Imbalanced sample fault diagnosis of rotating machinery using conditional variational auto-encoder generative adversarial network

Abstract

In many real applications of planetary gearbox fault diagnosis, the number of fault samples is much less than normal samples while fault samples are hard to collected in different working conditions, so many traditional diagnosis methods will get low accuracy. To solve this problem, a method based on conditional variational auto-encoder generative adversarial network (CVAE-GAN) is proposed for imbalanced fault diagnosis. Firstly, new method uses encoder network of conditional variational auto-encoder to obtain the distribution of fault samples, and then a large number of similar fault samples can be generated through decoder network. Secondly, the parameters of generator, discriminator and classifier may be continuously optimized using adversarial learning mechanism. Finally, the trained CVAE-GAN is applied for intelligent fault diagnosis of planetary gearbox. The experimental results show that CVAE-GAN can generate fault samples in different working conditions, which improve the fault diagnosis performance of planetary gearbox. The sample generating ability of CVAE-GAN is significantly higher than other methods in two cases of imbalanced dataset.