Performance assessment modelling of a low-speed two-stroke marine engine

Abstract

This thesis aims to develop a new method to assess, process, diagnose and predict the performance of a low-speed two-stroke (LS2S) marine engine. Performance monitoring and maintenance decision making are intertwined elements of machinery management and critical for trouble-free ship operations. With the evolution of sensor technology, improved data transfer capability, machine learning algorithms and better knowledge management, there is a gradual move away from fixed-interval maintenance actions towards predictive maintenance to extend the time between overhauls. In this study, a Bayesian Network (BN) model has been developed for performance assessment of a low-speed two-stroke marine engine used for ship propulsion. The overarching objective of the model is to evaluate the health of each cylinder of LS2S engine as performance can vary significantly from one cylinder to another. The results from combustion performance, wear assessment and post-combustion fouling combine to indicate an engine’s operational health. The BN model receives operational inputs from key engine performance parameters identified through a literature search. The conditional probability tables are mainly developed through expert judgements, and partially through various data sources. To address the challenge of converting the engine’s operational data into suitable prior probabilities for BN input, a fuzzy model is developed to accommodate and convert raw subjective data from domain experts, which has been further fine-tuned through the evidential reasoning approach. Moreover, the original BN network is expanded to integrate various features of the BN, fuzzy mode and evidential reasoning to produce a larger comprehensive BN-based decision making toolkit for diagnostics and health assessment. Finally, based on the comprehensive static BN, a novel predictive assessment tool using dynamic BN methodology for a low-speed two-stroke engine is proposed and applied through a case study. The models developed in this study are expected to be utilised by ship operators, with their graphical interface and quantitative outputs facilitating the engine performance management and supplementing existing engine monitoring methodologies. Secondly, with the frequent processing of engine data through this model, an engine operational profile can be developed and compared with the shop test data, calm weather operations, rough weather operations, cargo voyage, ballast voyage and so forth to create a bespoke predictive health assessment tool.