.ux.!l¢. ..
culvnbv.

 

t. :9
A z 3 .3!)-
Ex:

1.31:1

V (11::

.. iii-31:2.
h”! l. 21»! v.9...

II bdlv .
up... 490...!

1
.J...tt.ian I)
. 1; . .

. 2.3:.
3 Liza-33 a!
I. .v

.4 LitrllfV): . l
IIR’II 71.5.".
unsil:!....t I

I b
. t4: .Z-II:I.....I
.5: \co.‘ ‘

 

” “199»;

Date

                                                                                                                         

3 129300 8927

This is to certify that the

thesis entitled

KNOWLEDGE ACQUISITION FROM EXPERTS

IN CONCEPTUAL DESIGN
OF ENVIRONMENTAL ENGINEERING SYSTEMS

presented by

Lori Schutz—Riley

has been accepted towards fulfillment
of the requirements for

Masters degree m Environmental
Engineering

 

 

142,4; MJL

Major professor

March 30, 1993

 

07639

MS U is an Affirmative Action/Equal Opportunity Institution

 

 

LIBRARY
Michigan State
1 Unlverslty

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.

 

DATE DUE DATE DUE DATE DUE

 

m1—

 

m toe]

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU Is An Affirmative Action/Equal Opportunity Institution

chS-q

—-———__

KNOWLEDGE ACQUISITION PROM EXPERTS
IN CONCEPTUAL DESIGN

OF ENVIRONMENTAL ENGINEERING SYSTEMS

BY
Lori Schutz-Riley

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

MASTER OF SCIENCE
Department of Civil and Environmental Engineering

1993

ABSTRACT

KNOWLEDGE ACQUISITION FROM EXPERTS
IN CONCEPTUAL DESIGN
OF ENVIRONMENTAL SYSTEMS

BY

Lori Schutz-Riley

This thesis describes a methodology for knowledge
acquisition useful for knowledge representation in the
Hierarchial Classification tool. The domain is conceptual
design for environmental engineering systems, specifically
wastewater treatment. The conceptual design problem is
solved by decomposing expertise using design criteria.
Basic concepts are defined such as experts versus novices,
paradigms, and why there is a need for such an expert
system. Knowledge acquisition and knowledge representation
are discussed, with descriptions of the two knowledge
acquisition tools used. The repertory grid reveals how the
expert relates the features and design criteria to
technologies. Constraint based representation reveals how
the expert compares design criteria against waste features

independent of the technologies.

A methodology for knowledge acquisition and representation

a

is presented. The important design criteria of the

competing technologies are identified. Next, the features

 

relative to the design criteria are identified. Input from
three experts was translated into the knowledge
representation units, or matchers, within the Hierarchial

Classifier.

Copyright by
LORI SCHUTZ-RILEY

1993

To my husband Rick with love

ACKNOWLEDGMENTS

I am especially grateful to Dr. Craig S. Criddle and Dr.
William Punch, III for their inspiration, leadership and
encouragement during my graduate studies at Michigan State
University. I would like to express my sincere appreciation
to my other graduate committee members Dr. MacKenzie Davis
and Dr. Susan Masten for their guidance and contribution.
Special acknowledgements are due to the consulting firms of
Rose & Westra, Inc. of Grand Rapids, Michigan, and
Environmental Science & Engineering, Inc. of Williamston,
Michigan for their expertise provided in interviews for

knowledge acquisition.

vi

1.0.

2.0.

3.0.

BACKGROUND . . . . . ' . .

TABLE OF CONTENTS

INTRODUCTION . . . . . . . . . . .
1.2. OVERVIEW . . . . . . . . . .
1 O 3 O APPROACH O O O O O O O

2.1. DEFINITIONS OF TERMS . .
2.2. EXPERTS VERSUS NOVICES .
2.3. PARADIGMS . . . . . . . .

2.4. WHY WE NEED A COMPUTER-AIDED CONCEPTUAL

DE S I GN TOOL O O O O O O O O O

KNOWLEDGE-BASED SYSTEMS . . .

3.1. HIERARCHIAL CLASSIFICATION . . . . .
3. 2. KNOWLEDGE ACQUISITION . . . . . . . .
3.3. REPERTORY GRID TECHNIQUE . . . . . .
3.3.1. Components of the Repertory
3.3.1.1. Elements . . . . . . .
3.3.1.2. Constructs . . . . . .
3.3.2. Grid elicitation . . . . .
3.3.3. Grid Scoring . . . . . . .
3.3.4. Grid Analysis . . . . . . .
3.3.4.1. Difference Measure (di

3.3.4.2. Similarity Value (SV)
3.3.4.3. Cluster Analysis . . .
3.4. CONSTRAINT BASED REPRESENTATION . . .
3.4.1. Eliciting the CBR . . . . .
3.4.2. Scoring the CBR . . . . . .
3.4.3. Rules and Comments from CBR
3.5. KNOWLEDGE REPRESENTATION . . . . . .
3.5.1. Decision tree structure . .
3.5.2. Matchers . . . . . . . . .

3.5.2.1. Building the Matchers

3.5.2.1.a.
30502010b.

Primary Matcher
Secondary Matcher

3.5.2.2. Matcher questions and

confidence factors

3.5.2.3 Weight factors . . .
3.5.2.4. Degree of Match . .
3.5.3. Comment window . . . . .

vii

 

U.

O O O O O O O O OVO O O O O 0O O O O

r'd

1

\INH

12
12
13
19

22

7.0.

8.0.

9.0.

10.0.

METHODOLOGY . . . . . .
4.1. Elicitation . . . .
4.2. Display charts . .
4.3. Scoring . . . . . .
4.4. Display tree . . .
4.5. Transfer . . . .

ANALYSIS OF RESULTS .
5.1. EXPERT f 1 . . .
5.2. EXPERT f 2 . . .
5.3. EXPERT # 3 . . .
5.4. CLASSIFYING A CASE

DISCUSSION . . . . . . . .
6.1. KNOWLEDGE ACQUISITION
6.2. MULTIPLE EXPERTS . .
6.3. REASONING EXPLANATION
6.4. AUTOMATION . . . . .

ENGINEERING SIGNIFICANCE .
CONCLUSIONS . . . . . . .
FUTURE RESEARCH . . . . .
9.1. AUTOMATION . . . .

3:3: ASSEMBLER . . . . . .

BIBLIOGRAPHY . . . .

EXPANDING THE HIERARCHY

DRAWBACKS

Appendix A Excerpt from COMMERCIALIZATION

Appendix
Appendix
Appendix D KNOWLEDGE ACQUISITION FROM EXPERT # 3

Appendix

viii

E SAMPLE CASE CLASSIFICATION .

PLAN

B KNOWLEDGE ACQUISITION FROM EXPERT # 1

C KNOWLEDGE ACQUISITION FROM EXPERT # 2

58
58
61
62
63
67

72

84

98
106
111
112
115
118
119
121
123
127
127
127
128
129
133
141
150
162

170

Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table

Table
Table

Table
Table
Table
Table

Table

Table

Table

Table

Table
Table

1.
2.
3.
4.
5.
6.

9.

10.

11.

12.
13.

14.

15.

16.

17.

18.

19.

20.

21.

22.
23.

LIST OF TABLES

Activated Sludge design criteria . . . . .
Element elicitation for a repertory grid .
Construct elicitation for a repertory grid
Sample grid - scoring procedure . . . . .
Sample SV Table . . . . . . . . . . . . . .
Portion of a primary matcher, Expert # 1
Element 1. . . . . . . . . . . . . . . . . .
Design constructs with associated feature
constructs and similarity values (SV) . .
Sample secondary matcher for Expert # 1
-sludge age . . . . . . . . . . . . . . .
Sample comment window for node E 1,
conventional complete mix reactor. . . . . .
Repertory grid for activated sludge node
completed by Expert #1 . . . . . . . . . . .
Constraint based representation rules for
Expert # 1 . . . . . . . . . . . . . . . . .
Sample primary matcher for Expert # 1 . . . .
Secondary matcher for Expert # 1 Sludge age
> 15 days . . . . . . . . . . . . . . . . . .
Expert #1 - Design constructs with associated
feature constructs and similarity values(SV)
Repertory grid for activated sludge node
scored by Expert # 2 . . . . . . . . . . . .
Constraint based representation rules for
Expert #2 . . . . . . . . . . . . . . . . . .
Primary matcher for element 1 from Expert # 2
- conventional plug flow . . . . .
Expert #2 - Design constructs with associated
feature constructs and similarity values

(SV)
Repertory grid for contaminated ground water
scored by Expert # 3 . . . . .
Expert #3 - Design constructs with associated
feature constructs and similarity values

(SV) . . . . . . . . . . . . . . . . . . . .
Primary matcher for element 1 from

Expert # 3- . . . .
Secondary matcher from Expert # 3 . . . . . .
Questions and responses for a sample case. .

4

ix

58
61
63
64
68
68
69
71
74

81
82

82

83

87

93

95

96

100

102

103
103
106

Table

Table

Table

Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table

Table

24.

25.

26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

39.

40.

41.

42.

43.

44.

Questions and responses for the secondary

matcher regarding sludge age for a sample

case. . . . . . . . . . . . . . . . . . . . . 107
Questions and responses for the secondary matcher
regarding dissolved oxygen concentration for a

sample case . . . . . . . . . . . . . . . . . 107
Primary matcher with confidence factors

and degree of match for Element 1 -

Conventional completely mixed reactor. . . . 109
Degree of match for all primary matchers for

a sample case. . . . . . . . . . . . . . . . 110
Degree of match for all primary matchers

including weight factors for a sample case . 110
Environmental software for conceptual

design. . . . . . . . . . . . . . . . . . . . 137
Primary matcher for Element 1 from Expert # 1

- Conventional completely mixed reactor . . . 146
Primary matcher for element 2 from Expert # 1

— Step feed, plug flow reactor . . . . . . . 146
Primary matcher for element 3 from Expert # 1

- Deep shaft design . . . . . . . . . . 147
Primary matcher for element 4 from Expert # 1

- sequencing batch reactor . . . . . . . . . 147
Primary matcher for element 5 from Expert # 1

- pure oxygen complete mix reactor . . . . . 148
Primary matcher for element 6 from Expert f 1

- oxidation ditch . . . . . . . . . . . . . 148
Secondary matcher for Expert # 1.

- Sludge age > 15 days . . . . . . . . . 149
Secondary matcher for Expert # 1

- Plug flow regime . . . . . . . . . 149
Secondary matcher for Expert # 1

- Hydraulic retention time > 10 hours . . . . 149
Secondary matcher for Expert # 1

- High MLSS Concentration . . . . . . . . . 149
Secondary matcher for Expert # 1

- Dissolved 02 > 4 mg/l . . . . . . . . . . 149
Secondary matcher for Expert # 1

- Soluble BOD removal > 90 % . . . . . . . 149
Primary matcher for element 1 from Expert # 2

- conventional plug flow . . . . . . . . . 156
Primary matcher for element 2 from Expert # 2

- conventional completely mixed reactor . . . 156
Primary matcher for element 3 from Expert f 2

- step feed, plug flow . . . . . . . . . . . 157

 

Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table
Table

Table

45.

46.

47.

48.

49.

50.

51.

52.

53.

54.

55.

56.

57.

58.

59.

60.

61.

62.

63.

64.

65.

66.

67.

68.

Primary matcher for element

4 from Expert f
- deep shaft design . . . . . . . . . . . .
Primary matcher for element 5 from Expert #
- contact stabilization . . . . . . . . . .
Primary matcher for element 6 from Expert #
high rate . . . . . . . . . . . . . . . . .
Primary matcher for element 7 from Expert f
- pure oxygen . . . . . . . . . . . . . . .
Primary matcher for element 8 from Expert #
- extended aeration . . . . . . . . . . . .
Secondary matcher for expert # 2
- cost and space factors . . . . . . . .
Secondary matcher for Expert # 2
- Soluble BOD removal > 90 % . . . . . . .
Secondary matcher for Expert # 2
- Dissolved oxygen concentration > 4 mg/l .
Secondary matcher for Expert # 2
- Primary settling tank . . . . . . . .
Secondary matcher for Expert # 2
- Sludge age > 15 days . . . . . . . .
Secondary matcher for Expert # 2
- Plug flow regime . . . . . . .
Secondary matcher for Expert # 2
- Space loading . . . . . . . . . . . . .
Secondary matcher for Expert # 2
- Nitrification . . . . . . . . . . . . .
Secondary matcher for Expert # 2
- Suspended BOD Removal, ETC. . . . . . . .
Primary matcher for element 1 from
Expert # 3 - aerobic fixed film . . . . . .
Primary matcher for element 2 from
Expert # 3 - activated carbon . . . . . .
Primary matcher for element 3 from
Expert # 3 - ultraviolet oxidation of H202
Primary matcher for element 4 from
Expert # 3 - air stripping . . . . . . . .
Primary matcher for element 5 from
Expert # 3 - anaerobic GAC fluidized bed .
Secondary matcher for Expert # 3
- Removal rate to 1 ppb . . . . . . . .
Secondary matcher for Expert # 3
- low temperature . . . . . .
Secondary matcher for Expert # 3
- contaminant destroyed . . . . . . . . .
Secondary matcher for Expert # 3
- Flow rate > 15 MGD . . . . . . . . . . .
Primary matcher with confidence factors and

degree of match for Element 1 -
Conventional completely mixed reactor . .

X1

eNeNeNONON

157

158

158
159
159
160
160
160
160
160
160
161
161
161
161
167
167
168
168
169
169
169

169

171

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

69.

70.

71.

72.

73.

74.

75.

76.

77.

78.

79.

Primary matcher with confidence factors and

degree of match for Element 2 -

Step feed, plug flow reactor . . . . . . . . 171
Primary matcher with confidence factors and

degree of match for Element 3 -

Deep shaft design . . . . . . . . . . . . . 172
Primary matcher with confidence factors and

degree of match for Element 4 -

Sequencing batch reactor . . . . . . . . . . 172
Primary matcher with confidence factors and

degree of-match for Element 5 -

Pure oxygen complete mix reactor . . . . . . 173
Primary matcher with confidence factors and

degree of match for Element 6 -

Oxidation ditch . . . . . . . . . . . . . . 173
Primary matcher with weight factors,

confidence factors and degree of match for

Element 1 - Conventional completely mixed

reactor . . . . . . . . . . . . . . . . . . . 174
Primary matcher with weight factors,

confidence factors and degree of match for

Element 2 - Step feed, plug flow reactor . . 174
Primary matcher with weight factors,

confidence factors and degree of match for

Element 3 - Deep shaft design . . . . . . . . 175
Primary matcher with weight factors,

confidence factors and degree of match for

Element 4 - Sequencing batch reactor . . . . 175
Primary matcher with weight factors,

confidence factors and degree of match for

Element 5 - Pure oxygen complete mix reactor 176
Primary matcher with weight factors,

confidence factors and degree of match for Element
6 - Oxidation ditch . . . . . . . . . . . . . 176

xii

Figure
Figure
Figure
Figure

Figure
Figure

Figure

Figure

Figure
Figure
Figure
Figure

Figure
Figure
Figure
Figure
Figure

8.

10.
11.
12.

13.
14.
15.
16.
17.

LIST OF FIGURES

Hierarchy for conceptual design of
environmental engineering systems. . .

Refined search space of conceptual design

hierarChy O O O O O O O O O O O O O O

Flow diagram for proposed meth dology of
knowledge acquisition and representation .

Sample of constraint based representation

scoring. . . . ... . . . . . . . . . .
Beginning of FOCUS structure diagram .

Beginning of decision tree and example of

a pruned branch in a decision tree. .
Portion of original grid and CBR chart
offered for revision to resolve a

conflict in the tree structure . . . .
Constraint based representation scored
Expert #1. . . . . . . . . . . . . . .
FOCUS structure of Expert # 1. . . . .
Decision tree for Expert # 1. . . . .
Pruned decision tree for Expert f 1. .
Scored constraint based representation
Expert # 2. . . . . . . . . . . . .
FOCUS structure for Expert # 2 . . .
Decision tree for Expert # 2 . . . .
Pruned decision tree for Expert # 2
FOCUS structure for Expert # 3 . . .
Decision tree for Expert # 3 . . . .

xiii

 

eeeeeeI-heeeea’e

'<

O
H e e e e

34

59

63
65

66

67

76
79
80
80

89
90
91
92
104
105

1.0. INTRODUCTION

1.1. MISSION STATEMENT

The objective of this study is to develop a knowledge

acquisition methodology that will lead to knowledge

representation in the Hierarchial Classification tool. The

knowledge acquisition methodology outlined attempts to model

the map which an expert must traverse to reach a conceptual

design decision. The methodology must satisfy the following

criteria:

1.

The domain is environmental engineering conceptual
design, or the selection of technology(s) for

waste remediation and treatment.

The repertory grid is used as the knowledge

acquisition technique.

The process can be automated. A computer program

can be written to execute the methodology.

The knowledge captured from experts by the
methodology can be transferred into matchers of

the Hierarchial Classification (HC) tool.

When the HC tool classifies a case, the reasoning
can be explained as to why a node establishes or

rejects.

1.2. OVERVIEW

Environmental engineering design consists of three steps -
conceptual, preliminary and final design. Conceptual design
is the identification of appropriate technology or sequence
of technologies for the remediation of compounds in a waste
stream which exceed regulatory standards. Preliminary
design involves the selection and sizing of technology
components, while final design involves the equipment and

operational specifications.

Conceptual design in wastewater treatment can be compared to
medical diagnosis. Individuals share common symptoms when
affected with same disease organism. Medical diagnosis is
the task of identifying these shared symptoms, which then
dictates the prescribed treatment. Diagnosis in the case of
environmental engineering means identifying the
contaminating agents and developing an appropriate treatment
technology (or sequence of technologies) for the removal or
treatment of the undesirable compound(s). Using collective
analytical parameters, commonalities are observed within the
same industry or the same industrial process. Analytical
test results, site evaluation and economic factors all
contribute to the choice of technology implemented for

‘treatment.

Engineers responsible for the conceptual design task make

complex, difficult decisions when they evaluate systems for
wastewater treatment. This process requires the
understanding of diverse and changing legal requirements,
and complex chemical and biological interactions.
Frequently, their choice is among many competing
technologies. It is difficult to discriminate among
competing technologies in a rational manner since

engineering expertise is diffuse and specialized.

As more technologies develop, so does the dilemma
confronting engineers and regulators. Experts in conceptual
design of environmental engineering systems are faced with
the challenge of a developing and changing domain.

Selection of the most efficient and cost effective
remediation scheme to comply with regulatory standards
becomes complex and difficult. To maintain expertise in
this discipline, an expert must constantly incorporate new
knowledge in the areas of technology, regulations and basic
scientific research. The popular environmental solutions
change over short time periods due to emotional and
political influence. Since this domain is moving forward at
such a rapid pace, it is nearly impossible to maintain
expertise except in a very narrow scope. Therefore, in
response to new regulations and technological breakthroughs,
experts in conceptual design should ideally alter their

decision making process to incorporate the latest

developments. Unfortunately, their inability to do so
results in conflicts and omissions when recommending

remedial solutions.

One aspect of this dilemma centers on the conflict between
conventional and innovative methods of treatment.
Conventional methods are expensive, subject to increasing
restrictions, and are not effective for certain wastes.
Innovative technologies are likely to be cheaper, faster and

more efficient, but are not widely implemented.

A knowledge-based system could provide relief from this
dilemma for those making conceptual design decisions. An
expert system to assist environmental engineers with
conceptual design offers a mechanism to evaluate and select
the most effective and efficient remediation system.
Specifically, a system for computer-aided conceptual design
of environmental engineering systems could provide answers
to these problems by providing assistance to engineers in
the selection and design of technologies and regulators in

the evaluation of engineered designs.

Knowledge-based systems placed at the disposal of the
decision makers can enhance their effectiveness (Klein,
1990). The computer-aided design tool for conceptual design

has such potential. Data and models are used to recognize

and formulate the problem, and ultimately evaluate
alternative solutions. There is a need for a system which
can process data and numeric relationships, and by
reasoning, transform this data into opinions, judgments,
evaluations and advice. Knowledge-based systems (KBS)
fulfill this need by formalizing and automating the expert's

paradigm and decision-making process.

The proposed system employs two knowledge-based tools, the
hierarchial classifier and the assembler, to efficiently
select appropriate treatment technologies and to permit
rapid and efficient assembly of treatment trains that
satisfy the treatment goals. The hierarchial classifier
acts as a heuristic filter, eliminating technologies that do
not satisfy the goals of the problem. The assembler takes
the output of the classifier and presents a sequence of
technologies (as many as feasible and required) to satisfy a
set of goals while simultaneously satisfying any other

problem constraints.

This thesis describes a methodology for knowledge
acquisition useful for knowledge representation in the
Hierarchial Classification tool. The domain is conceptual
design for environmental engineering systems, specifically
wastewater treatment. First, basic concepts are defined

such as experts versus novices, paradigms, and why there is

a need for such a tool. Then, knowledge acquisition and
knowledge representation are discussed, with descriptions of
the two techniques used- repertory grids and constraint
based representation. A step-by-step methodology is
presented with three examples. Finally, a discussion and

concluding remarks close this presentation.

1.3. APPROACH

The product of this knowledge acquisition methodology is
knowledge representation at the design level of the pre-
determined hierarchy shown in Figure 1. Theoretically, any
of the designs which branch from the design level in the
hierarchy are appropriate solutions if the technology could
be designed and operated with unlimited funds, space and
supplies. However, not every alternative is practical to
implement. The goal of the expert system is to model an
expert's conceptual design decision for the most effective

alternative.

Complex problems are easier to solve if partitioned, or
decomposed into parts. First the smaller parts are solved,
and then recombined for a result or decision (Kidd, 1987,
Wright & Bolger, 1992). To model the decision-making
process, the manner of decomposition must be identified.
Kidd (1987) suggests that a methodology should specify a
strategic division of tasks. In the case of conceptual
design, the assumption is that the problem is decomposed on
the basis of design criteria. When selecting a technology,
in effect, the values of design criteria are selected which
will most effectively achieve the goal of treatment. The
conceptual designer assigns values to the design criteria
based on the nature of the waste, effluent standards, space

constraints and resources of the client. Then he selects

Table 1.

Activated

design criteria.

 

Design option

P103
FLO“

EXTENDED
AERATIOI

CONTACT
STABILIZATION

 

  

 

Design
Criteria

Units

  

 

 

 

 

  

 

  

  

 

Organic pgggg§_§gg 0.05- 0.2 - 0.05 - 0.2 - 0.6 1 - 0.4 -
loading pounds HLSS-day 0.06 0.4 0.2 2.5 1
Space pgggg§_§gg 3 < 60 20 - 40 < 15 25 - 50, 1.5 - 120
loading day-1000 ft 60 - 75 3.5

1 hydraulic hours 4 - 36 16 - 36 1 - 2 1 - 3

retention

  

 
 

 

 

due

5 - 15

12 - 30

20

 

 

 
 

WLSS in grams
. reactor liter

1.2 - 3

10

  
 

 

  

} Dissolved milligrgmg < 2 < 2 < 2 < 2 6 - 10
‘ oen liter

 

 

    
   

 

 

percent

85 - 95

   
 

 

 

 

percent

50 - 150

 

 

   

 

25 - 75

 

 

50 - 300

 

 

    
 

50 - 150

 

 

 

 

 

 

‘ HLSS a mixed liquor suspended solids, 800 = biochemical oxygen demand

 

   

 

 

 

 

 

IREAT COMPOUND
OVERSTANDARD
MEDIA

AIR

QHABAQTERIZA I IQN

 

 

I WATER l

I

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

COLLOIDAL
OR
DISSOLVED
I REDUCE OXIDIZE HYDROLYZE
INORGANIC
ORGANIC
MECHANISM J l \
PHOTO- l CHEMICAL BIODEGRADABLa OZONE 1
CHEMICAL OXIDATION OXIDATION
OXIDATION K
AGENT / \‘
GENERIC
IMETHANOGENS I METHANO- AEROBIC RSELSBZAETES LANXAE'EEDE
TROPHS HETEROTROPHS
DE I N \
....va
swoee gases:

 

 

 

 

 

 

Figure 1. Hierarchy for conceptual design of environmental

engineering systems.

the technology or method which incorporates the design
values assigned. When choosing from competing design
applications of, for example, the many variations on
activated sludge, the engineer may refer to a chart similar
to Table 1. The most effective technology design will be
the one with the best match to the design values assigned by

the engineer.

When the knowledge acquisition methodology reveals these
differences of design as the expert views them, then the
subsequent knowledge representation could model the process
by which the decision is made. Toward this goal, the
proposed methodology is based on the decomposition of the
conceptual design problem by design criteria values. A
four-step approach to knowledge acquisition and
representation is proposed by Kidd (1987). First, component
terms are elicited with a repertory grid in a familiar
language. Second, criteria that the expert uses to organize
concepts are identified. Next, a structure is formed
through analytical techniques such as cluster analysis.

This structure is represented and then transformed into
rules and a framework using constraint based representation
and the raw grid. Finally, the results are expressed in the

matchers within the Hierarchial Classifier.

To classify a case, select a treatment design for a waste

10

stream, an engineer could use an expert system. The
proposed expert system poses questions about the waste
influent and effluent and other relevant factors to
determine appropriate design values for remediation. These
questions are determined from two knowledge acquisition
techniques - the repertory grid and constraint based
representation (CBR). A map drawn of the internal decision
tree structure indicates the expert's internal organization
of knowledge. This expertise is then transferred into a
matcher, the knowledge representation unit, for each
technology represented as a goal state in the Hierarchial

Classifier.

In a fully functional expert system, comment windows will
explain the reasoning behind the decisions presented by the
expert system along with related design specifications.
Each expert may see these criteria differently, which will
be expressed in the final expert system built based on the

knowledge acquisition and representation.

11

2.0. BACKGROUND

2.1. DEFINITIONS OF TERMS

s e s t' - a knowledge acquisition
technique, expert responds to groupings of concepts
independent of context, identifies association and conflict
among constructs

QQDELIQQES - bipolar characterization Of elements in a
repertory grid, ie. design criteria

EIQEEDEE - represent the examples or instances within domain
in a repertory grid, ie. technologies

Ee;e§11eh_egg_;efine - process of eliminating inappropriate
solutions to the problem early in the hierarchy

EEEQILISQ - goes beyond just knowledge and experience by
incorporating the process by which the knowledge and
experience are structured and accommodated for use in
decision-making.

Wow - a step-wise problem solving

technique which acts as a filter eliminating categories
which are not appropriate for achieving the goal, hierarchy
of nodes or steps the end node is the solution

Kneglegge_eeggiei§ieg - communication and capture of

expertise - goal to obtain and express the rules which the
expert uses in the decision-making process

Megehe; - table of rules or questions in a particular
sequence, according to the answers to the questions,
applicability is based on a calculated degree Of match which
when compared to a pre-determined threshold determines if
node will establish

Reper;e;y_§;ig - a knowledge acquisition technique, designed
to help experts recognize and verbalize the breakdown of a
problem, provides a forum for the expression of the picture
of how a person views their world with a minimum of observer
bias

W - FOCUS - provides exhibition of

structure and pattern of the elements and constructs,
identify which elements and constructs are highly correlated
according to the expert

12

2.2. EXPERTS VERSUS NOVICES

Wright & Bolger (1992) offer an objective definition Of an
expert as "anyone who demonstrates significantly more valid
judgement than persons or systems not accredited with
expertise". Expertise is certainly more easily recognized
than defined (Hart, 1986). An expert's perspective reveals
an enlightening Observation - domain knowledge is not
adequate for expertise (Wright & Bolger, 1992). It is not
just what an individual knows, but how that knowledge is
absorbed, processed, accommodated and structured, which in
turn determines how they use what they know (Hart, 1986,
Wright & Bolger, 1992). Expertise goes beyond just
knowledge and experience by incorporating the process by
which the knowledge and experience are structured and
accommodated for use in decision-making. Therefore,
decision-making and expert judgement are skills which can be
learned through enhanced intuition and creativity (Wright &

Bolger, 1992).

A decision can be defined as an irrevocable allocation of
resources under the control of the decision maker (Henrion,
1991). Preferences describe a decision maker's relative
ordering of the desirability of possible states Of the
world. Successful decision analysis includes techniques for
structuring problems, computing implications, analyzing

sensitivities, and explaining results. Personal experience,

13

heuristics and paradigms along with knowledge collectively
narrow the search space for decision-making. When a person
has created an internal structure for knowledge storage and
retrieval, and can perform in a manner which others judge
successful, they are considered to have expertise, and are
recognized as an expert. It is this internal structure

which expert systems attempt to model and emulate.

Problem solving is the process of transforming an initial
situation into a desired situation, or a goal state found by
carving a path through the search space. Experts operate
from a search space which is their personal internal
(mental) representation of a problem where problem-solving
takes place (Klein, 1990). Individual behavior influences
problem solving depending upon the structure of the problem.
In a task with less structure, the individual experience has
a greater influence. Therefore, by lending more structure
to the task, the solution will be based more on Objective

knowledge, rather than subjective opinions.

Klein (1990) identifies three development stages from novice
to expert. The first stage, cognitive, involves the
encoding of declarative knowledge which is interpreted by
general procedures. Next, in the associative stage, the
novice compiles knowledge into procedures specific to the

performance of the task at hand. Attention is required in

14

the formulation of a solution to a problem, but knowledge is
no longer stored in declarative form. Finally, expertise is
achieved in the autonomous stage. Procedures become
automated, and decisions are made without attention to the
process itself. A recognized drawback, and sometimes noted
as the evidence of expertise, is that verbal reporting of
the thought process is lost. A knowledge engineering
paradox is that the more competent domain experts become,
the less able they are to describe the knowledge they use to
solve problems (Ford, 1991). Knowledge acquisition attempts
to model the autonomous stage by decoupling the declarative

domain 'facts' from the rules used by the expert.

Wright and Bolger (1992) go on to say that experts exhibit a
number of psychological characteristics which are not
expressed in novice decision-making behavior. Some
significant differences between novices and experts are 1)
the definition of the initial state, 2) how the operators
are used to navigate the problem space, and 3) what

information is considered relevant and important.

Experts develop a calculative plan which pre-selects what to
do at each stage with the shortest path to the goal with the
least difficulty. Operators, as rules and heuristics, guide
the search through the problem space. Some critics of

experts say that these very rules and heuristics which are

15

essential to their expertise often limit and restrict the
search space to the point that the best or even appropriate
solutions are overlooked. This is especially prevalent in
domains where the knowledge base is changing rapidly. Also,
some alternatives will be eliminated from the solution set
early on in the decision process depending on the sequence

that the heuristics are brought to bear on the problem.

The ability to simplify complex problems with their superior
pattern recognition abilities is another common trait shared
by experts across many domains. By making sense out of
chaos, they can overcome adversity by identifying and
adapting to exceptions. One strategy involves attacking
smaller pieces of a problem and making continuous
adjustments (Chorafas, 1990). This results in less critical
errors which are correctable, thereby avoiding big mistakes.
The way in which the problem is decomposed also differs from
experts and novices. Experts group together items which
share a commonality which is not recognized by the novice.
This is a crucial element of expertise which knowledge

acquisition techniques express.

There is a cognitive science difference between novices and
experts (Wright & Bolger, 1992). Experts develop a
perception of a limited domain from which they extract

information well. Experts predefine their informational

16

needs, where novices recognize them as they progress through
the problem-solving task. They can sense what is relevant

when making decisions through pattern-recognition.

Experts predefine the informational needs, and categorize
according to principles. First they select the equations
useful for the solution. This, in turn, prompts more
knowledge that determines the equations useful for the
arriving at the solution. Now the expert has the knowledge
or data required for a decision or conclusion. Described as
forward thinking, these processes known as protocol analysis
(Wright & Bolger, 1992), are automatic and functionally
independent. Protocol analysis for raw data is the skill
component of expertise which expert systems attempt to
model. Therefore, experts follow forward reasoning by
utilizing stored functional units of knowledge from the

given data to the goal.

In contrast, novices use means-end analysis (Wright &
Bolger, 1992). They examine the data, and choose the
appropriate equations, and attempt a solution. The
acquisition of appropriate schema for forward thinking is,
therefore, restricted. The novice will examine the goal
state and reason backward, then select the goal which best
fit the path. Deliberate reasoning with each step dependent

on the previous forecasts the‘direction to continue.

17

Novices develop patterns through deductive thinking during
the process Of evaluation, rather than recognizing the

pattern as experts do.

Judgment is an inferential cognitive process by which an
individual draws conclusions about unknown quantities or
qualities on the basis of available data (Wright & Bolger,
1992). However, Often not all the data required is
available. Experts and novices differ in the manner they
handle uncertainty. Sometimes the novice will tend to
immediately remove himself from the process, claiming the
situation is out of his knowledge domain. They express a

desire to avoid judgement mistakes.

Experts allow for the flexibility of uncertainty, and are
willing to risk a decision as they are confident in their
expertise. Experts will assume that they can synthesize a
judgment when knowledge is insufficient (Wright & Bolger,
1992). With uncertainty, there is the increased tendency to
generate additional alternatives which must be evaluated.

To handle uncertainty, experts allow intermediate degrees of
truth between true and false for factors being considered
(Henrion, 1991). Heuristic methods used in dealing with
uncertainty allow experts to continue through the search
space in pursuit of the solution.

Experts have a highly developed faith in their own ability

18

(Wright 6 Bolger, 1992), and are effective, efficient
problem-solvers (Hart, 1986). Their self confidence often
manifests itself in arrogant behavior. They know when to
adapt the decision strategy, and moreover realize when the
problem is outside the realm of their expertise. Timing is
the key to recognizing when the conditions have changed.
Experts have a strong sense of responsibility for their
decisions, and will seek help from others as they recognize
that isolation can lead to inferior decisions (Wright 6
Bolger, 1992). They acknowledge other experts, and respect
the contribution from them in areas where their own

knowledge is lacking.

2.3. PARADIGMS

Three keys to the future are anticipation, innovation and
excellence according to Barker (1992). All three are
necessary for success in the next century. One important
component of excellence is the capability Of how to do the
right thing the first time. Innovation is the way
excellence is incorporated into action. Anticipation
provides the information that allows for the excellent
innovation to be implemented properly. All are relevant in

the area of environmental engineering conceptual design.

Corporate and individual entities facing environmental

compliance search for the best, most cost effective and

19

reliable means to remediate and treat their waste problems.
They turn to experts in the regulatory and consulting
community for assistance in the appropriate technology and
design for their situation. These experts, as in many
domains, Operate within paradigms. A paradigm is defined by
Barker as a set of rules and regulations (written or
unwritten) that does two things: 1) it establishes or
defines boundaries; and 2) it dictates behavior inside those
boundaries in order to be successful. There is an
interrelationship of several paradigms which guide an
expert's decision making and ability to solve problems.
There are scientific paradigms as well as cultural,
specifically regulatory, in the case of environmental

engineering applications.

Paradigms are the heuristics or rules used in solving
problems. New technologies and regulations attempt to
change the paradigms from which we Operate. When paradigms
change, the rules change, and the problem solving exercise
results in different solutions. In environmental
engineering, both the scientific and regulatory paradigms
are changing rapidly. The regulatory paradigms are not
optional - laws must be obeyed. However, not so with the
scientific. They are subject to individual evaluation and
opinion and certainly are not all universally accepted.

Herein lies the controversy of appropriate treatment

20

technology. While the regulatory paradigm changes uniformly
across both the consulting and regulatory community, the
scientific paradigm occurs in all stages from the most

conservative to the most progressive.

Dealing with incomplete or uncertain data can lead to new
paradigms. Although we would assume that only experts have
the knowledge to develop new ideas, sometimes the novice
comes forward with revolutionary concepts. This is because
their search space is not so narrowed as to eliminate
possibilities from consideration. Unfortunately, the novice
will seldom gain acceptance, or even an audience, where the

expert will receive more immediate recognition and respect.

Every expert will, in the process of finding new problems,
uncover problems he cannot solve. And those recalcitrant
problems provide the catalyst for a paradigm shift, which
alters the rules of the game. This is necessary when there
are problems to solve, and there are no reasonable
solutions. In the environmental arena, the conventional
methods have solved many problems, however, there are
increasing hazardous waste situations which conventional
means are too expensive and inefficient. Innovative methods
are a solution, they just need to be implemented. The
proposed computer-aided tool for conceptual design assist in

the acceptance of new paradigms in the approach to

21

environmental problems.

2.4. WHY WE NEED A COMPUTER-AIDED CONCEPTUAL DESIGN TOOL
The conclusions of a commercialization plan for a computer-
aided conceptual design tool found in Appendix A indicate a
need for this product (Schutz-Riley, 1992). A survey of
competitive computer programs revealed an unfulfilled niche
for a decision-making program which could discriminate among
competing remediation technologies. The distinguishing
features of the proposed Hierarchial Classifier tool are
that it is an expert system with decision making capability,
draws knowledge from interfaced databases, and includes
innovative technologies. Unique to the proposed system is
its portability to all major PC'sdue to the use of the

Smalltalk 80 language .

Experts develop a framework from which they can be
successful. Two paradigms, the scientific and the
regulatory, make up the framework from which the conceptual
design expert Operates. Rules of the paradigm, or
heuristics, are applied to narrow the search space for the
sake of making it manageable. Although good decisions are
made from within this narrowed search space, often
applicable alternatives are removed from consideration early
in the decision process. The commercialization plan

(Schutz-Riley, 1992) identified five heuristics implemented

22

in the choice of remediation: 1) regulatory, 2) total cost,
3) past personal experiences, 4) knowledge about
environmental contaminants, and 5) knowledge about
technology alternatives. The order in which these
heuristics are applied to a problem can greatly affect the
solution. Above is the sequence imposed most often in
conceptual design decisions in environmental engineering,
where regulatory issues drive the process. Implied are the
state and local regulatory climates which dictate the
probability Of government approval for implementation. The
final decision is not based on the best scientific evidence,

but rather on political possibility.

While perhaps not scientifically sound, this is the path of
least resistance. Expertise is more easily acquired based
on the regulatory climate and past experience than on the
scientific evidence. A paradigm based on regulatory
heuristics is more stable and embraced by a wider audience.
The risk of being wrong is less when dealing with
uncertainty and incomplete data as is often the case in
environmental problems. The solution Of choice is the
alternative which appears most cost effective and achieves

regulatory approval in the least amount of time.

Consequently, environmental consultants have become expert

in preliminary design rather than conceptual design. That

23

is, they can optimize the design which is most acceptable
from the regulatory standpoint to fit the remediation
problem, rather than optimizing the choice itself. This
behavior is reinforced by the regulatory agencies and legal
community. Thus, expertise develops with the past
experience and regulatory heuristics influencing their
decisions. Consultants gain expertise in predicting the
regulator's judgement, and base their decisions on this line
of reasoning. Consequently, the best scientific solution is

often not even considered as a viable Option.

The scientific paradigm is unstable. Many paradigm pioneers
are challenging the rules of the scientific paradigm by
developing new technologies to deal with the hazardous
wastes which contaminate our earth. These researchers are
willing to take the risks necessary to invest their time and
energy into ideas for which the outcome is uncertain. They
have been successfully rewriting the rules of the scientific
paradigm at a very rapid pace. The fruits of their efforts
are many new and exciting innovative ideas for coping with
pollution especially in the realm Of environmental

biotechnology and advanced oxidation techniques.

While consultants attempt to overcome regulatory obstacles,
the scientific community attempts to overcome natural

physical, chemical and biological obstacles. Alternative

24

methods of remediation are emerging through innovative
technological development. However, these methods are not
readily implemented. One reason is that regulators often
lack the scientific expertise needed to evaluate these new
methods, and there may be insufficient experience to risk
implementation. Consultants are hesitant to recommend
innovative methods if regulatory approval is unlikely. The
implementation of promising new technologies can only be
expected when the new scientific paradigms are embraced by

the regulatory community.

One could suggest that experts simply change the way they
make conceptual design decisions by changing the sequence in
heuristics. Even though the regulatory world is changing,
the scientific world is changing at an even faster pace.
Therefore, the landscape of the expert's search space is
also changing in an attempt to incorporate new knowledge.
The re-evaluation Of the search space by replacing new rules
for old to accommodate this new information is very
difficult. Rather than take the more difficult path of re-
evaluation, many consultants instead will strengthen their
present paradigm to block this process. The result is a
less than optimum knowledge base from which decisions and

judgements are made.

What if there was a computer tool available which contained

25

the rules of the new scientific paradigm and knowledge to
fill the gaps of uncertainty? It is likely that those
making conceptual design choices would be willing to accept
a broader range of possible answers to remediation problems.
Simply by considering a change in the sequence in which the
heuristics are applied to a conceptual design study, the
probability of implementing new innovative methods improves.
If such a scientific-based paradigm begins with the
contaminant characteristics and technology alternatives, a
wider assortment of methods will be presented. Regulators
might recognize alternative methods to the conventional
treatment regimes for many remediation problems. In some
cases, more money may need to be spent in waste
characterization and bench-scale testing to verify a
decision, however the cost of implementation and Operation
has proven time and time again to be favorable. The result
is overall cost reduction while achieving the same or better

level of remediation.

There are many factors which are unpredictable and
undetectable with respect to technology performance. This
is especially true with biological treatment processes. SO,
of course, bench and pilot scale testing is always
recommended prior to full scale implementation. To reduce
the cost and enhance the effectiveness and applicability of

bench and pilot scale tests, the design parameters can be

26

estimated and defined with the assistance of an expert
system. The purpose of the proposed conceptual design tool
is to limit the alternatives using design criteria that are

most appropriate for a specific waste stream.

The following statement (Schutz-Riley, 1992) summarizes the
sentiment of potential customers of a knowledge-based expert

system for conceptual design.

"The problem with implementing innovative technologies
is not awareness that they exist, but the confidence
that they will perform. Detailed information regarding
design, costing, and performance is needed to overcome
this insecurity. until conventional methods become too
expensive, they will remain the method of choice."

27

3 . O . KNOWLEDGE-BASED SYSTEMS

Knowledge-based systems (KBS) are computer programs which
have a wide base of knowledge in a restricted domain, and
use complex inferential reasoning to perform tasks which a
human expert could do (Hart, 1986). Rich, (1991) refines
this definition to include only tasks executed by computers
which presently are performed better by people. Expert
system technology improves human decision making by
formalizing human expert knowledge (Henrion, 1991). KBS’s
have a wide base of knowledge in a restricted domain in
which they use complex inferential reasoning to perform a
task which human experts do well. They simulate expert
problem solving by processing data and numeric
relationships, and then by reasoning, transform data in to

opinions, judgement, evaluations and advice (Klein, 1990).

Experts are effective, efficient users of their knowledge
(Hart, 1986), and a good KBS will reflect these traits.
Building the KBS model is only the beginning. For an expert
system to be judged a success, it must also perform
successfully in providing acceptable answers to the same
problems currently solved by experts. Sound justification
for problem-solving process by the KBS will lend confidence

for it's use.

Experts develop a superior strategic awareness of

28

possibilities within the search space, or problem space
(Wright 6 Bolger, 1992). The structure of the task
determines the structure of the search space. Search is the
process of finding a path through the internal structure.
Experts calculate a plan which preselects what to do at
teach stage of the path, and ensures that successive steps
follow (Wright 6 Bolger, 1992). They select the plan with

the shortest path to the goal with the least difficulty.

An informed search is performed using heuristic search
functions that are operators which guide the search. A
heuristic is a rule of thumb, or judgmental technique that
leads to a solution some of the time, but provides no
guarantee of success. The purpose of heuristics are to
reduce the number of alternatives from an exponential number
to a polynomial number. This allows the designer to obtain
a solution in a tolerable amount of time. Instead of
exploring every possible situation, the search is narrowed
and only a limited number Of alternatives are examined.
This is a constrained rather than exhaustive search. The
risk is that the heuristic may eliminate the best answer to
the problem. However, a good heuristic will identify good,
workable solutions most of the time using heuristic
information, which is information about the problem used to

guide the search more efficiently.

29

An important characteristic of expert systems, KBS's, is the
ability to explain reasoning and justify the solution
provided. When a node establishes within the hierarchy, a
comment window indicates what data and criteria are
responsible for the matcher assigning a confidence factor
which exceeds the threshold, therefore establishing the
node. Additional design considerations, consequences and
anticipated events are also included in support of the final

decision.

3.1. EIERARCEIAL CLASSIFICATION

According to Parsaye 6 Chignell (1988), when the number of
eventual selections is large, the decision making will be in
a hierarchial form. This is certainly the case with
decisions in conceptual design. Hierarchial classification
efficiently compares a set of pre-enumerated categories with
particular situation to find those categories that "best"
apply. The hierarchial classification approach can and has
been used successfully in a number of problem-solving roles
including areas of medicine, nuclear power and chemical
engineering. As part of the conceptual design process,
hierarchial classification is used as a heuristic filter,
pruning categories which are not appropriate for achieving

the goals of the problem.

Categories are organized into a hierarchy shown in Figure 1.

30

The connected nodes at the next level down in the hierarchy
represent a subcategory of the node and the connected node
one level up in the hierarchy represents a super-category of
the node. Categories become more specific as the hierarchy
is traversed from top down. A node can be thought of as an
expert in determining if the category it represents is

relevant to the problem at hand.

Each node of the hierarchial tree contains a matcher. The
matcher is a structured representation scheme which relates
small chunks of knowledge together as a unit. This
simplifies processing operations since knowledge required is
usually contained within the node itself, or is easily
accessed as a unit from another Object through just a few
linkages. Execution of a matcher, or matching, is the
controlled sequence of operations to determine the best
alternatives according to stored rules which consumes a

large portion of the processing time of execution.

The function of the matcher is to calculate the 'degree of
match’ between the category and the present problem.
Questions about the problem are answered by the user
classifying a case. The questions within the matcher are
representative of expert's decomposition of decision-making
strategy. Questions reflect the heuristics which guide the

search through the internal structure of the expert. These

31

answers are the input to the matcher, which returns a
computed degree of match. For environmental technologies,
the degree of match depends upon the chemical and physical
characteristics of the contaminated media, the technology or
class of technologies under evaluation and other practical
considerations such as space and cost. Confidence factors,
assigned based on the results of knowledge acquisition, are

additive, and contribute to the calculated degree of match.

The path throughout a search space is defined by the nodes
that establish. The establish and refine process is a way
of eliminating inappropriate and inapplicable solutions to
the problem as early as possible in the hierarchy. The more
data and information known about the problem, the stronger
the confidence in the returned value. The degree of match
is compared to a pre-determined threshold to determine if
the node will establish or reject. This ultimately reduces

the number of goal states to be evaluated.

Figure 2 shows an example of a defined search space of
conceptual design. The shadowed boxes represent nodes which
have established. Further search toward the goal is limited
to sub-categories of shadowed boxes. The other paths will
not be considered, as they lead to inappropriate
conclusions. For example, at the mechanism level of the

hierarchy, the node for 'oxidized by ozone' did not

32

establish, which implies that this mechanism will not be
effective in achieving the remediation goal. Therefore, it
is not necessary to explore any of the sub-categories of
this node. By limiting alternatives, the overall search
restricted to potentially appropriate solutions, and is

therefore, more efficient.

33

 

 

SUSPENDED
MATTER

    
  
     

INORGANIC

 

M NI

 

PHOTO-
CHEMICAL
OXIDATION

 

 

 

 

 

\_

SULFATE
REDUCERS

 

 

 

 

 

MIXED
ANAEROBE

DESIGN
TRICKLING ‘ ROTATING
FILTER BIOLOGICAL PONDS
CONTACTOR

Figure 2. Refined hierarchy for conceptual design of
environmental engineering systems.

34

 

 

 

 

 

 

3.2. KNOWLEDGE ACQUISITION

Knowledge acquisition is largely a matter of communication,
beginning when experts in some domain determine they have
valuable information to share (Bradshaw, 1992). The goal of
knowledge acquisition is to Obtain and express the paradigm
from which the expert operates. Kidd (1987) proposes two
functions of knowledge acquisition, the elicitation and
analysis of data. Data elicitation refers to what data is
needed and how it will be used. Analysis refers to the

transformation of data into an interpretive framework.

Knowledge acquisition techniques provide a means of
capturing and representing expertise within the Hierarchial
Classifier. One difficulty in constructing expert systems
is that experts often cannot explain what they do or how
they use their knowledge to reach conclusions. Although
experts are skilled at pattern recognition and communicating
results, they lack the verbal protocols to identify the
decision process itself (Klein, 1990, Wright 6 Bolger,
1992). Tools such as the repertory grid and constraint
based representation overcome this difficulty and allow the

expression of expertise from some experts.

Kidd (1987) defines knowledge acquisition as the
transformation of data into an implementation formalism.

Kidd, along with other knowledge engineers (Gruber, 1987 and

35

Wright 6 Bolger, 1992) regard knowledge acquisition as the
bottleneck in building expert systems. Although many
dispute the label "bottleneck" when addressing the knowledge
acquisition issue, they agree that effective acquisition of
knowledge is crucial to building an effective KBS. Gruber
(1992) maintains that the design of a KBS should anticipate
the acquisition process and make it easy for experts to
express their knowledge. Further, some recognize the
problem of knowledge acquisition as a representation
mismatch. The primary contribution of acquisition design is
to enable the expression of knowledge in a more
comprehensible and accessible manner. Without the proper
model for expression, the expert will not fully be able to

divulge his knowledge in a representable fashion.

An analysis of successful knowledge acquisition tools
according to Gruber (1987) suggests that they satisfy two
primary requirements. First, they must identify the type of
knowledge to expect from the expert. Then, they should
provide a functional mapping from the user input to
representation in the knowledge based system. When the
underlying architecture supports representation in the first

step, then the second step is simplified.

A review of knowledge acquisition techniques for expert

systems is offered by Welbank (1987) and Neale (1988).

36

Klein (1990) identifies three strategies of knowledge
acquisition: 1) expert-driven, 2) machine-driven and 3)
knowledge engineering-driven. The expert-driven strategy is
Observed when the expert encodes their own expertise. Since
many experts cannot verbalize their decision making process,
this strategy is not effective as a general solution (Wright
6 Bolger, 1992). The second approach is machine-driven
techniques involving machine learning. A criticism of this
method is that the limiting Operating structure includes
irrelevant knowledge and omits necessary information for
appropriate results. The resulting knowledge representation
scheme do not produce decisions which accurately reflect

those of the expert.

Knowledge engineering-driven approaches include direct
interaction with experts to model the task and performance
of the decision process. Methods include the interview,
protocol analysis, repertory grid, constraint based
representation, and others (Chorafas, 1990). With complex
decision-making, there is a problem with observation
methods, such as interviews and protocol analysis according
to Wright 6 Bolger (1992). Experts Often arrive at
decisions quickly. Due to their rapid combinatorial
assessment of information, they seldom can provide details
of the process. Researchers of knowledge acquisition have

found that methods such as interviews only cover possible

37

considerations concerning a particular case, without
providing the scope for the rules which apply to the domain
of expertise. Methods such as repertory grids and constraint
based representation overcome these problems by addressing
the full scope of the decision domain, and are therefore
accepted as more reliable methods in the field of knowledge

engineering.

A combination of knowledge engineering-driven and machine-
driven techniques appears to offer the best strategy for
knowledge acquisition. Repertory grids and constraint based
representation can be used to gather the knowledge, and a
computerized program assists the expert in the arrangement,

modification and final representation of knowledge.

3.3. REPERTORY GRID TECHNIQUE

The repertory grid (also referred to as simply ’grid') is
based on the personal construct theory first introduced by
George Kelly in 1955. Kelly’s theory proposed that people
partition experiences into constructs which make up a model
of their world. They classify and categorize experience and
knowledge, and then develop theories about their world.
Their behavior and decisions are directed by these theories
because they anticipate events and act based upon their

expectations.

38

Shaw (1981) describes grids as "the seductive promise of
accurate measurement of subtle perceptions". Grids appear
simple at first, however, through proper design and
elicitation using skill and sensitivity, they can be a
powerful knowledge acquisition tool. The grid accomplishes
several tasks including eliciting distinctions, decomposing
problems, combining uncertain information, incremental
testing and integration of data types (Boose, 1989). They
are most applicable for analysis problems, or portions of
synthesis problems that can be reduced to analysis problems.
Boose (1989) discusses the use of several repertory grid-

centered tools.

Since grids provide a forum for the expression of the mental
map of how a person views their world with a minimum of
observer bias, they offer improved process efficiency and
faster knowledge base generation than other knowledge
acquisition tools (Hart, 1986). When experts attempt to
verbalize their decision-making process, they often have
trouble identifying the key point which drives the result.
In an attempt to isolate this point, they may become
confused and embarrassed because their thought process
suddenly appears without structure or basis. The grid
captures the structure of the problem from the expert’s
perspective in a way which he cannot verbalize. Grids are

particularly effective in domains where experts can detect

39

subtle nuances that differentiate concepts, but cannot
articulate the criteria by which they arrive at a decision
(Shaw 6 Gaines, 1983). Through grid analyses, the structure
and protocols are extracted in a manner which is both

comfortable and non-threatening for the expert.

3.3.1. Components of the Repertory Grid

The repertory grid is composed of elements and constructs.
The grid is designed to explore thought patterns by
eliciting constructs which define the elements of a person's

individual world.

3.3.1.1. Elements

Elements are the choices an expert must make, and are the
examples or instances within the domain of interest.
According to Hart (1986), elements must be representative of
the pool from which they are drawn. The choice is
determined by the reason for the investigation. Assuming
that expertise is in a narrow domain, the number of elements

will be limited.

Stewart 6 Stewart (1981) offer some guidelines regarding the
selection of elements. Elements should be specific,
discrete and precise as possible to allow for clear
construct elicitation. The elements do not necessarily need

to be evenly distributed across the search space, however

40

they should not be sub-sets of each other. The outcome of
the grid analysis is dependent upon one expert's
understanding and definition of the elements compared to

that of another expert.

3.3.1.2. Constructs

Constructs are distinguishing characterizations which are
shared by the elements to some degree. Diaper (1989)
describes constructs as mental "tools" which allow a person
to discriminate between elements in one's world. Constructs
are a way of transcending the obvious - lending structure to
our outlook of our world. A construct is a way of
describing how two or more things are alike, and therefore

different from a third or more things.

Kelly asserted that people never affirm anything without
simultaneously denying something else. Hence, constructs
are bipolar, allowing a matrix of the pattern of
interrelationships between elements. The bipolarity resides
in the construct itself, not in the two sets of elements
that are sorted by the construct (Hart, 1986). A detailed
discussion of the bipolarity of the construct is offered by

Diaper (1989).

3.3.2. Grid elicitation

The elicitation of elements and constructs is not trivial,

41

but rather complex and iterative. Although the process
appears simple, the design and elicitation Of grids requires
sill and sensitivity (Shaw, 1981). The knowledge engineer
must state the clear objective for analysis. Level and

extent of expertise will shape the grid to be analyzed.

There are three strategies for selecting elements for a grid
(Stewart 6 Stewart, 1981). Elements can be supplied by the
interviewer, which is helpful when several expert responses
are being compared. The other options involve the expert
supplying the elements, either directly or by elicitation

using a predetermined sequence of questions.

The recognition of elements leads to elicitation of
constructs. Kelly (1955) introduced a method for eliciting
constructs known as the triad method. This method is based
on the personal construct theory of how constructs are
formulated in a person's mind. The expert is asked to
compare three elements, and specify how two are alike and
thereby different from a third. The contrasting pole is
identified to ensure the constructs are bipolar. As the
triad comparison progresses, more constructs and elements
are added to the lists used to create a repertory grid. The
triad comparison continues until the expert does not

identify any new elements or constructs for the grid.

42

According to Hart (1986), two important factors must be kept
in mind with regard to grid elicitation. First, the
elements must be within the range of convenience of the
constructs, and secondly, judgment should not be forced onto
the expert. Kelly proposed that a construct always operates
within a context and that there are a finite number of
elements to which it can be applied. The range of
convenience of a construct is described by the number and
kind of elements that can be rated on it. Not every element
can be rated on every construct. Some constructs have a
very wide range of convenience, such as good-bad. A
construct such as high oxygen transfer rate versus low
oxygen transfer rate will not have as wide a range.
Therefore, some constructs will not be applicable or
relevant to all of the elements in a grid. The expert
should not be forced to make a judgement on the relationship

between elements and constructs where none exists.

3.3.3. Grid Scoring

Elements are rated according to each construct using one of
three scales. Elements can be ranked against each other, or
rated along a continuous scale. Rating scales have been
used from a two-point to a nine-point scale. There is
evidence to suggest that a seven-point scale is getting
close to most people’s limits of discrimination, and much

above five points is very difficult to examine visually

43

(Stewart 6 Stewart, 1981). Ranking may force the indication

of differences when in fact they don't exist.

A third technique calls for a positive or negative response,
without indicating range. For fast visual inspection, a two
point scale indicated by symbols of colors is useful. The
trade-off is simplicity and speed for detail. A dichotomy
rating scale allows for the expression of differences

without the notion of scale.

3.3.4. Grid Analysis

Analysis of a grid provides the exhibition of structure and
pattern of the elements and constructs. The cluster
analysis technique known as FOCUS is a way of measuring the
distance between a pair elements or a pair of constructs by
summing the absolute differences between ratings for all
pairs (Shaw, 1981, Hart, 1986). FOCUS identifies how the
elements and constructs relate to each other by calculating

the difference measures and similarity values.

3.3.4.1. Difference Measure (dij)

Elements are compared by calculating the difference in their
scored ratings. The sum of the differences in the scored
ratings of two elements for all of the constructs (dij) is

calculated by the following formula:

44

dij = :ElCl-EZCl: + :E2C2-E3C2: + ... :rn,1cj-Eicj:

where i number of elements

number Of constructs

u.
II

To calculate the difference between constructs, sum the
differences between two constructs for all elements (dji) as

follows:
dji = lElCl-E1C2: + :E2C1-E2C2: + :EiCj_1-E1Cj:

where i = number of elements

j = number of constructs

3.3.4.2. similarity value (8V)
The dij value reflects the dissimilarity in scores between
two elements or constructs. The maximum dij value is the
maximum distance in the rating scale times the number of
constructs. A dissimilarity fraction is calculated by the
formula: V
3..
k * l.

where maximum distance in the rating scale

= number of constructs or elements

haw

For example, the maximum distance in the rating scale, k,

45

for the rating scale -1 to +1 is k = 2. The maximum dij

value us k * l.

A similarity fraction, or similarity value (SV) is

calculated by the formula:

SV = 1 - dij

To remove the decimal point, multiply by 100. Two elements
or constructs are considered similar if the SV is greater

than 50 (Hart, 1986).

3.3.4.3. Cluster Analysis

The FOCUS technique is a means of identifying elements and
constructs that are similar. The original grid is evaluated
by calculating a difference measure (dij) and similarity

value (SV).

Correlations are made between elements or constructs, and
are visually displayed in a tree structure. Elements in the
completed grid can be FOCUSed to derive the tree structure
using the following procedure which is a modification of the
original FOCUS technique introduced by Shaw (1981) and
described by Hart (1986).

1. Calculate the difference measure (dij) between all

the scored ratings for the pairs of elements.
Create table of dij values.

46

7.

Calculate a similarity value (SV) for each dij.
Create a table of SV values.

Examine SV table for highest value greater than 50
(Hart, 1986). Fuse the two elements involved,
creating a new element. Assign a new element
number to the pair of elements which comprise this
SV. The new element will contain all the
constructs which the two fused elements scored the
same. If one element scored "0", use the "-1" or
"+1" from the other in the new fused element. If
one element scored "-1" and the other scored "+1",
then remove that construct score from the new
fused element. In subsequent dij and SV
calculations, the number of scored constructs, the
value of 1, will reflect the eliminated
constructs.

Excluding the two elements which were fused into a
new element, find next highest value in the SV
table.

a. If a higher SV exists between an unfused and
fused element pair, proceed to next step. If
not, then fuse this new pair, and assign a
new element number.

b. If there is a tie, that is, two pairs of
elements have the same SV value, the three
elements are fused into one.

Repeat step 1-2 using new fused elements with any
remaining elements. Adjust the i, j, and 1 values
accordingly when calculating the dij and SV.

Create a new raw grid, dij table, and calculate
new SV's from the new dij table. Build onto the
tree as in Step 3 above.

Repeat steps 5-6 until all elements are 'fused'.

3.4. CONSTRAINT BASED REPRESENTATION

Constraint based representation (CBR) Offers an effective

approach to knowledge acquisition, and is more flexible than

the interview or grid technique (Gammack, 1989). An expert

is allowed to respond to groupings of concepts rather than a

47

sequence of independent questions. When the outcome of a
decision procedure depends on features, Often the order of
presentation affects the outcome. This is true in the
conceptual design process, as priorities and existing
conditions often dictate the solution. Constraint based
representation provides explicit representation of the
probabilistic dependence and independence the design
constructs (Henrion, 1991). Unrealistic combinations are
removed from consideration, and strong associations are

given priority in the decision-making processes.

Questions are posed to the user in all knowledge acquisition
tools. In many expert systems, these questions may be
irrelevant to the final decision but they are required by
the structure and predetermined flow of control. Users are
put off by irrelevant questions, and will tend to discredit
the system when this occurs. Experts tend to restrict and
define the search space so that irrelevant questions are not
asked. This is observed especially in domains where

solutions are reached by data or event-driven strategies.

Elicitation of knowledge supporting the decision-making
process may be a problem using conventional techniques.
When expertise consists in rapid combinatorial assessment of
relevant information, Wright contends that these techniques

are ineffective in identifying heuristics to constrain the

48

search. In CBR, forward-chaining reasoning is constrained
to narrow the search space, and then backward chaining is

used to test the result (Gammack, 1989).

Constraint based representation provides a venue to capture
and represent this problem-solving approach. The number of
decisions the expert must make is reduced by pruning the
decision tree that represents the internal structure for the
expert. By grouping in this manner, the total number of
individual responses gleaned from the process are more then

the expert actually gave (Gammack, 1989).

3.4.1. Eliciting the CBR

From the design constructs elicited in the repertory grid
technique, a triangle chart is developed. The expert is
asked to set a range of probable values for each design

construct, and then divide this range into three sub—ranges.

3.4.2. Scoring the CBR

The expert is asked to compare two design constructs at a
time. For each of the three pre-selected sub-ranges, the
expert will indicate a strong relationship between the pair
of design values using a "+" for strong positive
relationship, and "-" for strong negative relationship. Not
every square will have a score. A score on the CBR

indicates extreme situations.

49

3.4.3. Rules and Comments from CBR

When an expert has scored a pair of design constructs as "+"
or "-" in the CBR, a strong relationship is indicated
regardless of other design criteria. The results of the CBR
chart are expressed in table form grouped by design
constructs for easy reference. Results of the CBR can be

used for:

1. Pruning the internal decision tree.

Design constructs can be revealed by constraint
based representation. If the tree contains a path
which, according the CBR, is not a consideration,
then the affected branches are deleted.

2. Conflict resolution

CBR can also be used to find any conflicts within
the repertory grid. After the tree is
constructed, the results of CBR can be used to
prune branches which violate the CBR rules which
have been identified by the expert. If a branch
is pruned which leads to an element goal state
from the grid, then the expert is asked to examine
the responses for that element in the grid and the
design constructs in the CBR. The changes made by
the expert can be used to corrected tree
structure, and CBR pruning continues until there
are no violations between CBR rules and the tree
structure.

3. Comment windows

Comments are written from the constraint based
representation as an if/then statement presented
in tabular form. These are summarized and
revealed in the comment window of each node as
they apply.

50

3.5. KNOWLEDGE REPRESENTATION

Knowledge representation is the model built of the expertise
resulting from knowledge acquisition. Knowledge must be
represented in such a way that objective expression of the
expert's decision-making processes are reflected in the

inferential reasoning of the knowledge-based system.

Models are devices used to attain or formulate the knowledge
about a problem, dependent on the context and purpose from
which they are derived (Bradshaw, 1992). The power of
models lies in their ability to function as tools for
thought, also known as cognitive artifacts. The way in
which knowledge is brought to bear upon a problem depends
upon the way in which knowledge is represented within the
model. In the HC model, the results of the knowledge
acquisition process are incorporated in matchers and comment

windows for the nodes of the hierarchial tree.

The internal structure from which experts solve problems can
be shown pictorially as a decision tree. A decision tree
assumes the decision process is serial, composed of
predecisional stages which contribute to the final decision.
Keren (1992) observes that a decision may take place at a
certain point in time, yet is usually preceded by several
predecisional stages. The predecisional stages are

represented in the HC tool in the primary and secondary

51

matchers.

3.5.1. Decision tree structure
Using the following five-step procedure, a decision tree
structure can be formed from grid analysis. Starting with

the node at the top of the cluster analysis (CA) diagram,

(1) identify the branch with lowest SV score
between two elements,

(2) list the design criteria constructs by which
these two elements differ,

(3) create two branches using poles of the
applicable design criteria,

(4) examine the next lowest SV and repeat steps
1-3.

(5) Continue steps 1-4 until the CA diagram is
redrawn with the appropriate design construct
at each branch Of the tree. If a design
construct was identified at a lower SV
branch, it should not be repeated in a branch
with a higher SV.
Branches that violate CBR rules can be pruned by applying
the above procedure. If the pruning results in the
elimination of an element, then the expert must re-examine

the responses responsible for this conflict between the grid

and the CBR chart. Any changes should then be incorporated.

3.5.2. Matchers
The purpose of a matcher is to represent the pattern of

decisions used by the expert.‘~Rules identified in the CBR

52

and grid are represented as matchers at each node within the
hierarchical classifier of the KBS. The user is presented a
series of questions, and the answers are the input into the
matcher. This input is represented as the confidence
factor, derived from the concept of certainty factors used

in the MYCIN program (Buchanan and Shortliffe, 1984). The

output is a degree of match between the answers given for '
each feature and the goal state being evaluated. a
Matchers are independent units based on local information e

 

(Punch, 1991). The output from one matcher does not affect
the other matchers in the hierarchy. Changes in the
queries, similarity values and weight factors within the
matcher do not necessarily influence the operation and
responses returned from others. This allows the expert to

make adjustments to each matcher independently.

3.5.2.1. Building the Matchers

Two levels of matchers are used in the proposed HC, primary
and secondary. The primary matchers calculate the degree of
match from user input. There are three possible answers to
the questions stored within the matchers - yes, no or don’t
know. A confidence factor is assigned to yes or no answers.
If the answer is "don’t know", then the secondary matchers
return a confidence factor to the primary matcher. The

value of confidence factors and degree of match ranges from

53

-1, indicating a low confidence factor or degree of match,
to +1, indicating a high confidence factor or degree of

match.

3.5.2.1.a. Primary Matcher

The primary matcher for a node is derived from the design
constructs shown in the FOCUS structure. These design
constructs make up the first column of the matcher and are
listed in order of ascending SV. Rows in the matcher
provide the response from the user for the corresponding
design criteria. The sign of the confidence factor (CF) is
taken from the tree structure for each node indicated by the
elements in the original grid. A confidence factor of 1 is
returned with a yes or no response to the question which

returns a +1 or -1 as appropriate.

3.5.2.1.b. Secondary Matcher

If the user response to a design construct is "don't know",
then by cascade condition, the user is presented with a
secondary matcher. The confidence factor computed from the
secondary matcher is returned to the primary matcher for the
corresponding design construct. The maximum absolute value

returned is 1.

3.5.2.2. Matcher questions and confidence factors

A four-step procedure is outlined to derive the matcher

54

 

questions and assign confidence factors.

1.

Create a table of dij and SV's for constructs as
was done for the elements. Consider the portion
of the table with design constructs on the
horizontal axis and feature constructs on the
vertical axis.

Examine the 1st branch in the tree from the
original node. Identify the design construct(s)
responsible for the split in this node. To do
this, go back to the grid of ’fused’ elements and
determine which design construct scored
differently. The SV value of the pair determine
the sequence of design constructs in the primary
matchers.

If no design construct is different, then ask the
user to compare the two fused elements and
identify a design construct for which they do
vary. Incorporate the new construct into the
original grid, score, reFOCUS and continue.

The secondary matchers are derived from the SV
table for constructs. For each design construct,
list the feature constructs with a SV greater than
60 in descending SV value. (An SV greater than 50
indicates similarity, (Hart, 1986). A value of 60
is chosen to reflect a strong degree of similarity
between constructs. The user of the system is
asked a question based on the feature. The
feature constructs are presented in order of SV,
with the greatest value first.

The confidence factor (CF) is calculated as a
function of the SV and the ratio of number of
rules considered to number of rules available.
The sign of the CF is taken from the grid score.
If the answer to the rule is yes, then the sign of
the SV is positive, if the answer to the rule is
no, then the sign of the CF is negative. If the
score was "0", ask the expert which sign would
apply to the design construct. (This response
does not change the score on the original grid,
only affects the construction of the secondary
matcher.) The expert has the option of deleting
feature constructs from the secondary matcher. A
formula algorithm (Boose, 1986) is used to
calculate the CF for each combination in the
table. If the computed CF has an absolute value
greater than 1, default to a value of 1 OR -1.
The confidence factor returned to the primary

55

3.5.2.3.

matcher is computed based on the user responses to
questions generated from the feature constructs.
The algorithm for calculating CF is as follows:

a)

b)

C)

CF

where: n

Begin with rule with largest SV with yes
(Y) answer. (A high SV indicates a
strong correlation between constructs.
The first rule in the CF computation has
the most influence, therefore it should
be the construct with the highest
correlation, or SV.)

Divide SV by number of rules available.
For each additional rule to be

considered, in sequence of greatest SV
first, continue:

= mend +[ (n/m) * new SV * (l-CFn-1) 1

number of rules considered
number of rules available
confidence factor
similarity value

111
CF
SV

Weight factors

The expert has the option to prioritize the design criteria

in each matcher independently. A weight factor from 0 - 1

may be
design
factor
weight
column

degree

3.5e2e‘e

of 1.

of match.

assigned. A weight factor of 0.1 means that such a
criteria is 10 % as important as one with a weight
The confidence factor is multiplied by the

factor for the value returned to the right-hand

in the matcher. This value is used to calculate the

Degree of Match

In} overall degree of match is calculated from the confidence

factors .

The degree of match is computed as a weighted

56

 

average of the values returned from each construct in the
matcher. This is based on the concept of additive value
function introduced by Klein (1990). The ratio of the
weighted average to the best possible value is the degree of

match, with a maximum absolute value of 1.

degree of match = Qfilfll + QEzflg +. . . n-I‘ln
n
where CF1 = confidence factor for design criteria 1
W1 = weight factor for design criteria 1
n = number of design criteria in matcher

Based upon a pre-determined threshold value assigned by the
user classifying a case, the computed degree of match will

cause the node to either establish or reject.

3.5.3. Comment window

A summary of the rules invoked to establish or reject each
node are provided in the comment windows. Any applicable
CBR rules are listed, along with other design criteria
selected from the primary matcher. A second window will
outline the feature constructs which relate to the primary
matcher. This provides the explanation for the expert

systems' reasoning in arriving at the particular goal state.

57

4.0. METHODOLOGY

A flow diagram for a proposed methodology of knowledge
acquisition in environmental engineering conceptual design
is shown in Figure 3. The responses from Expert # 1 are

used in the examples provided in the following sections.

4.1. Elicitation

This procedure is adapted from methods developed by Kelly
(1955), Kidd (1987), Shaw (1981), and Stewart 6 Stewart
(1981). In the following sections, instructions for
elicitation are provided. Underlined titles correspond to
the steps in flow diagram Of the proposed methodology shown

in Figure 2.

Qheoee noge to expang. Begin with a node whose children are

design considerations of the parent. Refer to Figure 1. In

the example, the node to expand is activated sludge.

Lis; Qesige epplications which make up the children nodes of

the node to be expanded. This is element elicitation. The
elements make up the headings of the repertory grid as in

Table 2.

Table 2. Element elicitation for_a repertory grid

 

 

   

 

I ELEIEITS seoueucrgc
AT

 

 

 

 

 

 

 

 

 

[Choose node to expand

I

 

 

 

I List elements 7 Iterate
I I

 

I List design constructs

II

I List feature constructs

I

I Create repertory grid

I

I Create CBR chart

I

 

 

 

 

 

DISPLAY
TREE

 
 
 
 

 

 

 

I Score the grid 3 Iterate
* ' I

 

I Score the CBR chart

I

I Create tree structure

I

I Prune tree

I

Build primary matcher

I

Build secondary matcher

I

I Create comment window

Figure 3. Flow diagram for proposed methodology of knowledge
acquisition and representation.

 

 

 

DISPLAY
TREE

 

 

 

 

 

 

 

 

 

59

  

  
   

Iterate

 

Iterate

 

Iterate

Lie;_;he_ezi§e;ie or parameters used in design. Compare two

design applications to a third, and name a design criteria
which is the same for the first two, but differs from the
third. This is elicitation of design constructs. The
constructs must be bipolar in nature, with an extreme pole
indicated on the grid. The list of design constructs is

placed in the first column of the repertory grid.

.Lie§_fiee§g;ee of the waste problem which describe the
influent, effluent, and operational characteristics of the
system. This is accomplished using the triad technique.
Compare two design applications to a third, and name a
characteristic which is the same for the first two but
differs in the third. This is elicitation of the feature
constructs. Add the feature constructs to the previously
elicited design constructs as shown in Table 3.

Example: Step feed and deep shaft have high cost of
construction, where the oxidation ditch has a
low cost of construction.

Conventional mix and pure oxygen systems can

tolerate shock loading, where step feed
cannot.

60

Table 3. Construct elicitation for a repertory grid
==

 
   

 

 

 

 

 

 

 

Ems CGIPLETE ST P DEEP SEOUENCIIIQ PURE OXIDATION
El! FEED SHAFT BATCH OXYGEN orrcu
REACTOR REACTOR
E 1
ms
Iiigh cost of 1
_ construction .

I Tolerates shock 2
. loading ‘

 

.Pl _- flow r me 3

‘ Space loading I I.
I > 1.5

E Sludge age 5
' >15 days

' hydraulic 6
retention
time > 10 hours

 

 

 

 

' Soluble 800 7

racval > 90 X
-— ---— — —-— --~~——— =I=III=II===

 

 

 

 

 

 

 

I
l

4.2. Display charts

gregte e repertery grid with the elicited elements and
constructs. The constructs are separated into two
categories, the design and feature constructs. Design
constructs are the parameters of the system where the value
is set by the engineer. Feature construct value are a
function of the waste stream and technology, and change only
as a result of changing a design construct. In other
workds, the feature constructs values are a function of the
design construct choices. Feature constructs are listed

first, followed by the design constructs in the final grid.

Creege e QBR ehart of design constructs. The constraint

61

based design chart presents every possible interaction
between pairs of relevant design constructs. Each
combination of constructs is divided into three partitions
of expression. For each design construct, choose a high,
low and intermediate level in appropriate units. The pairs
of constructs considered systematically by the expert. Both

incompatible and compatible combinations are indicated.

4.3. Scoring
Seoze ghe gepertery ggig by assigning a "+" if there is a
positive relationship between element and construct, a "-"
if there is a negative relationship between element and
construct, and "0" if neither extreme applies. Each box of
the grid must be scored. Typically, each of the design
constructs will score a "-" or "+", rarely a "0". If a "0"
is scored, the user will be presented with the box a second
time to verify the scores. See Table 4 for an example.
Example: According to this expert, a conventional
complete mixed reactor can tolerate an
influent with fluctuating flow rate, so the
expert scored a '+'. Deep shaft design
cannot tolerate such fluctuating flows, so a
'-' was scored. A sequencing batch reactor

can tolerate minor fluctuations, so the
expert scored a '0' in the appropriate box.

62

 

 

 

 

Tolerates
L'“‘“991!991

 

 

 

 

 

 

 

See;e_§he_§§3 chart by indicating a '+' for strong positive
relationship between design criteria for the ranges
indicated. Use '-' for a negative relationship. If there
is no direct or strong relationship, leave blank. See

Figure 4 for a sample of CBR scoring.

F1 F2 F3 N1 N2 N3 Ct C2 CS

 

< 70 % 81 + °/. soluble

BC“)
”0‘90 9" 82 removal

>90% 33 .- +
<1d8y C1

 

 

5.15 daysC 2 Sludge
> 15 daysC 3

 

 

<50% N1
% TKN
so - 80 % N 2 removal

>90% N3

 

 

 

kav

Regime

F1 = Plug

F3 = Complete mix

Figure 4. Sample of constraint based representation
scoring.

4.4. Display tree

63

Q:ee;e_;he_§;ee structure from cluster analysis of the grid
by using a modification Of the FOCUS technique introduced by
Shaw (1981). From the FOCUS analysis, the internal
structure of the expert's internal decision tree is formed.
The elements are focused and fused according to the
procedure outlined previously. See Table 5 for a sample SV

table.

Example: Examine the SV table from the original grid
from Expert # 1 from calculations shown in
Appendix B.

Table 5. Sample SV Table

ELEMENTS 2 3 1 E Q

1 48 58 72 72 82
2 86 64 60 50
3 70 70 60
4 60 82
5 54

E 2 and E 3 with an SV of 86 will be fused to become E
7. E 1, E 4 and E 6 with a score of 82 will be fused
to become E 8. The next highest SV is 72, but since
this score is with fused elements, ie. E 2 and E 4,
they will not be fused at this level. Fused grid # 2
is created.

The FOCUS structure diagram has SV on the vertical axis and
elements on the horizontal axis. For expert f1, begin as
shown in Figure 5. Continue to fuse elements as shown in
Appendix B to create a complete tree structure based on the

design constructs.

64

SV
86--

 

E2 '53

Figure 5. Beginning of FOCUS structure diagram.

A decision tree is created from the FOCUS tree. Begin with
the lowest SV in the FOCUS tree. This SV was calculated
from the dij of two elements. Review the scored constructs
between the two elements and identify the design constructs
which scored differently. Identify the design constructs
responsible for each branching in the tree. Construct a

decision tree structure representation as shown in Figure 6.

EXample: The branch between E 2 and E 3 is due to a
different score on design constructs 19 -
soluble BOD removal > 90 t. This is found in
the original grid.

65

 

    

 
    

 

I HRT>10houra I

II

Scluble 800 removal >90 °/.

llllll  

 

 

 

 

'N Y .11 .II

Figure 6. Beginning of decision tree and example of a
pruned branch in a decision tree.

£19ne_;ne_§;ee by eliminating branches indicated
nonapplicable from CBR rules, see Figure 6.
Example: From the CBR chart, the expert indicates th.
a sludge age < 15 days will not result in
soluble BOD removal > 90 %. Consequently,
any branch indicating > 90 % removal of

soluble BOD will be pruned below the branch
where sludge age is determined.

IgegeEe to resolve conflicts. If a pruned branch is also an
element from the original grid, there is a conflict in the
responses from the expert. The expert will be presented the

portions of the grid and CBR which correspond to the

66

conflict in the tree. The expert will have the opportunity
to change a response, or override the "prune-effect" of the
CBR.

Example: The element #1 is at the end a pruned branch.
Figure 7 is the display offered to the expert
for revision. The tree is adjusted
accordingly per the experts response.

F1F2F3 616263

 

 

 

 

 

 

 

 

 

.1015 II c Mounts
300
“$33 [mfl
.ms.ul .. ’
. city Ci
4 Design criteria 1
' Attribute ;' $166:wa £00.06
I aIDGeyeOD
; Soltble coo
Flow
We

Figure 7. Portion of original grid and CBR chart offered
for revision to resolve a conflict in the tree structure.

4.5. Transfer

The results from the knowledge acquisition tools are
transferred into the knowledge representation units called
matchers. A primary matcher and associated secondary
matchers are created for each element in the grid. The
sequence of the questions within the matchers are identified

from the similarity values calculated in grid analysis.

agile primary matchegs, Identify design construct(s) which

govern each node, beginning at the top of the tree. Build a
primary matcher for each element with design constructs

listed in the sequence from the tree structure top-down. A

67

confidence factor of 1 or -1 is assigned to the yes and no
responses. The sign of the confidence factor is taken from
the original grid score. A portion of a primary matcher is

shown in Table 6.

Example: For goal state for E 1, the sign for the
response to C 17 is negative, and for C 23 is
negative.

Table 6. Portion of a primary matcher, Expert # 1,

 

 
    
  
  
  

 

 

 

 

 

 

 

 

 

Element 1.
=r-L
oestci comm Imam r I creamer confluence
nuns anus: \MflE

17-Sludge Age > 15 days +1 -1

23-Nitrification occurs +1 -1

users: or urea

s O a c e The feature constructs are used

to build the secondary matcher. List the feature constructs
which correspond to each design construct along with the

appropriate SV. Table 7 is a list for sludge age.

Table 7. Design constructs with associated feature
constructs and similarity values (SV).

mm ms “W

C 17 - Sludge Age Recycle rate

Fluctuating flow

Shock loads, toxic

Risk of short circuiting
Sludge deuctering capability

      
 

 

3%333

 

 

68

Table 8. Sample secondary matcher for Expert # 1
-sludge age

 

 

...... I2I.I.I;|L«
m
5V 4 ++++

 

 

The sign of each SV in the secondary matcher, shown in Table
8, is taken from the fused grids as was the case for the
primary matcher. If both of the fused elements score "0"
for any associated constructs, there are two options. Since
a score of "0" indicates there is not a strong relationship
to either pole of the construct, this construct is not be a
discriminating factor. The expert can specify to either
omit the feature construct from the secondary matcher, or
assign a positive or negative sign as appropriate.
Confidence factors are computed using the formula algorithm
as a function of the SV and the number of constructs being
considered. The computed confidence factor is returned to
the primary matcher as the value for the appropriate design

construct used to calculate a degree of match for the node.

mening_;he_me§ehegee The matchers are presented to the

expert for fine tuning. Any conflicts will be presented to
the expert for resolution. Matchers, both primary and
secondary, can be changed independently to reflect the
decision-making process of the expert. These changes

include the addition or deletion of feature constructs from

69

secondary matchers, and changing the weight factors in the
primary matchers. In the course of fine tuning, the expert
may realize additional feature or design constructs which
should be included in the original grid, or changes to the
CBR. These changes can be made, and subsequently
incorporated into the knowledge representation. Automation
of the methodology becomes very helpful in this phase of the‘

methodology.

W
Q;ee§e_explene§ieg_§e§le for each goal state based on CBR

table and matcher results.

Example: Comment window for E 1 for matcher completed.
The confidence value for the matcher is
computed by adding the confidence values
returned for each design construct, and
dividing by the number of constructs
considered. In this case, the value 0.35 is
returned, indicating a moderate positive
response that E 1 technology would be
appropriate. Note that the user did not
answer yes or not to the questions about
nitrification or flow regime. The
corresponding hidden matchers were presented,
and the indicated response returned.

The comment window for this completed matcher would look as

follows:

70

Table 9. Sample comment window for node E 1, conventional
complete mix react9rs

    

i sludge age > 15 days
1 hydraulic residence time is > 10 hours

TKN removal > 90 X
- Space loading > 75 lb 300/ cu. ft-d
. High HLSS concentration
= Soluble 800 removal > 90 x

71

5.0. ANALYSIS OF RESULTS

Three experts participated in the evaluation of the
methodology. The first two were geared toward the
generating matchers for the design 'activated sludge' as
shown in the hierarchy displayed in Figure 1. Expert # 1 is
an environmental consultant who specializes in wastewater
conceptual design. Expert # 2 is an expert in biological
remediation processes. These two were presented the same
pre-prepared grid. Each made adjustments according to their
individual priorities. The CBR was developed from their

final grid.

Expert # 3 is an expert in conceptual design specializing in
groundwater remediation. An original grid was elicited from
Expert # 3, however, due to time limitations, a CBR was not
completed. This expert was not given a pre-determined
hierarchy, rather asked to begin with a waste stream instead
of a design category. The competing technologies were
identified, and grid elicitation continued as outlined in

the methodology.

A sample case of a wastewater with overstandard BOD
(biological oxygen demand) is classified according to the
results from Expert # 1. Calculations of secondary matchers

confidence values and degree of match are explained.

72

5.1. EXPERT I 1

The repertory grid and constraint based representation chart
scored by Expert # 1 are shown in Table 10 and Figure 8
respectively. Expert # 1 selected 6 design technologies,
elements, as he considers himself to be an expert in
distinguishing among these treatment technologies. The
calculations of the FOCUS analysis are presented in Appendix
B. Figure 9 shows the FOCUS structure derived from these
calculations, and Figure 10 shows the decision tree. Each
level of branching in the tree is determined by a design
construct, or a group of feature constructs. The decision
tree structure for Expert #1 includes the possible
solutions, or elements from the repertory grid. The
location in the tree is based on the design construct
values. Several other possible design combinations are also
identified which the map of the search space suggests that

would be possible alternatives.

The CBR results are presented in Table 11. Based on these
rules, branches of the decision tree can be pruned. The
pruned tree for Expert # 1 is shown in Figure 11. A
conflict is identified when pruning the tree for Expert # 1.
From the CBR, a sludge age of greater than 15 days will have
soluble BOD removal of less than 90 %. Element 1 was scored
as having a sludge age less than 15 days but, is at the end

of a pruned branch. In an automated situation, the expert

73

Table 10. Repertory grid for activated sludge node completed
-13yrylzxgpexrjgujfl..

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

\ ‘ "_“__ _ ’T
swans l WLETE
‘ 1 mx
, l
'I FEAR!!! ms l
3 High cost of 0
construction 1 I
Low power irput 2 3 0 0 0 0 0 0
for aeration :
Low risk of short I + - + O 0 +
‘ circuiti - flow 3 I
High operational ‘ 0 - - - + O
1 costs ‘ 1
Lou space 0 0 + O + 0
‘ requirement 5
tolerates + . - 0 + +
1‘ fluctuati - flow 6
Sludge fouling and - - - - - -
I bulking problems 7
tolerates shock + - - 0 + +
loadi » toxics 8
3 High recycle + ‘ ' 0 ‘ 1'
rate 9
Biodegrades I - - - - - -
suspended son 10 l
1 High oxygen 0 0 0 0 0 +
! transfer rate 11
‘ Yolerates - - - - - -
‘ intermittent flow 12 -
‘ sv: > 100, Sludge ‘ - - - - - -
settleabi l ity 13
sludge easily + + + + - +
deuatered 1!.
High biological 0 0 0 0 0 0
‘ solids prochcmi _ 15 _ l_ __

 

74

 

 

 

 

 

 

 

 

 

 

 

Table 10 (cont'd).
ELEIEITS CDHPLEIE SIEP DEEP s B R* PURE 02 ox.
MIX FEED* SHAFT CWLETE DITCH“
MIX

DESIEI CDHSTIUCTS

High MLSS - + + + - +

concentration 16

Long sludge age, + - - + - +
> 15 days 17

Dissolved oxygen + - + + + +
> 4 lull 18

Soltble 300 + - + + + +
removal > 90 X 19

complete mix + - - + + +
flow regime 20

plug flow regime - + + + - -

21 ,

Long HRT > 10 - - - . . +
hours 22

Nitrification - - - + - +
occurs 23

Denitrification - - - + - +
occurs 24

Primary settling + + + + + +
tank required 25

 

 

 

 

 

 

 

*SIEP FEED = Step feed, plug flow reactor
*8 B R I Sequencing Batch Reactor
*0! DlTCH 8 Oxidation Ditch

75

.F # tmaxm B 388 5sz898. 889 Efizmcoo .m 229m

5.: 22950 . on.
2338.25 . an.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

so: as... . ....
9:52 26E
REES m2 .\ 8A
212 o\. «2 $88
+ PZ ob om v
59.8 me SE v A
82820 0 BE 3
. + 5 :9: o
.2: 968. 3 me A
ddudm. 83m 3 mnmm
S m. v
ww._s_ 9. 868. + 9 5o
ddmdm. £590 ~> o.o.w.o
S 8.?
5.0. m. 87
325 m. 878
: 8v
00 can + no :8 m. A
$25 mo :8 mg
. a.» + 6 E. F v
0 we: no 9:2 A
c2229 «0 a; 3
9.3925 + 5 a; o v
. a. + + mm 26.
mm.=2 3:8 + mm 5285
8283.... + n. ... + .m 82
+ + + -- 8 $8 A
o, + + .9 pm on v
aqua—d ..flfljwduaauzmze Emu...

 

 

 

 

76

would be presented this conflict as shown in Figure 7 in the

methodology section.

Expert # 1 first discriminates according to sludge age, then
nitrification, followed by flow regime, and so on. This
sequence is transferred into the primary matchers. A sample
is shown in Table 12. The complete compliment of primary

matchers is presented in Appendix B.

A sample of a secondary matchers is shown in Table 13. For
each design construct, the dij and SV calculations reveal
the feature constructs with an SV value of over 60,
indicating a high correlation between them, see Table 14.
This is the basis for the secondary matchers. Expert # 1
associates the design construct long sludge age with recycle
rate, fluctuating flow, shock loading, short circuiting of
flow and sludge dewatering. These are the factors which the
expert considers when deciding on sludge age. A sample of
a secondary matcher is shown in Table 14, while a complete
compliment of secondary matchers is found in Appendix B.
Rules derived from the scored CBR are used in the comment

windows as shown in Table 11.

77

62-

72

5V Similarity Value

68w ’

82»
est»

 

ACTIVATED

 

 

 

SLUDGE
C17
023
E
. 9 E 8
C16
020
C21
E
7
C20
C18 021
019 C24
E 2 E 3 E 5 E 1 E 6
Step Deep Pure Complete Oxidation Sequencing
feed shaft oxygen mix ditch
ELEMENTS reactor
CQNSIBQCIS

016 -high MLSS concentration
017 -sludge age > 15 days
C18 -disso|ved oxygen > 4 mg/l
C19 -Soluble BOD removal > 90
%

020 -complete mix flow regime
021 -p|ug flow regime

022 -hydraulic retention time >
10 hours

023 -nitrification occurs

024 -denitrification occurs

 

Figure 9. FOCUS structure for Expert # 1.

78

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Y
7 Sludge age> 15 days
N Y I Nitrification occurs 1 N Y
N
N Y Y N Y N Y
[Plug flow regime
v N . N
N Y N Y N Y N Y N Y N Y
HRT>10hourS j
[Soluble BOD removal >90 %
NYNYNYNYNY NYYN NYNYNYNYNYNYNYNY

        

 

 

 

 

Convantonal sequencing
Step teed Deep shaft complete mix batch
plug flow reactor reactor
reactor '

 

 

 

 

 

 

 

Figure 10. Decision tree for Expert # 1.

79

 

Pruned branches

-It flow regime is complete
mix, then “(N-removal < 50

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

%
-li sludge age > 15 days, N
then soluble 800 removal > y
90% ,
Fludge age> 15 dayj
N Y Nitrification occurs N Y
N Y N I Y N Y N ’ Y
I . I
/ [Plug flow regime /
K Y N < N
N Y N Y N/ \ , Y N Y N Y N/ \Y
I \ AHRT 10h i I \
I \ > ours J I \ Y
i l l l R l
‘ I \ I I\ S l bl BOD 1 90°/ 1 H
\ I \ I |\ on e remova> o H
\ I \ l I \ I I | \ / \ / I I

 

 

NYNYNYNYNYNYNYNYNYNYNYNYNYNYNYNY

 

 

 

 

Conventional sequencing
Step teed Deep shaft complete mix batch
plug flow reactor reactor
‘ reactor

 

 

 

 

 

Figure 11. Pruned decision tree for Expert #‘1.

80

Table 11.

 

Constraint based r r

esentation rules for E rt I 1.
has the value THEI additional I

 

> 15 days

< 1 day

SVl > 100

TKH removal > 90 %

high HLSS concentration
Soluble 300 removal > 90%

low HLSS concentration
Soluble BOD removal < 70%
TKH removal < 50 %

 

 

Completely mixed

TKN removal < 50 %

 

TKN removal > 90 %

Space loading > 75

Lou HLSS concentration
Soluble 000 removal > 90%

Space loading < 15
Soluble 000 removal < 70%

 

 

Organic loading 0.6-1
Sludge age < 1 day
Soluble 000 removal < 70%
Plug flow regime

Dissolved oxygen 0 mg/l

Sludge age > 15 days
HRT > 10 hours

 

HRT < 3 hours

Sludge age < 1 day
Organic loading 0.6-1
Soluble BOD removal > 90%

SV! < 50

HRT > 10 hours
Soluble 800 removal < 70%
Sludge age > 15 days

 

Organic loading < 0.05

 

 

 

 

 

Hydraulic Retention Time > 10 hours -
(HRT) .
< 3 hours -
TKH removal < 50 % -
> 90 % -
HLSS concentration High -
Medium -
Lou -

Table 11 (cont'd).
I Soluble 800 removal < 70 % -
> 90 % -

a

Low HLSS concentration
HRT < 3 hours

Sludge age < 1 day
Organic loading 0.6-1
Space loading> 75
Dissolved oxygen 8 0 mull
TKH removal < 50 %

High HLSS concentration
HRT > 10 hours

Sludge age > 15 days
Organic loading < 0.05
Space loading < 15
Dissolved oxygen Z-b mg/l

 

81

 

 

Table_12. quole orima . matcher f°'_§mf_tt I 1i_.m.__

EESIGI ODISTIUCTS

HEIGHT
FACTOR

 

I St 'fO A.e > 15 days

 

' Nitrification occurs

 

Pl 2 flow regime

 

HRT > 10 hours

 

Denitrification occurs

 

High HLSS concentration

 

1 Dissolved 02 > 6 mg/l

 

 

I Soluble 800 removal §_30 %

DEEHEE 0F HATCH

 

 

 

 

 

Table 13. Secondary matcher for Expert # 1
Sludge age > 15 days

 

run-e 2'9

 

 

 

 

 

82

Table 14. Expert #1 - Design constructs with associated
feature constructs and similarity values (SV).

 

 

 

 

DESIGI ms FEATIIIE ms SV
c 17 -Sludge Age Recycle rate 92
Fluctuating flow 75
Shock loads, toxic 75
Risk of short circuiting 67
Sludge deuatering capability 67
c 23 -Hitrification Recycle rate 92
Fluctuating flow 75
Shock loads, toxic 75
Risk of short circuiting 67

 

 

 

 

 

 

 

0 21 -Flou regime Fluctuating flow 83
Shock loads, toxic 83

Sludge deuatering capability 83

Operational costs 75

Recycle rate 67

c 22 -Hydraulic retention time Recycle rate 92
Fluctuating flow 75

Shock loads, toxic 75

Risk of short circuiting 67

Operational costs 67

Sludge problems 67

Suspended solid biodegradation 67

intermittent flow 67

Sludge settling, SVI 67

c 26 -Denitrification Recycle rate 75
Sludge problems 67

Suspended solid biodegradation 67

Intermittent flow 67

Sludge settlinLSVl 67

c 16 -HLSS concentration Sludge deuatering capability 83
Recycle rate 75

Risk of short circuiting 67

Sludge problems 67

Suspended solid biodegradation 67

Intermittent flow 67

Sludge settling, SVI 67

67

c 18 -Dissotved oxygen concentration Risk of short circuiting 83
Fluctuating flow 75

Shock loads, toxic 75

Space requirement 67

Sludge devatering capacity 67

 

  
  
 
 
  

Table 16 (cont'd).
Risk of short circuiting

c 19 -Soluble 800 removal rate Fluctuating flou

Shock loads, toxic

Space requirement

Sludge deuatering capacity

 

 

 

$3338

83

5.2. EXPERT f 2

The repertory grid and constraint based representation chart
scored by Expert # 2 are shown in Table 15 and Figure 12
respectively. Expert # 2 selected 8 design technologies,
elements, as he considers himself to be an expert in

distinguishing among these treatment technologies.

Calculations for the FOCUS analysis from Expert # 2 is
presented in Appendix C. The resulting FOCUS tree and
decision tree are presented in Figures 13 and 14
respectively. Table 16 shows the rules resulting from the
CBR scored by Expert # 2. These rules were used to prune

the decision tree shown in Figure 15.

The decision tree is developed from the FOCUS structure. As
shown in Figure 13, the lowest SV is 31, and is due to a
collection of feature constructs relating cost factors.
Recall that similarity is indicated by a SV value greater
than 50. In an automated system, Expert # 2 would be asked
for an additional design construct which would discriminate
the Elements 13 and 14. The next discriminating design
construct is soluble BOD removal rate, then dissolved oxygen
concentration and so on. The sequence of design constructs
for Expert f 2 is reflected in the sample primary matcher in
Table 17. The complete compliment of primary matchers is
presented in Appendix C. The secondary matchers for each

84

design construct are built according to Table 18 as was done
for Expert # 1. A complete compliment of secondary matchers

is found in Appendix C.

A score of "0" on the grid presents an interesting
interpretation on the decision tree structure. Element 1
was scored "0" in the constructs which discriminate the
first branch of the decision tree. Since a score of "0"
indicates neither strong positive or negative relativity,
and due to the binary nature of the tree, Element 1 is found
as a goal state in two places in the tree. Classifying a
case where plug flow is a possible alternative would depend
on the confidence value returned by the secondary matcher.
However, analysis indicates that either branch of the tree
could lead to the choice of Element 1 as a solution.
Element 2 was scored "0" on two levels (Table 15).
Therefore, there are four possible goal states on the

decision tree.

When two elements at the base of the tree are not separated
by a design construct, this could indicate that one is a
sub-element of the other. In this example, Expert # 2
indicates that Elements 1 and 3 are not discriminated by a
design construct (Figure 14 and 15). This expert may view
step feed and conventional plug flow as closely associated,

with one being a sub-grouping of the other. Or, upon

85

review, he may add a design construct. In this case, the
grid would be re-analyzed and presented once again

incorporating this additional construct.

Automation of the methodology would allow iterations to
refine the decision tree and resulting matchers. Expert # 2
had an opportunity to review the initial results of the
knowledge acquisition exercise and indicated some changes he
would make for the next iteration. He would add ’uniform
oxygen supply' as a design construct to distinguish between
step feed and plug flow. Recall these two elements were not
separated by design construct, so appeared as sub-elements
on the decision tree. Expert # 2 would change the response
on Element 3 from "+" to "0" for the plug flow regime design
construct. He also would add sequencing batch reactor as an
element. Other changes include designating dissolved oxygen
concentration as a feature construct rather than a design
construct. Other additional design constructs include MLVSS

concentration and oxygen transfer efficiency.

86

Table 15. Repertory grid for activated sludge node scored by

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Expert # 2.
ELEIEITS I Plgg CSTR. STEP DEEP c.s.* HIGH
1 fig! re mun iui
FEATURE COISTRUCTS ‘
High cost of i o o o + o o
I construction 1 I
I
Lou power input 0 0 O + - s
. for aeration 2 l
‘ Low risk of short + 0 0 + O 0
circuiti - flow 3
High operational ' O O 0 + 0 0
costs 4 i
‘ I
Lou area ' O O O + + +
‘ requirement 5 ‘
‘ Tolerates _ O + 0 O O O
I fluctuating flow 6 i
I Removes I + O 0 - + -
suspended BOD 7 1
Tolerates shock ! - + - + + 0
loading, toxics 8 I
High ales E - o o + + +
1 concentration 9 I
‘ High recycle i O O O - + -
rate 10 i
‘ Few sludge bulk/ ' O - O O + +
1 foamingyproblems 11 I
I
‘ High oxygen i O 0 o + o o
; transfer rate 12
Tolerates 0 + 0 0 0 0
intermittent flow 13
SVl < 100, Sludge + - 0 + + .
settleability 14
Ease of 0 0 O - - +
I operation 15
Lou biological ' o o o o + -
solids reduction 16 ‘ I
____ E

 

 

 

 

 

 

 

 

 

 

87

 

I Table 15 (cont'd). . *

 

ELEIEIYS ‘ PLU CSTR STEP QEEP 0.5. HIGH PUR§ §.A,
FL FEED 5 FT RA? 9:

 

 

  
  
   

Single stage 0 - + - O - - +
denitrification 17

Long sludge age, - - - - O - - +
> 15 dgys 18

 

 

Dissolved oxygen - - - + - - + -
> 4 l 19

 

space loading, - - - + - + + -
>1.s k ace/m3 o 20

 

 

 

 

 

flow regime 21
plug flow regime + - + + + + + +
22 I

> 10 hr. Hydraulic - - - - - - - +
retention time 23

I Soluble DOD + 0 + + . - + +
neaoval > 90% 24
Single stage + + O - - - + +
nitrification 25

 

Primary settling I + + + - + + + -
26

 

 

 

 

 

 

 

complete mix - + - - - . . D I

 

 

 

PLUG FLOU 8 Conventional Plug Flow Reactor
CSTR 8 Completely Hixed Reactor

0.8. 8 Contact Stabilization

E.A. 8 Extended Aeration

88

.m a toaxm .2 cosmocomoao. comma E5528 cocoow .§ 939....

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

9:59
32...
3359 o+ + oz ooooA
o o++ .....z $8.8
tho ooo .2 Sow
combs +o. o 3:31
oozoomo o o o No 35 to
.o+ 5 ooEmv
o-e=oEOm o. ..o. ..+ 3 m:
9.63. oomow + o o - o o 3 meow
ooo +oo 5 ouv
$423589. 1. ..-. 9 of
mfiomo. 0.590 ++o -oo ~> ammo
.o+ +oo S moov
Sm +oo oo+ .o+ ooo o. 87
macaw o++ o+o ooo o++ o. 878
..o ++o +o. .++ : ooh
OOoaw o++ +o- -o+ ++- mogeomwA
cacao o++ ..o ..o oo- «0:333
. +o. +-- +0- .-+ 5 camopv
09:: .1. .o+ .o+ ..o mo ago?
5:56. o+. o+o o+o ooo No egos
0:39.??— .o+ +o. +0- --+ PC Eznv
O++ O+O 00+ +++ +++ COO mm 30—
Eoo=c_c_mo__oo o+o ooo ooo oo+ oo+ ooo mm 5285
ooocooooo o. ..o +oo ..o ..o ..+ 5 ..o...
..o. ++. ++. .o+ .o+ .o+ ooo o++ om .\ooA
_o>oEo100m ooo +++ I: o++ o+o o+o ooo ++o mm smooo
o_oo_omo\o +o. +oo +oo oo.. oo. 5 o5.
{warning women. |..

 

 

 

 

89

 

SV Similarity Value

 

Activated

 

 

Slud e
31-
01
04
40 (:5
50 011
E 14
69-- 018 C19 E 13
71-- E 12 023 E11
026
-. 017 E 10
83 021
C22
88-- E 9
020
018 026
025 025
017
E 1 E 3 E 2 E 8 E 5 E 6 E 4 E 7
Plug Step Complete Extended Contact High Deep Pure
Flow Feed mix aeration stabilization rate shaft oxygen
reactor
ELEMENIS
CQNSIBUQIS

 

017 -single stage denitrification

018 -sludge age > 15 days

019 Dissolved oxygen > 4 mg/I

020 -space loading > 1.5 IbBOD/cu it-d
021 complete mix flow regime

022 -plug flow regime

C23 -hydraulic retention time > 10 hours
024 -solub|e BOD removal > 90 %

025 -single stage nitriication

026 -primary settling tank required

 

 

Figure 13. FOCUS structure of Expert # 2.

90

 

E 1 - Conventional Plug Flow
E 2 - Complete mix reactor

E 3 - Step teed, plug flow

E 4 - Deep shalt

E 5 - Contact stabilization

E 6 - High rate

E 7 - Pure oxygen

E 8 - Extended aeration

 

 

 

N

Cost. space factors ,

 

 

 

Soluble 800 N Y
7 removal > 90 %

N E: ’
N Y Ysmdgaage N Y N Y
<15days I

N

 

 

 

NYNYNYNYNYNYNY

 

fig flow regimel

I r I II I Y
|\'/ Il/R A

NYNYNYNYNYNYNYNYNYNYNYNY YN NYNY

[3:1 It; {glam

Figure 14. Decision tree for Expert # 2.

 

 

 

 

 

 

 

 

 

 

 

 

 

91

 

E 1 - Conventional Plug Flow
E 2 - Complete mix reactor

E 3 - Step feed. plug flow

E 4 - Deep shaft

E 5 - Contact stabilization

E 6 - High rate

E 7 - Pure oxygen

E 8 - Extended aeration

 

 

 

|\
l\

 

Cost. space factors

 

 

 

 

Soluble BOD
removal > 90 %

 

 

 

 

Sludge age N
< 15 days

 

 

 

 

 

Plug flow regirlel

lH

 

 

 

Suspended BOD removal

 

 

H/l|\/

 

 

YNYNYNYNYNYNYNYNYNY

l

 

Esl ‘ ‘
E6 E5 52 E

 

 

 

 

mm
001

 

 

 

 

 

E8 52

I —l —>

 

Figure 15. Pruned decision tree for Expert # 2.

92

 

 

 

 

N Y
Y N Y
N Y N’ Y N,
/
I
I [Y
i I
It It
I\ ll
\ I\ ll
NYNYIYN mm

 

 

 

Table 16. Constraint based representation rules
for Expert #2.

 

 

 

 

fl
has the value THEI additional
I Soluble 800 removal < 70 % High suspended in influent
HRT < 3 hours
Sludge age < 1 day

SVI > 100 *
Organic loading > 1*:
Space loading > 75
Dissolved oxygen < 2 mall
Complete mix flow regime

Sludge age > 5 days

SVl < 100 *
Organic loading 0.2 - 1,5
Space loading 20 - 75
intermediate to complete mix regime

  

> 90 % Lou suspended solids in influent
HRT > 3 hours

Sludge age > 5 days

SVI < 50 *

Organic loading < 0.;

Space loading < 20

Intermediate to plug flow

 

Soluble 800 removal < 70 %
HRT < 3 hours

SVI > 100 *
Organic loading > 1,:
Space loading > 75

TKH removal < 50 %

Sludge age < 1 day

 

Soluble 800 removal > 80 %

Lou suspended solids in influent
HRT > 10 hours

SVI > 50 *
Organic loading 0.2-1.5,

Space loading 20 - 75

TKH removal < 50 %

, 5 - 15 days

Soluble 800 removal > 80 %
Hedium suspended solids in influent
HRT > 10 hours

SVl > 50

Organic loading > 1.5
Space loading > 75
TKH removal > 50 %

> 15 days

 

 

 

 

 

93

‘ Table 16 (cont'd).

1F Desi: Criteria

has the value

THE! additional

 

T Hydraulic Retention
‘ Time, HRT

< 3 hours

3-8 hours

> 10 hours

Soluble 800 removal > 80 %

Lou suspended solids in influent
Sludge age < 1 day

Organic loading > 1.:

Space loading > 75

TKH removal < 50 %

Soluble 800 removal > 80 %

Hedium suspended solids in influent
Organic loading < 0,;

Space loading < 20

Soluble 800 removal > 80 %

High suspended solids in influent
Sludge age 5-15 days

SVl < 50

 

. Flow regime

Plug flow

Completely mixed

Soluble 800 removal > 90 %
SVl < 50
TKH removal > 50 %

Soluble 800 removal 80-90%
SVl > 50

 

Space loadi 3
lb 8005/ 1 ft -d

 

< 20

20-75

 

94

- TkH removal < 50 %___ .

 

Soluble 800 removal > 90 %

Medium suspended solids in influent
HRT < 3 hours

SVl > 100 .

Organic loading < 0.05

TKH removal > 90 %

Soluble 800 removal 80-90%

Lou suspended solids in influent
HRT of 3-8 hours

Sludge age 5-15 days

SVl < 100 *

Organic loading > 0.2

TKH 50 - 90 %

Soluble 800 removal < 70 %

Lou suspended solids in influent
HRT < 3 hours

Sludge age < 1 day

SVl < 50 .

Organic loading > 0.2

      

 

 
   

THEN additional

 

  
  
   
      

 

. Table 16 (cont'd).
IF Desi- Criteria
Primary settling tank Lou
required, suspended
solid
1 concentration in
‘ influent
Hedium
High

 

- Soluble 800 removal > 90 %
HRT < 8 hours

Sludge age 5-15 hours

SVI < 50

- HRT 3-8 hours *
Organic loading < 0.3
- Space loading < 20

Soluble 800 removal < 70 %
HRT > 10 hours

Sludge age < 5 days

SVl > 100 *
Organic loading < 0.;
Space loading < 20

TKN removal < 50 %

:* Organic loading kg 800659 V§S-d
Space loading lb BOD/1 ft -

  

 

 

 

 

 

 

 

 

 

 

 

 

  

 

  
 

 

  

Table 17. Prime matcher for element 1 from Exggrt 8 2 - conventional plug flow
DESIGN COISTRUCTS HEIGHT Y I SECONDARY
FACTOR IATCHER
Cost factors, etc. 0 O
ISoluble 800 removal > 90 % +1 -1
I Dissolved 03 > 4 ”Ellll, -1 +1
Primary settling tank required +1 -1
> 15 days -1 +1
Plug flow regime +1 -1
Space loading > 1.5 leOD/cu ft-d -1 +1
Nitrification single stage +1 -1
Suspended BOD removal, etc. ._70 0

 

 

 

 

 

 

DEGREE OF HATCH

95

  

 

 

table 18.

and similarity values (SV).

Expert #2 - Design constructs with associated feature constructs

 

-Soluble BOO removal

Risk of short circuiting
Suspended BOD removal

Oxygen transfer rate

Sludge settleability, SVI
Biological solids production
Operational costs
Construction costs

 

C 19

-Oissolved Oxygen concentration

Construction costs
Operational costs
Shock loads, toxic
Oxygen transfer rate

 

C 23

-flydraulic retention time

Ease of operation

 

C 18

-Sludge age

Ease of operation

Biological solids production
Intermittent flou

Recycle rate

Suspended BOO removal
Shock loads, toxic

 

I C 22

-Flou regime

Risk of short circuiting
MLSS concentration

Area requirement

Suspended BOO removal

Oxygen transfer rate

Sludge settleability, SVI
Biological solids production
Shock loads, toxic

 

c17

-Denitrification

Ease of operation
Intermittent flow

Biological solids production
Fluctuating flow

Suspended BOO removal
Recycle rate

 

96

 

l
l
l

gaaassa assesses assess a sass aassssa a?

l

 

Table 18 (cont'd).
DESIGN CD'STIUCTS

 

C 20 -Space loading

Area requirement
Construction costs
Oxygen transfer rate
HLSS concentration
Operational costs

Power input for aeration
Shock loads, toxic

 

C 26 -Primary settling tank

Area requirement
Suspended BOO removal
Recycle rate

Sludge settling problems
Operational costs

 

 

C 25 -Nitrification

 

Removes suspended BOD
Biological solids production
Operational costs

Ease of operation
Fluctuating flou

Oxygen transfer rate
Intermittent flow

Sludge settleability, SVI

 

masseuse asses E82$$$$Dl 5

 

 

97

film-

 

5.3. EXPERT # 3

The repertory grid scored by Expert # 3 is presented in
Table 19. Calculations for the FOCUS analysis from Expert #
3 is presented in Appendix D. The sequence of design
constructs for primary matchers and SV values for secondary
matchers were derived from Table 20. The FOCUS structure
and decision tree are shown in Figures 16 and 17
respectively. A sample primary matcher is shown in Table
21, a secondary matcher in Table 22, with a complete

compliment presented in Appendix D.

The elements elicited from Expert # 3 differ from those
found the other two examples in that they are at various
levels in the hierarchy in Figure 1. Air stripping,
oxidation by hydrogen peroxide (Héoz) and activated carbon
are mechanisms, where aerobic fixed film and anaerobic
granular activated carbon (GAC) are from the level of design
and agent. Nonetheless, the analysis did produce the
internal mapping of the five elements according to the

expert’s responses.

The small number of elements, only 5, reflects the narrow
search space of this expert. The expert also commented that
the cost and regulatory component weigh heavily in the
ultimate decision. This supports the statement that

environmental consultants rely more on the preliminary

98

design phase than conceptual design. The paradigm of
operation is narrow, with only conventional technologies
offered as alternatives. If the preliminary design results
show that the conceptual design alternative is
inappropriate, due to cost, regulatory or other factors,
then the expert may extend beyond his current search space
for additional design alternatives. This is further
evidence that consultants rely more on the preliminary
design phase than conceptual design. The paradigm of
operation is narrow, with only conventional technologies
offered as alternatives. If the preliminary design results
show that the conceptual design alternative is
inappropriate, due to cost, regulatory or other factors,
then the expert may extend beyond his current search space

for additional design alternatives.

99

 

 

romd water scored by Emrt A 3.

 

   

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ACTIVATED ULTRAVIOLET J3, HAERNI
m OXIDATIQ STRIPP _G_A_l;
FLUIDIZED
BED REACTOR
I FEATlIE m
I High maintenance 1 + + - +
requirmnt
High power 2 - + . .
remiroment !
High operating 3 + + - +
costs
Highly chlorinated l. ‘ + + - + l
hydrocarbons i
Toxic by-prochcts 5 i 0 - 0 +
of biodegradation '
I High lead 6 - - + +
concentration
Lou mobility of 7 ! - - + +
contuninant in soil .
High capital 8 ' - + . + I
investment ‘
I Ho close 9 I + - + +
electricity ;
I BTEX concentration 10 I + + - -
< 10 ml) ‘
BTEX concentration 11 E - - - +
> 10,000 ppb
[TDS > 30,000 mg/l 12 - - - -
inorganic
Contaminant density 13 - - + -
> water
pH adjustment 14 - - - +
required
15 + + 4»

Iron removal
r ired

 

 

 

 

 

 

100

 

I Table 19 (cont'd)

  

AERQIC

FIXED FILM

ACTIXATEQ
£5829!

HLI§A¥19L§I

OX!DATIQ

Ala
STRIPP no

 

555589915
£55
FLUIDIZED
sea a ACTOR

 

F

 

Removal rate to

1 PPb

 

‘ Flow rate >

15HGD

 

‘ Long residence

E time

 

Contaminant

‘ destroyed

 

Lou taperature

 

 

 

 

 

 

 

 

11)].

Table 20.

I
1

C 16 -Ramoval rate to 1 ppb

FEAJUIE CDISTIUCTS

BTEX concentration < 10 ppb
High maintenance requirement
High operating costs

High chlorinated hydrocarbons
Density of contaminant > water

High power requirement
High capital investment

Expert #3 - Design constructs with associated feature constructs
__ and similarit values (SV).

 

‘ C 20 -Low temperature

BTEX concentration < 10 ppb
High maintenance requirement
High power requirement

High operating costs

High chlorinated hydrocarbons
High lead concentration

High capital investment

TDS > 30,000 mg/l inorganic
Density of contaminant > water

 

‘ Contaminant destroyed

High power requirement

High capital investment

BTEX concentration > 10,000 ppb

pH adjustment required

High maintenance requirement

High operating costs

High chlorinated hydrocarbons
Toxic by-product of biodegradation
Low mobility of contaminant in soil
108 > 30,000 mg/l inorganic
Density of contaminant > water
Iron removal required

 

Flow rate > 15 HGD

 

High lead concentration
Electricity not in close proximity
BTEX concentration < 10 ppb

 

888 888888888888 888888888 888888

 

1132

 

Table 21. Primary matcher for element 1 from Expert # 3 -
aerobic fixed film
-- IL

 

 

j DESIU ms Elm ”Elm
' rm FACTC

 

Removal rate to 1 ppb

 

.' Low temperature
5 Containant desth
Flow rate > 15 HOD

 

 

 

 

 

 

 

 

°N£UEE§[E;EEflEEil

[cease U MIC"

Table@2:_Secondary_matcher from Expert# 3

 

 

FHWUE 12 1 3 9 1; 2 Q
. ms
l
8V ‘ + + + + - - - I

 

103

S V Similarity value

TREAT

 

 

 

538' C16
C 20 E 8
54-
' C 17
65- C ‘9 E 7
70- E 6
C 4
C17
C19
E 2 E 3 E 4 E 5 E 1
Anaerobic
Activated Ultraviolet Air G AC Aerobic
carbon oxidation stripping fluidized fixed
Oi H202 bed film
ELEMENIS
92135130913

016 -removal rate to 1 ppb

C17 -flow rate > 15 MGD

C18 -Iong residence time

C19 -contaminant chemically destroyed
CZO -Iow temperature

 

 

 

 

Figure 16. FOCUS structure for Expert # 3.

104

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Y
[Removal to 1 ppb
N Y Low temperature N Y
N Y N Y N Y N Y
Contaminant destroyed
Y N . N
N Y N Y N Y N Y N Y N Y
Flow rate > 15 MGD Y
Highly chlorinated hydrocarbons F
: /||\/|l\/|‘ l

 

 

 

 

 

 

 

 

 

 

NYNYNYNYNYNYNYNYNYNYN NYNYNYNYNY

 

 

  
    

 

   

 

Air Anaerobic ACtNatOd
stripping GAC carbon 33%;?” Activated uv / H202
fluidized bed " '°" carbon Oxidation

 

 

 

 

 

 

 

 

 

 

Figure 17. Decision tree for Expert # 3.

105

5.4. CLASSIFYING A C38!

The following is an explanation of how a matcher calculates
a degree of match when classifying a case. The matchers for
this example are taken from the knowledge acquisition from
expert #1. The responses to the questions a representative
of a wastewater with overstandard soluble BOD (biological
oxygen demand) levels.

The primary matchers for the nodes from Expert # 1 are
displayed in Appendix B. Table 23 shows the questions
presented to a user of the KBS to determine appropriate
activated sludge design, and the responses to our sample

wastewater.

Table 23. Questions and responses for a sample case.
: m _ ______a _l___, . _._un _ ______m__--_______ _* _ 1

 

 

 

 

 

 

 

 

 

. SUN PRESEITE DY TIE M RE"' ‘ l
1 Is the sludge age over 15 days? DOH'T KHOH? :
i Is nitrification occur? no i
1 Is the flow regime plug flow? YES

; Is the hydraulic retention time over 10 hours? YES

3 Does denitrification occur? no |
? Is the concentration of HLSS high? H0 1
E Is the dissolved oxygen concentration over lo mg/l? DOH'T KHGI? I
1' the WM Witt__________t _ {*5 ._ _ __ _ J

For the two questions with the response of 'don't know', the
secondary matcher will invoke and the user will be presented
additional questions. Table 24 shows the questions which

would be presented if the answer to the sludge age is 'don't

106

know”, and the responses for the sample case, and table 25

for dissolved oxygen concentration.

Table 24. Questions and responses for the secondary matcher

regarding sludge age for a sample case.

 

 

 

 

 

 

 

 

 

 

GENIUS ASKED REGADIB SLIDE AE m SV [Em

Is the recycle rate high? 9 .92 YES

Is the flow fluctuating? 6 .75 YES

Are shock loads characteristic of the flow? 8 .75 DOH'T KHOH?
Is there a low risk of short circuiting? 3 .67 DOH'T KHOH? II
Is the sludge easily dewatered? 14 .67 DON'T KHOH? fl

 

 

Table 25. Questions and responses for the secondary matcher
regarding dissolved oxygen concentration for a

sample case.

 

In;—

 

 

 

 

 

 

 

 

 

 

OUESTIEIB asses RECARDIIC DISSDlVED OIYGEI CDISTRUCT SV RESPOISE
WT um

Is there a low risk of short circuiting? 3 .83 DON'T KHOU

Is the flow fluctuating? 6 .75 YES

Are shock loads characteristic of the flow? 8 .75 DOH'T KHOH I
Are there space limitations? 5 .67 YES “
Is the sludge easily dewatered? 14 .67 DON'T KHOU ll

 

 

The confidence factor from the secondary matchers are

calculated with the formula algorithm.

Consider the

calculation of the confidence factor for sludge age. There

are two of the five questions with a response. Begin with

the response with the highest SV is from construct 9, where

SV = 0.92. Since n=1, m=5 and CFn,1== O, the CF is computed

as follows:

CF1 = o + [ (1/5 * 0.92 + (1-0) 1

CF1 = 0.18

107

 

The second highest SV is 0.75, so to compute the confidence
factor for two responses, n=2, m=5, Canl== 0.18 and SV =
0.75. Therefore:

CF2 = .18 + [ (2/5) * 0.75 * (1-0.18) ]

CF2 = 0.43

Then calculate the confidence factor for dissolved oxygen

concentration using the formula algorithm in the same

manner a
CFl = 0 + [ (1/5) * 0.75 * (1-0) ]
CF1 = 0.15

CF2 = 0.15 + [ (2/5) * 0.67 * (1-.15) 1

CF2 = 0.3a

In this sample case, each question from the primary matcher
has a response, and therefore a confidence factor. The
degree of match for each primary matcher can be calculated
from the confidence factors by weighted average. An example
is shown in Table 26 for Element 1, conventional complete
mixed reactor. Answers to primary matcher questions are in
bold, and from the secondary matcher are indicated in the
appropriate column. The degree of match is calculated as:

degree of match = 0.34 -1 -1 +1 +1 +0.38 +1
8

4‘

degree of match 0.35

108

Table 26. Primary matcher with confidence factors and degree
of match for Element 1 - Conventional completely
mixed reactor.

 

 

 

 

 

    

   

 

 

 

DESI“ m S Y I SEW
MTCIEI
Sludge Age > 15 days +0.43
Nitrification occurs +1 -1
Pl flow r ime -1 +1
HRT > 10 hours +1 -1
Denitrification occurs -1 +1 +1.0

 

 

lHigh MLSS concentration -1
Dissolved 03f> 4 mall

 

  
  

 

 

 

 

L“Soluble BOD removal > 90 X

 

 

A degree of match is calculated for each primary matcher
representing each element and can be found in Appendix E.

Table 27. shows the results of these calculations.

 

109

 

 

Table 27. Degree of match for all primary matchers for a

sample case.

 

! Activated sludge technology

 
    
 

Establish 7
threshold 8 +0.5

 

  

 

  

 

: Deep shaft design

' Conventional completely mixed 1 +0.6 YES

; reactor

i Step feed, plug flow reactor -0.35 NO
-0.06 NO

  

 

   

NO

 

 

i Sequencing Batch Reactor

, Pure oxygen complete mix reactor

 

 

Winch

 

 

 

 

Weight factors in the primary matcher allow the expert to

prioritize the design criteria, which in turn influences the

confidence factor values.

Table 28 shows the same

classification as Table 27 but includes weight factors. A

complete display of primary matchers is in Appendix E.

Table 28. Degree of match for all primary matchers including

weight factors

for a sample case.

 

 

 

 

 

 

 

 

 

 

 

Activated sluhe technology Eleunt 9 Degree of latch Estwlish 7
threshold = +0.5
Conventional completely mixed 1 +0.35 YES
reactor
“ Step feedu plug flow reactor 2 -0.42 N0
“ Deep shaft design 3 -0.02 NO
H Sequencing Batch Reactor 4 +0.19 YES
Pure oxygen complete mix reactor 5 +0.21 YES
u Oxidation ditch 6, +0.10

110

 

6.0. DISCUSSION

The methodology described in this thesis satisfies the two
primary requirements of a successful knowledge acquisition
exercise according to Gruber (1987) by establishing the type
of knowledge and functional mapping. The type of knowledge
is identified as the design criteria and features used for
conceptual design. The functional mapping from the user is
represented in the matchers by way of the grid and
constraint based rules. The structure of how the expert
decomposes the problem is expressed decision tree with a
limited number of possible goal states. Confidence factors
when compared with a pre-determined threshold are used to

eliminate inappropriate goal states from the final decision.

Conceptual design decisions are decomposed based upon design
criteria. The engineer chooses the combination of design
values which he believes will result in satisfactory
treatment of a waste stream. Therefore, primary matchers
are composed of the design criteria, represented as design
constructs in the proposed methodology, by which the expert
bases his decisions. The user, when classifying a case,
will select the value for the design criteria, which the
matcher computes into confidence factors, and ultimately a
degree of match. This degree of match, compared to a pre-

selected threshold, determines if the node will establish or

111

 

reject.

Frequently, the user will not have sufficient information to
determine the value of the design constructs. The secondary
matcher contains the feature constructs which, according to
the expert, have a high correlation to the design construct.
The secondary matcher will calculate and return a confidence
value to the primary matcher that indicates the appropriate
design construct value based on those features of the waste,
technology and effluent standards which are related to that

particular design criteria.

Not every question in the matchers, both primary and
secondary, need be answered when classifying a case. The
more information included in the evaluation of each node,
the higher the confidence factors, and therefore the higher
the degree of match. This is analogous to human decision-
making processes. The more information an expert engineer
knows about a waste stream and the technologies available,
the more confidence he will have in his decision of

conceptual design.

6.1. KNOILSDGB ACQUISITION DRAIBACRS
Building expert system models serves several functions.
First, it is a way of formulating domain knowledge. Also,

it is a means of communicating about the problem space from

112

'which the expert operates. Models bring a common language
to the table of discussion, allowing for deeper
understanding of the decision-making process among experts.
According to Bradshaw, et. al. (1992), knowledge acquisition
usually involves "inventing new languages for modeling
previously unarticulated experience.". They provide a
platform for mutual agreement, and consequently, respect

and acceptance.

Another important benefit of building and using an expert
system is not merely the recommended outcome, but the
improved insight for decision makers. Expertise is
developed from understanding why one goal state is chosen
over another. The recognition of the assumptions and
underlying uncertainties which lead to the conclusion are
the building blocks of expertise in trained individuals.

One assumption relevant to the methodology described is that
the "Gold Standard" of the expert must be accepted (Punch &
Sticklen, 1991). The Gold Standard means that there is one
answer which is clearly the proper choice from the options
available for a given set of circumstances. If the Gold
Standard is not accepted, then the results of this knowledge

acquisition methodology will not be received favorably.

This knowledge acquisition exercise may not be able to

uncover the automatic processes of some experts. Knowledge

113

 

 

 

 

and judgement is stored differently among experts. For
some, knowledge acquisition may have the onion effect. When
all the layers are pealed away and examined, they may not
fit back together. 0r, there may be nothing in the middle
holding the layers together. In effect, the basis for the
experts decision making capability may not be able to be
represented from these methods. Perhaps the expert has
decomposed the problem into so many partitions that these
techniques are not able to recognize and reassemble the

parts.

A hindsight bias will tend to impede the knowledge
acquisition process and must be overcome if one is to
achieve effective capture and representation of expertise.
According to Fischoff (1982), the very outcome which gives
the feeling of understanding about the past may prevent
learning anything from it. There is a strong tendency to
establish the criteria for evaluating decisions after the
corresponding events have occurred and the outcomes are
known. With the wisdom of hindsight, one is free to ignore
all the "noise" created by irrelevant or unreliable cues.
Knowledge based on hindsight may occasionally hinder the
development of appropriate decision processes. Attempts to
modify and reduce this bias have been met with relatively

little success (Fischoff, 1982).

114

A drawback of the FOCUS technique is the equal weight of all
constructs considered in analysis. This is seldom the case
in design decisions. There should be a way to reflect these
differences among constructs in the repertory grid. The
proposed methodology allows the matchers to be reviewed and
tuned using weight factors. Priorities imposed with weight
factors in primary matchers address this problem once the
grid has been scored. Secondary matchers could also include
a weight factor, which could be multiplied with the SV to
reflect the expert's prioritization. However, this does not
overcome the inherent bias to the FOCUS technique. Shaw
(1981) has several examples of automated FOCUS programs

which may address this issue.

6.2. MULTIPLE EXPERTS

The proposed methodology represents the mapping of a single
expert. The results of the tree representing an expert's
map may not satisfy another expert's view. Since the grid
technique is subjective, the result is a mapping of the
individual expert. One criticism of the grid technique is
it's subjectivity. That is, two experts considering the
same problem can produce different solutions. Newell &
Simon (1972) provide evidence that this variation in view is
what distinguishes novices from experts, and experts from
among themselves. If several experts in a domain are

presented with a problem, they~each produce a solution based

115

on their own paradigm. These results may or may not agree,
reflecting the inconsistency among experts in the same
field. The grid will capture the internal structure of the
paradigm which could explain these differences based on the
rules or even the sequence of rules by which the expert
reasons the problem. This can be viewed as a positive
attribute, as many experts systems are built to capture the
expertise of an individual rather than the common pattern of

decision-making generally accepted within the domain.

There may indeed be disagreement about the design criteria
for technologies. Even when presented with the same grid and
CBR chart to complete, two experts may score the tools
differently, resulting in conflicts when attempting to join
them. For example, expert # 1 scored "+" for E 1 regarding
long sludge age greater than 15 days. Expert f 2 scored "-"
for this design criteria construct. Therefore, we would not
expect agreement in their final representations. Has the
technique failed? Quite the contrary, the representation
accurately models the decision tree structure of each
expert, whether scientifically accurate or not. If the
results from two or more experts are to be combined into a
single expert system, there would need to be a method for

conflict resolution of this type.

Conflicts may arise between responses in the grid versus in

116

the CBR chart scored by the same expert. There are several
reasons why these discrepancies may occur. First, the CBR
chart asks the user to relate two design criteria and the
effect they have on each other exclusive of other aspects of
a particular technology. The grid, on the other hand, asks
the user to compare attributes of the technology as
functional unit. The contradictions may be a result of the
knowledge acquisition technique itself, which forces the
expert to view a problem as they have not done before. This
may cause confusion. When forced to answer a question which
is unnatural to their thought processes, they can respond
erroneously. Out of context, their judgement may be clouded
or altered by a new framework. They may not be reporting
accurately. By comparing the two representations when a
conflict is expressed, the expert can review and hopefully
resolve the contradiction by altering one or both responses

to agree.

The proposed methodology can be used to compare the decision
process among multiple experts by comparing the resulting
decision trees. For Expert # 1, the first discrimination is
based on removal rate, a regulatory requirement, much the
same as with expert #2 where removal rate was the second
level in the tree. This indicates that the paradigms
overlap in the conceptual design decision process. Insight

as to the point and extent of overlap can be identified from

117

the analysis described in the methodology. For example,
knowing which design constructs are influenced by cost
feature constructs in the repertory grids can help experts
focus on how to reduce costs. When these design criteria
are minimized, then so should be the associated costs,

resulting in less expensive remediation.

Incorporated in computer programs such as ETS and Aquinas,
they can be used directly by experts with very little
assistance from the knowledge engineer, thus lending to
efficient knowledge acquisition. For multiple expert input,

these programs are an invaluable asset.

This methodology can be used to build one expert system with
the input from multiple experts. The experts involved would
gain consensus in eliciting and scoring the grid and CBR.
The combined knowledge would be used to form one decision
tree and one set of primary and secondary matchers. Since
experts recognize their own limitations respect others'
abilities, this should be possible. Obstacles such as the
potential language and ego barriers may present difficulties

which would need to be resolved among the experts.

6.3. REASONING EXPLANATION
The KBS built following the proposed methodology contains

the ability to explain the decision process. The reasoning

118

behind the rejection or establishment of nodes is expressed
in the matchers and comment windows. The decomposition is
exposed for examination by critics, and the point of
discrepancy can be isolated. Often the expert would not be
able to verbalize this distinction. When asked to explain
his reasoning, the response is often "I just know". They
cannot describe how they know. This knowledge acquisition
technique allows for this distinction due to the explicit

decomposition of the problem.

6.4. AUTOMATION

Automation of this method will greatly improve the quality
of knowledge acquisition and representation. The expert
will be able to view the decomposition of knowledge and make
adjustments during the building process. Conflicts will be
resolved, and gaps filled which exist in the examples
included here. Experts have trouble verbalizing the
decomposition, however, they are able to examine parts of a
problem and reassemble components in an intelligent fashion.
The skill of pattern recognition is called upon in the
building or knowledge representation phase. The expert will
quickly identify nonsensical relationships and conflicts
which may not have been realized in the knowledge

acquisition phase.

Re-evaluation of the grid results through the automation of

119

the methodology will provide the expert with a hindsight
view of their expertise. They will not be able to ignore
the "noise" from irrelevant conclusions inferred from their

decision tree and CBR results (Wright & Bolger, 1992).

120

7.0. ENGINEERING SIGNIFICANCE

The search for innovative techniques is motivated when
conventional technologies fail or are too expensive. The
incentive to stay with conventional techniques for
regulatory approval is overwhelming in the United States
today. According to the Michigan Department of Natural
Resource employees interviewed in the development of the
commercialization plan (Appendix A), only by addressing the
regulatory community and their strict adherence to the tried
and true methods will these new scientific technologies be

implemented.

The hesitancy to approve innovative technologies is
justified by the lack of reliable performance and
specification data. Often these data exist, but not in a
readily accessible format. A knowledge-based system could
assist regulators by providing a resource database which
includes information about new technologies. Classifying
cases would direct attention toward these new methods, where
the relationship between waste and treatment may not

otherwise be recognized.

A knowledge based system to assist in the conceptual design
process for environmental systems could expand the
implementation of new technologies which are not well known

or understood. The KBS applies reasoning free from

121

regulatory biases, therefore the applicability of these
treatment regimes is more visible. The knowledge acquisition
process delineates the criteria important in the decision
process. The performance data required to adequately place
the new technology in the internal decision tree will be
specified. Therefore, the developers of these technologies
will have guidance as to the testing required for their
design to gain acceptance and implementation. Incorporating
an innovative technology into an existing KBS will clarify
what information is needed to objectively compare the
innovation to conventional competitive technologies.
Limitations and expectations will be evaluated on the same
criteria and scale. Perhaps if a technology is indicated as
a potential remedy for a problem by the KBS, the regulators
and consultants will have more confidence in recommending

its implementation.

122

0.0. CONCLUSIONS

Knowledge is the key to ease the gridlock faced by the
scientific and regulatory communities. The applicable
knowledge is expansive and continually growing. The extent
of regulations from all governmental levels is also
increasing in complexity. Therefore, a computerized tool is
required in environmental conceptual design to ensure the
incorporation of all of this knowledge into the very

important decisions being made today and tomorrow.

As a result of these factors, together with the extreme time
constraints under which consultants and regulators are
obliged to operate, conceptual design decisions are
frequently made in a far from optimal manner. We cannot
expect the experts who recommend and approve remediation
solutions to keep up with this new knowledge, and
incorporate all into their heuristic vocabulary. Experts
compare to a wastewater to one they have seen before, take
the treatment regime which worked before and apply it to the
new wastewater. With modeling and estimating techniques,
they judge whether it will work again. There is tendency to
design around the restrictions, weed out any rejection
criteria, and use scientific or risk analysis to verify
their judgement. A more efficient manner could involve the
assistance of a KBS as a decision support tool. This would

allow a broader scope of potential technologies to choose

123

 

from, thus resulting in optimized effective conceptual

design.

Computer-aided tools can assist incorporating new rules
which establish the scientific paradigm used when performing
the search through the alternatives. Once these have been
determined, the consultant can invoke their experience
regarding cost, regulatory approval and design expertise.
Consultants can apply their clever presentation and
negotiating skills to a well-designed innovative treatment

alternative to gain approval from regulators.

Although a tool which only assists in the decision making
process may be helpful to the engineer or regulator, there
is reason to believe that the use of this tool will not
alone promote the implementation of new technologies.
Choosing the appropriate technology is important, but a need
was identified for a tool which would assist in the costing
and final design of the selected method of treatment. A
costing and/or design tool is the product which will best
promote the implementation of new technologies. If the
regulatory community would adopt such as KBS tool as
guidance or policy in the conceptual design process, perhaps
the implementation of innovative technologies would

increase.

124

 

In spite of the numerous environmental software products
available, the market is not growing as quickly as in other
domains. Lack of standards and buyer skepticism are noted
for the lack of acceptance of these new products. Gregory
B. Baecher, CEO of ConSolve, the Lexington, Mass., vendor of
SitePlanner software is quoted saying "Engineers and
geologists still tend to work with 79 cents worth of colored
pencils. The consulting industry needs to get into the 20th
century." (Rubin, 1992) This applies not only to the
modeling software, but to decision tools as well. The bulk
of the cost of new software is the expense of demonstrating
to customers that they need the system. Vice president of
CMZH Hill, Inc. says "It's hard for many to make the
paradigm shift." The conceptual design tool described in

this paper will surely face the same battle.

From this methodology, the potential of developing new
design strategies exists. A user may classify a waste from
this representation and arrive at a goal state which has not
been identified by the expert as a specific technology.
According to the rules derived from the grid and CBR, this
combination of the design criteria indicates possible
remediation success. This goal state may also represent a
known technology which is not apparently part of this user's
paradigm. The conceptual design tool may direct users to

designs which they may not otherwise consider, but may

125

 

indeed be most appropriate for the situation.

126

 

9.0. PUTURE RESEARCH

9.1. AUTOMATION

Automation of this methodology could greatly enhance and
improve the implementation of knowledge acquisition. When
the expert can view the changes in tree structure resulting
from different responses in the grid and CBR, he can develop
a more accurate representation of his search space. With
multiple experts, automation would provide timely feedback

when resolving conflicts in expertise.

9.2. EXPANDING TEE EIERARCEY

Two examples from this study were applied to the design of
activated sludge processes. The methodology should be
executed with experts in other areas of environmental
engineering processes. Matchers need to be developed for
each node in the hierarchy, regardless of the level, to
complete the HC tool. The theory is that whether the
process be physical, chemical or biological, the method

should apply. However, this has not been tested.

9.3. ASSEMBLER

The proposed methodology is designed for knowledge
acquisition and knowledge representation within the
Hierarchial Classifier only. Recall that the proposed
computer-aided conceptual design tool is comprised of two

components - the Classifier and an Assembler. The assembler

127

would be responsible for sequencing the technologies should
more than one be required to meet regulatory standards.

Many times pretreatment is necessary for effective
remediation. The knowledge and decision support for both of
these very important aspects of conceptual design must be
developed to be incorporated with the HC for a complete

system.

128

 

10.0. BIBLIOGRAPHY

Barker, Joel Arthur, ture d e° d s ve n e new

peregggpe_pf_epeee§§, William Morrow & Co, Inc., New York,
NY, 1992

Boose, John H., Expertise Trepsfef for Expeft System Design,

Elsevier, Amsterdam, 1986.

Boose, John H., s n tor G id- en ed Knowled
Aegpfsition Ipels fo; Decision Support, Proceedings of the
Twenty-Second Annual Hawaii International Conference on
Systems Sciences, Vol. III: Decision Support and Knowledge
Based Systems Track, PP 221-20, vol. 3, 1989

Boose, John H., Using Repertory Grid-Centered Knowledge
Acquisition Tools for Decision Support, L333, Vol 73 (0073-
1129), p211-220, 1989.

Bradshaw, Jeffrey M., Peter D. Holm, and John H. Boose,
Sharable Ontologies as a Basis for Communication and
Collaboration in Conceptual Modeling, Efpceedipgs pf phe 7th

Benff Knowledge Aegpisitiep fez Kppwledge-Based Systems
ngksnpp, October 11-16, 1992

Buchanan, B. G. and E. H. Shortliffe, Bple;_§eeeg_§xpepf
Systepe; The MXQIN Experimenfs of the Stanford Heuristie

prgfeppipg_£ppjeep, Addison-Wesley, Menlo Park, California,
1984.

Chorafas. Dimitris N-. Kne21edge_Engiaeeringi_Knegledge
ecgpisition, knowledge representation. the Role of fpe

knewlegge epgipeer apd Dopaips feftfle fpf Al
Ipplemeptefign, Van Nostrand Reinhold, New York,1990.

Criddle, Craig, Biological Process in Environmental
Engineering, ENE 804, Michigan State University, Winter 1992

Diaper, Dan, ow ed ' 'ta 'on: 'nc s ec s
' , John Wiley & Sons, NY, 1989.
Findler, Nicholas V., t'f’ ' te ' e e h
fog Information ang Facf Beffievel-Ap Applieetiop ip negicel

Kppglegge_£;peeeefpg, The Massachusetts Institute of
Technology Press, 1991.

Fischoff, B., For those condemned to study the past:
Heuristics and biases in hindsight. In D. Kahneman, P.

Slovic, & A. Tversky (Eds.), Judgpept ppgez upeeffainfy;
Heufiefics egg piases. Cambridge: Cambridge University

Press, 1982.

129

 

Ford, Kenneth M., Frederick E. Petry, Jack R. Adams-Webber &
Paul J. Chang, An Approach to Knowledge Acquisition Based on
the Structure of Personal Construct Systems, TEEE, Vol 3,
No. 1, p 78-88, March 1991.

Fransella, Fay and Don Bannister, A Mappai for Repertory
grig_Teeppigpe, Academic Press Inc., San Diego, CA, 1977.

Gammack, John 6., Steven A. Battle, Robert A. Stephens, A
Knowledge Acquisition and Representation Scheme for
Constraint Based and Parallel Systems, IEEE, 1030-1035,
1989.

Gray, N. F., Activated Sludge: Theory and Tractice, Oxford

University Press, Oxford, 1990.

Green, Paul E & Yoram Wind, New Way to Measure Consumers’

Judgements, fiervarg Busipeee Review, July/August 1975.

Gruber, Thomas R. 8 Paul R. Cohen, Design for acquisition :
principles of knowledge-system design to facilitate

knowledge acquisition. Inf1.11.8aazuashine_§sudie§. 26: 143-
159, 1987.

Guariso, G. and H. Werthner, Env ent e 's' t
Syereme, Ellis Horwood Limited, London, England, 1989.

Hart, Anna, now e Ac isitio or E ert S s ems, McGraw
Hill Book Co., New York, 1986.

Heekerman. David 8.. 2r2babilissisléimilari§x_Ne;12rk§. The

Massachusetts Institute of Technology Press, 1991.

Henrion, Max, John S. Breese, and Eric J. Horvitz, Decision

Analysis and Expert Systems, The AI Magegipe, p 64-91,
Winter 1991.

Keren, Gideon, Improving Decisions and Judgments, the
desirable versus the feasible, from ed. Wright, George and

Fergus Bolger, Expertise apg Decision Spppgrt, Plenum Press,

NY, 1992.

Kidd, Alison L., now e c 's tion e t s-

Prastical_nandbeek. Plenum Prose. NY. 1987-

Klein, Michael and Leif B. Methlie, Experf_§yefepe;_A

Dseisien__seeer§_aeer_acn Addison-Wesley. NY 1990-

Metcalf & Eddy, Inc., W w t ' '

Qiepeeei_epg_3epee, McGraw Hill, Inc., New York, 1991, 3rd
ed.

130

 

Mital, Sanjay and Clive L. Dym, Knowledge Acquisition from
Multiple Experts, The AT Magezine, Summer, 1985.

Neale, I. M., First generation expert systems: A review of
knowledge acquisition methodologies. The Knowledge
v w, 3, 105- 145.

Parsaye, Kamran and Mark Chignell, Expert Systems for
Experfe, John Wiley & Sons, New York NY, 1988.

Patterson. Dan w., Iatr2dust1_a_t2_Art1fisial_lntelligease

and_fixnert_§x§tems. Prentice Hall. 1990-

Punch, William and Jon Sticklen, Introduction to Artificial
Intelligence, CPS 841, Michigan State University, Fall 1991

Reynolds, Tom D., Unit 0 e a 'ons an o sses 'n
' o e t n 1 er' , PWS-Kent, Boston, MA, 1982.

Rich, Elaine and Kevin Knight, hrfificiai Inteiiigenee, 2nd
ed., McGraw-Hill Inc., New York, NY, 1991

Rubin, Debra K. and Mary B Powers, How Green is our

software, Ehgiheering Recorg news, October 26, 1992.

Schutz-Riley, Lori, Coppercieiieatiop Piep, MTA 810,
Michigan State University, Summer 1992.

Shaw, M.L.G., Rece Adva ces 'n erso o ct
Teghppiegy, Academic Press, London, 1981.

Shortliffe, Edward H. and Bruce G. Buchanan, A Model of

Inexact Reasoning in Medicine, Mathemetieei hicsciepces, 23:
351-379, 1975.

Sprague, Ralph H. Jr. and Eric D. Carlson, fiuiiding

Effeetive Decision Support Systems, Prentice Hall, NJ,
1982.

Stewart, Valeria, Andrew Stewart & Nickie Fonda, Business
Appiieetions of the Reperfpry Grig, McGraw Hill Book Co.,
London, England, 1981.

Thomas, Laurie F. & E. Sheila Harri-Augstein, Se -

oun a s v sat e
peyghplpgy, Routledge & Kegan Paull, Boston, MA, 1985.
Tzeng, Chun-Hung, st' f a ' ' a -
Tree_§eergh, Springer-Verlag, 1988.
Welbank. M. v'ew of K ow e c is' ion ch ' s

Experf_§y§fepe. Ispwich Matlesham Consultancy Services.,

131

 

1987.

Wright, George and Fergus Bolger, E
§pppprf, Plenum Press, NY, 1992.

132

e t'se a d e 's o

Appendix A

Excerpt from

CONNERCIALIZATION PLAN

prepared by
Lori Schutz-Riley
MTA 810
Summer 1992

APPENDIX A

Excerpt from COMMERCIALIZATION PLAN

§pppery pf competitive prodpefs.

Environmental software is available to assist in the tasks
of modeling, completing forms for governmental agencies,
design, costing and decision-making. Some are available
free of charge from the Environmental Protection Agency
(EPA) and other branches of the United States Government.
Others are commercial endeavors where prices range from a
few hundred into the thousands. Pollution Engineering
magazine reviews environmental software each year the
January issue. They site the prime concerns for users of
environmental software are flexibility, compatibility and

maintainability.

This investigation was limited to software useful in the
conceptual design task. A list of features compares and
contrasts five tools to the MSU (Michigan State University)
tool presented in Table 29. The distinguishing features of
the MSU tool are that it is an expert system with decision
making capability, it draws knowledge from interfaced
databases, and includes innovative technologies. Unique to
the MSU tool is the portability of the Smalltalk 80 language
to all major PC's, although RAAS will be portable to PC-DOS

133

SOOI‘I .

Two of the software tools, RREL and VISITT, are databases
developed by the EPA. RREL is compound oriented, while
VISITT is technology oriented. VISITT, according to
developer Linda Fiedler, is not intended to be a decision
making tool. By design, the decision of which technology to
use is dependent upon the expert judgement of the user.
VISITT was released in June 1992, and annual updates are
scheduled. Only seven weeks after release circulation
exceeded 1,500 copies. This indicates a demand for such a
product is real. The product is delivered at no cost and
only contains vendor information supplied by the vendors
themselves. In other words, this is not a comprehensive,
unbiased compilation of technologies. A customer survey was
executed as orders were taken, and a copy will be sent to us
in the next month. The customer base at this time appears to

be a mix of the groups we interviewed.

The RREL database contains relevant data about contaminants
necessary to make such decisions. A few of the consultants
we interviewed were familiar with RREL, and used it when
they encountered a compound new to them. This tool is
especially useful to the inexperienced who are performing

the conceptual design task.

134

Neither RREL or VISITT share the distinguishing features of
the MSU system outlined at the beginning of this discussion.
Their initial purpose and function are different from the
MSU system as they do not make decisions. However, they do
address innovative technologies. They are accepted by their
audience; the success is yet to be seen. Both will be
helpful tools in building the MSU conceptual design tool,

however do not occupy the same marketing niche.

CORA is an expert system combined with a costing tool. The

 

primary goal in development of CORA was to assist in
developing cost estimates in the pre-feasibility stage of
site analysis at Superfund Sites. The 42 technologies
included in the database are conventional methods only, so
that accurate costing could be performed, and approval was
certain. The last update was in 1990, and there are no

plans for future development.

NETAC's ETAP program uses a database of only innovative
technologies. The heuristic search is manual, performed by
trained technicians. A weakness of NETAC is that the
possibility of the same biases seen in consultants'
recommendations could develop in the technicians performing

these searches. A strength is that the database is

 

continually updated with new technologies and support data

for technologies already included. Preliminary cost and

135

design information is included with the recommendation of

the appropriate technology when available.

The RAAS system is the closest to the proposed MSU system we
found as shown in Table 29. RAAS is an object oriented
expert system which includes 90 conventional and innovative
technologies. Although the initial goal was to address
Superfund and Hazardous sites, the program is applicable to
industrial wastes as well. RAAS is in the final alpha
testing stage, and will begin beta testing soon. This
project is years ahead of MSU in development of a total

conceptual design tool.

RAAS system mimics the process of the RI/FS phases of
remedial action. Included in the system are site, compound
and technology databases, a modeling module, the capacity to
conStruct treatment trains and address side streams produced
from the original treatment technology. Whatever the
defined goal, the search through technology objects within
the RAAS application will attempt to construct treatment
trains to satisfy that goal. RAAS is expandable to include
technologies developed in the future. A costing tool to

compliment RAAS is under development at this time.

136

Table29. _Environmental software for conceptualdesign.

 

 

 

 

 

 

 

 

 

 

 

    
  
 
  
 
  
 

 

 

 

 

 

 

 

 

 

 

“MW

1 EEFEY‘E‘ _. ___ffxfl_fxfl_flf __, f__ _“,f x_m_ _ f_w__mf i

f interactive * s i * e i

, user-friendly * * l e e *

expert system * e ’ e l

;database 1 e e *

based on waste * * s a e

; based on site * e ‘ * at

based on * * ‘ * *

technology 4
explained * s i *
reasoning

'decision making * s g a

I cost evaluation * g * * * i

l multi-constituent * * l s ‘

waste

Superfund sites * g *
Hazardous waste i *

only ‘ -

domain : air * i s *

l wastewater * * i s * \
groundwater * e ’ e e * * *
soil/sediment * M e e *
sludge * 1

portable * _“‘_f_ 1‘” __ _ _‘ _ _

 

 

 

 

137

2.0531131121152115

One goal of the MSU conceptual design tool is to promote the
implementation of innovative technologies in environmental
engineering systems. By using an unbiased, decision making
computer program, the conceptual designer will recognize
opportunities to implement new strategies which otherwise
would be overlooked and not considered. Based on interviews
with consultants, regulators and environmental lawyers,
awareness was not the primary reason that these innovative
methods were not being used. The reason is that there is
not enough supportive costing and design data available to
ensure competent design, permit approval and eventual

success in the implementation.

The more experienced consulting firms did not feel that this
tool would be useful. Consultants with 10 to 15 years of
experience use their judgement to make technology choices.
They spend very little time in the conceptual design phase.
Cost is the driving force to motivate engineers to search
beyond their personal knowledge base for innovative
solutions. Small, inexperienced consulting firms, however,
could use the expert system to make technology decisions as
they lack the experience to make decisions based on their
own judgement. Consultants tend not to select new
technologies over conventional methods because the DNR

(Department of Natural Resources) and other regulatory

138

agencies take more time to approve permits. The cost of
fixing or replacing a faulty system is also motivation to
only recommend conventional methods. Consultants hesitate
to use new technologies unless the government will pay for
the efforts made to correct the problem if the new

technology does not work.

The regulatory agencies do not perform the conceptual design
task. However, they are responsible for reviewing the
environmental systems recommended by consultants and
industry who do perform this task. When a new or innovative
technology method is involved, the review period is extended
due to the research required to evaluate the design and
technology selection. With a conventional choice, past
experience expedites the process. Often with a new
technology choice, data is incomplete and unclear, resulting
in the request for additional information. Only through
experience is the DNR acquainted with new methods, and this
is a very slow process. A tool such as this could verify
and validate the decision to implement innovative methods,
however, this process would need to be mandated by a higher

authority before use would be guaranteed.

The expertise of lawyers is first with the law and the
process of law. The priority of environmental engineering

is secondary, therefore a knowledge tool would be a asset to

139

 

second guess and verify a recommendation. A large company
with an in-house environmental department expressed interest
in this tool. The knowledge base is useful in that they
often employ inexperienced engineers to make the technology

selection, and rarely go outside the company for advise.

140

Appendix B

KNOWLEDGE ACQUISITION IRON EXPERT # 1

Appendix B

KNOWLEDGE ACQUISITION FROM EXPERT I l

HAY FRP‘I'Y FXPEY 1
WIGIIAL GRID

ELEMENTS 1 z 1 2 Q 1
O - - cost of construction
1
1

..
....

‘

power irput for aeration
risk of short circuiting flow
iii operationel costs

ow space rewireumnt

lerates fluctuating flow

ludge bulking/foaming problem
olerates shock loading, toxics
igh recycle rate

iodegrades suspended solids
1 oxygen transfer rate

erates intermittent flow

I > 100, r sludge settlability
udge easi y dewatered

biological solids prediction
NLSS concentration

s
lved oxygen > 4

le am removal >10 s

lete mix flow regime

flow r Ime

h raulic retention time, > 10 hours
trif cation occurs

24 Denitrification occurs

25 Prinry settling tank rewind

0
do‘

ddo‘

o

«m-er-zr-r-

0 0
0 0
0 0
$303.8“

ii"

0 D

3

58

N N ....d-b-Dd
aaufiagoowom533=300~oueu~
a a
o
”Naps-ada-a-a-aa-aad
N-I OONOMO~UNAOODVOUDN~
3’...
8"-

. 0

80-10
.e. e-
8

0
0
0
1
1
1
1
0
1
1
1
0
1
1
1
1
1
1
1
1
1
1

“£5:ng
Conventional coapletely mixed reactor
2 Step feed, plug flow reactor
3 Deep sheft
4 Seqnncing batch reactor
5 Pure oxygen couplets mix reactor
6 Oxidation ditch, extended aeration

d

141

__2- 3 4 45.x:g '_§- 4 5 9 "5’ 5 §-_5- §_::i

 

EC.

20‘1100020100202200122220
0011050110100000000102000
20°2110110000202200020220
2005‘2022010000020010222°
0012020200000202000220000
20,0110110000000200200220
2021020220100000222102220
o012120200000202022220000
205001011ooooooozzzzoozzo
o020100000000000022000000
10°00 0°00 0100002000120020
alOsleldlooozoooozoozoooozzoo
el011010110000002000022020
1001120220000002200222200

al021020220000002222222200

7 5 7

7 18 20 25 15 15 20 20 9 23

14 14 9

26 21

dij

FOCUS OF THE ELEMENTS

SIMILARIIY VALUES (SV)

Exams 2 3

DIFFERENCE MEASURES (dij)

Q

5

A

 

STJL

3

2

 

flflifls

umwufi
nwmw
nan

8
5

123‘s

14»2

2° 012022010000 2 2 2 o

2012100020‘0020 2000 2 o

11.—r 00 20202000002020 220000

 

. C o C o o

7 1o 111‘1'101‘101‘ 1.11111
. . . . . . . . . . . . .

o o o o o o u c c o o o

RENT

dij 14 15 16

SIHILARITY VALIES (SV)

DIFFERENCE flEASURES (d1 1)

IT

14 15

5

16

143

LT

F

LENEIT

.._

-9—

$1175

20 1 0 20100 0 2

23‘saraooazuuuuwmwmauaua

O c o u o u

20

10

dij

144

FMS OF THE WSYRUCTS OF EXPE" I 1

d1] VALIES

,‘Oa‘szs7252226
1' Cl C!!!
"“O'S‘s3‘5““
"646.36316566‘6
96“.3‘30I‘5“66
76‘W69097OSOOZO
355352724767775
762643035050046
0| cl ‘0!
762643035050046
cl 1' ‘1!
"64633631656646
766WO9097353326

121-‘56,...0‘23‘5

F

123‘567U9012
1' 1|

‘

13
14
15

A

an
wanna
038%
annunnuua
1 a
anew
r nuns a
69.05 a
7 055% a
1 a

145

Table 30. Primary matcher for Element 1 from Expert 3 1 -
Conventional completely mixed reactor

 

 

   

 

 

 

 

 

 

 

DESIGN CONSTRUCTS HEIGNT Y N SECONDARY CONFIDENCE
FACTOR NATCNER FACTOR
> 15 days +1 -1
Nitrification occurs +1 -1
Plug flow regime -1 +1
HRT > 10 hours +1 -1
Denitrification occurs -1 +1
High MLSS concentration -1 +1
Dissolved 02 > 4 mg/l +1 -1
Soluble 800 removal > 90 x +1 -1

 

 

 

 

 

 

DEGREE OF IATCN

 

 

 

 

 

 

 

 

 

 

Table 31. Primary matcher for element 2 from Expert # 1 -
Step feed, plug flow reactor
DESIGN OONSTRUCTS NEIGNT Y N SECONDARY CONFIDENCE
FACTOR NATCNER FACTOR

Sludge Age > 15 days -1 +1

Nitrification occurs -1 +1

Plug flow regime +1 -1

HRT > 10 hours -1 +1

Denitrification occurs -1 +1 W
High MLSS concentration +1 -1

Dissolved 027> 44mgll -1 +1 "
Soluble 800 removal > 90 X -1 +1

 

 

 

 

 

 

DEGREE OF IATCN

 

146

Table 32. Prime matchr foeemenlt 3 frE t -_slunft_des_'~

 

 

 

 

 

 

 

 

DESIGN CONSTRUCTS NEIGNT Y N SECONDARY CONFIDENCE
FACTOR NATCNER FACTOR

Sludge Age > 15 days -1 +1

Nitrification occurs -1 +1 :
l Plug flow regime +1 -1 +
1 NRT > 10 hours -1 +1 ;
1 Denitrification occurs -1 +1 '
‘ Nish NLSS concentration +1 -1 .
. Dissolved 02 > I. Tug/l +1 -1 i

Soluble BOD removal > 90— X ‘

 

 

 

 

 

beans: or up —---_

Table 33. Prime matcher for element 4 from* E . rt I 1 - s ffl~ ing batch_reactor_‘

 

 

 

 

 

 

 

 

, DESIGN CDNSTIUCTS NETGNT Y N SECONDARY CONFIDENCE
‘ FACTORS NATCNER FACTOR
I Sl -~e A- > 15 days +1 -1
; Nitrification occurs +1 -1
’ Pl 2 flow regime +1 -1
j HRT > 10 hours -1 +1
; Denitrification occurs +1 -1
Eh Nigh NLSS concentration +1 -1
Dissolved 034? 4 mg/l +1 -1
Soluble BOD removal > 90 X +1

 

 

 

 

 

 

 

_DEGREE OF NATCN

 

1117'

Table 34. Prime matcher for element 5 from Expert # 1 - pure oxygen cggglete mix reactor

 

 

 

 

 

 

 

DESIGN CONSTIOCTS NEICNT Y N SECONDARY CONFIDENCE
FACTOR NATCNER

Sludge Age > 15 days -1 +1

Nitrification occurs -1 +1

Plug flow regime -1 +1

HRT > 10 hours -1 +1

Denitrification occurs -1 +1

 

  

Ni-h NLSS concentration

 

 

 

Dissolved 0, > 445g!l

 

I

 

 

  

I Soluble BOD removal > 90 X

   

 

 

 

 

 

new or m_cu ----—

 

 

 

 

 

 

 

 

 

 

 

 

Table 35. Prime matcher for element 6 from Expert # 1 - oxidation ditch

DESIGN CONSTRUCTS NEIGNT Y N SECONDARY CONFIDENCE

FACTOR IATCNER FACTOR
> 15 days +1 -1

Nitrification occurs +1 -1

Plug flow regime +1 -1

NRT > 10 hours +1 -1 I
[Denitrification occurs +1 -1 I
I High NLSS concentration +1 -1 I
' Dissolved 037> 4 mg/l +1 -1 I

 

g Soluble BOD removal > 90 X

 

 

 

 

 

 

oases: or urea ---—-

Ju4£3

Table 36. Secondary matcher for Expert # 1 - Sl e a e > 15 days

 

 

FEA‘lllE 2 Q Q i 1.4.
mm
3V + + + + +

0.92 0.75 0.75 0.67 0.67

Table 37. Seconds matcher for Exggrt # 1 - Plug flow regime

 

 

 

 

 

 

  

d
‘~
1“
O

 

 

 

 

 

Table 38. Secondary matcher for Expert # 1 - Hydraulic retention time > 10 hours

 

 

rum 2 Q 5 1 5 Z w l; E
m
SV + + + + + . .

 

 

 

 

 

 

 

 

 

EEO.” 0.75 0.75 0.67 0.67 0.67 0.67 0.67 0.67

 

Table 39. Secondary matcher for Expert # 1 - High HLSS Concentration

  

 

Table 40. Secondary matcher for Expert # 1 - Dissolved o, > 4 mg/l

 

 

FEMIIE 1 5 5 l 2 I 13 H
mm
3' + - - + +
0.83 0.75 0.75 0.67 0.67

 

 

 

Table 41. Secondary matcher for Expert # 1 - Soluble 800 removal > 90 X

 

mums 1 9 § 5 .11

 

 

 

 

 

 

 

 

149

 

Appendix c

RNOILBDGB ACQUISITION RESULTS FROM EXPERT f 2

FOCUS 0F ELEMENTS

14'1125

 

3 4 5 6 7 8 -§- 4 5 6 7 .48 "i' 5 AIQL 7 §-§- 6::;7

 

NPONNc—OOOPPPPPNONNNNPONOON
PPPPNPNPOPNOPPONNNONPONNNN
PNv-v-OONe-ONPPONNNOONOOOONNO
Pv-v-FNPOOOFNOPFNOI—v—OOPONNNN
FOPI—OOQOONPPOOOOPFNNOOONNO
ONOOOONv-OOOOONNNv-v-ONOOOOOO
NPONNI-NOOPFPPPNPNNNNPONONO
ONOOOONOOOOOOOOc—OOOOOOOONN
POPv-OOOPONPPONNPOONOOOONON
v-NPPOONOONI—POOOe-o-PNNOOONON
FOG-www.-NPOFOFO—PONOOPONOFN
PPPFPOFNPPOPOPFPNONNDOOGv-O
OPOOFOPPFPPOOPPFNOONOOONPO
O'OOv-Qc-Nv-c-e-cOv-v-e-c-v-OOOOONv-O
v-v-e-c-e-Oc-Nv-PGPOPI-GNONNOOOOFN
POPPPOPO'OOOOPPFNNOOPNNPON
Pe-e-e-o-v-e-e-e-e-I-I-v-e-e-v-e-Pv-v-e-e-e-v-v-e-
OPOOv—PPPPPNOPOPPOOONNNOPNO
OPOOv-Pi-Ov-I-NOv-Ne-v-v-POONNOU-NO
C-e-I-v-v-PF'OPPPPFNPOOONNNNOFNN
OOOOOPONOOPOFPOGNOOONNOPFO
POOPPPOCMNOPOPPPPPNOOPONQON
e-v-OPFOONNPOPOOFPPONNOOOOOO
OPPOFONPNFPOONPFPOONOOONNO
OPFOPOONNPPOOOPPOPOOOQONNO
PPOPPONNNPOPOOPOFONNOOOONN
OOPOOOPOPOOOOPOOPOQOOOOOPO

0°!-OOPFNPOFOPNOOPOOONNOPOO

. ~nvn°~~°ezenza°~~°samnaaa

IS I

15 17 25 24 32 28

14 28 22 21 26 21 22 16 17 21 21 22 19 9 31

17 21

16 6 22 16 21

di]

MEN?

NMQU‘QN

24 32
28

22 16 17 21 21

14 28 22 21 26 21

NM‘U‘ON

 

 

:682113121139114'41-fi

MGIIOOZ

'- OO-v-O NNv-Ov-Oo-c-c-NOIUNOOOO
N GNNP OOP'PP'NONNOUNv-ONO
0'00vaONNOPOPOPPONOOPGNOON
'- v-00 Po~ww°NNNOONO°OON
OP'O'ONPNP'OOPPFNOONGOONNO
Pvt-0'0"-NPOPNO'PGNNNONPONNNN
'- Pv-OO OGNP—OOOOP'NNOOON
Ot-C-Ov-OONNv-v-OO'PP'PQOOOONNO
v"-Pv-Nv-OOO'NOI-PNOC-POOPONNNN
ONOOOON'OOOOONNNFC-ONOOOOOO
'- e-c-e-o- Owwao-N'e-QQNNNNQC-
GOO-GGC-o-No-OI-OPPOONOOONNOPOO
wove-FO'GU-ooeac-wI-NNOOU-NNU-ON
0'90?!-v-v-v-o-NOv-OPPOOONNNOPNO

OPOOC-c-v-Ov-u-flcv-NPO-v-v-GONNOPNQ

~ncn¢~ooe:~e:eseeezzanxna

'

'- v-v-e-O v-e-POC-Ov-o-o-o-u-o-e-Pv-e-e-
o a o o n o

°°'°°9"'°°°°°°°""'F""
. S 8 8 S S O

I 8 0 8 0 O 8

8 8 0 0 0 I I 8 0 I

0 I 8 8 8 0 8 0

ooocov-Ov-OOv-Ov-v-oov-Pwe're-POP!-
I I O I

gg[~n*r*~~°2zeozcatsezzanana
8

I52

22212116231581815522118192518

d1]

58 60 60 69 68
71 52 65 66
38 60 59

63 45

59

NGOOO

23

18
25
18

18 15

21
19

25
32

22 21 21 16
15

2
5
6
8
9

0 1011 21201 00 0220
II 0511- 255552“ 022 00
1 1100 002110 002 0002
1001100 20000111. 200 2002

an 1121. 001201 220 102222

41 12 'anj

 

z 0221 00111112022221020

23456769”fluuufiuwuwmvnauaa

0010011 1.0101100 111 1111
o o o o o o o

o o o o o o o

o o o o o o

O o o o o o .

4.19—1.14

““““"

252314111813

d1]

‘83
111

1
‘

001
“

153

FUSEDGIIDI4

2 022‘ 21a!

0 a o o .

.1}_16..

‘1‘1

aslsaracoazsuuuwmwmuuauau

mi!

11
31

dij

154

FOCUS 0F CONSTRUCTS

Emmmfli.JL;E_fiL31Jn_flLlLJE_fL3§

 

171419

8333£33333333333
°3K33333°°33333R

3°£33m333nn3333

56
50
44

44
5056
38 31
69 75

56

50 69
56
69
50
50 69

50 56 56 56
50 69
44 63
38 81
63 56 56
38 56 31

3333333333
°°3 333335333383
3333333333333333
33333333333333E3

"NMQU‘ONQOOPNFHC who
e-II- e-e-e-

OOQNOONOOO'QOOOBQ
OOQQONIflth-e-mklflnln
F'-

PNOFNNNOONPONPQN
'- e-e-e- e- '-
honkmsnonksnhmom
OKPONNOOOOOOOOU‘O
.- e- e- v- e- .-
mOOIn‘tOgOInONmOOQO
mQQmNOOIflNPOIflONON

'- v- '-
OOOONNO‘OOOOQOOmQ
e- e-e- e- '-

OOQOPOQOGOONIBOQIB
'- e-e-

PNMQU‘ONfloczNMQmO

155

gagggg .ILJ1;E_flL3LJR_fl;flLJLJ&

3R 3 33
E 33
3 R3 3

3333K
3 3 3
fl
3 3 3
3 3
33 3E 33 3
3 3 3 3
33 3 3 33
33 3 3 E3

8 3 338

PNM'IflONOOOC-Nn~flh°
e-v-e-e-e-e-e-

 

' Cost factors, etc.

 

’ Soluble BOD removal > 90 X

 

Dissolved 0, > 4 m!!l

 

‘ Primary settling tank required
‘ Sludge age > 15 days
‘ Pl . flour ime

- Space loading > 1.5 lbDOD/cu ft-d

 

 

 

 

' Nitrification single stage

  

 

Sus- 3 - BOD removalfl~etc.

Table 43.

DEEREEfiSF IITCU

 

 

 

 

 

Prima matcher for element 2 from Ex rt I 2 - conventional c

 

 

   

 

 

 

 

 

 

 

 

 

Cost factors, etc. 0 D
Soluble BOD removal > 90 x 0 0
Dissolved 02 > 4 mg/l -1 +1
Primary settling tank required +1 -1
Sludge age > 15 days -1 +1
fl Plug flow regime -1 +1
m Space loading > 1.5 leOD/cu ft-d -1 +1
" Nitrification single stage +1 -1
Sus BOD removal etc. 0 0

 

 

DEGREE 0f HITCH

 

1£5€5

 

 

 

Table 44.

   
 
 

 

Prima matcher for element 3 from Egrt # 2 -

st: feedI Bl: flow

 

 

 

 

 

 

 

 

  
 

 

    

 

   

 

  

 

 

 

  

 

 

 

 

 

 

    
 

  
 

 

 

   

 

 

 

 

 

 

 

DESIE ms IEIGNT T I SEW
rm MEIER
Cost factors, etc. 0 0
Sollble aoo removal > 90 1: +1 -1
Dissolved 02 > 4 mill -1 +1
Primary settling tank required +1 -1
Sludge age > 15 days -1 +1
Pl flow regime +1 -1
Space loading > 1.5 lthD/cu ft-d -1 +1
Nitrification single stage 0 0
etc. 0 0
MQEE N M1011
“a, f 2 ' $211M“. _
DESIH ms EIGHT Y I EMMY
rm MTCIER
1 Cost factorsJ etc. 0 D
Sollble soo removal > 90 11 +1 -1
Dissolved 02 > 4 I!“ +1 -1
Primary settlinuank required -1 +1
Sludge age > 15 days -1 +1
Plug flow regime +1 -1
Space loading > 1.5 lechu ft-d +1 -1
Nitrification single stag; -1 +1
Sus 3m removal etc. 0 0

 

DEQEE G M3011

 

 

157

 

 

 

 

 

 

"b“ ‘6- "1"" "“1"” f°" “em" 1 :11!" ',°"‘,'c “Nubia”

 

 

 

 

 

  

 

 

 

  
     

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

*1“ ms 1mm Y I SEW ounoace
FACTG “TUE! rm

Cost factors, etc. 0 0
I Soluble Ba) removal > 90 X -1 +1
I Dissolved 02 > 4 I!“ -1 +1

Primary settlim tank required +1 -1

Sludge age > 15 days 0 0

Pl flow r ime +1 -1

Space loading > 1.5 lbDw/cu ft-d -1 +1
Nitrification single stage -1 +1
1 Sue. . removla,etc. 0” # _ W _¥
Jew M__ ________- _a,_‘_-_ fly, _ *
Table 47. Prime matcher for element 6 from E rt # 2 - high rate

DESIH ms 516'? Y I SEW

rm MTCIEI

Coat factors, etc. 0 0

Solible aoo removal > 90 X -1 +1

Dissolved 03 > 4 mg/l -1 +1

Primary settling tank required +1 -1
_S_l1.£ge age > 15 days -1 +1

Plug flow regime +1 .1

Space loading >1.5 lb Dm/cu ft-d +1 -1

Nitrification sfle sgge -1 +1

 

 

 

 

 

 

 

 

 

Table 48.

  
  

 

Prime matcher for element 7 from Bart I 2 - are an

 

 

 

 

 

 

 

 

  

 

  

 

  

*1“ ms 1£16IT Y I SEW
rm MTG“
Cost factors, etc. 0 D
Soluble son removal > 90 X +1 -1
Dissolved 03> fll +1 -1
Primsettling tank required +1 -1
Sludge age > 15 days -1 +1
ime +1 -1
Space loading > 1.5 lbDw/cu ft-d +1 -1
. Nitrification single stge +1 -1
L_ ;=_° =_=J.===°

 

 

 

 

    

W m

 

 
 

 

 

 

 

 

Table 49. Primary matcher for element 8 from Expert # 2 - extended aeration

 

 

 

 

 

 

 

 

  

 

  
 

 

  

 

  

 

 

 

DESIH ms IEIGIT Y I sacrum “Hm
rm MTCIEI PM
Cost factors, etc. 0 0
Solible 300 removal > 90 X +1 -1
Dissolved 02 > 4 mil -1 +1
Prim settang tank required -1 +1
Sludge age > 15 days +1 -1
Plug flow regime +1 -1
Space loading >1.5 lb Mlcu ft-d -1 +1
Nitrification single stage +1 -1
Suspended 8m removal, etc. 0 = 0

 

 

 

 

 

M“ N MTCI

159

 

 

(.-

11

Table 50. Secondary matcher for expert I 2 - cost and space factors

 

i 1.1.

 

 

 

 

 

 

 

   

(_ ___

 

 

Table 52. Seconda matcher for Ex.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

160

(construct 10 omitted, 0 score)

Table 56. Secondary matcher for Expert 8 2 - Space loading (constructs 1, 4 and 12 omitted, 0 score)

 

 

FEATIIIE 5 2 3 8

ms

W + + + -
.75 .69 .63 .63

 

 

 

 

 

Table 57. Secondary matcher for Expert # 2! - “trifle-union (constructs 13 and 6 omitted, 0 score)

 

 

PERM Z i 16 12 1.4 12 9

ms

sv a» + + - + + +
.88 .75 .75 .69 .63 .63 .63

 

 

 

 

 

 

 

 

 

Table 58. Secondary matcher for Expert I 2 - Suspended soo Removal, ETC.

 

 

mums 1 1 1.4.

comm

SV + + +
.8 .8 .8

 

 

 

 

 

 

161

Appendix D

KNOWLEDGE ACQUISITION FROM EXPERT f 3

ORIGINAL GRID roa EXPER'I' I 3

 

EBEHENTg 1 2 3 4 5

1 -1 1 1 41 1 High maintenance requirement

2 1 -1 1 -1 -1 High power requirement

3 -1 1 1 -1 1 High rating costs

4 -1 1 1 -1 1 High c lorinated hydrodrocarbons.
5 1 0 -1 0 1 Toxic by-products of biodegradation
6 -1 -1 -1 1 -1 High lead concentration

7 1 -1 -1 1 1 Lou mobilit of contaminant in soil
8 -1 -1 1 -1 1 High capita investment

9 1 1 -1 1 1 Not in close prox1mity of electricity
10 -1 1 1 -1 -1 8TEX concentration < 0

11 1 -1 '1 -1 1 8TEX concentration > 10, 0 ppb

12 1 -1 -1 -1 -1 TDS 3 30,000 mg/l, inorganic

13 -1 1 1 -1 1 Density of contaminant > water

14 1 -1 -1 -1 1 pH adjustment required

15 1 1 1 1 1 Iron removal required

16 * -1 1 1 -1 -1 Removal rate to 1 ppb

17 * -1 1 -1 1 -1 Flow rate > 15 ISO

18 * 1 0 1 0 1 Long residence time

19 * 1 -1 1 -1 1 Contaminant chemically destroyed
20 * -1 0 0 -1 -1 Lou temperature

EL§N§NTS

1 Aerobic fixed film

2 Activated carbon

3 Ultraviolet oxidation of H202
4 Air stripping . .

5 Anaerobic GAC fluidized bed

162

FOCUS OF ELEMENTS

d

-l.'.hi'. s l

20221202002022002120

7.
5"5

02002020222002020001

3 22221222220020022121

5

00001022022002022121

4

20220220020020020001
~+ 02001002200000002120
5 zzzzooozooozzooooooo
IQ 02001200002202002120
3 20222022222222020001

22221020022222022121

 

23‘567890123‘56789

I NT

22

17

19 29

16 14 12 17

27

dij

385
55‘
‘ 8
5%
M- 2
V
V-
'I
III
R T
A I
u .m. 2345
H
972
112
4 Eu
2
1
2355
s
a .1
N
cl
"

163

 

CU ATIONS

d1

FUSE! 0810 i 2

5

ELEMENTSTs

7

1

6

6
:_7___7

 

EIENTS

202 22 202 0 al
120 002 2002120

2 22122 2002002 al al

123‘567090123‘56709”
1111

111 111 1111111
. n o a o

1 11111 1111111 1 o
c o o o c a

o o o c o a o a u o o o .

123‘567090123‘567090

11111111112

12 13

19

dij
SIMILARITY VAL

m

SV

TABL

di'

7
13

NT

35

164

7013

L
LEHENTS

FUSED GRID O 3

Shag“?

6

:5

T8

9 8

1
2
Z
2

0201

123‘567890123‘56789”

1 1 11 ‘1 1 1
o u o

1 11111 1111111 ‘ o
c o o o c .

123‘567890123‘567890
11112

10

dij

8V

165

FMS OF MSTRUCTS

i

CNSTRUCTS

@Z“‘6‘84826“68

9".“‘8‘2662‘oQ2‘

1

8t66628£45646£62

1

W68666268“86686

 

67‘2286058086286
1| cl

123‘567890123‘5

VA
CUISTRUCTS

wwwwwwww

w wwww mm wwwmw
n wmwm mm» m mm»
n w mm

n mm» m m m

123‘567890123‘5

166

 

; Removal rate to 1 ppb

 

Lou teuperature
J Contaminant destroyed

 

 

' Flow rate > 15 H.

 

 

 

 

 

 

Highly chlorinated hydrocarbons

 

T_l_e. Watcher for eemenlt 2__ from rt _# - acvatited carbon__

 

 

 

 

 

 

 

 

 

 

 

I DESI“ warms IEIGIT Y I swam" GII‘IIEHIE—l
i rm MTCIEI new:

E Removal rate to 1 ppb +1 -1

1 Lou teuErature O 0

1 Contaminant destroyed -1 +1 1:.
5 Flow rate > 15 MGD +1 -1 i
i

 

Tabe 61. Prr matcher for el_t 3 rt # 3 - ultraviolet ition of II -

 

mm Y I SEMY WHOM
FIE"! ”TUE! FACTCR

 

I Removal rate to 1 ppb +1

 

. Low temperature 0

 

Contaminant destroyed

 

Flow rate > 15 MGD

 

 

 

 

 

 

DEREE N MTCII

167

Table 62.

   
  

  

Prilna matcher for element I. from E

rtl3-airst

SEW
MICE!

   
   

 

  

Removal rate to 1 ppb

+1

 

 

  

Lou teuperature

+1

 

 

Contaminant destroyed

+1

 

  
 

Flow rate > 15 H.

-1

 

  

Highly chlorinated hydrocarbons

DEflEE 0F MTCII

 

 

 

 

 

 
   

Table 63. Primary matcher for element 5 from Egrt # 3 - anaerobic GAC fluidized bed

 

 

 

 

 

 

 

06310 ms lEIGIT Y I SEW GIFIDEKE
me MTCIER rm
Removal rate to 1 ppb -1 +1
I Low temperature -1 +1
ll Contaminant destroyed +1 -1
I Flow rate > 15 H6!) -1 +1
Highly chlorinated hydrocarbons +1 r4

 

 

 

 

 

DEGEE N MTG!

 

 

 

168

 

Table 66.

 

 

Secondary matcher for Expert I 3

 

- Removal rate to 1

 

 

______

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 65. Secondary matcher for Expert # 3
- lou t rature
EE
FEATURE 19 g 3
ms
SV + - 9-
.80 .60 .60
Table 66. Secondary matcher for Expert 3 3
- contaminant deatr ed
“a
FEAT“! 3 Q n 110. 1 9. 2 Z J; 11 1:
ms
SV + - + + - - + + + + +
.80 .80 .80 .80 .60 .60 .60 .60 .60 .60 .60 .60
Er ==
Table 67. Secondary matcher for Expert 8 3
- Flow rate > 15 H60
FEATURE Q 2 19 l
COISTIUCTS
sv - + + }
.80 .60 .60

 

 

 

 

169

 

 

Appendix E

SAMPLE CASE CLASSIFICATION

170

Table 68. Primary matcher with confidence factors and degree of match for Element 1 -
_f __Li’L‘d "“3“" _ .-

  
   
  
   
   
  
  
  
 
  

 

 

 

i Nitrification occurs

 

; Plug flow regime

 

I HRT > 10 hours

 

. Denitrification occurs

 

2 High HLSS concentration

 

Dissolved 02 > 6 mg{l

 

 

 

 

 

: Soluble 800 removal > 90 X

LDEGREE 0F HATCI

 

Table 69. Primary matcher with confidence factors and degree of match for Element 2 -
Step feed, plug flow reactor

   
  
  
 
 

 

 

 

   

 

 

 

 

   
 

 

   

 

 

 

 

DESIGI COISTIUCTS Y I SECDIDARY

IATCIEI
Sludge Age > 15 days -0.43 -0.63
Hitrification occurs -1 +1 +1.0
Pl flou r ime +1 -1 -1.0
HRT > 10 hours -1 +1 -1.0
Denitrification occurs -1 +1 +1.0
High HLSS concentration +1 -1 -1.0
Dissolved O, > QAmgll
Soluble 800 removal > 90 x -1 +1 -1.0

 

_ ___—___..- _._._._._ __,._____. -— _. _.‘_-_ _ _, ___ ___..4

W—M

 

T
L

 

171

Table 70. Primary matcher with confidence factors and degree of match for Element 3 -
0 ;T 8h'ft_993‘90 _..

 

 

 

 

 

 

 

 

 

 

 

 

I IESIEI CDISIIUCTS Y I SECDIDAIY CDIFIDEICE

; MIC” rm

1 Sl '3» A-7 > 15 days ~0.63 -0.63

E Nitrification occurs -1 +1 +1.0 ,

; Plug flow regime +1 -1 -1.0

{ HRT > 10 hours -1 +1 -1.0 I

: Denitrification occurs -1 +1 +1.0

I High HLSS concentration +1 -1 -1.0 I

1 Dissolved 02 > «#5911 0.33 +0.38 '
Soluble 800 removal > 90 1 +1 -1 +1.07

 

 

1 1
'1—m-1

 

 

 

 

 

 

 

 

 

 

Table 71. Primary matcher with confidence factors and degree of match for Element 4 -
S . - ing batch reactor
Ir iiiiiii 7 — fi—Vn r~~-~~— :—
DESTGI COISTIUCTS Y I SECDIDARY CDIFTDEICE
; IATCHER FACTOR
i Sludge Age > 15 days 0.63 +0.43 '
1 nitrification occurs +1 -1 -1.0
T Pl . flow r ~ime +1 -1 -1.0
; HRT > 10 hours -1 +1 -1.0 I
|
’ Denitrification occurs +1 -1 -1.0
: High MLSS concentration +1 -1 -1.0
! Dissolved 03 > 4 mg{l 0.38 +0.38
i Soluble 800 removal 3_?0 x +1 -1 _+1.o_

 

 

 

 

T __E ___ _nh________m__- ___fl_

1372!

Table 72. Primary matcher with confidence factors and degree of match for Element 5 -
Pure o ,‘FQt‘VT‘*_'°QCt°r_4,

 

 

 

Nitrification occurs

 

Plug flow regime

 

HRT > 10 hours

 

" Denitrification occurs

 

 

I High MLSS concentration

! Dissolved 03 > 6 mg/l

 

 

 

 

 

I soluble Boo removal > 90 x

DEGREE 0F IATCI

Table 73. Primary matcher with confidence factors and degree of match for Element 6 -
Oxidation ditch

 

 

 

 

 

 

 

 

 

 

 

 

 

 

DESIGI CDISTIUCTS Y I SECDIDARY CDIFIDEICE
IATCIEI FACTOR
Slugs Age > 15 days 0.43 +0.43 ‘
Nitrification occurs +1 -1 -1.0
I Pl . flow regime +1 -1 -1.0 J
‘ m > 10 hours +1 -1 +1.0 ‘j
; Denitrification occurs +1 -1 -1.0 I
1 High HLSS concentration +1 -1 +1.0
J Dissolved 03 > 6_mg{l 0.38 +0.38 [
Mel M_______ ____ +10
assess or 1mm 0.10 l

 

173

l‘ u“-

Table 7%. Primary matcher with weight factors, confidence factors and degree of match for
_Element 1 - Cogzgntional completely mixed reactor__

 

 

 

 

   
  
  
  
  
 
 

 

 

 

 

 

' DESI. ms IEIGII‘I' Y I SEW C(IFIDEICE 1:

FACTOR IATCIER FACTOR ‘

Sl '30 Are > 15 days 0.8 +0.43 0.36 .

‘ Nitrification occurs 0.1 +1 -1 -0.1 l

Pl . flow regime 1.0 -1 +1 +1.0 I
HRT > 10 hours 0.5 +1 -1 +0.5

Denitrification occurs 0.1 -1 +1 +0.1 I
High NLSS concentration 0.5 -1 +1 +0.5

Dissolved 02 > 4 mgll 0.5 +0.38 +0.19 I

|

 

 

 

 

 

 

 

 

 

 

Soluble 800 removal > 90m§__um __ 0.3 ____mw__+1 __f1 ________________________+0.3______a
,1__,, a; _ a _ a. --

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 75. Primary matcher with weight factors, confidence factors and degree of match for
___ ___—“5}. '_____9°‘°____ -_ . ,,____ __-_
DESIGI CCISTIUCTS HEIGHT Y I SECOIDARY CCIFIDEICE I
new: names rm '
I Sludge Age > 15 days 0.5 -o.l.3 -o.22 l
{ Nitrification occurs 0.1 -1 +1 +0.1 ‘
Pl 3 flow r ime ' 1.0 +1 -1 -1.0 1
HRT > 10 hours 0.5 -1 +1 -0.5 '
" Denitrification occurs 0.1 -1 +1 +0.1 '
" High NLSS concentration 0.5 +1 -1 -0.5 E
" Dissolved 02 > §:gg{l 0.8 -0.38 -0.3 ’
1. soluble ooo removal__ _x________________ -1 +1 -1.0 5
DEGREE or HATCH -o.s2 J

 

1'74

Table 76. Primary matcher with weight factors, confidence factors and degree of match for

Element 3 - Deg: shaft design

  
 
 

 

 

 

 

 

 

 

 

 

  
  

 

 

 

 

 

15161" T I
FACTOR
0.5
Nitrification occurs 0.2 -1 +1 +0.2
Plug flow regime 0.5 +1 -1 -0.5
“AHRT > 10 hours 0.5 -1 +1 -0.5
Denitrification occurs 0.5 -1 +1 +0.5
n High NLSS concentration 0.5 +1 -1 -0.5
“ Dissolved 02 > 44mg!l 0.9 0.38 +0.36
Soluble 800 removal ?_90¥§47 70.3¥W

 

 

  
    
   

 

 

 

 

 

 

    
 

 

 

 
 

 

 

 

 

 

Table 77. Primary matcher with weight factors, confidence factors and degree of match for
Element 4 - S i batch reactor
Y
IATCIER FACTOR
"Sludge Age > 15 days 0.5 0.43 +0.22
Nitrification occurs 0.5 +1 -1 -0.5
Plug flow regime 0.5 +1 -1 -0.5
HRT > 10 hours 0.8 -1 +1 -0.8
“Denitrification occurs 0.1 +1 -1 -0.1
Hieh HLSS concentration 0.5 +1 -1 -0.5
Dissolved 02 > 4 m!!l 0.5 0.38 +0.19
SollbleBm removal > 90 X

 
  

 

  

DEEHEE OF IATCI

 

175

Table 78. Primary matcher with weight factors, confidence factors and degree of match for

 

 

Nitrification occurs

 

Plug flow regime

 

1 HRT > 10 hours

 

' Denitrification occurs

 

High NLSS concentration

 

Dissolved 024? ‘-E§!l

 

 

 

 

 

 

Soluble 800 removal > 90 X

 

DEGREE OF IATCI

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 79. Primary matcher with weight factors, confidence factors and degree of match for
__- , A ___-___ --__ _ ,_ __ __ __ I
DESIGI COISTIUCTS NEIGIT Y I SECOIDARY COIFIDEICE :

FACTOR IATCIER EACTOR 1
Sludge Age > 15 days 1.0 0.43 +0.43 }
Nitrification occurs 0.4 +1 -1 -0.4 {
Plug flow regime 0.5 +1 -1 -0.5 1
HRT > 10 hours 0.5 +1 -1 +0.5 fi
Denitrification occurs 0.4 +1 -1 -0.4 L

..fliflh HLSS concentration 0.5 +1 -1 +0.5 f

H Dissolved 02 > 44mg(l 0.5 0.38 +0.19 l
Soluble 800 removal > 90_N _“ 91§______ +1 -1 ---____-f°'5 __h

 

 

1 mm m __ ___.fi-#- g!’

7‘_ _—____— 4

 

13765

Appendix A

» Excerpt from

COKHBRCIALIZAIION PLAN

prepared by
Lori Schuts-Riley
ETA 810
Summer 1992