基于代价敏感直推式学习的故障诊断方法.pdfVIP

学兔兔 第31卷 第5期 仪 器 仪 表 学 报 Vo1．31 No．5 2010年5月 Chinese Journal of Scientific Instrument Mav 2010 基于代价敏感直推式学习的故障诊断方法 吴 薇 ，胡静涛 (1 中国科学院沈阳自动化研究所T业信息重点实验室 沈阳 110016；2 中国科学院研究生院 北京 100039) 摘 要：针对故障诊断领域存在的不考虑误诊断代价以及提出泛化能力强的诊断规则难等问题，提出了一种代价敏感直推式 学习故障诊断方法。基于Kolmogorov算法随机性理论和代价敏感学习最小期望误分类代价准则提出了代价敏感直推式分类机 制，并在此基础上设计了用于故障诊断的CsTCM—kNN算法。通过旋转机械轴系故障代价敏感诊断实验，验证了该方法能够有 效地降低误诊断代价，且保证较高的诊断准确率。 关键词：故障诊断；代价敏感；直推式学习；算法随机性理论 中图分类号：TP18，TP206 文献标识码：A 国家标准学科分类代码：520．2099 Fault diagnosis based on cost-sensitive transduction inference Wu Wei ’-．Hu Jingtao ( Key Laboratory Industrial Informatics，Shenyang In,st．ofAutomation，Chinese Academy D厂Sciences，Shenyang 110016，China； 2 Graduate School ofthe Chinese Academy ofSciences，Beijing 100039，China) Abstract：Most researches on fault diagnosis pursue to minimize the error rate without considering the misclassifica— tion cost，and are also difficult to propose enough diagnosis rules with good generalization ability．In this paper，a new fault diagnosis method based on cost—sensitive transduction inference is presented，which solves the two problems mentioned above．A cost—sensitive transduction classification machine is proposed based on the Kolmogorov algorithm randomness theory and minimum expected misclassification cost principle，and then a CsTCM-kNN algorithm for fault diagnosis is designed．Experiment results of cost—sensitive fault diagnosis of a rotating machine show that the pro— posed method can reduce misclassification cost effectively with high diagnosis accuracy． Key words：fault diagnosis；cost—sensitive；transduction inference；algorithm randomness theory 论更倾向于正常状态