Ship Diesel Engine Fault Diagnosis Using Data Science and SVM Classifier

Abstract

Main engine is the most significant element of reliability structure of the motor vessel. Its failure excludes the ship from the operation and leads to the most dangerous situation - loss of manoeuvrability. Therefore, accurate fault diagnosis should be key element of the modern maintenance strategies. In mechanical engineering the vibrations and acoustic signals recorded during the object operation is the most meaningful data used to identify the state of reliability. The diagnostics of damage to a marine diesel engine presented in the article is therefore carried out based on vibration and noise signals collected on each of the engine cylinders. It is proposed to identify the engine reliability state and the type of failure state using supervised machine learning. At first, the vibrations and noise signals were recorded during normal operation and during operation in five fault engine conditions. The recordings were then divided into parts by masking with a time window of calculated length. To model real situations where the number of available input data is limited, the number of signals were increased using shuffling and overlapping of masking windows. For each of received signal the characteristic values were calculated. Among the calculated values, the most significant features were selected. The obtained set of labelled samples was divided into learning and testing subsets. A subset of the learning data was used to identify the state of reliability and type of engine failure state, thanks to the use of machine learning. Eventually, the quality of the proposed identification method was checked using a subset of testing data.