An intelligent multi-element fault diagnosis method of rolling bearings considering damage degrees and sensor abnormity under small samples

Abstract

Aiming at the intelligent fault diagnosis problem of rolling bearings, a novel diagnosis method considering damage degrees and sensor abnormity under small samples is proposed. A complex fault mode simulation scheme with a total of 18 states is designed for rolling bearings, including a single element fault, double elements fault, and all elements fault with damage degrees of slight and heavy and the loose threaded connection of the used sensor. The variational mode decomposition (VMD) is used to decompose the original vibration signals and reconstruct the denoised signals, the reconstructed signals are converted into the grayscale images, and then processed by local binary pattern (LBP) to enhance the image texture features. Under small samples, an improved deep convolutional generative adversarial network (DCGAN) through upsampling, activation function optimization, Dropout addition and model architecture adjustment is used to expand the grayscale texture image (GTI) samples. The improved DCGAN converges the fastest in all states, and the final MMD values are all below 0.5. For the different sample expansion ratios, the residual neural network (ResNet) as the fault diagnosis model is used to verify the effectiveness of DCGAN sample expansion method in improving the accuracy of fault diagnosis. The results show when the original number of samples is 100, the optimal expansion ratio is 1:1. And the fault diagnosis accuracy of ResNet with DCGAN sample expansion is increased by 6.81% from 85.97 to 92.78%, which proves that the proposed method can not only effectively distinguish the fault modes from a single element to all elements with different damage degrees of rolling bearings, but also identify the sensor abnormity with a high accuracy. This work provides an effective way for the intelligent diagnosis of complex fault modes of rolling bearings under small samples.