基于相对特征和多变量支持向量机的滚动轴承剩余寿命预测  

Remaining Life Predictions of Rolling Bearing Based on Relative Features and Multivariable Support Vector Machine

作　　者：申中杰[1] 陈雪峰[1] 何正嘉[1] 孙闯[1] 张小丽[1] 刘治汶[1] 

机构地区：[1]西安交通大学机械制造系统工程国家重点实验室,西安710049

出　　处：《机械工程学报》2013年第2期183-189,共7页Journal of Mechanical Engineering

基　　金：国家自然科学基金(51035007;51225501);博士点优先发展(20110201130001);教育部长江学者创新团队资助项目

摘　　要：为解决有限状态数据下滚动轴承剩余寿命难以估算的问题,提出一种基于相对特征和多变量支持向量机(Multivariablesupport vector machine,MSVM)的剩余寿命预测的新方法。该方法利用不受轴承个体差异影响的相对方均根值(Relative rootmean square,RRMS)评估轴承性能衰退规律,运用相关分析选取敏感特征作为输入,构造兼顾多变量回归和小样本预测双重优势的MSVM模型预测轴承剩余寿命。与单变量支持向量机相比,MSVM克服了结构简单、信息匮乏等缺点,实现小样本数据潜在信息的最大挖掘。运用仿真数据和轴承全寿命试验数据对预测模型进行检验,结果表明MSVM可在小样本条件下利用尽可能多的有效信息获得准确的预测结果,具有较强的工程使用价值和通用性。Novel prediction method is proposed based on the relative features and multivariable support vector machine （MSVM） to estimate the rolling bearing remaining life under limited condition data. The relative root mean square （RRMS） with ineffectiveness of the bearing individual difference is used to assess the performance degradation, and sensitive features are selected as input by correlation analysis. Meanwhile, MSVM is structured to predict the remaining life, which has the advantages of multivariable prediction and the small samples prediction. Unlike univariate SVM, MSVM overcomes the simple structure and the lack of information, and excavates the potential information of small sample as much as possible. The simulation and the bearing run-to-failure tests are carried out to inspect the prediction model, and the results demonstrate that MSVM can utilize the effective information as much as possible for the more precise results with the practical values and generality.

关 键 词：剩余寿命预测 相对方均根值 性能衰退评估 多变量支持向量机 

分 类 号：TH17[机械工程—机械制造及自动化] TP18[自动化与计算机技术—控制理论与控制工程]