MULTIPERIOD DESIGN OFAZEOTROPIC SEPARATION

SYSTEMS

Kenneth H. Tyner

and

Arthur W. Westerberg

OVERVIEW

• Problem Description

• Problem Challenges

• Related Research Issues

• Solution Approach

• Conclusions

PROBLEM DESCRIPTION

• Design An Optimal Separation Plant

• Multiple Feeds– Flowrate

– Composition

– Operating Time

• Azeotropes

A

B

CAz

F1

F3F2

PROBLEM DESCRIPTION

A

B

CAz

F1

F3F2

F

A

B

C

Az

PROBLEM DESCRIPTION

A

B

CAz

F1

F3F2

F

A

B

C

PROBLEM DESCRIPTIONFEED 1 FEED 3FEED 2

PROBLEM DESCRIPTIONFEED 1 FEED 3FEED 2

PROBLEM DESCRIPTIONFEED 1 FEED 3FEED 2

PROBLEM DESCRIPTIONFEED 1 FEED 3FEED 2

PROBLEM DESCRIPTIONFEED 1 FEED 3FEED 2

PROBLEM CHALLENGES• Highly Combinatorial

– Separation Pathways– Process Units– Task Assignment

• Difficult Subproblems– Large Models– Highly Nonlinear– Recycle Streams– Shared Equipment

INITIAL RESEARCH THRUSTS

• Synthesize Designs

• Evaluate Designs

• Optimize / Modify Designs

AZEOTROPIC SYNTHESIS

A

B

CAz

F

F

A

B

C

Az

AZEOTROPIC SYNTHESIS

A

B

CAz

F

A

B

C

Az

F

AZEOTROPIC SYNTHESIS

A

B

CAz

F

F

A

B

C

SIMULATION

ZeroSlack

S

S

S

SIMULATION

Solve / Optimize

Initialize

ModifyLibrary

REVISED RESEARCH THRUSTS

• Collocation Error Detection

• Scaling

• Solver Design

SIMULATION

Solve / Optimize

Initialize

ModifyLibrary

SOLUTION APPROACH

• Approximation– Separation Task– Column Design and Operation

• Shortcut Costing

• Autonomous Agents

ECONOMICS

Cost = F( Feed, Distillate, Trays, Reflux )

ECONOMICS

Cost = F( Feed, Distillate, Trays, Reflux )

Separation TaskContribution

ECONOMICS

Cost = F( Feed, Distillate, Trays, Reflux )

Separation TaskContribution

Column Design and OperationContributions

TASK APPROXIMATION

• Variables:– Compositions

– Flowrates

• Relations:– Mass Balance

– Lever Rule

– Geometric Objects

A

B

CAz

F

D / F

D

B

COLUMN APPROXIMATION

• Cost = F(Feed, Distillate, Trays, Reflux)

• Reflux = F(Trays, Feed Location)

COLUMN APPROXIMATION

• Cost = F(Feed, Distillate, Trays, Reflux)

• Reflux = F(Trays)

• Optimal Feed Location = F(Trays)

COLUMN APPROXIMATION

• Reflux = C1 * exp(-C2 * Trays) + C3

• Opt Feed Loc = C4 * Trays + C5

– Numerical Difficulties

• Gilliland Correlation

DATA COLLECTION

• Fix Trays and Task• Find Optimal Reflux

0

2

4

6

8

0 0.2 0.4 0.6 0.8

Feed Location

Ref

lux

DATA COLLECTION

8

10

12

14

16

18

20

22

30 35 40 45 50 55 60 65

Trays

Ref

lux

00.050.10.150.20.250.30.350.40.450.5

Fee

d L

ocat

ion

DATA COLLECTION

A

B

CAz

Store InDatabase

CalculateParameters

SIMULATION

F

A

B

C

Az

A

B

CAz

F

Database

SIMULATION

F

A

B

C

Az

A

B

CAz

F

Database

SIMULATION

ZeroSlack

S

S

S

ASYNCHRONOUS TEAMS

• Independent Software Agents

• Shared Memory

Trial Points

Newton Solver Gradient Solver

ASYNCHRONOUS TEAMS

• Independent Software Agents

• Shared Memory

Trial Points

Newton Solver Gradient Solver

ASYNCHRONOUS TEAMS

• Independent Software Agents

• Shared Memory

Trial Points

Newton Solver Gradient Solver

ASYNCHRONOUS TEAMS

• Independent Software Agents

• Shared Memory

Trial Points

Newton Solver Gradient Solver

ASYNCHRONOUS TEAMS

• Independent Software Agents

• Shared Memory

• Advantages– Scalable– Ease of Creation / Maintenance– Cooperation

ASYNCHRONOUS TEAMS

• Applications– Train Scheduling– Travelling Salesman Problem– Building Design

ASYNCHRONOUS TEAMS

ProblemDescription

ApproximationData

Designs

Database

DesignAgents

ApproximationAgents

MINLP DESIGN AGENT

• Fixed:– Separation Pathways– Intermediate Streams

• Variable:– Task Assignment – Number of Columns– Column Dimensions– Operating Policy

MINLP DESIGN AGENT

• Fixed:– Separation Pathways– Intermediate Streams

• Variable:– Task Assignment– Number of Columns– Column Dimensions– Operating Policy

MINLP DESIGN AGENT

• Fixed:– Separation Pathways– Intermediate Streams

• Variable:– Task Assignment– Number of Columns– Column Dimensions– Operating Policy

TASK ASSIGNMENT

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

20 30 40 50 60

Trays

Dia

mete

r

TASK ASSIGNMENT

$500,000.00

$600,000.00

$700,000.00

$800,000.00

$900,000.00

$1,000,000.00

$1,100,000.00

1 2 3 4 5 6 7

PATH SELECTION

• Sequential Selection

• Genetic Algorithm

• Active Constraint

MINLP DESIGN AGENT

• Fixed:– Separation Pathways– Intermediate Streams

• Variable:– Task Assignment– Number of Columns– Column Dimensions– Operating Policy

ASYNCHRONOUS TEAMS

ProblemDescription

ApproximationData

Designs

Database

DesignAgents

ApproximationAgents

GENERAL BENEFITS

• Alternative to Hierarchical Design

• Persistent Data

• Scenario Analysis

• Human Agents

MULTIPERIOD DESIGN OFAZEOTROPIC SEPARATION

SYSTEMS

Kenneth H. Tyner

and

Arthur W. Westerberg