prada, ISA, UVA 37
Control valve (linear)

Spring d2x dx
m 2= (p − p 0 )A + ∆p vS − kx − k v
x dt dt
Diafragm
x ∆p v
Aire q = Cv
L ρ
Aire
p x position of stem from
abre
Friction
equilibrium position
L range of stem
Líquid displacement
p air pressure
Prof. Cesar de Prada, ISA, UVA 38
RC circuit
R I2
Infinite
I1 impedance
V E
C I2 =0
I1-I2

1 1 d V d I1 1
V = I1 R + ∫ (I1 − I 2 )dt V = I1 R + ∫ I1dt = R + I1
C C dt dt C
1 1 dE 1
E = ∫ (I1 − I 2 )dt E = ∫ I1dt = I1
C C dt C

d I1 1 1 dV
=− I1 +
dt RC R dt
dE 1
= I1
dt C
Prof. Cesar de Prada, ISA, UVA 39
Circuits
R1 L I2

I1
V C E
R2
I1-I2

d I1 1
dt C ∫
V = I1 R 1 + L + (I1 − I 2 )dt + (I1 − I 2 )R 2

1
E=
C ∫ (I1 − I 2 )dt + (I1 − I 2 )R 2

Prof. Cesar de Prada, ISA, UVA 40


DC Motor
R L T external torque
V I k2ω e.m.f.

ω dω
J = k 1I − fω − T
dt
Independent T dθ

excitation dt
dI
V = RI + L + k 2 ω
dt

Prof. Cesar de Prada, ISA, UVA 41


Distributed parameters processes

Ti F
∆x
Prof. Cesar de Prada, ISA, UVA 42
Distributed parameters process
Ts

T(x,t) Ti-1 Ti Ti+1 F

x ∆x

The pipe is divided into small volumes of width ∆x


where T can be assumed to be constant
Energy balance on every volume
Limit when ∆x → 0

Prof. Cesar de Prada, ISA, UVA 43


Distributed parameters process
Ts
r
T(x,t) Ti-1 Ti Ti+1 F

∆x
Energy balance d πr 2 ∆xρc e Ti
= Fρc e Ti −1 − Fρc e Ti + 2πr∆xU(Ts − Ti )
dt
Partial d Ti F (Ti −1 − Ti ) 2 U(Ts − Ti )
= 2 +
differential dt πr ∆x rρc e
equations d Ti F (T − Ti ) 2 U(Ts − Ti )
lim = 2 lim i −1 + lim
∆x → 0 d t πr ∆x →0 ∆x ∆x → 0 rρc e
∂ T( x, t ) F ∂T ( x , t ) 2 U(Ts − T ( x , t ))
=− 2 +
∂t πr ∂x rρc e
Prof. Cesar de Prada, ISA, UVA 44
First principles models
• Described by sets of algebraic and differential
equations, often non-linear.
• They can be used for a wide range of aims
• They have a wide range of validity within the one
fix by the associated hypothesis
• They require a certain modelling background
• They may be difficult to operate with
• Quite often they are solved by simulation
Prof. Cesar de Prada, ISA, UVA 45
Simulation: EcosimPro
• Facilitate the description of the model, and provides the
methods to solve it
• Modelling / Simulation languages
• What happens if…?
• Based on object oriented software technologies
• Advanced numerical methods and functionalities
• ESA: European Space Agency
• C++ code generator with a development and execution
environment
• Library / Component / Partition / Experiment
• Open system

Prof. Cesar de Prada, ISA, UVA 46


EcosimPro

Prof. Cesar de Prada, ISA, UVA 47


Graphical environment

Prof. Cesar de Prada, ISA, UVA 48


Simulation

Prof. Cesar de Prada, ISA, UVA 49


Model Linearization
• To facilitate the analysis and computations, we use
a linear approximation of the non-linear equations
• Easier to operate, but with a limited validity range

dh d∆h
A =q−k h A = β∆q − α∆h
dt dt

Prof. Cesar de Prada, ISA, UVA 50


Linearization of a function f
Taylor series expansion in an operating point u0, y0,
z0, ….
f ( u , y, z ) = 0 f (u 0 , y 0 , z 0 ) = 0
∂f ∂f ∂f
f ( u , y, z ) = f ( u 0 , y 0 , z 0 ) + (u − u 0 ) + (y − y0 ) + (z − z 0 ) + ...
∂u 0 ∂y 0 ∂z 0

∂f ∂f ∂f
∆u + ∆y + ∆z = 0 ∆u = u − u 0 ∆y = y − y 0 ∆z = z − z 0
∂u 0 ∂y 0 ∂z 0

Linear equation in the new variables ∆u, ∆y, ∆z


Prof. Cesar de Prada, ISA, UVA 51
Linearized model of the tank
dh
q A −q+k h = 0
dt
f (h , h , q ) = 0 h 0 , h 0 , q 0
∂f   ∂f ∂f
(h − h 0 ) + (h − h 0 ) + (q − q 0 ) = 0
h ∂h 0 ∂h 0 ∂q 0
∂f ∂f k ∂f
=A = = −1
∂h 0
 ∂h 0 2 h 0 ∂q 0
F
d∆h k
Deviation Variables A + ∆h − ∆q = 0
dt 2 h 0
∆h = h - h0
∆q = q - q0 Linear differential equation

Prof. Cesar de Prada, ISA, UVA 52


Simulation / Validation
5.0 8
4.8
7
4.6
6
4.4 h h
h_l h_l
4.2 5

4.0 4
0 10 20 30 40 0 10 20 30 40
TIME TIME

Responses of the non-linear and linearized models for 2 step changes in q


5.5 7.0
5.4
6.5
5.3
6.0
5.2
q q
5.1 5.5

5.0 5.0
0 10 20 30 40 0 10 20 30 40
TIME TIME

Prof. Cesar de Prada, ISA, UVA 53


Linearized model of the tank
q d∆h k
A + ∆h − ∆q = 0
dt 2 h 0
A 2 h 0 d∆h 2 h0
+ ∆h = ∆q
h k dt k
d∆h
τ + ∆h = K∆q
F dt
A2 h 0 2 h0
Deviation Variables τ= K=
k k
∆h = h - h0
The value of the model coefficients depends
∆q = q - q0
on the linearization point
Prof. Cesar de Prada, ISA, UVA 54
Linearized models
Variables u and y represent u ( t ) = U( t ) − U 0 ( t )
changes above and below
y( t ) = Y( t ) − Y0 ( t )
the operation point U0 , Y0

U
U0 Y
U Y Y0
t Process
t
The range of validity is constrained to a certain extent around
the linearization point
Prof. Cesar de Prada, ISA, UVA 55
Liquid flow in a pipe
Non-linear L d q ∆p 0 1 fL
= − ( 2 2 + 2 )q 2 − gh
differential A dt ρ a Cv A
equation
 , q, ∆p 0 , a ) = 0
f (q
∂f ∂f
 −q
(q 0) + (q − q 0 ) +
∂q
0 ∂q 0
∂f ∂f
+ ( ∆p 0 − ( ∆p 0 ) 0 ) + (a − a 0 ) = 0
∂∆p 0 0 ∂a 0

a h
q
∆pv
∆p0
Prof. Cesar de Prada, ISA, UVA 56
Linearized model of the flow
d q A ∆p 0 1 fL
= [ − ( 2 2 + 2 )q 2 − gh ]
dt L ρ a Cv A
d ∆q A ∆ (∆p 0 )  1 fL   2 
= [ − ( 2 2 + 2 )2q  ∆q +  3 2 q 2  ∆a ]
dt L ρ  a Cv A 0  a C v 0
1 d ∆q
+ ∆q =
A 1 fL  dt
 ( 2 2
+ 2
) 2 q 
 L a C v A 0
1  2 
= [∆ (∆p 0 ) +  3 2 ρq 2  ∆a ]
 1 fL  a Cv 0
ρ
 2 2
( + 2
) 2 q 
 a Cv A 0
d ∆q
τ + ∆q = K1∆ (∆p 0 ) + K 2 ∆a
dt

Prof. Cesar de Prada, ISA, UVA 57


Changes in the operating point
d ∆q
τ + ∆q = K1∆ ( ∆p 0 ) + K 2 ∆a
dt
 
1 
 q 

τ= K2 =  2 2 
A 1 fL   a (β a 2 C 2 + 1 + fLa Cv 
 (β + + ) 2q   )
 0
v
 L a 2 2
C v A 2
0 A2

q
τ increases in operating points
corresponding to large openings
K2 decreases in operating points
corresponding to large openings
t
Prof. Cesar de Prada, ISA, UVA 58
Linearized model of the tank
dT V2
Ah = q (Ti − T ) + si Ti y h = cte.
dt ρc e R
f (T , T, q, V) = 0
∂f   ∂f ∂f ∂f
(T − T0 ) + (T − T0 ) + (q − q 0 ) + (V − V0 ) = 0
∂T 0
 ∂T 0 ∂q 0 ∂V 0

Ti d∆T 2V0
q Ah = −q 0 ∆T + (Ti0 − T0 )∆q + ∆V
dt ρc e R
Ah d∆T (T − T ) 2V0
+ ∆T = i 0 0 ∆q + ∆V
q 0 dt q0 ρc e Rq 0
V R T d∆T
τ + ∆T = K1∆q + K 2 ∆V
dt
Prof. Cesar de Prada, ISA, UVA 59
Similarities
a q h

∆pv
∆p0
Ti d ∆q
q τ + ∆q = K1∆ ( ∆p 0 ) + K 2 ∆a
dt

d∆T
τ + ∆T = K1∆q + K 2 ∆V
dt
V R T

Prof. Cesar de Prada, ISA, UVA 60


Linearized model of the reactor
d cA − E RT
V = Fc Ai − Fc A − Vke cA CAi Product A
dt
d cB − E RT F
V = − Fc B + Vke cA
dt

Two equations
A⇒B
f1 (c A , c A , F, c Ai ) = 0 CA CB T

f 2 (c B , c B , c A , F) = 0
F

Prof. Cesar de Prada, ISA, UVA 61


Linearized model (1)
d cA − E RT
V = Fc Ai − Fc A − Vke cA Operating point:
dt F0 , c A 0 , c B0 , c Ai 0
Taylor series expansion.....
d∆c A −E
V = −(F0 + Vke RT0 )∆c A + (c Ai 0 − c A 0 )∆F + F0 ∆c Ai
dt

d∆c A F0 −E (c − c ) F
= −( + ke RT0 )∆c A + Ai 0 A 0 ∆F + 0 ∆c Ai
dt V V V
d∆c A
= a11∆c A + b11∆F + d11∆c Ai
dt
Value computed at the operating point
Prof. Cesar de Prada, ISA, UVA 62
Linearization point
dc A −E
If the linearization point V = Fc Ai − Fc A − Vke RT c A = 0
dt
corresponds to an
dc −E
equilibrium point: V B = −Fc B + Vke RT c A = 0
dt
If cAi0 = 8 y cA0 = 0.8 ⇒ cB0 = 7.2
If F0 = 26.66 and V = 80 ⇒ ke-E/RT = 2.999
d∆c A −E
V = −(F0 + Vke RT0 )∆c A + (c Ai 0 − c A 0 )∆F + F0 ∆c Ai
dt
d∆c A F0 −E (c − c ) F
= −( + ke RT0 )∆c A + Ai 0 A 0 ∆F + 0 ∆c Ai = −3.332∆c A + 0.09∆F + 0.333∆c Ai
dt V V V
Prof. Cesar de Prada, ISA, UVA 63
Linearized model (2)
d cB − E RT
V = − Fc B + Vke cA
dt
Using the Taylor series expansion…

d∆c B −E F0 c B0
= ke RT0
∆c A − ∆c B − ∆F = 2.999∆c A − 0.333∆c B − 0.09∆F
dt V V

d∆c B
= a 21∆c A + a 22 ∆c B + b 21∆F
dt
d∆c A
= a11∆c A + b11∆F + d11∆c Ai
dt

Prof. Cesar de Prada, ISA, UVA 64


Linear states space model
d∆c A
= a11∆c A + b11∆F + b12 ∆c Ai
dt
d∆c B
= a 21∆c A + a 22 ∆c B + b 21∆F
dt

 d ∆c A 
 d t   a 11 0  ∆c A   b11 b12   ∆F 
  =     +    
 d ∆c B  a 21 a 22   ∆c B   b 21 0  ∆c Ai 
 d t 
∆c A 
∆c B = (0 1)
dx
= Ax + Bu

 ∆c B 
dt
y = Cx

Prof. Cesar de Prada, ISA, UVA 65


Isothermal chemical reactor
 d ∆c A 
 d t   − 0.33 0  ∆c A   0.09 0.333   ∆F 
 =   +   
 d ∆c B   3 − 0 .33 ∆
 B  
c − 0. 09 0 ∆
  Ai 
c
 d t  Product A
CAi
F
∆c A 
∆c B = (0 1) 

 B
c Isothermal
reactor
A⇒B
CA CB T

F
Prof. Cesar de Prada, ISA, UVA 66
State space models
d∆h k 1
q =− ∆h + ∆q
dt 2A h 0 A
d∆h
= α∆h + β ∆q
h dt
∆h = 1.∆h

d ∆q
a q τ + ∆q = K 1 ∆(∆p 0 ) + K 2 ∆a
h dt
∆p0 d ∆q − 1 K K 2  ∆(∆p 0 )
= ∆q +  1 
dx dt τ  τ τ   ∆a 
= Ax + Bu
dt ∆q = 1∆q
y = Cx
Prof. Cesar de Prada, ISA, UVA 67
Linear states space models
x state variables: If its value is
dx
= Ax + Bu known at an initial time instant,
dt as well as the values of the
y = Cx + Du input u(t) along time, then the
value of the outputs y(t) is
determined along time
t
Analytical
solution: x(t ) = e A( t −t0 ) x(t 0 ) + ∫ e A(t − τ ) Bu (τ)dτ
t0

Prof. Cesar de Prada, ISA, UVA 68


Many equivalent representations
dx
= Ax + Bu u y
dt
y = Cx

z = Px x = P -1z = [PAP -1 ] z + [PB]u


dz
dz dt
= PA (P -1z) + PBu
dt y = [CP -1 ] z
y = CP -1z There exist many equivalent
formulations of the model from the
input /output point of view
Prof. Cesar de Prada, ISA, UVA 69
Eigenvalues
= [PAP ] z + [PB]u
dx dz
= Ax + Bu -1

dt dt
y = Cx y = [CP -1
]z
A − λI = 0 PAP −1 − λI = 0
PAP −1 − λPP −1 = 0
Equivalent
representations P(A − λI)P −1 = 0
have the same P A − λI P −1 = 0
eigenvalues A − λI = 0

Prof. Cesar de Prada, ISA, UVA 70


Impulse response model
t
y( t ) = Ce At x (0) + ∫ Ce A ( t −τ ) Bu( τ)dτ
0

if u is a unit impulse and the initial state is zero :


t

y(t) = ∫ Ce A ( t −τ )
Bδ( τ)dτ = Ce B = g( t )
At Impulse response

0 t

δ(t) g(t)
y( t ) = ∫ g ( t − τ)u (τ)dτ
0

y( t ) = ∫ g( t − τ) u( τ)dτ
0
Prof. Cesar de Prada, ISA, UVA 71
Impulse response model
t
y( t ) = ∫ g ( t − τ)u (τ)dτ t−τ=σ dτ = − dσ
τ = 0 ⇒ σ = t
0


y ( t ) = ∫ g ( t − τ ) u ( τ ) dτ τ = t ⇒ σ = 0
0 t
y ( t ) = ∫ g ( σ ) u ( t − σ ) dσ
0
δ(t) g(t)

y ( t ) = ∫ g ( σ ) u ( t − σ ) dσ
0

Prof. Cesar de Prada, ISA, UVA 72

You might also like