A Convolution Neural Network-Based Approach for Ship Engine Fault Diagnosis Leveraging Transfer Learning

In Facing the complexities of diagnosing ship engine faults in marine environments, this study presents a novel approach that combines Convolution Neural Network (CNN) with transfer learning. We began by obtaining a substantial annotated data set from a marine engine simulator to train an initial CNN model. This model was then adapted to the target domain with limited annotated data through fine-tuning. Comparative evaluations with traditional methods like CNN, K-Nearest Neighbors (KNN), decision trees, and random forests showed that our transfer learning-enhanced CNN model achieved highest recognition accuracy. This outcome confirms the efficacy of our approach and its significant potential for diagnosing ship engine faults, especially in data-scarce scenarios.