Twin Network-based Bearing Fault Diagnosis Method Based on Transfer QCNN
It is of great significance for bearing fault diagnosis to reduce the risk of damage in rotating machinery and further im-prove economic benefits.Deep learning is widely used in bearing fault diagnosis,but deep learning models are prone to noise interfer-ence during training and testing,leading to performance degradation.Moreover,because the operating conditions of bearings change frequently,it is difficult to collect the data in different conditions.To address this issue,a bearing fault diagnosis method based on transfer quadratic convolutional neural network(QCNN)and Siamese network is proposed.Firstly,the QCNN is pre-trained to ob-tain the parameters of the model with strong discrimination.Then,the pre-trained parameters are transferred to the QCNN as a sub-network in the Siamese network.And then,the Siamese network is trained to obtain the model.Finally,the test data and fault data are combined to form the input of data pairs to the model,which obtains the fault type of the test data.This method combines the QCNN with the Siamese network,where the quadratic neurons of the QCNN have powerful feature extraction capabilities,and the Si-amese network is trained with the shared weights and relative relationships,which alleviates the impact of noise and the data of imbal-anced operating conditions.Experimental results show that compared to the traditional machine learning models and QCNN,the pro-posed method has a better performance in dealing with noise and imbalanced operating condition data.
万方数据
维普
