Implementing unsupervised learning algorithm for marine engine data clustering applications

Raptodimos, Yiannis and Lazakis, Iraklis (2018) Implementing unsupervised learning algorithm for marine engine data clustering applications. In: Proceedings of the 2018 Smart Ship Technology Conference. Royal Institution of Naval Architects, London.

Full text not available in this repository.Request a copy from the Strathclyde author

Abstract

Data preparation and processing is of great importance in a ship condition monitoring tool, as inaccurate and misinterpretation of data can significantly affect the condition monitoring accuracy and performance. Data for performance parameters related to the case study of a Panamax container ship main engine are clustered using an artificial neural network, the Self-Organizing Map (SOM). Neighbouring clusters are compared through a distance metric to examine the existence of data similarities. Additionally, the SOM has a supplementary functionality of identifying data clusters exceeding thresholds, consequently providing diagnostics connected to a Failure Mode and Effects Analysis (FMEA) for the main engine, providing valuable insight and information regarding potential faults. The SOM model is validated through actual data extracted from the case study. Moreover, simulated data representing data exceeding alarm levels for the engine fuel oil system demonstrate the capabilities of the SOM clustering process in combination with the associated FMEA results.