Multichannel Vibration Signal Fusion Based on Rolling Bearings and MRST-Transformer Fault Diagnosis Model
Xiaoguang Wu, Huadong Peng, Xingyue Cui, Tianwen Guo, Y. Zhang
Abstract
Bearings are essential components in rotating machinery, and the health of bearings significantly influences the overall performance of the mechanical system. To address the constant-speed bearing fault diagnosis, we propose a multi-channel vibration signal fusion method based on bearings, known as the CWTMap, along with the fault diagnosis model Mobile Residual Soft Thresholding Transformer(MRST-Transformer). The MRST-Transformer comprises an enhanced Residual Shrinkage Building Unit with channel-wise (RSBU-CW) and a shallow Cross VisionTransformer (Cross Vit). Specifically, the CWTMap combines the multiple vibration signals from the bearing, providing the intelligent fault diagnosis model with more comprehensive information. Subsequently, the MRST-Transformer incorporates the inverted residual structure from MobileNetV2 in its feature extraction phase, significantly reducing the number of parameters while enhancing the feature extraction capability. In addition, the number of layers in the Cross Vit is reduced to one layer. Experimental results demonstrate that the version with adjusted layer numbers outperforms the unadjusted version, with an average increase in classification accuracy of 1.42%.