上海交通大学学报

上海交通大学学报（自然版） ›› 2012, Vol. 46 ›› Issue (09): 1440-1444.

• 自动化技术、计算机技术 • 上一篇    下一篇

基于蚁群SVDD和聚类方法的旋转机械故障诊断  

杜文辽1,2，李安生2，孙旺1，李彦明1，刘成良1   

  1. （1. 上海交通大学 机械系统与振动国家重点实验室， 上海 200240； 2. 郑州轻工业学院 机电工程学院， 郑州 450002）  
  • 收稿日期:2011-11-15 出版日期:2012-09-28 发布日期:2012-09-28
  • 基金资助:

    国家自然科学基金资助项目(61175038)，上海市科技创新行动计划资助项目 (11dz1121500，11JC1405800)，机械系统与振动国家重点实验室项目资助(MSVMS201103,

Fault Diagnosis of Rotating Mechanism Based on Ant Colony SVDD Algorithm and Cluster Method

 DU  Wen-Liao-1, 2 , LI  An-Sheng-2, SUN  Wang-1, LI  Yan-Ming-1, LIU  Cheng-Liang-1   

  1. (1. State Key Laboratory of Mechanical System and Vibration, Shanghai Jiaotong University, Shanghai 200240, China; 2. School of Mechanical and Electronic Engineering, Zhengzhou University of Light Industry, Zhengzhou 450002, China)
  • Received:2011-11-15 Online:2012-09-28 Published:2012-09-28

摘要： 摘要： 
针对典型故障样本缺乏而使常规机器学习方法无法直接应用的难题，提出了一个基于支持向量数据描述（SVDD）新异类检测与基于Davies Bouldin指数（DBI）的K均值聚类方法相结合的旋转机械故障诊断框架.首先，针对正常状态样本建立SVDD模型，并利用蚁群算法对SVDD模型参数进行优化；然后，当拒绝样本数目累积到设定的阈值时，利用K均值聚类方法对其进行处理而获得能够进行标记的类别，其中，K均值聚类的类型数目由DBI辅助确定；最后，针对所标记的各类样本，分别建立SVDD模型并进行训练，将SVDD分类器按照二叉树形式构建系统状态的完整诊断模型.同时，利用滚动轴承多故障模式样本进行训练测试，以验证所提出算法的有效性.结果表明，所提出算法的训练速度为常规网格搜索算法的近10倍，DBI能够有效确定聚类的数目，对样本状态的识别率达到100%. 关键词： 
蚁群支持向量数据描述； K均值聚类； Davies Bouldin指数； 旋转机械； 故障诊断 中图分类号：  TP 183
文献标志码：  A    

Abstract: For the absence of typical fault samples, the general machine learning methods can not be used directly. A hybrid fault diagnosis scheme for rotating mechanism was proposed combining the SVDD algorithm with the Davies Bouldin index (DBI) K-cluster method. Firstly, the SVDD model is constructed for the samples in the normal condition, and the ant colony algorithm is utilized to optimize the SVDD parameters. Then, when the number of rejected samples reaches a given threshold, the K-cluster method is employed to classify these samples and the labels are obtained; furthermore, the number of the classification is determined in accordance with the DBI. Finally, the one class samples are trained with SVDD individually, and the SVDD classifiers are joined to a complete diagnosis model based on a binary tree structure. For the multifault mode samples of rolling element bearing, the speed of training is nearly 10 times greater than that of the grid search approach, while the DBI is verified to determine the number of clusters, and the recognition rate for the bearing samples reaches 100%.

Key words: ant colony support vector date description (SVDD), Kcluster method, Davies Bouldin index, rotating mechanism, fault diagnosis