Intelligent Fault Diagnosis of Bearings under Variable Working Conditions and Small Samples with Generative Adversarial Network

Intelligent fault diagnosis of bearing based on data drive has been a hot research field in recent years and achieved lots of results. However, current research mainly faces: 1) It is a great challenge to develop an effective intelligent diagnosis method in practical industrial scenarios because of the lack of fault signals in small samples; 2) It has poor adaptability to intelligent fault diagnosis under variable working conditions. Aiming at the above problems, an intelligent fault diagnosis method for bearings under variable working conditions and small samples based on generative adversarial network is proposed. Firstly, the signal highly similar to the actual fault signal is generated through generative adversarial network training and this part of the signal can be used as training data to solve the problem of deficient small sample fault dataset. Then, the similar fault characteristics learned from the data of a certain working condition through domain confrontation training are transferred to the target working condition. Finally, fault diagnosis is realized on the target domain data by the classifier trained on the fault features. The proposed method is evaluated through the Case Western Reserve University (CWRU) bearing dataset with the result show that it has high fault classification accuracy and transferability under the condition of small samples and variable working conditions.