9/30/2019

Ali M. Sahlodin
Department of Chemical Engineering
AmirKabir University of Technology
1397 S.H

F, C1
 Fixed inlet concentration

dC2 F
(t )   C1  C2 (t )  Ordinary-differential
F, C2
dt V equation

 Variable inlet concentration

dC2 F
(t )   C1 (t )  C2 (t )  Differential-algebraic equation
dt V
C1 (t )   (t )

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

2

1
 9/30/2019

 Fully-implicit DAE
 t ), t )  0
F(x(t ), x( F
Singular
 x

Implicit ODE F
 Non-singular  t )  f(x(t ), t )
x(
 x

 Semi-explicit DAE
 t ), y(t ), t )  0
F(x(t ), x( F
Non-singular
G(x(t ), y(t ), t )  0  x

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

3

 Conservation of mass/energy
 Hydraulic relations
 Reaction kinetics
 Thermodynamic relations
 Total volume
…

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

4

2
 9/30/2019

FV

Fin P, T
FL
Q

Sahlodin et. al. 2016, AIChE J.

All algebraic equations

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

5

 Method of lines
 Convert PDE to system of ODEs by discretizing
the space

 Example 1: one-dimensional heat

equation
T  2T

t x 2

T j T j 1  2T j  T j 1
  0, j  2,, N  1 ODE System
t ( x) 2

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

6

3
 9/30/2019

 Example 2:

y z  2 y
b  2
t x x
 z
2
Discretization of the spatial derivative  a
results in a DAE x 2

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

7

 System of ODEs
 t )  f(x(t ), y(t ), t ,  )
x(
Singular perturbation problem
 y(
 t )  g(x(t ), y(t ), t ,  )

 If 0    1 (variables y have fast dynamics)

 t )  f(x(t ), y(t ), t ,0)

x(
Semi-explicit DAE
0  g(x(t ), y(t ), t ,0)

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

8

4
 9/30/2019

x  x 2  x 3
 What is stiffness?
 No agreed-upon definition!
 Depends on differential
equation, initial conditions,
and numerical method

 Presence of both fast and slow

dynamics
 Non-stiff solvers must take very
small steps

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

9

 DAEs are similar to stiff ODEs in the limit.

 Recall singular perturbation problem

 t )  f(x(t ), y(t ), t ,  )
x(  t )  f(x(t ), y(t ), t ,0)
x(
 y(
 t )  g(x(t ), y(t ), t ,  ) 0  g(x(t ), y(t ), t ,0)

 Stiff ODE methods can be used to solve DAEs

(discussed later).

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

10

5
 9/30/2019

1. Accuracy of numerical solution

2. Stability of numerical solution
3. Constraint satisfaction
4. Initialization

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

11

dx
 Numerical solution of ODEs (t )  f ( x, t )
dt

h2 
x(tk )  x(tk 1 )  hf ( xk 1 , tk 1 )  f ( xk 1 , tk 1 )  ...
2
•Forward Euler
x(tk )  x(tk 1 )  hf ( xk 1 , tk 1 )  O(h 2 )

•Backward Euler
x(tk )  x(tk 1 )  hf ( xk , tk )  O(h 2 )

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

12

6
 9/30/2019

F, C1
 Tank example: Case I
 Dynamic Simulation problem (known input)
dC2 F
(t )   C1 (t )  C2 (t ) 
dt V F, C2
C1 (t )   (t )

Order of error O(h 2 )

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

13

F, C1
 Tank example: Case II
 Dynamic design problem (known output)
dC2 F
(t )   C1 (t )  C2 (t ) 
dt V F, C2
C2 (t )   (t )

Order of error O ( h)

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

14

7
 9/30/2019

F, C1
 Tank example: Case III
 Tanks in series with known output

dC2 F
(t )   C1 (t )  C2 (t ) 
dt V F, C2
dC3 F
(t )   C2 (t )  C3 (t ) 
dt V
C3 (t )   (t )

F, C3
Order of error O (1)

Reducing step size will not improve order of accuracy!

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

15

F, C1
 Tank example: Case I

dC2 F
(t )   C1 (t )  C2 (t ) 
dt V
C1 (t )   (t ) F, C2

How close is this to an ODE?

dC1
Differentiate the system until (t ) appears.
dt
One differentiation is enough.

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

16

8
 9/30/2019

F, C1
 Tank example: Case II

dC2 F
(t )   C1 (t )  C2 (t ) 
dt V
C2 (t )   (t ) F, C2

dC1
Differentiate the system until (t ) appears.
dt
Two differentiations are needed.

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

17

F, C1
 Tank example: Case III
dC2 F
(t )   C1 (t )  C2 (t ) 
dt V
dC3 F
(t )   C2 (t )  C3 (t )  F, C2
dt V
C3 (t )   (t )

dC1
Differentiate the system until (t ) appears.
dt F, C3
Three differentiations are needed.

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

18

9
 9/30/2019

 The minimum number of times a DAE system or

part of it must be differentiated in order to
arrive at an implicit ODE

F0 : F(x(t ), x(
 t ), t )  0

d
F1 :  t ), t )  0
F(x(t ), x(
dt

d2
F 2 :  t ), t )  0
F(x(t ), x(
dt 2
d j 1
F j 1   t ), t )  0
F(x(t ), x(
dt j 1
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.
19

 Find index
x
mxT 0
L
y
my mg T 0
L
x 2  y 2  L2 0

What is the index?

Wikipdia by Ruryk

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

20

10
 9/30/2019

 Semi-explicit DAE
 t ), y(t ), t )  0 F
F(x(t ), x( Non-singular
 x
G(x(t ), y(t ), t )  0

is index 1 IFF
g
Non-singular
y
 Example: a CSTR with the rate equation
written separately.

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

21

 High-index DAEs (v>1) can be problematic.

 ODE solvers with high-index DAEs may
 Converge poorly
 Converge to wrong results!
 Fail to converge

 Problem formulation can affect the index.

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

22

11
 9/30/2019

 Use high-index DAEs solvers

 No general-purpose solver available
 Solvers available for specialized forms
 Reduce DAE index to <=1.

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

23

 Convert a high-index DAE (v>1) to a low-index

DAE (v<2).
 Simplify computational complexity

 How to do it?
 Successive differentiation
 Change of variables
 Regularization

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

24

12
 9/30/2019

x2  x1  f1  0
x3  x2  f 2  0
x3  f3  0

Define z1  x1 and substitute above.

Obtain the index again.

Applicable when time derivative of a state variable does not appear in

the system
Not easily automatable

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

25

 t )  f(x(t ), y(t ), t ,0)

x(  t )  f(x(t ), y(t ), t ,  )
x(
0  g(x(t ), y(t ), t ,0)  y(
 t )  g(x(t ), y(t ), t ,  )

•Converts DAEs to stiff ODEs (we know how to solve ODEs!)

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

26

13
 9/30/2019

x1  x3  f1  0
x2  x1  f 2  0
x2  f3  0 1 apparent algebraic equation

Substitution will reveal two additional constraints

This issue will be discussed further.

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech.

27

14