A fault diagnosis method for ship motors based on a multi-source data three-level fusion strategy

Three-phase asynchronous motors are widely used in various applications, such as ship propulsion and auxiliary power systems. Accurate monitoring of their operating conditions is of vital importance for ensuring the safety of ship navigation. However, most of the existing methods are limited to single-level fusion strategies, which makes it difficult to explore the signal features in multi-source data fully. For this reason, a diagnostic method based on a multi-source data three-level fusion strategy (MDTFS) is proposed. Secondly, different types of features are extracted using graph convolutional networks and improved convolutional neural networks. The features are fused, and an attention mechanism is applied to redistribute the weights of each channel to enhance key features. Finally, a decision-level fusion scheme based on entropy weighting is designed to diminish the influence of bad diagnostic channels and improve the effectiveness of the final decision. The experimental results show that MDTFS can effectively diagnose various types of motor faults, with a maximum accuracy of 98.12%. Thanks to the excellent diagnostic performance and maintenance support capabilities demonstrated by MDTFS in marine three-phase asynchronous motors, it is expected to improve the diagnostic accuracy of marine motors and enhance vessel safety in future practical applications.