• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2015, Vol. 51 ›› Issue (20): 1-8.doi: 10.3901/JME.2015.20.001

• 智能制造和RFID技术专栏 •    下一篇



钱士才, 孙宇昕, 熊振华   

  1. 上海交通大学机械系统与振动国家重点实验室 上海 200240
  • 出版日期:2015-10-15 发布日期:2015-10-15
  • 基金资助:

Support Vector Machine Based Online Intelligent Chatter Detection

QIAN Shicai, SUN Yuxin, XIONG Zhenhua   

  1. State Key Laboratory of Mechanical System and Vibration, Shanghai Jiao Tong University, Shanghai 200240
  • Online:2015-10-15 Published:2015-10-15

摘要: 为了检测车削过程中的颤振,提出一种颤振在线智能检测方法。使用最小二乘一类支持向量机,训练出描述特征矢量集的超球面,通过计算被测样本与超球面的距离来判断其是否颤振。基于相干准则和分块矩阵求逆,构造了在线稀疏结构的最小二乘一类支持向量机,将特征信息存储于特征库(字典)中,通过更新特征库实现检测模型的在线进化。在颤振检测的应用中,首先使用小波包分解,得到第三层节点能量的比例作为特征矢量,以离线数据构造特征矢量作为输入,训练得到初始检测模型以及特征库,在线检测中不断更新特征库,实现检测模型的在线进化。试验结果表明,在车削颤振识别中,在线进化的检测模型的识别效果更好,颤振预报准确率高达至99.04%,优于离线模型的预报准确率96.74%。

关键词: 颤振, 特征库, 相干准则, 在线进化, 最小二乘

Abstract: In order to detect chatter in the process of turning, an online intelligent chatter detection method is proposed. In this method, least squares one class support vector machine(LS-OC-SVM) is used to extract a hyper plane as an optimal description of training objects. Chatter is detected by computing the distance between the sample to be tested and the hyper plane. Sparse online LS-OC-SVM is proposed based on coherence criterion and partitioned matrix inversion, so that features information can be stored in the feature library which is also called dictionary. The detection model can be evolved continuously through the online update of feature library. In the application of chatter detection, firstly, feature vector is constructed for chatter detection based on node energy ratios of the third level of wavelet packet decomposition. Then, initial detection model and feature library are trained by using offline feature vectors as input. In the online detection scheme, the detection model is evolved while feature library is updated. The experimental results show that the online evolution model performs better than offline model in the cutting chatter detection. Chatter detection accuracy of the online evolution model is 99.04%, which is better than offline model whose detection accuracy is 96.74%.

Key words: chatter, coherence criterion, feature library, least squares, online evolution