posted on 2010-11-19, 10:25authored byAndrew J. Trenchard
The development of computer based tools, to assist process plant
operators in their task of fault/alarm diagnosis, has received much
attention over the last twenty five years. More recently, with the
emergence of Artificial Intelligence (AI) technology, the research
activity in this subject area has heightened. As a result, there are a
great variety of fault diagnosis methodologies, using many different
approaches to represent the fault propagation behaviour of process
plant. These range in complexity from steady state quantitative models
to more abstract definitions of the relationships between process
alarms.
Unfortunately, very few of the techniques have been tried and
tested on process plant and even fewer have been judged to be
commercial successes. One of the outstanding problems still remains
the time and effort required to understand and model the fault
propagation behaviour of each considered process.
This thesis describes the development of an experimental
knowledge based system (KBS) to diagnose process plant faults, as
indicated by process variable alarms. In an attempt to minimise the
modelling effort, the KBS has been designed to infer diagnoses using a
fault tree representation of the process behaviour, generated using an
existing fault tree synthesis package (FAULTFINDER). The process is
described to FAULTFINDER as a configuration of unit models, derived
from a standard model library or by tailoring existing models.
The resultant alarm diagnosis methodology appears to work well
for hard (non-rectifying) faults, but is likely to be less robust when
attempting to diagnose intermittent faults and transient behaviour.
The synthesised fault trees were found to contain the bulk of the
information required for the diagnostic task, however, this needed to
be augmented with extra information in certain circumstances.
History
School
Aeronautical, Automotive, Chemical and Materials Engineering