摘要
针对机械故障诊断中故障类样本不易获取以及样本分布不均的问题,提出了基于一类超球面支持向量机(SVM)的故障诊断方法,该方法只需要对正常类样本进行训练。试验分析了异常类样本缺失对一类超球面支持向量机性能的影响,并提出模型参数优化选择方法,以提高分类模型的推广能力。分析了不同训练结果的分类能力,并对一类超球面支持向量机与一类超平面支持向量机的分类结果进行比较,验证了前者的正确性和有效性。
In order to solve the pratical problem in fault diagnosis of machine, which includes data insufficiency and imbalanced data constitution, the method of fault diagnosis based on one-class hyperspherical SVM is presented in this paper. For one-class hyperspherical SVM, only normal class samples are needed for training purpose. The influence on performance of one-class hyperspherical SVM for lacking of abnormal class samples is analysed, and optimization selection for model parameters is presented to improve generalization performance of classification model. Classification ability of different training result is analysed. Classification result of one-class hyperspherical SVM and hyperplane SVM are compared. The result illustrates effectiveness of one-class hyperspherical SVM.
出处
《振动工程学报》
EI
CSCD
北大核心
2008年第6期553-558,共6页
Journal of Vibration Engineering
基金
国防科技重点实验室基金资助项目(51457050103JB3502)
关键词
故障诊断
一类超球面支持向量机
互信息
匀幅
fault diagnosis
one-class hyperspherical support vector machine
mutual information
even amplitude
