基于指数平滑和改进增量SVR的齿轮寿命预测研究  

Research of Remaining Useful Life Prediction of Gear based on Exponential Smoothing and Improved Incremental SVR

作　　者：刘珍翔 陈晓慧[1] 

机构地区：[1]重庆大学机械传动国家重点实验室

出　　处：《机械传动》2016年第5期30-34,共5页Journal of Mechanical Transmission

基　　金：国家自然科学基金重点项目(51035008);机械传动国家重点实验室自主项目(SKLMT-ZZKT-2012MS 02);重庆市自然科学基金(CSTC;2009BB3365)

摘　　要：针对齿轮剩余使用寿命预测这一问题进行了研究,提出了一套基于二次指数平滑和改进增量支持向量回归(Support vector machine for regression,SVR)的齿轮剩余寿命预测新方案。该方案利用主成分分析法(Principal component analysis,PCA)筛选组成融合指标集,再进行二次指数平滑处理(Double exponential smoothing,DES)作为输入,构建基于增量拉丁超立方采样(Latin hyper-cube sampling,LHD)和Direct搜索算法的改进增量SVR预测模型。与基于原始特征值的标准SVR模型相比,新模型克服了信号随机性和突变性的干扰,并获取了最新状态信息的内在演变趋势,寻优效果和预测性能得到明显改善。用齿轮全寿命周期试验数据对新模型进行验证分析,结果表明,改进增量SVR预测模型获得了更可靠、更稳定的预测结果,具有一定工程实用价值。The remaining life prediction problem of gear is studied and a new set of gear life prediction program is put forward based on double exponential smoothing（ DES） and improved incremental support vector machine for regression（ SVR）. The new program utilizes principal component analysis（ PCA） to filtrate fusion index set,and then using DES to process fusion index set as the input of improved incremental SVR. Meanwhile,the improved incremental SVR prediction model is constructed based on incremental Latin hyper- cube sampling（ LHD） and Direct search algorithm. Compared with the standard SVR prediction model based on original characteristic values,the interference of signal randomness and mutation of the new model sre overcome,furthermore,the latest evolution trend information of the gear status is obtained,the optimization effect and prediction performance is obviously improved. The validation analysis of the new model is carried out by using the full life- cycle test data,the results show that the improved incremental SVR prediction model obtained a more reliable,more stable predictions,it has certain practical value in engineering.

关 键 词：二次指数平滑 融合指标集 改进增量SVR 寿命预测 

分 类 号：TH132.41[机械工程—机械制造及自动化]