1/10/2021

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

 Stream tearing
 Criterion 1: minimum tear streams
 Criterion 2: non-redundant tear set
 Criterion 3: minimum tear variables
 Criterion 4: easy initialization
 Solving recycle loops

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 2

1
 1/10/2021

 Choose q<0 to reduce maximum Eigenvalue

max
x k
 
x k 1

Assume  min  0 AND  min   max

max

q max
 1
 Bound q to avoid instability (qmin<q<0)
 What if min/max Eigenvalues are far apart?
 Possible divergence
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 3

 DEM uses a single relaxation parameter for all

variables.
 Wegstein’s Method: use a separate relaxation
parameter for each tear variable

si g ( xik )  g ( xik 1 )
qi  , si 
si  1 xik  xik 1

 Suitable for subsystems with low interactions (e.g., single

recycle without reactor).
 Again, bound q to avoid instability (qmin<q<0)
 Apply relaxation every 2-5 iterations (for stability)

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 4

2
 1/10/2021

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 5

 Converged solution

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 6

3
 1/10/2021

 Direct substitution does not apply to design spec. blocks

(why?)
 Solve f(x)=0 iteratively through successive linearization

f ( x( k ) )
x( k 1)  x( k ) 
f '( x( k ) )

Requires derivative calculations Courtesy of Benoit Chachuat

Fast (quadratic) convergence if started close to solution

Divergence possible if started too far from solution
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 7

f ( x)  exp( x)  x  2  0
Courtesy of Benoit Chachuat

Converged in only 3 iterations!

Observe quadratic convergence

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 8

4
 1/10/2021

Initial guess should be sufficiently close to solution,

or expect:
 Divergence
 Oscillatory iterations

Courtesy of Benoit Chachuat

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 9

 Exact derivatives are often costly or impossible to

calculate in SM simulation.
 Why impossible?
 Iterative, possibly ad hoc procedures used to solve each block
(instead of closed-form functions).
 No propagation of derivatives through the blocks (even if
calculated within each block)

Courtesy of Benoit Chachuat

Each block is a black box in SM simulation Finite difference!

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 10

5
 1/10/2021

 Derivative Approximation using the past iterate

Courtesy of Benoit Chachuat

( x( k )  x( k 1) ) f ( x( k ) )
x( k 1)  x( k ) 
f ( x( k ) )  f ( x ( k 1) )

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 11

f ( x)  exp( x)  x  2  0
Courtesy of Benoit Chachuat

 Pros
 Increased stability compared to Newton method
 Only 1 function evaluation per iteration
 Cons
 Slower convergence (superlinear, not quadratic)
 Two initial guesses needed

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 12

6
 1/10/2021

 We cannot use direct substitution here!

 Manipulated and specified variables are not the same!

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 13

 We cannot use direct substitution here!

 Manipulated and specified variables are not the same!

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 14

7
 1/10/2021

 f1 ( x1,..., xn )  0


 f ( x ,..., x )  0
 n 1 n

1
x ( k 1)  x ( k )   J ( x ( k ) )  f ( x(k ) )
 

Courtesy of Benoit Chachuat

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 15

 High computational cost

 Jacobian evaluation
 Jacobian inversion

 Broyden’s method
 Approximate Jacobian by finite difference
 Update inverse of Jacobian
( s k  ( J k 1 ) 1 r k )( s k )T ( J k 1 ) 1
( J k ) 1  ( J k 1 ) 1 
( s k )T ( J k 1 ) 1 r k
s k  x k  x k 1 , r k  f ( x k )  f ( x k 1 )
0 1
 Initialize with ( J )  I
 Only 1 function evaluation per iteration!

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 16