MCBA-MVACGAN: A Novel Fault Diagnosis Method for Rotating Machinery Under Small Sample Conditions

In complex industrial scenarios, high-quality fault data of rotating machinery are scarce and costly to collect. Therefore, small sample fault diagnosis needs further research. To solve this problem, in this work is proposed a minimum variance auxiliary classifier generation adversarial network based on a multi-scale convolutional block attention mechanism. Firstly, the multi-scale convolutional block attention mechanism is designed to extract multi-scale information and perform weighted fusion to enhance the ability of the model to capture effective features. Secondly, the minimum variance term is designed to minimize the variance of sample distribution, so that the generated samples are distributed more evenly in the feature space, avoiding the problem of pattern collapse. Finally, the objective function is reconstructed by independent classification loss to improve the ability of model data generation. Experimental results on CWRU and gearbox datasets validate the effectiveness and reliability of the proposed method.