Objectives In the realm of marine engineering, the reliable operation of marine engines is of utmost importance for ship safety and efficient operation. However, the complex and harsh marine environment subjects marine engines to various issues, such as component corrosion, pipeline blockages, and poor lubrication. These problems not only degrade engine performance but also increase maintenance costs and pose a threat to the safety of ships at sea. Therefore, developing an effective fault diagnosis method, especially one with cross-operating condition diagnostic capabilities, is crucial. The primary objective of this study is to enhance the diagnostic performance of marine engine fault diagnosis models under cross-operating conditions. By addressing this issue, we aim to provide a more accurate and reliable fault diagnosis solution for marine engines, thereby improving the overall safety and reliability of ships.

Methods To achieve this goal, a novel intelligent fault diagnosis framework based on model fine-tuning technology is proposed. This framework utilizes the deep learning model ResNet-18 as the feature extractor for engine faults. First, a pre-trained model is constructed using source domain data. The pre-training process involves minimizing the gap between the model's predicted values and true labels through a loss function, with the Adam optimizer employed for gradient descent and parameter optimization. A 1D-CNN is used as the core computational module, which consists of convolutional layers, batch normalization, activation functions, and pooling layers. After pre-training, a fine-tuning strategy is adopted. Specifically, the feature extraction layers are frozen while the classifier layers are fine-tuned using target domain data. This approach optimizes the feature distribution between the source and target domains and enhances the model's performance under cross-operating conditions. The effectiveness of the proposed framework is verified using fault data from the lubrication and intake/exhaust systems of marine engines.

Results In the cross-operating condition fault diagnosis tasks for the engine lubrication system and intake/exhaust system, the proposed framework achieves diagnostic accuracies of more than 94% and more than 98% respectively. Comparative experiments with the VGG-11 model as a feature extractor demonstrate that the ResNet-18-based framework has a significant advantage in diagnostic accuracy. The ResNet-18 model shows higher stability and robustness in different transfer tasks, with a lower variance in classification accuracy.

Conclusions In conclusion, the proposed intelligent fault diagnosis framework based on model fine-tuning technology effectively improves the cross-operating condition fault diagnosis performance of marine engine models. It significantly enhances the model's generalization ability, enabling it to adapt to different operating conditions. However, the current study has some limitations. Future research can focus on integrating Agent technology to improve the model's interaction and text understanding capabilities, enhancing the model's ability to identify and diagnose unknown fault types, and conducting experiments on cross-device transfer diagnosis for diesel engines. This will further improve the practical application value of the fault diagnosis technology for marine engines.