A novel fault transfer diagnosis method using CBAM-MobileNetV2 and its application to mine hoist main bearings under variable working conditions

The mine hoist is a key equipment in mining operations, with the main bearing being a crucial component that operates under harsh conditions and is susceptible to frequent failures. Therefore, diagnosing faults in the main bearing during its operational phase is of utmost importance. However, a single diagnostic model may not be universally applicable due to the complex working conditions of main bearings in coal mining. Aiming to address this challenge, this paper introduces a novel fault transfer diagnosis method using CBAM-MobileNetV2. At the algorithmic level, this study integrates the CBAM attention mechanism into the lightweight convolutional network MobileNetV2. This integration facilitates prioritization of fault features by the network, enhancing its feature extraction capabilities. At the data level, the study uses the CWRU dataset for network training to establish a transfer diagnostic model. The toplevel parameters of the model are finetuned, and variable condition transfer diagnoses are performed, transferring from the rolling bearing data of the CWRU laboratory to bearing data from a mine hoist simulation test bench. Comparative experimental analyses with three other lightweight convolutional neural networks reveal that the proposed fault transfer diagnosis method offers high accuracy and robustness, addressing fault diagnosis needs across various operational conditions.