Developing Artificial Neural Network (Ann) Based Predictive Maintenance Techniques for Marine Propulsion Systems

The reliability of marine propulsion systems (MPS) is fundamental to the operational efficiency, safety, and economic viability of maritime activities. This research focused on developing an Artificial Neural Network-based Predictive Maintenance model for early fault detection in MPS, aiming to minimize breakdowns and downtime. The model was developed using a feedforward neural network architecture. It implemented the model combining a reliability framework with field data collected on the case study – Tugboat Abun for risk‑informed maintenance prioritization towards reliability of the MPS of the vessel. The model prediction was an average of 95 per cent for the injectors of the starboard propulsion plant of the vessel, with a mean square error of 12.13 and a mean absolute error of 3.13, indicating the model's predictions were, on average, very close to the actual values. The training progression of the Multi-Layer Perceptron network utilized for the study showcased its best validation performance at epoch 3 with a mean square error of 3.746 x 10-6. The average combined system reliability was 58.82 per cent, indicating a reliable system. These results indicate that the developed Artificial Neural Network predictive model using the Multi-Layer Perceptron network could predict marine machinery failure with a reasonable degree of accuracy for enhanced reliability and minimal downtime.