Symbolic representation of knowledge for the development of industrial fault detection systems

In critical infrastructure, such as nuclear power generation, constituent assets are continually monitored to ensure reliable service delivery through pre-empting operational abnormalities. Currently, engineers analyse this condition monitoring data manually using a predefined diagnostic process, however, rules used by the engineers to perform this analysis are often subjective and therefore it can be difficult to implement these in a rule-based diagnostic system. Knowledge elicitation is a crucial component in the transfer of the engineer’s expert knowledge into a format suitable to be encoded into a knowledge-based system. Methods currently used to perform this include structured interviews, observation of the domain expert, and questionnaires. However, these are extremely time-consuming approaches, therefore a significant amount of research has been undertaken in an attempt to reduce this. This paper presents an approach to reduce the time associated with the knowledge elicitation process for the development of industrial fault diagnostic systems. Symbolic representation of the engineer's knowledge is used to create a common language that can easily be communicated with the domain experts but also be formalised as the rules for a rule-based diagnostic system. This approach is then applied to a case study based on rotating plant fault diagnosis, specifically boiler feed pumps for a nuclear power station. The results show that using this approach it is possible to quickly develop a system that can accurately detect various types of faults in boiler feed pumps.