基于支持向量数据描述的剩余寿命预测方法  

Remaining useful life prediction based on support vector data description

作　　者：武千惠 黄必清[1] WU Qianhui;HUANG Biqing(Department of Automation,Tsinghua University,Beijing 100084,China)

机构地区：[1]清华大学自动化系,北京100084

出　　处：《计算机集成制造系统》2018年第11期2725-2733,共9页Computer Integrated Manufacturing Systems

基　　金：国家重点研发计划资助项目(2018YFF0214705);江苏省重点研发计划资助项目(BE2016049)~~

摘　　要：为了解决工业设备关键部件的故障预测问题,基于支持向量数据描述(SVDD),定义了表征设备部件健康状态的退化指数,并由此提出一种剩余寿命预测方法。首先利用小波包分解从历史传感器状态监测数据中提取特征向量;然后通过粒子群优化算法选择能够使训练集退化指数取值的变化趋势更加接近指数规律的核函数参数,进而利用目标部件处于健康状态的特征向量训练SVDD模型,得到相应的超球面;最后通过待测样本点和SVDD超球面间的距离计算退化指数,确定目标部件的健康状态并预测其剩余寿命。最后通过实验验证了所提剩余寿命预测方法的有效性。To solve the failure prognostics problem of industrial equipment,based on Support Vector Data Description(SVDD),a Degradation Index(DI)for representing the operation state of target component was defined,and a method for remaining useful life prediction was proposed.In this method,feature vectors were extracted from historical monitoring data of sensors by applying Wavelet Packet Decomposition(WPD).The particle swarm algorithm was employed to select an appropriate parameter of kernel function,which could make the trend of DI of training set closer to the exponential function.A SVDD model was trained to obtain the hypersphere by using features extracted from the healthy state.The distance between sample points and hypersphere was used to calculate DI,which was adopted to determine the operating state of component and furthermore to estimate its remaining useful life.The effectiveness of the proposed method was verified with the experimental data on bearings' accelerated life tests provided by FEMTO-ST institute.

关 键 词：支持向量描述 粒子群优化 轴承 故障预测 剩余寿命 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]