From: AAAI-96 Proceedings. Copyright © 1996, AAAI (www.aaai.org). All rights reserved.
Botond Virginas
University of Portsmouth, Department of Information
Locksway Road, Southsea
PO4 8.TFHampshire, United Kingdom
botond@sis.port.ac.uk
Introduction
Several accidents during the last decade have emphasized the
need to properly analyse and manage the low probability high
consequence risks associated with plant operations in the
nuclear, chemical and other industries. Formal risk assessment
methods are vital tools for this task, and include a number of
qualitative and quantitative techniques.
The increasing need to perform risk analyses induces a
growing interest in standardised analysis procedures and
corresponding computerised supporting tools.
The amount
and type of information handled in risk assessment calls for the
application of knowledge-based systems.
The research investigates how knowledge-based systems
techniques can be used in conjunction with conventional risk
assessment methods in order to develop a Knowledge-Based
Risk Assessor (KoBRA) supporting the various phases of risk
assessment.
‘The KoBRA environment takes the form of a
toolkit containing a variety of tools performing the different
tasks involved in risk assessment. The following research area
problems have been iden tificd:
A safety oriented process model needs to be developed for
a given type of process within a certain application
domain. Different knowledge representation
formalisms
have to be investigated.
Different reasoning strategies have to be explored for
the construction
of different logic models. A mixed
programming language approach will be examined.
A platform concept will be analysed.
The research concentrates mainly on:
Review of the risk assessment field
Evaluation
of the existing computer support for risk
assessment
Design of a risk assessment framework within an overall
risk management strategy
Design of an integrated set of computer aided risk
assessment tools
Implementation and testing of KoBRA
Review of the work completed
A general review of the risk assessment field has been
completed. In parallel, computerised tools supporting various
phases of the risk assessment process have been explored It
has been decided that an application domain from the-chemical
1374
SIGART/AAAI
Science
process industry will be chosen and the toolkit will be tested on
several particular processes within that domain.
The design of the risk assessment framework has been
completed.
A substantial
part of KoBRA
has been
implemented on a Sun SPARC workstation. The different risk
assessment tools are being developed and tested incrementally.
A mixed language programming environment (POPLOG) is
used for the implementation.
As the development of the different tools from KoBRA
continues, so a safety oriented process model is being built
incrementally. This model takes the form of an object oriented
database with functional and structural links between the
objects. Generic objects describe what is known in general
about a particular type of system or component. The user builds
his process description with process specific instantiations of
these objects. The objects are represented with a hybrid
formalism using rules and frames. The different tools in the
toolkit perform the v<ariousrisk assessment tasks on this model.
A platform concept will be used for testing KoBRA . A
pilot system will be built for one chosen process maintaining a
distinction between process-specific and process-independent
(but still domain-specific)
knowledge. The process-specific
knowledge will then be removed from the pilot system leaving
the knowledge that will form the basis of the platform. The
platform will then be used to build a subsequent pilot system
for another process.
References
1. Center for Chemical Process Safety 1989. Guidelines .for
Chemicd Process Qucrntitcrtive Risk Andy.&. New York,
N.Y.: American Institute of Chemical Engineers.
2. Apostolakis, G.E. ed. 1990. The Role and Use of Personal
Computers in Probabilistic Safety Assessment and Decision
Making. Relicrhility Enginc~ering X System Sqfety30. Elsevier
Applied Science.
3. Poucet, A. 1992. Knowledge Based Systems for Risk
Assessment and Monitoring. In Computer Applications in
Ergonomics, Occupational Safety and Health, 29-36. Elsevier
Science Publishers B .V.
4. Barett, R.; Ramsay, A.; and Sloman, A. 1986. Popll: A
Pructicul Lungutrge ji,r Art@kid Intelligence. Chichester,
England: Ellis Hot-wood Limited.