Fault diagnosis and restoration of ship structure monitoring signals based on machine learning

To prevent signal faults from causing misjudgements of structural health and even catastrophic accidents in ship monitoring systems, a fault diagnosis and restoration method is proposed by using machine learning. The method integrates advanced technologies including wavelet transform, IPSO-BP neural network, CNN, event-triggered sampling based on predictive zero-crossing instant, and adaptive segmented least squares algorithm, focusing on efficient identification and rectification of the typical signal faults. Based on the monitoring data of a model test, an impact analysis on the fault diagnosis and restoration method is carried out, covering aspects like noise resistance, fault severity, fault occurrence time, and ship monitoring position. The result indicates that the proposed method has a total diagnostic rate of no less than 98 % at different monitoring positions, and its noise-resistance ability is superior to that of the LSTM and Random Forest algorithms. Moreover, it delivers outstanding restoration effects. Compared to traditional mean segmentation, it reduces RMSE by 73.86 % for bias faults, 75.49 % for drift faults, and 19.55 % for impulse faults. This method can effectively enhance the stability of ship structural health monitoring systems, providing critical technical support for intelligent ship navigation safety.