Machine learning algorithms in shipping: improving engine fault detection and diagnosis via ensemble methods

Abstract

Detection and diagnosis of marine engines faults are extremely important functions for the optimized voyage of any sea-going vessel, as well as the safe conduct of navigation. Early detection of these faults is a prerequisite for reliability: incidents of engine breakdowns can be avoided, since the timely resolving of these faults can ensure the non-interrupted tempo of the sail. Avoiding malfunctions could also improve the ship’s overall environmental “footprint” and even ensure reduced fuel consumption. Initial results of the analysis at hand were presented during the 3rd International Symposium on Naval Architecture and Maritime (INT-NAM 2018), in Istanbul-Turkey. Further exploring the use of machine learning algorithms in shipping and by elaborating more on that effort, an evaluation of intelligent diagnostic methods applicable for a two-stroke slow-speed marine diesel engine is taking place, with the aim to facilitate effective detection and classification of occurring faults. This research was carried out via the cost-free Weka data mining tool, which was used to analyze the data of the engine’s operating parameters that were found outside of the prescribed boundaries. The proposed method is based on the construction of an ensemble classification model “AdaBoost”, which further improves the performance of a basic Simple Cart classifier. During the related experimental activities, the overall recorded performance was 96.5%, clearly establishing this method as a very appropriate choice.