Predictive unsupervised organisation in marine engine fault detection

This paper utilises topological learners, the Self Organising Map in combination with the K Means algorithm to organise potential engine faults and the respective location of faults, focussing on a 12 cylinder 2 stroke marine diesel engine. This method is applied to reduce the numerosity of the data presented to a user by selecting representative samples from a number of clusters to enable efficient diagnosis. The novelty of the approach centres around the sparsity of the dataset compared to the majority of fault diagnosis techniques, and the potential for improved safety and efficiency within the marine industry compared to existing diagnosis systems. The accuracy of the SOM and K Means, as well as the Neural Gas algorithm is compared to the standard accuracy of the K Means algorithm to validate the algorithm's performance and application to this domain, where it can be seen that topological learners have much potential to be applied to the field of fault diagnosis.