Implementation of a self-learning algorithm for main engine condition monitoring

Ship machinery maintenance can be seen as a convoluted optimisation problem where managerial, economic, and technical aspects are considered for vessel operation to remain sustainable and profitable. Recent literature shows that condition monitoring of ship systems has been tackled on single and independent component level. However, there has recently been a research tendency towards holistic system modelling. In this respect, this paper presents a methodology for intelligent, system-level modelling for the monitoring of main engine performance utilising data acquired through noon-reports. The proposed methodology will train a main engine expected-performance model. Training is based on one-class Support Vector Machine (SVM) classifier. Newly-acquired data are compared against model output and the probability of belonging to the same performance profile as used for model training is estimated. This will eventually lead to increasing ship operability and income through operational enhancement and minimisation of ship downtime. Two case studies are included, one utilising main engine data and one diesel generator data. Both are complemented by a sensitivity analysis, showing successful results in the recognition of deviant, abnormal ship machinery conditions.