基于最小期望代价的汽轮机代价敏感故障诊断  

Cost-sensitive fault diagnosis based on minimum expectedcost criterion for steam turbine

作　　者：刘金福[1] 于达仁[1] 

机构地区：[1]哈尔滨工业大学能源学院,哈尔滨150001

出　　处：《仪器仪表学报》2008年第12期2599-2604,共6页Chinese Journal of Scientific Instrument

摘　　要：在故障诊断领域,粗糙集已成为一种有效的不一致信息处理工具,然而当故障诊断存在明显的诊断代价差异时,粗糙集方法由于没有考虑诊断代价,不能取得满意的效果。本文提出了一种基于最小期望代价的汽轮机故障诊断方法,首先利用粗糙集方法进行故障诊断规则的提取,然后采用最小期望代价准则进行诊断决策的综合评价,克服了传统方法不考虑故障诊断代价的不足。该方法不仅实现简单,而且通用性强,能够被容易地应用于其他类型的学习器,从而考虑诊断代价。通过汽轮机振动的代价敏感故障诊断实验表明,在故障诊断存在不一致的情况下,本文提出的方法倾向于将故障诊断为高代价故障,从而能够达到降低诊断代价和风险的目的。In the realm of fault diagnosis, rough sets have been a powerful tool to deal with the inconsistent information. However, when the misdiagnosis costs of different faults are unequal, rough sets can not acquire a satisfied result due to not considering the misdiagnosis costs. This paper proposes a cost-sensitive fault diagnosis method for steam turbine based on minimum expected cost criterion. This method employs rough sets to extract the fault diagno- sis rules and employs the minimum expected cost criterion to evaluate the decisions of fault diagnosis, which overcomes the shortcomings of the classical method not considering the diagnosis costs. The proposed method is simple and universal, and can be easily used in other learning machines to consider the diagnosis costs. Experiments on the cost-sensitive fault diagnosis for the vibration of a steam turbine show that when the outputs of fault diagnosis are inconsistent, the method proposed in this paper is inclined to output the high-cost fault and decreases the overall costs of fault diagnosis.

关 键 词：汽轮机 故障诊断 诊断代价 最小期望代价准则 粗糙集 

分 类 号：TK38[动力工程及工程热物理—热能工程] TP277[自动化与计算机技术—检测技术与自动化装置]