Image recognition of bearing fault based on parameterized AlexNet with interpretable decision fusion

A bearing fault diagnosis technique based on interpretable decision fusion and parameterized AlexNet is developed in order to address the issues of the standard depth model of rolling bearing fault detection, specifically regarding the network's weak anti-interference capabilities and the challenging selection of training parameters. Firstly, to improve the anti-noise performance of the network, Gaussian white noise is implanted into the original data. The original data undergoes continuous wavelet transform processing to create a two-dimensional time-frequency map, which is then fed into a convolutional neural network to diagnose bearing faults. The Seagull optimization algorithm is utilized to optimize the training parameters, resulting in obtaining network training parameters with robust feature extraction ability. Then we train multiple networks to make decision fusion, and finally give the interpretability of the networks based on Grad-CAM. Verified by the case of Western Reserve University's open-bearing fault data set, the method proposed in this paper has a good effect on fault image recognition accuracy and interpretability and can be popularized in engineering applications.