Systems Dynamics

Prof. Cesar de Prada

Dpt. Systems Engineering and Automatic Control
EII, University of Valladolid, Spain
prada@autom.uva.es
 Outline

 Introduction
 Continuous and discrete dynamics
 Stability of equilibrium points
 Bifurcation analysis of non-linear systems
 Introduction to chaotic behaviour
 Application examples
 Introduction

Systems dynamics study the

time evolution of process
models. Generally, the type of
trajectories followed by the dx
states depend on the value of = f ( x , u , p) x(0) = x 0
dt
the actions u and the initial
conditions x0, but also on the
structure of the mathematical
model representing the process
and the value of the model
parameters p
 Continuous and discrete dynamics

Continuous processes are

represented normally by
ODEs, DAEs or PDEs dx
involving real variables that = f ( x , u , p) x(0) = x 0
dt
change continuously over time
taking any value in a given
range.
Sampled or discrete systems x (k + 1) = F( x (k ), u (k ), p)
are represented normally by
k = kT k = 0,1,2,...
difference equations involving
variables that change only at x ( 0) = x 0
certain time instants
 Example: Chemical reactor
A→B Reactant
dc A −E
cAi
V = Fc Ai − Fc A − Vke RT c A TT
F
dt
Tr Coolant
x = (c Ai − c A ) / c Ai TT T Tri
Fr Reactor
Mass balance x, cA Jacket

Energy balance
Coolant

dT − E RT
Vρc e = Fρc e Ti − Fρc e T + Vke c A ∆H − UA ( T − Tr )
dt
d Tr
Vr ρ r c er = Fr ρ r c er Tri − Fr ρ r c er Tr + UA ( T − Tr )
dt
 Systems dynamics

The local study of systems

dynamics, and in particular dx d∆ x
= f ( x , u , p) → = A∆x + B∆u
stability, can be made using dt dt
the eigenvalues of the
linearized model around the
considered point.

d xe
Points specially important are = f ( x e , u , p) = 0
dt
the equilibrium points.
The steady-state points are given by
the solution of this set of equations
 Systems dynamics

 The numerical value of the A

and B matrices of the linearized A ( p ) x e − λI = 0
models, as well as the
equilibrium points, depend on λ Eigenvalues of A
the parameters p
If Real(λi ) > 0 unstable point.
Real negative λi creates
dx
= f ( x , u , p) overdamped dynamics.
dt
Imaginary negative λi creates
d∆ x underdamped dynamics.
= A(p) x ∆x + B(p) x ∆u
dt e e
Real(λi ) = 0 creates
oscillations.
 Autonomous systems

For a given input trajectory

u(t), the systems dynamics dx
= f ( x , u , p) x(0) = x 0
only depends on the initial dt
point x0. E.g. systems under
closed loop control.

dx
Autonomous system: = f ( x , p) x(0) = x 0
dt

Dynamics of the autonomous

system can be study as a
function of the initial point and
parameters p
 Bifurcations
The numerical value of the A dx
= f ( x , p)
matrix of the linearized model, as dt
well as the equilibrium points, d∆ x
depend on the parameters p = A(p) x ∆x
dt e

Changing p, it may happens that

the eigenvalues of A, or the
number of equilibrium points, A ( p ) x e − λI = 0
change in such way that the new
type of dynamics is created (stable λ Eigenvalues of A
vs. unstable, limit cycle,..). This is
called a bifurcation. Then, p is a
bifurcation parameter.
 Example 1

Possible equilibrium points

dx
= f ( x , µ) = µx − x 3
dt
If µ ≤ 0, there is only one real
solution xe = 0 f ( x e , µ) = µx e − x 3e = 0
If µ > 0, there are three different xe = 0
equilibrium points: x e (µ − x ) = 0 ⇒ 
2

µ =
e 2
 x e
xe = 0, - µ1/2 , µ1/2
The value µ = 0 is a bifurcation
point for the system because the
number of equilibrium points This is called a
changes between 1 and 3 at µ=0 pitchfork bifurcation
 Dynamics at the equilibrium points

Example: Non linear system dx

= µx − x 3
with a parameter µ dt
 Linearized system at d ∆x
= (µ − 3x e2 )∆x
equilibrium point xe dt

 Eigenvalues |A - λI | = 0 µ − 3x e2 − λ = 0
λ = µ − 3x e2

 Equilibrium points: f ( x e , µ) = µx e − x 3e = 0
xe = 0
x e (µ − x ) = 0 ⇒ 
2

µ =
e 2
 x e
 Dynamics at the equilibrium points

 If µ < 0 → xe = 0, λ < 0 x e (µ − x e2 ) = 0 ⇒
The origin is a stable overdamped xe = 0

equilibrium point for any initial µ = x 2
e
condition
λ = µ − 3x e2

µ = -2
red: x0 = 2
Black: x0 = -1
 Dynamics at the equilibrium points

x e (µ − x e2 ) = 0 ⇒
 If µ > 0 → three equilibrium points
 xe = 0, λ = µ >0 unstable point xe = 0

µ = x 2
 xe = µ1/2, λ = µ - 3µ = -2µ < 0 e

stable overdamped equilibrium λ = µ − 3x e2

 xe = - µ1/2, λ = µ - 3µ = -2µ < 0
xe = 2
stable overdamped equilibrium
The origin is unstable and
each of the two stable
overdamped equilibrium
points are reached depending xe = − 2
on the initial point x0 µ=2
purple: x0 = 0.001
green: x0 = -0.001
 Example 2

x 1 = x 2 + x1 (µ − x12 − x 22 )
Possible equilibrium points x 2 = − x1 + x 2 (µ − x12 − x 22 )

0 = x 2 e + x1e (µ − x12e − x 22 e )
They should satisfy:
Substituting x1e and x2e in 0 = − x1e + x 2 e (µ − x12e − x 22 e )
the other equation:
0 = x 2 e + x 2 e (µ − x12e − x 22 e )(µ − x12e − x 22 e )
(0,0)’ is the only x 2 e (1 + (µ − x12e − x 22 e ) 2 ) = 0 ⇒ x 2 e = 0
equilibrium point
0 = x1e (1 + (µ − x12e − x 22 e ) 2 ) ⇒ x1e = 0
 Stability

− x 2 e x1e = x12e (µ − x12e − x 22 e )

(0,0)’ is the only equilibrium
point, but notice that a trajectory x 2 e x1e = x 22 e (µ − x12e − x 22 e )
given by µ = x1e2 + x2e2 also 0 = ( x12e + x 22 e )(µ − x12e − x 22 e )
satisfies the equilibrium x1e = 0, x 2 e = 0
µ = x12e + x 22 e
x 1 = x 2 + x1 (µ − x12 − x 22 )
x 2 = − x1 + x 2 (µ − x12 − x 22 ) For xe = (0,0)’
 µ 1
 ∆x 1  µ − 3x12e − x 22 e 1 − 2 x1e x 2 e   ∆x1  A= 
∆x  =  − 1 − 2 x x 
µ − x12e − 3x 22 e  ∆x 2 
 − 1 µ 
 2 
eig(A) = µ ± j
1e 2 e
 Stability of xe
eig(A) = λ = µ ± j

 If µ < 0, µ = -2
x10 = 1
underdamped x20 = 1
stable
equilibrium
point for any
initial Phase plane
condition
 If µ > 0, (0,0)’ µ=2
unstable x10 = 0.001
equilibrium x20 = 0.001
point
 Limit cycle
For xe = (0,0)’
 If µ < 0, stable equilibrium
 µ 1
point for any initial A= 
condition − 1 µ 
 If µ > 0, unstable eig(A) = λ = µ ± j
equilibrium point
 µ = 0 is a bifurcation point.
The system changes
dynamics from a stable to
unstable equilibrium point
and the trajectory moves to
a cycle limit. This is called a
Limit cycle: Periodic
Hopf bifurcation
isolated trajectory
 Limit cycle
For points in the
trajectory satisfying
 ∆x 1  µ − 3x12e − x 22 e 1 − 2 x1e x 2 e   ∆x1 
∆x  =  − 1 − 2 x x 
µ − x12e − 3x 22 e  ∆x 2 
µ = x12e + x 22 e
 2  1e 2 e

 − 2 x12e 1 − 2 x1e x 2 e 
A= 
 − 1 − 2 x x
1e 2 e − 2 x 2
2e 
− 2 x12e − λ 1 − 2 x1e x 2 e
=0
− 1 − 2 x1e x 2 e − 2 x 2 e − λ
2

(−2 x12e − λ )(−2 x 22 e − λ ) − 4 x12e x 22 e − 1 = 0

Solutions are always λ2 + 2λ ( x 22 e + x12e ) + 1 = 0

stable on this trajectory λ2 + 2λµ + 1 = 0
Stable cycle limit λ = −µ ± µ 2 − 1 Re al(λ ) < 0
 Chaotic behaviour

Generally, it is possible to
dx
predict the future behaviour of = f ( x , u , p) x(0) = x 0
model states as a function of its dt
initial value.
Nevertheless, certain systems If there are no stable
have such huge sensibility to equilibrium points and
the initial conditions, that it is possible cycle limits are
impossible to predict its long unstable, the solution may
term trajectory. This is called a wander never repeating
chaotic behaviour. trajectory and showing a
chaotic behaviour
 Lorenz equations
x 1 = σ( x 2 - x1 )
x 2 = rx1 - x 2 - x1x 3
T2
x 3 = − bx 3 + x1x 2
x1 turning speed of the convective
rolls
x2 temperature difference between
ascending and descending currents
T1 > T2 x3 distortion of vertical temperature
profile from linearity
Convention rolls due to a σ Prandtl number
temperature difference in a fluid r Rayleigh number/ critical
which density decreases with
Rayleigh number
temperature
b geometric factor
 Lorenz equations

Solution for x0 = (0,1,0) σ = 10, b = 8/3, r = 28

Comparison with the solution for x0 = (0, 1.01, 0) σ = 10, b = 8/3, r = 28

Lorenz equations

Comparison between the solution for

x0 = (0, 1, 0) σ = 10, r = 28,
b=2.6666667 and
x0 = (0, 1, 0) σ = 10, r = 28,
b=2.6666666667