An improved generative adversarial network for fault detection of bearing

A method employing the Convolutional Wasserstein Conditional Generative Adversarial Network (CW-CGAN) is proposed to improve fault detection under imbalanced sample conditions. Real fault signal datasets are collected and pre-processed, utilizing the CWGAN-GP model to generate additional samples and expand the training set. A Convolutional Neural Network (CNN) is developed to facilitate automatic fault feature extraction and classification. Experimental results indicate that the augmented CNN model excels in validation and test sets, accurately classifying various fault types. The effectiveness of the model is further substantiated by the visualization of the confusion matrix. This method presents an effective solution for addresses data imbalance issues in fault detection, with future work aimed at optimizing model architecture and training algorithms to enhanced accuracy and real-time performance.