CBAM-CRLSGAN: A novel fault diagnosis method for planetary transmission systems under small samples scenarios

Abstract

In various and complex industrial scenarios, the fault data acquisition of planetary transmission system is expensive and unavailable, thus intelligent fault diagnosis under small fault samples scenarios is a challenging task. Therefore, we propose a Convolutional Block Attention Mechanism Conditional Regularized Least Squares Generative Adversarial Network (CBAM-CRLSGAN) method for intelligent diagnosis of planetary transmission systems. First, the diversity of original samples is increased by an overlapping segmentation strategy. Then, a novel data augmentation method is proposed via incorporating the CBAM module and conditional regularized least squares loss function into least squares generative adversarial network, which enables the proposed method to extract data features efficiently and improve training stability. Finally, the real and obtained fake samples are input into the classifier to realize fault diagnosis. The experimental results on the planetary transmission system test rig show that the proposed CBAM-CRLSGAN can obtain superior diagnostic performance with a diagnosis accuracy of 99.35%.