[3F2] An expert Bayesian network solution for intelligent machine fault diagnosis
M Gibson¹, K Taylor¹, B Brown² and B Stephen²
¹Ailsa Reliability Solutions Ltd, UK
²University of Strathclyde, UK
Vibration-based fault analysis remains the standard approach to condition monitoring of rotating machines. However, with increasing complexity and volumes of data, condition monitoring engineers are turning to artificial intelligence (AI) to improve maintenance productivity and accuracy. This paper describes an engineering solution combining the AI methodologies of expert systems and Bayesian networks, providing close-to-real-time online machine fault diagnosis. The solution involves capturing machine fault domain knowledge as encoded heuristics rules in an expert system. A Bayesian network then permits diagnosis with uncertain machine data, where data anomalies or multiple faults still result in the correct identification of fault(s). The solution has been designed and implemented to operate in a cloud platform and results have been verified using Ailsa Reliability Solutions Limited (ARSL)’s Research & Training Skid, where real faults were introduced for system validation. The solution was tested using operational machine data from an industrial site, captured using a variety of technologies: portable vibration analysers and online fixed-wired vibration sensors. This provides condition monitoring engineers with a rapid means of assimilating large amounts of heterogeneous performance evidence to diagnose machine faults in the field.