. ....f... . ....:.I:
4T2? ~ ....‘. («’2‘

:21;
.s...._.

3:1:1.
2.

I ‘
~.~_ \~,. .c.
.2

Va

«7:3.

‘ :1
‘tv

. 5.1.1:

: . 1:. . .
N? .

3.: a

i .3
r a

. . t ‘

, 1.:7. : _: :

nauruqqnu

n: .Eguvu

L... c. ‘1.
3:14.». I i 3.:

I ‘31:; u

.T:§..::
.23... .\.:

 

MICHIGANS

l l"Willi!!!“Illll(lllllllllllllllll

0897 8664

was an

 

 

 

 

 

 

 

 

 

This is to certify that the

thesis entitled

Investigation of Hierarchical Classifaction for
Life Cycle Analyses

presented by

James Howard Benda

has been accepted towards fulfillment
of the requirements for

M.S. degreein Chem. Engr.

 

Date

0-7 639 MSU is an Affirmative Action/Equal Opportunity Institution

 

LIBRARY

Mlchlgan State
University

 

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

M=SU Is An Affirmative A=ctionlEqual Opport=unity Institution
c:\c|rc\datadue. ans-9.1

     
     
   
   
   
  

   

INVESTIGATION OF HIERARCHICAL CLASSIFICATION FOR
LIFE CYCLE ANALYSES

By

James Howard Benda

A THESIS

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

MASTER OF SCIENCE

Department of Chemical Engineering

1993

, ABSTRACT

INVESTIGATION OF HIERARCHICAL CLASSIFICATION FOR
LIFE CYCLE ANALYSES

By

James Howard Benda

Today, landfills are overflowing with valuable and misdirected
material. Life Cycle Analysis (LCA) is a tool which can provide
environmental impact-information to improve products to utilize
their entire life potential and prevent premature burial in landfills.
Wide acceptance has been limited. however, for several reasons.
such as: cost. data aggregation. and data limitation. The purpose
here is to investigate the development of an Expert System for LCAs
which can provide Life Cycle services. This Expert System, a
Hierarchical Classifier (HC). can be used to initiate LCAs by
identifying the processes involved in a products life cycle.
Currently. this classification system cannot provide the impact
information for an entire Life Cycle Analysis. This classification
system was constructed using a simplified LCA for paper, which
indicated the feasibility of using the HC if several modifications are
added. These are: 1) an ability to quantify environmental impacts.

2) a knowledge acquisition module. and 3) an End-User Interface.

To Biika and Isaac.

ACKNOWLEDGMENTS

I would like to thank my Mom and Dad. and my brothers. Steve
(and his wife, Ann). and Karl (and his wife, Diane). Through whom.
I have received repeated assistance, Without which. this work would

have been an impossibility.

 

 

TABLE OF CONTENTS

List of Tables ........................................................................... vii
List of Figures ........................................................................ viii
List of Symbols. Abbreviations. and Nomenclature ............... ix
Introduction .............................................................................. 1
Chapter 1: Life Cycle Analysis ................................................ 2
1.1 Characteristics ................................................................ 2
1.2 The Need for LCAs ........................................................... 6
1.2.1 Legislation promoting DCAs ................................... 8

1.3 Difficulties ....................................................................... 9
1.3.1 Problems Associated with LCAs ............................. 9

1.3.2 Diaper Report Example ........................................ 13

1.4 Current Work ................................................................ 15
1.4.1 Frankhn and Associates. Ltd. .............................. 16

1.4.2 Canadian Standards Association ......................... 17

1.4.3 RADS Group ........................................................ 17

1.4.4 European Community ......................................... 18
Chapter 2: Paper Life Cycle Analysis .................................... 20
2.1 Paper Products .............................................................. 21
2.2 Paper Life Cycle ............................................................. 23
2.3 Paper Raw Materials ...................................................... 23
2.4 Paper Production ........................................................... 23
2.4.1 Energy Use .......................................................... 25

2.4.2 Releases for Paper ............................................... 25

2.4.3 Releases for Printing ............................................ 27

2.5 Paper Waste Management ............................................. 28
2.5.1 Landfill ................................................................ 28

2.5.2 Incineration ......................................................... 29

2.5.3 Recycling ............................................................. 30
2.5.3.1 Recycled Fibergrade ................................... 30

2.5.3.1 De-lnking ................................................... 34

2.5.3.2 Energy Savings Through Recycling ............ 35

2.5.4 Composting ......................................................... 36

Chapter 3: The Expert System-Hierarchical Classification. 38

3.1 Expert Systems ............................................................. 39

3.2 Hierarchical Classification ............................................. 42
3.3 Rationale for Use of Expert Systems .............................. 47
3.4 The Computer System Experimental Setup ................... 48
3.4.1 The Hierarchical Classifier .................................... 48
3.4.2 HC Individual Window Descriptions ...................... 48
3.4.3 Failure Handling .................................................. 55
3.4.4 Updating the Knowledge Base .............................. 56
3.5 Difficulties in Use; Recommendations for Further
Development of the Hierarchical Classifier .................... 56
3.6 Limitations in the use of Hierarchical Classification ...... 59
Chapter 4: Paper Product Classification System ................. 61
4.1 The System ................................................................... 61
4.2 The Table Matchers ....................................................... 64
4.3 Sample Results ............................................................. 65
4.3.1 Corrugated Medium Example .............................. 66
4.3.2 Other Results ...................................................... 68
Chapter 5: Conclusions/Recommendations for Further
Research ...................................................................... 69
5.1 Benefits of Using HC for LCA ......................................... 69
5.2 Limitations of using Hierarchical Classification ............. 71
5.3 Future Work Recommendations .................................... 71
5.3.1 Higher Level Expert System ................................. 71
5.3.2 System Scope ...................................................... 73
5.3.3 Knowledge Acquisition and Domain Expertise ..... 74
5.3.4 Continual Validation of System Performance ....... 75
5.4 Impacts of This Research .............................................. 75
List of References for LCA ..................................................... 77
List of References for Paper ................................................... 80
List of References for Expert Systems .................................. 83
Appendix A: Snapshots of Working Windows ....................... 85

Table 1:
Table 2:
Table 3:

Table 4:
Table 5:
Table 6:
Table 7:
Table 8:
Table 9:
Table 1 0:

LIST OF TABLES

Environmental Impacts of Diaper Usage ..................... 13
Sources of paper pulp for US Paper Mills .................... 23
US Pulp, Paper, and Paperboard Industry Estimated

Fuel and Energy Use ................................................. 25
Releases for Paper Industry ....................................... 26
Specific Chemical Releases for Paper Industry ............ 26
Releases for Printing Industry .................................... 27
Specific Chemical Releases for Printing Industry ......... 27
Paper As Received by Weight Basis, MSW .................. 28
Change in Energy Consumption due to Recycling ........ 35
Sample Basis for Waste Management Rules ............... 65

Figure 1:
Figure 2:
Figure 2:
Figure 3:
Figure 4:
Figure 5:
Figure 6:
Figure 7:
Figure 8:
Figure 9.

Figure 1 0:

Figure 11:
Figure 12:

Figure 13:
Figure 1 4:
Figure 15:
Figure 16:
Figure 17:
figure 18:
Figure 19:
Figure 20:
Figure 21:
Figure 22:
Figure 23:
Figure 24:
Figure 25:
Figure 26:

LIST OF FIGURES

Stages of Ly'e Cycle Analysis ...................................... 3
Micro-Analysis Boundary ............................................ 4
EPA 1995 Goals ......................................................... 6
Stages of Paper Life Cycle Analysis ........................... 22
Paper Production Schematic ...................................... 24
Recycled HighGrade De-Inking Flow ......................... 31
Recycled Mixed Paper Flow Diagram ......................... 32
Recycled Pulp Substitutes Use .................................. 32
Recycled OldNewspapers flow .................................. 33
Recycled OldCorrugated Carton ................................ 33
Sources of Solid Waste from Deinking ....................... 34
The Development and Use of An Expert System ......... 40
Example of a Classification Scheme for a

Products LCA ............................................................ 43
Filter for Process Steps ............................................. 44
’Iiansportation and Energy Filter ............................... 45
Launcher .................................................................. 49
ClassUier Window .................................................... 49
DataBase Window .................................................... 5O
DataBase Variable Define Window ........................... 51
SubSpecialistRelation Window .................................. 51
TableMatcher Window .............................................. 52
Table Editor .............................................................. 53
Graphing Capability .................................................. 54
System ’Ii‘anscript .................................................... 55
Classification system for paper ................................. 62
Table Matcher Example ............................................. 63
Potential LCA Expert System Application ................... 72

LIST OF SYMBOLS. ABBREVIATIONS AND NOMENCLATURE

ADL Arthur D. Little

AI Artificial Intelligence

APA American Pulpwood Association

API American Paper Industries

CL Carl Lehrberger

CSA Canadian Standards Association

EPA Environmental Protection Agency

ES Expert Systems

HC Hierarchical Classification

ISO International Standards Organization
KBS Knowledge Based System

LCA Life Cycle Analysis

MSW Municipal Solid Waste

OOP Object Oriented Programming

PLA - Product Lifecycle Analysis

RDF Refuse Derived Fuel

REPA Resource Environmental and Profile Analysis
SETAC Society of Environmental Toxicology and Chemistry
A Logical And

V Logical Or

[1] Reference number 1

INTRODUCTION

The goal of this research is to obtain a better understanding of Life
Cycle Analysis (LCA), its application along with its difficulties, and
to determine the feasibility of using a Knowledge-based systems
approach to carry out these analyses. This task was accomplished
by investigation of Life Cycle Analysis methods, a case study of
paper, and an experimental implementation of this study with
Hierarchical Classification. Currently, problmes associated with
Life Cycle Analyses has made them difficult to use. The ease and

power of todays computers could be used to help eleviate problems I
generally associated with LCAs. The first Chapter of this Thesis is a
study of Life Cycle Analysis. This section discusses the general
character of LCAs, and the problems associated with them. The
second Chapter is a simplified LCA of paper to better understand
the process of conducting LCAs. This section is an LCA essentially
done by hand. The third Chapter entails the study of Expert
Systems, and more specifically, Hierarchical Classification. The
fourth Chapter describes the experimental LCA classification for
paper. The final Chapter is a compilation of conclusions along with

recommendations for future work.

CHAPTER 1: LIFE CYCLE ANALYSIS

INTRODUCTION

This Chapter introduces the concept of Life Cycle Analyses (LCA)
and presents the general character along with the associated
problems in these LCA implementations. The need for LCAs is also

presented.
1. 1 CHARACTERISTICS

The first LCA of products was undertaken in 1969[21]. The concept
is attributed to Harry Teasley with The Coca—Cola Company. The
purpose of his study was to determine which of several different
beverage containers produced the least effects on natural resources

and the environment. Since that time much work has been done to

develop LCAs, but as of yet their potential applicability and
technique of application is still unclear.

LCA is essentially an environmental and energy audit (accounting
procedure) that focuses on the entire life cycle of a product from
raw material acquisition to final product disposal, rather than on a
single manufacturing step or environmental emission. According to
Franklin Associates, Ltd., LCA is a quantified inventory of resources
used and releases to the environment in the entire life cycle of a
product; "cradle to grave." LCAs are analyses that can identify
opportunities to reduce environmental impact. The key is in
defining the system boundary and the assumptions that are
involved in defining this boundary. Much work has been done by
Franklin Associates in developing the research protocol and
underlying assumptions in LCA [17,18,19,21,24]. Figure 1 is a
generalized stage flow diagram for a complete Life Cycle Analysis.

     
 
   
   

Acquiring
Relining

Raw Materials Shipping

Recycling

      
   

Composting
Process

  

Production Process

Raw Materials Storage - Product Analysis
Processes Packaging Analysis

Environmentally Benign
Use
Reusable?

Landfill

 
    
   

Consumption

'ecycleable?

Dis osal
p No

Incinerate

Figure 1: Stages of Life Cycle Analysis[12]

Where the analysis consists of each step being monitored for

materials and energy flows is as in Figure 2.

Energy

  
  

      
  

Material Process

Modified Material
Step

  

Waste
figure 2: Micro—Analysis Boundary

Such an analysis consists of five major sections:
1. Raw Materials Acquisition
Product Manufacturing / Packaging Manufacturing
Consumer Use
Recycle, Compost
Final Disposal.

9‘99”.“

LCAs are useful for product designs. For example, in order to
reduce the volume of waste (municipal solid waste - MSW) a plan
could be designed to collect and recycle a particular product or the
product could be redesigned to biodegrade in an appropriate waste
management infrastructure. However, these changes often result
in the overall process consuming more energy or producing
wastes/ emissions in some step. The net result of the change
towards recycling or biodegradation, then, may do more harm than
good to the environment. Although recycling is thought of as the
"green thing to do", there is associated risks. The risk of recycling

needs to be examined to determine if risk reduction is also needed

for recycling. [36, page 3]. LCA helps quantify all these effects and
permits informed decision-making on the proposed change. LCAs
would also help in identifying significant areas for improvement in

energr usage and waste reduction throughout the entire process.

Franklin Associates, Ltd. claim [19] Life Cycle Analyses can provide:

1) Quantitative profiles of environmental impact - baseline
data,

2) Comprehensive report on environmental issues,

3) Analysis of alternatives for dealing with
environmental issues,

4) Comparison of alternative improvement strategies, and

5) Comprehensive analysis of product, packaging, and

processes from cradle to grave.

LCAs can also provide information to evaluate waste management
options, can help reformulate products and can provide an overall
view of the process such that environmental decisions can be made.
LCAs are sometimes viewed as one input into a broader decision
making process such as corporate decision making to [19]:

1) Support strategic planning,

2) Support corporate policy and product planning,

3) Guide public relations policy and action,

4) Identify opportunities for suppliers and customers to

reduce environmental impact.

1.2 THE NEED FOR LIFE CYCLE ANALYSES

In order to develop new technologies, modify existing ones, or
redesign products to meet the new environmental criteria, Life Cycle
Analysis (LCA) or Cradle to Grave Analysis of the technology has to
be performed. In today's rush by industry to make changes in
product design that would enable their products to meet
requirements for their ultimate disposability or recyclability, LCA
assumes great importance.

Environmental product planning and design have become a
necessity. All processes produce wastes, thus, the characteristics
of the wastes and their impacts on the environment must be one of
the basis to determine alternate designs of products and processes.
A recent EPA report Characterization of Municipal Solid Waste in
the United States[36], presents EPA's stated goal of managing 25%
of America's municipal solid waste through source reduction and
recycling by 1992 (now updated to 1995). Composting will play a
key role in attaining this goal.

1992 Goals

    

Source Reduction
Reuse of products

Recycling

       
    

(11%)

Composting

———————— Landfilling
(80%)

Figure 3: 1 992 EPA Goals Extended to 1 995

With the majority of a landfill occupied by materials that should not
be there (40% paper, for example), planning is necessary on the
front-end design of products to keep them out of the landfill. This
is especially true in the paper industry, where planning today will
be essential to increasing domestic paper recycling in the mid
1990's.

Another EPA publication The Solid Waste Dilemma: An Agenda
for Action[26] has hit upon the need for LCAs.

As a nation, we generated about 160 million tons of solid
waste last year; by the year 2000, we are projected to
generate 190 million tons. This deluge of garbage is
growing steadily and we must find ways to manage it
safely and effectively. Eighty percent of garbage is
landfilled.

Until recently, little or no thought had been taken of the final
disposability of the product during the product design process.
Industry is starting to respond to environmental and economic
pressures. Manufacturers of products have had, as of yet, no direct
incentive to design products for waste management because they
usually are not directly responsible for the ultimate costs of waste
managementl36]. The way business is conducted is starting to
change. With that change comes the need for LCAs. Soon, every
manufacturer wiH need to do LCAs of their products. If not due to

governmental laws then due to economic pressure. The diaper

industry has seen good examples of the latter. If an opponent does
an LCA and claims their product superior, the only way to disclaim

the results is to conduct an LCA to try to repudiate the first results.

1.2.1 Legislation Promoting Life Cycle Analysis

Governments are moving towards promoting the concept and
implementation of LCAs. The ISO 9000 series, which promotes
environmental decision-making, is being accepted internationally.
Canada is developing a standard specifically for Life Cycle Analyses.
And an example of the US trend is seen from CONEG (Coalition of
Northeastern Governors) Source Reduction Task Force, which has

recently published this "Model Legislation"[24]:

It is the intent of this Act to: 1) assist the state in its
goal of promoting effective and efficient solid waste
reduction, reuse, and recycling programs that maximize
public and private sector cooperation; 2) encourage the
use of packages and packaging components that are
source reduced, reusable, made of recycled materials or
are recyclable while also minimizing the potential
dislocation of manufacturing, wholesale and retail
companies engaged in or dependent on the packaging
industry that distribute their products in (insert state
name); and 3) control the extent to which packages and
packaging components contiibute to the decline in local
landfill disposal capacity.

The EPA will foster workshops for manufacturers and educators to

promote the design of products and packaging for effective waste

management, and will identify economic, regulatory and possibly
legislative incentives for decreasing the volume and toxicity of
waste. The EPA will also take steps to facilitate Federal
procurement of products with source reduction attributes.
Industry should conduct waste audits, and determine ways to

decrease the volume and toxicity of materials used in

manufacturing[26] .

1.3 DIFFICULTIES WITH LIFE CYCLE ANALYSES

A proposed LCA system will have to face and overcome a number of
challenges in the domain of product development/assessment. A

number of problems exist in the application of LCAs that have

prevented universal acceptance and application.

1.3.1 Problems Associated with LCAs

The problems generally associated with LCAs are 1) Data
Limitation, 2) Assumptions, 3) Lack of a Standard, 4) Cost, 5) Lack

of a Common Currency, 6) Boundary Definitions, 7) Weighting
Factors, and 8) Aggregation.

1) Data Limitation. Data are not always available as in the case

where manufacturers have proprietary equipment or processes.

10

This information is essential for the most correct and up-to-date
solutions. The problem of data limitation has been noted in a
number of articles [10,32 e.g.]. The lack of manufacturers' input
has been a major problem for current LCAs due to the
manufacturers' fear of divulging proprietary information. This
problem may be remedied in one of two ways (or both). The first is
that the governments (EPA, CSA, etc.) could require disclosure of
relevant information to be used solely for the construction of the
data bases required to run an LCA The government would not
have to explicitly make such a requirement, but could, in their
usual way, make it uneconomical not to cooperate in such a
capacity. Second, perhaps in combination with the first is that
proprietary information could remain locked/ hidden in the data
base, where the providing manufacturers are not at risk. This lack
of information availability, however, remains a major stumbling
block for most analyses. The EPA is making strides towards an

accessible database needed to increase the applicability of LCAs[26].

2) Assumptions. Another problem in this domain is the
representation of the allowable assumptions. For some processes,
the quantitative data is known and attainable while for others a
best guess is as far as can be done. It is often easy to assume a
case out of reality, which makes the assessment no longer useful.
Generally, these oversimplifying assumptions have plagued LCAs in
the past, since there are many external forces acting on a products

life cycle. This can be overcome by an extensive knowledge base on

 

 

11

the processes themselves and the components involved. With this
information, a product manufacturer can interactively 'see' the
impact a design will have on the environment and can interactively

judge whether assumptions are necessary or viable.

3. Lack of a Standard. One of the greatest difficulties is the lack of
a standard for LCAs. Marketing people have used this fact to their
advantage, claiming green superiority over other products in the
same class, only to be rebuffed by another LCA determining
completely opposite results. In this sense, LCAs can currently be
manipulated for any purpose. Some feel this to be a benefit for
LCAs, but it merely serves to injure the integrity of such studies.
According to Dr. Baumgartner of the Ecological Economic Research
Institute in Germany (Institut fuer Oekologishe Wirtshaftsforshung

-IOVV)[11]:

...marketing products might become interesting exactly
because their does not exist unanimity about the quality
of results. This is the stuff of advertisement. If need
be one can always find somebody who produces an LCA
where one‘s own product comes out on top. The
freedom they now have to make these definitions allows
them to fulfill their clients' frequent expectation that an
LCA will show their product to be best.

Not only does this render LCAs untrustworthy, but it also points
out that without a standard, LCAs are just another marketing
weapon. An example of opposing results of two studies considering
the life cycle of both single-use and reusable diapers shows how the

analyses of the same products can differ (see Section 1.3.2).

12

4) Cost.[16] The workhours required to carry out an LCA currently
costs time and money. This cost has made LCAs unattractive to

some.

5) Lack of a Common Currency.[l6] According to Norman Dean,
there exists no common unit for comparing different environmental
impacts. He claims that it is difficult, for example, to compare the
impacts of a ton of carbon dioxide with a ton of benzene, much less
compare the loss of an endangered species with human exposure to
a carcinogen. But the purpose of an LCA is not to compare, but to
present. The final decision lies on the user. This difficulty is
indirectly associated with the lack of a standard problem, but deals

more with the comparability of products.

6) Boundag Definitionsl12] Altering the boundary of a problem
will change the results of a problem. The study may still be termed
an LCA (there is no standard), but may omit certain steps which
dramatically affect the outcome.

7) Weighting Factors. A very large problem also exists in weighting
factors from other media using LCA[16]. For example, should
ozone-depleting chemical release be weighted against global—

warming—contributing-gases release?

13

8) Aggregation. Both ecological and environmental assessment
instruments have to deal with the problem of aggregation, that is,
the reduction of the potentially large number of impacts to a

number that is manageable by decision-makersllO].

1.3.2 Diaper Example

Two seemingly contradictory reports on reusable vs. disposable dia-

pers were issued [9,23]. Table 1 summarizes the data from the two

reports.

Reusable Dia ers Dis osable Dia ers

     

 

Raw Materials A_DL (A ADL g.
*Renewable 0.0047 lbs 0.0046 lbs 0.254 lbs 0.216 lbs
'Non-renewable 0.038 lbs

Water Consumgtion 1405 lbs 2.3 lbs
1.69 gal 3.779 gal 0-278 gal 5986 93'
My (Btu) (Btu)
‘Renewable sources 175.2 43.8
'Non-ren. sources 752.9 230.2
Total 928.1 2,030.6 274.0 31455-5
__p___'___'_ l' 1 l' l I' I
Atmos heric Emtssrons 1 0A3 10A3 '10"3 107(3
' Particulate Matter 3 0 lbs 0.45 lbs o_035 lbs 1.23 lbs
* Nitrogen Oxides 1.8 lbs 1.32 lbs 0,071 lbs 1.18 lbs
" Sulfur Oxides 3.8 lbs 2.29 lbs 0.082 lbs 2.29 lbs
' Carbon Monoxide 0 4 lbs 0.81 lbs 0.094 lbs 2.76 lbs
' Hydrocarbons 1.2 lbs Mbs 0.080 lbs 1MB
Total 10.2 lbs 5.61 lbs 1.082 lbs 8.82 lbs

Waste Water Effluents (1003) (*10A3) (107(3) (107(3)

* Total Sus. Solids 0.152 lbs 1.796 lbs 0.082 lbs 1939 lbs

* Chemical Oxy. Dmd. 0.047 lbs 4.226 lbs .......... 1.227 lbs

' Biological Oxy. Dmd. 0,141 lbs 1.887 lbs 0.035 lbs 1-5 le

Process Solid Waste 0.0369 lbs 0.0040 lbs 0.024 lbs 0.014 lbs

Post Consumer Waste 0.0028 lbs 0.055 lbs 0.261 lbs 0.428 lbs

Table 1: Environmental Impacts of Diaper Usage (Arttuir D. Little and Carl
Lehrburger)

l4

(ADL numbers compiled by Arthur D. Little and Associates and were per
85 diaper changes but have been normalized to per single diaper use; CL
numbers are from Carl Lehrberger et. al. and were per 1000 diaper
changes, but have been normalized also).

Direct comparison of results show contradictions, even though both

research teams evaluated the same products, reusable and

disposable diapers:

l) A 300% Energy difference-
CL: single-use diapers use 70+% more energy
ADL: reusables use 230% more energy
2) An 800% Water-usage difference (per diaper change basis)-
CL: single-use diapers use greater volumes of water:
reusable, ~3.8 gal, disposable, ~6.0 gal
ADL: reusable, ~l.78 gal, disposable, ~0.27 gal
3) A 14% Materials difference

CL: reusables use 72% fewer raw materials
ADL: reusables use 86% fewer raw materials

Thus, different studies of the same subject may arrive at
contradictory conclusions as a result of varying assumptions and
boundary conditions set for LCA. The results are not always black
and whitel31]. As stated earlier, the key is in defining the system
boundary and the assumptions that go into this boundary. A
report in Germanyl32] about the same comparisons also talks of
several other reports along the same line that differ. There is thus
a need for a standard, an analyzing technique broad based enough
to allow applicability to a wide spectrum of products within their
Life Cycle boundary. The analyzing tool needs to judge the en-

15

vironmental merits/ demerits of a product or process. This tool
would be capable of drawing together reports such as those used
in the above example and provide practical, sound environmental

product and process information.

1.4 CURRENT WORK

Life Cycle Analyses have been attempted under many names, but
are essentially LCAs. Some of these include: Environmental
Audits, Cradle-to-Grave Analysis, Life Cycle Analysis, REPA
(Resource Environmental Profile Analysis), Eco-Balance, PLA
(Product Lifecycle Analysis), among others; publications range near
130 [10].

To facilitate the need for LCAs and overcome the difficulties
associated with LCAs, a standard system should be developed.
Currently, a number of attempts at such a standard exist. These
next two sections touch upon groups carrying out research in this
area. Some believe that LCAs should be left mainly to human
judgment or subjective evaluation, citing that even a cursory review
of the existing LCA documents show them to contain many
subjective judgments[16]. Until recently, LCAs have been used to
study and as of yet not to predict the environmental impact of

products to aid in design decision-making.

16

Several EPA reports talk of using "integrated waste

management"[3,26].

In this approach, systems are designed so that some or
all of the four waste management options (source
reduction, recycling, combustion and landfills) are used
as a complement to one another to safely and efficiently
manage municipal solid waste. Recycling (including
composting) is the preferred waste management option
to further reduce potential risks to human health and
the environment, divert waste from landfills and
combustors, conserve energy, and slow the depletion of
nonrenewable natural resources.

Franklin Associates, Ltd., The European Community, Canada,
Batelle Institute, Ecological Economics Research Institute, Sweden,
Recycling Algorithm for the Development of Strategy, are some of

the groups helping to develop an LCA characterization and

standardization.

1.4.1 Franklin Associates, Ltd.

Franklin Associates, Ltd. have done extensive research into
developing the protocol for Life Cycle Analysesll7,18,19,2l,24].
William Franklin was one of the pioneers of the subject back in
the early 70's while working for Midwest Research Institute. He
then started working on Resource and Environmental Profile

Analysis (REPA). Currently, Franklin Associates, Ltd., are known

17

for their work in LCAs. A detailed description of their approach

can be found in their reports cited earlier.

1.4.2 Canadian Standards Association

Canada is currently working towards an LCA Standard

[Environmental Life che Assessment, CSA-Z760, Draft # 5B].

CSA views the life cycle assessment process as an iterative
process, consisting of four interrelated stages: 1) Initiation, 2)
Inventory, 3) Impact Analysis/Assessment, and 4) Improvement
Analysis. The standard CSA-Z760 is a generic life cycle model to
facilitate system formulation and boundary definition, and is
based on the technical contributions of the Society of
Environmental Toxicology and Chemistry (SETAC) and the on—
going work carried out by the US EPA.

1.4.3 RADS group

RADS[25,55,56], Recycling Algorithm for the Development of
Strategy (working with SEPA-Systems Engineering for Pollution
Avoidance). The work being done is to develop a model that will be
an input-output model similar to that used for assessing the
industrial output of the US industrial sector including such things

as: inventory or landfill accumulation options, economic and

18

logistical subroutines, and suggestions on future efforts. This
work is to create a more or less materials and energy flow diagram
for all products. Although, this does not deal directly with LCAs,

the work is related and useful.

1.4.4 European Community

The European Community is working out an Eco labeling
standardization. A part of the criteria will be a Life Cycle Analysis.
They are working to develop the standards for such an analysis.
One benefit of having a single standard is that companies will not
be afraid of being audited continually by different groups doing
similar but different tests. Hence, the European Community as
well as the European Chemical industry is promoting the concept
of ISO 9000 certification of health, safety, and environmental
(HSE) management systems, of which Life Cycle Analysis (Eco-

Balances) plays a role.

In the near future, an LCA audit will be required for every product
and process, if not due to government regulation then due to the
markets (consumers) demand. Life Cycle Analysis can provide the

essential information necessary to improve each product and

19

process environmentally and economically. However, due to the
difficulties listed in section 1.3, LCAs should not be used to
compare the environmental merits or demerits of competing
products, but to present information regarding the impacts of the
individual products.

The characteristics of an Expert System, which are explained in
chapter 3, can be exploited to provide a tool to perform such

analyses.

CHAPTER 2: PAPER LIFE CYCLE ANALYSIS

INTRODUCTION

The purpose of this chapter is to gain understanding of a Life Cycle
Analysis by applying a sample analysis to paper, and to have a
sample product line with which to run the Expert System evalu-
ation. Paper was chosen due to its high environmental profile.
Municipal Solid Waste consists of more than 40% pa-
per/paperboard products. These products are presented in this
chapter along with the energy requirements and releases for their
manufacturing. The difficulties with such LCAs are presented in

Section 1.3.

20

21

2. l PAPER PRODUCTS

Paper can be broken down into a myriad of sub—categories[54]
1) Printing and Writing Papers
NewsPapers
Groundwood Paper
Coated Printing and Converting Paper
Book Paper
Bleached Paper
Writing and Related Papers
2) Packaging and industrial converting paper
Wrapping Paper
Shipping Paper
Bag Sack Paper
Other Converting Paper
Glassine Greaseproof Paper
3) Tissue and other creped paper
Sanitary Paper
Tissue Paper
4) Paperboard
Linerboard
Corrugating Medium
Folding Carton
Tube, Can, and Drum Paperboard
Other Bleached Kraft
Other Bleached Paperboard
Container Chip Fill Paperboard
Combination Paperboard
Combination N onbendable Paperboard
Special Combination Paperboard
5) Wet Machine board

6) Construction paper and board

22

2.2 PAPER LIFE CYCLE

The manufacture of paper in its life cycle runs along the somewhat
generic life cycle from raw materials acquisition to disposal is

illustrated in Figure 3:

Raw Materials §Ha~°flin9

Shipping

Raw Materials Storage
Processes

Environmentally Benign

   
     
  
  

  
  
 

Recycling Composting

Process

   

  

Production Process

Product Analysis
Packaging Manufacture
Yes

lncinerate

Figure 4: Stages of Paper Life Cycle Analysis

      
     
 

   

Consumption
Use

  

Yes
Recycleable?

Dis osal
p No

 

Paper is first harvested, shipped to production where any of the
various products listed in Sect. 2.1 are produced. The packaging is
made, if necessary (whether purchased or manufactured on site).
The final product is sent to the market/ consumption stage where it
serves its useful purpose. It is either reused, recycled or thrown
away. Depending on which choice has been made, the paper will
either be transported to the landfills, whether via incinerators or

directly, or be shipped to be recycling facility to start back into some

23

products raw materials. Section 2.6 shows the recycling rates for

1990, and the utilization for each waste stream.

2.3 PAPER RAW MATERIALS

The raw materials acquisition is mainly derived from the forest
trade along with recycled paper and other wastes from the lumber

industry:

41.4% Round Logs from private forests.
29.0% Waste Paper
26.0% Chips, sawdust, and other waste products

from lumber operations including public
and private lands.
3.3%* Round logs from public forests.
0.3% Misc., cotton. linens, etc.
Total 100%
* Estimated between 2% and 3.3%

Table 2: Sources of Paper Pulp for US Paper Mills. 1992 estimates[55]

2.4 PAPER PRODUCTION

The total paper manufactured for 1990 was 78.78 million tons.
(39.36 million tons paper, 39.42 million tons paperboard) [63].

The manufacture of paper products is essentially generic for most of
the paper grades, but differs nearing the final production stage in

the web modification and converting.

24

The five steps to make paper as shown in Figure 4 are:
l) Pulping
2) Stock Preparation
3) Paper Making
4) Web Modification
6) Converting

    

Digester

PaperMaklng

Web Modification 1 |
and/or Converting . ________________________

figure 5: Paper Production Layout (For process descriptions see [38,54])

25

2.4. 1 Energy Use

The Energy utilized by the Paper as a whole consists of 44% from
outside sources and 56% from self generated sources. The Energy

use is for the entire Paper industry.

 

Source Est. Use Units Billion Btu's % of Total
Purch. Electricity 46,6730 MMKWH 158,772.5 6.5%
Purch. Steam l7,490.l MM lbs 19,0854 0.8%
Coal 13,9694 M tons 338,051.6 13.7%
Residual Fuel Oil 23,9638 M 42 gal 150,854.7 6.1%
Distillate Fuel Oil 1427.1 M 42 gal 6898.2 0.3%
Liquid Propane Gas 34,5756 M gal 3350.1 0.1%
Natural Gas 393.4196 MMCF 401,682.9 16.3%
Other Purchased Energy 5413.9 0.2%
Energl Sold (-33,581.1)

Total Purchased Fossil Fuel and Energy 1,050,528.2 44.0%
Hogged Fuel 29,7022 M tons 2247,5783 10.1%
Bark(50% moist) 15,0657 M tons 130.6944 5.3%
Spent Liquor 77,977 .3 M tons 967.0365 39.4%
Hydronr 3938.3 MMKWH 17,240.0 0.7%
Other Self-Generated Energy 12,9840 0.5%

Total Self Generated & Waste Fuels 1 375 533.2 56.0%
Total Energy 2,476,06l.4 100.0%

Table 3: US Pulp, Paper, and Paperboard Industry Estimated Fuel and
Energy Use [37]

2.4.2 Releases for Paper

The total releases for the Paper manufacturing industry are pre-
sented in Table 4.

26

Air Releases: 194,029,676.9 lbs
Surface Water Releases: 42,164,0208 lbs
Public Sewage Releases: 45,985,7226 lbs

Off—Site Releases: 21,144,661.3 lbs
Land Releases: 9,930,159.4 lbs
Underground Releases: 0 lbs

Table 4: Total Media Releases for Paper Industryl49]

The specific chemical releases are listed in Table 5.

Carcinogenic chemicals:

Dichloromethane 7 1 7,000 lbs
Styrene 41 200 lbs
Tetrachloroethylene 204,000 lbs
Formaldehyde l , 1 30,000 lbs
Chloroforrn 20,500,000 lbs
Chromium 36,189 lbs
Asbestos (friable) 1,425,729 lbs

Isopropyl alcohol (manufacturing)473,000 lbs
polychlorinated biphenyls (PCBs) 764,000 lbs

Di- (2-ethylhexyl) phthalate 26,000 lbs
Ethylene Oxide 31,400 lbs
Other Chemicals:
Methanol 1 18,334,484 lbs
Toluene 36,055,332 lbs
Hydrochloric Acid 27,038,003 lbs
Sulfuric Acid 23,312,163 lbs
Acetone 17,751,587 lbs
Ammonium sulfate (solution) 13,064,014 lbs
Chlorine 10,024,5 14 lbs
Methyl ethyl ketone 8,244,205 lbs
Chlorine dioxide 6,137,251 lbs

Table 5: Specific Chemical Releases for Paper Industryl49]

27

2.4.3 Releases for Printing

The total releases for the printing industry are listed in table 6.

Air Releases: 55,349,146.0 lbs
Off-Site Releases: 4,819,06l.6 lbs
Public Sewage Releases: 749.3610 lbs
Surface Water Releases: 6092.4 lbs
Land Releases: 0 lbs
Underground Releases: 0 lbs

Table 6:- Total Media Releases for Printing [49]

The specific chemical releases for the printing indusz are listed in
Table 7.

Carcinogenic Releases:

Dichloromethane 352,000 lbs
Styrene 28,800 lbs
Tetrachloroethylene 1 95,380 lbs
Lead 5,960 lbs
Chromium 2,680 lbs

Isopropyl alcohol (manufacturing) 1 ,24 1,000 lbs
polychlorinated biphenyls (PCBs) 24,100 lbs

Di-(2-ethylhexyl) phthalate 4,080 lbs
Other Chemicals:
Toluene 42,789,389 lbs
1,1,1-Trichloroethane 4,522,055 lbs
Methyl ethyl ketone 3,906,677 lbs
Xylene (mixed isomers) 2,004,817 lbs
Glycol ethers 1,893,662 lbs
Acetone 863,576 lbs
Methanol 822,857 lbs
Dichloromethane 351,262 lbs
Methyl isobutyl ketone 328,537 lbs

Table 7: Specific Chemical Releases for Printing Industryl49]

28

2.5 Paper Waste Management

The main waste management practice today is set for Landfill, In-

cineration, Composting, and Recycling.

Constituent %l-l2 %Volitile Matter %Fixed arbon %Ash HHV

Newsprint 23.0 68.5 6.93 1.54 6565
Corrugated 17.0 71.4 9.58 1.95 6594
Other Paper 30.6 ' 58.8 6.80 3.8 5430

Chemical Analysis:

0 H2 00C %H %N 000 %S 0001 0 o h
Newsprint 23.0 37.0 3.53/6.11 <.10 348/552 0.16 - 1.54
Corrugated 17.1 39.0 4.98/6.90 <.10 36.7/51.9 0.25 - 1.95

OtherPaper 30.6 30.7 4.26/7.69 <.10 30.4/57.5 0.12 0.12 3.80

Table 8: Paper As Received Basis by Weight [28]

Source reduction and recycling are the preferred options for manag—
ing solid waste. Combustion and landfilling should be used only

when the preferred options are unavailable or insufficient.
[3,26,56,57]

2.5. 1 Landfill

A landfill essentially buries waste for posterity sake[62]. The landfill

has been a last ditch effort to entomb waste for generations to

come. The landfill is one of the greatest indicators of the need for

29

LCA's. Americans have been too generous in our waste production,
and have been under zealous in waste recovery.

The landfill requires the least amount of preparatory work. Usually,
waste will enter as is. This option should not be available for paper.
Currently, of the 175 million tons of waste annually [2], 30-40% is
paper [36].

2.5.2 Incineration

Incineration has gained importance due to the lack of or diminish—
ing landfill space. Although burning waste can provide energy, the
greatest benefit derived from incineration is the waste volume re-
duction. But along with these benefits came many negative aspects.
In fact, an earlier study of Incineration Units by Michael Braungart
(EPEA, FRG) stated that the construction of an Incineration plant
produced more waste by volume than it would ever reduce over its
useful lifetime (referred to in [44], unlike in Braungart's report, this
thesis does not consider the environmental impacts of the capital
equipment manufacture, though this may gain in importance later

in the development of Life Cycle Analysis).

The characteristics of the Refuse Derived Fuel (RDF) are as fol-
lows[28]:

Higher Heating value 6100 BTU / lb (14,152 kJ / kg)
Ash content 10— 1 2%
Moisture Content 18-22%

Particle Size 91% less than 3/4 in (1.90m)

30

Percent RDF produced per ton of MSW 54%
If MSW enters an Incinerator without preparation, the energy value
is much less. Usually, to increase the fuel value, the waste is sepa-
rated, but this adds processing steps and decreases the overall en-

ergy benefit from incineration.

2.5.3 Recycling

The paper industry has always been known for its recycling prac-
tice. Now, American paper companies have set a goal of 40% paper
recovery by 1995 [40]. Recycling waste paper is one of the oldest,
and fastest growing segments of the recycling industry. Recycling
one ton of paper conserves approximately 3 cubic yards of landfill
space [54]. But, in order to understand the impacts that paper re-
cycling has on the environment, we must be aware of the effects,
both positive and negative of the paper recycling process.

When recycled paper leaves the recycling stage it enters the raw
materials stage of either the same or a new product. The main im-
pact of recycling lies in the acquisition of raw material where recy-
cled fiber displaces raw fiber. Figures 6-10 illustrate where the

recycled material is utilized.

2.6.3.1 Recycled Fibergrade

Each wastepaper grade has unique physical and chemical proper-

ties and contaminantsl44]. In 1990, 29.28 Million tons of paper

31

was recycled. Totals for the various grades of recyclable paper
are[42]:

Waste Paper Fibergrade Amount Recycled

High Grade: 2,925,500 tons
Mixed Paper: 3,510,600 tons
Pulp Substitutes: 3,218,050 tons
Old NewsPapers: 6,143,550 tons

Old Corrugated Cardboard: 13,460,000 tons

This recycled fiber is utilized in various ways. Figures 5-9 show

where each waste fiber is used.

 
     
   

Expon
965,400 tons
Tissue
,404,200 ton
Y

 

   
  

Rec cled
Boxboard

146 300 tons

High Grade De-inking
Paper

2,925,500 tons

   
    
    
   

Container
Board
58,500 tons

  

Print/Write
Paper
351 100 tons

Figure 6: Recycled High Grade De-inking Paper Flow

32

 
 

Export
1,088,300 ton
Tissue
351,100 tons
NewsPaper
105,300 tons

Rec cled
oxboard

1,474,000 ton

 
   

 
  
     
    
    
   

Mixed Paper
3,51 0,600 tons

  

Construction
Board
491 500 tons

figure 7: Recycled Mixed Paper Flow Diagram

  

Export
406,000 tons

     
     
 
 
   
 
      
 
 
      
      
   
     

  

Tissue
643,600 tons

Print/Write
Paper
1,029,800 ton ~

Recycled
Boxboard
708,000 tons

Construction

Board
32,200 tons

  
  

  

Pulp Substitutes
3,218,000 tons

  

  

Container
Board
257,400 tons

isc. Packaging
Papers
128,700 tons

Figure 8: Recycled Pulp Substitutes use

 

33

 
 

Import
23,000 tons
Expon
1,290,000 ton

Construction
Board
245,700 tons

   

     
    
  

  
     
 

Recycled
Boxboard

1,413,000 ton

Tissue
491,500 tons
Non-Paper

End Uses
921,500 tons

NewsPapers
1,781,600 tons

Figure 9: Where Old Newspapers are Used [also 41]

Recycled NewsPapers
6,143,550 tons

 
      

  

Expon
2,691,000 ton

Construction board
and Misc.
403 700 tons

  
     
 
   
   
 
    
   

  

Container Board

13'460’000 tons Container Board

6,863,000 tons

  

Recycled
Boxboard

3,500,000 tons
Figure 10: Recycled Old Corrugated Containers

34

2.5.3.2 De-inking

De-inking processes clean recycled paper. Simple repulping allows
for a higher yield, but the cleaning process is not complete enough

to allow the pulp to be used in many high quality papers.

Wastepaper
receiving

     
   
 
  
    
 
    
    

  

Inspection
and Grading

To bleaching

Cleaning and
Screening

Relects
Pallets .
Trash
Contaminated Talllngs
Bales screen rejects
W' Plastic , '
"8. Paper Clips Rejects Sludge
Plastic
Wood Rubber Bands Ink Fines
S les F'be' ‘ Coatin s
Wet-strength paper tap . g
M a] Glue Fillers
9 Dirt lnk
Rags Latex Latex

Figure 1 1: Sources of Solid Waste from De-inkirtg (adapted from 45).

The Solid Waste generated at a De-inking plant [45]:

* Rejected Wastepaper. Unusable paper or con-
taminated paper.

* Screen Tailings. (rejects) can consist of fiber and
water, plastic, glue, staples, rubber bands, pa-
per clips, and other debris.

* Sludge. The de-inking sludge is generated in
large quantities. It consists of the inks, fillers,

35

toners, coatings, adhesives and other materials
washed out of the paper during the cleaning
process. Sludge from newspaper de-inking
usually consists of fiber and ink and is lower in
ash since newsprint grades little fillers and clay
(for specific chemical releases see [45,55,671

* Misc. Solid Wastes. This is the solid waste gen-
erated in wastepaper receiving and handling:

baling wire, pallets and boxes.

2.5.3.3 Energy Savings through Recycling

Many paper processes realize significant energy savings when waste
paper is used. Bark, spent liquor, and other by-products, however,
are consumed for fuel at many virgin fiber mills. For this reason,
some recycling processes actually consume more energy from out-

side sources[55].

End Product Virgin Fiber % Change in Energy use

Tissue paper 0% - 57%
Printing & Writing 16% — 35.9%
Newsprint 0% — 21.6%
Packaging Paper 70% - 7.6%
Corrugated Board 0% ~ 2.5%
Construction Board 65% + 1.7%
Box Board 0% + 39.6%
Liner Board 75% + 150.9%

Table 8: Change in energy consumption due to recycling

 

36

Source: US. EPA, Office of Technology Assessment. 1983 Study. Fig-
ures include only energy demand from outside sources. Energy supplied
from by-products are not included. These figures do not include a full
life cycle analysis. This only entails production Energy savingsl55].

2.5.4 Composting

Composting is one of the oldest solid waste disposal methods
known to man which converts solid organic material into a humus
like mixture. This compost: (1) has a lower bulk volume than the
original waste, (2) is stable and (3) has the potential of being recy-
cled for a multitude of uses without destruction of its innate high
energy value[53].

In the basic process, organisms break down the available biode-
gradable organics into simpler, more stable compounds and carbon
dioxide. The organisms self-generate heat, which has been deter-
mined-to kill possible pathogens [50]. Since during their processing
period the anaerobes generate offensive odors that are difficult to
control in a composter, the normal practice is to use aerobic

composting. (An overview of considerations can be found in [8])

The difficulty associated with the de-inking process in not consid-
ered a problem for Composting [1]. Namely, the same chemicals are
present but are not concentrated as is the case for recycling during
the de-inking stage. Paper should not be composted alone, how-
ever, due to its lack of Nitrogen content, which is necessary for

composting. Generally, paper is co-composted with other wastes;

37

paper can act as a bulking agent, as a dryer, or as a carbon-content

contributor, the other wastes can supply the nitrogen content
[53,47,11.

Composting Process description[28,8]:

(1) preparation of the feedstock,
sorting of organic and inorganic fractions
and adjusting the feedstock if necessary, the
carbon/nitrogen ratio should be adjusted to
about 30:1.

(2) decomposition,
Windrow decomposition: prepared solid
wastes are placed in windrows in an open
field. The windrows are turned by mechani-
cal means to insure uniform reaction, and
aid in aeration for a period of about 5 weeks.
(There are also controlled environment closed
composters).

(3) curing, (stabilization of the material)
2-4 weeks to 2-3 months

and (4) finishing or product preparation.
may include screening to recover the bulking
agent and fine grinding to remove oversize
material, blending with various additives,
granulation, bagging, storage, and shipping.

 

 

 

CHAPTER 3: THE EXPERT SYSTEM - HIERARCHICAL
CLASSIFICATION

INTRODUCTION

The use of Hierarchical Classification as the Expert System to
accomplish LCAs has been studied. The aim of this work is to
better understand the applicability of using a hierarchical classifier
to carry out life cycle analyses. This was accomplished by
attempting to classify the life cycle of paper products. This chapter
presents the necessary background of Hierarchical Classification
and an introduction to the breakdown of the paper scenario, which

chapter 4 will present in more detail.

38

39

3. 1 EXPERT SYSTEMS

The Expert Systems (ES) area centers on the construction of
problem solving in such areas as process design, fabrication
control, plant trouble shooting, etc. In the engineering domains,
these systems solve problems that would otherwise be solved by
human 'experts' (hence, "Expert Systems") by drawing upon past
experiences and 'back of the envelope calculations.’ Thus, these 0
systems must contain expertise such that they can mimic the

performance, skill and robustness of a human expert [74].

For engineering problems, 'exact' answers are often unattainable,
and when found are frequently 'over exact.’ ES try to work their
way around this problem by using heuristic 'rules of thumb' which
an expert would have applied in the same situation were exact
data unattainable, and thus provide a sufficient answer to tough
problems. This is especially useful where exact answers are
difficult or impossible to obtain, such as in LCA, due to various

factors such as random inputs.

The development and use of an Expert System is shown in Figure

12.

Extends
and tests

Toolbuilder I
EXpe" i EXPERT
System . SYSTEM
Building I .

Tool

    
     

Staff
Figure 12: The development and Use of an Expert System (Adapted from
[77]).

The Expert System has four main users. As a system, the most
visible user is the End-User (hence the necessity of an explanatory
interface). The system must not only be equipped with this usable
interface, but should also be able to provide the reasoning behind a
particular solution to demonstrate the basis for a decision. This is
one area that separates ES from conventional algorithms. Expert
Systems contain the ability to explain how an answer was deduced.
(As of yet, the Hierarchical Classifier cannot do this without going
into the inner workings of the system). The Clerical Staff must also
be able to use the system to input large amounts of data. If
information (lmowledge) is required to be input into the system, it
must be provided in a 'computer-understandable' manner. They
must update or complete the knowledge base in an accurate
manner (this input stage is sometimes unnecessary due to the vast
databases available, but whether working on the database or
directly on the knowledge base, the clerical staff must provide

correct information). Once the Building tool is available, the

41

knowledge engineer works most interactively with the ES. The
Expert System is built through interviews with experts in the field
of study, and is extended and tested. Feedback is given to the
toolbuilder to aid in the further development of the system.

The basic elements of an expert system as described by Rich [73]

are:

1) these systems derive their power from a
great deal of domain specific knowledge,
rather than a single powerful technique

2) in successful systems, the required
knowledge is about a particular area and
is well defined.

3) usually derived with the aid of one or
more experts.

4) the transfer of knowledge takes place
gradually through many interactions
between the expert and the system.

5) the amount of knowledge required
depends on the task.

Expert Systems are used for problem solving, which is a search
through potential solutions, guided by heuristic rules. These
should point to a destination without exhausting every possible
avenue by eliminating the study of undesirable or unnecessary

paths.

42

For the purposes here, the Expert System Building tool is the
hierarchical classifier which has been built by Dr. Sticklen of the
Computer Science AI/KBS department at MSU. This work has
been focused on the knowledge Engineering, Expert System,
Domain Expert loop, with a feedback to the Toolbuilder as to

useability and wishes. (Enclosed in dotted box in Figure 12).

3.2 HIERARCHICAL CLASSIFICATION

Life Cycle Analysis (LCA) can be easily broken into a hierarchy,
wherewith an Expert System could be applied. Expert Systems are
particularly useful in complicated but routine problems.
Hierarchical classification is a problem-solving technique that
efficiently compares a set of pre-enumerated categories with a
particular situation to find those categories that 'best' apply. As
part of the conceptual LCA process, hierarchical classification can
be used as a heuristic filter, pruning categories which are not
appropriate for achieving the goals of the problem (determining the
processes involved with the problem and the associated
environmental loading). This is essential in accomplishing Life
Cycle Analyses. As previously mentioned, one of the largest
difficulties with LCAs is the vast number of directions in which an
LCA can go. A hierarchical classifier can help reduce the number of

paths one must investigate.

43

In hierarchical classification, the categories are organized into a
hierarchy in which the children (the connected nodes at the next
level down the hierarchy) represent a sub-category of the node and
the parent (the connected node one level up the hierarchy)
represents a super-category of the node (Figure 13 illustrates an

abbreviated and greatly simplified hierarchy for a products life

cycle) .

G nProd ct

       

omponents Packaging

- recess/RMAc<\quistion VET; Op7rati< Recycle Disposal

Purchase Extraction Processing Waste Use Reuse
Handling

Processes Compost Recycle

Mining Farming _
Reaction

Bioremediation Landfill Incineration

Figure 13: Example of a classification scheme for a products life cycle
analysis.

As is shown in Figure 13, a generic product (GenProduct) can be
made up of five (any number, say, cap, bottle, label etc.) separate
components, A, B, C, D, and an associated (usually) packaging.
Each one of these components has children nodes for processes:
RM(Raw Materials) Acquisition, Manufacturing, Operation,
Recycling, Disposal (Only component B is expanded in Figure 13 for

the sake of clarity). Some of these first generation nodes will also

have children: i.e. the Recycling node has both Compost and
Recycle as children. As can be seen, these categories become more
specific as the hierarchy is traversed from the top towards the

lower, more knowledge-specific nodes.

Each node in the hierarchy is responsible for determining the
process steps involved for its category relating to the current
problem. For example, the RMacquistion node in Figure 13 is
responsible for determining the processes involved for acquiring raw
materials for manufacture of component B of GenProduct. Each
process node will likewise be queried for information about

environmental impacts, as shown in Figure 14.

Transportation Energy

\/

Materials —PProcess —P Materials

1

Waste
/1\
Air Land Water

Figure 14: Filter for Process Steps

Each of these subprocesses or Flows have their own impacts which
are imbedded in their respective filter. Such as Transportation and

Energy:

45

       
   

Transportation Energy
#Btu's
#Miles
0” Coal
Truck Plane Camel Nuclear Natural Gas
Ship Train Pneumatic
Hydroelectric Waste to Energy

Figure 1 5: Transportation and Energy Filter

Each node can be thought of as a specialist in the field which the
node represents (see [70]). Higher level nodes act as managers to
the lower-level specialists. The RMacquistion knows nothing about
farming or mining but can refine itself and essentially 'get' the
information from the Extraction specialist. Not all nodes, however,
are relevant to all components. In the Processing, Reaction
Injection Molding(RIM), for instance, may be important for plastics,
but will not be needed for cotton products. The general idea is that
each node requires a list of features that are important in
determining whether the category it represents is relevant to the
present system or not, and a list of patterns that map combinations
of features to confidence values. This essentially classifies the
product with each node. For a given product, confidence values will
depend on the physical properties of the product, the technologies
normally used and other practical considerations. The output value
will be a numerical value within some specified interval, for
example, -3 to 3. Positive values would indicate confidence in the
categories' applicability while values less than or equal to zero

indicate a low level of confidence.

46

Consider the manufacture node in Figure 13. As previously noted,
this node determines the process steps for product B for its
manufacturing process. Each node includes a list of patterns that
map to some confidence value. At the deepest level of the hierarchy
specific confidence values are chosen and sent back up the

hierarchy for the department's manager to report to the top node.

Before a manger sends the exploration down to its specialists it will
use a technique called "establish-refine." A node will establish by
applying its local pattern-match knowledge to determine that the
confidence value is 'good-enough.’ That is, in dealing with plastics,
one can be fairly confident of the use of a RIM machine, so a
confidence value is given of 2. Once established, a node will
attempt to refine itself by asking its more detailed sub-categories
(its direct children) to establish themselves. If later a match is
found to conflict with the value of 2, that value will then be removed
by the 'advice' of the specialists. In this way the most detailed
categorical description can be determined. (In the HC used for this
research, a value of 2 or greater Establishes, 1 or zero Suspends,

any minus value is Rejected).

Pruning is a large advantage gained by the establish~refine
technique. This prevents the system from wasting time. Whether
or not a given node will establish depends upon the pattem-match
knowledge encoded in the node. A category that rules out or rejects

with a low degree of confidence does not ask its daughter nodes to

47

establish, thus pruning the search space or cutting down the
search required. In order to solve tough problems efficiently, it is
important to eliminate some of the details of the problem until a
solution that addresses the main issues is found. Then an attempt
can be made to fill in the appropriate details. This will become

clearer in the paper products example.

3.3 Rationale for Use of Expert Systems

Expert Systems can offer depth and completeness to the LCAs. The
general idea of the Expert System is that it can attempt to mimic
the problem solving behavior of an expert. Since LCAs are based
upon the availability of knowledge from many various experts, these
systems can act as those experts. Research in LCAs using
conventional algorithms exist [76,19]. A nice feature of using an
ES, however, is the ability to more or less 'discuss' results with the
user. The fact of being an Expert System does not preclude the
exclusion of conventional algorithms, but rather adds a dimension
to the software capability. An ES can utilize such algorithms if

necessary and 'knows' when this necessity arises.

LCAs lead to complicated system of solutions, in determining all
environmental impacts of a product and its processes. However, by
breaking the solution search into sub-problems, such as the five
parts of a products life cycle, one can then deal with each of these

problems separately and more handily.

48

3.4 THE COMPUTER SYSTEM EXPERINIENTAL SET-UP

The Knowledge Based System Building Tool, a hierarchical
classifier, has been constructed by Dr. Jon Sticklen at Michigan
State University. The classifier is built using Egg and SmallTalk
version 4.0, object oriented programming. This section describes

somewhat the set-up and the use of the system.

3.4.1 The Hierarchical Classifier

The operating system used was the Sun/OS, with the host
"Pleiades" in the AI/KBS lab at Michigan State University. The
basic building tool uses ’I_‘igE/SmallTalk to set up interfacing
windows to facilitate the construction of the Expert System.
Appendix A shows the working screens on X-windows, taken by

snapshots. The individual windows will be discussed shortly.

3.4.2 Hierarchical Classifier Individual Window Descriptions

To work with the Hierarchical Classifier, there are essentially nine

windows of importance: 1) The Launcher, 2) Classifier Window, 3)
DataBase Window, 4) VariableDefine Window, 5) SubSpecialist-

49

Relation Window, 6) TableMatcher Window, 7) Table Editor, 8)
Graphing Tree, and 9) The System Transcript.

El Launcher

Browsers

>
Utilities >
Changes >
>

>

   

  
 
 

 

     
  
 

Special
Tigre (tm)
Quit

  

figure 16: Launcher

Normally, when using Egg/SmallTalk or programming in
SmallTalk the Launcher is essential. For the use here, this was
only needed to save the image. The save provided with the system
proved to have difficulties, and hence to avoid problems the

Launcher was used.

cases
new case save case
p43 V
inspect case BrowseDB

classifier

 

Figure 1 7: Classifier Window

50

This is the main classifier Window. All other windows of the
classifier are direct descendants. The database and SubSpecialists
windows come directly from here; wherefrom all other sub-windows

originate. A new case name must be entered for each new

classification attempt.

-

loaded wse:

save case P31 V

load case current time:

GtOrdinalOrderV .

GtStringVar II

It
Inspect DB Destroy DB

figure 18: DataBase Window

 

The database stores all of the current values of the variables within
the classifier. The variable type is defined here (Numerical, string,
Yes/No, etc.). The database allows for the definition of the process
variables. The choice define variable from the operate window

brings up the Variable Define Window.

51

    

E Var Define Window

variable name: percRecyc

W1... pressrem
.lear lower llmlt 1E0 afterentry

of the percent recycled if it
is known.

 
     
   

figure 19: DataBase VariableDefine Window

The question which will prompt the user for a value (answer) is
input (an example: What percent of the product is recycled?) All of
the variables in the tables must be defined in order to be used by

the system.

El Pap.HC:SubSpecialistRelation.Pap.HC

 

figure 20: SubSpecialistRelation Window

52

This window is a sub-window of the classifier window and is a
compilation of all of the nodes Within the classifier. Highlighting a
node (a name in the window drawn above) will call up the daughter
nodes in the right-adjacent window. This highlighting can be done
down to the lowest nodes. Highlighting a node also makes that
node current such that its TableMatcher may be called upon from
its operate menu (activate by selecting the menu located 'under' the
arrow just above it). To initially run the classifier once a new case
name has been chosen, the run-establish/refine option must be
chosen in the operate menu while the highest level node is selected,

in this case it is the Pap.HC.Top (shown).

 

figure 21: TableMatcher Window

The TableMatcher is a sub—window of the SubSpecialist window.
When the classifier tries to establish at a node, it calls on its
SubMatcher, or knowledge holder for that node. This Matcher has

the information needed, or knows how to ask the user for the

53

necessary information to determine whether or not the node will

establish or not.

l§l Table Editor

isSanitorTlssue
TissuePaper
SanitPaper

 

figure 22: Table Editor

The Table is a sub-window of the SubMatcher. These tables contain
the pattern—match knowledge for the system. This contains the
rules of the system as a more or less weighted if then ruling. For
example, in the above table, the first line says: "if {isSanitor’I‘issue}
= 'Yes' and TissuePaper has a confidence of 3, then the confidence
is 3 (very high)" Since TissuePaper has a confidence value attached
to it, the table has a sub-table which must be queried to determine
its value and then return a confidence of 3 in order for this table's
first line to return with a value of three. The order of the table is
important. The sub-table will not be called unless it is either first
or the previous rules have proven correct. Here, the classifier will
enter the node on the SubMatcher and ask it to refine. The node
checks its matching table to try to return a confidence value. When
the problem solver enters the table, it will start with the first line

54

and solve from left to right. This is important in some cases to
eleviate wasted search and refining. If, for example the first two
boxes were switched in the above table, the problem solver would
first try and find a confidence for TissuePaper (via its table
information). The problem solver would have wasted that time in
search if it comes back with a confidence of 3 but then finds out
that the variable isSanitor’l‘ issue is 'No'.

Most of the difficulties in use came from the Table Editor(see 3.5:

Difficulties in Use).

E] Tree Graph

 

figure 23: Graphing Capability

The graphing capability exists for both the table Matcher and the
SubSpecialist relation. (Here, Figure 22 shows the SubSpecialist
relation just down to the third level.) This window is brought up by
depressing the tree button in the bottom of the Matcher or
SubSpecialist windows (see Figures 19 and 20). This function is

55

particularly useful to prevent the knowledge engineer from
becoming too lost while interactively building the Expert System.
The nodes, however, do not update dynamically. That is, to see how
changes have affected the set up, a new window has to be opened.
Also, although the nodes are buttons (implemented with Tigre) they
do nothing on depression. It would be particularly useful to have
the SubSpecialist move to select (find and highlight) the appropriate
node selected on the tree graph. This would save the steps required

to locate that node when in the SubSpecialist alone.

System Transcript g
_
Recycle attained a confidence value of 2. -- ESTABLISHED

NewObject attained a confidence value of 0. -- SUSPENDED
Compost attained a confidence value of 0. -- SUSPENDED

lncinerate attained a confidence value of 0. —- SUSPENDED
Landfill attained a confidence value of -3. -- REJECTED

 

figure 24: System Transcript

The System Transcript is mainly the output window for results.
This window will dynamically indicate what value of confidence each

node receives as the classifier runs through towards a solution.

3.4.3 Failure Handling

When conflicting knowledge is incomplete or inconsistent a node

cannot establish and a failure occurs at the node. The Hierarchical

Classifier identifies that failure as a confidence value of zero and

 

56

aborts continuing down the leg of that node. The problem solver

will then jump back one level to the next highest node and try to

establish one of its other children nodes.

3.4.4 Updating the knowledge base

Updating the system will be necessary as technology progresses.
This will have to be done via the table Matchers. The rules will have
to be checked periodically and altered accordingly. There is no easy
way to update or check these data/rules. A recommended addition

to the system is the Update/Check Data modules (Discussed in
Section 3.5).

3.5 DIFFICULTIES IN USE; RECOMMENDATIONS FOR
FURTHER DEVELOPMENT OF THE HIERARCHICAL
CLASSIFIER

The current Hierarchical Classifier runs nicely but could use a few

refinements.

1. It is difficult to run new cases. Currently, to run a new
classification, a new case name has to be entered in Figure 16, the
highest level node in Figure 19 must be highlighted,
SupSpecialistRelation's operate menu must be activated, "run"

highlighted, and finally "establish/refine" must be chosen.

57

2. A case cannot be edited. To run a similar case with but a few

changes, the entire system has to be run again. The capability
should exist such that the user could alter the information in one

section and find global response.

3. The "Quit" button is ineffective or missing. It would be nice
have a "Quit" button on windows. Some windows have "Quit"
button already, but its function is as of yet limited. For example, in
the table Matcher when the "Quit" button is selected, the system
prompts the user to remind that the changes since the last save will
be lost (even if no changes were made). Upon verification, the
system does not leave the table, and the user has to use the window
menu to close the window anyway. The DataBase define variable

window needs a quit button.

4 Saving the database requires a lot of time due to all of the

conformations.

5 The DataBase saves only with a direct data base save
command. Upon simply choosing "Quit" in the Launcher, a prompt
notifying that the DataBase has been updated since last save and
would not be saved along with would be a helpful safety feature.

Or, the save image could have the option to save the database.

6. The Matcher is a SortedCollection not an OrderedCollection

which poses a problem in some instances. That is, if nodes would

58

like to be placed in a particular order, they cannot be, as SmallTalk

sorts the list on input

7. In the Table Editor:

- the 'set condition' selection cancel leads to an error message

- when viewing the TableMatcher, if the mistake is made to
select to edit a table, cancel is not one of the options (only: Simple
Matcher or Data Attribute). After the mistake, a fake choice has to
be made, the table must pop-up, and the it has to be closed.
"Cancel" could be added to the pop-up menu.

- Need <OR> capability. Currently, the only way to do an
<OR> statement is to create a new line. Thus, the table could end
up looking like a diagonal array, with values along the diagonal, and
question marks elsewhere. As an example ,

if (A A (B V C) then 3
could be entered instead of:
if (A A B) then 3
and if (A A C) then 3.
as is the case in Figure 21 where the first two lines could be
consolidated as above. For small tables this is not a problem, but
as the information base becomes larger, this will be a great

hindrance.

8. No help has been installed for any of the windows. The help
button is available, but it leads to a prompt indicating that no help

is available.

59

9. The tree graph is a great visualization tool, however,

- it would be nice to have it selfUpdate, that is dynamically
update from any changes in the SubSpecialist Relation.

- it would be nice to be able to click on a node and have the
SubSpecialist relations window move to that node. As this tree
graph shows all nodes, it would not be difficult to locate a particular
node, instead of having to search the Matcher (as is currently
necessary). This will be particularly useful as the size of the

mapping becomes larger.

10. It is difficult to Update the knowledge of the system.

To update or add knowledge data/ rules into the system, a previous
understanding of the system is necessary in order to get into the
workings of the classifier. A data acquisition /data check module
could be developed as part of a user shell.

3.6 LIMITATIONS OF USING HIERARCHICAL CLASSIFICATION

Hierarchical Classification can be used mainly as the pruning
method to consolidate a search. It cannot however, carry out the
Life Cycle Analysis (LCA) alone. The pure classifying system simply

classifies. It does not quantify impacts.

It cannot store variables, nor can it manipulate them. This

capability will be essential for summing environmental impacts. To

60

allow this system to be suitable for Life Cycle Analyses, three

changes/ additions are needed.

1) An Explanatogg capabilipl is necessagr.
This system could tell what to do, but not why. A higher level

system should be capable of extracting the proper information at
the level of specificity the user requires.

2) The end-user interface will need to be developed.
Currently, an end user needs to have prior knowledge of the
building of the system to be in position to use it effectively. The
final product should allow a user—friendly atmosphere.

3) A knowledge Acquisition Module is needed
This module is essential to allow the ease and accuracy of the

knowledge base construction.

CHAPTER 4: PAPER PRODUCT CLASSIFICATION SYSTEM

INTRODUCTION

The experimental set-up of the classification system for paper is
presented. A simplified system is shown as an example. The
decision process is shown for one case to illustrate the operation of

the Hierarchical Classifier throughout the problem solving.

4. 1 The System

The simplified Classification system for paper is as follows:

61

 

figure 25: Classification System for Paper

Where Paper is the highest level node and the first descendants are
the nodes which represent the main breakdown of a life cycle. This
abbreviated hierarchy for paper is to illustrate the type of network
an LCA must deal with. The highlighted portion shows a
classification example traverse through a network of nodes. The
first node in bold print here is Manufacture. The true classifier has
been set up to start with RawMaterials, but this is just an example
layout. To traverse through the network to the 'lowest' node (lowest
in the tree shown in Figure 24, but highest in regard to specificity),
the top agent, Paper, queries for a description of a product. The
first answer it must establish is that the system is for a paper
product. In this system, this answer is trivial, but when this

system is expanded to include other higher level nodes and other

63

types of products, it will need to establish itself as being paper, as
opposed to tin, say. Once paper is established, the query is made to
determine what type of paper product: Paper or Paperboard? Then
further refine to the specific type of paper product (while still in the
top node). The table Matcher is set up as follows:

Pa er

   
    
  
   
    
 
    
     
 

CrepedPaper
PrintWrite

GroundePap CoatedPrirrting
BookUnCoared

NEWSPflPer BleachedBristols

SanitPaper TissuePap

PaperBoard

ConvertPaper

 
  
  

Bagsukpap WrappingPaper WetMachineBd COHSLPaPBd

Glassine ConstPap Constbd

ShippingSackPap

VegtableParchmerlt lnsulath
GreaseProofPap Pb ard Hardbd
BleachedPapbd ChllmFiIIbd
FoldingCanon CombBcndabd
LirlerBoard C ombPade
CombNonbendBd Unbleachedbd

TUbeCflnDrum CorrugatingMedium

figure 26: Table Matcher Example

As can be seen in Figure 25, the table Matcher not only establishes,
it refines as well. The outcome of this node will not only be that the
system is paper, but also what kind of paper. In Figure 25, the

darkened lines follow the determination that the system is dealing

64

with CorrugatingMedium. Each one of these nodes have a table
associated with them to determine the confidence value for that
product. All of these 'nodes' are associated with the refinement of
the top classifier node Paper.

This continues down until a leaf node (the lowest level) establishes.
The system will then halt and query the user if it should continue.
In the case of the LCA, at least four leaf nodes need to establish
before completion (one for each main branch: RMAcquisition,
Manufacturing, Use/ Operation, and WasteManagement). The
system has been set up such that each of these nodes will establish
(as it is assumed that all products have at least these processes
involved). Hence, once it is done establishing the processes
involved with the acquisition of raw materials the system will then
instruct the user that a leaf node has established, and the option
remains whether or not to continue. In order to complete the
classification, at least three other nodes will also have to refine and
establish. This system, then, is made up of essentially four

separate, but similar and interlinked, sub—classification systems.

4.2 Table Matchers

It has been attempted to wherever possible, establish the rules for
the table Matchers from information in the literature. Where no
specific rules were located, rules of thumb were sought out. An
example of such is shown in Table 10 abbreviated from the Solid

Waste Handbook [28]:

65

Reuse Recycling Suitability Suitability as Suitability for

Potential uitabili forl stora e Incinerator fuel Lan lllin
Low High/Variable Not Suitable High/Variable Not Suitable

Table 1 0: Sample Basis for Waste Management Rules

With this information, the Landfill node was set with the condition:
if the component is paper, then the confidence of using a landfill is -
3, hence the Landfill node will be rejected every time paper is
involved [see also 49]. Whereas, Recycling and Incineration would
establish with a high value of confidence. Where such information
was unattainable a best hypothesis was set for this experimental

system.

4.3 Sample Results

As a whole, the results obtained seemed fair. Three sample results
are provided to demonstrate the classification runs. The first
example is a walk through the interactive procedure to complete an
analysis. The next two examples simply state the waste
management outcomes, as they are the most interesting (the other
nodes simply establish the leafs of RawMaterials, Manufacturing
and Use/ Op, which are relatively generic). In an entirely built
system, the specific processes involved in each step would be

explicitly identified.

66

4.3.1 Corrugated Medium Example

The following is a sample of a classification run. The first item is
the question asked by the system, with possible responses in "< >".
The second (in italics) is the user-response given to identify this
example as Corrugated Medium: The unindented lines are the
responses of the computer after user-entry of the answer to the

question just asked.

SAMPLE CLASSIFICATION
1) Is this system to classify Paper? <Yes, No, Unknown>

Yes. .
2) Is this Paper or PaperBoard? <Enter a String or Unknown>
PaperBoard.
3) What type of PaperBoard? <choose from a list>
ComtgatingMedium.
Paper.HC.Top attains a confidence of 3 - ESTABLISHED
RMacquistion attains a confidence of 3 - ESTABLISHED
Farming attained a confidence of 3 - ESTABLISHED
Forestry attained a confidence of 2 - ESTABLISHED
Timber attained a confidence of 2 - ESTABLISHED
4) A Leaf Node has Established. Is the classification
complete?<Yes, No>
No ‘
5) How much of Raw Material is from Recycled Paper?

67

80%
FromRecycle attained a confidence of 3 - ESTABLISHED
6) Is the waste stream presorted? <Yes, No, Unknown>
Yes.
[PreSort Establishes}
7) A Leaf Node has Established. Is the classification
complete?<Yes, No>
No.
{Manufacturing nodes establish}
8) A Leaf Node has Established. Is the classification
complete?<Yes, No>
No.
9) How many times can the consumer reuse it? <value>
5.
{Use/ Op nodes establish)
10) A Leaf Node has Established. Is the classification
complete?<Yes, No>
No.
WasteManagement attained a confidence of 3 - ESTABLISHED
Recycle attained a confidence of 2 — ESTABLISHED
NewObject attained a confidence of 0 - SUSPENDED
ToRawMat attained a confidence of 2 - ESTABLISHED
11) A Leaf Node has Established. Is the classification
complete?<Yes, No>
No.
Compost attained a confidence of 2 — ESTABLISHED
Incineration attained a confidence of 1 - SUSPENDED

68

Landfill attained a confidence value of -3 - REJECTED

4.3.2 Other Example Results

These results seem fair estimations of what could be done with the
products (only the waste management node results are shown for
the following two examples):
For NewsPaper:
WasteManagement attained a confidence of 3 — ESTABLISHED
Recycle attained a confidence of 3 - ESTABLISHED
NewObject attained a confidence of l - SUSPENDED
ToRawMat attained a confidence of 2 - ESTABLISHED
Compost attained a confidence of 2 - ESTABLISHED
Incineration attained a confidence of 1 - SUSPENDED
Landfill attained a confidence value of -3 - REJECTED

Which is essentially the same as for the CorrugatingMedium.

For Tissue Paper
WasteManagement attained a confidence of 2 - ESTABLISHED

Recycle attained a confidence of —2 — REJECTED
Compost attained a confidence of 2 - ESTABLISHED
Incineration attained a confidence of 0 — SUSPENDED
Landfill attained a confidence value of -3 - REJECTED

CHAPTER 5.0 CONCLUSIONS/ FUTURE RESEARCH
RECOMMENDATIONS

In the near future, an environmental audit will be required for every
product and process, if not due to government regulation then due
to the markets (consumers) demand. Life Cycle Analysis can
provide the essential information necessary to improve each
product and process environmentally and economically. With
further development, the characteristics of an Expert System can be

exploited to provide a tool to perform such analyses.

5.1 BENEFITS OF USING HIERARCHICAL CLASSIFICATION .
FOR LIFE CYCLE ANALYSES

Hierarchical Classification (HC) can help solve some of the problems
associated with Life Cycle Analyses (LCA).

69

70

1) Data Limitation. Although this has been a major stumbling
block in the past, HCs can help to circumvent it. Proprietary data
need not be viewable or reachable within a system. Also, if the
information is unattainable: the Expert Systems will provide the

'best' answer with incomplete information.

2) Lack of a Standard. The developed system can aid in providing
the standard for carrying out LCAs. This system will then have to
be tested completely by the governing bodies to implement such a
standard, and in the long run, through interactive development —

this system could be the LCA standard.

3) Cost. An LCA done using an HC utilizes the information of
thousands of experts for the cost of one: the Expert System.

4) There exists no common currency. This is lesser tackled by the
HC, than it is by the nature of LCAs. LCAs are not to criticize, but

to present environmental impacts.

5 and 6) Boundam Conditions and Aggregation. Boundary
Conditions are sometimes set to simplify the problem. With the aid

of computing power and methods used by the HC to reduce

computing time, the complexity of a problem is not a problem.

7) Fuzzy Rules Some experts cannot give precise quantitative data

about processes, but can give estimates and rules of thumb which

71

govern the operation of these processes. This fuzzy, non—precise

type of data is generally implementable in Expert Systems.

5.2 LIlVIITATIONS OF USING HIERARCHICAL CLASSIFICATION
FOR LIFE CYCLE ANALYSIS

Hierarchical Classification can be used mainly as the pruning
method to consolidate a search. It cannot however, carry out the
LCA alone. The pure classifying system simply classifies. It does
not quantify impacts. Ultimately, the goal of an LCA is to do just
that, quantify impacts.

5.3 FUTURE WORK RECOMlVIENDATIONS

A system needs to be developed such that exploits the
characteristics of a knowledge rich system and can provide the user

with appropriate life cycle information.

5.3.1 Higher Level System

A final package should consist of a user-shell which will provide a
simple interface for the user. One that could conduct an LCA by
only asking pertinent questions. This systems LCA algorithm could
be based on the best available methods, such as from CSA Z760, or

72

form Franklin Associates, Ltd. Such a system could take the form

as in Figure 26:

 
      
  

TransportDataBase ‘
WasteManageDataBase

  
   

Conduct LCA
Update Data

Figure 27: Potential LCA Expert System Application

Where the HC is a sub-part of the entire system. The Hierarchical
Classification system could be run by the "Conduct LCA" button.
The classifier will determine the processes involved for the desired
product input by the user. Then via access to the appropriate
databases, the system could return a quantified Life Cycle Analysis
in a User-acceptable (i.e. understandable) manner. Or, once the
process information has been established by the HC, a
mathematical model such as the Manufacturing Systems of Koenig

and Tummala could be adapted.

The Data manipulation buttons ("Update Data", and 'View Data")
will be of great use need to update or viewe the Rules or Data for

accuracy. These buttons would lead to questions about the

73

information to update or view and would alter show accordingly.
The nice feature about this is that this higher level will take care of
the manipulation of the hierarchy such that no prior knowledge
about the inner workings of the system is necessary, only the
knowledge about the information architecture. This system will
lead the user with appropriate questions to assure correct input of
data. Since this system's integrity is an issue, data input should
only be allowed through various security checks (password levels for
example). This could help prevent hackers or desperate

manufacturers from adulterating the system.

5.3.2 System Scope

The system needs to include not only paper products but also other
major product components such as: ferrous and non-ferrous
metals, composites, plastics, and cloth. The Higher Level System
mentioned above should consist of a database also to determine the
end impact values. This system should provide as concise and
subjective a result as possible. It might be suitable to develop eco-
balance modules for important subsystems such as energy
production, waste treatment and transport. Standard modules for
the production of important materials encompassing all the life
cycle stages up to the final production stage could also be

calculated and provided as input into product eco-balances.

74

5.3.3 Knowledge Acquisition and Domain Expertise

Once the basic system has been built, the task is long from over.
The knowledge acquisition phase is a long and arduous task. This
system should be able to classify a wide spectrum of products. For
a new product, for example, a design engineer is faced with the
same basic questions, i.e. how will this design affect the
environment, or is full compostability the best design for a
particular product? In order to accomplish such a decision making
process, the information in the system must be extensive. Not only
must the information be attainable to the system, it should be
correct (the acquisition module should have a feasibility checker,

though it cannot check for correctness).

Hence, the knowledge acquisition phase will involve interviews with
many experts (essentially corresponding to the nodes represented
in Figure 24, Chapter 4). This system must attempt to imitate such
expert knowledge. Some engineers may know about the
manufacturing of Polymethylmethacrylate but know nothing about
paper bag manufacturing. So, in order to make the manufacturing
nodes as 'smart' as possible, as much expertise from as many

sources as possible needs to be included.

75

5.3.4 Continual Validation of System Performance

The system will have to be tested for performance throughout its
continued development. This will provide information as to system
capabilities, limitations and areas in need of improvement.

These tests can be made using LCAs that have been previously
studied. These results can then be compared with those obtained
from the system.

The Expert System construction should include continual checking
of the decision making rules involved in the system. These rules
should be discussed and verified by the appropriate experts (as
mentioned in Section 5.3.3, a data acquistion module cannot check

for correctness).

5.4 IMPACT OF THIS RESEARCH

This research introduces a new, more scientific approach to Life
Cycle Analysis. In the past, Analyzers oversimplified life cycles to
make data more manageable. With computer-aided design, this
unruly amount of data can be easily managed and studied in an
efficient manner.

There are nearly 135 public LCAs are available. Without a set
standard for carrying out such a study, conclusions can vary
greatly, confusing an already difficult question. This application
could help define the standard to minimize the grayness around
environmental decisions. A tool such as a Computer-Aided Life

76

gflcle Analyzer would minimize the amount of time needed to
evaluate the impacts associated with a product. Such studies will
then aid designers in developing as environmentally sound a
product as possible, and will also help legislators determine and

develop environmental regulations.

LIST OF REFERENCES

1.

2

77

LIST OF REFERENCES FOR LIFE CYCLE ANALYSIS

64 Questions About Composting. Minneapolis, MN: Buhler Inc.

. Composting and the Solid Waste Stream. : Solid Waste
Composting Council. -

. Decisilogn8-S13/Iakers Guide To Solid Waste Management. : EPA, Nov.

. ISOZIEC Strategic Advisogg Group on Environment [SAGE] Sub—
Grou on Environmental Pe orrnance Standards".
Nggghbrook, IL: Underwriters Laboratories, Inc., 31 Jan.

. "Life cle Assessment Newsletter Supplement." SETAC News
an. 1992,

. Policies and S stems of Environmental Im act Assessment.
Environmental Series 4. New York: nited Nations, 1991.

"Post—Project Analysis in Environmental Im act Assessment."
Environmental Series 3. New York, 19 0.

, . Yard Waste Composting: A Study of Eight Programs. : EPA, Apr.

1989.

. A.D.Little. Disposable vs. Reusable Diapers: Health,
Environmental and Economic Comparison. , Mar. 1990.

10. Baumgartner, Thomas, and Frieder Rubik. Evaluation of Eco-

l

Balances. Heidelberg, FRG, Apr. 1992.

1. Baurrlr gamer, Thomas Dr. Letter to Narayan, Ramani. 5 May

12. Benda, James, and R. Narayan. "Life Cycle Anal sis, A

Marketing Nightmare." Cellulose '91 (1992 :

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

78
Paper.
Braunschweig, Arthur Dr. Letter to D. Rais. 23 Nov. 1991.

Brookes, Warren. "Greens aper over olystyrene's benefits."
The Detroit News 18 eb. 1991, omment: 11A.

Chynoweth, Emma. "Health/Safe Model: ISO 9000."
Chemical Week (30 Sept. 199 ): page 62.

Dean, Norman L. Letter to SETAC Foundation for
Environmental Education. 1 Feb. 1992.

Frankfin Associates, Ltd. Com arative Ener and

Environmental Anal sis o T ree Interior Pac ‘n
Materials. , Nov. 1990.

Prepared for Free-Flow Packaging corporation.

--—. ife gflcle Analysis, Brochure
---. Life gflcle Analysis: A Planning Tool for the 90s.

Hocking, Martin B. "Paper Versus Polgstyrene: A Com lex
Choice." Science 51 (1 Feb. 19 1): page 504-50 .

Hunt, Robert G., Jere D. Sellers, and Franklin William E.
Resource and Environmental Profile Analysis: A Life (flcle

Environmental Assessment For Products and Procedures.
Massachusetts Institute of Technology: Elsevier Science
Publishing Company, Inc.

Koch, Dianne G. "Lifecycle Methodology as it relates to the
Environmental Quality of Packa ing." EnvironPak 11
Conference. Washington, DC., 6 cc. 1990.

Lehrberger, Carl, J. Mullen, and C. V. Jones. Diaper

Eggilronmental Impacts and Life ggcle Analysis. : Jan.
1 .

Levy, Mike, and William Franklin. Life Cycle Analysis and

25.

26.

27.

28.

29.

30.

31.

32.

33.

34

35

79

Complu‘pg with CONEG's Model Packaging Standards
Legis ation. ,Apr. 1992.

Maffia, Gennaro. "Recycling Al orithm for the Development of
Strategy." Cellulose 9 . ew Orleans,. England: Elis

Harwood. Paper #204.

Munici al Solid Waste Task Force US EPA Agenda for Action. :
E A, Feb. 1989.

Rees, William E. "Economics, Ecolp’gy, and the Limits of
Conventional Anal sis." Air& aste Mana ement
Association 41.10 Oct. 1991): 1323-1327.

Robinson, William D. The Solid Waste Handbook. USA: John
Wiley & Sons, 1986.

Rosenberg, Victor, Cyrus Ghalambour, Peter R cus, et al. Pro-
Cite. Com uter software. 2.0 ed. Personal ibliographic
Software, nc., 1991. 2 disks.

Society of Environmental Toxicology and Chemistry. Lifeg flcle
Analysis Jan. 1992,

Stipp, David. "LCA Measures Greenness but Results May not be
Black and White." Wall Street Journal

Stoetin , P., and F. Rubik. Oekobilanzen von BabyWindeln.,
Fe eral Republic of Germany

Strauss, Ste hen. "The Haze Around Environmental Audits."
Techno ogy Review (20 Apr. 1992): page 19-20.

. Tillman, A., H. Baumann, E. Eriksson, et a1. Life—cycle analyses

of selected acka in materials. Goetborg, Swe en:
Zetterqvist Jlfikeri $.1992.

. US. EPA. Bibliography of Municipal Solid Waste Management
Alternatives. as ' gton, DC., Aug. 1989.

80

LIST OF REFERENCES FOR PAPER

36. Characterization of Munici a1 Solid Waste in the United States:
1992 Update. Executive Summary ed. : US. EPA, July

1992.

37. Fact Sheet on 1991 Energy Use In The US. Pulp and Paper
lndustgg
1972 1990 and 1991 ener use corn arison. A
Summagg. New York, NY.: %erican Paper Institute, 9
Sept. 19 2.

38. The Paper Handbook, 3rd ed. Portland, Oregon: Boise Cascade
Paper Group, 1976.

39. Paper Recycling: The Induspgy's 40% Recovery Goal 1992
Pro ress Re ort. : American Paper Institute.
T o e goal for 1995 of 40%) recycling.

40. Paper Recycling: the Art of the Possible 1970-1985. : Midwest
Research Institute, 1972.

41. Pulpwood Statistics. Washington, DC., Nov. 1989. 89-A-l2.

42. Wheiggbrades of Recyclable Paper Go and What They Make. ,

43. Apotheker, Steve. "Market Trends of Old Newspapers."
Resource Recycling (July 1992): page 25.

44 Braungart, Michael, Special Wastes: Prevention, Reduction,
and Disposal. Proceedin s of an international symposium
sponsored by Ciba-Geigy irnited

45. Carroll, Robert C., and Thomas P. Ga'da. "Mills Considering
New Deinking Line Must Answer nvironmental Questions:
Physical and chemical pro erties of various wastepaper
grades have to be analyze before mills can permi and

81

handle process wastes." Paper Recycling: Strategies,
Economics, and Technolog. Editor. Ken L. Patric . San
Francisco: Miller Freeman. Chapter 29.

46. Casey, J. P. Pulp and Paper, Chemisgg and Chemical
echno ogy, 3rd e . 01. II. New ork: John Wiley & Sons,

1980.

47. De Baere, Luc, Six Winfried, Richard Tillinger, et a1.
'Waste agper Improves Biowaste Composting." BioCycle
(Sept. 9 2):

48. El Nawawy, A. S., E. El-Rayes, R. D. Al Hussaini, et a1.
"Bioconversion of Cellulosic Wastes." Perspectives in

Biotechnolofl and Applied Microbiology.

 
 
 

: National and Local Pers ectives.
C.: US. E.P.A., 1991.

49. Toxics in the Communi
EPA. Was ington,

50. Goldfitgéré) Nora. "Adding Paper to the Mix." BioCycle (Aug.

51. Iannazzi, Fred D., and Richard Strauss. "Changing markets for
recycled paper." Resource Recycling (Apr. 19 2): page 81.

52. Ince, Peter J ., and Joanne T. Alig. 'Waste Pa er Recycling and
the Future Timber Market.‘ Resource ecycling (Apr.
1992): page 123.

53. Jeris, John S., and Raymond W. Re an. "Controlling
Environmental Parameters for ptimum Composting:
Parts I-III." Compost Science (Jan. 1973):

54. Kline, J. E. Pa er and Pa erboard Manufacturin and
Converting Fundamentals. California, U.S.A.: Miller
Freeman lications, Inc., 1982.

55. Larson, Nancy, Executive Director. Environmental Impacts of
Paper Recycling. , 16 Sept. 1992.

56. Maffia, G. J. Waste Not - A Review of Municipal Solid Waste

82
Processing: Existing and Potential. Diss. New Hampshire:
Thayer SC 001 0 Engineering, 1986.

57. Maffia, Nancy L., Principal Investigator. Algorithm for the

Develo ment of Rec clin Strate on Local and State-Wide
Levels. , 11 Feb. 1991.

58. Misner, Michael. "The Emerging World of Deinking." Waste Age

(June 1992):

59. NicoltleggéJoseph C. Letter to Energy Coordinators, API. 9 Sept.
US. Pul and Paper Industry's Energy Use — 1972, 1990,
and 19 l.

60. ---. Letter to Enegpy Coordinators. 9 Sept. 1992.
US. Pulg an aper Industry Ener Efficienc : Calendar
Year 19 1. A Comparison. lnclu es Annual nergy
Report. Energy broken down to Energy Use per ton.

61. ~-—. Letter to Ener Coordinators, API. 9 Sept. 1992.
Annual Re 0 of In-Plant Electric Generation, Electricity
Sales to U ‘lities and Cogenerated Electricity by the US.
Pulp and Paper Industry.

62 Narayan, Ramani. Environmentally Degradable Plastics --
Rationale & Design, First International Co erence on
nggage Degrada ility, Chicago Illinois, 24—25 October

63. Powell, Jerry. "How are We Doing? The 1991 Report." Resource

Recycling (A r. 1992):
Production aprrd Recovery in selected grades of Paper.

64 Smith, Stephen C., Bruce A Low, and Robert V. Herman.

Desi n and Performance of a Rotatin -Drum Com oster for
Evaluation of Degradabilig of NonWovens and Otfier
Materials.

65. Tracyi Ken Dr., Vice President of Environmental Technology,
1 Sggn Products. "Testimony of Dr. Ken Tracy." , 5 June

83

66. US. EPA.

67. Usherton, Judy. "Rec cle paper and sludge.

68

69

70.

71.

72.

73.

74

Bibliography of Municipal Solid Waste Management
Alternatives. as 'ngton, DC., Aug. 1989.

Has ublications about the various waste management
solu ‘ons listed.

'1

Resource

Recycling (Mar. 992 pgs. 95-100.

. Van Derveer, Paul D., and Kenneth E. Lowe. Fiber Conservation
and Utilization. California, U.S.A.: Miller Freeman
Publications, Inc., 1975.

LIST OF REFERENCES FOR EXPERT SYSTEMS

. The T igre Programming Environment. : ParcPlace Systems,

Brown, David C., and B. Chandrasekaran. Desi n Problem
Solvin : Knowled e Structures and Control Strate ies. San
Mateo, California: lVlorgan Kaufmann Publishers, Inc.,

1989.

Chynoweth, Emma. "Health/Safe Model: ISO 9000."
Chemical Week (30 Sept. 199 ): page 62.

Finn, Gavin A., and Kenneth F. Reinschmidt. "Expert Systems
in Operation and Maintenance." Chemical Engineering
(Feb. 1990):

Rich, Elaine, and Kevin Knight. Artificial Intelligence. 2nd ed.
New York, NY.: McGraw-Hill, Inc., 1991. Shelves.

. Settle, Frank A. Jr., Michael Pleva, Clair Booker, et al. "An
information System for Selective Methods of Chemical
Analysis." American Laboratory (Mar. 1987):

84

75. Tummala, R. La], and Bruce E. Koenig. Classes of Process

Models and Their Characteristics Useful for the Design of
Manu acturing Systems.

76. ----- Ente rise Models for the Desi n and Mana ement of
Manufacturing Systems. Submitted to Joint US7German

Co erence on New Directions for Operations Research in
Manufacturing 15 May 1991

77. Waterman, D. A A Guide to E ert S stems. Reading, Mass:
Addison Wes ey Publishing Co., 986.

APPENDIX A

UNIX HIERARCHICAL CLASSIFICATION WINDOWS

This appendix contains the working windows for operation of the Hierarchical
Classifier

ElLauncher — —— barnard:

1:"‘11

@800

an
i
5
3

Hi

i

E Pap.HE:SubSpectalistRelatron.Pap.ll£

@088

a
g:
91

 

 

 

 

 

 

lEl Launcher E '

 

 

[3 System Transcript

. Usellequlrements attained a confidence value of 3. - ESTABLISHED

‘ UsePollutlon attainedaconfidence value of -l. — SUSPENDED

. Electridtyllequred attained a confidence value of -l. - SUSPENDED

: OtherProdReuu'red attained a confidence value of -l. — SUSPENDED

‘ orherProcessNeeds attainedaconfidence value of—l. - SUSPENDED

i WasteManagelnentattained atonfidena value of 3. — ESTABLISHED

‘ Recycle attained a confidence value of 2. — ESTABUSHED

' NewObject attained a confidence value of 0. — SUSPENDED

. toPaperllN attained a confidence value of 0. — SUSPENDED
Compost attained a (onfidence value of 0. —- SUSPENDED

' Incineratll aminedaconfidono value of o. -— SUSPENDED

 

 

 

'IICHIGQN STQTE UNIV.

ll lll lllll llllllll lll lll

31293008978664