变工况下磨削振纹的深度迁移智能辨识方法

Deep Transfer Learning Method and Its Application in Grinding Chatter Marks Identification under Variable Working Conditions

磨削振纹是磨削加工过程中常见的缺陷之一,直接影响零件的表面质量和加工精度,对磨削振纹进行在线监测、智能辨识并及时采取措施具有十分重要的意义。然而实际加工过程中磨削工况多变,数据分布特征存在差异,传统的机器学习方法均以训练数据和测试数据具有相同的分布为前提,将不再适用。针对这一问题,提出变工况下磨削振纹的深度迁移智能辨识方法。首先根据磨削振动信号的特征频率建立基于边频带特征频率的振纹在线监测指标,并对典型工况下的磨削振动信号进行分析以获取源域样本标签信息。然后在此基础上构建基于加权最大均值差异算法的领域自适应网络(Weighted maximum mean discrepancy based domain adaptation network,WDAN)以提取域不变特征。最后开展变工况试验验证网络的有效性,结果表明在磨削参数多变的情况下,该网络优于卷积神经网络以及其他常用领域自适应网络,能够有效提高振纹识别精度。

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.