9/22/2021

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

Solving recycle loops (cont’d)

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

1
 9/22/2021

Line search
 Obtain the optimal step length

1
x ( k 1)  x ( k )    J ( x ( k ) )  f ( x(k ) ) 0  1
 

 Line search by optimization is expensive.

( k 1)
 Instead find α such that f ( x )  f ( x( k ) )

For complex flowsheets: Broyden’s method with line search

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 3

Turton et al. Analysis, Synthesis, and Design of Chemical Processes

Newton method: converged in

fewest iterations, but highest
cost per iteration

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

2
 9/22/2021

Solution times

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

Nested: inner loop is solved and converged

at each iteration of the outer loop.

Simultaneous: all loops are solved
simultaneously
 Advantageous when interactions among loop are
strong

Turton et al. Analysis, Synthesis, and Design of Chemical Processes

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

3
 9/22/2021

Infeasible flowsheet: no steady-state solution exists

 Missing purge stream
 Missing makeup stream
 Wrong stream/unit specifications

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 7

Done as an inner iteration loop

 Bubble point at given T
 Bubble point at given P (iterate on T)
 Dew point at given T
 Dew point at given P
 Flash at given P and T (PT)
 heaters/coolers
 Flash at given P and H (PH)
 Valves, heaters/coolers

Handling of phase changes

 Trial and error (assume phase regime, solve,
repeat if no feasible solution)

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

4
 9/22/2021

Optimization creates an outer iteration loop

Analytical first- and second-order derivatives

not available
 Must be obtained by perturbation

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

Pros
 Easy modeling (problem formulation)
 Customizable solvers for each block (e.g.,
columns)
 Easy troubleshooting
 Established software available

Cons
 Efficiency/convergence issues in case of many
recycles and large interactions
 Inefficient/unreliable for optimization (why?)

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

5
 9/22/2021

Collect all equations and solve simultaneously

 Process model
 Design specs

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

Modify to avoid inverting Jacobian

x k 1  x k   k d k
J k d k  f ( xk )

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

6
 9/22/2021

Idea: break a large problem into

smaller subproblems
 Smaller subproblems are solved more
easily.

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

Construct the incidence matrix

Find permutation matrices P and
Q s.t.
DAEPACK block Solver
Manual (J.E. Tolsma)
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 13

Choose appropriate solver for each block.

 Quasi-Newton method
 Bisection
 Homotopy continuation
 Write your own solver!

DAEPACK block Solver

Manual (J.E. Tolsma)

No derivative calculation needed for off-

diagonal entries  faster solution
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 14

7
 9/22/2021

Ind. Eng. Chem. Res., Vol. 39, No. 6, 2000

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

Solve the following system using BLT.

f1 : x1x2  x3  0
f 2 : x22  10  0
f3 : 2 x1  x3  0

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

8
 9/22/2021

Minimize (norm of) the residual values

minn g ( x)
x

Apply Newton-type methods to

g '( x)  0

Advantages
 State-of-the-art optimization algorithms
 Systematic reduction of residuals to zero
 Ability to handle bounds on variable
 Facilitate convergence
 Handle multiple steady-state solutions
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 17

Initialization
 Good initial guesses being the biggest challenge
 Typically an SM solution required.

Derivatives
 Analytical differentiation is possible!
 But not always
 User-defined subroutines (black boxes)
 unsupported blocks
 Property packages/flash calculations (sometimes available)

Apply perturbation in these cases

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 18

9
 9/22/2021

1. Use Aspen components/property package

2. Build EO case in subflowsheet (HYSYS)
3. Build the SM simulation (good initial guess)
4. Switch to EO ad solve again
5. Perform simulation, optimization, parameter estimation,
etc.

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 19

Pros
 Suitable for highly interactive processes (many recycles
and design blocks)
 Suitable for optimization and model calibration

Cons
 is harder to troubleshoot
 Needs more expertise
 Needs good initial guess (solving very large system)

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 20

10
 9/22/2021

Combine good features of SM and EO

No unique definition (different ways to combine)

Example I
 Apply EO to highly interacting parts
 Apply SM to other parts

Example II
 Apply quasi-Newton for simultaneous solution of
recycles

Example III
 Apply EO to external variables
 Input/output streams, design specs
 Apply SM to internal variables
Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 21

Formulation I
 Tear all streams

2( N C  2) N CS  N d Eqs

Chen & Stadtherr (1985).

AlChE Journal Vol. 31, No.
11.

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 22

11
 9/22/2021

Formulation II
 Eliminate unnecessary eqs

Chen & Stadtherr (1985).

AlChE Journal Vol. 31, No.
11.
( NC  2) NCS  N d Eqs

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 23

Formulation III
 Only tear a subset of streams

Chen & Stadtherr (1985).

AlChE Journal Vol. 31, No.
11.
( NC  2) Nt  N d Eqs

Copyright © Ali M. Sahlodin, Dept. of Chemical Engineering, AmirKabir Univ. of Tech. 24

12