Fault Identification of Marine Rotating Machinery Based on Feature Fusion and Model Fusion

Modern vessels possess intricate structures and operate in harsh environments. The safety of their machinery and equipment presents a significant hazard. This paper proposes a fault identification method based on the fusion of features and models. Initially, the audio of ship mechanical equipment is processed to extract and fuse identification features, followed by training using an ensemble method of multiple neural network models. Ultimately, by analyzing the training loss and accuracy under various data samples and network models, the optimal data processing and network model fusion methods are established. Experimental results demonstrate that the classification accuracy of this method can reach 91%.