基于紧密度FSVM新算法及在故障检测中的应用  

A FSVM based on affinity and its application in bearing fault detection

作　　者：陶新民[1] 徐晶[2] 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 [2]黑龙江科技学院数力系,黑龙江哈尔滨150027

出　　处：《振动工程学报》2009年第4期418-424,共7页Journal of Vibration Engineering

基　　金：黑龙江省博士后科学基金资助项目(LBH-Z08227);哈尔滨工程大学校科研基金资助项目(002080260735)

摘　　要：针对传统的模糊支持向量机(FSVM)算法对边缘噪声敏感的不足,提出一种基于非线性紧密度和K最近邻方法(KNN)相结合的FSVM算法。该方法在计算样本隶属度大小时既考虑样本与类中心的距离,类中样本的紧密度,同时还考虑与其他类样本间的关系,其中紧密度的计算采用非线性数据分布描述方法进而使计算的隶属度更精确。实验结果同传统FSVM及其他改进的FSVM算法进行比较,对于国际标准测试数据及轴承故障检测问题,结果验证了建议算法具有很强的鲁棒性及高效的检测性能。Since SVM is very sensitive to outliers and noises in the training set, a novel fuzzy support vector machine （FSVM） algorithm based on affinity among samples is proposed. In the proposed method, the fuzzy membership is dependent on not only the relation between a sample and its self cluster center, but also the relation between a sample and those non-self samples. A method defining the affinity among self samples is considered using a nonlinear sphere with minimum volume while containing the maximum of the samples. Then, the suzzy membership is defined according to the position of samples in nonlinear sphere space. The affinity among non-self samples is described by K-nearest neighbors （KNN）. Compared with other FSVM algorithms, this method can effectively distinguish between the valid samples and the outliers or noises. In the benchmark data and bearings fault detection application, Experimental results show that the proposed FSVM algorithm is more effective and robust than the traditional FSVM and other modified FSVM algorithms.

关 键 词：故障检测 模糊支持向量机 K近邻方法 紧密度 

分 类 号：TH165.3[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]