Predictive Maintenance in Maritime Operations: A Data-Driven Approach for Early Warning Systems

This study outlines a comprehensive analytical methodology applied to a vessel’s operational data to establish a rigorously validated, data-driven baseline for the Marine and Health Usage Monitoring System (MHUMS) algorithmic modules. By correlating a continuous three-month dataset of sensor telemetry with historical alarm logs, the primary objective was to develop and train a robust predictive model. This study addresses the critical challenge of converting continuous marine sensor telemetry into actionable early warnings by establishing a validated predictive baseline. The proposed baseline predictive architecture achieved a Recall of 0.924 and an area under the ROC curve (AUC) of 0.939 for a 15 min forward prediction window, proving the feasibility of transitioning from reactive diagnostic monitoring to proactive early warnings.