变工况下磨削振纹的深度迁移智能辨识方法

Grinding chatter marks are one of the common defects in the grinding process, which directly affect the surface quality and machining accuracy of the parts. It is very important to monitor and identify the grinding status intelligently, and take timely measures. However, in actual machining process, the grinding conditions are variable, and the data distribution characteristics are different. The traditional machine learning methods are based on assumption that the training and testing data have the same distribution, thus will no longer be applicable. Aiming at this problem, a deep transfer learning method is proposed for grinding chatter marks identification under variable working conditions. First, an online monitoring indicator of grinding chatter marks is established based on the characteristic frequency of the sideband, and then the grinding vibration signals under the typical working conditions are analyzed with the indicator to obtain labels of source domain. On this basis, the weighted maximum mean discrepancy based domain adaptation network is proposed to extract domain-invariant features of the source and target domains. Finally, variable-condition experiments are carried out to verify the effectiveness of the method. The results show that the method is superior to convolutional neural networks and other commonly used domain adaptation methods, which can effectively improve the accuracy of grinding chatter marks identification.