The solvent selection expert system for azeotropic and extractive distillation

by Zuyin Yang
A thesis submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in
Chemical Engineering
Montana State University
Copyright by Zuyin Yang (1994)
Abstract:
Azeotropes, or close-boiling mixtures often preclude conventional distillation as a method of
separation. Instead, extractive or azeotropic distillations are commonly used to separate close or
azeotropic boiling mixtures. For the design of these separation units, selecting suitable solvents is an
indispensable step. The traditional method for solving this problem is through experimentation which is
time-consuming and expensive.
A new approach is offered by combining heuristic knowledge and numerical methods for solvent
screening in the form of a knowledge-based expert system. The prototype expert system takes as inputs
a small amount of physico-chemical property data for pure components and mixtures, and the processes
which are suitable for the separation of a close boiling or an azeotropic mixture are selected. Then the
expert system proposes solvents from the solvent databases based on the requirements of the process
under consideration. The heuristic knowledge and numerical knowledge from different sources for
selecting solvents for mass-based separation are organized through ac task-oriented framework as the
representing strategy. The structured heuristic search hierarchy has been developed as the
problem-solving strategy that an expert follows.
At present, the prototype expert system is limited to the preliminary selection of promising solvents for
azeotropic and extractive distillation. The prototype expert system is applied to screen numerous
solvents for the separation of xylene isomers and ethanol-water mixture. The result for the separation
of xylene isomers agrees with experimental results to date. For the ethanol-water mixture to be
separated by extractive distillation, the expert system confirms some solvents suggested in the literature
and proposes more solvents for further experimental verification. For the methyl acetate-methanol
mixture, the expert system can propose promising solvents based on Berg's solvent classification and
Sheibel's rule, but cannot accurately rate the solvents.

THE SOLVENT SELECTION EXPERT SYSTEM FOR AZEOTROPIC

AND EXTRACTIVE DISTILLATION

by
Zuyih Yang

A thesis submitted in partial fulfillment

of the requirements for the degree

Doctor of Philosophy
in
Chemical Engineering

MONTANA STATE UNIVERSITY

Bozeman, Montana
M a y , 1994

J)2n%
ii

APPROVAL

of a thesis submitted by

Zuyin Yang

This thesis has been read by each member of the thesis

committee and has been found to be satisfactory regarding
content, English usage, format, citations, bibliographic
style, and consistency, and is ready for submission to the
College of Graduate Studies.

// /ff/

Date
Approved for the Major Department

// /

f?c/
Major Department

Approved for the College of Graduate Studies

Date Y

Graduate Dean

ill

STATEMENT OF PERMISSION TO USE

In presenting this thesis in partial fulfillment of the

requirements
University,

for

doctoral

degree

at

Montana

State

I agree that the Library shall make it available

to borrowers under rules of the Library. I further agree that

copying

of

this

thesis

is

allowable

only

for

scholarly

purposes, consistent with "fair use" as prescribed in the U.S.

Copyright Law. Requests for extensive copying or reproduction
of this thesis

should be referred to University Microfilms

International, 300 North Zeeb Road, Ann Arbor, Michigan 48106,

to whom I have granted "the exclusive right to reproduce and
distribute my dissertation for sale in and from microfilm or
electronic

format,

along

with 'the

right

to reproduce

and

distribute my abstract in any format in whole or in part."

Signature _
Date

iv

ACKNOWLEDGEMENTS

The author wishes to acknowledge the partial financial

support

for

the

research

project

by

Glitsch

Technology

Corporation.
The author is especially indebted to Dr. Lloyd Berg and
Dr. Ronald W. Larsen, the advisors for this project, for their
guidance.

Dr.

Warren

Scarrah and

all

the thesis

committee

members are acknowledged for their help. Thanks are extended

to

the

faculty

and

staff ' of

the

Chemical

Engineering

Department at Montana State University for their encouragement

and

help.

Dr.

John

Sears,

Department

Head

of

Chemical

Engineering, is acknowledged for his support of this project.

Ms.

Randy Wytcherly

and Mr.

Joseph Gentry

of Glitsch

Technology Corporation are acknowledged for their suggestions

and useful discussion for this project.
Finally,

the

Administrative
Engineering,

author wishes

Assistant

of

to

thank Ms.

Department

Jane Curtis,
of

Chemical

for her moral support during my stay at Montana

State University.

TABLE OF CONTENTS
Page
APPROVAL..................... .... ............. ..... .......
STATEMENT OF PERMISSION TO USE. . ................. .........

ii
iii

ACKNOWLEDGEMENTS.... .............. .......... ....... .....

iv

TABLE OF CONTENTS.... .......... ... ... ..... ..... . .......

LIST OF TABLES.......... .... ......... ...... ..............viii

LIST OF FIGURES................................... '........

ABSTRACT. . .'..... ...................... ; .................. xiii

1. INTRODUCTION. ...... . ,.............. .................

Techniques for Difficult-to-Separate Mixtures ....

Significance of Selection of Solvents.............
Expert System Approach.........................
Problem Specification........
The Objectives and the Scope of Research...........

I
3
5
5
. 6

2. HISTORICAL BACKGROUND......
Terminologies for Azeotropic and
Extractive Distillation...... ;.................

8
8

Azeotropes and Close-Boiling Mixtures........

Azeotropic and Extractive Distillation
Processes...........................

18

The Brief History of Azeotropic and

Extractive Distillation..........................
An Overview of. Solvent Selection Methods...........

27
28

For Azeotropic Distillation.... ............. .

, 29
For Extractive Distillation. ....................
36
The Selection of Azeotropic or Extractive
Distillation.... ............

42

vi

TABLE OF CONTENTS Continued

Page
3. SOLVENT SELECTION CRITERIA............................

47

Separation Enhancement Characterization............

47

Separation Factor............ . *.... . .........

Relative Volatility................
Selectivity....................
Activity Coefficients at Infinite Dilution....
Thermodynamic Models for the
Estimation of ACID. . . .........................

47
49
52
56

Easy Reversibility. . .................

59.
66

4. EXPERT SYSTEMS AND SELECTION PROBLEMS................

67

Introduction to Expert Systems........

Expert System Shell LEVEL 5 OBJECT......
Expert System Applications
in Chemical Engineering. ..................
The General Theory of Selection Problems......
Selection-related Expert Systems
in Chemical Engineering...........................

67
70

75

5. KNOWLEDGE BASE............. ......... . . ................

79

Knowledge Acquisition................................

79

Knowledge Involved in the Solvent

Selection Problem............................
Knowledge Used in the Current Expert System...

79
82

71
73

Types of Knowledge.............................

86

Knowledge Representation. Strategy..................

89

Implementation Language
Object-oriented Programming. ........
Features of an Object..........................
Procedural Knowledge
Rules and Methods Representation............
Search Space Solvent Database
^
and Related Objects..........................
S t r u c t u r i n g K n o w l e d g e .........................................................

90
92
96
98

106

vii

TABLE OF CONTENTS Continued

Page
6. EXPERT SYSTEM IMPLEMENTATION.........
Selection Hierarchy...............
Hydrocarbon Branch..........................
.Aqueous System Branch................. ........
Nonaqueous Nonhydrocarbon Branch..............
Azeotropic Branch. . .............................
Application Examples...............
Example I. Selection of Solvents for
Separation of Xylene Isomers.............
Example 2. Selection of Solvents for
Separation of Water from Ethanol.........
Example 3. Selection of Solvents
for Separation of Methyl acetate
from Methanol.......... . . .................

109
109
114
118
12 0
122
124
124
131
13 6

7. SUMMARY.............. .................'....... .......... - 140

i
Conclusions..................... .................... 14 0
Recommendations for Further Studies................
141
NOMENCLATURE...... ....................................... . 143
ABBREVIATIONS............

146

LITERATURE CITED..........................................

148

APPENDICES......;......................

158

APPENDIX A. INDUSTRIAL SOLVENTS

FOR THE EXPERT SYSTEM DATABASE................
APPENDIX B. SOLVENTS RECOMMENDED
BY DR. LLOYD BERG.... ....................
APPENDIX C. QBASIC PROGRAM FOR CALCULATING ACID
FOR THE EXPERT SYSTEM. . ........................
APPENDIX D. RULES AND. THERMODYNAMIC MODELS
FOR THE EXPERT SYSTEM........... .....'......... .

159
164
166
171

viii

LIST OF TABLES
Table

Page

1. Solvent Selection Based on Scheibel's Rules..........

37

2. Heuristics and Rules for Selection of

Separation Methods.....................................

44

3. Experimental Methods for Determining A C I D ............

59

4. Binary Activity Coefficient Correlations

at Infinite Dilution............................... . ..

60

5. Models for Estimation of ActivityCoefficients at Infinite Dilution

Using Solubility Parameters...........................

62

6.. Deviation Between Experimental and

Predicted Activity Coefficients
at Infinite Dilution...................................

64

7. Predicted Infinitely Dilute Activity

Coefficients by UNIFAC for Aqueous Systems...........

65

8. The Basic Features of LEVEL 5 OBJECT.................

72

9. Application Areas of ES and Their Functions,........

73

10. Progress of the Knowledge for Solvent Selection

for Azeotropic and Extractive Distillation...........

83

11. Rules Used in the Current Expert System.............

85

12. The Physico-Chemical Properties

of Solvent Methanol.................. ................

103

13. Critical Temperatures Estimated

by Group Methods......................................

103

14. Solvent Classes Based on

Chemical Function Groups ..............................

104

15. Correlations for Calculating Induction Parameter....

116

ix
LIST OF TABLES Continued
Table

Page

16. Correlations for Calculating the Nonpolar

Solubility Parameter........... ........ ..............

117

17. Physical Properties of Xylene Isomers. ...... ........

125

18. The Selected Solvents by SSS

with Their Selectivities. . . . ...................... .

131

19. Properties of Components and Azeotrope

at I Atm for Ethanol and Water. .......... .......... .

133

20. Activity Coefficients at Infinite Dilution

for Water-Ethanol with Different Solvents............
21. Solvents for Water-Ethanol Based upon Selectivity...

134
135

22. Solvents for Water-Ethanol Based upon

Relative Volatility...............................
23. Properties of Components and Azeotrope
at I Atm for Methanol and Methyl Acetate.... .......
24. Partial Solvent List Selected for Separating
Methanol from Methyl Acetate. ..........................

136
136
137

LIST OF FIGURES
Figure

Page

1. Process Development Steps for the Mass-based

Separation and the Scope of This Research............

2. Liquid Mole Fractions and Vapor Mole

Fractions for Minimum-Boiling Mixtures
of Acetone and Methanol...............................

10

3. Liquid Boiling Points and Vapor Condensation

Temperatures for Minimum-Boiling Azeotropic
Mixture of Acetone and Methanol......................

11

4. Pressure and Liquid Compositions

for Minimum-Boiling Azeotropic Mixture
of Acetone and Methanol. ...............................

12

5. Liquid Mole Fractions and Vapor Mole

Fractions for Maximum-Boiling Azeotropic
Mixture of Acetone and Chloroform. . ...................

13

6. Liquid Boiling Point and Vapor Condensation

Temperatures for Maximum-Boiling Azeotropic
Mixture of Acetone and Chloroform....................

14

7. Pressure and Liquid Mole Fractions for

Maximum-Boiling Azeotropic Mixture of
Acetone and Chloroform.......

15

8. Liquid Mole Fractions and Vapor Mole

Fractions for Heteroazeotropic Mixture
of Water and n-Butanol................................

16

9. Liquid Boiling Point and Vapor Condensation

Temperatures for Minimum-Boiling Heteroazeotropic
Mixture of Water and n-Butanol........................

17

10. Liquid Mole Fractions and Vapor Mole

Fractions for Close-Boiling Mixture
of Propylene and Propane....... .....................

19

11. Liquid Mole Fractions and Vapor Mole

Fractions for Close-Boiling Mixture of
n-Hexane and Benzene

20

xi

LIST OF FIGURES Continued

Page
12. Liquid Mole Fractions for Low Relative
Volatility Mixture of Water and Acetic A c i d .........

21

13. Simplified Flow Diagram for an Extractive

Distillation Process to Separate Toluene
and Methyl Cyclohexane...............................

23

14. Azeotropic Distillation Process for

Separating Ethanol and Water
by Using Pentane as the Entrainer...................

25

15. A Typical Residue Curve Map Showing.

a Distillation Boundary..............................

32

16. The Algorithm for Determining

the Residue Curve Map and
the Selection of Azeotropic Entrainers..............

34

17. Procedure for the Selection of Extractive

Solvents Based on UNIFAC and U N IQUAC Models........

41

18. Relative Volatility Required

for Different Separations with Equal Cost ...........

46

19. Vapor-Liquid Equilibrium Curves

for the Methanol Acetone System
with Solvents and with No Solvent...................

48

20. NTP Requirements for Different

Relative Volatilities................................

51

21. NTP and Column Cost vs. Relative Volatility........

53

22. Effect of Ratio of Solvent

to Mixture on Selectivity............................

54

23. Effect of Temperature on Selectivity................

55

24. Effect of Solvent Selectivity

on Cost for Low Selectivity..........................

57

25. Effect of Solvent Selectivity

on Cost for High Selectivity.........................

58

26. Integration of Expert System Components

68

xii
LIST OF FIGURES Continued
Page
27. Knowledge Involved for Solvent
Selection Problems. . . ............................ .
28. Knowledge Components and Classification...... .

81
^

88

29. Illustration of Object Concept. ......................

94

30. Object "Domain" for the Driving Module..............

95

31. An Example of Creating a R u l e....

99

. .........

32. An Illustration of Creating a Method.................

IdO

33. An Example of Inheritance............................

107

34. Selection Hierarchy for Solvent Selection...........

ill

35. The Flowsheet of the Expert System........

113

'

36. Selection Procedure and Results for p-Xylene

and m-Xylene..... .....................................

129

37. Selection Procedure and Results for o-Xylene

and m-Xylene........... ........ .......................

132

38. Selection Procedure and Results for

Methyl Acetate and Methanol........... ...............

138

xiii

ABSTRACT

. Azeotropes, or close-boiling mixtures often preclude

conventional distillation as a method of separation. Instead,
extractive or azeotropic distillations are commonly used to
separate close.or azeotropic boiling mixtures. For the design
of these separation units, selecting suitable solvents is an
indispensable step. The traditional method for solving this
problem is through experimentation which is time-consuming and
expensive.
A new approach is offered by combining heuristic
knowledge and numerical methods for solvent screening in the
form of a knowledge-based expert system. The prototype expert
system takes as inputs a small amount of physico-chemical
property data for pure components and mixtures, and the
processes which are suitable for the separation of a close
boiling or an azeotropic mixture are selected. Then the expert
system proposes solvents from the solvent databases based on
the requirements of the process under consideration. The
heuristic knowledge and numerical knowledge from different
sources for selecting solvents for mass-based separation are
organized
through
a c task-oriented
framework
as
the
representing
strategy.
The
structured heuristic
search
hierarchy has been developed as the problem-solving strategy
that an expert follows.
At present, the prototype expert system is limited to the
preliminary selection of promising solvents for azeotropic and
extractive distillation. The prototype expert system is
applied to screen numerous solvents for the separation of
xylene isomers and ethanol-water mixture. The result for the
separation of xylene isomers agrees with experimental results
to date. For the ethanol -water mixture to be separated by
extractive distillation, the expert system confirms some
solvents suggested in the literature and proposes more
solvents for further experimental verification. For the methyl
acetate-methanol mixture, the expert system can propose
promising solvents based on Berg's solvent classification and
Sheibel1s rule, but cannot accurately rate the solvents.

CHAPTER I

INTRODUCTION

Techniques for Difficult-to-Separate Mixtures

Separating azeotropic or close-boiling mixtures into pure

components

is a task commonly encountered

industry. Sometimes,

in the

chemical

an azeotrope can be negated through an

ordinary distillation by swinging pressure. However, that is

not always the case.

Generally,

pressure-swing distillation

cannot be applied to separating close-boiling mixtures because

the pressure change will result in the same change in boiling
points for each of the components in the mixture. If pressure
swing

distillation

cannot

apply

because

the

azeotrope

composition does not vary much with pressure or because the

required

pressure

leads

to

product

degradation,

there

are

still many separation techniques available and some of them

are being used widely in chemical industry to accomplish the
tasks

mentioned

above.

The

most

following:
azeotropic distillation
extractive distillation

, solvent extraction

commonly

used

are

the

These

reactive distillation

"salt" distillation

techniques

component.

The

have

in

common

application

of

the

addition

azeotropic

of

and

third

extractive

distillation depends on how much the relative volatility of

the two constituents in the mixture is altered by adding xa
third component.
more

components

In solvent extraction,
is

treated

by

adding

a mixture of two or
a

component

that

preferentially dissolves one or more of the components in the

mixture.

Reactive distillation

involves the reversible and

preferential reaction of the added component with one of the

components in the mixture. In "salt" distillation, the added
component dissociates ionically in the solution and changes
the

azeotrope

composition.

Among

them,

azeotropic

distillation, extractive distillation and solvent extraction

have drawn the greatest attention in industry for separating
azeotropic or close-boiling mixtures.
Sucksmith[I ] has pointed out that extractive distillation
is worth considering even when the conventional distillation
is

feasible

conventional
mixtures

for

some

close-boiling

distillation

usually

requires

to
many

fixed cost for the process,

separate
stages,

mixtures.
some

Using

close-boiling

which means

a high

and a high reflux ratio, which

results in a high operating cost for the process. Extractive

distillation

often

uses

less

energy

than

conventional

distillation does for many close-boiling mixture separations.

3
Significance of Selection of Solvents

The

development

of

separation

process

for

the

difficult-to-separate mixture consists of several steps[2] as

shown in Figure I. After establishing the need for one of the
aforementioned
industry,

techniques

screening

the

becomes

an

entrainers)

for

tough

promising

separation

solvents

inevitable

task

(also

step

in

in

called

the

whole

development process.
The

initial

consists

of

phase

three

of

designing

principal

steps:

separation

First,

process

separation

technique is determined based on a given separation problem;

second,

potential

solvents

are screened,

then a separation

sequence is synthesized for each selected solvent. Once the

separation

technique

is

chosen,

the

second

step

becomes

critical for the process design, as pointed out by Laroche et

a l .[3 ], because an economically optimal design made with an
average solvent can be much more costly than an average design
using the best solvent.
The

selection of potential

extractive

distillation

and

solvents

solvent

for azeotropic or

extraction

largely empirical and experimental, and thus

consuming and

expensive.

Over the years,

has

been

is very time-

the

criteria,

or

rules for selection of promising solvents or entrainers have

been

established,

and

the

knowledge

about

predicting

behavior of liquid solutions has been accumulated in the

the

PILOT
RUN?

Scope of the research

DETERMINING
SEPARATION
METHOD

DETAILED
ESTIMATE

UCENSE

AGENT
SCREENING

EFFECTIVE
AGENT?

BASIC
ENGINEERING

NO / ADDITIONAL \N O
VSCREENING? /

STOP

DETAILED
ENGINEERING

^d-STOP
CONSTRUCTION

ECONOMIC
EVALUATION

PRODUCTION

Figure I Process development steps for the mass-based separation and the
scope of t h i s .research [2]

5
literature. But the knowledge is represented as both empirical
rules and procedural algorithms, it is not complete, and there
are

sometimes

contradictions

within

this

problem

domain

knowledge.

Expert System Approach

The

1980's and the beginning of

1990's

witnessed the.

emergence of artificial intelligence.(Al), which has led to

knowledge-based system (also called expert system) application
to many disciplines. The main advantage of a knowledge-based
system approach

lies

in the

fact

that

it can make

use of

qualitative models based on the available knowledge about the

system and heuristics proposed by the experts who work in the
field. As pointed out above, the knowledge for the selection
of

promising

entrainers

for

azeotropic

and

extractive

distillation is both qualitative and quantitative. This work

took advantage of a knowledge-based system approach to combine
these two kinds of knowledge to establish an expert system
capable of identifying potential solvents.

Problem Specification

The development of an expert system for the selection of

proper solvents for azeotropic and extractive distillation is
an

interdisciplinary

endeavor,

combining

aspects

of

both

6
chemical engineering and artificial intelligence (Al). These
two

diverse

fields

contribute

very

different,

but

deeply

interrelated, perspectives to the problem of the selection of

solvents

for

distillation.

azeotropic

distillation

and

extractive

In a broad sense, the problem of the selection

of solvents for azeotropic and extractive distillation can be

defined generally as follows:
Given (I) an azeotrope, a close-boiling mixture, or other
difficult-to-separate. mixture
aqueous solution,

such

as

dilute

acetic

acid

(2) physical property data of the mixture,

and (3) a portfolio of solvents with their physical property

data; first,

find an appropriate

separation method between

azeotropic distillation and extractive distillation; second,

choose

solvents

markedly
mixture;

for

enhancing
third,

effectively

negating the

t h e ' separation

predict

their

of

the

effectiveness

azeotrope

or

close-boiling
or

rate

their

separation-enhancing ability.

The Objectives and the Scope of Research

We

are

concerned

here

with

speeding

the

preliminary

solvent screening process' for the separation of azeotropes or

close-boiling mixtures that are often encountered in chemical
industry using knowledge available from various sources. We
developed and illustrated a prototype expert system that deals
with

selecting

a proper

separation

technique

based on the

7
characteristics of the mixture to be separated. We then focus
on

the

solvent

selection

process

for

the

separation

of

binary azeotrope or binary close-boiling mixture by azeotropic

or extractive distillation.

CHAPTER 2

HISTORICAL BACKGROUND

Terminologies for Azeotropic and Extractive Distillation

In this section, some terminologies and concepts related

to the selection of solvents
distillation

are

for azeotropic and extractive

introduced

as

the

basis

for

further

discussion.

Azeotrope and Close-boiling Mixtures

Dif ficult-to-separate
classified

into

two

mixtures

types:

can

azeotropes

be

and

generally-

close-boiling

mixtures.
Azeotrope, from the Greek "to boil unchanged," literally
means

that

composition

the
as

vapor
the

boiling

liquid.

An

from

liquid

azeotrope

has

forms

the

same

because the

mixture comprises structurally dissimilar compounds and its

vapor-liquid equilibrium behavior deviates considerably from
Raoult's law. Azeotropic systems may be classified broadly in
relation to the characteristics of the azeotrope (maximum- or
minimum-boiling), number of components in the system (binary,

9
ternary, or quaternary), and whether one or more liquid phases
are

formed

(homogeneous

or

heterogeneous).

In

line

with

deviations from Raoult's law, azeotropes can be divided into

positive

azeotropes

and

negative

azeotropes.

Positive

azeotropes are characterized by a minimum-boiling temperature

at

constant

pressure,

i .e .

a maximum

in the

total

vapor

pressure at constant temperature. Acetone and methanol form a

typical

positive

azeotrope

system

and

its

vapor-liquid

equilibrium (X-Y curve), temperature-composition at constant

pressure plot (T-X or T-Y curve), and pressure-composition at
constant temperature plot

(P-X or P-Y curve)

are

shown

in

Figures 2, 3, and 4, respectively. Negative azeotropes have a

maximum

boiling

temperature

and

minimum

total

vapor

pressure. A representative system forming a negative azeotrope

(maximum boiling) is chloroform and acetone; its X-Y curve, TX or T-Y curve, and P-X or P-Y curve are given in Figures 5,6,
and 7, respectively. The term azeotrope gives no indication of
whether one, two, or more liquid phases are formed. To make it
clear,

the terms "homoazeotrope" and "heteroazeotrope" were

introduced[6].

Typical

vapor-liquid

equilibrium

data

binary homoazeotropes are represented in Figures 2 to 7.

typical

heteroazeotrope

is

the

water-n-butanol

system

for
A
for

which an X-Y curve, T-X (or Y) curve, and P-X (or Y) curve are
represented in Figures 8 and 9, respectively.

It should be

noted that most azeotropes commonly encountered in industry

are binary, minimum-boiling azeotropes, though many

Pressure: 760 mmHg

A cetone( I )-M ethanol(2)

0.6

0.2

Figure 2 Liquid mole fractions and vapor mole fractions for

minimum-boiling mixtures of acetone and methanol[4]

11
66

Pressure: 7 6 0 mmHg
A cetone( I )-M eth a n o l(2 )

a z e o tro p e

Acetone m ole fractions

Figure 3 Liquid boiling points and vapor condensation
temperatures for minimum-boiling azeotropic
mixture of acetone and methanol[4]

12

T em perature: 45 C
A cetone( I )-M eth an ol(2)

mmHg

a z e o tro p e

Pressure,

V + L

A c e to n e m o le f r a c t io n s

Figure 4 Pressure and Liquid Compositions

for Minimum-Boiling Azeotropic Mixture
of Acetone and Methanol[4]

Pressure 760 mmHg

Acetone( I )-Chloroform(2)

Figure 5 Liquid mole fractions and vapor mole fractions for maximumboiling azeotropic mixture of acetone arid chloroform[4]

14

P re s s u re : 7 6 0 m raH g
A cetone( I )-C h lo ro f o rm (2 )

T em perature,

A zeotrope

acetone mole fractions

Figure 6 Liquid boiling point and vapor condensation

temperatures for maximum-boiling azeotropic
mixture of acetone and chloroform!4]

15

T e m p e r a tu r e : 4i? C

Pressure,

mmHg

A ceto n e( I )-C h lo ro fo rm (2 )

A zeotrope

Acetone m ole fractions

Figure 7 Pressure and liquid mole fractions for maximum

boiling azeotropic mixture of acetone and
chloroform[4]

1
Pressure: 760 mmHg
Water(1 )-n-Butanol(2)

0.2

Figure 8 Liquid mole fractions and vapor mole fractions for

heteroazeotropic mixture of water and nbutanol[4 ]

17

T em perature,

Pressure: 7 6 0 mmHg
Water( I )-n -B u ta n o l(2 )

v + L

Water m ole fra ctio n s

Figure 9 Liquid boiling point and vapor condensation
temperatures for minimum-boiling heteroazeotropic
mixture of water and n-butanol [4]

18
maximum-boiling binary, ternary and even quaternary ones are
known to exist.
A

close-boiling

mixture

is

difficult

to

separate

by

conventional distillation because components in the mixture

have very similar physical properties and boil very closely
together.
propane

Examples
system

of

or

such

systems

include

aromatic-paraffinic

the

propylene-

hydrocarbons.

These

close-boiling mixtures have relative volatilities close to

unity.

Figures

10

and

11 present the

typical vapor-liquid

behavior of these kinds of solutions.

Other solutions that

form regions of very low relative volatility (such as acetic

acid and water)
pure-component
solution
acetic

vary

can fall into this category even though the

boiling
quite

widely.

acid and water

equilibrium data

points

of

components

Vapor-liquid

is shown

for Figures

the

in Figure

2 through

in

the

equilibrium

for

12

12

(Vapor-liquid

is adapted

from

[4]).

,
From

Figures

2 through

12,

it

can

be

seen

that

the

separation of these mixtures by means of ordinary distillation

is

either

theoretical

impossible
plates

(for

(for

azeotropes)

close-boiling

or

needs

mixtures)

so
that

many
the

capital investment for equipment is prohibitively high.

Azeotropic and Extractive Distillation Processes

Azeotropic

and

extractive

distillation

are

separation

processes in which a third separating entrainer (also called

Temperature: 310.95 K
Propylene(1 )-Propane(2)
Diff of B.P.(2) and B.P.(1): 5.95 K

Figure 10 Liquid mole fractions and vapor mole fractions

for close-boiling mixture of propylene and propane[4]

Pressure: 760 mmHg

n-Hexane(1 )-Benzene(2)
Diff. of B R.(2) and B.P.(1): 11.4 K

Figure 11 Liquid mole fractions and vapor mole fractions for

close-boiling mixture of n-hexane and benzene [4 ]

Pressure: 760 mmHg

Water(1 J-Acetic Acid(2)
Diff. of B R.(2) and B P ( I ) : 18 K

Figure 12 Liquid mole fractions for low relative volatility

mixture of water and acetic acid[4]

22

solvent

or

an

agent)

is

deliberately

added

to

the

distillation column to improve the relative volatility of the

feed components.
Extractive

distillation

refers

to

those

processes

in

which a high-boiling solvent is added to a tray in a column to

alter the relative volatilities of components in the main feed
to the column. The solvent is generally selected such that it
boils at a temperature so far above the boiling points of the
feed

components

impossible. Also,
untreated

that

the

formation

of

a new

azeotrope

any troublesome azeotropes present in the

feed disappear

in the presence

of

a well-chosen

solvent.

The absence of azeotropes plus the fact that

solvent

can

be

is

recovered

by

simple

distillation

the
makes

extractive distillation a less complex and more widely useful

process than azeotropic distillation. The generally accepted
definition of extractive distillation is given by Benedict and
Rubin[5]:
Distillation in the presence of a substance that is
relatively non-volatile compared to the components to be
separated and, therefore, is charged continuously near
the top of the distilling column so that an appreciable
concentration is maintained on all plates of the column.
The typical configuration for an extractive distillation
process is shown in Figure 13, in which methyl cyclohexane and
toluene are to be separated[6]. These two components do not
form an azeotrope, but their relative volatility is less than
1.01

at

low

concentrations

of

toluene.. The

volatility

methyl cyclohexane relative to toluene is enhanced by the

of

Makeup
solvent

Freshfeed

Extractive
distilation
column

Toluene
product
Solvent
recovery
column

Recycle solvent

Figure 13 Simplified flow diagram for an extractive distillation process

to separate toluene and methyl cyclohexane[6]

24
addition of a solvent.

This allows

separation of these two

components with fewer stages than would be required in simple

distillation. The solvent chosen is less volatile than the two
components and, in order to maintain a high concentration of
solvent throughout most of the column, must be introduced to
the extractive distillation column above the fresh-feed stage.
It is usually introduced near the top stage. Recovery of the
solvent is relatively simple for extractive distillation as
compared with azeotropic distillation. The solvent chosen does
not form an azeotrope with non-solvent material in the bottom
product from the extractive distillation tower,

and solvent

recovery can be accomplished by simple distillation.

Azeotropic

distillation

refers

to

those

processes

in

which a component (called solvent or entrainer) is added above

the main feed tray to form (or nearly form) an azeotrope with
one or more of the feed components.

The azeotrope

is then

removed as either a distillate or a bottom product. Azeotropic

distillation can also refer to a process in which a solvent is
added to break an azeotrope that otherwise would be formed by
components

in the feed.

Here,

the process

is distinguished

from extractive distillation because the solvent appears in

the distillate, from which, it must be separated and recycled
back

to

flowsheet

the
for

top

section

of

the

column.

azeotropic. distillation

of

A
an

representative
ethanol-water

mixture using pentane is shown in Figure 14[6]. Nearly pure

ethanol [78.3C] cannot be obtained from a dilute mixture with

Fresh Feed

Decanter

Azeotropic
Distillation

Distillation

Ethanol Product

W ater

Figure 14 Azeotropic distillation process for separating

ethanol and water by using pentane as entrainer[6]

26
water

[IOO0C]

by

simple

distillation

at

I atm.

because

minimum-boiling homogeneous azeotrope [78.15C] is formed. As

discussed by Black et a l .[7], the separation can be made by
low-pressure

[<0.1135

atm]

simple

distillation,

extractive

distillation with gasoline or ethylene glycol, or azeotropic

distillation with n-pentane, benzene, or diethyl ether. Figure
14 shows the azeotropic distillation process with n-pentane as
the entrainer. In the azeotropic column, nearly pure ethanol
is taken

as a bottom product

and a heterogeneous

boiling azeotrope is taken as a distillate,

decanted

into

a pentane-rich

reflux

minimum

condensed,

and

stream and an aqueous

ethanol stream that is separated by distillation to produce a

bottom product of nearly pure water and a distillate that is
combined with the reflux from the decanter.
It should be noted that either extractive distillation or
azeotropic distillation can be used to separate both close
boiling mixtures and azeotropes.
Heterogeneous
azeotropic

azeotropic

distillation.

distillation

Doherty

et

vs.

homogeneous

a l .[8]

classified

azeotropic distillation into the homogeneous,

in which the

added solvent or entrainer alters the relative volatility of

the two azeotropic constituents without inducing liquid-liquid
separation, and the
alters

the

heterogeneous, in which the added solvent

relative

volatility

of

the

two

azeotropic

constituents and induces liquid-liquid separation. Homogeneous

azeotropic

distillation

covers

the

traditional

extractive

27
distillation for the separation of an azeotrope. Laroche et
a l [I ]

argued

that

heterogeneous

azeotropic

distillation

suffers from a major drawback: Operating such a column can be

very

difficult

because

upsets

can

induce

phase

separation

inside the column, leading to severe efficiency losses. Thus

homogeneous azeotropic distillation which does not have the
above

disadvantage

heterogeneous

is

easier

azeotropic

to

operate,

can

distillation while

outperform

separating the

same azeotrope, and represents an economically attractive way

of separating azeotropes.

The Brief History of Azeotropic

and Extractive Distillation

The

first

successful

distillation

of

an

azeotropic

mixture is credited to Young(1902)[9], who in 1903, received

a

patent

for

anhydrous

his

alcohol.

landmark

event

in

industry.

In

1908

distillation

batch

distillation

Young's
the

process

invention

history

of

Guillaum[10]
for

the

process

can be

the

of

making

regarded as

chemical

patented

removal

for

an

processing
extractive

fusel

oil

from

fermentation alcohol. Since then, research on azeotropic and

extractive

distillation

has

received

periodic

attention,

usually as the result of some crisis in world history.

For

example, World War I caused a tremendous increase in demand

for butanol that was met by the introduction of an azeotropic
distillation process.

Acetic acid demand also

climbed,

and

28
engineers

found

that

substantial

increased production rates

from

conventional

in

energy

and

could be realized by converting

distillation

distillation process.

savings

Then again

process

to

an

azeotropic

during World War

IIz the

greatly increased demand for rubber was met by the development

of a synthetic rubber industry. This industry relied heavily
on the use of azeotropic and extractive distillation for the
recovery of high-grade butenes and butadiene.
Currently,

azeotropic

and

extractive

distillation

are

used for such widely diverse applications as dehydration of

acetic acid, formic acid, ethanol, isolation of isoprene from
a C5 hydrocarbon
production,
aliphatic

cut,

2-propanol production,

deemulsification
and

aromatic

of

oi l ,

hydrocarbons

and
or

vinyl

acetate

separation
alkanes

of
and

alkenes[16].

An Overview of Solvent Selection Methods

Over the years, theoretical methods, empirical heuristic

rules

and experimental methods have been developed for the

selection of promising solvents or entrainers for azeotropic

and extractive distillation to separate azeotropes or close
boiling mixtures. The following gives only a brief description
of the empirical guidelines and some thermodynamic bases for
the selection process.

29
For Azeotropic Distillation
For azeotropic distillation,

Berg[11] has proposed the

general heuristic criteria to screen entrainers based on the

liquid

classifications

using

hydrogen

bonding

ability

and

polarity as attributes of liquids:

1.

To

separate

closely

boiling

pair

or

a maximum

azeotrope, choose an entrainer such that:

a) The entrainer forms a binary minimum azeotrope with
only one component, or
b) The entrainer forms a binary minimum azeotrope with
each

component

but

one

minimum

azeotrope

boils

at

significantly lower temperature than the other, or

c ) The entrainer forms a ternary minimum azeotrope that
boils

at a significantly lower temperature than any binary

azeotrope.

The

ratio

of

compositions

of

the

original

components in the ternary must be different from their ratio

before the entrainer is added.
2. To separate a minimum azeotrope, choose an entrainer
such that:
a) The entrainer forms a binary minimum azeotrope with
one component that boils at a significantly lower temperature
than the original minimum azeotrope, or
b) The entrainer forms a ternary minimum azeotrope that
boils

at a significantly lower temperature than any binary

minimum azeotrope and in which the ratio of compositions of

the original components is different from their ratio in the

30
binary minimum azeotrope.
Berg has also listed the following desirable properties
of a solvent or entrainer for hydrocarbon separations[11]. The
entrainer: .
1. Should boil within a limited range

(0-30C)

of the

pure-component boiling points of the hydrocarbon mixture to be

separated.
2. Should form, on mixing with the hydrocarbon, a large
positive

deviation

from

Raou l t 1slaw

to

give

minimum

azeotrope with one or more of the hydrocarbon types in the

mixture.
3. Should be soluble in the hydrocarbon.
4. Should be easily separable from the azeotrope.
5. Should be inexpensive and readily obtainable.
6.

Should

be

stable

at

the

temperature

of

the

distillation.
7.

Should

separated

and

be

non-reactive with the

with

the

materials of

hydrocarbon

construction

being

of

the

equipment.
To make recovery easy,

the ideal solvent or entrainer

will usually boil I O 0C to 3 0 C below the hydrocarbon mixture

providing

large

temperature

hydrocarbon-entrainer

difference

azeotrope

and

between

the

the

unaffected

hydrocarbon.
'

In

addition,

Treybal[12]

has

listed

the

following

attributes.expected for a promising solvent or entrainer:

31
1. If the entrainer forms a new azeotrope with only one
of the constituents of the binary mixture to be separated, it
should preferably be the constituent present in the minority
to reduce the heat requirements of the process.
2.

The

entrainer

new

content,

azeotrope
to

should

reduce

the

preferably
amount

of

be

lean

in

vaporization

necessary in the distillation.

3.

The

new

azeotrope ' should

preferably

be

of

the

heterogeneous-liquid type, which then greatly simplifies the

recovery of the entrainer.
4. Should be nontoxic.
5. Should be of low latent heat of vaporization.
6.

Should

be

of

low

freezing

point

to

facilitate

storage and outdoor handling.

7.

Should

be

of

low

viscosity

to

provide

high

tray

capacity.
Doherty et al. [13] have introduced a method for screening
solvents for homogeneous azeotropic distillation based on the
so-called simple distillation residue curve map analysis. A
residue

curve

map

is

a triangular

diagram

(with

the pure

components at each vertex) which shows the locus of the liquid

phase

composition

as

it varies

with

time

during

a simple

distillation process. For illustration, the residue curve map

for the acetone-chloroform-benzene system is shown in Figure
15. The sketching procedure for the residue curve map needs
only a knowledge of binary and/or ternary azeotropes along

32

Benzene

Simple distillation boundary

Acetone

Binary azeotrope

Chloroform

Figure 15 A typical residue curve map showing a distillation

boundary[13]

33
with

their

respective

compositions,

boiling

point

temperatures

and

and was described by Doherty et a l .[14][15].

According to Doherty and Caldarola's criterion[16], based on

the analysis and classification of the residue curve maps, in
the case of a minimum boiling azeotrope, any.component "which
does not produce an internal distillation boundary between the
two components to be separated" can be used as an entrainer.
Based

on

the

results

from

directed

graph

theory

and

an

adjacency, matrix representation, Foucher et a l . [16] described

the

extended

determination

rules
of

and

the

an

automatic

structure

of

the

procedure
simple

for

the

distillation

residue curve maps for ternary mixtures as shown by Figure 16.

Their

procedure

requires

only

knowledge

of

boiling

temperatures and approximate compositions of any azeotropes in

the mixture.

They have claimed that their procedure enables

the rapid identification of feasible entrainers,

but their

approach is limited to homogeneous azeotropic distillation.

Stichlmair et a l .[18]

have also tackled the entrainer

selection problem and his results were given as follows:

1.

If

the

binary

mixture

exhibits

a maximum

boiling

azeotrope, the entrainer must a) be a high-boiling substance

or b) form new maximum boiling binary azeotropes.
2.

If

azeotrope,

the

binary

mixture

the

entrainer

must

exhibits
aj

be

a minimum
a

minimum

boiling
boiling

substance or b) form new minimum boiling binary azeotropes.

By examining homogeneous azeotropic distillation, Laroche .

34

Figure 16 The algorithm for determining the residue curve map

and the selection of azeotropic entrainers[16]

35
et a l .[19] have found that homogeneous azeotropic distillation
can behave in a very unusual manner. The following lists some
strange features for homogeneous azeotropic distillation: '
1. Increasing reflux in a given column does not always
increase

separation.

In

fact,

in many

cases,

there

is

no

separation at all at infinite reflux.

2. Meeting the same specifications with a large number of
trays sometimes requires higher internal flows.
3. Sometimes, the separation is feasible, but neither the
direct nor the indirect sequence is possible.

Indeed, there

are cases where we can recover the intermediate boiler as a

pure distillate product, but not the light boiler. There are
also cases where we can recover the intermediate boiler as a
pure bottom product, but not the heavy boiler.
On

the

basis

of

the

above

observations,

they

have

analyzed the assumptions behind Doherty et al.'s rules and

Stichlmair

et

al.'s

existing entrainer

rules,

and

they

selection rules

have

found ;that

the

for the minimum boiling

azeotrope are conflicting. Thus, they proposed the following

entrainer

selection

procedure

for

homogeneous

azeotropic

distillation:
1. Eliminate all chemicals that introduce additional
azeotropes.
2. By comparing the corresponding equivoIatiIity curve
diagrams,

select the best (or best few) candidates in each

of the following classes:

36
a) Heavy entrainers that send the lighter azeotropic
component to the top of the extractive column;
b) Heavy entrainers that send the heavier azeotropic
component to the top of the extractive column;

c) Light entrainers that send the lighter azeotropic

component to the bottom of the extractive column;
d) Light entrainers that send the heavier azeotropic
component to the bottom of the extractive column.
3. Keep all intermediate entrainers.
4.

Design,

sequence(s )
entrainers.

cost

and

optimize

corresponding
The

best

to

entrainer

the

the

feasible

separation

remaining

candidate

yields

the

lowest

total

annualized cost.

For Extractive Distillation

Scheibel[20] has pointed out that
1 . The entrainer must not form an azeotrope with either
of the components of the mixture to be separated.
2. The entrainer must be sufficiently less volatile than
either of the components.
3.

The

entrainer must

have

a different

effect on the

partial pressure of each of these components in the solution

at

given

concentration.

This

means

that

selectivity

is

different from unity.

4. The following well-known properties of mixtures can be
made

use

of

for

selecting a

solvent,

first,

members

of

37
homologous series show little or no deviations from

Raoult's

law; second, when two compounds show deviation from Raoult's

law, one of these compounds shows the same type of deviation
from

any

member

of

the

homologous

series

of

the

other

component. Thus, any member of the homologous series of either

component

of

an

azeotrope

might

be

used

to

separate

the

azeotrope.
It is only necessary to choose such an entrainer that
satisfies the above conditions (I),

(2), and, (3).

These principles are best illustrated in Table I for the

selection of solvents for separation of the acetone-methanol
azeotrope.
The solvents listed in the left-hand column of Table I
are in the homologous

series with acetone and deal with it

while those in the right-hand column are in homologous series

with methanol and probably deal with methanol.

Table I . Solvent Selection Based on Scheibel's Rules.

(Adapted from [20])
Acetone
B .P . 56.4 C
Solvent

Methanol
B.P. 64.7 C
B .P .C

Solvent

B.P.C

Methyl- ethyl ketone... ...79.6

Ethanol....... ...78.3

Methyl n-propyl ketone ..102.0

Methyl isobutyl ketone ..115.9
Methyl n-amyl ketone.. ..150.6

Propanol......
Butanol........
Amyl alcohol...
Ethylene glycol

...97.2
. .117.8
. .137.8
. .197.2

38
B e r g [12] has suggested that all solvents can be divided
into five classes based on their potentialities for forming
hydrogen bonds. The criteria proposed by Berg for successful
extractive

distillation

solvents

are

that

they

boil

considerably higher than the components being separated, form

no

minimum

azeotropes

with

the

components, and

be

highly

hydrogen-bonded liquids, that is. Class I or Class II of the

hydrogen bond classification.

(See Appendix C, part C for the

classification in detail).
For hydrocarbons, Tassios[21] has proposed a procedure as
follows:
1.

Compare

the

molar

volumes

of

the

close-boiling

hydrocarbons to be separated, and prepare a list of potential

entrainers:
a) If the difference amounts to 5% or more, consider the
entrainers with high polar cohesive energy; and
assume that the optimum entrainers will also have a
high polar cohesive energy;
b) If the molar volumes are too close, select as
potential entrainers those having high electron
affinity.
2.

Consult the literature to see whether experimental

data might be available on some systems under consideration;

3. If any of the close-boiling hydrocarbons belong in the
three groups: saturated hydrocarbons, olefins, or aromatics;

use the Weimer-Prausnitz method to estimate the selectivity of

39
the potential entrainers.
4.

For

generally

more

reliable

values,

and

for

hydrocarbons not included in the classes, use the PierottiDeal-Derr(P D D ) method to calculate entrainer selectivity.
5. Compare the potential entrainers for temperature and
concentration effects.
Treybal[13] has stated that the general requirements of
a satisfactory extractive-distillation solvent or entrainer
are:
1. High selectivity, or ability to alter the vapor-liquid
equilibria of the original mixture sufficiently to allow its
easy separation, with only small quantities of entrainer.
2. High capacity, or ability to dissolve the components
in the mixture to be separated.

It frequently happens that

substances that are incompletely miscible with the mixture are

very selective; yet if sufficiently high concentrations of an
entrainer

cannot

be

obtained

in

the

liquid

phase,

the

separation ability cannot be fully developed.

3. Low volatility in order to prevent vaporization of the
entrainer

with

the

overhead

product

and

to

maintain

high

concentration in the liquid phase.

4. Separability; the entrainer must be readily separated
from the mixture to which it is added,

and particularly it

must form no azeotrope with the original substances.

5. The same considerations of cost, toxicity, corrosive
character, chemical stability, freezing point, and viscosity

40
apply to entrainers as to agents for azeotropic distillation.
Tassios [22] has also pointed out ..that the entrainer has
physical, and

chemical

effects

on

the

constituents

of

the

binary mixture. The chemical effect is due to the formation of

a complex of loosely bound aggregates. This phenomenon, called,
solvation,

is

considered

the

result

of

Lewis

acid-base

interaction. Usually, the physical effect is larger than any

chemical

effects,

and

usually,

the

physical

and

chemical

effects complement one another.

Kolbe

et

a l . [23 ] have

proposed

procedure

for

the

selection of entrainers for extractive distillation using both

the. UNIFAC model and UNIQUAC as shown, in Figure 17.
Yeh[24] has applied the polarity diagram to compare the
affinity of an entrainer. for key components

in binary and

ternary mixtures, and used this chemical effect to explain the

behavior of DMFA and DMSO in alcohols. Furthermore, the MOSCED
model

developed

selectivity

of

by

Eckert

entrainers

et
and

a l . [25]
then

was

used

entrainers

to
were

predict
ranked

based on predicted selectivity, but the MOSCED model cannot be

applied to aqueous systems.
The selection of solvents or entrainers for extractive
distillation by their capacity is another important issue.
Knapp et a l .[26] used minimum entrainer flows in extractive
distillation to

second-tier rate a set of entrainers. Their

approach depends upon analyzing bifurcations of the finite

difference equations describing the middle section of the

41

List of s o lv e n t s differing
in functional groups

Calculation of s e l e c t i v i t y
at infinite dilution us i n g
UNIFAC and DDB

P r e - s e le c t io n o f s u i t a b l e
solvents

Calculation of equilibrium
diagrams at d i ff e r e n t
con stan t solvent concentrations

^VLE data
a v a i l a b l e in
\
DDB
Calculation of
UNIQUAC parameters
from UNIFAC

Evaluation of
UNIQUAC parameters

Plot of y - x

diagrams

S e le c t io n of the " b e s t
solvent

Figure 17 Procedure for the selection of extractive solvents

based on UNIFAC and UNIQUAC m o d e l s [23]

42

column for a mixture with a given VLE model.

In summary, selecting entrainers and ranking them for
azeotropic
inexact

and

extractive

activity,

based

experimentation.

distillation

to

some

Furthermore,

extent

the

is

complex

and

on

experience

and

underlying

theory

for

entrainer selection is not adequate to allow the prediction of

unique, complete and certain answers.

Selection of Azeotropic and Extractive Distillation

In many

industrial

mixtures

the

key

components

under

consideration can be separated by extractive distillation and

azeotropic distillation by the selection of a proper solvent
or entrainer, and an economic comparison of the processes is
usually required

for the final decision.

Another

consideration in the selection of a process

important

is the thermal

stability of the components, particularly that of the heavy

key

component

higher

and

heavier

temperature

than

key

component which

the

heavy

key

boils

at

component. . Van

Wink l e [27] has pointed out that if one of these components is

unstable or tends to decompose, polymerize, or otherwise react
at a higher
distillation

temperature level, it is better to use azeotropic

to

keep

bottom product at

the

bubble-point

its lowest point.

temperature

A heavy

of

the

solvent would

increase the bubble point above this temperature and could

have

a deleterious effect on the somponents in the bottom

product.

43
Gerster[28] has indicated that if the feed is a close
boiling

hydrocarbon

properties

pair,

the

differences

of the feed components

in

the

physical

is usually comparatively

small, so that an agent is reguired to improve the relative

volatility
achieved

over

best

the

in

entire

height

extractive

of the

distillation

column;
where

this

the

is

agent

enters at, or near the top of the column, and is discharged at

the bottom of the column. He has also noted that azeotropic
distillation is particularly useful when the feed component
selected to come overhead as an azeotrope with the solvent is
present in the feed in small amounts since in such an instance
the amount of solvent or entrainer needed to be circulated is
small, resulting in only small additional steam costs because
of the presence of the solvent, and in a low solvent recovery
cost.
Barnicki et a l .[29] have classified the separations of
liquid mixtures into two types: dilute and bulk. A separation
is considered dilute when the total distillate or bottoms of
a potential

distillation operation

is

less than

5% of the

feed, such as the recovery of acetic acid from dilute acetic

acid-water mixtures . In addition, a large distillate-to-bottom
(D/B)

ratio

distillation

has
than

greater
small

effect

(D/B)

on

ratio.

the

economics

Mixtures

of

composed

a
of

mostly low-value, low-boiling components to be separated from

a small amount (less than 10-15%) of low value, high-boiling
component require a relatively large amount of energy to

44
Table 2. Heuristics and Rules for the Selection of Separation
Methods.
Favored method

Conditions

Extractive
distillation

generally.favored; closeboiling mixtures.

Azeotropic
distillation

very high-boiling azeotropes or close-boiling

mixtures (above 170 C) ; a low concentration
(<1015%) of the component that forms a
minimum boiling azeotrope with the solvent; in
this case, the minimum boiling azeotrope will
go overhead[29].
a dilute solution (between 1% and 5%) of a
high boiling, polar compound; extractive or
azeotropic distillation would require
vaporization of large amounts of the feed;
temperature sensitive components in the feed;
the separation of aliphatic or olefin
hydrocarbon from aromatics[29].

Solvent
extraction

vaporize

the

distillate.

85-90%
In

of

these

the

feed

cases,

that

will

extractive

appear
and

in

the

azeotropic

distillation can be eliminated as potential methods, solvent

extraction is favored.
Some components may decompose or react unfavorably at the
temperature needed for distillation. Moreover,

the freezing

point of a component may be too high for distillation to be

carried

out

at

an acceptable

temperature

and pressure.

In

these cases, extractive, or azeotropic distillation cannot be

used. Again, solvent extraction is favored.
Treybal[12]
generally

has

considered

distillation since
because

the

shown that extractive distillation

process

(I)

more

desirable

than

is

azeotropic

there is a greater choice of agents

does

hot

depend

upon

the

accident

of

45
azeotrope

formation and

(2)

smaller quantities

of an agent

must be vaporized.
Figure 18 gives the relative volatility for azeotropic
distillation, extractive distillation, and solvent extraction
that

are. required

comparison

is

to

based

give

on

the

concentration

and

four times

distillation

and

in

solvent

equal

plant

assumptions
as much

cost[30].
of

liquid

extraction

as

67%
in
in

This

solvent

extractive
ordinary

distillation. The ordinate represents the relative volatility

required in extractive distillation or in solvent extraction
and

the

abscissa

distillation.

the

relative

volatility

for

ordinary

Sofvert EdJaction (SE)

Extractive Dlstffladon (ED)

RV for o rdinary distillation

Figure 18 Relative volatility required for different

separation with equal cost[30]

47

CHAPTER 3

SOLVENT SELECTION CRITERIA

Although

the

final

selection

of

solvents

must

be

determined by means of an economic evaluation in which all

variables and criteria are considered for the whole process
including

the

preliminary

recovery process, screening

stage

should

consider

two

solvents

at the

aspects: separation

enhancement and easy reversibility.

Separation Enhancement Characterization

The separation enhancement is a major concern for the

separation aided by adding a solvent. Figure 19 shows that the
successful

separation of acetone-methanol

azeotrope

can be

carried out through distillation by modification of its vaporliquid equilibrium by adding DMSO and M E K [24]. The degree of
separation enhancement can be expressed by either relative
volatility or selectivity.

Separation Factor
The degree of separation that can be obtained with any

P ressure: 630 mmHg

With DMSO

No solvent

With MEK

wt% of acetone in liquid phase

Figure 19 Vapor-liquid equilibrium curves for the methanol-acetone

system with solvents and with no solvent [24]

49
particular separation, process is indicated by the separation
factor:

ocsij

The separation factor

(I)

between components i and

Cts l j

is the

ratio of the mole fractions of those two components in product

I divided by the ratio in product 2. An effective separation
is accomplished to the extent that the separation factor is
significantly different from unity. If
of

components

component

component

i and

tends
does,

has

to

Cts l j

= I, no separation

been accomplished.

concentrate

and component

in

product

If

Cts i j

I more

I,

>

than

tends to concentrate

in

product 2 more than component i does. On the other hand,

if

a ^ < I, component j tends to concentrate preferentially in

product I .and component.i tends to concentrate preferentially
in product 2. By convention, components i and j are generally
selected so that

Cts i j

, defined by Equation (I), is greater than

unity.

Relative Volatility
For

vapor-liquid

system,

the

separation

factor

is

written as
a

_ Vil/Xll _ y iP i

(2 )

where a is the relative volatility, x is the liquid phase mole

fraction,

is

the

vapor

phase

mole

fraction,

is

the

50
activity

coefficient,

and

is

the

pure

component

vapor

pressure. The goal of the selection of solvents is to change

the

relative

volatility,

Cti j ,

as

far

away

from

unity

as

possible. The initial design phase for distillation involves

determining the number of theoretical plates

(NTP at finite

reflux) reguired because of its impact on the column cost. The

connection between the minimum number of theoretical plates
Nmin (at infinite reflux) required for a specified separation
purity

and

relative

volatility

is

established

through

the

Fenske equation[27],
_ log [(XiZx7
.)JZ(XiZ^ ) jb]1
mln
I ^ T

(3>

where d means the distillate product while b means the bottom

product.

The

equilibrium
volatility,

value
stage.

Cti j ,

of
It

Nmin. includes
should

be

the

noted

reboiler

that

the

as

an

relative

may be approximated as the geometric means of

the relative volatility values at the top and bottom of the

column:
a av=

1/2

(4)

If there is a wide variation of relative volatility between

top and bottom,

an additional value,

ctF, can be used in the

average:
a av= (a da FJ,) 1/3

(5)

where F represents the feed. Figure 20 shows the relationship

between the minimum number of theoretical plates required for
C

Number of theoretical plates

Relative volatility
Figure 20 NTP requirements for different relative volatilities [27]

52
different separation purity and relative volatility at total
reflux. It can be seen that the number of theoretical plates
approaches infinity as the relative volatility becomes closer
to

unity.

That

is

the

situation

for

the

presence

of

an

azeotrope or close-boiling mixture. The number of theoretical

plates

decreases

increases

to about

rapidly
1.25.

when

the

Figure

relative

21 presents

volatility

the number of

theoretical plates and column cost as functions of relative

volatility. It can be seen that the column cost has a similar
relation

to

relative

volatility

as

does

the

number

of

theoretical plates.
Selectivity
Since the ratio

(PicZPj) is nearly constant for small

temperature changes, the only way the relative volatility is

affected is by introducing a solvent which changes the ratio
(Yi/Yj)

This ratio, in the presence of the solvent, is called

the selectivity (Slj) :

[ Y j / Y j ] solvent

The

activity

coefficients

concentrations
composition.

in
A

the

high

of

the

solvent

selectivity

()

components

depend
gives

on
a

at

the
high

finite
mixture
relative

volatility, making an easier separation possible. Sij increases

with increasing solvent concentration as shown in Figure 22,
usually becoming a maximum at 100% solvent[21], and falls with

rising temperature as Figure 23 shows. The preliminary

C o lu m n cost

Separation Purity: 99%

~-- a

2.5

R e la tiv e volatility

Figure 21 NTP and column cost vs. relative volatility[24]

4.2
T em perature @298 K

0.2

0.4

0.6

0.8

Ethanol mole cone, (solvent free basis)

Figure 22 Effect of ratio of solvent to mixture on
selectivity

4
S o l v e n t : M ixture = 1:1
Ethanol(1)-Water(2)-2-Nitropropane(3)

Figure 23 Effect of temperature on selectivity

56
selection of solvents can be based upon the limit selectivity,
which is defined as the ratio of the activity coefficients of
the key components when each alone is present in the solvent
at infinite dilution,

(7 )
Kumar and Prausnitz[31 ] have discussed the annual process cost
for the separation of propylene-propane system based on the
limit selectivity at infinite dilution. As Figures 24 and 25
show, the annual cost falls with increasing selectivity, and
the

total

extractive

distillation

process

cost

is

approximately the same as that for binary distillation for

propylene-propane at a selectivity of 2.6. A further increase
of the limit selectivity makes extractive distillation process
more economically attractive than conventional distillation.

Activity Coefficients at Infinite Dilution

Comparing the

limit

selectivity of different

solvents

according to Eguation 7 depends upon activity coefficients at

infinite

dilution

experiment,
theoretical

or

(ACID).

estimated

ACID

using

thermodynamic models,

can

be

obtained

semi-empirical

and

from
semi-

or compiled ACID data in

text[32] or database[33] form. Table 3 lists several good

experimental techniques available for determining ACID.
The references listed in Table 3 contain experimentally
measured ACID values for some binary systems.

Xs = 0.9

Ul

300

1.5

16

1.7

1.8

i----- r

1.9

2.1

I-------------

2.2

Selectivity, y - C3H Zy - C3HS

Figure 24 Effect of solvent selectivity on cost for low selectivity [3 1 ]

Xs = 0.9

1000

Act. Coeff. infinite dilute cone, for C3H8 = 6.0

Total Cost

800 -

Total For
Fractional

CD

600 -

Distillation

Operating Cost

400 Fixed Cost

Selectivity, yC3H8 Iyxc3h6

Figure 25 Effect of solvent selectivity on cost for high selectivity [31]

59
Table 3. Experimental Methods for Determining ACID.
Method

Author

Reference

Dew Point Measurement

Trampe and Eckert

[34],

Dynamic Gas Chromatography

Pescar and Martin

Shaffer and Daubert

[35]
[36]

Gas Stripping

Leori et a l .
Richon et al.

[37]
[38]

Differential Ebulliometry

Scott
Trampe and Eckert

[39]
[40]

Headspace Chromatography

Hassam and Carr

[41]

Thermodvnamic Models for the Estimation of ACID

Activity

coefficients

at

infinite

dilution

are

so

valuable in solvent evaluation for mass-aided separations that

many efforts, empirical and theoretical,
address this problem.

have been made to

Their application can frequently save

extensive experimental or computational effort.

Many

equations

proposed . for

correlating

activity

coefficients with composition and temperature can be extended

to

the

infinite

dilution

situation.

The

correlations.

applicable to binary systems at infinite dilution are listed

in Table 4.
These types of equations require empirical interaction
parameters between two components that can be found in the
DECHEMA Vapor-Liquid Data Collection for many binary systems.
Unfortunately,

much

such

information

solvent selection applications.

is not

available

for

60
The importance of a knowledge of y has led to the
development of different prediction methods. The solubility
parameter approach developed by Scatchard-Hildebrand[47] has
Table 4. Binary Activity-Coefficient Correlations
at Infinite Dilution.

Margules[42]
van Laar[43]
Wilson15[44 ]

1In 2~Ap^
I In Agi -Aqg

NRTL25 [45]

^12

(otzI2 2 ^
)^ ^ 12

^"21

UNIQUAC35 [46]

Inrq/rg+gq [ (z/2 )ln(q1rg/q^/r1)- Inr21+!Tqg

]+Iq tqlg/rg

Inr g/rq+qgf ( z / 2 ) l n ( q grq/qq/rg)-Inr12+!UgqJ+ig-rgiq/rq

I) A1j = V jZV1 exp [ - ' (A11 - Ai i JZRT]

V 1 molar volume or pure liquid component i
^ interaction parameter between component i and j; Aij=Aji
2) r 7 <9ti * 9)/*

. .
gij interaction parameter between component i and j; gij=gji
CL nonrandomness parameter; Qri j =Otj i
if

3) r.i j

U ij

r i

been

exp[-(uij - Ujj)ZRT]
interaction parameter between component i and j; U i j = U j i
relative van der Waals volume of component i
relative van der Waals surface area of component i
= zZ2(ri-qj)-(ri-l)
coordination number; z = 10

widely

extensions

used;

have

and

been

modifications,

proposed

by

many

improvements
authors.

and

Ye h [24]

61
summarized all the models developed along this line to 1986.
Modifications continue to be proposed. Table 5 lists all the
expressions established so far based on solubility parameters.
The most accurate prediction method for y" according to
solubility parameter approach is the MOSCED model proposed by
Thomas and Eckert[56], which gives an average error of 9.1%
for

3357

three

values

of

adjustable

parameters

are

y* The

original

parameters.

difficult

Among

to obtain

MOSCED model
them,
or

contains

acidity/basicity

estimate.

Also,

the

MOSCED is limited in application now to monofunctional species

for which certain pure component data are available. Moreover,
the MOSCED cannot apply to aqueous systems. An improved model
using solvatochromic

parameters

for

estimating

the

MOSCED

acidity/basicity parameters, developed by Howell et a l .(1989)

[57], has reduced the number of adjustable model parameters
from

to

(one

per

compound,

plus

two

per

class).

Nevertheless, this is achieved at the price of increasing the

overall average error from 9.1% to 11% (fitted data) and from
13% to 15% (predicted data). An empirical modification of the
MOSCED model with the addition of three adjustable constants
per class was proposed by Wu et a l .[58] as follows:
In Ym1 = V1/RT [ (X1-X2)2+q12q22(T1-r2)2Zc2+
( Ct1-O tg ) ( P 1(Sg ) /

( A 1- A g ) + C g ( T 1 - T g ) ( Ot1 -O tg ) + C g ]

+In (V1ZVg) ^+!-(V/Vg) ^

(8)

Where C1, C2, and C3 are empirical constants for one type of
system and are obtained by regression from experimental y

62
Table 5. Models for Estimation of Activity Coefficient at
Infinite Dilution Using Solubility Parameters.
Scatchard-Hildebrand1^ [47]
In Yto1 = V1ZRT(S1-S2)2
Scatchard-Hildebrand-Florv-Huqqins [48]
In Yra1 = V1ZRTCS1-S2^ l n ( V 1ZV2)+I-V1ZV2
Prausnitz-Ander son-Weimer2)[49]
In Y001 = V1ZRT[ (A1-A2)2+r22-2W12]+ 1 n (V 1ZV2)+ 1-V1ZV2
W12 =

C t 22

Helpinstill-Van Winklef50]
In Ym1 = V1ZRT[ (A1-A2)2+ (T1-12)2-2W12]+In(V1ZV2)+1-V1ZV2
W12 = Cft1 -t2)2
Hsieh-Newmanf51][52]
In Y011 = V1ZRTf (A1-A2)2+(t1-t2)2+(a1-a2) (T1-U2) ]+In(V1ZV2)+1
- V 1ZV2

Hansenf53]
In Ym1 = V1ZRTC (A1-A2)2+(t1-T2)2+(a131-a232)2]+ln(V1ZV2)+l-V1ZV2
Tiissen-Billet-Schoenmakers f5 4 1

In y", = V1ZRTI (X1-X2)2+(T1-r2)2+2(a:1-ci2)(B1-B2)3+ln(V./V2)+l

-v,/v2
Karqer-Snvder-Eon[551
In Ym1 = V1ZRTf (A1-A2)2+ (T1-T2)2+2 (Q1-O2)(P1-P2)+2 (Q1-K2) (T1-T2) ]
Thomas-Eekert's MOSCED model[56 I
In Ym1 = V1ZRTf (A1-A2)2+q12q22(Tl-T2)2ZC2+ (a1-2) (I3I-I32)/^] +
in(V1ZVp)aa+l-(V1/Vp)ak
__________________________
data.

The

difficulty

modification

less

in

useful

obtaining
to

general

constants
systems

makes

although

this
they

claim that the average correlation and prediction relative

63
deviations were reduced from 7.97% (original MOSCED model) to
4.62%

(Equation(S)).

The

difficulty

original MOSCED to aqueous

systems

of

application

is not

of the

overcome by the

empirically modified MOSCED model.

There is another approach based on the so-called group
contribution

to

predict

activity

coefficients

at

infinite

dilution. In the 1960's Pierotti et a l . [59] suggested the use

of the molecular structure as a basis for making quantitative
estimates

of

activities

in

mixtures

of

nonelectrolytes.

Starting from these considerations. Deal and Derr in 1969 [60]

proposed

the

ASOG

method

for

VLE

calculations.

Several

modifications have been introduced by various authors

Kojima

and

Tochigi[61]) .

Similar

approaches

have

(e.g.
been

developed by Fredenslund et al. [62] who in 1975 presented the

UNIFAC method.
improved

From 1975 onwards various modifications and

tables

of

the

published.

The

UNIFAC

prediction

of

interaction

model

V L E ; however,

is

when

parameters

convenient
the

method

have

tool
is

been

for the
used

for

predicting y, results of poor quality are obtained[56].

This insufficiency and the incapability of the original
UNIFAC method to provide quantitative and reliable estimates
of heats of mixing have led to new modifications of the UNIFAC
expression in which temperature is fitted to V L E , y and hE
data

simultaneously.

At

present,

versions of the UNIFAC expression,

there

are

two

modified

developed in Dortmund by

Weidlich and Gmehling [63], and developed in Lyngby, Denmark,

64
by Larsen et a l . [64], respectfully. Table 6 shows a comparison
of ym results predicted by the Dortmund modified version of
UNIFAC to the results of the original UNIFAC, as well as to
several other thermodynamic models.

Table 6. Deviation Between Experimental and Predicted

Activity Coefficients at Infinite Dilution.
(Adapted from [32])________________
AYm*t

AyVt
(%)

1.30
1.42
0.82

26.29
26.03
13.92

1.56
1.85
0.99

26.69
25.82
13.35

1.53

21.72

1.68

21.15

group contribution method

ASOG
UNIFAC
Modified UNIFAC
(Dortmund)
Modified UNIFAC
(Lyngby)

Hote: I

"1Y W " yW J'

Ay*

AY^elt
(%)

., ,

t 10,000 data points,

i 9,900 data point.

From Table 6, it is obvious that the Dortmund modified

UNIFAC

version

is

superior

to

other

group

contribution

methods. UNIFAC has an advantage that MOSCED does not have,

namely,

that

UNIFAC

has

the

ability

to

predict

coefficients at infinite dilution for aqueous

activity

systems. Table

7 shows that the deviation between experimental and

predicted

activity coefficients at infinite dilution by UNIFAC (Dortmund

version) are small and in the range acceptable for engineering
applications.

65
Table 7. Predicted Infinitely Dilute Activity Coefficients by
UNIFAC (Dortmund modification) for Aqueous Systems.
Solvent

Solute

Temp K

y"
Measured

y"
%Difference*
Predicted

Acetone

Water+

307.9
318.1
328.5

6.0210.13
5.68+0.1
0.3010.02

6.61
6.18
5.78

9.80
8.80
9.06

DMFA

Water+

353.7
363.0
367.7
383.0

0.9610.12
0.9910.05
1.0210.03
1.33+0.07

1.05
1.09
1.12
1.19

9.38
10.1
9.80
-10.5

Isopropanol

Water+

318.4
328.2
338.0

3.510.4
3.39+0.07
3.0+0.16

3.27
3.20
3.12

-6.57
-5.60
4.00

Methanol

Water+

307.8
317.9
327.9
337.1

1.610.2
1.810.2
1.7+0.25
0.76+0.05

1.63
1.67
1.69
1.71

1.88
-7.22
-0.59
-2.84

Nitromethane Water+

314.3
322.5
331.1
343.4

13.3+0.2
11.8+0.5
11.8+0.8
8.610.4

13.68
12.13
10.73
9.07

2.86
2.80
-9.07
5.47

Water

DMFA+

323.2
333.1

0.89+0.4
0.35+0.26

0.79
0.92

-11.2
-31.9

Water

Methanol*

323
373

1.93
2.75

1.94
2.46

0.52
-10.5

Water

Ethanol*

323
373

5.42
5.21

5.32
5.93

- 1 .85
13.8

Water

n-Propanol*

313
333
369

20.9
16.8
14.3

14.58
15.13
. 15.30

-30.2
-9.94
7.00

Water

sec!-Butanol*

333
367

39.7
32.1

36.74
34.53

-7.46
7.57

* %Dif f erence = (Pred. - Meas . )/Meas .>< 100,

t Measured data is from Ref.[65], $ Measured data is from
R e f .[66].

66

Easy Reversibility

Easy reversibility in this case means low distillation

energy

requirements

solvent that
azeotropic

It

of

is

ease

of

producing

can be recycled to the primary

distillation

vaporization
load.

and

the

column.

solvent will

desirable

that

low

reduce

the

solute-free

extractive or

latent
the

heat

reboiler

solvent-solute

of
heat

system

introduced into the solvent-recovery column do not form an

azeotrope or a close-boiling m ixture.
For

the

distillation

recovery

of

the

entrainer

from

azeotropic

process, Berg made the following comments[12]:

Pressure-swing rectification offers a simple means of

recovering the entrainer from its azeotrope, provided the
vapor pressure-temperature relation is such that the
azeotrope composition varies considerably with pressure.
If a solvent can be found capable of affecting a complete
separation by extraction alone and which is readily
separated from the extract, solvent extraction is an
attractive method of recovering the original entrainer.
The classical process of recovering the entrainer from
azeotropic

distillation

is

decantation

unit

plus

conventional distillation column. Stripping, absorption, and

solvent
recovery.

extraction

are

the

alternative

techniques

for

67

CHAPTER 4

EXPERT SYSTEMS AND SELECTION PROBLEMS

Introduction to Expert Systems

Experts,

as

used

here,

are

people

who

have

become

proficient at something from long-term experience, and whose

knowledge is therefore valuable within its specific domain.
Using decision-making heuristics,
judgements at crucial moments,

experts can make skillful

even in situations where the

information on which such decisions ought ideally to be based

is insufficient.

An expert system is an advanced computer program that

attempts to emulate human expert decision making. Waterman{67]
defined an expert system as:
A computer program that uses expert knowledge to attain
high levels of performance in a narrow problem, area.
These
programs
typically
represent
knowledge
symbolically,
examine . and
explain
their
reasoning
processes, and address problem areas that require years
of special training and education for a human to master.
By
experts,

capturing

and manipulating

the

knowledge

of human

expert systems provide advantages of availability,

consistency, and testability.

The key components of expert

systems and their relationships are shown in Figure 26.

An

Domain Expert
TransferofExpertlse

KnowtedgeEnglneer
Knowledge
Structure

Expert System Shell

External Interface*

Updates
Data Bases
Executable Programs

User Interface
(ConsutatioiVE^Jlanation)

User

Figure 26 Integration of expert system components[67]

69
expert system itself usually contains:
(1)

a knowledge base;

(2)

an inference engine;

(3)

a user interface;

(4)

external interfaces; and

(5)

data bases.

The knowledge base

contains

specific,

in-depth

information

about the problem at hand. That knowledge consists of facts,

rules, heuristics, semi-empirical models, etc.
To utilize the knowledge, an expert system relies on its
inference

engine.

The

inference

engine

uses

inference

mechanisms to process the knowledge and draw conclusions.

The user interface provides smooth communication between
the program and the user.
The external interfaces provide the facilities to connect
an inference engine with databases

and external

executable

programs.
The database is a repository of some facts for the domain
problem.
The

knowledge,

extracted

from

the

domain

expert, is

obtained through the process of knowledge, acquisition by the

system developer,
researchers.

the so-called knowledge engineer among Al

This

knowledge, provides

expert reasoning capabilities.

Al

programming

k n o w le d g e

base,

languages
create

or

The

the

developer

software

the u s e r

system

to

in te rfa c e ,

then

with

utilizes

construct
w rite

its

the

e x te rn a l

H -ti-

70
executable programs
combine
system.

if needed,

them to make
During

the

establish the database,

everything function
earlier

stages

of

as

an

Al

and

integrated

research,

two

programming languages, LISP and PROLOG, were developed to deal

with the symbolic representation and reasoning for Al research
work,

including

expert

system .development This

was

done

because conventional programming languages such as FORTRAN and

PASCAL lack the abilities to efficiently handle qualitative
knowledge. Then, the expert system developer who was equipped
with LISP and PROLOG often devoted too much time to organizing
and

constructing

researchers

the

realized

inference

that

the

engine.

Later

inference part

of

on,

Al

an expert

system is independent of the domain knowledge and therefore

can be separated from the knowledge base. Therefore, a variety
of software tools that contain the inference engine and other
facilities aimed at speeding up expert system development have
become commercially available

for meeting the needs of the

expert system building. Depending on the capabilities of the

development tools, the developer may also construct of modify
the

inference

engine.

Usually,

this

step

is not

necessary

because modern expert system shells provide the developer with

strong inference facilities.

Expert System Shell--LEVEL 5 OBJECT

As shown in Figure 26, expert systems use a combination

71
of

user

interface,

interfaces,

together

handling knowledge.
specific

inference

domain

called
The

the

expert

knowledge

mechanisms,

is

expert

and

system

system shell,
removed,

can

external

shell,

for

in which the
facilitate

the

construction of an expert system in any domain justified for

the expert system development by providing the basic control
structure,

knowledge

representation

methodologies,

and

constructs for the user interface creation. In this work, the

expert

system

Information

shell,

called

Builders,

Inc.

LEVEL
was

OBJECT

used

as

environment for our prototype expert system.

developed

by

development

LEVEL 5 OBJECT

is an application development environment that combines expert

system

technologies,

relational

database

object-oriented
models,

and

programming

graphical

(OOP),

development

and

debugging tools. To provide a graphics-based user interface,

LEVEL 5 OBJECT is run under Microsoft Windows. Table 8 shows
the basic features of LEVEL 5 OBJECT.

Expert System Applications in Chemical Engineering

Expert
financial

systems
management

have
and

been

successfully

planning,

applied

marketing,

to

patient

diagnoses, prospecting ore deposits, and military operations.

Expert
engineering.

systems

have

begun

to

penetrate

chemical

With ,the growing availability of powerful and

inexpensive hardware., using an expert system is now within the

IIM W

72
Table 8. The Basic Features of LEVEL 5 OBJECT.
1 . Platforms
IBM PC (WINDOWS)
2. Knowledge base components
Data types:
Compound; Color; Interval;
Multicompound; Numeric;
Picture; Rectangle;
Simple; String; Time
Knowledge representation:
Rules; Object-oriented data structure
3. Inference Engine
Reasoning mechanisms:
Forward chaining; Backward chaining;
Mixed-mode; Procedural; Object-oriented;
Uncertainty:
Confidence factors

grasp of the practicing engineer. Compared to typical chemical

engineering computing, expert systems are unique. Instead of
"crunching
knowledge,

numbers,"

they

contain

chemical

engineering

i .e ., facts, rules, and heuristics.

Having this

knowledge, they have the ability to reason and solve problems.

Table
systems in

lists

six

main

application

areas

of

expert

chemical engineering identified by Quantrille et

a l .[68] and their individual functions.

It can be seen that

many applications are involved with the selection task.

73

Table 9. Application Areas of ES and Their Functions.

Main area

Application characteristics

Fault diagnosis

Process troubleshooting, i .e .,
determining the origins of process
problems and recommending solutions.

Process control

Monitoring processes and improving

process control.

Process design

Designing processes, developing

flowsheets, specifying equipment,
ensuring process safety, assessing
process flexibility, estimating
cost, selecting and estimating
physical and thermodynamic
properties.

Planning and operations

Scheduling, developing procedures,

assessing safety concerns, executing
complex interrelated procedures, and
aiding maintenance.

Modeling and simulation

Using qualitative reasoning and

symbolic computing to model and
simulate chemical processes.

Product design,
development,
and selection

Recommending chemical formulations,

compositions, materials, process
procedures, selecting new or
existing products that achieve
specified objectives.

The General Theory of Selection Problems

Selection problems are of interest for chemical engineers

and

appropriate

for

expert-system

development.

Selection

itself is a complex problem-solving activity that incorporates

varieties . of

knowledge

and

problem-solving

tasks.

The

74
selection process involves searching in an object space and
testing objects using some constraints. The object space and
the initial requirements, which also form the constraints for
the selection process,

can be very large.

So,

exhaustively

testing attributes of each object for all the constraints is

computationally intensive and can become

intractable for a

complex, real world selection problem.

The result of a selection activity is a set of objects
which

satisfy

associated

both

with

specifications.

the

the
The

goals

definition

and
of

constraints
the

objects

that
and

are

their

inputs to the problem are constraints,

information about the attributes of the objects and knowledge

to be used to control the selection process.
Selection
complicated

problems

depending

can
on

range

the

from

simple

problem-solving

to

very

techniques

employed. For a simple selection problem such as selecting a

proper wrench from a wrench set to tighten a nut, table-look
up may be an adequate strategy.

But for a complex selection

problem such as selecting promising solvents

or entrainers

among a large set of solvents for azeotropic and extractive

distillation based oh both thermodynamic principles and expert
experiences,

such

table-look-up

strategy

becomes

inefficient.
Actually,

selection

processes

are

not

well-defined

problems from an Al viewpoint. A subset of selection problems

are

the

so-called

routine

selection

problems

that

are

Jl

75
discussed thoroughly by Gandikota and defined as follows[69]:
A routine selection problem has clearly stated goals and
well-structured selection procedures to achieve the
goals. The object space and the necessary knowledge for
making appropriate decisions to select from the object
space are well-defined.
The overall task of routine selection can be decomposed
into two

subtasks.

The

first

subtask,

called preliminary

selection uses a hierarchy of conceptual categories. At each

stage

of

the

preliminary

selection

there

is

knowledge

available to decide if a particular conceptual category can

satisfy

some

broad

initial

requirements.

The

result

of

preliminary selection is a set of abstract categories that are

used

as

called

inputs
critical

in

the

second

selection

subtask.

involves

The

second

refining

abstract categories using a selection critic.

the

subtask
set

of

The selection

critic consists of a set of conceptual nodes corresponding to

the tip nodes of the preliminary selection hierarchy.

These

conceptual nodes contain knowledge to test the suitability of

each abstract category to satisfy the full range of general to
very specific initial requirements. The output of the critical
selection

is

set of

selected

categories

ranked

by

some

order.

Selection-related Expert Systems in Chemical Engineering

Generally speaking,

the function of a selection expert

system is to help end users

choose products,

processes,

or

76
something

else,

alternatives.

often

One. of

from
the

large

or

complex

characteristics

of

sets

of

selection

applications is the extensive use of data entry forms and the

need for flexible access to databases. The object information
is often stored in an inventory database, and the end user is
prompted for desired features and constraints. The knowledge
base is then responsible for matching the constraints to the
possible alternatives in the database.
It can be
selection

seen

from the

processes

gain

following brief

much

attention

summary that
in

chemical

engineering.
Separation design.
gained

much

attention

Separation sequence syntheses have

among

chemical

engineers

as

Al

applications. The well-known systems for separation sequence

synthesis

such

as

EXSEP

developed

by

Liu

et

a l .[70],

an

adaptive heuristic system developed by Huang et a l .[71] , and

SAD developed by Barnicki and Fair [72]
the

need

to

select

distillation,

proper

[73] have in common

separation

method

adsorption or crystallization

such

at the

as

initial

stage of sequencing.
Thermodynamic model selection. CONPHYDE (CONsultant for
PHYsical

property Decisions),

University

by

developed

Banares-Alcantara

[74],

at

Carnegie-Mellon

is the

first

expert

system that selects the appropriate physical property model

for

vapor-liquid

interactive

equilibrium . calculations.

"question

and

answer"

session

to

It

uses

an

acquire

the

77
information

necessary .to

recommend

a model.

CONPHYDE

uses

rules and statistical inference techniques to draw conclusions

and make

recommendations.

What CONPHYDE

lacks,

however,

is

depth in its knowledge base. Gani and O'Connell[75] have also

developed

knowledge-base

system

for

the

selection

of

thermodynamic models. Their system contains deeper knowledge

than CONPHYDE does. In addition, they use a different problem
solving approach from that of CONPHYDE. Gani and O'Connell's
system gathers the knowledge about model selection and employs
a large index table, in which the best model is simply to be
looked up. Chen and Chang have also developed an expert system
(ESPEM) [76] for assisting in selecting suitable vapor-liquid
equilibrium thermodynamic models. In addition to giving advice
on

the

selection

of

supplies

an

numerical

calculation

proper

intelligent

thermodynamic

database

facilities

models,

manager

and

ESPEM

provides

to give the thermodynamic

properties needed for subsequent applications such as process

design.
Product or equipment selection. CAPS

[77] is an expert

system for plastic selection. The user is questioned about the

intended

use

of

the

final

product,

and

classifications of polymers to consider.

CAPS

recommends

CAPS is a forward

chaining rule-based system. DECADE (Design Expert for Catalyst

Development)[78][79] is a prototype expert system developed at
Carnegie-Mellon University for catalyst selection by BanaresAlcantara.

On

the

chemical

engineering

side,

DECADE'S

78
knowledge base is limited.

PASS

(Pump Application Selection

System), by Venkatasubramanian[80],

is an expert system for

the selection of a pump from one of the following categories:

rotary-screw pumps, rotary-gear pumps, positive displacement
diaphragm
piston

pumps, displacement-plunger

pumps,

and

dynamic

pumps, displacement-

centrifugal

pumps.

PASS

is

forward-chaining rule-based expert system. It uses production

rules

that

Hanratty

et

"fire"

when a given

a l .[81]

developed

set of information
an

expert

system

exists.
for

the

selection of an appropriate laboratory reactor for collecting

the experimental data for such diverse purposes as catalyst
screening, scale-up, troubleshooting, and kinetic modeling.

79

CHAPTER 5

KNOWLEDGE BASE

An expert system,
was

designed

to

named SSS

utilize

some

(Solvent Selection System)

of both

the

qualitative

and

quantitative knowledge described in Chapters 2 and 3 to select

promising solvents for extractive and azeotropic distillation.
SSS is intended to give both the expert and nonexpert
quick answers to the problem of solvent selection. The first
phase for the development of an expert system involves the
knowledge

acquisition,

that

is,

garnering

the

information

related to the problem domain at hand from both the literature

and experts in the field. The second phase is concerned with
the knowledge representation approaches which are based on
knowledge characteristics and reasoning strategies applied in
the system. This chapter discusses these two phases in detail.

Knowledge Acquisition

Knowledge Involved in the Solvent Selection Problem

The knowledge acquisition in this work was conducted in
tw o

p a rts:

(I)

g en eral

k n o w le d g e

a c q u isitio n

in

o rd er

to

80
obtain an understanding .of the problem domain and (2) specific
knowledge

acquisition

to

determine

the

solvent

classes

on

which the selection is based and obtain detailed knowledge

about each solvent.
The objective of the general knowledge acquisition was to
obtain

an

overview

of

the

problem

domain.

With

this

understanding, a knowledge structure for the selection process

was developed leading to a system knowledge hierarchy that is
shown in Figure 27. The information for the general knowledge
acquisition
(textbooks

phase
and

was

obtained mainly

publications)

and

from the

through

literature

interviews

arid

discussions with Dr. Lloyd Berg who has worked in this field
for several decades.
The

knowledge

related with

the

selection

of

solvents

falls into four categories:

1) Separation methods.

From the foregoing overview, it

is known that the procedure and rules for the selection of

solvents

vary

for

different

separation

methods

such

as

azeotropic distillation, extractive distillation, and solvent

extraction.
2) The specific problem knowledge -- the characteristics
of the mixture to be separated.

The separation method which

should be used usually depends upon the characteristics of the

mixture to be separated. Also, the solvent selection process
is

based upon

both the

characteristics of the mixture and

the physico-chemical properties of each component in the

81

Separation,
IMethods

-Azeotropic Dist.
Extractive Dist.
Solvent Extra.
Bulk Attributes
1Thermal
Attributes

- Dilute Solution
Concentrated Solution
- Thermal Sensitive
- Thermal Stable
High Boiling
Normal Boiling
Azeotropic Mixture

Min. Azeotrope
. Max. Azeotrope

Zeotropic
iAttributes

Mixture
Characteristics

Solvent
Selection'
System

Close-boiling Mixture :
Hydrocarbon

.Chemical
Family

Aqueous
Nonaqueous
. Common Properties

i Physical
Properties

Solubility Parameters
Solvatochr. Parameters

Solvent
__ _
Characteristics

Chemical.
Family

Miscellaneous
Properties

Interaction

. Molecular Volume
Boiling Point
Melting Point
Heat of Vaporization
Critical temperature
Dispersion
Polarity
Hydrogen Bonding
Basicity
Acidity

Monoalcohol
Polyalcohol
-K eto n e

Hydrogen Bonding

I Physical
Properties

Homolog
Isomer
Aliphatic,open chain
Aliphatic,cyclic chain
Aromatic

Polar degree

-Strong
- Moderate
Poor
Strong
Moderate
Poor

Common Properties
Solubility Parameters
Solvatochr. Parameters
Corrosiveness
Toxicity

Selectivity
Solubility
Volatility
Recovery
Reaction

Figure 27 Knowledge involved for solvent selection problems

82
mixture.
3) Physico-chemical properties of solvents. The preferred
solvent

(from a solvent database)

should be selected for a

specific separation problem, and is largely justified by its

physico-chemical properties such as hydrogen bonding ability,
polarity, melting points, boiling points, and so on.
4) The interaction of the solvents and solutes.

The goal

of adding a solvent into the mixture to be separated is to

change the vapor-liquid equilibrium behavior of the mixture.
The changes made in the vapor-liquid equilibrium by adding the
solvent are good criterion for the selection of a solvent.
During
process,

the

second

part

of

the

knowledge

acquisition

the knowledge about the search space was gathered.

The search space for this problem consists of solvents and

specific

information needed for the

including

some

experience.

This

physico-chemical
knowledge

comes

selection of

properties
mostly

and

from

solvents,
practical

the

public

literature[ 82-86 ] . When particular physico-chemical properties

are not available, the group contribution method is utilized
to obtain an estimate.

Knowledge Used in the Current Expert System

The knowledge involved in the solvent selection problem
as described in the above section contains a large amount of
information.

If energy saving or economic factors are taken

into account, much more knowledge may be needed for tackling

83
the problem.
Table 10 shows the progress of the knowledge for solvent
selection for azeotropic and extractive distillation based on
the information provided in chapter 2.

Table 10. Progress of the Knowledge for Solvent Selection

for Azeotropic and Extractive Distillation.
Azeotropic Distillation

Extractive Distillation

Sheibel1s rules(1948)
Berg's heuristics(1969)
Tassio1s procedure(1972)
K o l b e 1s procedure(1979)
Treybal1s criteria(1980)
Y e h 1s polarity p l o t (1986)

Ewell'S' heuristics (1944)

Berg's criteria(1969)
Treybal's criteria(1980)
Stichlmair'sheuristics(1989)
Doherty's residue curve (1991)
Laroche's procedure(1992)

It should be noted that not all the knowledge described

in Figure 27 and Table .10 is built into the knowledge base in
the current prototype expert system.
In S S S , the rule proposed by Dr. Lloyd Berg
below)

(described

is used to select a separation method.

If the boiling point of any component in the mixture to

be

separated

is

above

170oC ,

then

prefer

azeotropic

distillation. Otherwise, the end user inputs either azeotropic

distillation

or

extractive

distillation

as

separation

method.
This

rule

was

also

suggested

by

Van

Winkle [27].

The

reasoning behind using this rule is that we can lessen thermal

decomposition,

polymerization,

or

other

t h a t m ig h t happen a t a h ig h e r te m p e ra tu re .

'I

chemical

reactions

84
For azeotropic distillation,
are

used

for

solvents

to

azeotrope,

illustration.
separate

only Berg's two rules[11]

Those

rules

closeboiling

govern

mixture,

selecting
a

maximum

and a minimum azeotrope by forming a new binary

azeotrope.

At

present,

the

knowledge

for

azeotropic

distillation has not been exploited very deeply due to time

limitations.
For extractive distillation,

it is well known that the

solvent should be highly selective, and of low volatility. As

described
criteria

in
for

combinations

chapter
selecting
of

characteristics
hydrocarbon
volatile),

2,

solvents.

rules
of

and

the

mixtures,

authors,

to

boiling

Tassios1s procedure,

proposed

In SS S , we

procedures

mixtures

Berg's

have

use

according
be

point

different
different
to

the

separated.

For

rule

(or

high solvency criterion

less
(or

high capacity), and high selectivity criterion are used for

solvent selection. For aqueous systems, chemical families with
high polarity are chosen, then Berg's boiling point rule is
applied,

Kol b e 's procedure is used to select solvents with

high selectivity. For a nonhydrocarbon, nonaqueous mixture, we

use Berg's rule to classify solvents into 5 classes, of which
four classes with high or medium hydrogen bonding ability are
chosen. Members of homologs of each component in the mixture
are chosen (Sheibel's rule). After that, Berg's boiling point
rule is applied. Then solvents are selected based on a high

selectivity criterion.

85
Table

11

lists

the

rules

and procedures

used

in the

current expert system.

Table 11. Rules Used in the Current Expert S y s t e m . _______

1. Rule for method selection
Berg's recommendation (also Van Winkle heuristic)
2. Azeotropic distillation
Berg's two rules (form a new binary azeotrope)
3. Extractive distillation
A. Hydrocarbon mixtures
Berg's boiling point rule (less volatile)
Tassios's procedure
High solvency (high capacity)
High selectivity
B. Aqueous mixtures
Chemical families (high hydrogen bonding ability)
Berg's boiling point rule.(less volatile)
Kol b e 's procedure
High selectivity
C. Nonhydrocarbon nonaqueous mixtures
Berg's 5 classes of solvents
(select solvents with high hydrogen bonding ability)
Members of homologs (Sheibel's rule)
Berg's boiling point rule (less volatile)
High selectivity

The high capacity criterion proposed by Treybal is not

always used in the current system because, for many mixtures,
calculation

of

solubility

inaccurate. .Treybal

also

is

listed

often
the

time

consuming

additional

required for solvents such as low cost,

and

attributes

low toxicity,

low

86
corrosiveness, chemical stability, low freezing point, and low
viscosity. In the present expert system, these criteria have
not been considered because data for the above properties are
not available for many solvents in the database.
One of .Sheibel's rule requires that the selected solvent
not form an azeotrope with any component in the mixture. If a
solvent boils 20C higher than components do, the chances of
forming an azeotrope are rare.

Types of Knowledge

There is a difference between theoretical and real-world

knowledge.

We

acquire

the

former by

studying

the

relevant

theory, which is under the form of general laws and axioms.

The latter is not completely acquired through theory; one also
needs to gain insight into problems that may arise in actual
situations
order

to

Therefore,
types

in

and

so

become

acquire

an

knowledge
knowledge

expert
has

problem-solving
in

given

been broadly

engineering

field:

capabilities

real-world
classified
Deep

and

in

domain.
into two
Shallow

Knowledge.
With Deep Knowledge, we mean the set of theoretical laws
and axioms that form the basis for abstraction capability, and
mathematical models developed to describe the behavior of the
domain system.

school.

It is the same kind of knowledge we learn at

87
Shallow
knowledge
mentors

Knowledge

will

to

not

be

provide

defines

all

acquired

other

knowledge.

from books,

insightful

suggestions,

but

will

Such
need

experience

to

acquire problem-solving capabilities, and intuition to tackle

unknown

situations.

Shallow

Knowledge

is

also

called

expertise, referring to the unique skills of an expert in a

given domain. The outcome of Shallow Knowledge (or expertise)
is

also

generally

referred to

as

heuristics

(or heuristic

knowledge). This term refers to the so-called rules of thumb,

meaning the judgment capabilities of the expert to grasp the
meaning of a given situation and to foresee the consequences
even when lack of data or the complexity of the domain do not
permit a correct or complete theoretical analysis of the
problem. Figure 28 sums up the knowledge components.
Knowledge, whether shallow or deep, must be extracted and
encoded into usable forms in order for an expert system to
solve problems. Knowledge extraction is performed to produce
the

organized,

experiences.
Before

indexed

These

encoding,

categories[87]:

are

deep

knowledge

then

encoded

knowledge

may

declarative

be

and
into

heuristics
the

Al

subdivided

knowledge

and

from

program.
into

two

procedural

knowledge. Each kind of knowledge has one or more associated

knowledge representation techniques.
Declarative

knowledge

refers

to

facts

and

assertions

while procedural knowledge refers to sequences of actions and

consequences. Declarative knowledge is associated with knowing

Knowledge Assembly

Shallow Knowledge

Deep Knowledge

Axioms

Theoretical

Mathematical

Laws

Models

Facts

Rules

Figure 28 Knowledge components and classification

Heuristics

89
what is involved in solving a problem. Procedural knowledge
isassociated with knowing how to apply appropriate problem
solving strategies to solve a given problem.

Knowledge Representation Strategy

The representation of knowledge is a basic issue in the

development of Expert Systems (ESs) for applications. In fact,
the performance of an ES relies heavily on the adequacy of its
knowledge base.
Al researchers have developed several different knowledge
representation techniques to represent the knowledge acquired
from the expert in expert systems while preventing ambiguities
in

the

problem

solving

procedures.

Some

techniques

are

suitable for a majority of problems typically encountered by

expert systems. There are, however, some problems that require
unique

knowledge

representation

knowledge

representation

networks,

(2)

Frames,

approaches.

techniques[87]

(3)

Production

The

are

rules,

(I)
(4)

major

Semantic
Predicate

logic, (5) Scripts, and (6) Object-oriented. Since the expert

system

shell

provides

us

which

we

w i t h . the

have

chosen

application

as

our

developing

development

tool

environment

through both rules and object-oriented programming (OOP), we

focus

our

discussion,

on

knowledge

object-oriented approach and rules.

representation

by

the

90
Implementation Language Object-oriented Programming
Object-oriented programming has
that

recently

can

assist

emerged as a

programming

methodology

in

software

development.

The object-oriented approach has been gaining

acceptance in the software community in part because of its

representational
approaches,
that

act

capabilities.

In

traditional

programming

a program is made of a collection of functions

upon

data;

this

data may be

organized

into data

structures. A problem is solved by repeatedly applying these

functions. By contrast, the object-oriented approach models a
domain as a collection of computational entities, referred to
as objects.

These objects define operations that are invoked

as a result of the

interaction of these objects with each

other. In other words, these objects contain not only a list

of properties

(data), but also

the

required procedures

to

manipulate them. In this approach, a problem is solved through

the

interaction

of

the

objects

that makes

up

the problem

domain.
One better-known prototype research program in chemical
engineering application that takes an object-oriented approach
is DESIGN-KIT by Stephanopoulos et a l .[88]. DESIGN-KIT is a
software-support
activities

such

configuration,

tool

developed

as:

flowsheet

and

operational

to

aid process-engineering

development,
analysis.

It

control-loop
is written

in

Common LISP, and is built on Intellicorp's software package,

KEE (Knowledge Engineering Environment).

u.

91
Another application in chemical engineering that takes an
object-oriented approach
Constantinou

et a l .[89]

is the
for

expert system developed by

estimation of physico-chemical

properties from molecular structure.

It should be noted that application areas that have the
following characteristics can benefit from an object-oriented
knowledge-based system development[90]:
Highly
complex

Applications
that
have
many
interrelated and interdependent rules or constraints that
are
typically
beyond
the
scope
of
conventional
programming systems, or those that reguire very large
maintenance efforts.
Frequently changing
Applications where policies,
information, and regulations change frequently. These
type of applications can cause major maintenance problems
if built with conventional tools.
Requiring consistency Applications where the rules
used to address cases are subject to inconsistent
interpretation. Codifying the rules ensures that they
will be interpreted consistently throughout the problem
domain.
Requiring expert knowledge Applications that deliver
expertise
to
people
who
do
not
have
it.
These
applications make an expert's opinion and knowledge
available to a wide audience and provide tracking of the
decision-making process.
Time-critical Applications, such as process control,
which must rapidly apply decision-making knowledge or
rules, or quickly identify patterns due to on-line, timecritical constraints.
Basic application areas that possess some of the above
characteristics

include

documentation,

decision

management,

intelligent

diagnosis/troubleshooting,

scheduling/constraint analysis, configuration/design, process

monitoring

and

control,

and

selection/classification.

Our

92
application falls into selection/classification category.

Features of an Object
Although no formal definition is available, an object in
a

full-fledged,

object-oriented,

multiparadigm

programming

environment is often found to exhibit the following features

[?1 ]:
1. Objects are divided into two categories:
instance. A class

class and

is a description of. one or more similar

objects sharing attributes; an instance is a specific example

or instance of a class.
2. An object class or instance is uniquely identified by
a name.
3. An object manifests itself as a frame, accommodating
slots as the frame's attributes. Slots in an object are not
ordered.
4. The domain of a slot is specified by facets

(slot's

attributes). The value for a slot could be an atomic value, a

list,

or

variable

pointing

to

another

data

structure,

depending on the domain being defined.

5. Slots are usually defined in a class;
inherited

across

the

class-subclass

and

they will be

class-instance

hierarchies. Various inheritance mechanisms can be specified

by slot facets.
6. Procedures,

known as methods, can be defined in an

object to manage the data stored in slots. Objects interact by

93
calling each other's methods through the mechanism of message
sending.
7.

Additional computations may be activated when a slot

is created or deleted or when its data are fetched or stored

by attaching active values, or demons, to the slot.
It

should

programming

be

emphasized

technique

is

that

the

discipline

object-oriented

tedious

to

master.

Fortunately, LEVEL 5 OBJECT provides a user-friendly interface

to facilitate the manipulation of information in a complex
object world.
Figure 29 delineates the structural breakdown of a LEVEL
5 object.
for

The LEVEL 5 class declaration provides a template

classifying

and

organizing

knowledge

within

an

application. A class describes the structure and behavior of

an

object

within

structure,

file,

database.

an

application,

is

characteristics

called

database

or

rules,

demons,

to

how

a record

or table defines the basic structure of a

class

fields

similar

defined

attributes,

columns.

and methods,

by

which

Behavior,

collection
are

defined

similar
by

of
to

facets,

is bound to the attributes.

An

object-oriented system therefore binds structure and behavior

in a single object

in order to

facilitate maintenance and

promote rapid development.

Both classes and their attributes are declared from the
Objects Editor provided by LEVEL 5. For example, the Object
Editor screen in Figure 30 shows the class domain with the

H Q m c-OT O

Figure 29 Illustration of object concept[90]

95

Objects

Class
new
X 7
V r

Attribute
Instance
Classes
dB3 SOLV_SUB1
domain

new

Instances

Zx the domain
[ Str ] comp_name[1] undeterm
[ N I crit_temp[1]
undeterm
[ N ] mol_vol[1]
undeterm
[ N ] heat_vap[1 ]
undeterm
[ N ] number of Components un
[ N ] boil_point[1]
undeterm
[ Str ] subdiv[1 j
undeterm
[ Str ] dassnam e[i ]
undeterm
[ Str ] moLformula[1] undeterm
[ N] i
undeterm
[ S ] ok
undeterm
[ S ] check existence undeterm
[ S J continue
undeterm
[ S ] input feed info undeterm
[ C ] feed category
undeterm

Facets

View

C L A S S domain
[ Str ] oomp_name
[ N ] criLtemp
[ N J mol_vol
[ N j heat_vap
[ N j number of Components
[ N ] boil_point
[ Str ] subdiv
[ Str j dassnam e
[ Str ] mol formula
[ N] i
[ S j ok
[ S I check existence
[S j continue
[S 3 input feed info
[C ]feed category
hydrocarbons
aq nonhydrpcarbs
nonaq nonhydrocarbs
[ N ] composition
[ N j component number

Figure 30 Object "domain" for the driving module

11

96
attributes comp_name, crit_temp, mol_volr heatvap, number of

components, boil point, subdiv, classname, mol_formula, i, ok,

check existence, continue,

input feed info, feed category,

composition, component number in MAIN.KNB that is a drive

program for the whole system.

A class declaration remains an

empty structure until values are added to its attributes.

In

order to add values, specific occurrences of the class must be

created.

An

occurrence

of

class

is

called

an

object

instance.
Using the database analogy, if a class is similar to the
file

structure

or

table

that

defines

database,

then

instances are similar to records in the file or rows in the

table. LEVEL 5 creates both temporary and permanent instances.
A. temporary instance occurs during a knowledge base session
when the values of attributes are pursued,

as, for example,

during rule processing. The lower left part of Figure 30 also

shows a temporary instance for domain class. When the session
ends,

any temporary instances no longer exist.

A permanent

instance can occur during development time as a result of an

instance declaration. Instances can be declared and edited in
the Objects Editor of LEVEL 5 OBJECT.

Procedural Knowledge

Rules and Methods Representation

Procedural knowledge is compiled knowledge related to the

performance of some tasks. The algorithmic knowledge to rank
the entrainers and heuristic knowledge for the selection of

97
entrainer will be utilized in our system.
knowledge used to make good judgements,
tricks, or "rules of thumb"

Heuristic is the

or the strategies,

used to simplify the solution of

problems. Heuristic is usually acquired with much experience.

The guidelines for selection of entrainers,

reviewed in the

section of History of Entrainer Selection Methods, have been

used largely as procedural knowledge.

Those guidelines have

been converted into the IF...THEN rule format for the expert
system.

The most common form of architecture used in expert

systems

is rule-based knowledge.

This

type

of

system uses

knowledge encoded in the form of production rules, that is,

IF*..THEN rules.

The rules have an antecedent or condition

part, the left-hand side, and a conclusion or action part, the

right-hand

side.

Each

rule

represents

small

chunk

of

knowledge relating.to the given domain of expertise. A number

of related rules collectively may correspond to a chain of
inferences that lead from some initially known facts to some
useful . conclusions.

When

the

known

facts

support

the

' f"

conditions in the rule's left side, the conclusion or action

part of the rule is then accepted as known.
As

shown

in

Figure

29;

values

can

be

assigned

to

attributes of an object by rules, methods, and demons that are

means for the representation of procedural knowledge. LEVEL 5
OBJECT provides us with flexible utilities to edit rules and
methods.

follows:

The general

form of a LEVEL 5 OBJECT rule

is as

98
RULE <name>
IF
<condition> .
AND <condition>
OR <condition>
THEN <condition>
ELSE <alternate conclusion>
LEVEL

OBJECT'S

conditions

are

connected

with

the

keywords AND or OR. All conditions joined by AND must be true

for the rule to fire. OR can be used to combine several rules
with

the

same

conclusion

into

single

ru l e . Figure

31

displays a representative rule in our system using LEVEL 5

OBJECT syntax.
Procedural

knowledge

can

also

be

represented by WHEN

NEEDED methods (for backward reasoning) or by WHEN CHANGED

methods

(for

CHANGED)

forward

method

reasoning).

contains a

A . WHEN

block of

NEEDED

(or

WHEN

developer-defined

procedural statements which the backward-chaining (or forward

chaining)

inference engine uses to determine an attribute's

value. Figure 32

gives an example of a WHEN CHANGED method

for this system.

Search Space
The
space

Solvent Database and Related Objects

selection process

and

testing

objects

involves
by

using

searching in an object
some

constraints.

The

object space and the initial requirements, which also form the
constraints
Therefore,

for the

selection process,

can be very

large.

the object space should be established first for

the selection problem.

99

Rules
O

Methods/Rules/Demons
Edit

Selection

Classes

XT- Berg c la s s I
X Z Berg c la s s Il
X Z Berg c la ss III

Lists
View
O All Rules
v for primary selected
v for reselected
v for sheibal's rule applied

X Z Berg c la s s IV
X Z Berg c la ss V
X Z critical selected

new instances

S a v e N ew O >

for s h e ib a l 1S rule a p p lied

FIND solvent list

WHERE ( solvent Ilstd a ssn a m e - classnam e[1]
OR solvent Ilstd a ssn a m e = c la ssn a m e [2 ])
AND solvent llst.com m onnam e <> com p_nam e[1]
AND solvent Iistoom m onnam e <> com p_nam e[2]
WHEN FOUND
MAKE sheibai solvent list
WITH d a s s n a m e := solvent Iistd a ssn a m e
WITH com m onnam e := solvent Ilst com m onnam e
WITH m olform ula:solvent list.molformula
WITH m elLpoint := solvent list.melt_point
WITH boiLpoint := solvent list.boil_point
WITH m oLvol := solvent list.moLvol
WITH refraindex := solvent Iistrefraindex
FIND END
THEN sheibal rule applied

Figure 31 An example of creating a rule

100

Methods/Rules/Demons

Methods

Edit

Classes

XZ dB3 SOLV_SUB1
x T r dom ain

Selection

O
V

Lists

View

All when changed methods

check existence

V continue
V
y

category
Input fe e d Info

fe e d

Save N e w < i>

c h e c k e x is t e n c e

WHEN CHANGED

new instances
A

th e dom ain

BEGIN
comp_name[1
LOWCASE( comp_name[1])
com pjnam e[2]L O W C A SE ( comp_namej2])
comp_nam[3
L0WCASE( comp_name[3])
com p_nam e[4]LO W CA SE( com pjnam e[4])
i 1
WHILE (I <= number of com ponents)
BEGIN
FIND dB3 SOLV_SUB I
WHERE comp nam e [ I ] - dB3 SOLVSUB I .comp name
LIMIT I
OR ccm pjnam e[ I ] = dB3 SOLV_SUB l.commonname
component number [ I ] : - 1
WHEN FOUND
bolLpolnt 11 ] := dB3 S0LV_SUB I .bolLpolnt
crlU em p [I J := dB3SOLV_SUB I. crlt_temp
moLvol [ i I
dB3 SOLV_SUB 1 .moLvol
heat_vap [ 11 := dB 3 SOLV_SUB I .h e a tv a p
dassnam e [ I ] : - dB3 SOLV_SUB.dassname
subdiv [ i ] : - dB3 SOLV_SUB I .subdiv
moUormula [ I ] : = dB3 SOLV_SUB.mol_formula
WHEN NONE FOUND
ASK show property input
FIND END
record OF dB3 SOLVI I

I:= 1+ 1
END
FORGET nam e composition window
name composition window.visibte := FALSE
main window.full name TRUE
main window.output := feed property
ASKfesd property
END

Figure 32 An illustration of creating a method

101
As a matter of fact,

hundreds of thousands of solvents and

their combinations exist which are suitable for extractive and

azeotropic distillation applications. Nonetheless, there are
just

few

separation

solvents
task

that

from

the

are

promising

viewpoint

for

of

both

specific

separation

feasibility, and economic optimization.

If

solvents

azeotropic

are

selected

distillation

task

for a special
through

extractive

experimentation,

or
one

should have many solvent alternatives to try. Similarly, for

an

expert

solvents

system
upon

prototype

to

which

expert

work,
the

system,

first,

we

selection

need

task

it is unwise to

is

database

based.

of.

For

include too many

solvents in the database since manipulating the database will

consume

great

reasoning

deal

process.

of

computer

Too

few

resources

solvents,

and

slow

however,

the

would

inadequately test the expert system and give no solution for

many problems. One should, then, make a compromise regarding
the number of solvents to be included in the database.
choice of which

The

solvents and what kind of properties of a

solvent should be included in the solvent database was based,

in this work, upon whether the solvent is used often in the
chemical

industry,

whether

the physical

properties

of the

solvent are easy to find from the literature or to estimate by .

available
needed
system,

methods,

for
and

the
the

whether

those

solvent-solute
suggestions

physical

interaction
given

by

properties

are

models

the

experts

in

who

have

102
experience, in this field.
Appendix

lists

part

of

the

solvent

list

for

the

separation of hydrocarbon mixtures by extractive distillation

selected
gives

the

from Industrial
solvent

list

Solvents Handbook [ 8 2 ] Appendix B

recommended

by

Dr.

Lloyd

Berg

for

extractive distillation. The solvent list in the database for

azeotropic distillation
i n [92].

The

number

of

is

from Horsely1s Azeotropic

the

solvents

and

the

Data-

scope

of

azeotropic systems is very limited in the present prototype,

since

creating

large

database

is

time-consuming

and

routine task. The current objective is to create a prototype

system

as

a template.. The

extension

of

the

small

solvent

database to a large one will be a part of future work.

The physico-chemical properties of a solvent form some
constraints on which the solvent selection process is based.
Therefore, these properties should be contained in the solvent
database. Table 12 gives a typical physico-chemical property
list

for methanol.

Most of these properties

come

from the

physical property handbooks, and published articles[82-86].

The

Joback

modification

of

Lydersen1s m e t h o d [86]

was

utilized for the estimation of critical properties of some

solvents that cannot be found in the literature. The estimated
critical temperatures of some solvents are listed in Table 13.
The solvents are classified
based

on

their

functional

into ten major categories

groups.

Subdivisions

are

made

according to the nature of the hydrocarbon residue such as

103
saturated,

aromatic, unsaturated;

open chain,

cyclic chain;

the number of times the characteristic atom or group occurs;

and the primary,

secondary,

or tertiary

nature

of amines.

Table 14 shows the solvent classification in detail.

Table 12. The Physico-Chemical Properties of Solvent Methanol.

Class name:
Molecular formula:
Molecular weight:
Melting point: (0C)
Boiling point: (0C)
Molecular volume: (cm3/mol)
Refractive index at 20 0C:
Critical temperature: (0C)
Heat of vaporization: (kcal/mol)
Subdivision:
Antoine constants:
A
B
C
Solubility parameters:
Dispersion (MPa1/2)
Polarity (MPa1/2)
Hydrogen bonding (MPa1/2)
Acidity

monoalcohol
CH40
32.04
-97.68
64.70
40.70
1.32841
239.43
8.4330
aliphatic, open
7.25164
1608.39
-31.070
15.10
12.30
22.30
0.62

Table 13. Critical Temperatures Estimated by Group Methods.

Compound
Vm at 25C
Tb
T
______________________________ (cm3/mol)____ (0C)_______ (0C)_____
Sulfolane
Dimethyl Sulfoxide
2-Pyrrolidinone
I-Methyl-2-Pyrrolidinone
Benzyl Alcohol
Methyl Salicylate
1,3-Butanediol
Propionic Anhydride
Ethyl Benzoate
Propyl Benzoate
Benzyl Benzoate
Benzyl Acetate

94.9
71.0
76.9
96.4
103.6
129.1
90.9
131.5
144.4
160.5
189.8
143.0

287.2
189.0
250.5
202.0
205.0
233.3
207.5
169.0
212.4
231.2
323.5
215.5

581.80
444.96
519.26
425.16
384.42
436.00
336.65
339.08
391.60
399.30
495.60
395.80

104
Table 14. Solvent Classes Based on Chemical Function Groups.

1
2

Water
Hydrocarbons
2.I saturated aliphatic
2.1.1 open chain
2.1.2 cyclic chain
2.2 aromatic
2.3 unsaturated aliphatic
2.3.1 open chain
2.3.2 cyclic chain
Hydroxy compounds
3.I monohydric
3.1.1 aliphatic
3.1 .1 .1 open, chain
3.1.1.2 cyclic chain
3.1.2 aromatic
3.1.3 phenols
3.1.4 unsaturated
3.1 .4.1 open chain
3.1.4.2 cyclic chain
3.2
polyhydric
Ethers
4.1 monogroup
4.1.1 aliphatic open chain
4.1.2 aliphatic cyclic chain
4.1.3 aromatic
4.2 polygroup
4.2.1 acetal
4.2.2 dioxane
4.2.3 ether alcohol
Carbonyls
5.1 aldehydes
5.1.1 saturated aliphatic aldehydes
5.1.2 aromatic aldehydes
5.1.3 unsaturated aldehydes
5.2 ketones
5.2.1 aliphatic ketone
5.2.2 aromatic ketone
Acids
6.1 aliphatic acids
6 .1.1 saturated acids
6 .1.2 unsaturated acids
6.2 acid anhydrides
6.3 aldehyde ethers

105
Table 14. Solvent Classes Based on Chemical
__________ Function Groups continued.
7. Esters
7.I monocarboxylic acid
7.1.1 saturated aliphatic
7.1.2 unsaturated aliphatic
7.1.3 aromatic
7.2 dicarboxylic acid
7.3 polybasic acids
1 .4 hydroxy acids
8. Halohydrocarbons
8.1 monochlorinated
8 . 1.1
8 .1.2
8.2 polychlorinated
8 . 2.1
8 . 2.2
8.2.3
9. Nitrogen compound
9.1 nitroparaffins
9.2 nitriles
9.2.1
9.2.2
9.2.3
9.3 amines
9.3.1

9.3.2
9.3.3
9.3.3
9.3.4

aliphatic
aromatic
aliphatic
aromatic
unsaturated

aliphatic
aromatic
unsaturated
primary monoamines
9.3.1.1 primary aliphatic
9.3.1 .2 aromatic
9.3.I .3
primary unsaturated
primary polyamines
secondary monoamines
tertiary monoamines
amino ethers

9.4 amides
10. Sulfur compounds

The solvent database files are built based on dBASE III

Plus. LEVEL 5 OBJECT is able to access these solvent database
files by the database management system that controls the
interaction between all objects in the knowledge base. This is
implemented through a built-in object called "dB3" in LEVEL 5
OBJECT that can extract the information from the dBASE III

106
Plus files and transfer it into a hybrid object that contains
the built-in attributes and developer-assigned attributes.
In other words, LEVEL 5 OBJECT has an interface to communicate
between dBASE III plus and the knowledge base. Figure 33 shows
the relationship between the built-in object
hybrid object
called

"dB3

solv sub" . This type of relationship is

inheritance

environment.

"d B 3 " and the

in

the

object-oriented

In the above example,

programming

the child class object,

"dB3 solv_sub", inherits all the attributes which the parent

class object dB3 possesses.

Structuring Knowledge

The solvent selection problem consists of the search and

selection

of

A l /knowledge

components. The

important

engineering viewpoint are

questions
(I)

from

an

can the search-

selection problem be decoupled or decomposed into tractable

subproblems
represent

and

and

(2)

what

structure

is

the

the

most

knowledge

effective
required

to

way

to

select

suitable solvents.
When designing the structure of the expert system, the
goal

is

strategy
possible.

to

mimic

of

the

Experts

as

much

expert,
employ

as

possible

although
many

the

problem-solving

other

strategies

are

different

strategies

and

variations in problem-solving. Even when the problem-solving

strategy is decided upon, there are still many different ways

107

Class dB3
access COMPOUND
read,
write,
read shared,
write shared
action COMPOUND
advance,
append record,
insert record,
delete record,
recall record,
close,
open,
pack
eof SIMPLE
record NUMERIC
size NUMERIC
index file NUMERIC
file nam e STRING
active SIMPLE

Parent Class

Child Class

I
Class dB3 SOLV SUB I
ac ce ss COMPOUND
read,
write,
read shared,
write shared
action COMPOUND
advance,
append record,
insert record,
delete record,
recall record,
close,
open,
pack
eof SIMPLE
record NUMERIC
size NUMERIC
index file NUMERIC
file nam e STRING
active SIMPLE_______ Z
classnam e STRING
com p nam e STRING
com m dnnam e STRING
molformula STRING
mol weight NUMERIC
melt point NUMERIC
boil point NUMERIC
mol volume NUMERIC
crit tem p NUMERIC
heat vap NUMERIC
subdiv STRING

Figure 33 A n Example of Inheritance

Inherited
Attributes

New
Attributes

108
to approach the design

of the

structure

for the knowledge

base. In our system, a task-oriented expert system design is

adopted. The task-oriented approach to expert system design
represents a strategy of explicit knowledge organization. This
method

is

based

on

the

following

premises

stated

by

Chandrasekaran[93]:
(1)

complex problem

can

be

decomposed

in terms

of

"generic tasks." A large problem may be composed of scores of

interrelated tasks.
(2)

The

domain

knowledge

is

available

to

encode

into

blocks of knowledge, each of which solves a single task.

(3) The tasks can be built into a structured hierarchy
that solves the overall complex problem.
A problem decomposed in this manner can be thought of as
a group of

"specialists", each working on a separate task.

Tasks at the upper levels of hierarchy are more abstract in

nature, while those at the lower levels are more concrete. The
behavior

is

reflected

in the

expert

who

focuses

first

on

broader issues in the problem and delays consideration of the

low level details until later.

The task-oriented method has

proven

diagnosis[94][95],

useful

in

malfunction

equipment

design[96], and equipment selection problems[97] in chemical

engineering.

109

CHAPTER 6

EXPERT SYSTEM IMPLEMENTATION

Selection Hierarchy

The

key

to

the

task-oriented

decomposition and knowledge

approach

structuring.

is

problem

From the overview

section, it is clear that an extensive body of information on

the selection of solvents is available, although much of it in
a form unsuitable for directly coding into tasks.
For extractive distillation, the goal of the selection is
to find solvents in order to achieve the desired separation
based on

criteria such as selectivity or relative volatility,

heat capacity, e t c .
We

faced

an

important

challenge

when

addressing

the

accuracy of an expert system for solvent selection. Without

carrying out rigorous multicomponent thermodynamic equilibrium
calculations,

how do

volatility

for

associated

with

we

determine

specific

the

selectivity

solvent?

developing

incorporation of quantitative or

or

Another

expert

system

relative
challenge
is

the

"deep" knowledge into the

systems. Systems using only qualitative or "shallow" knowledge

HO
tend to be inaccurate/ and in the presence of new situation,
may be unreliable. However, systems using deep knowledge often
require numerical models,
slowly to be practical.

which are cumbersome and run too

In this work, we demonstrate that a

proper compromise between accuracy and efficiency in an expert

system for solvent selection can be made through simplified
thermodynamic models. Specifically, in order to quantitatively
evaluate the selectivity, we use
0

the

Helpinstill-van

Winkle

model

to

estimate

the

selectivity for hydrocarbon mixtures;

0 the extended solubility parameter approach with the
solvatochromic

parameters

to

estimate

the

selectivity

for

selectivity

and

nonaqueous nonhydrocarbon mixtures;

0

UNIFAC

with

UNIQUAC

to

estimate

relative volatility for .aqueous mixtures.

Figure 34 presents the selection hierarchy in its present
form. The form of the hierarchy is guided by two principles.
First,

as

decisions
solely

few calculations are performed as possible.

in

on

quantitative

the

upper

levels

qualitative
information

of

the hierarchy

relationship's.
is

used

Second,

primarily

Most

are based
detailed
for

final

comparisons.
The hierarchy structure consists of four levels, namely,
(1) obtaining basic data for the separation problem;
/

(2) selection of suitable process;

(3) determination of selectivity and solvency by using

Ill

Level 1
B a s ic D a ta In p u t

Level 2
E x tra c tiv e

A z e o tr o p ic

D is tilla tio n

D is tilla tio n

Level 3

H y d ro c a rb o n

N onaqueous

A queous

n o n h y d ro c a rb o n

s y s te m s

Level 4

B o iling

S o lv e n c y

S e le c tiv ity

Figure 34 Selection hierarchy for solvent selection

112
proper thermodynamic models based on the mixture information.
(4)
and

identifying the promising solvents based on polarity

boiling

points,

ranking

potential

solvents

by

using

solvency and selectivity criteria.

Figure

35

gives

the

detailed

description

of

all

the

branches in the expert system in the present form.

At the start point, the basic information needed is the
number, names, and normal boiling points of components in the
mixture

to

be

separated.

The

end-user

is prompted

by the

expert system to enter the information about the mixture.

Then the separation method is determined based oh the
given information by using the following heuristic:
If the boiling point of one of the components
mixture

is

higher

than

170C,

then

prefer

in the

azeotropic

distillation; if the boiling points of all components in the

mixture are lower than 170C, then ask an end-user to specify
whether extractive distillation or azeotropic distillation is
used to separate the mixture.

We use this heuristic in our

expert system according to Dr. Lloyd Berg's recommendation.

Next, when extractive distillation is specified as the
separation method, the end-user will be prompted by the expert
system to enter

into one of the

following three branches:

Hydrocarbon, Nonaqueous nonhydrocarbon,

and Aqueous system.

If all the components in the mixture are hydrocarbons,

then Hydrocarbon branch should be entered. If one of the

C START )
Title Screen
m o u s e click to c o n tin u e
INPUT: Basic Information

INPUT: A zeotropic properties

R e c o m m e n d e d Solvents

CHECK: B.R > I 70

Azeo1r0pte M a t o n -----------APPLY: Berg's c u e s

INPUT: Separation M ethods

Extractive Distillation
INPUT: Mixture Classification
BRANCH: Mixture Class
Non-Hyd carbon,
Non-A queous

______CHECK: d a ta b a s e for
I ' C
? / solubility p aram eters

113

INPUT: so lu b ility f-^V

p a ra m e te rs
A queous
APPLY: C h em ical Family Rule
CALC: ACID
CALC: selecth/tties

CALC: ACID
CALC: solvency

d M

INPUT: ratio

INPUT: UNlQAC p a ra m e te rs
In term ed iate Selections
I

CALC: relative volatilities

CALC: selectivfties
< ^ ? ^ >

Preliminary Selections

Interm ediate Selections

CALC: solubilities
CALC: ACID
CALC: selectivities

CHECK: relative volatlll1y> 1.5

CHECK: selectivity> 1.5

"*| R e c o m m e n d e d Solvents

APPLY: Berg's B.R Rule

CHECK: solvency> 0 .0 2

APPLY: Berg's Classification Rule

APPLY: Scheibal's Rule

R e c o m m e n d e d Solvents

CHECK: selectivity > 1.5

R e c o m m e n d e d Solvents

Figure 35 The flowsheet of the expert system

114
components in the mixture is water, then enter into Aqueous
system branch. Otherwise, enter into Nonaqueous nonhydrocarbon

branch.
The reason for this type of classification is that no
general thermodynamic model can be successfully applied to all
cases.

The

solubility

HelpinstilT-Van

parameter

Winkle[50]

gives

approach

developed

satisfactory

estimating ACID, for hydrocarbon systems.

results

Therefore,

by
for

for the

time being, Helpinstill-Van Winkle model is used in the expert

system to calculate the selectivity for comparing solvents.
When

the

mixture

contains

water,

the

system

will

be

highly nonideal. The solubility parameter approach gives very

poor

results

for

aqueous

systems. So,

at

present,

we

UNIFAC, which is an external program written in QBasic

Appendix C ) , to calculate ACID,

use
(see

then enter the ACID values

into the database created in LEVEL 5 OBJECT.

For nonaqueous nonhydrocarbon systems, at the time this
work was proceeding, the most accurate method to estimate ACID
was

the

MOSCED

model [56]

proposed

by

Thomas

et

al.

The

limitation of this model is that many parameters in the model

are unavailable.

Because of this we chose to implement the

Karger-Snyder-Eon model [55] to calculate ACID.

Hydrocarbon Branch
If the mixture is hydrocarbons,

then the expert system

checks to see if the components of the mixture exist, in the

115
database

for

the

hydrocarbon

branch. If

so,

no

physico

chemical properties need to be entered by the end-user.

If

not, then the end-user needs to enter the properties listed

below

for

each

component

in

the mixture

by

following

the

instructions on the screen:

1.
2.
3.
4.
5.
6.
7.
8.

class name
molecular formula
melting point (C)
molar volume at 25C (cm3/mol)
critical temperature (0C)
heat of vaporization at 25C (Kcal/mol)
refractive index at 25C
subdivision

We classify hydrocarbon compounds into saturated, unsaturated,

and aromatic hydrocarbons as subdivisions.
Phase I. When all needed physico-chemical properties are
available, the expert system proceeds to select solvents from
the database based on I) the polarity criterion and 2) boiling
point criterion as follows:
1) Select highly polar solvents.
The

solvents

in

the

classes

of

monoalcohols,

polyalcohols, acids, aldehydes, esters, amines, organic sulfur

compounds,

phenols,

nitriles,

and

nitroparaffins

in

the

database are considered to be highly polar solvents.

2) If the boiling points of solvents are 20C higher than
the

maximum

mixture,

of

then

boiling
the

points

solvents

of

the

satisfy

components
the

boiling

in

the

point

criterion. If not, the solvents, are not selected.

At this point, the expert system will show the number of

116
selected solvents and the name of each selected solvent.
Phase 2. The Helpinstill-Van Winkle model (shown below)
is used to calculate ACID for each component in the mixture
with each solvent.
Iny1" = V1[ ( A 1 - A2) 2 + (T1 - T2)2 - 2Tjr12 ]/RT +
In(V1Zv2) + I - V1Zv2

(9)

A1 2 : The nonpolar solubility parameters of the solute

and solvent;
U1^2 : The polar solubility parameters of the solute and
solvent;
V1 2 : The molar volume of the solute and solvent;
Tjf12 : Induction parameter characteristic of the nature of
the solvents and the type of hydrocarbon.
Correlations for calculating Tjr12 are given in Table 13.

Table 15. Correlations for Calculating Induction Parameter.

Type of hydrocarbon
1IfTP
a saturated hydrocarbon
an olefin
an aromatic

0.399 (C1 0.3 8 8 (T 1 0.447 ( T 1 -

T2
T2
T2

)2
)2
)2

The nonpolar solubility parameter of a polar solvent is

obtained from its homomorph:
A2 = AUvZV
Table

16

lists

the

expressions

nonpolar solubility parameter.

(10)
needed

for

calculating the

117
Table 16. Correlations for Calculating the Nonpolar Solubility
Parameter.
Solvent Homomorph

X2

yy ^-0.127^-0.00161V . (0.25661.il(V)-0.3553)

n-paraffin
cycloparaffin
aromatic

225.3V" "35V T r0"7

1616V(0"0017V " -73V-.872V>/ t (0-003V + O-*)

where Auv is the energy of vaporization (kcal/mol), V is the

molar volume (cm3/mol) , and Tr is the reduced temperature.
The polar solubility parameters are then obtained by
difference:
T2

(11)

= S2 - X2

where 6 is the total cohesive energy density,

S2 = (AH - RT )/V

(12).

where AH is the enthalpy of vaporization (kcal/mol).

For hydrocarbon and solvent

solutions,

an approximate

expression can be used to measure solubility roughly for high

[98]:
solvency = 1/y*

(13)

It should be noted that this expression is approximatly used

only for hydrocarbons with high polar solvents. It cannot be
extended to other mixtures. If solvency is too small, then the
solvent

will

form

two

phases

with

solutes.

This

is

not

expected for extractive distillation. When a solvent satisfies

the following criterion, then the solvent will be kept for the
further consideration:

solvency > 0.02

(14)

118
After this criterion,

the expert system will show the

number of the selected solvents and the name of each selected

solvent.
Phase 3. We use selectivity criterion to select solvents,
selectivity is defined as[49]:
<o

co

co

s 12 = Yi /Ya

(15)

If selectivity of a solvent satisfies the following criterion:

selectivity > 1.5

(16)

then the solvent is recommended for the further verification

either by the economic analysis or by experimentation. The
list of solvents recommended will show on the screen along
with their selectivities. It should be noted that the value of
1.5 is a default which can be changed by the end-user.

Aqueous System Branch

If extractive distillation is specified and the mixture
to

be

separated

is

aqueous,

then

the

end-user

selects

solvent from each chemical family with high polarity, at the

same time, the solvent selected should satisfy Berg's boiling
point criterion. Then ACID for each component in the mixture
with

each

selected

solvent

are

calculated

by

using

the

external UNIFAC program (see Appendix for the detailed UNIFAC

equations). These

ACID

values

are

then

entered

into

the

database.
In the present version of S S S , the ACID database is for
methanol-water and ethanol-water

systems with

12 different

119
solvents.
Phase I . When the end-user enters into the aqueous system
branch using extractive distillation as a separation method,
the

expert

system prompts

the user to enter

the

ratio

of

solvent to mixture, actual composition of the mixture as the

feed stream, and ACID for each component in the mixture. Then
the

expert

system prompts

the

user to

enter

the

name

and

number of the chemical groups of each component by using a

table equipped with the different main group parameters for
the UNIQUAC mod e l .
With

ACID

data

available,

the

expert

system

selects

solvents by applying the selectivity criterion (selectivity >

1.5). With the UNIQUAC thermodynamic model, we can calculate
activity

coefficients

at

the

specified

composition

for

different ratios of the solvent to mixture (see Appendix D for

the detailed description of the required equations).
Phase

2.

The

solvents

which

pass

the

selectivity

criterion are examined by the following relative volatility

criterion:
relative volatility > 1.5

(17)

Relative volatility is defined as:

a 12 =

Yi

P 1 /Y 2

18)

where P012 are the saturation pressures of the components.

The expert system then displays its recommendations for
the promising solvents.

120
Nonacnieous Nonhvdrocarbon Branch
If

the

end-user

enters

the

nonaqueous

nonhydrpcarbon

branch, the expert system will check if the components in the

mixture exist in the database for hydrocarbon branch and in
the database for nonaqueous nonhydrocarbon. If the components
do exist in the database, the user does not need to enter the
physico-chemical properties
components

do

not

physico-chemical

exist

for the expert system.

in

properties

the

database,

(defined

below)

the

If the
following

need

fo

be

entered:
1.
2.
3.
4.

class name
molecular formula
molar volume at 25C (cm3/mol)
solubility parameters (MPa1/z)
a. dispersion parameter
b. polarity parameter
c. hydrogen bonding parameter
5. Kamlet basicity parameter
Where class name is defined in the same way as in Appendix A.
Solubility parameters,

as used here,

are Hansen's extended

solubility parameters[83]. Kamlet, Taft and others [99] have

developed

scales

of

hydrogen

bond

acidity/basicity

and

dipolarity/polarizability, based upon extensive spectroscopic

measurements used to determine solvatochromic parameters (n*,
q:k

t,

and

/3ki) . The

molecular

interactions

for

compounds

in

solutions are characterized by the three scales: the tt* scale

for dipolarity/polarizability, the aKT scale for hydrogen bond
donor

(HBD)

acceptor

strength,

(HBA)

and the

strength.

y3KT scale

Each of the

for hydrogen bond

scales

is based upon

spectroscopic measurements coupled with a linear free energy

11

121
relationship

approach

to

separate

the

effects

of

various

interactions in solution. The hydrogen bond acidity/basicity

parameters are used to evaluate basic and acidic solubility
parameters for Karger-Snyder-Eon eguation. Some values of /?KT
are taken from the literature[99]. If they are not available,
we use rules, provided in the literature[104] to estimate them.
Then

the

expert

system

enters

into

the

nonagueous

nonhydrocarbon branch.
Phase

I.

In

this

branch,

Berg's

classification

for

solvents is used as a guide to make a preliminary selection of

solvents

(see Appendix D for the detailed classification of

solvents) . Classes I, II, and III from the five Berg's classes
are selected on this level. Also, Sheibel's rule is used for
solvent selection in this phase[20]:
Any member of the homologous series of either component
of

the

mixture

can

be

used

as

an

agent

to

separate

the

azeotropic mixture by extractive distillation.

At

this

point,

the

screen

will

display

the

list

of

solvents based on the above heuristics.

Phase 2. The expert system uses the same boiling point
rule as described above (20C higher) to make a more critical
selection. After this rule is applied, the expert system will
display the number of selected solvents and their names.
Phase 3. Karger-Snyder-Eon equation is used to calculate
A C I D [55]:
In

Y i " = V / R T [ ( S ld -

(S2 d ) 2 +

((Slp -

(S2 p ) 2 +

122
I
( 5 Ia

"

5 2 a ) ( 5 Ib

"

5 2b )

] + In(V1ZV2) + I - (V1ZV2)

(19)

where <Sd, 5p, Sar Sh are the dispersion solubility parameter,

polar solubility parameter,

basic solubility parameter,

and

acidic solubility parameter, respectively. V is molar volume.

5d and 5P are adopted from Hansen's dispersion parameter and
polar parameter.

In this study,

it was found that a linear

relationship exists between the S^-value of Karger et a l .[55]

and the /?KT-value of Kamlet et al. [99] :
V iz2Sjb=GO-1 P^+ 2 . 45

'

(20)

As for acidic solubility parameter, 5a, the a-values given by

Kamlet et al.

are less accurate than /3KT-values.

Sa can be

obtained using the following equation:

2 M b = 5T -6d2-5p2=5A2

Then

the

selectivity

criterion

is

applied

(21)

to

solvents

as

follows:
selectivity > 1 . 5

(22)

The definition of selectivity is the same as described before.

After the selectivity criterion,

the expert system displays

the number of the recommended solvents and their names along

with their selectivities. .

Azeotropic Branch
. As for azeotropic distillation, the current database is
only for the methanol-acetone system for illustration. In this
branch, only Berg's rules[11] have been implemented to dat e .

123
If

the

azeotrope,

mixture

is

closeboiling

one

or

maximum

then select solvents which form a binary minimum

azeotrope with one component, or form binary minimum azeotrope

with

each

component

but

one minimum

azeotrope

boils

at

significantly lower temperature than the other.

If the mixture is a minimum azeotrope, then solvents that
form a binary minimum azeotrope with one component that boils
at a significantly lower temperature than the original minimum
azeotrope can be selected.

Here we subjectively assign as a

"significantly lower temperature," 15C.

When the end-user enters this branch, the expert system
prompts for the following information to be entered:
Is it a close-boiling mixture or azeotrope?
If

it

is

an

azeotrope,

is

it

minimum

or

maximum

azeotrope?
What is the composition of the azeotrope? What is the
boiling point of the azeotrope?
With the information available, the end-user can direct
the expert system into the aqueous branch.
If the mixture is a close-boiling or a maximum azeotrope,
then

select

solvents

from

the

azeotropic

database

for ,

component I and from the azeotropic database for component 2.

Then if solvents from the two databases are not identical in
terms of their common names and CAS names, add them to the
recommendation list; otherwise compare the boiling point of

azeotrope I

(component I in it) with_ that of azeotrope 2

124
(component 2 in it) , if the difference between the two boiling
points are greater than 15C, then add the solvent into the
recommendation list; otherwise discard the solvent.
If .the

mixture

is

minimum

azeotrope,

then

select

solvents from the azeotropic database for component I and from

the.azeotropic database for component 2. If solvents from the
two databases are identical, discard them.

If solvents from

the two databases are not identical, then compare the boiling
point

of

the

new

azeotrope

with .that

of

the

original

azeotrope. If the difference between the boiling point of the

new azeotrope and that of the original one is greater than
15C, then add the solvent into the recommendation list,

if

not, discard it.

Appendix

lists

all, rules

and

thermodynamic

models

applied in the systems.

Application Examples
Example I: Selection of Solvents for
Separation of Xvlene Isomers
Xylenes are dimethy!benzenes and exist as three distinct
isomers, depending upon the relative positions of the methyl
groups

on the

material

aromatic

ring.

p-Xylene

is an

important raw

for manufacture of terephthalic acid and dimethyl

terephthalate,

both used to manufacture polyester synthetic

fibers. o-Xylene is used as a raw material for the manufacture

of

phthalic

anhydride

which

is

used

in

turn

for

the

125
manufacture of dioctyl phthalate and other phthalates, which
are used as plasticizers for polyvinyl chloride. m-Xylene is
almost entirely used for gasoline blending and conversion into
the

other

isomers

through

isomerization.

p-Xylene

has

the

greatest demand in relation to the availability of the various

isomers from refinery streams. Various properties of the three
xylene isomers are shown in Table 17. From the table, it can
be seen that the boiling points are quite close together. From
the boiling points
respect

to

pressure

and the changes

it

can

be

in boiling points with

computed

that

the

relative

volatility of p-xylene to m-xylene is 1.02[100] , which is so

slight that separation by conventional distillation is out of
the question.

First, we specified as a test a separation problem that

a mixture of m-xylene and p-xylene with the feed concentration
of 50% m-xylene.

The mixture is close-boiling.

The boiling

points are 139.5C and 138.7C, respectively.

Table 17. Physical Properties of Xylene Isomers.

o-Xylene
144.2
O
in
N
I

Boiling point, C
Freezing point, C
Molecular weight
Heat of Vaporization, KJ/kg
Density at 293 K, Mg/m3

106.16
347.00
0.8802

m-Xylene
139.5
-47.7
106.16
343.00
0.8642

P-Xylene
138.7
13.5
106.16
340.00
0.8610

The first task is to determine which separation process

will be used, since the boiling points of both components are.
lower than 170C, and compositions do not fall into a category

126
favoring

liquid-liquid

extraction,

both

extractive

and

azeotropic distillation processes are favored. The end-user

can choose any option of extractive distillation or azeotropic
distillation.

For

purposes

of

this

example,

extractive

distillation is arbitrarily specified at this point.

Both components are hydrocarbons, so the end-user directs
the program into the branch in which hydrocarbon mixtures are
handled.
Phase

I.

In this

module,

the

first

screen

shows

the

solvent list obtained by using Berg's boiling point rule and

polarity

criteria

as

shown

in

the

form

of

a WHEN

NEEDED

METHOD,
' WHEN NEEDED
BEGIN
IF number of components = 2 THEN
boil_point[3] := 0.0
boil_point[4] := 0.0
IF number of components = 3 THEN
boil_point[4] := 0.0
i := 0
FIND dB3 SOLV SUB I
WHERE boil_point OF dB3 S0LV_SUB I >= 20.0 + MAX
(boil_point[I ]/ boil_point[2]
boi!j o i n t [3] , boil_point[4])
AND (classname OF dB3 SOLV Su b I = "monoalcohols"
OR
classname OF dB3 SOLV "s u b I = "polyalcohols"
OR
classname OF dB3 SOLV "s u b I
"acids"
OR
classname OF dB3 SOLV "s u b I
"esters"
OR
classname OF dB3 s o l v ""s u b I = "aldehydes"
OR
classname OF dB3 SOLV "s u b I
"amines"
OR
classname OF dB3 SOLV "s u b I = o r g s u l f u r c o m p "
OR
classname OF dB3 s o l v ""s u b I = "phenols"
OR
classname OF dB3 SOLV "s u b I = "nitriles"
OR
classname OF dB3 s o l v ""s u b I = "nitroparaffins")
WHEN FOUND
i := i + I
solv coeff.solv nam e [i ] ':= dB3 S0LV_SUB I .commonname
solv coeff.boil point[i] := dB3 S0LV_SUB l.boil_point

solv coeff.crit temp[i]

dB3 S0LV_SUB I.crit^temp
solv coeff .mol vol [ij := dB3 S0LV__SUB l.mol_vol

127
solv coeff.heat of v a p [i ] := dB3 SOLVJSUB l.heat_vap
solv coeff.subdiv[i] := dB3 S O L V S U B l.subdiv
MAKE solv attri show
WITH solv number := i
WITH solv name := solv coeff.solv n a m e [i ]
WITH boil point := solv coeff.boil point[i]
WITH crit temp := solv coeff.crit t e m p [i ]
WITH mol vol := solv coeff.mol vol[i]
WITH heat vap := solv coeff.heat v a p [i ]
WITH subdiv := solv coeff.subdiv[i]
FIND END
total number OF solv coeff := i
record OF dB3 SOLVJSUB I := I
full screen OF main window := FALSE
title OF main window := "AUTO Solvent Selection System"
ASK SolvAttri display
END
the

system

comes

up

with

3 0 potential

solvents

from

the

database containing 145 solvents.

Phase 2. The second screen gives the solvent list after
considering solvency criterion as shown below,
WHEN NEEDED
BEGIN
solvency solvents.solv number := O
FOR idac.gammal2[i] <= 50.0 AND idac.gamma22[i ] <= 50.0
THEN
BEGIN
solvency_solvents.solv number := I +
solvency_solvents.solv number
solvency_solvents.solv name[solvency jsolvents.solv number]
:= solv coeff.solv nam e [i ]
solvency_solvents.gammaI2[solvency solvents.solv number]
:= idac.gammal2[i ]
solvency solvents.gamma22[solvency_solvents .solv number]
:= idac.gamma2 2[i ]
solvency_solvents.selectivity[solvency_solvents.solv
number] := idac.gammal2[i]/idac.gamma22[i]
MAKE solv list
WITH solv number := solvency_solvents.solv number
WITH.solv name := solvency solvents.solv name
END
solvency solvents.total number :=
solvency solvents.solv number
END
ASK solvent listjsolvency

END

128
the program cuts one solvent off from 3 0 potential solvents on
the previous screen and ends up with 29 potential solvents.
Phase 3. The third screen presents the solvent list after
considering the selectivity criterion which is also executed
by a WHEN NEEDED METHOD as the following,

'

WHEN NEEDED
BEGIN
FORGET solv name
FORGET selectivity
selectivity solvents.solv number := 0
FOR ( i= I TO solvency_solvents.total number]
BEGIN
IF solvency_solvents.selectivity[i] > = 1 . 2
OR I.0/solvency_solvents.selectivity >= 1.2 THEN
BEGIN
selectivity_solvents.solv number := I +
selectivity_solvents.solv number
selectivity_solvents.solv name
[selectivity_solvents.solv number] :=
solvency solvents.solv n a m e [i]
selectivity_solvents.selectivity
[selectivity_solvents.solv number] :=
solvency_solvents.selectivity[i ]
MAKE solvent list
WITH solv number := selectivity_solvents.solv number
WITH solv name := selectivity_solvents.
solv name[selectivity_solvents.solv number]
WITH selectivity := selectivity_solvents.
selectivity[selectivity solvents.solvnumber]
END
selectivity_solvents.total number :=
selectivity_solvents.solv number
END
ASK solvent list selectivity
END

the total number of potential solvents shown on the screen is

0, that is, all of the 145 solvents in the current database
have

been

excluded

as

potential

candidates

for

this

very

difficult separation, based upon four rules. Figure 36 shows

th e

s e le c tio n

p ro c e d u re and th e

num ber o f s o lv e n ts

a t each

129

145 solvents in the database

high polarity rule

and Bergs boiling point rule

V
30 solvents selected

solvency criterion

29 solvents selected

selectivity criterion (1.2)

7

0 solvents selected

Figure 36 Selection procedure and results for p-xylene

and ra-xylene

130
selection stage for the p-xylene and m-xylene system.

This

negative result in fact agrees with experimental results to

date. Dr. Lloyd Berg has run about 5,000 experiments but ended
up with no promising solvents for the separation of p-xylene
and m-xylene.
We also tested m-xylene and o-xylene system.
Phase I. After the primary selection based on polarity
criterion and Berg's boiling point rule,

the expert system

comes up with 35 potential solvents from 145 solvents in the

database.
Phase

2.

After

the

solvency criterion

is applied,

32

potential solvents are displayed on the screen.

Phase 3. The final recommended solvents after considering
the selectivity criterion are 16, which are given along with
their selectivities in Table 18. Table 18 also shows relative
volatilities for several solvents obtained from experiments by
Dr. Lloyd Berg[101][102]. As we can see, Dimethyl sulfoxide.
Dimethyl acetamide,

and I-Octanol show some promise by both

the expert system and experiments. I,4-Butanediol and

propylene glycol were proposed as a co-solvent with Dimethyl
sulfoxide

in the

literature[102].

Other solvents

listed in

Table 18 should be verified further by experiments. Figure 37

shows the selection procedure and the number of solvents at
each stage.

131
Table 18. The Selected Solvents by SSS with Their
Selectivities.
Computed
Experimental
Solvent
Selectivity
Relative Volatility
cyclohexanol
sec-heptyl alcohol
1-octanol
ethyl hexyl alcohol
propylene glycol
I,3-butanediol
I ,4-butanediol
dibutylamine
dimethyl acetamide
dimethyl sulfoxide
butyric acid
valeric acid
isovaleric acid
hexanoic acid
octanoic acid
propionic anhydride

1.28
1.45
1.38
1.42
1.31
1.34
1.26.
1.52
1.35
' 1.24
1.36
1.35
1.37
1.33
1.33
1.44

-1.38 [101]

1.33 [102]
1.30 [102]

Example 2. Selection i
of Solvents for Separation
of Water from Ethanol
Water and ethanol form a minimum azeotrope which contains
89 mol% ethanol and 11 mol% water:
separated

by

conventional

The azeotrope Cannot be

rectification

regardless

of

the

number of theoretical plates used. Table 19 shows the boiling

points of ethanol, water, and the minimum-boiling azeotrope at
one atmosphere.
In

order

azeotropic

to

produce

distillation

or

ethanol

free

extractive

of

water,

distillation

either
can

be

applied. For the design of these p r o c e s s e s a suitable solvent

should be determined first.
We use this example to illustrate the use of the expert

system to screen solvents for the separation of an aqueous

132

145 solvents in the database

high polarity rule

and Bergs boiling point rule
'

'

'

'

35 solvents selected

solvency criterion

. V
32 solvents selected

selectivity criterion (1.2)

I
16 solvents selected

Figure 37 Selection procedure and results for o-xylene

and m-xylene

133
Table 19. Properties of Components and Azeotrope at I Atm for
Ethanol and Water.
n.b.p., vC
Azeotrope
composition
Ethanol
Water
Ethanol-Water
azeotrope

78.3
100.0
78.15

89 mol% ethanol

mixture by extractive distillation.

At the beginning, the expert system asks for information
about the mixture. The number of the components is specified
as t w o , and then the names of both components in the mixture
are entered by the end user as water and ethanol. After that,
the components are checked against the database, the result of
the

checking step

shows

that both

components

exist in the

database, therefore the end user does not need to type in the
physical properties of the components in the mixture. Then, we
specify the separation method as extractive distillation. At
this point, the system asks the end user whether the mixture
is

aqueous

.nonhvdrocarbons,

hydrocarbons.

Obviously,

nonaqueous
this

nonhydrocarbs,

mixture

is

or

aqueous

nonhvdrocarbons. The expert system then chains to the branch

dealing

with

reguires

more

agueous

nonhydrocarbon

information

about

mixtures. This

the

mixture.

The

branch
mole

fractions of the azeotrope are entered as 0.11 and 0.89 for

water

and

ethanol,

respectively.

Activity

coefficients

at

infinite dilution for water as solute and ethanol as solvent

and for ethanol as solute and water as solvent are 2.35 and

134
5.42[65],

respectively.

The

ratio

of

any

solvent

to

the

mixture is entered as one.

The
solvents

database
for

in

the

different

representative

of

expert

system

mixtures.

different

contains

Each

chemical

several

solvent
group.

is
In

this

prototype system, activity coefficients at infinite dilution

for some mixtures with different solvents are calculated by an
external program written in QBasic. Table 20 shows activity
coefficients

at

infinite

dilution

for

water-ethanol

with

different solvents.

Table 20. Activity Coefficients at Infinite Dilution

for Water-Ethanol with Different Solvents
Solvent
Solute
Y~i
Y~2
ethylene glycol
acrylonitrile
dimethyl sulfoxide
dimethyl formamide
furfural
2-nitropropane

The

expert

water
ethanol
water
ethanol
water
ethanol
water
ethanol
water
ethanol
water
ethanol
system

uses

0.836
1.3601
23.128
3.6965
0.229
0.5912
0.75
1.5473
11.2
5.5041
24.434
6.8454
infinite

0.732
1.4114
49.317
2.8836
0.047
0.3022
0.501
1.8599
88.834
4.5449
391.07
5.6948
dilution

activity

coefficients to calculate selectivity by Equation

(6), then

the selectivity criterion is applied in the form of a WHEN

CHANGED METHOD:
BEGIN
solvent number OF control := 0
FIND solv selectivity
WHERE solv selectivity.selectivity >= 1.5

135
OR ( 1/solv selectivity.selectivity ) > = 1 . 5
WHEN FOUND
control.solvent number := control.solvent number + I
MAKE pri solvent
WITH number := control.solvent number
WITH name := solv selectivity.solvent name
WITH selectivity := solv selectivity.selectivity
FIND END
ASK first level solv display
END.

Then

the

solvents

along with

their

selectivities

are

displayed on the screen, as illustrated in Table 21.

Table 21. Solvents for Water-Ethanol Based upon Selectivity.

Solvents
Selectivity
ethylene glycol
acrylonitrile
dimethyl sulfoxide
dimethyl formamide
furfural
2-nitropropane

With

the

selectivity

0.6146
6.2567
0.3873
0.4847
2.0348
3.5694

of

different

solvents

for

the

mixture, the expert system uses U N IQUAC model to calculate

the relative volatility for each solvent. After the relative
volatility criterion,

which is very similar in form to the

selectivity criterion, is applied, the expert system comes up

with the proposed solvent list as shown in Table 22.
Ethylene glycol

is already used in industry.

Dimethyl

formamide has been suggested in the literature on the basis of

experimental vapor-liquid equilibrium data[103]. Furfural and
2-nitropropane need to be further verified by experiment.

136
Table 22.

Solvents Selected for Separation of Water-Ethanol

Based upon Relative Volatility.

Solvent

Boiling Point, 0C

Relative volatility

197.3
153.0
161.8
120.25

2.327
1.940
5.521
5.885

ethylene glycol
dimethyl formamide
furfural
2-nitropropane

Example 3. Selection of Solvents for Separation

of Methyl acetate from Methanol

Methyl

acetate

azeotrope which
methyl

contains

acetate.

conventional

and

The

methanol
18.7

minimum-boiling

wt.% methanol

azeotrope

rectification

form

cannot

regardless

and

be
of

81.3 wt.%

separated
the

number

by
of

theoretical plates used. Table 23 shows the boiling points of

methanol, methyl acetate and the minimum boiling azeotrope at
one atmosphere.
In
methanol

order
and

to

produce

methyl

pure

acetate

methanol
solution,

from

the

either

50

wt.%

azeotropic

distillation or extractive distillation can be used. We use

Table 23. Properties of Components and Azeotrope at I Atm for

Methanol and Methyl acetate.
n.b.p., 0C

Methanol
Methyl acetate
Methanol-MethyI acetate
azeotrope

Azeotrope
composition

64.3
56.3
54

18.7 wt.% methanol

extractive distillation to accomplish the separation task. The

137
expert

system

will

be

directed

, to

the

nonaqueous

nonhydrocarbon branch. The selection process comprises three

selection stages: primary selection, critical selection, and
reselection. At the primary selection stage, the expert system
selects

solvents

moderate

of

hydrogen

homologies

of

the

several
bonding

same

chemical
ability

chemical

classes with
or

polarity

classes

as

components in the mixture to be separated.

selection
accprding

stage,
to

Berg's

reselection stage,
select solvents.
calculation

the

of

solvent

boiling

selection

point

rule.

those

high or
and

the

for the

At the critical
is

carried

Finally,

the selectivity criterion

at

out
the

is applied to

We call this stage "reselection" because

selectivity

involves

using

model which is a time-consuming procedure.

deep knowledge as a refining selection step.

thermodynamic

This stage uses

Figure 38 shows

the selection procedure and the number of solvents at each

selection stage for the methanol and methyl acetate system.
Table 24 lists part of the solvents recommended by the expert
:

system.

From

Table 24. Partial Solvent List Selected for Separating

Methanol from Methyl Acetate.
Solvent

Selectivity (Methanol/MA)

Ethylene glycol
Propylene glycol
Glycerol
Ethyl formate

143.6
1424.0
7875.2
47.2

n=Propyl acetate

33.7

Ethyl butyrate

29.8

138

145 solvents in the database

primary selected
84 solven ts
critical selected
38 solven ts
reselected
30 solven ts

Figure 38 Selection procedure and results for methanol

and methyl acetate

139
the table,

we can see that the selectivity values are not

reliable. This is due to inaccurate estimation of solubility

parameters
mod e l .

and

The

solubility
and/or

inadequacies

in

the

current

current thermodynamic model

approach

molecules

interaction,

is

in which
divided

is based upon the

interactions
into

dipole interaction,

three

thermodynamic

between

parts:

solvent

dispersion

and hydrogen bonding.

The

total interaction is the summation of the three interactions

which are expressed by three solubility parameters: dispersion
parameter, polarity parameter, and hydrogen parameter.

Some

of these available parameters are not accurate and the current

estimation procedure is not reliable.
Much attention and many research efforts have been given
to

modifying

solubility
models

are

selectivity

the

thermodynamic

approach,

and

continuing
calculation

more

to

advanced

appear

might

models

in

be

the

based
and

more

upon

accurate

literature.

reasonably

the

reliable

The
and

acceptable if these more advanced thermodynamic models are

incorporated into the expert system.
Another way to overcome the

inadequacy

is to use the

Wilson or NRTL equations to calculate the selectivity.

But

this approach suffers the disadvantages that for many solventsolute systems there.are not empirical interaction parameters
available since they are estimated based on the experimental
vapor-liquid equilibrium data.

140

CHAPTER 7
SUMMARY

Conclusions

1. A knowledge-based system

(expert system)

utilizing both

qualitative and quantitative approaches has been developed in

this work and applied to helping the solvent selection for
separation of xylene isomers.by extractive distillation. The
expert

system

comes

separating p-xylene

up

with

no

from m-xylene.

promising

solvents

for

This result agrees with

experimental results to date. For separating o-xylene from mxylene, the expert system proposes several solvents which have
appeared in the patent literature. Other solvents proposed by
the expert system need to be verified by experiment.

2. For the etha.noI-water mixture to be separated by extractive

distillation,

the

expert

system

confirms

some

solvents

suggested in the literature and proposes additional solvents

for further experimental verification.

3. For the methyl acetate-rmethanol system, the expert system

can make selection of solvents for extractive distillation but
the expert system gives unreliable selectivity results due to
inadequacies in the current thermodynamic m o d e l .

141

Recommendation for the Future Development

Admittedly, the prototype expert system presented here is

not complete and far from perfect.

First of all, there were

numerous subjective decisions which went into the assignment

of values within the system. For instance, the selectivity (or
relative volatility) attribute was assigned a clear-cut value
of

1.5

as

criterion;

while

in

practice,

range

of

selectivities is usually considered. Therefore, fuzzy logic or

even confidence factors might be introduced into the system to
take

these

uncertainties

into

account

for

the

solvent

recommendation.
Second, at present, the theory that can predict activity
coefficients for aqueous systems is still under development
and

far

from

accurate,

with

the

exception

of

the

UNIFAC

method, which is moderately accurate for the prediction.

our

system,

UNIFAC

is

used

to

estimate

infinitely

In

dilute

activity coefficients and UN IQUAC to calculate selectivity and

relative volatility at

finite concentrations

temperatures.

the

One

of

disadvantages

and different

of . the

current

prototype system is that the calculation of infinitely dilute

activity coefficients and selectivity at finite concentrations
is separated from the expert system and the subroutine program
is written in the QBasic language.

T h i s .calculation routine

should be converted into a wiNDQWS-compatible external program

142
which

can

communicate

directly

with

the

LEVEL

OBJECT

programs.
Third, the solvent database for azeotropic distillation
is

quite

primitive,

extractive

and

distillation

even

are

the

not

large,

promising solvents are very limited.

expert system possesses,

solvent
so

databases
the

for

choices

of

The more knowledge the

the more difficult problems it can

tackle. Great effort should be made to expand these databases.

Fourth,

some

solvatochromic

parameters

that

are

not

available have been estimated manually based on rules of thumb

published
system

might

estimate
system.

in

is

the
be

literature
more

made

The

successfully
interaction.

references

efficient

and

automatically,

solvatochromic
applied
This

in

as

[103].

more
a

the

of

of

of

expert
if

the

approach

prediction

direction

powerful

part

parameter

The

the

expert

has

been

solute-solvent

development

is

worth

consideration. '
Fifth,
extraction

the

selection .of

differs

in

solvents

significant

for

ways

from

liquid-liquid
the

process

required for extractive and azeotropic distillation, but, to

some extent,

they are similar.

The expert system presented

here can be extended to solve the solvent selection problems

for

liquid-liquid

development

extraction

approach,

new

because,
module

by

for

using
the

modular

selection

of

solvents for liquid-liquid extraction can be attached easily

into the system.

143

NOMENCLATURE

144

NOMENCLATURE

Az Bf C

= Antoine's vapor pressure correlation constants

= constant

= enthalpy of mixing (Kcal/mol)

Nmin

= minimum number of theoretical plates

= system pressure (Atm)

P0i

= saturated vapor pressure of component i (Atm)

= gas constant (m3Pa/mol/K)

= selectivity

= absolute temperature (K)

Tb

= boiling point (0C)

Tc

= critical temperature (0C)

Tr

= reduced temperature

= internal energy (Kcal/mol)

= molar volume (cm3/mol)

= mole fraction in the liquid phase

= mole fraction in the vapor phase

aKT

= acidity scale

aF

= relative volatility at feed

Ot1J

= relative volatility

asij

= separation factor

145
basicity scale
Y

activity coefficient

incremental change

nonpolar solubility parameter (cal/cm3)1/2

TI*

dipolarity scale

polar solubility parameter (cal/cm3)1/2

polar-polar and polar-nonpolar induction energy

(cal/cm3)1/2

Subscripts
c

= critical conditions

i,j

= component i, component j

1,2

' = component number

Superscripts
oo

= infinite dilution

= vaporization

ABBREVIATIONS

ABBREVIATIONS

Al
ACID

artificial intelligence
activity coefficients at infinite dilution

DMSO

dimethyl sulfoxide

ED

extractive distillation

ES

expert system

MEK

methyl ethyl ketone

NTP

number of theoretical plates

RV

relative volatility

SE

solvent extraction

SSS

solvent selection system

Temp.

temperature

LITERATURE CITED

149

LITERATURE CITED
1. Sucksmith, I., "Extractive Distillation Saves Energy,"
Chem. E n g . , p. 91, June, (1982).
2. Berg, L., R. W. Wytcherley, and J. C . Gentry, "Industrial
Applications of Extractive Distillation," AICHE Spring
National Meeting, Houston, April, (1993).
3. Laroche, L., N. Bekiaris, H. W. Andersen, and M. Morari,
"Homogeneous Azeotropic Distillation: Comparing
Entrainers," Can. J. Chem. E n g . , 69:1302, (1991).
4. Hirata, Mitsuho, O h e , Shuzo, and Nagahama, Kunio, Computer
Aided Data Book of Vapor-Liquid Equilibria, Tokyo,
Kodansha Limited Elsevier Scientific Publishing Company,
1975.
5. Benedict, M., and L. C . Rubin, "Extractive and Azeotropic
Distillation, I . Theoretical Aspects," Trans. AICHE,
41:353, (1945).
6. Perry, R . H., Don W. Green, and James 0. Maloney, Perry's
Chemical Engineers' Handbook, 6th Edition, New York,
McGraw-Hill, Inc., 1984
7. Black, C., "Distillation Modeling of Ethanol Recovery and
Dehydration Processes for Ethanol and Gasohol," Chem. Eng.
Proa., 76(9):78, (1980).
8. Doherty, M. F . and G. A. Caldarola, "Design and Synthesis
of Homogeneous Azeotropic and Extractive Distillation,"
Ihd. Eng. Chem. Fundam., 24:474, (1985).
9. Young, S., "The Preparation of Absolute Alcohol from Strong
Spirit," Trans. Chem. Soc., 81:707, (1902).
10. Guillaum, E., U. S . Patent 887,793,
Alcoholic Liquid," May 3, (1908).

"Rectifying of

11. Berg, L . "Selecting the Agent for Distillation," Chem.

Eng. Prog., 65:52, (1969).
12. Treybal, R. E., Mass-Transfer Operations, 3rd ed., New
York, Mcgraw-HilI, Inc., (1980).

150
13. Doherty, M. F., "The Presynthesis Problem for Homogeneous
Azeotropic Distillation Has a Unique Explicit Solution,"
Chem. Ena. Sci.. 24:1985, (1985).
14. Van Dongen, D . B., and M. D . Doherty, "Design and
Synthesis of Homogeneous Azeotropic Distillations. I.
Problem Formulation for Single Column," In d . Eng. Chem.
Fundam., 24:454, (1985).
15. Doherty, M. F., and Perkins, J. D., "On the Dynamics of
Distillation Processes I. The Simple Distillation of
Multicomponent Non-reacting, Homogeneous Liquid Mixtures, "
Chem Eng. Sci., 33:281, (1978).
16. Doherty, M. F., and Calddrola, G. A . ,"Design and Synthesis
of Homogeneous Azeotropic Distillations III. The
Sequencing of Columns for Azeotropic and Extractive
Distillations," Ind. Eng. Chem. Fundam., 24:474, 1985.
17. Foucher, Etienne R., M. F . Doherty and M. F . Malone,
"Automatic Screening of Entrainers for Homogeneous
Azeotropic Distillation," Ind. Eng. Chem. R e s . , 30:760,
(1991).
18. Stichlmair, J., James R . Fair, and Jose L . Bravo,
"Separation of Azeotropic Mixtures via Enhanced
Distillation," Chem. Eng. Prog., 85:63, (1989).
19. Laroche, L., N . Bekiaris, H . W. Andersen, and M. Morari,
"The Curious Behavior of Homogeneous Azeotropic
Distillation Implications for Entrainer Selection,"
AIChE Journal, 38:1309, (1992).
20. Scheibel, E . G . ,"Principles of Extractive Distillation,"
Chem. Eng. Prog., 44:929, (1948).
21. Tassois, D . P., "Choosing Solvents for Extractive
Distillation," Chem. Eng. , 7,6:118, (1969).
22. Tassios, D. P., "Rapid Screening of Extractive
Distillation Solvents," Advances in Chemistry Series 115,
R. F . Gould, Ed. American Chemistry Society, Washington,
DC, p p . 46-63, (1972).
23. Kolbe, B . "Selection of Solvents for Extractive
Distillation Using Predicted and Correlated VLE data," I .
Chem. E . Symposium Series No. 5 6 , 1.3/23, (1979).
24. Yeh, An-I, "A Study of the Reversing of Relative
Volatilities by Extractive Distillation," Ph.D. Thesis,
Montana State University, Montana, April, (1986).

151
25. Thomas, E . R ., and C . A. Eckertz "Prediction of Limiting
Activity Coefficients by a Modified Separation of
Cohesive Energy Model and UNIFACz" Ind. Eng. Chem.
Process D e s . Dev., 23(2):194, (1984).
26. Knapp, J. P . et a l .,"Minimum Entrainer Flows for
Extractive Distillation: A Bifurcation Theoretical
Approach," submitted for publication, (1993).
27. Van Winkle, M., Distillation, McGraw-Hill, Inc., New York,
1967.
28. Gerster, J. A., "Azeotropic and Extractive Distillation,"
Chem. Eng. Prog., 65:43, (1969).
29. Barnicki,' S . D., and James R. Fair, "Separation System
Synthesis: A Knowledge-Based Approach. I. Liquid Mixture
Separations," Ind. Eng. Chem. Res . , 29:421, (1990).
30. Souders, M., "The Countercurrent Separation Process,"
Chem. Eng. Prog., 60, No.2, pp 75, (1964).
31. Kumar, R., John M. Prausnitz, and C . J. King, "Process
Design Considerations for Extractive Distillation:
Separation of Propylene-Propane," Advances in Chemistry
Series 115, R. F . Gould, Ed. , American Chemistry Society,
Washington, DC, pp.16,(1972).
32. Tiegs, D., Gmehling, J., et al., Activity Coefficients at
Infinite Dilution, DECHEMA Chemistry Data Series, V o l . IX,
2 Parts, DECHEMA: Frankfurt, 1986.
33. Bastos, J. C., M. E . Soares, and Augusto G . Medina,
"Selection of Solvents for Extractive Distillation. A Data
Bank for Activity Coefficients at Infinite Dilution," Ind.
Eng. Chem. Process Des. D e v . , 24:420, (1985).
34. Trampe, D . B., and C . A. Eckert, "A Dew Point Technique
for Limiting Activity Coefficients in Nonionic
Solutions," AIChE J . , 39, 1045(1993).
35. Pecsar, R. E., and J. J. Martin, "Solution Thermodynamics
from Gas-Liquid Chromatography," An a l . Chem., 38, 1661
(1966).
36. Shaffer, D. L., and T. E . Daubert, "Gas-Liquid
Chromatographic determination of Solution Properties of
Oxygenated Compounds in Water," A n a l . Chem., 41, 1585
(1969).
37. Leroi, J. C., and J. C . Masson, "Accurate Measurement of
Activity Coefficients at Infinite Dilution by Inert Gas

152
Stripping and Gas Chromatography," Ind. Eng. Chem.,
Process D e s . D e v . , 16, 139(1977).
38. Richon, D., P . Antoine, and H . Renon, "Infinite Dilution
Activity Coefficients of Linear and Branched Alkanes from
C1 to Cg in n-Hexadecane by Inert Gas Stripping, " Ind. Eng.
Chem., Process Des. Dev . , 19, 144(1980).
39. Scott, L . S . "Determination of Activity Coefficients by
Accurate Measurement of Boiling Point Diagram," Fluid
Phase Equilib., 26, 149(1986). .
40. Trampe, D . M., and C . A. Eckert,"Limiting Activity
Coefficients from an Improved Differential Boiling Point
Technique," J. Chem. Eng. Data., 35, 156(1990).
41. Hassam, A., and P . W. Carr, "Rapid and Precise Method for
the Measurement of Vapor/Liquid Equilibria by Headspace
Chromatography," A n a l . Che m ., 57, 793(1985).
42. Margules, M., Sitounqsber. A k a d . W i s s . Wien, Math.
Naturwiss. Klasse, 104:1243, (1895).
43. van Larr, J. J., Z . Phvs. C h e m ., 185:35,

(1929).

44. Wilson, G. M., "Vapor-Liquid Equilibrium. XI. A New

Expression for the Excess Free Energy of Mixing," J. Am.
Chem. Soc., 86:127, (1964).
45. Renon, H., and J. M. Prausnitz, "Local Compositions in
Thermodynamic Excess Functions for Liquid Mixtures, " AIChE
Journal, 14:135, (1968).
46. Abrams, D . S., and J. M. Prausnitz, "Statistical
Thermodynamics of Liquid Mixtures: A New Expression for
the Excess Gibbs Energy of Partly or Completely Miscible
Systems," AIChE Journal, 21:116, (1975).
47. Hildebrand, J. H., J. M. Prausnitz, and R. L . Scot,
Regular and Related Solutions, Van Nostrand Reinhold, New
York, (1970).
48. Flory, P . J., Principles of Polymer Chemistry, Cornell U.
P . , Ithaca, N. Y . , (1953).
49. Weimer, R. F., and J. M. Praunsnitz, "Screen Extraction
Solvents This Wa y , " Hydrocarbon Process Petr. Re f . ,
44:237, (1965).
50. Helpinstill, J. G., and M. van Winkle, "Prediction of
Infinite Dilution Activity Coefficients for Polar-Polar
Binary Systems," Ind. Eng. Chem. Process. Des. Dev.,

153
7:213,

(1968).

51. Hsieh, C . K. R., "Molecular Thermodynamics and High

Pressure Kinetics of Polar Reactions in Solutions," Ph.D.
Thesis, University of Illinois, Urbana, IL, (1973).
52. Newman, B . A., "Molecular Thermodynamics of Polar
Reactions," Ph.D. Thesis, University of Illinois, Urbana,
IL, (1973).
53. Hansen, C . M., "The Universality of the Solubility
Parameter," Ind. Eng. Chem. Prod. Res. Dev., 8:2, (1969).
54. Tijssen, R., H. A. 'H. Billet, and P . Schoenmakers, "Use of
the Solubility Parameter for Predicting Selectivity and
Retention in Chromatography," J. Chromatoqr., 122:185,
(1976).
55. Karger, B . L., L. R. Snyder, and C . Eon, "Expanded
Solubility Parameter Treatment for Classification and Use
of Chromatographic Solvents and Adsorbents, " A n a l . Chem.,
50(14):2126, (1978).
56. Thomas, E . R., and C . A. Eckert, "Prediction of Limiting'
Activity Coefficients by a Modified Separation of Cohesive
Energy Model and UNIFAC," Ind. Eng. Chem. Process D e s .
Dev., 23(2):194, (1984).
57. Howell, W. J., Karachewski, A. M., Stephenson, K. M.,
Eckert, C . A., Park, J. H., Carr, P . W., and S . C . Rutan,
"An Improved MOSCED Eguation for the Prediction and
Application of Infinite Dilution Activity Coefficients,"
Fluid Phase Equilib. 52,: 151, (1989).
58. Yunzhong, W u , Wang Shaokun, Hwang Derong, and Shi Jun, "An
Empirical Modification of the MOSCED model," The Canadian
Journal of Chemical Engineering, 70:398, (1992).
59. Pierotti, G. J., C . H . Deal, and E . L . Derr, "Activity
Coefficients and Molecular Structure," Ind. Eng. Chem.,
51:95, (1959) .
60. Derr, E . L., and C . H . Deal, "Analytical Solutions of
Groups. Correlation of Activity Coefficients through
Structural Group Parameters," Inst. Chem. Eng. Symp. Se r .
(London), 3(32):40, (1969).
61. Kojima, K., Tochigi K., Prediction of Vapor-Liquid
Equilibria by the ASOG Method, Elsevier, Amsterdam, 1977.
62. Fredenslund, Aa., Jones R. L., Prausnitz, J. M., "GroupContribution Estimation of Activity Coefficients in

154
Nonideal Liquid Mixtures,11 AIChE Journal, 21:1086, (1975).
63. Weidlich, U., and J. Gmehling, "A Modified UNIFAC Model.
I. Prediction of V L E , hE, and y," Ind. Eng. Chem. Re s . ,
26:1372, (1986).
64. Larsen, B . L., Rasmussen, P., and A. Fredenslund, "A
Modified UNIFAC Group-Contribution Model for the
Prediction of Phase Equilibria and Heats of Mixing," Ind.
Eng. Chem. Res., 26:2274, (1987).
65. Wilson, G . M . , "Infinite Dilution Activity Coefficients
Estimation of One Binary Component From the Other," AIChE
Symposium Series, No. 140, 70:120-152, (1974).
66. Bergmann, Diane L., and C . A. Eckert, "Measurement of
Limiting Activity Coefficients for Aqueous Solutions by
Differential Ebulliometry," Fluid Phase Equilib.,
63:141, (1991).
67. Waterman, D. A., A Guide to Expert System, Addison-Wesley,
Reading, MA, 1986.
68. Quantrille, T . E., and Y. A. Liu, Artificial Intelligence
in Chemical Engineering, Academic Press: San Deigo, CA,
1991.
69. Gandikota, M. S., "Expert Systems for Selection Problem
Solving Using Classification and Critique," M.S. Thesis,
The Ohio State University, Columbus, 1988.
70. Liu, Y. A., T. E . Quantrille, and S . H. Cheng, "Studies in
Chemical Process Design and Synthesis. 9. A Unifying
Method for the Synthesis of Multicomponent Separation
Sequences with Sloppy Product Streams," Ind. Eng. Chem.
R e s ., 29:2227, (1990).
71. Huang, Y. W., and L. T. Fan, "An Adaptive Heuristic-Based
System for Synthesis of Complex Separation Sequences,"
Artificial Intelligence in Process Engineering, Academic
Press, Inc., (1990).
72. Barhicki, S . D., and James R. Fair, "Separation System
Synthesis: A Knowledge-Based Approach. I. Liquid-Mixture
Separation," Ind. Eng. Chem. Re s . , 31:1679, (1992).
73. Barnicki, S . D., and James R. Fair, "Separation System
Synthesis: A Knowledge-Based Approach. 2. Gas/Vapor
Mixtures," Ind. Eng. Chem. R e s . , 31:1680, (1992).
74. Banares-Alcantara, R. "Development of a Consultant for
Physical Property Predictions," M.S. Thesis, Carnegie-

155
Mellon University, Pittsburgh, PA (1982).
75. Gani, R., and J. P . O'Connell, "A Knowledge Based System
for the Selection of Thermodynamic Models," Comput. Chem.
En g . , 13:397, (1989).
76. Chen, C .- L ., and J.-Y. Chang, "Expert System for Selection
of Phase Equilibrium Modes," Int. J. System Sci.,
23(6):885, 1992 .
77. Venkatasubramanian, V., "CAPS: An Expert System for
Plastics Selection," Knowledge-Based System in Chemical
Engineering, V o l . Ill, CACHE Case Study Series, CACHE
Corporation, Austin, TX (1988).
78. Banares-Alcantara R., A. W. Westerberg, E . I . K o , and M.
D . Rychener, "DECADE: A Hybrid Expert System for Catalyst
Selection-I. Expert System Considerations," Comput. Chem.
En g . , 11:265, (1987).
79. Banares-Alcantara R., A. W. Westerberg, E . I . K o , and M.
D . Rychener, "DECADE: A Hybrid Expert System for Catalyst
Selection-11. Final Architecture and Results," Comput.
Chem. E n g . , 11:265, (1987).
80. Venkatasubramanian, V., "PASS: An Expert System for Pump
Selection," Knowledge-Based System in Chemical
Engineering, V o l . II, CACHE Case Study Series, CACHE
Corporation, Austin, TX (1988).
81. Hanratty, Peter J. et a l ., " Knowledge Representation and
Reasoning in the Presence of Uncertainty in an Expert
System for Laboratory Reactor Selection," In d . Eng. Chem.
Res., 31:228, (1992).
82. Flich, E . W., Industrial Solvents Handbook, 4th Edition,
Noyes Data Corporation, Park Ridge, New Jersey, U.S.A.,
1991.
83. Barton, Allan E . M., ed., Handbook of Solubility
Parameters and Other Cohesion Parameters, CRC Press, Inc.,
(1983).
84. Riddich, John A., and William B . Bunger, Organic Solvents:
Physical Properties and Methods of Purification,
Techniques of Chemistry, V o l . II, 3rd Edition, WileyInterscience, New York, 1970.
85. Stephenson, Richard M., and Stanislaw Malanoski, Handbook
of the Thermodynamics of Organic Compounds, Elsevier
Science Publishing Co., Inc., New York, 1987.

156
86. Reid, Robert C., J. M. Prausnitz, and B . E . Poling, The
Properties of Gases & Liquids, 4th Edition, McGraw-Hill
Book Company, 1987.
87. Barr, Avron and Edward A. Feigenbaum, The Handbook of
Artificial Intelligence, Volume I, HeurisTech Press,
Stanford, California, 1981.
88. Stephanopoulos, G., J. Johnston, T . Kristicos, R.
Lakshmanan, M. Marvrovouniotis and C . Siletti, "DESIGNKIT: An Object-Oriented Environment for Process
Engineering," Comput. Chem. E n g . , 11:655, (1987).
89. Constantinou, L., Prickett, S . E., and M. L .
Mavrovouniotis, "Estimation of Thermodynamic and Physical
Properties of Acyclic Hydrocarbons Using the ABC Approach
and Conjugation Operators," Ind. Eng. Chem. R e s . , 32:1734,
(1993).
90. LEVEL 5 OBJECT User's Guide, Information Builders, Inc.,
(1992).
91. Stefik, M. and D. G. Bobrow, "Object-oriented Programming:
Theme and Variation," Al m a g . , 6:40, (1985).
92. Horsely, L. H., Azeotropic Data-III. Advances in Chemistry
Series 116, American Chemical Society, (1973).
93. Chandrasekaran, B., "Generic Tasks in Knowledge-Based
Reasoning: High-Level Building Blocks for Expert System
Design," IEEE Expert, Fall, 23, (1986).
94. Davis, J. F., S . K . Shum, B . Chandrasekaran, and W. F.
Punch, "III. A Task-Oriented Approach to Malfunction
Diagnosis in Complex Processing Plants," Presented at NSFAAAI Workshop in Al in Process Engineering, Columbia
University, New York, March, (1987).
95. Ramesh, T . S., S . K. Shum, and J. F . Davis, "A Structured
Framework for Efficient Problem Solving Diagnostic Expert
Systems," Comput. Chem. Enq . , 12:891, (1988).
96. Myers, D . R., Davis, J. F., Herman, and D. J. Still, "An
Expert System for Distillation Design," Comput. Chem.
Eng., 12:959, (1988).
97. Gandikota, M. S., "Expert Systems for Selection Problem
Solving Using Classification and Critique," M.S. Thesis,
The Ohio State University, Columbus, (1988).
98. Yalkowsky, S . H. and S . Banerjee, Aqueous Solubility:

157
Methods of Estimation for Organic Compounds, Marcel
Dekker, New York, (1992).
99. Kamlet, M. J., Abboud, J.- L ., Abraham, M. H., and Taft, R .
W., "Linear Solvation Energy Relationships. 23. A
Comprehensive Collection of the Solvatochromic Parameters,
TT*, a, and 3, and Some Methods for Simplifying the
Generalized Solvatochromic Equation," J. O r q . Chem.,
48:2877, (1983).
100. King, J., Separation Processes, Second Edition, McGrawHill Book Company/ New York, (1980).
101. Berg, L., U. S . Patent 5,094,723, "Separation of m-Xylene
from p-Xylene or o-Xylene by Extractive Distillation with
Alcohols," Mar. 10, (1992).
102. Berg, L. and An-I Yeh, U. S . Patent 4,676,875.
"Separation of m-Xylene from o-Xylene by Extractive
Distillation," J u n . 30, (1987).
103. Shealy, G. S., Hagewiesche, D., and Sandler, S . I.,
"Vapor-Liquid Equilibrium of Ethanol-Water-N,Ndimethylformamide," J. Che m . Eng. Data, 32:366, (1987).
104. Hickey, J. P . and D . R. Passino-Reader,"Linear Solvation
Energy Relationships: 'Rules of Thumb' for Estimation of
Variable Values," Environ. S c i . Technol., 25:1753,
(1991).

158

APPENDICES

159

APPENDIX A
INDUSTRIAL SOLVENTS FOR THE EXPERT SYSTEM DATABASE

160

Appendix A. Industrial Solvents for the Expert System

Database
Class
Water

CAS name
water
methanol
ethyl alcohol
I -propanol
2-propanol
I -butanol
2-butanol
2-methyl-1 -propanol

Monoalcohols

2-methyl-2-propanol
1-pentanol
2-pentanol
3-pentanol

Ketones

n-propanol
isopropanol
n-butanol
sec-butanol
isobutanol
tert-butanol
n-amyl alcohol

Mol. formula
H2C
C1H40
C2H60
C3H80
C3H80
C4H10O
C4H100
C4H100

2-ethyl-1-hexanol

sec-amyl alcohol
3-pentanol
Isobutylcarbinol
dimethylethylcarbinol
hexahydrophenol
n-hexyl alcohol
methylamylcarbinol
heptylcarbinol
2-ethylhexyl alcohol

benzyl alcohol
2-propen-1-ol
1,2-ethanediol
1,2-propanediol

benzyl alcohol
allyl alcohol
ethylene glycol
propylene glycol

C4H10O
C5H120
C5H120
C5H120
C5H120
CSHI20
C6H120
C6H140
C7H160
C8H180
C8H180
C7H80
C3H60
C2H602
C3H802

1,3-propanediol
1,3-butanediol
1,4-butanediol
1,2,3-propanetriol
methyl ketone
2-butanone

isopropylene glycol
1,3-butylene glycol
1,4-butylene glycol
glycerol

C3H802
C4H10O2
C4H10O2
C3H803

acetone
methyl ethyl ketone
diethyl ketone

C3H60
C4H80

3-methyl-1 -butanol
2-methyl-2-butanol
cyclohexanol
1-hexanol
2-heptanol
1-octanol

Polyalcohols

Common name
water
methanol
ethanol

3-pentanone
2-hexanone
3-hexanone

methyl n-butyl ketone

ethyl propyl ketone

2-heptanone
4-methyl-3-penten-2-one
4-methyl-2-pentanone
methylphenyl ketone
3-methyl-2-butanone
2,6-dimethyl-4-heptanone

methyl n-amyl ketone

mesityl oxide
methyl isobutyl ketone
acetophenone
methyl isopropyl ketone
diisobutyl ketone

C5H100
C9H180

3,5,5-trimethyl-2-cyclohexene-1 -one

isophorone

C9H140

C5H100
C6H120
C6H120
C7H140
C6H10O
C6H120
C8H80

161
Appendix A. Industrial Solvents for the Expert System
Database -- Continued

Class
Ketones

Amines

CAS name
cyclohexyl ketone
4-hydroxy-4-methyl-2-pentanone
1-aminopropane
2-aminopropane
1-aminobutane
1-amino-2-methylpropane
2-aminobutane
diethylamine
triethylamine
di-n-propylamine
2,4-dimethyl-3-azapentane
di-n-butylamine
aminocyclohexane
1,2-ethanediamine
2-aminoethanol
phenylamine
n,n-dimethyIformamide
n,n-dimethylacetamide
azole
pyrrolidone
1-methyl-2-pyrrolidone
azine
diethylene oximide
cyclopentane
n-pentane
cyclohexane
n-hexane

Hydrocarbons

Common name
cyclohexanone
diacetone alcohol
propylamine
isopropylamine
butylamine
Isobutylamine
sec-butylamine
diethylamine
triethylamine
dipropylamine
diisopropylamine
dibutylamine
cyclohexylamine
ethylenediamine
monoethanolamine
aniline
dimethylformamide
dimethylacetamide
pyrrole
2-pyrroIidinone
n-methylpyrrolidone
pyridine
morpholine
cyctopentane
pentane
cyclohexane
hexane

Mol. formula
C6H100
C6H1202
C3H9N
C3H9N
C4H11N
C4H11N
C4H11N
C4H11N
C6H15N
C6H15N
C6H15N
C8H19N
C6H13N
C2H8N2
C2H70N
C6H7N
C3H70N
C4H90N
C4H5N
C4H70N
C5H9CN
C5H5N
C4H90N
C5H10
C5H12
C6H12
C6H14

n-heptane

heptane

C7H16

n-octane
n-nonan
n-decane
n-dodecane

C8H18
C9H20

benzene
methylbenzene
1,2-dimethylbezene
1,3-dimethylbenzene

octane
nonane
decane
dodecane
benzene
toluene
m-xylene
o-xylene

C10H22
C12H26
CS HS
C7H8
C8H10
C8H10

1,4-dimethylbenzene
1-hexene

p-xylene
I -hexene

C8H10
C6H12

1-heptene
1-octene
1-nonene

1-heptene
1-octene
1-nonene

C7H14
C8H16
C9H18

162

Appendix A. Industrial Solvents for the Expert System

Database -- Continued
Class

CAS name
1-decene

Hydrocarbons

Acids

cyclohexene
styrene
methanoic acid
ethanoic acid
methylacetic acid
butanoic acid
2-methylpropionic acid
pentanoic acid
3-methylbutanoic acid
caproic acid
caprylic acid
propenoic acid
ethanoic anhydride

2-methyl-1 -propyl methanoate

methyl ethanoate
ethenyl ethanoate
acetic ester

propionic anhydride
methyl formate
ethyl formate
propyl formate
butyl formate
isobutyl formate
methyl acetate
vinyl acetate
ethyl acetate

C4H802
C4H802

propyl acetate
2-propyl ethanoate

n-propyl acetate
isopropyl acetate

butyl acetic ether

2-methyl-1 -propyl ethanoate

butyl acetate
isobutyl acetate

C5H1002
C6H1202
C6H1202
C6H1202

ethyl methanoate
propyl methanoate
butyl methanoate

pentyl acetate
amyl acetate
isopentyl acetate
isoamyl acetate
2-methyl-1 -propyl-2-methylpropanoate isobutyl isobutyrate
benzyl ethanoate
benzyl acetate
ethyl propanoate
ethyl propionate
ethyl butanoate
ethyl butyrate
methyl benzenecarboxylate
methyl benzoate

Halohydrocarbons

Mol. formula
C10H20
C6H10
C8H8
C1H202
C2H402
C3H602
C4H802
C4H802
CSHI002
CSHI002
C6H1202
C8H1602
C3H402
C4H603
C6H10O3
C2H402
C3H602
C4H802
CSHI002
C5H1002
C4H602

propanoic anhydride
methyl methanoate

Esters

Common name
1-decene
cyclohexene
styrene
formic acid
acetic acid
propionic acid
butyric acid
isobutyric add
valeric acid
isovaleric acid
hexanoic acid
octanoic acid
acrylic acid
acetic anhydride

ethyl benzenecarboxylate
propyl benzenecarboxylate
benzyl benzenecarboxylate
dichloromethane
methylene trichloride
carbon tetrachloride
1,1-dichloroethane

ethyl benzoate
propyl benzoate
benzyl benzoate
methylene chloride
chloroform
tetrachloromethane
ethylidene chloride

C7H1402
C7H1402
C8H1602
C9H1002
CSHI002
C6H1202
C8H802
C9H10O2
Cl 0H12O2
Cl 4H1202
C1H2CI2
C1H1CI3
Cl CW
C2H4CI2

163
Appendix A. Industrial Solvents for the Expert System
Database -- Continued
Class

Halohydrocarbons

Nitriles

Organic sulfur
compounds

Aldehydes

Nitroparaffins

Phenols

CAS name
1,2-dichloroethane
1,1,1 -trichloroethane
1,1,2,2-tetrachloroethane
1,1,1,2,2-pentachloroethane
o-dichlofobenzene
ethyl nitrile .
ethyl cyanide
n-propyl cyanide
phenyl cyanide
vinyl cyanide
methylsuffinylmethane
tetramethylene sulfone
2-furancarboxaldehyde
acetaldehyde
propionaldehyde
butyraldehyde
2-methylpropanal
pentanal

Common name
ethylene chloride
methlychloroform
tetrachloroethane
pentachloroethane
1,2-dichlorobenzene
acetonitrile
propion'rtrile
butyron'rtrile
benzon'rtrile
acrylonitrile
dimethyl sulfoxide
sulfolane
furfural
ethanal
propan al
butanal
isobutyraldehyde
n-valeraldehyde

Mol. formula
C2H4CI2
C2H3CI3
C2H2CI4
C2H1CI5
C6H4CI2
C2H3N
C3H5N
C4H7N
C7H5N
C3H3N
C2H60S
C4H802S
C5H402
C2H40
C3H60
C4H80
C4H80
C5H100

nitrocarbol
n'rtroethane
1-nitropropane
2-nitropropane
nitrobenzol

nitrocarbol
nitroethane
1-nitropropane
2-nitropropane
nitrobenzene

C1H302N
C2H5Q2N
C3H702N
C3H702N
C6H502N

carbolic acid
1-methyl-2-methylbenzene

phenol
o-cresol

C6H60
C7H80

1-methyl-3-methylbenzene
1-methyl-4-methy!benzene

m-cesol
p-cresol

C7H80
C7H80

164

APPEKfDIX B
SOLVENTS RECOMMENDED BY DR. LLOYD BERG

165

Appendix B . Solvent Recommended by Dr. Lloyd Berg for

the Expert System Database

CA S n a m e

C om m on nam e

Mol. form ula

1,4 -d ic y a n o b u ta n e

adiponitrile

C 6H 8N 2

2,3-dim eth y l p h e n o l

2 ,3-dim ethyl p h e n o l

C 8H 10O

2,4-dim eth y l p h e n o l

2 ,4-dim ethyl p h e n o l

C 8H 10O

3 ,4-dim eth y l p h e n o l

3 ,4-dim ethyl p h e n o l

C 8H 10O

3,5-dim eth y l p h e n o l

3 ,5-dim ethyl p h e n o l

C 8H 10O

2-n itro to lu en e

2 -n itro to lu en e

C 7H 7N 02

4 -n h ro to lu en e

4 -n itro to lu en e

C 7H 7N C 2

2 -h y d roxyb ip h en y l

2 -p h en y l p h e n o l

C 12H 10O

4-h y d ro x y b ip h en y l

4 -p h en y l p h e n o l

C 12H 10O

2 -nitrophen o l

2 -n itro p h en o l

C 6H 503N

4 -n r trophenol

4 -n itro p h en o l

C 6H 503N

2-butyl a c e ta te

se c-b u ty l a c e ta te

C 6H 1202

2 -m eth o x y eth a n o l

eth y le n e glycol m ethyl eth e r

C 3H 802

2 -eth o x y eth an o l

eth y le n e glycol ethyl e th e r

C 4H 10O 2

2 -b u to x y eth an o l

eth y le n e glycol butyl e th e r

C 6H 1402

d ie th y le n e glycol m ethyl eth e r

d ie th y le n e glycol m ethyl eth e r

C 5H 1203

d ie th y le n e glycol ethyl eth e r

d ie th y le n e glycol ethyl e th e r

C 6H 1403

1,3 -d io x o lan e-2 -o n e

eth y le n e c a rb o n a te

C 3H 403

1,2 -p ro p an e d io l cyclic c a rb o n a te

p ro p y le n e c a rb o n a te

C 4H 603

dim ethyl c a rb o n a te

dim ethyl c a rb o n a te

C 3H 603

diethyl c a rb o n a te

diethyl c a rb o n a te

C 5H 10O 3

1,4 -d io x an e

p -d io x an e

C 4H 603

m ethyl o -h y d ro x y b e n z o a te

m ethyl salic y late

C 8H 803

ethyl o -h y d ro x y b e n z o a te

ethyl sa lic y late ,

C 9H 10O 3

c y c lo d o d e c a n o l

c y c lo d o d e c a n o l

C 12H 240

h e p ta n o ic a c id

h e p ta n o ic a c id

C 7H 1402

n o n a n o ic a c id

p e la rg o n ic a c id

C 9H 1802

d e c a n o ic ac id

ca p ric ac id

C 10H 20C 2

166

APPENDIX C
QBASIC PROGRAM FOR CALCULATING ACID FOR THE EXPERT SYSTEM

167
Appendix C . QBasic Program for Calculating ACID for the Expert
System.
1 REM

This is an external program to calculate activity

. coefficient at infinite dilution
for the expert system
2 REM The example system is ethanol-water-ethylene' glycol
6 REM enter volume and area data into lines 10-40
7 REM enter interaction parameters into 50-100
8 REM the temp, is set
9 REM ethanol is component I; water is component 2; ETHYLENE
GLYCOL is component 3
10 DATA I, I, I, 0.6325, 1.0608:
'CH3=1,CMP=I:ethanol,NU(I)=IfR(I)fQ(I)
11 DATA 2, I, I, 0.6325, 0.7081:
'CH2=2,CMP=I :ETHANOL,NU(2)=1,R(2),Q(2)
12 DATA 3, I, I, 1.0630, 0.8663:
fOH=3,CMP= I :ETHANOL,N U (3)=I ,R (3),Q (3)
16 DATA 4, 2, I, 1.7334, 2.4561:
'water=4,CMP=2:WATER,N U (4)=1,R(4),Q(4)
18 DATA 5, 3, I, 2.0880, 2.4000: 'DOH=S,CMP=3:ETHYLENE
GLYCOL,N U (5)=1,R(5),Q(5)
20 DATA -I, 0, 0, 0,
0 : 'The condition to terminate input
UNIFAC parameters
49 REM The following is the data array for interaction
parameters
50 DATA 0,0,0, 0,0,0, 2777.0,-4.6740,0.1551e-2,
1391.3, -3.6156,0.1144e-2, 897.70,0,0
52 DATA 0,0,0, 0,0,0, 2777.0,-4.6740,0.1551e-2,
1391.3, -3.6156,0.1144e-2, 897.70,0,0
56 DATA 1606.0,-4.746,0.9181e-3, 1606.0,-4.746,0.91816-3,
0,0,0, -801.9, 3.824, -.007514, 499.8,-2.4100,0
58 DATA -17.253,0.8389,0.9021e-3, -17.253,0.8389,0.9021e-3,
1460,-8.6730,0.1641e-I, 0,0,0, 372.50,-0.9091,0
60 DATA 28.170,0,0, 28.170,0,0, -468.8,2.421,0,
-368.80,0.7775,0, 0,0,0
70 REM INPUT COMPOSITIONS
86 DATA le-40,1, I,le-40
89 DATA -I .
98 temp = 323
100 DIM NGP(SO), CMP(SO), NU(50), R(50), Q(50)
105 DIM p s i (30, 30), AA(30, 30), BB(30, 30), CC(30, 30)
H O DIM R R (50), Q Q (50.) , PHP(SO) , PH(50) , TH(50) , LNGC(SO) ,
x(50), LNGR(SO) .
120 DIM g(50), CAPX(50), CAPT(SO), LNCAPG(50), SSUB(50, 50)
130 N = 0 : ENT = 0
140 FOR I = I TO 100
150 READ NGP(I), CMP(I), NU(I), R(I), Q(I)
155 REM VOL-AREA IS READ
160 IF NGP(I) = -I THEN 220

168
162
170
180
190
200
210
220
230
240
250
260
270
280
290
300
305
310
320
330
340
350
360
362
370
380
390
~400
410
420
430
440
450
460
470
480
490
500
510
520
522
530
540
550
560
570
580
590

REM occurs when entry of Table is complete

ENT = ENT + I: 'ho. of entries in vol-area Table
CMP = CMP(I)
IF NGP(I) < N THEN 210
N = NGP(I)
NEXT I
PRINT N
REM read interaction parameters
FOR j = I TO N
FOR k = I TO N
READ AA(j, k ) , B B (j , k ) , CC(j, k)
psi (j , k) = E X P (-(AA(j, k) + B B (j , k) * temp + CC(j, k) *
temp A 2) / temp)
NEXT k
NEXT j
FOR j = I TO CMP
R R (j ) = 0
QQ (j ) = 0
L (j ) = 0
FOR k = I TO ENT
IF j <> CMP(k) THEN 370
R R (j ) = R R (j) + R(k) * NU(k)
QQ(j) = Q Q (j) + Q(k) * NU(k)
REM RR(J) and QQ(J) are volume & area parameters for pure
components
NEXT k
L (j) = 4 * RR(j) - 5 * QQ(j) + I
NEXT j
FOR L = I TO CMP
FOR j = I TO CMP
x(j) = 0
NEXT j
X(L) = 1
GOSUB 1000: 'pure component group activity coefficient are
calculated
NEXT L
'
PRINT "TEMP
CONC(X)
ACT. COEFF."
PRINT
FOR SET = I TO 100
FOR I = I TO CMP
READ X(I)
IF X(I) = -I THEN 810
"
REM this condition occurs when all data have been
processed .
NEXT I
'
GOSUB 1000: 'Group activity coefficients are calculated in
the mixture
SUMl = 0
SUM2 = 0
sum3 = 0
FOR j = I TO CMP
SUMl = SUMl + x (j ) * R R (j) " (3 / 4)

169
600
610
620
630
635
640
650
660
670
680
690
700
710
720
730
740
743
745
746
747
750
760
770
780
790
800
810

SUM2 = SUM2 + x(j) * R R (j )

sum3 = sumS + x(j) * QQ(j)
NEXT j
FOR k = I TO CMP
PHP(k) =.R R (k ) " (3 / 4) / SUMl
P H (k) = RR(k) / SUM2
TH(k) = Q Q (k) /-sum3
LNGC(k) = I! - PHP(k) + LOG(PHP(k)) - 5! * QQ(k) * (I!
P H (k) / T H (k) + LOG(PH(k) / T H (k )))
NEXT k
FOR I = I TO CMP
LNGR(I) = 0
FOR j = I TO ENT
IF I O CMP(j) THEN 730
LNGR(I) = LNGR(I) + NU(j) * (LNCAPG(N G P (j )) - SSUB(I,
N G P (j )))
NEXT j
NEXT I
sumc =. x(l) + x(2)
x(l) = x(l) / sumc
x(2) = I - x(l)
x(3) = x(3) / sumc
FOR I = I TO CMP
g(I ) = E X P (LNGR(I) + LNGC(I))
PRINT temp, x(I), g(I)
NEXT I
PRINT
NEXT SET
END

1000 FOR I = I TO ENT

1010 SUMM = SUMM + (x(CMP(I))) * NU(I)
1020 NEXT I
1030 FOR j = I TO ENT
1040 CAPX(j) = 0: 'CAP X(J) IS THE G P . MOLE FRACTION
1050 FOR k = I TO ENT
1060 IF j O NGP(k) THEN 1080
1070 CAPX(j) - CAPX(j) + N U (k) * x(CMP(k))
1080 NEXT k
1090 CAPX(j ) = CAPX(j) / SUMM
1100 NEXT j
1110 SUMM = 0
1120.A = 1
1130 FOR I = I TO ENT
1140 IF NGP(I) < A THEN 1170
1150 A = NGP(I)
1160 SUMM = SUMM + Q(I) * CAPX(A)
1170 NEXT I
1180 A = I
1190 FOR j = I TO ENT
1200 IF NGP(j) < A THEN 1230
1210 A = NGP(j)

170
1220
1230
1240
1250
1260
1270
1280
1290
1300
1310
1320
1330
1340
1350
1360
1370
1380
1390
1400
1405
1410
1420
1430
1440
1450
1460

CAPT(A) = Q(j) * CAPX(A) / SUMM

NEXT j
A = I
FOR I = I TO ENT
IF NGP(I) < A THEN 1440
A = NGP(I)
SUMl = 0
FOR j = I TO N
SUMl = SUMl + CAPT(j) * psi(j. A)
NEXT j
SUM4 = 0
FOR j = I TO N
sum3 = 0
FOR k = I TO N
sum3 = sum3 + CAPT(k) * psi(k, j)
NEXT k
SUM4 = SUM4 + CAPT(j) * p s i (A, j ) / sum3
NEXT j
LNCAPG(A) = Q(I) * (I - L O G (SUMl) - SUM4)
REM LNCAPG is the gp. activity coeff.
FOR M = I TO CMP
IF X(M) = I THEN SSUB(M, A) = LNCAPG(A)
NEXT M
NEXT I
SUMM = 0
RETURN

171

APPENDIX D
RULES AND THERMODYNAMIC MODELS FOR THE EXPERT SYSTEM

172
Appendix D . Rules and Thermodynamic Models for the Expert
System

I. Rules for the Method Selection:

The separation method is determined based on the

given information by using the following heuristic[11]:
If the boiling point of one of the components
mixture

is

higher

than

170

0C,

then

prefer

in the

azeotropic

distillation; if the boiling points of all components in the

mixture are lower than 170 0C z then ask an end-user to specify
whether extractive distillation or azeotropic distillation is
used to separate the mixture.

II. Rules and Thermodynamic Models for the Selection of

Solvents for Extractive Distillation
A.

Hydrocarbon system:

Berg's ruleril]
Boiling point of a solvent for extractive distillation should
be greater than the highest boiling point of components in the
mixture to be separated.
Tasso's ruleF22]
If the mixture is composed of hydrocarbons, choose a solvent
with high polarity.
Infinite Dilute Activity Coefficients^50]
lnYi = V1[ (X1 - A2)2 +

(T 1

T2 ) 2

In(V1Zv2) + I - V1Zv2

- 2i|r12 ]/RT +
(I)

173
= c CrI - rE )2

(2)

: The nonpolar solubility parameters of the solute and

solvent;
T1^2 : The polar solubility parameters of the solute and
solvent;
V1^2 : The molar volume of the solute and solvent;
i|rl2 : Induction parameter characteristic of the nature of
the solvents and the type of hydrocarbon
If it is a saturated hydrocarbon, then
f12 =0.399 (T1 - r2 )2

(3) .

If it is an olefin, then
Iji12 =0.388

(T 1

- T2 )2

(4)

If it is an aromatic, then
Ijr12 =0.447

(T 1 -

T2

)2

(5)

The nonpolar solubility parameter of a polar solvent is

obtained from its homomorph
X2 =

AUv/V

(6)

If it is a n-paraffin, then

^yVy

j j

y*"0 .1270-0.00161Vyip (0.2566Ln(V)-0.3553)

(7 )

If it is a cycloparaffin, then
AUv/V = 225. S V 0-35ZTr0'7

(8)

If it is an aromatic, then
A y V /Y

where

= 1616V<0-0017V " - 7 3 ) e

AUv is

the

energy

( - 0 .0 0 8 7 2 V ) / r p (0.003V

of

+ 0 .4 )

vaporization,

^g ^

is

the

molar

volume, and Tr is the reduced temperature.

The polar solubility parameters then being

obtained by

174
difference:
( 1 0 )

where 6 is the total cohesive energy density

62 = (AH - RT )/V

( 11 )

where AH is the enthalpy of vaporization.

Solvency
solvency = 1/y"

( 12 )

Solvency > 0.02

If

solvency

is too

small,

then

the

solvent will

form two

phases with solutes.

Selectivity^49]
S"12

(13)

Check Selectivity > 1.5

The value of 1.5 should be changeable.
would

like .to

selectivity,
infinitely

narrow

he

can

dilute

the

range

change

1.5

concentration,

to
to

If an end user

solvents
a

with

larger

selectivity

higher

value.

tends

to

At
be

maximum.

B . Aqueous System

Chemical Family Rule

Select one to two solvents from different chemical families
with high polarity.
Infinite dilute activity coefficients
Use the UNIFAC thermodynamic model to calculate infinite

175
dilution

activity

coefficients

for

the

solvents

selected

above. In the modified UNIFAC model, the activity coefficient

is the sum of a combinatorial and a residual par t : [63]
In Y1 = lnYiC + Iny/

(14)

The combinatorial part can deal with compounds very different

in size:
In Yc1= I-Vf1 +InV' 1 -Sq1(I-Vf1ZF1 +In(V1
V F 1))

(15)

The parameter V f1 can be calculated by using the relative van

der Waals volumes, Rk, of the

different groups.
ri/4
(16)

(17)

(18)
(19)

Qi= Z v ^ e k

The

residual

part

can be

obtained

(20)
by using

the following

relations:
' Inyf=Ev^ (InTjc-Inr^)

(21)

Jc

inrt=Ct( i - m (Ee1T m k )

(22)

whereby the group area fraction :6m and group mole fraction Xm
are given by the following equations:
(23)
n
(J)

(24)

EEv
J n

( J)

176
In the modified UNIFAC model temperature-dependent parameters
are introduced to permit a better description of the real
Behavior(activity coefficients) as a function of temperature.
W nm=Bxp ( - anm+b^

+C^ T 2 )

(25)

Selectivity
The definition is the same as in the hydrocarbon system.
Select UNIOUAC Parameters
Select area and volume parameters for U N IQUAC model from
the interface.
the

infinite

These values are stored in the database. Use

dilute

activity

coefficients

calculated

from

UNIFAC to calculate the interaction parameters for UNIQUAC.

At infinite dilution the U N IQUAC expression for a binary
system can be written in the following form
I n y r = I n ^ +Q1 ( 5 l n - ^ ^ - l m ; 21+ l - T 12) +J1- - ^ I 2
Z2
UarI
x2

(26)

Inya=In^+Qf2 (Sln-^ y --InT12+I-T21) +I2- - ^ l 1

i
^r2

(27)

After rearrangement we get

Yi

(28)

Y2

(29)

lnT2I+^i2=Ai-ln

intia+t 2i=A2-In

%
So that we are able to calculate the U N IQUAC parameters r12 and
Z21 from Y1" and Y2"/

where

A1=A (j +inll-Ilj )+SinA A +i

<2i

r2

r2

(30)

g2r1

A1 and A2 contain only the relative van der Waals parameters

177
A2= - (I + I n - ^ . + 5 l n - ^ : + l

(31)

r. and g.. These parameters can be calculated from the van der

Waals

group

parameters

Rk and

Qk that

are

stored

in

the

database.
The activity coefficients at feasible concentrations are
calculated by UNIQUAC mode l .
Relative volatilities
otI2 =

P 1/Y2 ^ 2

Y-i

(32)

where P012 are the saturation pressures of the components.

Relative volatility > 1.5
This value should be changeable.

C. Nonhvdrocarbon, Nonaqueous System

Berg's Classification Ruler11 1

Water,

glycol,

glycerol,

amino alcohol,

hydroxylamine,

hydroxy acids, and amides are class I .

Alcohols, acids, phenols, primary and secondary amines,
nitrocompounds with alpha-hydrogen atoms, nitriles with alphahydrogen atoms' are class II.
Ethers,

ketones,

aldehydes,

esters,

tertiary

amines

(including pyridine type), nitrocompounds and nitriles without

alpha-hydrogen atoms are class III.
Class

IV

consists

of

liquids

composed

of

molecules

containing active hydrogen atoms but no donor atoms. These are

molecules

having two

or

three

chlorine

atoms

on

the

same

178
carbon as a hydrogen atom,, or one chlorine on the same carbon
atom and one or more chlorine atoms on adjacent carbon atoms.
All other liquids are class V. These are liquids having
no hydrogen bond forming capabilities.
Solvents

in

classes

I,II,III

are

selected because

of

strong or moderate hydrogen bonding ability.

Scheibal's rule[20]
Any member of the homologous series of either component
of the mixture can be used to separate the azeotropic mixture
by extractive distillation.
Berg's Boiling Point Ruler11]
The same as for the hydrocarbon branch.
Infinite Dilution Activity Coefficients
Modified Karger-Snyder-Eon equation combined with
solvatochromic

parameter

approach

is

used

to

calculate

infinitely dilute activity coefficients[55]:

-In YV = V/RT[(S1d - S2d)2 + (S1p - S2p)2 +
2 (^ia ^2a) (^lb ^2fc>) ]

(33)

where Sd, Sp, Sa, Sb are dispersion solubility parameter, polar

solubility parameter, basic solubility parameter, and acidic
solubility parameter, respectively. If the differences due to
molecular volumes

are taken into account,

then equation 33

becomes
in y ,- = V/RT[(61d - S2d)2 + (S1d - S2p)2 +

(1. - S2aX 5Ib - 52b>l + In(V1Zv2) + I - (V1ZV2)

(34)

In this study, the values of the dispersion solubility

179
parameters and polar solubility parameters are adopted from
Hansen's solubility parameter values. As for basic solubility
parameter,

6b, Kamlet et a l . [96] used spectrometric methods

to obtain two solvatochromic parameters, a and 3. The a-scale

of the solvent provided a measure of proton-donor capacity,
and the

3-values

quantified the ability of the

solvent to

accept a proton. Kamlet et a l .[99] have prepared an extensive

table

of

values

for

solvents,

and

Hickey

et

a l .[100 ]

documented rules of thumb for estimations of a-scale and 3scale. Karger et al. provided a limited number of 6g
and

6b values

for

solvents. One

assumed equal values

shortcoming

is

of Sg and 6b for alcohols,

that

they

while most

alcohols do not behave this way. In this study, it was found

that

linear

relationship

exists

between

the

6b-value

of

Karger et a l . and the 3-value of Kamlet et al.:

V 1/28jb=6 0 .i P + 2 .45

(35)

As for acidic solubility parameter, Sg, the a-values given by

Kamlet

et

al.

are

less

accurate

than

3-values.

Sg can be

obtained using the following equation:

2 6 ^ = 8 ^ -8 ^ -8 /= 8 ^

(36)

Selectivity
The definition is the same as in the hydrocarbon system.
Selectivity > 1.5
The reason that selectivity is greater than 1.5 is the same as
for the hydrocarbon branch.

180
III. Rules for the Solvent Selection for Azeotropic
Distillation

As for azeotropic distillation, the current database is

only for methanol-acetone

system for illustration.

In this

branch, Berg's rules[11] are used for the time being:

If

the

azeotrope,

mixture

is

close-boiling

one

or

maximum

then select solvents which form a binary minimum

azeotrope with one component, or form binary minimum azeotrope

with

each

component

but

one

minimum

azeotrope

boils

at

significantly lower temperature than the other.

If

the

solvents

mixture

that

form

is
a

minimum

binary

azeotrope,

minimum

then

azeotrope

select

with

one

component that boils at a significantly lower temperature than

the original minimum azeotrope.
Here

we

subjectively

assign

significantly

lower

temperature as 15 0C .
When the end-user enters this branch, the expert system
prompts the following information to be entered:
Is it a close-boiling mixture or azeotrope?
If

it

is

an

azeotrope,

is

it

minimum

or

maximum

azeotrope?
What is the composition of the azeotrope? What is the
boiling point of the azeotrope?
If the mixture is a close-boiling or a maximum azeotrope,
then

select

solvents

from

the

azeotropic

database

for

component I and from the azeotropic database for component 2.

181
Then if solvents from the two databases are not identical in
terms of their common names and CAS names,

add them to the

recommendation list; otherwise compare the boiling point of

azeotrope

(component

I in

it)

with

that

of

azeotrope

(component 2 in it)., if the difference between the two boiling

points are greater than IS0C , then add the solvent into the
recommendation list; otherwise discard the solvent.
If

the

mixture

is

minimum

azeotrope,

then

select

solvents from the azeotropic database for component I and from

the azeotropic database for component 2. If solvents from the
two databases are identical,

discard them.

If solvents from

the two databases are not identical, then compare the boiling
point

of

the

new

azeotrope

with

that

of

the. original

azeotrope. If the difference between the boiling point of the

new azeotrope and that of the original one is greater than
IS0C, then add the solvent into the recommendation list,
not, discard it.

if

MONTANA STATE UNIVERSITY LIBRARIES

1762

10254043 O

UHCA/OMAHA

NE,