an integrated framework for optimal solvent selection and...
TRANSCRIPT
An Integrated Framework for Optimal Solvent Selection and Separation Process Design Based on
Economic and Controllability Criteria
Panos Seferlis and Athanasios PapadopoulosPanos Seferlis and Athanasios Papadopoulos
Department of Mechanical EngineeringAristotle University of Thessaloniki (AUTh)
Thessaloniki, Greeceand
Chemical Process Engineering Research Institute (CPERI)Centre for Research and Technology – Hellas (CERTH)
Outline
• Introduction - Motivation
• Proposed integrated solvent, process and control design framework
• Solvent selection method
• Optimal process design• Optimal process design
• Controllability analysis using nonlinear sensitivity analysis
• Illustrative example
• Concluding Remarks
Aristotle University of Thessaloniki CPERI
Introduction
• Existing solvent design methods generally enable optimal solvent design in terms of economic performance in the process it is utilized.
Solvent Design
• Solvents are designed using CAMD methods
• Solvents that satisfy certain
Process Design
Optimal Solvent/Process
• Solvents that satisfy certain performance constraints (thermodynamic or process) are introduced in process design
• Iterations between solvent and process design are allowed to refine results
Aristotle University of Thessaloniki CPERI
Introduction - Motivation
• Processes are inherently dynamic environments susceptible to external disturbances
• Processes and the associated control system should be able to compensate for disturbances and maintain the satisfaction of predefined specifications
• Solvent selection and process design considers the static • Solvent selection and process design considers the static controllability properties of the system that has notbeen addressed systematically in available published work
• Ulas and Diwekar (2006) proposed an integrated solvent, process and control design scheme, with the focus maintained on the optimal control profiles of a batch distillation column.
Aristotle University of Thessaloniki CPERI
• “Ability of the control system to perform satisfactorilyin spite of process disturbances that persistently influence the operation of the system ” Postlethwaite, Skogestad
• Controllability is considered a property that is inherent
Controllability Analysis of Processes
• Controllability is considered a property that is inherent of the solvent and process design
• Solvent and process design imposes limits on the dynamic performance of the control system
• Controllability is considered independent of the selection of a control algorithm and its tuning
Aristotle University of Thessaloniki CPERI
Integrated solvent, process and control system design
An integrated framework for the design of solvents based on their economic and controllability process performance
Solvent screening for optimum economic process performance
Multi-objective solvent synthesis
Solvent screening based on static controllability
Chart of economic solvent performance
Representative discrete solvent options
Clustering of optimal solvent set
Conceptual process design
Rigorous process design
Non-linear sensitivity analysis
Chart of economic and controllability
solvent performance
Aristotle University of Thessaloniki CPERI
Advantages of the proposed method
• At the economic solvent screening stage:
– No potentially useful solvent options are excluded from the solution space
– Solvent-process design decisions are based on economic – Solvent-process design decisions are based on economic process performance criteria
– Major solvent-process design drives are rapidly captured using conceptual process models
Aristotle University of Thessaloniki CPERI
Advantages of the proposed method
• At the solvent screening stage based on static process controllability:
– The entire economic solvent performance chart can be screened using rigorous process models
– The rigorous models rapidly provide realistic process behavior informationbehavior information
– Nonlinear sensitivity analysis provides accurate information about the solvent-process behavior under multiple parameter variations for specified control objectives
Aristotle University of Thessaloniki CPERI
1st stage - Solvent and process design
The first stage is based on the approach by Papadopoulos and Linke (2005,2006)
CAMD Solvent Design
Optimal Process and Solvent
System
Set of Optimal Solvents
Process Design
Multi-objective optimization
Solvent Clusters
Solvent and ProcessDesign Interface
Aristotle University of Thessaloniki CPERI
Multi-objective solvent design
)(),.....(),( 21 xfxfxf p
SatP∞γ
Set of Optimal Solvents
Optimize
∞γ
Process related objective functions
f1•••
••f2
fp
Multi-objective optimization
SatA
SatB
SA
SB
P
Pf ⋅= ∞
∞
,
,1 γ
γ
B
A
SA
SB
MW
MWf ⋅= ∞
∞
,
,2 γ
γ
•
•••
•••f1
f2
Pareto optimalset
sMWf =2
Solvent related objective functions
∞= SBf ,1 γ
fp
Pareto optimal front
Aristotle University of Thessaloniki CPERI
Clustering of optimal solvent set
Solvent Clusters
•••• ••
•
Pro
pert
y 1• • •
••• •
• •
Step 1: Partition optimal solvent set into groups of molecules with similar properties
Process Synthesis:•Models•VLE/LLE•Objective Function
Pro
pert
y 1• •• •
••
••• •
Property 2Step 2: Introduce a representative
molecule from each cluster into
process synthesis
Aristotle University of Thessaloniki CPERI
Economic solvent performance decision chart
New cluster1Iteration 1 Iteration 2 Iteration N. . .
• Subsequent clustering iterations result in a decision chart that combines: – Statistical solvent property information– Economic process performance of representative molecule from
each cluster– Decisions on the emerging clustering paths are based on a
probability function that combines economic and statistical information
New cluster1
New cluster1
Original set
Cluster 1
Cluster 2
Cluster 3
Cluster N
New cluster1Optimum cluster
.
.
.New cluster2
New cluster N
New cluster2
New cluster N
New cluster2
New cluster N
Aristotle University of Thessaloniki CPERI
Clustering probability function
)T/E∆exp(P new−=
oldnew TaT ⋅=
SSWSSB
SSB1a
+−=
• Unifies statistical and economical clustering information under one index
• Uses the Simulated Annealing probability function to quantify the SSWSSB
1a+
−= function to quantify the uncertainties involved in clustering decisions
• Temperature parameter is reduced based on the information included in the clusters
• ∆E: Difference between tested cluster center OF and best cluster center OF
• SSB: Distance between cluster centres
• SSW: Distance within cluster points
Aristotle University of Thessaloniki CPERI
2nd stage – Rigorous process design
• Stage 1 models are short cut distillation models
• Stage 2 models are accurate but compact in size
• Orthogonal collocation on finite element models are
Top product
Side productSide feed
Heat stream
Side heat
Collocation points
finite element models are employed
• Columns section sizes are continuous variables
• Polynomial approximation of molar and enthalpy flows within column sections is implemented
Bottoms product
Side productSide feed
Heat duty
Elementbreakpoints
Aristotle University of Thessaloniki CPERI
2nd stage - Process and controllability properties for representative solvents
• The second stage is based on the integrated process and control system design method by Seferlis and Grievink (2001, 2004)
MV, u CV, y
disturbances, εεεε
y2u2
For each candidateset of MV
For each candidateset of CV
y1
Process
u1
For a given disturbance scenario calculate:
• the steady-state variation for the MVs to maintain the CVs within predefined levels
Aristotle University of Thessaloniki CPERI
• Large variation in the steady-state position of the MVs required for the compensation of relatively small in magnitude disturbances implies lack of “power” for the system
Controllability evaluation
Solvents, process design and control structure performance is evaluated and rank ordered
Aristotle University of Thessaloniki CPERI
Optimization framework with objective function that:• penalizes deviations from setpoint for CV
• penalizes usage of available MV
( ) ( ) ( ) ( )( )( )
ssuT
ssspyT
spuy
0εd,u,y,x,h
uuWuuyyWyy
=
−−+−−=
s.t.
fMin,
Disturbance sensitivity control problem
( )( )
ulululopt xxx,uuu,yyydd
0εd,u,y,x,g0εd,u,y,x,h
≤≤≤≤≤≤=≤=
,
s.t.
Weights in the objective function determine:• relative importance of control objectives
• cost related preference of system resources for corrective actions
Aristotle University of Thessaloniki CPERI
Solution method
• Solve the parameterized Karush-Kuhn-Tucker (KKT) optimality conditions
• gA: active inequalities
• ∆ε∆ε∆ε∆ε: disturbance relative variation0
gh
f
=
∇+∇+∇
=
gµhλ
FA
TT
• θθθθ: direction of perturbation
• ζ : free continuation parameter
0ghA
=
−
=
ζθ∆ε
F
( ) ( ) ( ) ( )( ) ( ) ( ) ( )
( ) ( ) ( ) ( )( )
p
i*i,p
*i,p
*i,p
ii
p
*i
*i
*i
ii
p
*i
*i
*i
iSC 0f
0fζfζw
0y
0yζyζw
0u
0uζuζwζΩ ∑∑∑
−+
−+−=
Controllability index
Aristotle University of Thessaloniki CPERI
Derive sensitivity information of process variables w.r.t. disturbances
Disturbance sensitivity control problem
( )[ ]Trefopt
εε εXP ∇=
• Disturbance sensitivity decomposition (SVD) for the determination of key directions, θ, of variability
• Equivalent to the worst case scenario for disturbance influence
Aristotle University of Thessaloniki CPERI
• Integrated solvent/process/control design for the separation of a Cyclohexene-Benzene mixture
Illustrative example
• Design a solvent/ extractive distillation system for the separation of a Cyclohexene-Benzene mixture
• Multiple objectivesMaximizing: Minimizing:
• Molecules that create azeotropic mixtures are not allowed
Maximizing: Minimizing:Relative volatility (aB,A)
Solvent power (Sp)
Heat of vaporization (Hv)
Molecular weight (MWs)
Process Cost
Aristotle University of Thessaloniki CPERI
• Identified in total 109 solvent molecules
1st Stage - Economic solvent screening
Cl Nm COF(k$/yr) P1 1 1810 0.002 14 1337.1 0.993 23 1404.2 0.974 34 1689.1 0.92
Cl Nm COF(k$/yr) P1 8 1359.8 0.942 6 1316 0.993 6 1315.8 1.004 1 1580 0.00
(2,3,5)
Iteration 1 Iteration 2
4 34 1689.1 0.925 6 1315.8 1.006 31 1742.4 0.88
4 1 1580 0.005 1 1537 0.006 21 1377 0.96
• Optimum solvents are included in clusters 2 and 3 of iteration 2 based on the high probability values
• We only screened 12 molecules in terms of process performance out of the 109 molecules
Aristotle University of Thessaloniki CPERI
2nd Stage - Rigorous process design
StgSfl
(kmole/hr) RQD
(MJ/hr)QB
(MJ/hr) H(m) D(m) Xd Rv(%)74 78 8.4 1430 1946 47 1.8 0.77 75.68 50.4 0.2 24.2 1752 7.1 0.45 0.96 97.9
24 54.2 0.3 43.6 2078 16.7 0.77 0.8 99.927 70.9 0.3 48.2 2439 19 0.77 0.72 1008 49.11 0.28 34.9 1422 7.1 0.48 0.96 99.4
17 67 0.5 65.4 3147 13.2 0.53 0.91 99.9
• Detailed design information is obtained through rigorous OCFE process models
• Such information can not be obtained by use of conceptual models
Aristotle University of Thessaloniki CPERI
Recalculation of probability
Economic solvent screening
Rigorous economic solvent screening
Cl Nm P (COF)
1 1 0.002 14 0.993 23 0.974 34 0.925 6 1.006 31 0.88
P(ROF) StgSfl
(kmole/hr)R
QD (MJ/hr)
QB (MJ/hr)
H(m) D(m) Xd Rv(%)
0.00 74 78 8.4 1430 1946 47 1.8 0.77 75.60.95 8 50.4 0.2 24.2 1752 7.1 0.45 0.96 97.90.94 24 54.2 0.3 43.6 2078 16.7 0.77 0.8 99.90.89 27 70.9 0.3 48.2 2439 19 0.77 0.72 1001.00 8 49.11 0.28 34.9 1422 7.1 0.48 0.96 99.40.85 17 67 0.5 65.4 3147 13.2 0.53 0.91 99.9
• The probability function calculation utilizing rigorous models provides similar solvent ranking as in 1st stage
• However, the ranking at this stage involves higher certainty regarding the solvent-process design decisions
Aristotle University of Thessaloniki CPERI
Recalculation of probability
Economic solvent screening
Rigorous economic solvent screening
Cl Nm P (COF)
1 1 0.002 14 0.993 23 0.974 34 0.925 6 1.006 31 0.88
P(ROF) StgSfl
(kmole/hr)R
QD (MJ/hr)
QB (MJ/hr)
H(m) D(m) Xd Rv(%)
0.00 74 78 8.4 1430 1946 47 1.8 0.77 75.60.95 8 50.4 0.2 24.2 1752 7.1 0.45 0.96 97.90.94 24 54.2 0.3 43.6 2078 16.7 0.77 0.8 99.90.89 27 70.9 0.3 48.2 2439 19 0.77 0.72 1001.00 8 49.11 0.28 34.9 1422 7.1 0.48 0.96 99.40.85 17 67 0.5 65.4 3147 13.2 0.53 0.91 99.9
(2,3,5)
6 31 0.88
Cl Nm P(COF)
1 8 0.942 6 0.993 6 1.004 1 0.005 1 0.006 21 0.96
P(ROF) StgSfl
(kmole/hr)R
QD (MJ/hr)
QB (MJ/hr)
H(m) D(m) Xd Rv(%)
0.51 20 67.6 0.2 22.1 2457.9 14.9 0.59 0.96 95.20.90 8 41.7 0.2 23.7 1892.9 7.1 0.47 0.96 95.61.00 8 39.13 0.283 34.9 1422.7 7.1 0.48 0.96 99.40.00 220 34.7 12.2 2081 2685 139.9 5.5 0.77 99.70.00 31 97.4 0.54 69.9 2099 21.3 0.85 0.92 99.90.70 21 74.4 0.2 20.2 2954 15.1 0.6 0.96 94.1
Aristotle University of Thessaloniki CPERI
Optimal solvent ranking
ID Molecule Conceptual model OF Rigorous model OF
S1
HC=O|
FCH2-O-C-(C=O)H|
HC=O
1 4
S2
HC=O|
H(C=O)-C-O-CH3|
HC=O
2 1
HC=O
S3
HC=O|
Cl-C-O-CH2 -(O=C)H|
HC=O
3 2
S4
HC=O|
H(C=O)-CH-O-CH-Cl|
HC=O
4 3
S5
HC=O|
H(C=O)-CH-CH-Cl|
HC=O
5 5
Aristotle University of Thessaloniki CPERI
Controllability analysis
Controlled variables:
top product cyclohexane purity level 0.96 (0.94)
cyclohexane % recovery
Manipulated variables:
reboiler duty (bound at +40% of base design point)
reflux flow (bound at +40%)
Aristotle University of Thessaloniki CPERI
reflux flow (bound at +40%)
solvent feed flow (bound at +40%)
w=[1000, 10, 10, 10, 100]
disturbances: feed composition, temperature
Nonlinear sensitivity analysis - I
5
10
15Feed temperature
% c
hang
e
-30
-20
-10
0Cyclohexane at feed
40
60
80
100Benzene at feed
0 5 100
5
zeta0 5 10
-50
-40
zeta0 5 10
0
20
zeta
Aristotle University of Thessaloniki CPERI
Disturbance characterization along the major direction of variability of the process system
0 2 4 6 8 10 12-50
-40
-30
-20
-10
0Condenser duty
% c
hang
e
0 2 4 6 8 10 12-10
0
10
20
30Reboiler duty
% c
hang
e
Reflux ratio Solvent flowrate
purity bound active
solvent flow bound active
Nonlinear sensitivity analysis - II
0 2 4 6 8 10 120
10
20
30
40Reflux ratio
zeta
% c
hang
e
0 2 4 6 8 10 120
10
20
30
40Solvent flowrate
zeta%
cha
nge
Aristotle University of Thessaloniki CPERI
• Variation of key process (manipulated) variables for disturbances along major directions of variability
0.95
0.955
0.96
0.965Distillate xd
93
94
95
96
97CY Recovery
Nonlinear sensitivity analysis - III
0 5 100.935
0.94
0.945
zeta0 5 10
90
91
92
93
zeta
Aristotle University of Thessaloniki CPERI
• Variation of key controlled variables for disturbances along major directions of variability
Controllability index
15
20
25
Co
ntr
olla
bilit
y in
dex
, om
ega
S1S2S3S4S5
0 2 4 6 8 100
5
10
disturbance magnitude parameter, zeta
Co
ntr
olla
bilit
y in
dex
, om
ega
Aristotle University of Thessaloniki CPERI
8
10
12
14
Con
trol
labi
lity
inde
xS2
S1
Economic - controllability performance chart
34 36 38 40 42 44 460
2
4
6
Economic objective function value, k$/yr
Con
trol
labi
lity
inde
x
S4
S3S5
Concluding remarks
• The effects of solvent selection on the optimal process design and static controllability properties have been investigated
• Solvents are evaluated and rank ordered based on economic and static controllability performance criteria
• Static controllability is performed using nonlinear sensitivity analysis and rigorous process models
• Static controllability is performed using nonlinear sensitivity analysis and rigorous process models
Aristotle University of Thessaloniki CPERI