A Fault Diagnosis Method Based on Multi-Layer Spiking Neural Network

For large-scale industrial systems, both efficiency and high reliability are essential. Undetected faults may lead to severe losses. Fault diagnosis serves as an effective approach to prevent such incidents. Machine learning and deep learning methods have been widely applied in this field due to their inherent advantages. With the emergence of the third-generation neural networks, Spiking Neural Networks (SNNs) have also become promising tools for fault diagnosis. Owing to their bio-inspired mechanisms, SNNs exhibit excellent capabilities in processing time-series data, making them particularly suitable for analyzing vibration signals. In this study, a multilayer SNN model with multi-synapse architecture is employed to diagnose wind turbine blade faults using a vibration dataset containing multiple fault types. Appropriate encoding methods are adopted to convert extracted features into spike times, which are then used to train the network through specific algorithms. Ultimately, the method achieved an accuracy of 91.18% in the classification task corresponding to the three conditions of healthy, crack, and erosion. Compared with some existing methods, it also demonstrates certain advantages.