基于支持向量机的机械零件剩余寿命区间估计  

Remaining Useful Life Interval Estimation for Machine Parts Based on SVM

作　　者：王健[1] 孙志礼[1] 于震梁[1] 柴小冬[1] 

机构地区：[1]东北大学机械工程与自动化学院,辽宁沈阳110819

出　　处：《东北大学学报（自然科学版）》2016年第7期974-978,共5页Journal of Northeastern University(Natural Science)

基　　金：国家科技重大专项(2013ZX04011-011)

摘　　要：为提高机械零件剩余寿命估计精度,提出了一种基于支持向量机(support vector machine,SVM)的剩余寿命区间估计模型.简要介绍SVM的线性及非线性理论,分析SVM输入变量与输出变量间的统计关系,将机械零件性能退化指标和剩余寿命分别作为SVM输入变量及输出变量.假设输入变量与残差相互独立且残差分布类型已知,采用极大似然法估计残差的分布参数,在此基础上推导一定置信水平下SVM输出变量置信区间.将均方误差作为SVM预测误差的衡量指标,应用变步长网格搜索法确定SVM参数.通过实例说明所提模型能够准确对机械零件剩余寿命进行区间估计,具有较强的工程应用价值及通用性.To improve the accuracy of remaining useful life estimation for machine parts, aninterval estimation model was proposed based on the SVM （ support vector machine ）. The lineartheory and nonlinearity theory of SVM were briefly introduced, and the correlation between inputvariable and output variable was analyzed. Degraded index and remaining useful life of machineparts were treated as input variable and output variable, correspondingly. It was assumed thatinput variable and residual error were independent and the residual error＇ distribution pattern wask nown . Distribution parameters of residual error were estimated by means of the M L E （ maxi mumlikelihood estimation）. Then the confidence interval of SV Moutput variable was obtained under acertain confidence level. The MSE （ mean squared error） was used to measure the prediction of SVM . The SVM parameters were gotten by the means of variable step size grid search. Anumerical example was presented to show that the proposed model can estimate the remaininguseful life confidence interval precisely with the engineering application values and generality.

关 键 词：剩余寿命 支持向量机 区间估计 机械零件 置信区间 均方误差 

分 类 号：TB114.3[理学—概率论与数理统计]