Leveraging BART for intelligent fault diagnosis of marine diesel engines: a language model approach

Abstract Accurate fault diagnosis of marine diesel engines is essential for ensuring operational safety and optimizing maintenance strategies. This study proposes a novel fault diagnosis method based on the pre-trained large language model BART (Bidirectional and Auto-Regressive Transformers). Experimental evaluations on a publicly available fault dataset under different signal-to-noise ratio (SNR) conditions demonstrate that the BART model achieves high classification accuracy and exhibits strong robustness against noise interference. Notably, the model maintains stable performance in low-SNR environments and achieves near-perfect accuracy in high-SNR scenarios. Compared with traditional classification methods, the proposed approach not only enables accurate fault identification but also generates specific, actionable maintenance recommendations. This dual capability enhances result interpretability and provides valuable decision-support insights for predictive maintenance applications. Overall, this study underscores the potential of large language models in advancing intelligent fault diagnosis for the maritime industry.