Research on Data Enhancement of Rolling Bearings Based on His-VAEGAN

Due to the poor working conditions of diesel generators and the precise structural characteristics and installation positions of rolling bearings, it is difficult to collect vibration samples to a great extent, resulting in the insufficiency and imbalance of sample data used for feature analysis. This paper proposes a method based on energy histogram optimization of Variational Autoencoding Generate Adversarial Network (VAEGAN) to expand sample data set. Specifically, on the basis of the classic VAEGAN, adding the classification to the generator (VAE) to improve and enhance the category attributes of the generated samples. Then, drawing the energy histogram of the real and generated samples, and using the non-intersection core of the energy histogram to update the network to improve the quality of signal generation. Through experiments and index analysis, the effectiveness of the algorithm for sample generation is verified.