Lecture 5:

Nonlinear Systems

Dominik Bongartz

Systems Analysis of Chemical Processes – H06T4A

 Lecture Outline: Systems Analysis of Chemical Processes

Part 1: Time domain

L1: Modeling & state-space representation

L2: Linear algebra review 5.1: Solvability of nonlinear systems

L3: Analytical solution of linear systems

L4: Steady states & stability of linear systems 5.2 Steady states of nonlinear systems
L5: Nonlinear systems

Part 2: Frequency domain 5.3: Stability & linear approximation of nonlinear systems
L6: The Laplace transformation

L7: The transfer function 5.4: Special behavior of nonlinear systems

L8: Gain, poles, & zeros of transfer functions

L9: Graphical representation of transfer functions

L10: Common transfer functions

3
 Part 5.1:
Solvability
of Nonlinear Systems

4
 Intended Learning Outcomes

After this lecture part, you….

• know some difficulties associated with solving nonlinear dynamic models

• can give an example how they can play out in practice

5
 Solution of Dynamic Models

Linear time-invariant
model

"̇ # = % " # + ' ((#)

+ # = , " # + - ((#)
" 0 = "!

Solution: " # = / "# "! + 0 / " #$%

'( 1 21
!

6
 Solution of Dynamic Models

Linear time-invariant Nonlinear

model model

"̇ # = % " # + ' ((#) "̇ # = 3 " # , ( #

+ # = , " # + - ((#) + # = ℎ " # ,( #
" 0 = "! " 0 = "!
#
No closed-form solution L
Solution: " # = / "# "! + 0 / " #$%
'( 1 21 à Numerical integration in simulator
!
Solution may not even exist, or may not be unique LL

Theoretical remedy: Under certain conditions on 3 and 6, K

can guarantee a unique solution for some interval of time
In practice: no solution à error message of simulator, and
a) there is a mistake in your model, or J
b) you have reached the validity limit of your model

7
 Example: Nonlinear Model with Solution for Limited Time

"̇ # = " − 1
" 0 = 0.5

Simulation result (numerical integration):

"& à Solution exists for a certain time

à Integration fails once " t < 0

E.g., error message in simulator Dymola:

8
 Intended Learning Outcomes

After this lecture part, you….

ü know some difficulties associated with solving nonlinear dynamic models

ü can give an example how they can play out in practice

9
 Part 5.2:
Steady States
of Nonlinear Systems

10
 Intended Learning Outcomes

After this lecture part, you….

• can name the cases that can occur regarding existence and uniqueness of steady states

• can compute steady states for simple nonlinear systems

• can give an example for a system with multiple steady states

11
 Steady States of Dynamic Models

Linear time-invariant
model

"̇ # = % " # + ' ((#)

% "' = −' ('

Linear equation system

à For given (' :

• Unique steady state
• No steady state
• Infinitely many steady states

à Easy to check: det %

à Easy to compute steady states

12
 Steady States of Dynamic Models

Linear time-invariant Nonlinear

model model

"̇ # = % " # + ' ((#) "̇ # = 3 " # , ( #

% "' = −' (' 3 "( , (( = 0

Linear equation system Nonlinear equation system

à For given (' : à For given (' :

• Unique steady state • Unique steady state
• No steady state • No steady state
• Infinitely many steady states • Infinitely many steady states
• Finitely many (but > 1) steady states
à Easy to check: det % à In general very hard to check
à Easy to compute steady states à Harder to compute a steady state

13
 Example 1: Steady States in Nonlinear Models

2"& (#) ⋅ (1 − "& (#))

"̇ # =
"& (#) − ") (#)

Steady states: "̇ # ≡ 0 ")

⇔ 3 "' , (' = 0
Steady states
2"',& ⋅ (1 − "',& ) 0
⇔ =
"',& − "',) 0 "&

⇔ "',& = 0 ∧ "',) = 0 ∨ ("',& = 1 ∧ "',) = 1)

0 1
⇔ "' = ∨ "' =
0 1

14
 Example 2: Steady States in Nonlinear Models

Build model: F(#)

̇ = G(t)
H I G(#)
̇ = H 6 sin(F(#))

Bring to state-space ") (#)

"̇ # = +!
sin(" (#)) ,
representation: + &
"

F # 6
" # ≔ , N≔
G t I

Stead states: "',)

N& 0
sin("',& ) =
N) 0

⇔ "',) = 0 ∧ sin "',& = 0

OP à Infinitely many steady states
⇔ "' = , O∈ℤ à Not all equally likely to observe (stability!)
0

15
 Example 3: Steady States in Nonlinear Models

Combustion chamber

Stead states: “cold” / “non-ignited” “hot” / “ignited”

• No reaction • Combustion reaction taking place

• Outlet: fuel + air • Outlet: mostly combustion products
• S,-. = S/0 + excess air
• S,-. ≫ S/0

à Which one will be observed depends on history (dynamic model!)

16
 Intended Learning Outcomes

ü After this lecture part, you….

ü can name the cases that can occur regarding existence and uniqueness of steady states

ü can compute steady states for simple nonlinear systems

ü can give an example for a system with multiple steady states

17
 Part 5.3:
Stability & Linear Approximation
of Nonlinear Systems

18
 Intended Learning Outcomes

After this lecture part, you….

• can describe name differences between the stability properties of nonlinear and linear

• can linearize a nonlinear system around an operating point

• can analyze stability steady states of nonlinear systems via linearization

19
 Reminder: Stability of Steady States

Definition: Stability
A steady state "' is called stable if for every neighborhood of "' with radius U > 0 there is a neighborhood
with radius V > 0 such that for all initial values "! with ∥ "! − "' ∥ < V, the solution "(#) fulfills the condition
∥ " # − "' ∥ < U for all # > 0. A steady state that is not stable is called unstable.

Put differently, …

stable means:
“If you start close enough to the steady state, the
solution will stay arbitrarily close to that steady state.”

unstable means:
“There are solutions that start arbitrarily close to the
steady state but still move far away from it.”

20
 Reminder: Asymptotic Stability of Steady States

Definition: Asymptotic Stability

A steady-state solution "' is called asymptotically stable if it is stable and additionally all solutions for which
∥ "! − "' ∥ < V also fulfill lim "(#) = "' .
#→2

Put differently, …

asymptotically stable means:

“If you start close enough to the steady state, the
solution will stay arbitrarily close to that steady state
AND eventually it will go towards the steady state”

21
 Stability in Linear vs. Nonlinear Models

Linear time-invariant
model

• Stability properties the same for all steady states

à property of the system

• Asymptotically stable: solutions from all initial

points converge to steady state

• Unstable: all solutions outside the stable

subspace go to infinity ( "(#) → ∞ for # → ∞)

22
 Stability in Linear vs. Nonlinear Models

Linear time-invariant Nonlinear

model model

• Stability properties the same for all steady states • Steady states of a system can have
à property of the system different stability properties

• Asymptotically stable: solutions from all initial • Asymptotically stable: solutions may only
points converge to steady state converge to steady state from certain initial
points (neighborhood with V < ∞)

• Unstable: all solutions outside the stable • Unstable: solutions may go elsewhere
subspace go to infinity ( "(#) → ∞ for # → ∞) (but have "(#) < ∞):
• to other steady states
• to limit cycles (see Part 5.4)

23
 Example: Stability of Steady States in Nonlinear Models

2"& ⋅ (1 − "& ) ")

0 1
"̇ # = ⇒ "' = ∨ "' =
"& − ") 0 1

"&

24
 Example: Stability of Steady States in Nonlinear Models

2"& ⋅ (1 − "& ) ")

0 1
"̇ # = ⇒ "' = ∨ "' =
"& − ") 0 1

“Experimental stability analysis”: "&

0
à "' = looks
0
unstable
"& ")
0.01 0.01
"! = "! =
0 0
0 0
"! = "! =
0 0

−0.01 −0.01
"! = "! =
0 0

# #

25
 Example: Stability of Steady States in Nonlinear Models

2"& ⋅ (1 − "& ) ")

0 1
"̇ # = ⇒ "' = ∨ "' =
"& − ") 0 1

“Experimental stability analysis”: "&

1
à "' = looks
1
asymptotically stable
"& ")

1.9
"! = 1.9
1 1 "! = 1
"! = 1 "! =
1 1

0.1
0.1 "! =
"! = 1
1

# #

26
 Example: Stability of Steady States in Nonlinear Models

2"& ⋅ (1 − "& )
"̇ # =
"& − ")

27
 Key Idea for Stability Analysis: Linear Approximation

3(")

à Can locally get good linear approximation

" of nonlinear function

28
 Stability Analysis of Nonlinear Models Via Linear Approximation

0. Compute stationary operating point ("' , (' )

1. Introduce deviation variables around operating point

2. Linearize around operating point

3. Analyze stability of linear system and transfer back to nonlinear system – if possible!

29
 Stability Analysis of Nonlinear Models: Step 1 – Transform to Deviation Variables

Original
Nonlinear model

"̇ # = 3 " # , ( #
+ # = ℎ(" # , ( # )

• Consider stationary operating point ("' , (' )

• Corresponding output +(

Deviation in (: Δ( # ≔ ( # − ('
Deviation in ": Δ" # ≔ " # − "'
Deviation in +: Δ+ # ≔ + # − +'
“deviation variables”
(= shift of coordinates)

30
 Stability Analysis of Nonlinear Models: Step 1 – Transform to Deviation Variables

Original Nonlinear model in

Nonlinear model deviation variables

"̇ # = 3 " # , ( # Δ"̇ # = 3 "' + Δ" # , (' + Δ( #

+ # = ℎ(" # , ( # ) Δ+ # = ℎ "' + Δ" # , (' + Δ( # − +'

• Consider stationary operating point ("' , (' ) • Still nonlinear, exact reformulation
• Corresponding output +(
• Key observations:
Deviation in (: Δ( # ≔ ( # − (' • Steady state at Δ", Δ( = 0,0
Deviation in ": Δ" # ≔ " # − "' • Often deviation from operating
Deviation in +: Δ+ # ≔ + # − +' point small (control!)

“deviation variables” à Can get good linear approximation

(= shift of coordinates)

31
 Stability Analysis of Nonlinear Models: Step 2 – Linear Approximation via Taylor Series

Δ"̇ # = 3 "' + Δ" # , (' + Δ( #

Taylor series around ("' , (' ):

_3 _3
̇
Δ" # = 3 "' , (' + " , ( Δ" # + " , ( Δ( # + quadratic & higher order terms
_" ' ' _( ' '

=0 Jacobi matrices of 3 ≈ 0 for small Δ" # , Δ((#)

(steady state) evaluated at operating point

_3 _3
⇒ Δ"̇ # ≈ "' , (' Δ" # + " , ( Δ( #
_" _( ' '

=: % =: '

32
 Stability Analysis of Nonlinear Models: Step 2 – Linear Approximation via Taylor Series

Δ"̇ # = 3 "' + Δ" # , (' + Δ( #

Taylor series around ("' , (' ):

_3 _3
̇
Δ" # = 3 "' , (' + " , ( Δ" # + " , ( Δ( # + quadratic & higher order terms
_" ' ' _( ' '

=0 Jacobi matrices of 3 ≈ 0 for small Δ" # , Δ((#)

(steady state) evaluated at operating point

_3 _3
⇒ Δ"̇ # ≈ "' , (' Δ" # + " , ( Δ( #
_" _( ' '

=: % =: '
_ℎ _ℎ
Similarly: ⇒ Δ+ # ≈ "' , (' Δ" # + " , ( Δ( #
_" _( ' '

=: , =: -

33
 Stability Analysis of Nonlinear Models: Step 2 – Linear Approximation via Taylor Series

Δ"̇ # = 3 "' + Δ" # , (' + Δ( #

Taylor series around ("' , (' ):

_3 _3
̇
Δ" # = 3 "' , (' + " , ( Δ" # + " , ( Δ( # + quadratic & higher order terms
_" ' ' _( ' '

=0 Jacobi matrices of 3 ≈ 0 for small Δ" # , Δ((#)

(steady state) evaluated at operating point

_3 _3
⇒ Δ"̇ # ≈ "' , (' Δ" # + " , ( Δ( #
_" _( ' ' Linear model in
deviation variables
=: % =: '
_ℎ _ℎ Δ"̇ # = % Δ" # + ' Δ((#)
Similarly: ⇒ Δ+ # ≈ "' , (' Δ" # + " , ( Δ( #
_" _( ' ' Δ+ # = , Δ" # + - Δ((#)

=: , =: -

34
 Stability Analysis of Nonlinear Models: Step 3 – Stability Analysis via Linear Approximation

Theorem: Approximation of Dynamics

Locally around the operating point ("' , (' ), the dynamics of the nonlinear system can be
approximated by an LTI model with
34 34 37 37
%= "' , (' , ' = ("' , (' ), , = ("' , (' ), D = ("' , (' )
35 36 35 36

à Can use for local stability analysis, controller design, …

35
 Stability Analysis of Nonlinear Models: Step 3 – Stability Analysis via Linear Approximation

Theorem: Approximation of Dynamics

Locally around the operating point ("' , (' ), the dynamics of the nonlinear system can be
approximated by an LTI model with
34 34 37 37
%= "' , (' , ' = ("' , (' ), , = ("' , (' ), D = ("' , (' )
35 36 35 36

à Can use for local stability analysis, controller design, …

Theorem: Stability Analysis via Linear Approximation

34
Let % = 35 "' , (' .

1. If all eigenvalues l8 of % satisfy Re l8 < 0, then ("' , (' ) is asymptotically stable.

2. If at least one eigenvalue l8 of % satisfies Re l8 > 0, then ("' , (' ) is unstable.
à No conclusion if all Re l8 ≤ 0 and at least one has Re l8 = 0!

36
 Example: Stability Analysis via Linearization

2"& ⋅ (1 − "& ) 0 1
"̇ # = ⇒ "' = ∨ "' =
"& − ") 0 1

37
 Example: Stability Analysis via Linearization

2"& ⋅ (1 − "& ) 0 1
"̇ # = ⇒ "' = ∨ "' =
"& − ") 0 1

_3& _3&
_3 _"& _") 2 − 4"& +0
⇒ = =
_" _3) _3) 1 −1
_"& _")
_3 2−4⋅0 +0 = 2 +0
⇒ "( = =: %
_" 1 −1 1 −1

⇒ l& = 2, l) = −1

38
 Example: Stability Analysis via Linearization

2"& ⋅ (1 − "& ) 0 1
"̇ # = ⇒ "' = ∨ "' =
"& − ") 0 1

_3& _3&
_3 _"& _") 2 − 4"& +0
⇒ = =
_" _3) _3) 1 −1
_"& _")
_3 2−4⋅0 +0 = 2 +0
⇒ "( = =: %
_" 1 −1 1 −1

⇒ l& = 2, l) = −1
>0
à unstable!

39
 Example: Stability Analysis via Linearization

2"& ⋅ (1 − "& ) 0 1
"̇ # = ⇒ "' = ∨ "' =
"& − ") 0 1

_3& _3&
_3 _"& _") 2 − 4"& +0
⇒ = =
_" _3) _3) 1 −1
_"& _")
_3 2−4⋅0 +0 = 2 +0
⇒ "( = =: %
_" 1 −1 1 −1

⇒ l& = 2, l) = −1
>0
à unstable!
à local dynamics: saddle

40
 Example: Stability Analysis via Linearization

2"& ⋅ (1 − "& ) 0 1
"̇ # = ⇒ "' = ∨ "' =
"& − ") 0 1

_3& _3&
_3 _"& _") 2 − 4"& +0
⇒ = =
_" _3) _3) 1 −1
_"& _")
_3 2−4⋅1 +0 −2 +0
⇒ "( = = =: %
_" 1 −1 +1 −1

⇒ l& = −2, l) = −1

41
 Example: Stability Analysis via Linearization

2"& ⋅ (1 − "& ) 0 1
"̇ # = ⇒ "' = ∨ "' =
"& − ") 0 1

_3& _3&
_3 _"& _") 2 − 4"& +0
⇒ = =
_" _3) _3) 1 −1
_"& _")
_3 2−4⋅1 +0 −2 +0
⇒ "( = = =: %
_" 1 −1 +1 −1

⇒ l& = −2, l) = −1
both < 0
à asymptotically stable!
à local dynamics: stable focus

42
43 Faculty, department, unit ...
44 Faculty, department, unit ...
45 Faculty, department, unit ...
 Intended Learning Outcomes

After this lecture part, you….

ü can describe name differences between the stability properties of nonlinear and linear

ü can linearize a nonlinear system around an operating point

ü can analyze stability steady states of nonlinear systems via linearization

46
 Part 5.4:
Special Dynamic Behavior of
Nonlinear Systems

47
 Intended Learning Outcomes

After this lecture part, you….

• can describe the concepts of limit cycles and chaos

48
 Limit Cycles

Definition: Limit cycle

A limit cycle is a closed trajectory to which at least one other trajectory converges as # → ∞ or # → −∞.

Limit cycle

Occur in real systems!

• Electric circuits (Van der Pol oscillator, …)
• Chemistry (Belousov-Zhabotinsky reaction, …)
• …

49
 Chaos

Description: Chaos
Solutions are extremely sensitive to initial conditions.

“When the present determines the future, but the approximate present
does not approximately determine the future”. E.N. Lorenz

Occurs…
• in complex real systems (e.g., weather)
• in some very simple models (e.g., Lorenz model)!
• only if p5 ≥ 3

50
 Intended Learning Outcomes

After this lecture part, you….

ü can describe the concepts of limit cycles and chaos

51
 Lecture Outline: Systems Analysis of Chemical Processes

Part 1: Time domain

L1: Modeling & state-space representation

L2: Linear algebra review 5.1: Solvability of nonlinear systems

L3: Analytical solution of linear systems

L4: Steady states & stability of linear systems 5.2 Steady states of nonlinear systems
L5: Nonlinear systems

Part 2: Frequency domain 5.3: Stability & linear approximation of nonlinear systems
L6: The Laplace transformation

L7: The transfer function 5.4: Special behavior of nonlinear systems

L8: Gain, poles, & zeros of transfer functions

L9: Graphical representation of transfer functions

L10: Common transfer functions

52
See you next week!