A Human-In-The-Loop System for Clustering and Visualization of Failure Modes In FMEA

Two important issues related to the practical implementation of Failure Modes and Effects Analysis (FMEA) are the lengthiness and intricacies of an FMEA worksheet. As such, to tackle the aforementioned issues, a new FMEA methodology which incorporates an HITL system is proposed. Risk ratings of failure modes, i.e., associated Severity, Occurrence, and Detection ratings are considered. In this thesis, the lengthiness and intricacies of an FMEA worksheet are first substantiated through three benchmark examples. Then, an HITL system incorporating clustering, cluster validation, and visualization is developed. Clustering serves to cluster failure modes such that failure modes of the same cluster share a higher degree of similarity as compared to those of other clusters. Cluster validation is used to determine the optimum number of clusters and the validity of the formed clusters. Last but not least, visualization is used to: (i) provide an overview of all failure modes in an easy to comprehend manner; (ii) allows FMEA users to access and analyse a greater number of failure modes simultaneously which ultimately leads to a better decision making process. The use of clustering, cluster validation, and visualization in FMEA are demonstrated through selected FMEA benchmark examples. Finally, the usefulness and practicality of the developed HITL system is demonstrated through a real-world FMEA case study pertaining to crop management in paddy. The specific focus is on issues related to investment yield and improving paddy FMEA users decision making process. Discussions held with paddy FMEA users reveals that the developed system is useful for paddy FMEA user