基于HOS奇异值谱的SVDD轴承故障检测方法  

Bearing fault detection using SVDD based on HOS-singular value spectrum

作　　者：陶新民[1] 杜宝祥[1] 徐勇[1] 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《振动工程学报》2008年第2期203-208,共6页Journal of Vibration Engineering

摘　　要：针对轴承故障检测中异常样本不易收集、数据分布不均以及阈值设定等问题,提出了基于支持向量数据描述(SVDD)的轴承故障检测方法。该方法只需对正常样本进行训练,以高阶统计矩阵奇异值谱为故障诊断特征,解决了高阶统计特征(HOS)数据冗余且受噪声影响的不足。实验分析了不同参数对检测性能的影响,并将本方法与多层感知机(MLP)方法及K均值聚类方法进行了比较,验证了方法的有效性和正确性。In order to avoid the practical application problems, including data insufficiency, imbalanced data constitution, and threshold setting which are often faced in bearing fault diagnosis application, The detection schemes based on the Support Vector Data Description（SVDD） and Genetic Algorithm are presented in this paper. In this model, only positive （normal） samples are needed for training purposes. The singular value spectrum of the Higher Order Statistics matrix is adopted as the discriminative characteristics for the bearings fault detection to solve the shortcomings that HOS is too abundant to make further intelligent detection. HOS-Singular Value Spectrum and HOS-Singular Value Spectrum Entropy are compared in experiments which show that HOS-Singular Value Spectrum is more effective than HOS-Singular Value Spectrum Entropy. Comparison of the performance of detection of SVDD with different kernel parameters is experimented. This hybrid approach is compared against Multi Layer Perception （MLP） and K-MEANS cluster detection techniques. The results illustrate effectiveness of the mentioned techniques with some conclusion remarks.

关 键 词：故障诊断 高阶统计特征 支持向量数据描述 奇异值谱 核参数 

分 类 号：TH17[机械工程—机械制造及自动化] TP306[自动化与计算机技术—计算机系统结构]