摘要
结合两类错误分类造成损失不等这一故障诊断特点,提出了基于最小风险的SVM方法。进行了三个方面的研究:a.在Platt提出的样本后验概率基础上,给出了修正的分类面以及修正的后验概率估算方法,并分析了该方法的合理性;b.将最小风险决策与SVM输出的后验概率有机融合,使该方法对故障的诊断更加敏感,减小漏判概率;c.以电液伺服阀故障诊断为例,对样本数据经K-L变换后进行可视化研究,分类结果表明了该方法的可行性。
According to the characteristics that the losses to the diagnosed equipment from the two types of errors are different , a support vector machine (SVM) technique based on Bayes minimum risk is put forward. Three phases have been discussed in this paper: a.The algorithm is designed to revise the traditional optimal hyperplane and the Platt posterior probability on doubtful classification area, and analyzing the rationality of the algorithm. b. A new method which combines the Bayes decision with minimum risk and the revised posterior probability is put forward, the method is more sensitive to the fault diagnosis and decreases missing fault probability. c.A diagnosis example on electro-hydraulic servo-valve is taken. After K-L transformation, a research on the visualization is proceeded. The results prove the feasibility of the technique.
出处
《振动．测试与诊断》
EI
CSCD
2006年第2期108-111,共4页
Journal of Vibration,Measurement & Diagnosis
基金
陕西省自然科学基金资助项目(编号:2004JC12)
关键词
最小风险
支持向量机
故障诊断
后验概率
电液伺服阀
minimum risk support vector machines fault diagnosis posterior probability electro-hydraulic servo-valve
