单类支持向量机在涡轮泵试车数据分析中的应用  

Application of One-Class Support Vector Machines to Fault Analysis of Turbopump Test Data

作　　者：胡雷[1] 胡茑庆[1] 秦国军[1] 邱忠[1] 

机构地区：[1]国防科学技术大学机电工程学院,长沙410073

出　　处：《振动．测试与诊断》2007年第2期95-97,共3页Journal of Vibration,Measurement & Diagnosis

基　　金：国家自然科学基金资助项目(编号:50375153);高等学校全国优秀博士学位论文作者专项资金资助项目(编号200434)

摘　　要：针对在火箭发动机涡轮泵的故障检测过程中缺乏故障样本的问题,应用单类支持向量机,为高速涡轮泵试车数据分析建立了一种新异类检测超球模型。该模型在仅对正常数据进行训练的基础上便可以进行分类工作,且为了提高训练效率,提出了一种分层式的快速训练方法。对火箭发动机涡轮泵试车数据分析的应用结果表明,该算法具有很高的训练效率,检测结果良好。To deal with the lack of fault samples in the fault detection of a Liquid Rocket Engine(LRE)turbopump,a detection model based on one-class support vector machines(OC-SVMs)was founded.As a novel detector,the model requires no negative training instances and is able to solve classification tasks on the basis of positive samples only.And a multi-layer high speed training strategy was proposed for the improvement of training efficiency.The analysis of LRE historical test data showed that the algorithm has a high training efficiency and can detect the faults of the LRE turbopump.

关 键 词：火箭发动机 涡轮泵 单类支持向量机 故障检测 新异类检测 

分 类 号：TP306[自动化与计算机技术—计算机系统结构] V434.21[自动化与计算机技术—计算机科学与技术]