Hybrid Digital Twin Framework for Predictive Maintenance and Operational Optimization of Marine Propulsion Systems

Abstract

Marine propulsion systems operate under highly variable environmental and mechanical conditions, making conventional threshold-based monitoring and calendar-based maintenance insufficient for early degradation detection and operational optimization. This study proposes a hybrid Digital Twin framework for a Panamax-class container vessel equipped with a MAN B&W ME-GI propulsion system. The framework integrates multi-rate operational and sensor data, time synchronization and preprocessing, a physics-based Mean Value Engine Model, machine-learning residual analysis, anomaly detection, Remaining Useful Life estimation, uncertainty quantification, and PMS/EMS-based decision support. A twelve-month simulated dataset was generated to represent variations in engine load, vessel trim, draft, sea state, hydrodynamic resistance, and progressive machinery degradation. The proposed framework was evaluated against a baseline Planned Maintenance System using Specific Fuel Oil Consumption, unplanned maintenance events, advisory latency, and false-alarm rate. Results show that the Digital Twin reduced Specific Fuel Oil Consumption from 184.6 to 176.9 g/kWh, equivalent to a 4.17% improvement. Unplanned maintenance events decreased from 11 to 9 per year, advisory latency was reduced from 20.0 to 8.5 seconds, and the false-alarm rate declined from 6.1% to 2.1%. The framework also enabled earlier identification of scavenge-cooler fouling, cylinder leakage, injector-response delay, turbocharger deterioration, and bearing degradation. These findings demonstrate that combining physics-based modelling with temporal residual analysis can improve propulsion-performance estimation, diagnostic reliability, and condition-based maintenance planning. However, further validation using real shipboard data and multiple vessel configurations is required before operational deployment.