a decision support...
TRANSCRIPT
A DECISION SUPPORT SYSTEM
FOR CHEMICAL INCIDENT INFORMATION
A Thesis
by
GAURAV SHARMA
Submttted to the Office of Graduate Studies of Texas ARM University
in Partial full'i llment of the requtrcments for the degree of
MASTER OF SC1ENCE
August 2002
Subject Major: Chermcal Engineering
A DECISION SUPPORT SYSTEM
FOR CHEMICAL INCIDENT INFORMATION
A Thesis
By
GAURAV SHARMA
Submitted to Texas A&M University in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
Approve as to style and content by:
. Sam Mannan (Chair of Committee)
H rry H. West (Member)
Marieua J. Treuer (Member)
Rayford G. Anthony
(Head of Department)
August 2002
Major Sub)cct: Chcmtcal Engineering
ABSTRACT
A Decision Support System for Chemical Incident Information.
(August 2002)
Gaurav Sharma, B. E. , Panjab University
Chair of Advisory Committee: Dr. M. Sam Mannan
Decision Support Systems (DSS) find extensive applications in business enabling
industry managers to make intelligent risk decisions. Although widely used in business
applications, the application of DSS to Process Safety Management has been lacking.
This thesis proposes the development of such a DSS based on chemical incident
information.
Chemical incident information is mostly qualitative in nature. Therefore, mathematical
and statistical analysis of this information is an extremely challenging problem. This
thesis introduces indices that quantify the qualitative nature of chemical accident
information. Weighted Scoring Method is the chosen decision aid for the DSS. Using
this decision aid, the various indices are finally consolidated into a single index that
serves to facilitate decision making for process sal'ety.
The proposed DSS is meant to be user specific. There is scope for the individual user to
use the DSS as pcr hts/her decision-making criterion.
ACKNOWLEDGEMENTS
I express sincere gratitude to my research advisor, Dr. M. Sam Mannan, for providing
academic and moral support for the successful completion of my research. I also thank
Dr. William Rogers for his valuable advice and encouragement. I appreciate the
suggestions and recommendanons proposed by my co-worker, Nir Keren. Finally, I
thank my family and friends for their steadfast belief in me,
TABLE OF CONTENTS
Page
l. INTRODUCTION .
1. 1 Process Safety Management. 1. 2 PSM and Incident Investigation.
2. INCIDENT INVESTIGATION .
. I
. 2
2. 1 Methodology of Incident Investigation . . 2. 2 Sequence of Events Leading to Industrial Incidents . . . 2. 3 Incident Investigation Techniques . 2. 4 Incident Databases . . 2. 5 Analysis of Incident Data. .
3. DECISION-MAKING AND PROCESS SAFETY MANAGEMENT . . .
4 . . . . . . 5
. 5
. 7
. 10
. . . . 1 2
4. DECISION SUPPORT SYSTEMS 16
4. 1 Introduction. .
4. 2 The Origin of Decision Support Systems 4. 3 DSS Development .
4. 3. 1 Specific DSS 4. 3. 2 DSS Developer 4. 3. 3 DSS Source.
16 18
. 20
. 21
. 21
. 21
5. RISK . 22
5. 1 Introduction. . 5. 2 Risk Decision-Making
. 22
. 23
6. DECISION AIDS. . 28
6. 1 Introduction . . 6. 2 Game Theory . 6. 3 Mathematical Programming .
6. 4 Goal Programming. 6. 5 Compromise Programming. 6. 6 Cost-Benefit Analysis. 6. 7 Voting Methods .
. 28
. 28
. 29
. 30 . 30 . 31 . . 33
Page
6. 8 Weighted Scoring Methods 6. 9 Screening/Ranking Methods. 6. 10 Nominal Group Technique 6. 11 Payoff Matrix Analysis.
34 36 37 37
7. CHOICE OF DECISION AID. 39
7. 1 Introduction 7. 2 State the Problem. 7. 3 Identify Distinguishing Aspects of the Problem . . . . . . . . . . . . . . . . . . . . . . . . . 7. 4 Study of Decision Aids . .
7. 5 Comparison of Problem Characteristics and Decision Aids. . . . . . .
39 39 41 42 44
7. 6 Rapid, Single-Entity Decision Aid for Chemical Incident Information . . 45
8. DECISION CRITERIA . . 47
8. 1 Introduction. 8. 2 Decision Criteria for Chemical Incident Information . . .
8. 2. 1 Environmental Impact. .
8. 2. 2 Dollar Damage 8. 2. 3 Litigation Cost 8. 2. 4 Employee Disenchantment 8. 2. 5 Government Action .
8. 2. 6 Company Disrepute 8. 3 Calculation of Consolidated Decision Index .
. . 47 . . . . . 48
. . 48 54 55 56 57 58 59
9. THE DECISION SUPPORT SYSTEiil . . . . 61
9. 1 Case Study for Decision Support System 9. 2 Decisions Based on Missing Data
. . 61 73
10. CONCLUSIONS AND FUTURE RESEARCH . . 74
LITERATURE CITED . 77
VITA. 80
vn
LIST OF TABLES
Table Page
On-Site Fatality Index Calculation. . . . .
Off-Site Fatality Index Calculation . . . .
On-Site Injuries Index Calculation . . . . .
Off-Site Injuries Index Calculation . . . .
. „, 49
, . „50
. . . . 50
. . . . 5 I
Hodge-Sterner Table with Toxicity Index Calculation . . . . . . . . . 52
Criteria for Determining Environmental Index . . . . . . . . . 53
Dollar Damage Index Values 54
Litigation Cost Index Values.
Employcc Disenchantment Index Values. . .
55
. . . . 56
10 Government Action Index Values . . . . . . . 57
Company Disrepute Index Values . . . . . . . 58
12 Calculation of Consolidated Decision Index . . . . . . . . . 60
13 Calculation of Environmental index . . . . . . . 64
14 Environmental Index for Case Study . . . . . . . . . 65
15 Dollar Damage Index for Case Study. . . . . . . . . 66
16
17
Litigation Cost Index for Case Study . . . . . . . . . . . . . . . . . . . . . .
Employee Disenchantment Index for Case Study . . .
. . . . 67
. . . . 68
Government Action Index for Case Study . . .
Company Disrepute Index for Case Study . . .
. . . . 69
. . . . 70
20 Default Values for 'Weights" of Decision Criteria. . . . . . . . 71
Table Page
21 Consolidated Decision Index for Case Study. . . . . . . . 72
1. INTRODUCTION
1. 1 Process Safet Mana ement
Process Safety Management (PSM) is one of the main areas of concern for the
Occupational Safety and Health Administration (OSHA). The occurrence of the Bhopal
tragedy in 1984 resulted in thousands of fatalities and prompted the development of
regulations in the United States to protect workplace employees and off-site public from
incident consequences [1]. The Clean Air Act Amendments of 1990 were promulgated
in this background and it directed OSHA to implement regulations to protect thc
workplace employees [I]. In response to that directive, OSHA promulgated the 29 CFR
1910. 119 Process Safety Management of Highly Hazardous Chemicals standard [2]. The
objcctivc of this standard is to establish minimum standards for the chemical process
industry (CPI) to utilize pnnciples of safety scicncc and cnginccring to reduce the
consequences of chemical incidents in the process industry [3].
Thc Process Safety Management (PSM) standard consists ot fourteen elements [3]:
l. Employee parttcipation
2. Process safety information
3. Process hazard analysis
This thesis conforms to Process Safety Progress.
4. Operating procedures
5. Training requirements
6. Contractors
7. Pre-startup safety review
8. Mechanical integrity
9. Hot work permits
10. Management of change
11. Incident investigation
12. Emergency planning and response
13. Compliance audits
14. Trade secrets
These fourteen elements are comprehensive in their treatment of chemical process safety
issues and are inter-dependent [3]. The holistic Process Safety Management (PSM)
Model encapsulating the fourteen elements of process safety clarifies the relationship
among the various cntittes of the OSHA standard [4].
1. 2 PSM and Inctdent Investi ation
One of the elements of OSHA's PSM standard is incident investigatton [2]. An incident
is defined as an event, which might have been an "accident or near miss" [5], that "did or
could have caused injury, or loss of or damage to property or environment*' [5]. This
definition of incident propounded by the Center for Chemical Process Safety (CCPS) is
comprehensive and serves to identify a wide variety of incidents.
2. INCIDENT INVESTIGATION
2. 1 Methodolo of Incident Investi ation
The methodology of incident investigation depends upon the complexity and depth
required for a particular incident [5],
The simplest type of incident investigation involves an informal inquiry carried out by
the personnel in-charge of the unit affected by or involved in the inctdent [5]. Such an
inquiry also includes the people injured in the mcident [5].
At a higher level, incident investigation includes a more formal inquiry involving safety
expetts [5]. This consists of brainstorming sessions involving seasoned safety
consultants and industry personnel [5].
The most sophisticated incident investigation involves a thorough understanding of
multiple-root causes of the incident [5]. The incident. inl'ormation collected through such
an analysis is input back to the Process Hazard Analysis (PHA) teams.
2. 2 Se uence of Events Leadin to Industrial Incidents
Many theories have been proposed to study the cause of industrial incidents [5]. In
general, industrial incidents go through a phase of I) initiation, 2) propagation, and 3)
eventual occurrence [5]. The initiation of an industrial incident may be triggered by
sources such as human error or organization error [5]. Multiple triggering events might
be an initiating event. Propagation involves breakdown of the control and safety systems
and may be caused by human error, design flaws or organization error [5]. The eventual
occurrence of an incident is the last step in the sequence of events leading to an incident
[5]. The severity of an incident will depend upon various factors like chemical inventory,
degree of human exposure, time of exposure, and energy released [5].
2. 3 Incident Investi~ation Techm ues
There are many incident investigation techniques that have been formulated by industry
safety experts. Diffcrcnt investigation techniques serve different purposes. Depending
upon the industry requirement, the choice of an appropriate investigation technique can
be made. If required, a choice of multiple investigation techniques can be done.
Guidelines for Investigating Chemical Process Incidents [5], a CCPS Publication divtdes
the investigation techniques into three subgroups:
1. Deductive Techniques
2. Inductive Techniques
3, Morphological Techniques
The deductive techniques aim to find root cause(s) of incidents [5]. This type of incident
investigation starts from occurrence of a specific incident [5]. Starting from that point,
the various initiahng and propagating events of that incident are determined [5]. Fault
Tree Analysis (FTA) is one of the tools used to accomplish this type of incident
investigation [5].
The inductive technique of incident investigation starts from a specific disrupting factor
in the process [5]. This deviation from normal operating conditions is comprehensively
studied [5]. The prospective initiating and propagating events are determined and a
general conclusion regarding the occurrence of that incident is reached [5]. Hazards and
Operability study is one of the tools used for this type of investigation [5].
The morphological technique of incident investigation involves analyzing the structure
and set-up of thc chemical plant and identifying potential hazards [5]. This technique of
incident investigation utilizes the experience of the industry personnel [5].
2. 4 Incident Databases
Once an incident has been investigated, it is imperauve to archive the incident
information. The purpose of storing this incident information is to provide the chemical
process industry (CPI) with information of previous incident information so that similar
incidents in the future can be prevented [5]. This type of informauon can be provided to
the following group of people I) Process Hazard Analysis (PHA) teams, 2) design
teams, and 3) operations and maintenance personnel. For this purpose, various chemical
incident databases have been constructed to store the incident information [5].
It has been observed that each mdustry sector within the CPI tends to develop its own
incident database over time. For example, different databases exist for pipeline incidents
(Reportable Incidents for Natural gas Transmission and Gathering Lines [5J, LNG Plant
Failure Rate Database [5], Pipe Failures m Land Based Pipelines [5], Pipeline Incident
Data [5]) as compared to olfshore incidents (Hydrocarbon Leak and Ignition Database
[5], Offshore Incident Data [5], World Offshore Incident Data [5], Platform Databank
[5J, IFP Databanks on Offshore Incidents [5]). There are databases that store incident
information exclusively for high frequency incident chemicals like ammonia (Ammonia
Plant and Related Facilities Safety [5]). Also, these incident databases are divided by
geography. Europe has its own Major Incident Repoiting System (MARS), which stores
incident information exclusively for member countries in Europe [5].
United States has numerous incident databases of which the RMP*Info Database
developed by the Environmental Protection Agency (EPA) is fairly widely used [6]. In
the year 1984, the release of methyl isocyanate in the Bhopal incident served as a
catalyst for development of a regulatory framework in the United States to protect
industry personnel and off-site public from chemical incidents [I]. This concern for plant
and public safety was manifested in the Clean Air Act Amendment of 1990 [I]. Section
112 ( r ) of this act set standards to mitigate the consequences of chemical incidents [I].
This section is at the heart of the Environmental Protection Agency's (EPA) Risk
Management Program [I]. The standards of the 112 (r ) section apply to all organizations
that carry quantities of chemicals above a certain threshold [I]. One of the requirements
ol' the RMP devised by EPA was that every facility must submit. a five-year (from June
21, 1994- June 21, 1999) incident history to EPA [I]. These incident data became the
RMP"'Info database [ I].
As mentioned above, there are a number of incident databases serving different. industry
sectors and geographical regions. This distributed model of incident databases has a
major drawback. Duc to the distnbuted nature of the incident databases, the information
and analysis derived from these databases is extremely limited in scope. In order to
overcome this shortcoming, there have been attempts to share information across the
industry. One such development is the development of the Process Safety Incident
Database (PSID) by the Center for Chemical Process Safety [7]. This incident database
was developed under an agreement between various industries to collect and share
incident data [7]. The incident data are collected in a central database and are available
to all participating companies for analysis [7]. Only those companies who volunteer
information can access incident data from other industries [7]. Thus, incident
information volunteering has an incentive. In fact, based on the yearly sales of a
company, there are a minimum number of accidents that must be submitted by each
company [7]. For example, a company having a sale of I billion USD has to submit at
least IO accidents per year.
Another incentive given to companies to volunteer information for this database is that
no company name is attached with an accident record. Date of the accident is not a
required field. As soon as data are entered into the database, the software destroys the
source name of that data. Thus, complete anonymity of the incident information is
preserved. The data stored in this database are always kept with CCPS and individual
companies access that data through forms and reports provided by the CCPS software
[7]. This particular incident database v:as built on the hnes of Exxon's Incident
Reporting and Analysis System (IRAS) [7].
10
2. 5 Anal sis of Incident Data
The incident data are mostly qualitative in nature. Therefore, mathematical or statistical
analysis of this data can be extremely cumbersome. Nevertheless efforts have been made
to accomplish the analysis of the incident data.
The classification of incident data and problems associated with it has been studied
comprehensively [8]. It was found during such a study that different incident data
analyses tend to find different classifications for the same data [8]. This can result in
confusion when comparing the same incident data from different sources. It was
observed that the root causes of incidents were often neglected during collection of
incident data [8]. Instead, the consequences of the incident were over-emphasized [8].
This study attempted to group incidents according to the keywords used to describe the
incidents [8]. It was observed that immense disparity existed during such a classification
because different people descnbcd the same incident in different ways [8].
Statistical analysis of domino chemical incidents versus non-domino chemical incidents
has been performed [9]. In this particular study, the severity of the incident is determined
by number ol' fatalities [9]. Thc frequencies versus number of l'atali ties curves are used
to cinry out the statistical and mathematical modeling of the incident data [9]. The
domino incidents have been defined as incidents whose occurrence causes a chain of
incidents to occur [9]. The comparative study of domino chemical incidents versus non-
domino is of significant interest [9].
The frequencies versus fatalities (F/N) graphs have been often used to analyze the
incident data. Such graphs are constructed for various industry sectors [10]. For
example, transportation incidents and industrial incidents have been studied exclusively
using the F/N curves [10]. The F/N curves are also used to assess the impact of process
safety management techniques on the process industry [10].
The RMP"'Info database containing the incident information has been comprehensively
studied in a working paper published by the Center of Risk Management and Decision
Processes, The Wharton School, University of Pennsylvama [1J. The incident
information from the RMP*Info database has been analyzed and various conclusions
drawn from it [1]. The top twenty chemicals tnvolved in most chemical incidents have
been determined [1J. The frequency of incidents versus the number of facilities reporting
those incidents has been calculated [1]. Also, the number of incidents for the various
industry sectors has been documented. Off-site fatalities and the number of fatahties on
each geographical region have been determtncd m this study (1].
3. DECISION-MAKING AND PROCESS SAFETY MANAGEMENT
Process Safety Management requires making intelligent decisions regarding process risk
and safety issues, which can be long-term strategic goals or short-term rapid risk
situations.
A long-term goal may include development of a systemic process safety paradigm for
your company. Within the systemic process safety model, it might be necessary to make
decisions regarding safe operating procedures, emergency response preparation, process
hazards analysis, pre-startup safety reviews, management of change and collecting
process safety information. All these issues require making long-term decisions.
Short-term decisions usually involve crisis situations. A sudden creation of an abnormal
situation in the process condition might warrant a rapid response to bring the process
back to its normal operations. An overpressure scenario inside a vessel, inadvertent
mixing of two or morc chcmtcals, or presence of contaminants will require rapid
decision making on part of the industry personnel.
Irrespective of whcthcr decisions to be made are long-term or short-term, it ts important
that any decision- making process involve studying and analyzing various types of data
from multiple sources. Three main sources of such information are a) hazard evaluation
procedures, b) risk assessment studies and c) chemical incident information. Hazard
evaluation procedures serve to provide the process safety personnel with information
regarding potential hazards in the process [11]. There are various types of hazard
evaluation methods and each method has its own technique. Some of these methods are
What-If Analysis, Hazards and Operability Studies (HAZOP), Fault-Tree Analysis
(FTA), and Failure Modes and Effects Analysis (FMEA) [11].
Risk assessment is a mathematical tool in the service of the process safety expert [12].
The steps in risk assessment involve determining source models, deriving dispersion
models, and doing consequence analysis of the incident scenario [12]. Principles of
industrial hygiene and toxicology are used in the consequence analysis [13]. Hazard
evaluation and risk assessment are the two main providers of information regarding any
process safety management decision [12].
The third main source of information that serves to facilitate process safety management
decisions is the chemical incident information [SJ. The idea behind using previous
chemical incident information is to revisit the mistakes made during an earlier incident
[S]. The root cause of the incident must be investigated and documented. This
inlormation can serve as a guide to decision makers [5].
Chemical incident information serves the same purpose as that of hazard evaluation and
risk assessment. At the same time there is a fundamental difference between the two.
14
A hazard evaluation/risk assessment approach is a "proactive" approach towards process
safety management [14]. This means that hazard evaluation/risk assessment studies
determine prospective hazards in the process and try to quantify them before an incident
actually occurs [14], For example, a what-if analysis postulates an abnormal situation in
the process [11]. This abnormal situation is then studied and consequences of such a
situation are determined [11]. If risk assessment is done following the hazard evaluation,
then the potential hazard will be quantified. Thus, hazard evaluation and risk assessment
studies are "proacuve" approaches to process safety management [14].
On the other hand, using chemical inc&dent information for process safety management
decisions is a "reactive" approach [14]. This means that measures for assuring safety of
the system are promulgated only after an incident has occurred and its root causes
determined [14]. The system is allowed to act on the judgment of the safety experts.
iVevertheless, the chemical incident information is an extremely important tool in the
hands of the safety experts and facilitates the decision making process.
The chemical incident information utilized is mainly the incident data stored in incident
databases. The incident data stored in databases suffer from thc following shortcomings:
a) Tools to analyze the incident data are absent from most databases. The plain-
vanilla information in the incident databases has hm&ted use, and interpretation of
this inl'ormation is left to the judgment of thc decision maker.
15
b) The incident information is extremely qualitative in nature. Therefore,
mathematical or statistical modeling of these data is extremely cumbersome.
Appropriate use of this information will require quantification of this information
in such a way that a mathematical decision aid can be applied to it
c) In some cases the information stored is incomplete. This incompleteness may be
the result of an incomplete inquiry of the incident. In other cases, incident data
may involve fields that cannot be captured during a post-incident investigation.
Despite the difficulties in using chemical incident data, the use of this information is
indispensable. This thesis addresses the concerns and drawbacks cited above, suggests
ideas to overcome these shortcomings and proposes the development of a decision
support system that will facilitate decision-making in process safety management.
The following groups of people in thc mdustry can use the chemical incident
information: a) process hazards analysis team, b) equipment design teams, and c)
operations and maintenance personnel. The proposed decision support aims to benefit
operators and maintenance personnel exclusively.
Thc operators in the chemical industry must make short-term rapid risk decisions. These
rapid risk decisions are m contrast to the long-term decisions to be made by process
hazard analysis teams or equipmcnt design groups. These rapid risk decisions are usually
crisis situations that occur during everyday operations of the chemical industry.
4. DECISION SUPPORT SYSTEMS
4. 1 Introduction
The concept of a decision support system (DSS) was first propounded in the 1970s by
Michael S. Scott [15]. Subsequently, this new tool in the computer industry became a
topic of interest for many researchers, and a definition for decision support systems was
formulated. Decision support systems (DSS) were defined as "interactive computer
based systems, which help decision makers utilize data and models to solve unstructured
problems'* [16]. This definition ol'Decision Support Systems comprises of three
significant terms a) interactive computer based system, b) data and models, and c)
unstructured problems.
An "interactive computer based system" implies that the system will be flexible enough
to adapt itself according to the needs of the user. This exchange of information between
the user and the system is a dynamic process as opposed to a static exchange in which
thc system simply feeds information to the user.
The second signi Bcant part of the definition is "data and models'*. Data are the basis of
the decision support system. With the advent of Relational Database Management
Systems (RDBMS), the data storage and access from different sources has become
extremely convement f16]. The storage and indexing of data in the database is an area of
17
study itself [17]. There are "best practices" to be followed when designing a database
model so that the manipulation of data can be made easier and more convenient [17].
Also, it is important to determine the sources for the data. Data collection should cover a
wide variety of sources [16]. This improves the efficacy of the DSS. Today there is a
huge software industry catering to the needs of database management itself. The DSS is
designed to extract useful data, process it and make inferences for future action. Models
are the algorithms that process the data to aid the decision making process [16]. These
models are computer programs that lie between data extraction and its final presentation
[16]. The models can be mathematical decision aids that process data so that intelligent
conclusions can be drawn from them [16].
The third important part of the definition is the use of the tetm "unstructured problems".
Unstructured problems arc problems that do not have a fixed methodology as their
solution [18]. On the other hand, structured problems have specific guidelines to be
followed to solve them [18].
This original definition of Decision Suppotx System was very specific, and some
software that were decision "facihtators" could not be categorized as DSS despite the
fact that they were enabling the decision making process [16]. Eventually, it was decided
that any computer-based system that atded decision-making could be labeled as a
Decision Support System [16]. This resulted in a wide variety of programs to bc labeled
as DSS [16].
4. 2 The Ori in of Decision Su ort S stems
There are different schools of thought regarding the evolution of Decision Support
Systems [16]. The most popular theory regarding the origin of DSS traces its evolution
from Electronic Data Processing to Management Information Systems and finally to
Decision Support System [16]. Electronic Data Processing (EDP) was the earliest form
of electronic data processing [16]. EDP was useful for file-processing and data storage
of the departments in the organization [16]. This system was inefficient because there
was no integration between the different files [16]. Each data file was independent and
had no hnks to the other data files [16], which resulted in requirement of greater storage
space thereby slov;ing down thc speed of the computers.
Eventually, EDP transformed itself into Management Information Systems (MIS) [16].
MIS was the next step after EDP, and its emphasis was the integration of the different
data-files [16]. The integration resulted in a holistic approach to data management [16].
Storage space requirement was considerably reduced and data processing became faster
[16]. At the same time, RDBMS transformed the way data were to be stored and handled
[17]. RDBMS propounded best practices from stonng and indexing data [17]. This
made data handling extremely efficient and elfective. The birth of MIS was facilitated
by the advent of RDBMS and increased processor speed.
MIS is now in a phase of transformation and the next stage of development is the
Decision Support Systems. These systems have a decision-based approach to data
handling, storage and processing. Once MIS has integrated the various data sources,
DSS uses the data to improve decision-making in the indusny.
DSS are being extensively used in the field of management. In the United Kingdom, a
Decision Support System to benefit the retail industry has been proposed [19], In this
particular case, Geographical Information Systems (GIS) were used to develop the DSS
[19]. GIS have been used to collect and store demographic data of various geographic
locations [l9]. These demographic data are of immense use to marketers and retail
industry in making decisions regardtng promotions and other customer related business
[l9]. GIS served as the data provider for the development of the DSS [l 9].
A software that simulates the consequences of an industnal incident has been developed
by Civil Protection Department, Lombardy, italy [20]. This simulation software acts as a
decision support system for the environment expctts [ZOJ. Thts particular software uses
consequence-modeling equations similar to the ones proposed by the Environmental
Protection Agency (EPA) [20]. These equations are analogous to the equations
suggested for risk assessment in the CCPS pubhcatton, Chemical Process Quantitative
Risk Analysis.
20
A DSS for the incidence response logistics (IRL) has been developed in Europe [21].
This system provides decision support in the event of a road incident [21]. Since, road
incidents can cause traffic disrupnons, a real-time DSS will help the incident response
teams in making quicker decisions [21]. This particular DSS contains mathematical
algorithms that help in the management of response units in the most effective manner
[21].
The examples of Decision Support Systems given above illustrate the use of these
systems in three &hfferent areas. There are numerous other applications of DSS in many
different areas of engineering and science. The application of DSS to utilize and analyze
previous incident information has not been implemented. The utilization of previous
chemical incident information is extremely useful for process safety management
decisions. Therefore, the development of a DSS that gives the safety experts guidance
for deciston-makmg can be an important tool in improvmg the safety performance of the
industry.
4. 3 D~SS D
A Decision Support System itself is a software application [16]. An application-
development software suite has to be used to develop a decision support system [16].
There are three lcvcls that clearly descnbe the development architecture of a DSS [16].
This is the actual DSS [16]. The functionality of this DSS depends upon the
requirements of the end user. This DSS can be a custom-made or a pre-designed
application. Mostly, the DSS being used are custom made. Eventually, as the DSS
market becomes more mature there will be pre-designed applications available. The
customer uses the Specific DSS after it has been implemented. After its implementation,
the customer must be be trained regarding the functionality of the DSS and how it has to
be operated.
4. 3. 2 ~DSS D
This is the software used to design the specific DSS [16]. This is essentially an
application development package [16].
4. 3. 3 DSS Source
DSS Source is the software tool used to design either the DSS Developer or thc Specific
DSS itself [16]. DSS Source is usually a programming language environment.
22
5. RISK
5. 1 Introduction
The Merriam-Webster Collegiate Dictionary defines risk as "possibility of loss or
injury" [22]. Risk in process safety management has a mathematical definition and is
defined as [13]:
Risk = Consequence * Probability of Consequence
The purpose of using a mathematical definition for risk is to speak about potential
hazards m the process industry in the language of numbers. The use of numbers to
quantify and compare risk gives a more accurate picture of the hazards in the plant. Risk
assessment involves calculating the risk as per ibis definition [12].
The nsk calculated using the above relation is time independent. This means that this
definition of risk does not differentiate between immediate and long-tenn risks. An event
that will occur the next moment will have the same nsk as an event that will occur the
next year provided the consequence and probability of consequence are the same, But,
an immediate risk will require expeditious decision- making. On the other hand, a long-
tetrn decision can be made after more investigation and at a laLer Lime. Thus, it is
23
imperative that decision makers use their judgment to differentiate the immediate risks
from long-term risks.
A textbook definition of immediate risk for chemical process safety is " a risk associated
with immediate effects of episodic events such as fire, explosion, and toxic material"
[23]. Decisions for this type of risk have to be urgent. This requires that decisions need
to be supported by tools that can process data and information faster than a human.
The application of decision support systems is ideal for such rapid risk situations. The
rapid risk situation requires analyzing information in a prompt manner. It cannot be
expected from a human being or a group of human beings to do hazard evaluation / risk
assessment / analyze previous chemical incident information in an urgent risk situation.
Therefore, decision support systems for such sttuations are needed.
As mentioned earlier, process safety management decisions require three types of
information a) hazard evaluation studies, b) rtsk assessment, and c) previous chemical
incident information. This thesis proposes the development of a decision support system
for rapid risk dectsions using the previous chemical incident information.
5. 2 Risk Decision-Makin
The decision-making under risk conditions can be divided into multiple steps [23]. These
steps are [23]:
24
1. Define the Risk Situation
2. Consider alternative decisions to solve the problem.
3. Select a decision tool to screen the alternatives.
4. Calculate risk associated with each alternative.
5. Select an alternative with the lowest risk.
6. Implement the decision.
Let us study each step in detail.
Ste l. Define the risk situation
This first step involves determining the risk situation that requires decision-making. For
example, an unexpected deviation in the process conditions may bc a rapid risk situation
that will require decision-making. On the other hand, prospective hazards determmed
during hazard evaluation studies and risk assessment analyses are long-term risk
situations [23]. These long-tetm risk situations might. not require an immediate decision
but a decision is nevertheless required. The long-term risk situations must not bc given
second priority. This thesis proposes a DSS that provides decision support for rapid risk
si tu ati on s.
25
Ste 2. Generate Alternative Decisions to Solve Problem
This step involves establishing alternatives to solve the risk situations. There are many
different ways to solve a risk situation. The alternatives selected in this step should cover
a wide variety of solutions [23]. All possible alternatives should be included irrespective
of their feasibility [23]. The selection of alternatives can be accomplished through a)
brainstorming in groups, b) using expert advice from safety consultants, c) using advice
from experienced operators in the company, or d) using previous incident scenarios as
examples. Thus, a wide spectrum of opinion must be gathered in this step [23]. This
thesis proposes the development of decision support system that generates alternatives
exclusively from previous chemical incident information.
Ste 3. Select a Decision Tool
There are numerous decision tools available to evaluate the different alternatives. This
step involves selecting an appropriate deciston tool to analyze the alternatives. The
selectton of the decision tool depends upon the a) type of problem we are solving and b)
the person who is solving it.
Different risk situations require different decision tools. For example, a cost-benefit
analysts might be ideal for analyzing a financial risk situation, but such an analysis tool
26
may not be suited to a process safety situation. Cost might not be a top priority in
maintaining safety of the working environment.
Decision makers in different capacities have different priorities. A manager in the
company will have different criteria for making a risk decision as compared to a safety
expert. Each person tends to make a decision that justifies his/her responsibility to the
job.
This thesis explores the different dectsion tools that can be used for process safety
dectsions. Finally, one decision tool is selected and analysis of the prevtous chemtca)
incident information is done using that decision tool.
Ste 4. Select the alternative with lowest risk
Once thc decision atd has been selected, each alternative will be screened using that
decision aid [23]. The screening process will involve calculating thc risk associated with
each alternative [23]. The calculation of risk will involve using data and information.
The source of thts data wil) be a) hazard evaluation studies, b) risk analysis, or c)
previous chcmtcal incident information. The alternattve for which the risk is lowest will
be the ideal one [23]. This thesis explores the selection of rapid risk situation alternatives
using the previous chemical incident information.
27
Ste 5. Im lement the decision
In the final step, the safety experts must implement the alternative with the lowest risk. It
might be possible that a group of alternatives are selected instead of a single alternative
[23]. This step involves taking concrete steps to execute the decision alternative selected,
To cite an example, in a rapid risk situation such a concrete step may be the shutdown of
the plant unit and evacuation. In case of a long-term decision, the alternative may be the
construction of safety barriers or redesigning of the entire process.
28
6. DECISION AIDS
6. 1 Introduction
Decision aids are at the heart of the decision support system. The decision aid
determines the algorithm used to screen alternative decisions in case of a risk situation.
Therefore, the selection of the right decision aid is imperative for a robust decision
support system. In our case, the right decision aid will be one that offers the functionality
to screen alternatives based on process safety principles. This thesis stuches various
decision aids before selecting one for the proposed DSS.
Some of thc popular decision aids are:
6. 2~G» Th
The game theory decision aid is used under conditions of competitive decision-making
[23]. For example. if there are stakeholders in a situation with conflicting interests, the
decision scenario becomes one of game theory [23]. John Von Neumann first introduced
the game theory in his hook 'The Theory of Games and Economic Behavior* [24]. Game
theory has been applied to situations ol' strategic decision-making [24]. For example,
geopolitics involves using the game theory decision-making paradigm. There are various
competing stakeholders in the geopolitical scenario. At the same time, some stakeholders
29
have mutual interests and some have opposing priorities, The formation of alliances and
associations under such a competitive situation requires application of game theory.
The application of game theory to rapid risk decision-making is limited because a rapid
risk decision scenario does not usually involve competing stakeholders. Thus game
theory does not justify the requirements of a DSS for rapid risk decisions.
6. 3 Mathematical Pro rammin
Mathematical programming is a powerful decision-making tool for optimization
problems in engineering and science. In this decision aid, there is a particular variable
that is maximized or minimized v hich is known as the objective of the problem [23].
The dcvclopmcnt of this decision aid begins by creatmg a mathematical model
illustrating the dependency of the variable to be maximized/minimized on other
variables [23]. The values of these variables are constantly changed until the objective is
achieved. Once the optimal solution in terms of the variable values has been determined,
the decision can be made.
This decision aid is used in situations where there is absolute clarity regarding the
objective of the problem [23]. In case of risk situations, there arc various vanables that
need to be controlled. The optimal value for these variables depends upon the a) decision
criterion of ihe user of the system and b) the particular risk situation. For example, a risk
situation decision might require a trade-off between process safety and cost incurred. In
this case, the decision to take a particular course of action depends upon the judgment of
the decision-maker. Although, safety must be maximized there are other variables that
need to be given priority as well. The objective of the problem in case of risk situations
is unclear. Therefore, the use of mathematical programming is not suited to solve such a
problem.
6. 4 Goal Pro rammin
Goal programming is similar to mathematical programming [23]. The difference
between the two is that goal programming serves to justify many objectives [23]. For
example, a goal-programming problem might involve not only maximizing/minimizing
one variable but also making certain that a second variable remains above a certain value
and a third variable remains constant. Thus, goal programming is more intensive in its
approach and can handle more complex problems as compared to mathematical
programming.
6. 5 Com romise Pro rammin
This decision aid is an extension of mathematical programming [23]. Compromise
programming involves treatment of various objcctivcs simultaneously [23]. The
objectives in this case mighl. be Ihe maximization of diffcrcnt variables [23]. At the same
time, there might be a certain priority in which the variables have to be maximized [23].
Therefore, the user of the system must communicate to the programmer the "weights" he
wants to be given to the different objectives [23]. The objectives of the users may be in
conflict with one another. The compromise-programming algorithm must have the
functionality to include the user's preferences.
This particular decision tool has been used in framing a policy for aquaculture
development [25]. Aquaculture development has a number of factors that the policy
maker must consider, e. g. , earnings from the aquaculture business, food production
levels, and contammation [25]. The maximization/ minimization of these objectives
must be treated in a mathematical model as per the "weights" assigned to each objective
by the user [25]. Such a program was mstrumcntal m decision suppot1 to the policy
maker for aquaculture development [25].
6. 6 Cost Benefit Anal sis
Cost benefit analysis (CBA) is one of the most popular decision aid tool used in the
industry. It has been extensively used in the financial industry to screen alternatives. The
genesis of the cost-benefit analysis was in the 1900s when there was a requirement to
study thc feasibility of infrastructure projects [23]. In this decision aid, each alternative
is evaluated according to the benefits it accrues versus the cost that is spent in
implcmcnting thai alternative [23].
32
The cost and benefits in this decision aid are strictly represented in monetary terms [23].
In some cases, the costs and benefits are non-monetary entities [23]. The conversion of
these non-monetary entities into dollar value can be a challenging task [23]. Also, the
monetary value of entities changes with time [23), which must be accounted in the cost
benefit analysis as well [23].
Cost-benefit analysis is a highly objective decision tool [23]. The inputs to the decision
tool are objective entities and there is no room for subjectivity of the decision maker
[231.
This particular decision aid is not suited for process safety management decision-
maktng. One of the challenges in applying CBA to PSM is that the benefits of a safer
process plant cannot be represented conclusively in terms of dollars. For example,
implementing a PSM decision might result in avoidance of an unknown incident. But if
the incident has not occurred at all then the prospective property damage cannot be
calculated. Today simulation software is heing developed to account for damage done
during an incident. These programs simulate the occurrence of an incident before it
occurs. Snll, the number of ways an incident can occur are infinite. Therefore,
calculating the monetary benefit of avoiding an incident is an impossible task. Also,
incidents involve loss to human life. Putting a price on the human life can be a highly
controversial practice.
33
6. 7V~ti M th 1
This decision aid is one of the simplest tools used to make decisions [23]. In the simplest
form of voting, the choice of one alternative from the many alternatives available is
made by having a vote among the participants [23]. The alternative that gets the
maximum number of votes is chosen as the course of action [23]. The fairness of this
type of a voting method is debatable [23]. For example, it is possible that the selected
course of action may have the vote of less than the majority of the participants [23].
Beside this simple voting method, there are many other different ways of voting that can
facilitate decision-making [23]. To cite an example, there is a modified form of voting
method in which the participants offer their votes by ranking the various alternatives in
order of preference [23]. Points are given to each alternative according the position in
which they are ranked [23]. The alteniative obtaining the least number of points is the
selected course of action [23). Besides this modified form of voting, there are many
other voting methods [23]. One of thc main benefits of the voting method is that it is
quick and easy [23].
This decision aid cannot be applied to process safety management decisions. The reason
for its non-applicability is that it is subjective, and its cffcctivcncss depends upon the
personal judgment and experience ol' the participants. Such a subjective decision aid can
result in a biased decision.
34
6. 8 Wei hted Scorin Method
The weighted scoring method is one of the powerful decision aids available. This
method is applied when there are multiple criteria for decision-making [23]. The
selection of the various criteria has to be done by the user [23]. Each criterion has its
own weight [23]. This weight of the criteria has to be decided by the user of the decision
aid [23]. The weight of the criterion can be determined by an analysis of the priority of
the person making the decision [23]. Once all the criteria for decision-making have been
determined, each alternative is weighed against the various criteria and points are given
for each criterion [23], These points are multiplied by the corresponding weights and
then the weighted points are summed for each alternative [23]. The alternative that gets
the maximum points is the chosen course of action [23].
There are three different variations of the weighted scoring method [23] namely a)
Kepner-Tregoe (KT) Decision Analysis, b) Analytical Hierarchical Process, and c)
Simple Multiattribute Rating Technique (SMART) [23]. In the KT decision analysis
process the decision ciitcria are divided into "musts" and "wants** [23]. The criteria
marked as "musts" have to be satisfied by the altcmative [23]. Thc "wants'* are given
weights in a lashion similar to the weighted scoring methods [23]. Each alternative is
weighed against the criteria marked as "wants*' and points are given for each criterion
[23]. These points are multiplied by the respective weights of all criteria [23]. Finally, all
the calculated points for each criterion marked as "wants*' arc summed [23]. The
35
alternative getting the maximum number of points AND satisfying all "musts" criteria is
the selected course of action [23].
The Analytical Hierarchical Process (AHP) is a modified version of the simple weighted
scoring method [23]. In this decision aid the decision criteria are structured in a
hierarchical fashion [23]. Thus, there is a top, middle, and bottom level of hierarchy for
the decision criteria [23]. Once the hierarchy has been established, the criteria in the
bottom level are given weights [23]. Each decision alternative is weighed against each
criterion and points are given [23]. Finally, all the respective points are multiplied by
their corresponding weights and summation is done [23]. The altetnative that has the
greatest number of points is the selected course of action [23].
The AHP decision aid has been used in the selection of multimedia authoring system
[26]. Multimedia authoring systems (MAS) are used to develop multimedia information
systems (MMIS) [26]. There are many MAS available in the market today [26]. The
selcctton of the apt MAS depends upon diffcrcnt criteria such as user requirements and
technology compatibility [26]. In the case study undertaken, a group decision was
required and AHP was the chosen decision aid [26]. The criteria were divided into four
levels [26]. The criteria at the bottom level were given weights and three alternative
products werc compared [26]. This case study exhibits the robustness of the AHP
decision tool for group deci sion-making [26].
36
The Simple Multi-Attribute Rating Technique (SMART) is another modified version of
the weighted scoring methods [23]. In this decision tool, the various alternatives are
studied and their atmbutes are noted [23]. The attributes of the alternatives are then
listed in order of importance [27]. The relative importance of the different attributes is
quantified by taking ratio estimates [27]. These ratio estimates then act as the weights for
the scoring process [23]. Next, all the alternatives are compared against each other on
basis of these "relatively important" attributes using the weighted scoring method [23].
The alternative that accumulates the greatest score is the selected course of action.
6. 9 Scrccnin Rankin Methods
This decision tool is a simple decision aid used in the industry today [23]. This method
ol decision-making works by eliminating the alternatives that do not satisfy the
minimum criteria set by ihe decision maker [23]. Once some of the alternatives have
been eliminated, the remaining alternatives are screened on the basis of a single criterion
[23]. The downside of this decision aid is that it can easily result in a fallacious decision,
since aH decision criieria are not treated comprchcnsively [23].
37
6. 10 Nominal Grou Techni ue
This decision aid is used essentially for group decision-making [23]. The decision-
making process involves brainstorming sessions in a group of experts and professionals
[23] participating expert is asked to contemplate and generate solutions by
himself/herself for the problem under discussion [23]. Different ideas contemplated by
the participants are documented and these ideas are then open for discussion within the
group [23]. During this discussion, the alternative courses of action for solving a
problem are considered [23].
Disputes among different ideas are discussed [23]. Once, all the alternatives have been
studied and decided upon, a secret vote is camed out [23]. The participating people vote
for the various alternatives in order of personal preference [23], This voting results in the
elimination of some alternatives [23]. The remaining choices are thrown for discussion
again and finally one alternative course of action is selected [23].
6. 1 l Pa off Matrix Anal sis
The payoff matrix analysis is one of the most robust decision aids that have been in use
for the past many years [23]. This decision tool is in the form of a matrix [23]. The
columns of the mattix represent the possible alternatives and the rows represent the
prospective outcomes [23]. Thc intcrsectton of thc columns and rows are populated by
38
values assigned by doing a thorough decision analysis [23]. At the same time, each
possible outcome is accompanied by a probability of those outcomes [23].
The mrdn advantage of the payoff matrix analysis is that it is capable of treating
uncertainty along with the value of the possible outcomes [23]. There is no software
programming required to implement this decision aid [23]. Therefore, its application is
easy and cost-effective [23].
39
7. CHOICE OF DECISION AID
7. 1 Introduction
After studying the various decision aids, it is required that we select a decision aid to
solve the problem facing us, that is, to use previous chemical incident information to
make process safety decisions. Selecting the right decision aid is a decision in itself.
Therefore, it a required that we systematically study the various decision aids available
and follow a methodology outlined by the CCPS Publication Tools for Making Acute
Risk Decisions with Chemical Process Safety Applications to make our decision [23].
This selection decision is extremely important for the efficacy of the DSS that we wtsh
to develop.
The following steps are recommended by the CCPS publication mentioned above [23]:
7. 2 State the Problem
This step involves a clear statement of the problem that we are trying to solve [23]. In
order to do so, we need to find out thc followmg [23]:
a) Resources that are available for so]vina the roblem 23: In our case our
resources are the previous chemical incident data, reactivity data, equipment
failure data, opinion from experts, database application tools, intellectual
40
property in the form of books and journals, and industry guidance through
forums and symposiums. There is no time constraint for us to collect these
sources of data as long as this is done in a reasonable time.
b Problem Com lexit: Our problem is to use previous chemical incident
information in making process safety decisions. The problem is complex in
nature since most of the chemical incident data are qualitative in nature. To
develop mathematical or statistical models on qualitative data is an extremely
cumbersome task. Therefore, before we actually begin to use the previous
chemical incident data, we must convert its quahtanve nature to a quantitative
form.
A significant problem when dealing with incident information is the fact that in
some cases the informatton is missing. There can be no decision support based
on missing information. Therefore, we need to find ways to tackle this problem.
Another aspect of the problem complexity is the fact that we wish to develop a
DSS for rapid risk decisions. This implies that such a DSS would be useful to
operators and maintenance personnel under acute risk situations. The time period
of response for such a system must be extremely small. Therefore, indexing of
the incident data and related reactivity and equipment failure data must be
according to the "best practtccs" of RDBMS. Such indexing ol' the data based on
'best practices" will enable cxpcditious data retrieval and shall reduce the
response time considerably.
Lastly, it must be emphasized that we are attempting to develop a DSS based
solely on incident information. Incident information is only one of the many
criteria upon which safety decisions are based. Since, we are limiting ourselves
to incident information, we must extract the maximum information from this
source so that it compensates for the risk assessment analysis and hazard
evaluation studies that are not considered.
c) Grou Decision Makin: This factor determines whether the decision to be made
has to be done in a group or by the single person. In our case, we establish that
rapid risk decisions do not have the response time for group decisions. Group
decisions will be needed where decisions for long-term safety polices are desired.
In case of rapid risk decisions, the response time for a decision is extremely
small. Therefore, we conclude that the decision maker should be a single person.
d) uantification of Data: This is an important factor in our case. As mentioned
before, the incident information is highly qualitative. Therefore, we need to
detetmine criteria to quantify this information.
7. 3 Identif Distin uishin As ects of the Problem
In this step we determine the main aspects of our problem [23J. We have already stated
the different aspects of our problem, c. g. , resource availability, problem complexity,
group decision-makmg, and need for quantification [23]. Thcsc arc thc main aspects of
42
our problem. The choice of the appropriate decision aid depends upon whether the
chosen decision aid can help to resolve these issues during decision-making,
7. 4 Stud of Decision Aids
This step involves studying the different decision aids in detail. Since we have already
done that, we don't need to elaborate on that issue. The different decision aids must be
evaluated by the following criteria. Each criterion corresponds to the issues we had
considered while analyzing our problem [23]:
a) Resource Re uirement: Each decision aid has its own resource requirement [23].
Some decision aids require greater mput that others in terms of data and
mformatton, time, and money, We must compare our resource availability from
Step I and then make a decision regarding selection of a decision aid [23].
As mentioned in Step I, thc resources available to us are previous chemical
incident information, intellectual property in the form of books and journals,
database application software, and expert advice. Thus, we conclude that the
resources available to us are extensive. There is no limitation on the availability
of resources
b) Problem Com lexit: The depth of complexity of a problem decides which
dectston aid is most suited for solving that problem [23]. In our case, we have an
extensive availabtlity of data. Also, the relationship between various entities of
43
the data is known, e. g. , relation between incident data and reactivity data. One of
the main complexities we face while dealing with the chemical incident
information is that the information is qualitative in nature. We must develop
criteria to convert the qualitative nature of the information into a quantitative
one. The various other complexities of our problem have already been stated in
Step l.
We conclude that our problem is of moderate complexity. The weighted scoring
method is highly recommended for problems that are moderate in nature [23].
c) Grou decision-makin: In our case, we have already stated that our decision aid
will be used mostly by a single entity. Therefore, we don*t need a decision aid
that rcquircs group decision-making. Thcrc arc very few decision aids that
provide for single entity decision-making. Therefore, we shall conccntratc on
decision aids that provide for decision-making in a small group of people. This
might be the case in a rapid risk decision where the DSS might be used by a
group of people rather than a single person.
quantify information and to what extent [23]. Since our information i» highly
qualitative in nature, wc require a decision aid that can provide moderate to
extensive quantification.
7. 5 Com arison of Problem Characteristics and Decision Aids
This step involves comparing the problem requirements with the decision aid
characteristics. We need to match our problem with the most appropriate decision aid. In
order to do so, it is recommended that we determine the problem class under which our
problem falls [23].
The following problem classes are the recommended classes for consideration
a) Ra id Sim le Decision: These are decisions for low-complexity
problems [23]. Also, the resource availability in this case is low [23].
There is no requirement for group decision-making and mostly data are
quantitative [23]. Since this problem class does not describe our problem,
we reject this problem class.
b) A Ra id Grou Decision: This type of decision is made when the
resources available are low and the decision has to be made in a group
[23]. Since, we are working to develop a DSS for use by a single entity,
wc shall relect the use of this decision aid.
c) A Lon -Term Grou Decision: In this case the resources available for
decision analysis are plentiful, but the decision has to be made in a group
[23] and therefore we rcjcct this problem class as well.
d) A Ra id Sin Ie-Entit Decision: In this problem class, the resources
available are low, the quantification requirement is high and the decision
45
aid has to be used by a single entity [23]. This problem class seems
appropriate for our requirement.
e) A Ra id uantitative Grou Decision: As the name suggests, this type of
problem class involves participation of a group of people [23]. Therefore,
we reject its use.
A Lon -term uantitative Decision with Plentiful Resource Availabilit:
This problem class deals with situations that involve long-term strategic
decisions [23]. There is a large resource availability to solve such
problems [23]. The time needed to deal with these problems is not a
constraint [23]. Our proposed DSS facilitates decision making for rapid
risk decisions. Thcrcfore, this problem class is rejected.
In the end we conclude that the rapid single-entity problem class is tdeally suited for our
needs. The next step is to look for decision aids that qualify for this problem class.
7. 6 Ra id Sin lc-Entit Decision Aid for Chemical Incident Information
After studying decision aids available for decision-making, this thesis proposes to use
the weighted scoring method as a decision aid for decision analysis. The weighted
scoring method is ideal for facilitating decision-making where multiple decision criteria
exist and each decision criterion has its own "weight". Also, the quantification
requirement for our DSS is quite high. Therefore, we need a system that can help us
46
quantify the qualitative information to a reasonable extent. The weighted scoring method
helps us to accomplish this.
47
8. DECISION CRITERIA
8. 1 Introduction
We need to determine the various decision criteria for the chosen decision aid, that is,
the weighted scoring method. In choosing the different decision criteria we have studied
the consequences of incidents. In the decision support model that we wish to utilize, the
consequences of the previous chemical incidents help us to assess the potential risks of
similar future tnctdents. Once the risks have been assessed based upon the previous
incident data, we can take measures to avoid thc prospective circumstances leading to
those risks.
The decision criteria are derived directly from the incident information that has been
collected m incident databases. Most of the incident databases already have the
inl'ormation required for quanttftcation of the decision criteria. In the cases where the
information is missing, it serves to guide developers of future incident databases to
capture particular information that can be used for thc specified decision criteria of the
DSS.
We have selected six decision criteria for the decision support system. Before we
analyze the six criteria, we must know a few facts about the criteria selected.
48
Each criterion has two quantities attached to it: I) index and 2) weight. The index of the
criteria is the value given to that criteria based upon severity of similar previous
incidents that have occurred. Each criterion is given an index of I to 10. The greater the
index, the more severe are the consequences of that incident for that criterion.
The weight of each criterion is the "importance" that the decision maker gives to that
criterion. We feel that this value of the "weight" must be user defined, which allows the
DSS to become user specific. This means that the DSS can be used by people having
different "weights" for the various decision-making criteria. The sum of "weights" given
to the six criteria must equal 100.
8. 2 Decision Criteria for Chemical Incident Information
The following are decision criteria sclcctcd for the decision support system:
8. 2. 1 Environmental Impact
This criterion determines the environmental impact of a chemical incident. Depending
upon the severity of the chemical incident, each incident has been given an
environmental impact index (EII). The greater thc environmental damage, the higher the
EII. The EII depends upon different factors, which have been derived directly from the
previous chemical incident information. Each factor has been given a certain 'weight"
49
depending upon the "seriousness" of that factor. The sum of the weights given to each
factor equals 100. The factors chosen to determine the EDI are:
a) Number of On-Site Fatalities: In this case, the number of on-site fatalities
determines the severity of the incident. N, „, ;„r, ~sa„ is the designated
symbol for this index. This index has been given a weight of 0. 3. The
index has been given values according to the following criteria in Table
Table l. On-Site Fatality Index Calculation
Number of On-Site Fatalities N on-sac fatasscs
=0
&1&(=3
&3 10
b) Number of Off-Site Fatalities: This criterion is similar to the criteria
mentioned above. The difference between the two factors is that the off-
site fatalities imply greater incident severity. Therefore, the chosen index
for this criterion should be stricter. This factor has also been given a
"weight** of 0. 3. The chosen symbol for this index is
N, rs, . u, a„sa„and the following Table 2 lists the values of the chosen
index.
50
Table 2. Off-Site Fatality Index Calculation
Number of Off-Site N oa-site fatalities
Fatalities
=0
10
c) Number of On- Site Injuries: This factor accounts for the on-site injuries
that occurred during a chemical incident. The "weight" given to this
factor is 0. 1. The symbol used for this factor is Non „«, „„„n„and the
following Table 3 lists thc values given to this index.
Table 3. On-Site Injuries Index Calculation
Number of On-Site Injuries Non-sile intones
) land&=3
) 3 and&=5
)5&&se7
10
d) Number of Off-Site Injuries: This criterion has a "weight" of 0. 1 and
determines the severity of an incident based upon the number of off-site
injuries. Table 4 lists the indices given to various situations.
Table 4. Off-Site Injuries Index Calculation
Number of Off-Site Injuries N off sire injuries
=0
&I &&=3
&3&&en5
&5 10
e) Toxicity of Chemical Relcascd: The toxicity of the chemical released is
an important factor to determine the severity of the environmental impact
of an incident. The toxicity of the chemical released must be determined
using the chemical reactivity database. Thus, a relational database link
between the incident database and chemical reactivity database is
proposed.
Thc variable used to determine the toxicity of the various chemicals is
Lethal Dose (LDqts) value [13J. LDqtt for a chemical is defined as a level
ol dose that causes death to 50% of thc population exposed to the dose
52
[13]. Therefore, lower the value of LDss, greater is the toxicity of the
chemical [13]. Based on this fact, we give a Toxicity Index (TI) to each
chemical. The Hodge-Sterner Table assists us in determining the TI for
each chemical. The Hodge-Sterner is provided in Table 5 below along
with the corresponding values for the Toxicity Index.
Table 5. Hodge Sterner Table [13] with Toxicity Index Calculation
LDqs Value Degree of Toxicity Toxtctty Index (TI)
& 1. 0 mg Dangerously Toxic 10
1. 0 — 50 mg
50 — 500 mg
Seriously Toxic
Highly Toxic
10
10
0. 5 — 5 gm Moderately Toxic
5 — 15 gm Slightly Toxic
& 15gm Extremely low toxicity
Thus, depending upon the category each chemical falls into, we shall asstgn a value for
Tl to that chemical. This value will be a calculated field and will be stored in the table
containing reactivity informatton of the chcmtcals. Thc "wetght'* assigned to the
Toxicity Index is 0. 2.
The following Table 6 summarizes the different criteria and their respective "weights"
for determining the Environmental index.
53
Table 6. Criteria for Determining Environmental Index
Criteria Symbol Weight
On-Site fatalities N on-site fatalities 0. 3
Off-Site Fatalities N off-site fatalities 0. 3
On-Site Injuries N on-mte mjanes 0. I
Off-Site Injuries
Toxlclty
N off-site injnrics 0. 1
0. 2
The Environmental Index is calculated by using the formula below:
EI = (N on site faialines ' 0 3) + (N off sne famliiies
' 0. 3) + (N on site nijoncs "' 0- l ) + (N ott site mlaiies
0. 1) + (Tl " 0. 2)
The value of El obtained above will be based on a scale of 10. The greater thc scvcrity of
the incident, the closer the value of EI to 10. Also, this number will be unique to each
incident and the El value will be stored in the incident database. Its use will be explained
in the subsequent sections of this thesis.
54
8. 2. 2 Dollar Damage
This criterion determines the dollar damage during occurrence of a chemical incident.
This is an extremely objective criterion, and generation of index values for it is a
relatively easy task. The dollar damage of property is defined as the dollar cost of the
physical assets destroyed or damaged during occurrence of a chemical incident. Such an
assessment is usually performed after the incident has occurred and we propose that this
information be stored in the incident database.
The following Table 7 exhibits the dollar damage index values for various dollar damage
scenarios.
Table 7. Dollar Damage Index (DDI) Values
Dollar Damage (in 1)SD) Dollar Damage Index (DDI)
Null
( 1, 000
1, 000 — 10, 000
10, 000 — 100, 000
100, 000- 500, 000
500, 000 — 1, &N)0, (y00
1, 000. 000 — 10, 000, 000
& 10, 000, 000 10
55
The value of the DDI is stored in the incident database along with the actual dollar
damage value.
8. 2. 3 Litigation Cost
Litigation cost is a significant burden on the industry after an incident has occurred.
Litigation cost is defined as the total dollar value of the cost incurred during various
post-incident incident-related litigation cases. We propose to inculcate this criterion in
decision-making for future rapid risk decisions. The generation of a Litigation Cost
Index (LCI) is proposed in Table 8 below.
Table 8. Litigation Cost index (LCI) Values
Litigation Cost (USD) Litigation Cost Index
Null
( 1, 000
1, 000 — 10, 000
10, 000 — 100, 000
100, 000 — 500, 000
500, 000 — 1, 000, 000
1, 000, 000 — 5, 000, 000 9
& 5, 000, 000 10
56
8. 2. 4 Employee Disenchantment
It has been observed that the employee morale is extremely low after an incident has
occurred. Employees leaving their jobs for other careers exhibit this disenchantment. We
propose to record this disenchantment by generating an Employee Disenchantment
Index (EmDI).
Disenchantment is a qualitative factor. We wish to quantify this factor by analyzing the
number of employees in proportion to the total number of employees leaving the
particular company for alternative careers within three months of the incident
occurrcncc. Thc chosen values for the Employcc Discnchantmcnt Index are given in
Table 9
Table 9. Employee Disenchantment Index (EmDI1 Values
Propotiion of Employees leaving
the Company within 3 months of
Employee Disenchantment. Index
(EmDI)
incident occurrcncc
Null
&= 0 and (0 01
10
57
8. 2. 5 Government Action
Every chemical incident occurrence is usually investigated by a government agency. If
in the course of the investigation it is observed that there was non-compliance with a
government regulation, the government might choose to penalize the company for the
lapse. The degree of punishment depends upon the level of non-compliance.
We propose to develop a Government Action index (GAI) to quantify this criterion. The
following Table 10 exhibits thc development of this index.
Table 10. Government Acl. ion Index Values
Goverrunent Action Government Action Index iGAI)
Null
A Simple Warning
Penalty & l0, 000 USD
Penalty &= 10. 000 USD & & 20, 000 USD
Penalty &= 20. 000 USD & & 30, 000 USD
Penalty &= 30, 000 USD &. & 40, 000 USD
Penalty &= 40, 000 USD 10
58
8. 2. 6 Company Disrepute
The occurrence of an incident results in company disrepute in public opinion. This
criterion is highly qualitative, so this thesis proposes the quantification of this qualitative
information.
We quantify this information by establishing the geographical extent to which the
incident news is reported. The following Table 11 illustrates the quantification of this
tnformation in terms of a Company Disrepute Index (CDI).
Table 11. Company Disrepute Index Values
Incident News Reporting Company Disrepute Index
Null
Local News
Regional News
State News
National / International News 10
59
8. 3 Calculation of Consolidated Decision Index
Each of the decision criteria selected have their own significance in the eyes of the
decision maker. The decision criteria are assessed on a scale of 10 points. The "weight"
that each decision criterion will be given depends upon the perception of the decision
maker. This thesis assumes that every person has his own priorities in making decisions.
The DSS being developed will have the robustness to capture the personal decision-
making criteria of each decision maker. At the same time, to expedite the process of
using the DSS, default values are assigned to the "weights" in the Decision Support
Form. Such functionality is introduced to expedite the process of using thc DSS.
The following Table 12 displays the methodology of calculation of the consolidated
decision index:
Table 12. Calculation of Consolidated Decision Index
Weight Decision Criteria Decision Decision Decision
Alternative- l Alternative-2 Alternative-3
Wt Environmental Mt
Damage
W2 Dollar Damage Ma
Ws
W4
Litigation Cost
Government Action 14 J4
Ms
M4
Ws Employee Morale Ms
Ws Company Disrepute
Con soli dated OO, ~O, OO, oO,
Decision Index
9. THE DECISION SUPPORT SYSTEM
To illustrate the use of the decision support system, we shall undertake a simulated case
study. This case study will help the reader understand the working of the decision
support system based on chemical incident data.
9. 1 Case Stud for Decision Su ort S stem
In this case study, we have taken data from the RMP*Info database. Some data items
that we need are not captured at all in the RMP*Info database. For these data items that
are not captured by the RMP'"Info database, we have made the following two choices I)
entered *Information Not Available* or 'Null * for these fields or 2) assumed a reasonable
value for that field. We shall use the superscript 'assumed' for assumed values. In either
case, thc data fields that are not captured by the RMP*Info database have been marked
with an asterisk.
We choose a rapid risk decision scenario for our case study. Let us assume there is an
overpressurc scenario mside a reactor of type R. The primary chemical inside this
reactor is C. The operator observes the ovcrpressure inside the vessel indicated by the
pressure gauge. At that point, he is required to make a rapid risk decision, He wishes to
make his rapid risk decision based upon previous chemtcal incident data only.
62
A decision support system proposed by this thesis is tailor-made for such a rapid risk
situation. The following steps illustrate how previous incident data will be used to
facilitate this decision.
Ste 1. Database Modelin
1. Firstly, we determine what data are required for facilitating decision-making.
These are the decision criteria we have mennoned above.
2. Then we need to determine how these collected data must be stored. Each
data field his placed in the appropriate table.
3. Finally, the relationship among the different tables has to be established and
inculcated in the database model.
In our case, we have determined four main entities that describe a chemical incident.
These entities are Incident Type, Chemical Involved, Equipment Affected, and Primary
Decision Taken. The attributes that defme the six indices we have chosen have been
mentioned in earlier sections. These attributes must be collected in the database before
the DSS can be used effectively.
The mam entities mentioned above, that is, Incident Information, Chemical Involved,
Incident Equipment and Incident Decision must be stored m separate tables. There must
be a relational model between these entitics, and thc database model suggested is based
63
on 'best practices' of RDBMS. This relational model helps to remove anomalies that
might occur during incorrect storage of the incident information.
Ste 2. Im lementin the Decision Aid Model
The Decision Aid Model involves development of the various indices mentioned in the
earlier sections. This model is programmed into the database application for processing
the incident information. The execution of the decision aid model is facilitated by user-
friendly forms and reports.
Let us apply the decision aid model for incident information involving chemical "C" and
reactor "R*' and an "Overpressure" incident scenario. We consider ten incident cases in
which the above-mentioned conditions existed. The incident decision taken during these
ten cases were different each time. We shall denote them by Dl, D2, D3. . . . . . DI 0.
1. Calculation ol'Environmental Index
Incident Chemical: C
Incident Equipment": R
Incident Scenario": Overprcssure
Both Incident Equipment and Incident Scenario are not explicitly captured by the
RMP*Info Database. These are assumed values. Table 13 lists values for on-site and off-
site fatalities and injuries taken from the RMP *Info Database.
64
Table 13. Calculation of Environmental Index
Accident
No
No of On-
site deaths deaths iujuries
No of off-site No of oa-site No of off.
site
injuries
Loss Dose Decision Taken'
Dl-Shutdown of Uait
D2- Shutdown of Plant
D3-Stop Steam Supply
D4-Decrease Steam
Supply
DS-Increase Stcam Supply
Dd-Stop Flow of Coolant
D7-Decrease Flow of
Cool'slit
DS- hicicasc Flog of
Coolant
D9- Contmue Noimal
Operations
10 D10- Evacuate Umt and
Release Pressure Valve
We observe that all values I'or falahlies and injuries taken from the RMP*Info Database
are 0. This table must be converted into an index table based on the decision model for
65
environmental index mentioned in the earlier section (see Table 1- Table 6). The index
table looks like the Table 14.
Table 14. Environmental Index for Case Study
Incident
No
Nee-ssse
dead s
N. rr. . s. N
seesedeees sllelsillsles
Toxtctty
Index
Environmental
Index
Decision
Taken
10 DI
10 D2
10 D3
10 D4
10 D5
10 D6
10 D7
10 Ds
10 D9
10 10 DI 0
66
2. Calculation of Dollar Dama e Index
Let us take the values given for the property damage in the following Table 15 and
create an index for dollar damage. The index is calculated using the dollar damage index
reference from an earlier section (see Table 7).
Table 15. Dollar Damage Index for Case Study
Incident No Dollar Damage Dollar Damage
Index
Decision
Dl
D2
7 000 assume
D4
DS
D6
D7
assume
D9
10 60, 000 assume D10
67
3. Calculation of Liti ation Cost Index
This index is calculated using the reference Table 16 for litigation cost (see Table 8).
Table 16. Litigation Cost Index for Case Study
Incident Litigation Cost* Litigation Cost Decision
No Index
Null Dl
110, 000"'" ' D2
50, 000"'" ' D3
4p ppp ussUluc D4
happ ppp USSUttlu D5
120 000 D6
Null D7
ppp CSSllmC D8
60, 000 uscumc D9
10 45 ppp Cecum Dl0
4. Em lo ee Disenchantment lndcx
The reference Table 17 for this index is given in an earlier section (see Table 9).
68
Table 17. Employee Disenchantment Index for Case Study
Incident Proportion of employees Employee
leaving company within Disenchantment Index
3 months"
Decision
0 assume Dl
0 assume D2
Nu11 D3
1 assume 10 D4
0 00 1 assume D5
0 002 asses)e D6
0 assume D7
0 assuulc D8
2 assume 10 D9
10 0 assulue D10
5. Government Action Index
This index is generaled in the following manner (see Table 101 in Table 18.
69
Table 18. Government Action Index for Case Study
Incident No Penalty* Government Action Decision
Index
Null Dl
20, 000 D2
2pp ppp assume 10 D3
60, 000 assume 10 D4
56 ppp assume 10 DS
24 ppp assume D6
60 ppp assume 10 D7
4 ppp assume D8
12 ppp assume D9
10 23 ppp assume DIP
6. Com an Disre ute Index
This tndex is calculated using the following index Table 19 (sce Table 11).
70
Table 19. Company Disrepute Index for Case Study
Incident
No
Media Coverage of
Incident*
Disrepute Index Decision
LOCal assume Dl
Information Not Available 0 D2
Local D3
Regional "'" ' D4
Regional D5
Local "'" ' D6
LOCal assume D7
International """ ' 10 Dg
International """ ' 10 D9
10 International "" ' 10 D10
Ste 3. Consolidation of Indices
Once the six indtces, that is I) environmental index, 2) dollar damage index, 3) littgation
cost index, 4) government action index, 5) employee dtsenchantment index, and 6)
company disrepute index, have been calculated we must consolidate these indices into a
single index that shall serve as a guide for dectsion making.
Each of the indices mentioned above correspond to certain decision criterion. Each
decision maker has his/her own priority and wishes to assign a particular "weight" to
each decision criterion. The DSS developed provides the functionality in which the user
can enter the "weight" that each decision criterion is given.
At the same time, we realize that a rapid risk situation might require a prompt decision.
For this reason, we have assigned default values to the "weights" of the decision criteria.
These values can be changed by the user at any time of the decision making process. The
following Table 20 exhibits the default values assigned to thc "weights'* of each decision
criterion.
Table 20. Default Value for "Weights" of Decision Criteria
Decision Criteria Default Weight %
Environmental Index 30
Dollar Damage Index
Litigation Cost Index 10
Government Action Index 10
Employee Morale Index 10
Company Disrepute Index IO
Taking these values for the specific "weights", the decision support system calculates a
final consolidated index for each decision taken based on Table 12 methodology. The
decision that has the lowest index is the most desirable are based on the selected decision
criteria. Table 21 below exhibits the calculation of the consolidated index:
Table 21. Consolidated Decision Index for Case Study
Incident No Decision Consolidated Index
Dl 1. 5
D2 2. 9
D3 3. 1
D4 3. 7
D5 3. 4
D6 3. 0
D7 2. 5
Dg 4. 7
D9 4. 1
10 D10
The consolidated index provides the decision maker support for rapid risk decisions. A
lower index decision implies greater compatibility with the decision maker's criteria and
73
therefore is the recommended course of action. In our case, Dl, that is, 'Shutdown of
Unit', is the best decision to take under the given scenario.
9. 2 Decisions Based on Missin Data
We have used the data from RMP*Info database to populate the fields in the tables of
our DSS. But it was observed that there are unavailable data in the RMP*Info database.
This presented a problem since the calculation of all decision criteria indices became
impossible. To resolve this problem, it was decided that the missing data would be
assigned a value of 'Null* in the database. Whenever the index generation algorithm
encounters the 'Null' value, it assigns a value of zero to that particular decision criterion
index. The final Decision Support Report presents to the user the Consolidated Decision
Index along with the value for the individual indices. Thcrcforc, a value of '0' for any
decision criterion index implies that that decision criterion was not included in the
calculation of the Consolidated Decision Index. The user of the system must be aware ol
this fact before using the DSS and the Decision Support Report directs the user to take
thts particular condition into account.
74
10. CONCLUSIONS AND FUTURE RESEARCH
1. The DSS proposed in this thesis is an intelligent system. It not only decides the
best decision from various decisions available but it also selects the decisions
that must be screened to arrive at a single decision. What this means is that the
user does not must provide the DSS with a set of decisions that he wants to
screen. The system itself selects the set of decisions. This is a powerful
functionality in the system. To use the system, the user must enter the
abnormality in his/her process. The system captures relevant data from the
previous incident database, chemical reactivity database, incident decision
database, and equipment database, These captured data is subjected to the
mathematical model of thc decision aid and a variety of decisions are screened.
ln this sense, the proposed DSS is an intelligent thinking system.
Thc mtegration of such a system with process control models can provide
effective process safety in the wake of an abnormality in the system. Future
research can focus on such mtegration.
2. In spite of thc robust functionality there are certain issues that must be examined
bel'ore using this system. Firstly, it must be mentioned that this system is based
solely on previous chemical incident infotmation. Therel'ore, any decision that is
selected will be one derived directly from analysis of chemical incident
information only. Such analysis will be devoid of a risk assessment or hazard
75
assessment study for the scenario. Therefore, it is recommended that such a DSS
be used in tandem with risk assessment decision support systems and hazard
analysis studies.
Future research can focus on the integration of such DSS based on chemical
incident information, together with risk assessment studies and hazard analysis
practices. Such an integrated system can provide better decision support than
using chemical incident solely.
3. Thirdly, for the DSS to be effective it is important that such a system be trained
first. Feeding as much data to the system as possible does the trains the system.
The larger the set of data provided to the system, the better will be the decision
support furnished to the user. Also, as the amount of data stored in these
databases increases, the system becomes more intelhgent in its decision-making
task. The increase in mtelligence is due to increase in "experience" of the system
in treating incident information. Therefore, the efficacy of the system m making
decisions depends greatly on the amount of data being stored. In fact, once
significant amount of previous incident information has been fed to the system,
the system becomes capable of "mimicking*' the human decision-making ability.
A negative side of this is that if previous incident data are completely absent for a
rapid risk situation, the system will be unable to provide decision support.
76
3. Lastly, it needs to be mentioned that this proposed DSS facilitates decision-
making based on the six specific criteria determined. It is assumed that there are
no other decision-making criteria. In the event that there are other decision
criteria that the user wishes to include, this DSS will must be modified to
inculcate these criteria in the decision aid. This modification will require
software programming and can be done only by a programmer. This limits the
use of the DSS in the sense that the user cannot choose the decision criteria he
wants to include besides the six criteria selected by the programmer,
Future research can provide the user functionality to increase the decision
criterion of the proposed DSS. Providing user-friendly wizards and tools to the
database application can help achieve this objective.
77
LITERATURE CITED
Kleindorfer, P. R. , Feidman, IL, Lowe, R. A. , "Accident Epidemiology and U. S. Chemical Industry: Preliminary Results from RMP*Info, " Working Paper 00-01- 15. Center for Risk Management and Decision Processes, The Wharton School, University of Pennsylvania. Revised March 6, 2000.
OSHA Preambles: Process safety management (29 CFR 1920. 119), 111. Summary
and explanation of the final rule, www. osha-sic. gov, Accessed: March 2002.
Mason, E. , "Elements of process safety management: Part I, " Chemical Health ifi Safety, 8, 4, pp 22-24, July / August 2001.
Mannan, M. S. , Makris, J. , and Overman, H. J. , "Process Safety and Risk Management Regulations: Impact on Process Industry,
" Encyclopedia of Chemical Processing and Design, Marcel Dekker, Inc. , New York, NY, 2002.
Guidelines for investigating Chemical Process incidents, CCPS Publication, New
York, NY, 1992.
AI-Qurashi, F. , Sharma, G. , Rogers, W. J. , Mannan, M. S. , "Application of relational chemical process safety databases for lowering mean failure rates, ' Process Safety Prot;res», 20, 4, pp 280-285, December 2001.
Vaughan, R. , Kelly, B. , CCPS Process Safety Incident Database (PSID), www. aiche. org. Accessed: Feb 2002.
Lortie, M. , Rizzo, P. , '*The classification of accident data, " Safety Science, 31, 1,
pp 31-57, March 1998.
Kouniotis, S. P. , Kiranoudis, C. T. , Markatos, N. C. , "Statistical analysis of domino chemical accidents, '* Journal of Hazardous Materials, 71, 1-3, pp 239- 252, January 2000.
10. Richard, P. W. , "Improved F/N graph presentation and criteria". Journal of Loss Prevention in Process Industries, 5, 4, pp 239-247, 1992.
11. Guidelines for Hazard Evaluation Procedures, CCPS Publication, New York, NY, 1992.
12. Chemical Process Quantitative Risk Analysis, CCPS Publication, New York, NY, 1989.
78
13. Crowl, D. , Louver, J. F. , Fundamentals of Process Safety Management with
Applications, Prentice Hall International Series, Englewood Cliffs, NJ, 1990.
14. Svedung, I. , Rasmussen, J. , "Graphic representation of accident scenarios:
mapping system structure and the causation of accidents, " Safety Science, 40, 5,
pp 397-417, July 2002.
15. Morton, S. M. S. , "Management Decision Systems: Computer Based Support for Decision Making,
" Division of Research, Harvard University, Cambridge, MA, 1971.
16. Sprague, R. H. Jr. , '*A framework for development of decision support systems, " Decision Support Systems: A Data-Based, Model Oriented, User-Developed Discipline, (Edited By: House, W. C. ), Petrocelli Books Inc. , New York, 1983.
17. Fleming, C. C. , Halle, B. V. , Handbook r&f'Relational Database Management, Addison Wesley Publishing Company, MA, October 1999.
18. House, W. C. (Edited By), Decision Support Systems: A Data-Bused, Model Oriented, User-Developed Discipline, Petrocelli Books Inc, New York, 1983.
19. Nasirin, S. , Birks, D. F. , "DSS implementation in the UK retail organizations: a
GIS perspective, " Information & Management, In Press, Uncorrected Proof, Accessed: March 2002.
20. Quaranta, N. , DeMartini, A. , Bellasio, R. , Bianconi, R. and Marioni, M. , "A
decision support system for the simulation of industrial accidents, '* Enviro&&mental
Modelling & Sofnvare, In Press, Uncorrected Proof, Accessed: March, 2002.
21. Konstantinos, G. Z. , Konstantinos, N. A. , Vasilakis, G. M, "A real-time decision
support system for roadway network incident response logistics, *' Transpr&r&ation
Researr h Part C: Emerging Technolr&gies, 10, I, pp 1-18, February 2002.
22. Mer&da&n-Webster Collegiate Dictionary, www. m-w. corn.
23. Tools fo& Making Acute Risk Decisir&ns with Chemical Process Safety Applications, CCPS Publication, New York, NY, 1995.
24. Game Theory: Ari Introductr&r) Sketch, http//: william-
king. www. drexel. edu/top/eco/game/game. httnl. Accessed: March, 2002.
79
25. El-Gayar, O. F. and Leung, P. , "A multiple criteria decision making framework for regional aquaculture development, " European Journal of Operational Research, 133, 3, pp 462-482, 16 September 2001.
26. Lai, V. S. , Trueblood, R. P. and Wong, B. K. , "Software selection: a case study of the application of the analytical hierarchical process to the selection of a
multimedia authoring system, " Information & Management, 36, 4, pp 221-232, October 1999.
27. Poyhonen, M. and Hamalainen, R. P. , "On the convergence of multiattribute
weighting methods, " European Journal of Operational Research, 129, 3, pp 569- 585, 16 March 2001.
80
Gaurav Sharma was born in Chandigarh, India on April 29, 1976 to Mrs. Chandra
Kanta and Dr. Gopal Krishan. He received his Bachelor of Engineering in chemical
engineering from the Department of Chemical Engineering & Technology, Panjab
University, Chandigarh, India in 1999. Gaurav joined Texas A&M University, College
Station, TX in the Fall of 2000 to pursue his Master of Science degree in chemical
engineering. His research was carried out under the supervision of Dr. M. Sam Mannan
at the Mary Kay O' Connor Process Safety Center. Gaurav will be employed by
Granherne Inc. in their Houston, TX office.
Permanent Address:
4500 Aberdeen Drive,
Amarillo, TX 79119
IIIIIIINlllll A14S29 74S'981