Development of a Predictive Maintenance Framework for Marine Propulsion Systems in the Nigerian Offshore Fleet Using Machine Learning

Predictive maintenance, enabled by machine learning analysis of condition monitoring data, represents a transformative approach to managing the reliability and availability of complex marine propulsion systems. This study develops an original predictive maintenance framework tailored to the operational characteristics and data infrastructure of marine propulsion systems aboard vessels in the Nigerian offshore support and tanker fleets. The research employs an engineering design science methodology, developing the framework through iterative cycles of data collection, model development, and validation. Condition monitoring data including vibration signatures, lubricating oil spectrometric analysis results, thermal imaging records, and fuel system performance parameters were collected from seventeen offshore vessels over a thirty-month period, yielding a dataset encompassing more than four hundred machinery degradation events. Multiple machine learning algorithms were evaluated including random forest classifiers, long short-term memory neural networks, and support vector machine models, with performance assessed using cross-validated prediction of failure events within fourteen-day and thirty-day forecast horizons. The long short-term memory model achieved the highest predictive accuracy, with a precision of eighty-seven percent for propulsion system failures within the fourteen-day horizon. The framework makes an original theoretical contribution by incorporating a transfer learning module that enables model adaptation to new vessel types with limited historical data, addressing the data scarcity challenge specific to the Nigerian context. Keywords: predictive maintenance, machine learning, marine propulsion, offshore fleet, condition monitoring