Rotating Machinery Fault Classification using IWGAN-GP and Small Gray Images

Abstract

Listen

Translate article

Sufficient fault samples of rotating machinery are difficult to be acquired in the application scenarios. The limitation of sample numbers may result in low accuracy and over-fitting in fault classification. Thus, rotating machinery fault classification under small samples is a challenging job. To address the above problem, a fault classification method based on improved Wasserstein generative adversarial network with gradient penalty (IWGAN-GP) is proposed in this paper. Firstly, small two-dimensional (2D) gray images are obtained by converting one-dimensional (ID) raw vibration signals to explore 2D features. Secondly, IWGAN-GP is constructed by using 2D convolution layers to design generator and discriminator to fully extract the features of the 2D gray images. Finally, the generated samples are used for assisting small real samples to achieve fault classification. In order to verify the effectiveness of the proposed method, comparative experiments of bearing fault classification are carried out under different ratios between the generated samples and the real samples. The results show that the proposed method can effectively improve the performance of classifying various fault types of bearing with small samples.