Research on bearing fault classification based on deep learning

In response to the problem of low classification accuracy due to the scarcity of bearing fault samples in industrial sites, this paper studied an accurate bearing fault classification method based on deep learning under the condition of small sample size, and proposed a meta-learning model that integrated the improved convolutional neural network and prototype network (referred to as IC-PN). Firstly, a time-domain - frequency-domain joint enhancement strategy is designed to expand the small sample data set; secondly, classification is achieved through "feature extraction - prototype generation - distance measurement", and the parameters of the IC-PN model are optimized using the round-robin training method. Finally, experiments are conducted on the CWRU bearing dataset. The results show that the classification accuracy of the IC-PN model in 1-time, 5-times and 10-times small sample scenarios is 92.3%, 96.7% and 98.1% respectively, which is 8.5%-12.2% and 3.2%-5.6% higher than that of the traditional convolutional neural network and the original prototype network respectively. Moreover, in the cross-condition environment, its classification accuracy remains above 90%, successfully solving the problem of small sample bearing fault classification.