Bayesian and machine learning-based fault detection and diagnostics for marine applications

Marine maintenance can improve ship safety and safeguard the profitability of the shipping by leveraging predictive maintenance, Machine Learning and Data Analytics tools. This paper aims to enrich the existing literature, by developing a novel framework for ship diagnostics based on the assessment of real-time operational data. The novel framework is structured around the calculation of the probability of different faults, with tangible effects on ship safety and ship operations. Moreover, the framework can identify the root-cause of developing faults without using black-box Neural Networks, nor complicated and time-consuming physics-based models. This research effort is based on the novel integration of Machine Learning-based Fault Detection, Exponentially Weighted Moving Average control charts, and diagnostic networks based on Bayesian Networks. The Bayesian Network also allows the examination of the rate of development (fault profile) of various faults and failure modes. For validation purposes, the case study of a marine Main Engine is used to examine faults in the engine’s Air Cooler and Air and Gas Handling System. It is concluded that any simultaneous abnormal deviations in the Main Engine’s Exhaust Gas Temperature are more likely to be caused by a fault in the Air and Gas Handling System.