# 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

WANG Jian, SUN Zhi-li, YU Zhen-liang, CHAI Xiao-dong

Abstract: To improve the accuracy of remaining useful life estimation for machine parts, an interval estimation model was proposed based on the SVM (support vector machine). The linear theory and nonlinearity theory of SVM were briefly introduced, and the correlation between input variable and output variable was analyzed. Degraded index and remaining useful life of machine parts were treated as input variable and output variable, correspondingly. It was assumed that input variable and residual error were independent and the residual error’s distribution pattern was known. Distribution parameters of residual error were estimated by means of the MLE (maximum likelihood estimation). Then the confidence interval of SVM output variable was obtained under a certain 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. A numerical example was presented to show that the proposed model can estimate the remaining useful life confidence interval precisely with the engineering application values and generality.

Key Words: remaining useful life    SVM(support vector machine)    interval estimation    machine parts； confidence interval； MSE (mean squared error)

SVM在小子样预测等问题上表现突出，在时间序列预测、结构可靠性分析、回归分析、经济金融等领域有广泛的应用，机械零件剩余寿命预测研究中也被较频繁使用.Emmanuel等[6]通过分析轴承工作过程中振动信号提取退化信息，建立基于SVM的轴承剩余寿命点预测模型；Loutas等[7]，Caesarendraa等[8]以非线性退化零件或系统为研究对象，将贝叶斯方法应用到SVM建模过程中.

1 剩余寿命

 图 1 ZA-2115双列滚动轴承RMS曲线Fig.1 RMS curves of a ZA-2115 double row bearings

2 支持向量机模型2.1 支持向量机

 (1)

Cε分别为惩罚因子和不敏感因子，ξnξ*n为松弛因子.根据Karush-Kuhn-Tucker条件，以上问题可转化为凸二次优化问题.

 (2)

 (3)

2.2 输出变量区间估计

ζn(n=1,…,N)为样本采用极大似然法估计σ，极大似然函数为

，得

，则

tn(n=1,2,…,N)为传感器采集数据时间点，各时间点对应的机械零件的退化指标为x(tn),剩余寿命为y(tn)，若将x(tn)作为输入变量，将y(tn)作为输出变量，即可训练SVM得到x与y的对应关系模型，进而使用该模型实现机械零件剩余寿命区间估计.

3 参数选择

3.1 均方误差及交叉验证法

SVM模型推广能力及预测精度可由其在验证集T上的预测均方误差(mean squared error，MSE)度量，

3.2 变步长网格搜索算法

1) 初步确定γC的取值范围.为保证最后搜索到最优(γ,C)，初始范围应适当大些，如γC范围都是[10-15,1015].

2) 设定搜索步长.在双对数直角坐标系中设定搜索步长，如0.1.这样就在双对数直角坐标系中生成了一个二维网格，每个交叉点都对应一组(γ,C)取值，对每一组(γ,C)采用交叉验证法计算其对应的MSE值.在双对数直角坐标系中绘制等高线.

3) 缩小γC范围.由步骤2)中等高线缩小搜索范围，在双对数直角坐标系中缩小步长，重复步骤2)直至γC的范围满足预先设定的精度要求.

4 实例

 图 2 RRMS曲线Fig.2 Curves of relative root mean square

ε=0.5，初始γC范围都是[10-8,108].将S2∪S3作为训练集，采用3.2中改进步长网格搜索确定(γ,C)，得到最佳(γ,C)为

S4作为验证集预测4号轴承剩余寿命，并得到其置信度p0=0.9的置信区间，如图 3a所示.分别以S2∪S4和S3∪S4作为训练集计算(γ̂ 2,Ĉ 2)(γ^2,C^2)和(γ̂ 3,Ĉ 3)(γ^3,C^3)，进而得到回归模型2和回归模型3并估计σ2,σ3，再分别以S3和S2为验证集得到置信度p0=0.9的剩余寿命区间估计结果，如图 3b3c所示.表 1为3个回归模型参数估计结果.

 图 3 区间估计结果与真实值比较Fig.3 The comparisons between interval estimations and true values(a)—4号轴承； (b)—3号轴承; (c)—2号轴承.
 表 1 γ,C,σ估计值Table 1 Estimations of γ,C and σ

5 结论

 [1] Sikorska J Z, Hodkiewicz M, Ma L. Prognostic modelling options for remaining useful life estimation by industry[J]. Mechanical Systems and Signal Processing, 2011, 25 (5) : 1803 –1836. (1) [2] Si X S, Wang W B, Hu C H, et al. Remaining useful life estimation—a review on the statistical data driven approaches[J]. European Journal of Operational Research, 2011, 213 (1) : 1 –14. (1) [3] Sanders D A, Jasper G J, Alexander G. Improving ability of tele-operators to complete progressively more difficult mobile robot paths using simple expert systems and ultrasonic sensors[J]. Industrial Robot, 2010, 37 (5) : 431 –440. (1) [4] Si X S, Wang W B, Hu C H, et al. A Wiener-process-based degradation model with a recursive filter algorithm for remaining useful life estimation[J]. Mechanical Systems and Signal Processing, 2013, 35 (1/2) : 219 –237.(1) [5] Su C, Shen J Y. A novel multi-hidden semi-Markov model for degradation state identification and remaining useful life estimation[J]. Quality and Reliability Engineering International, 2013, 29 (8) : 1181 –1192. (1) [6] Emmanuel R, Rafael G. Remaining useful life estimation by classification of predictions based on a neuro-fuzzy system and theory of belief functions[J]. IEEE Transactions on Reliability, 2014, 63 (2) : 555 –566. (2) [7] Loutas T H, Roulias D, Georgoulas G, et al. Remaining useful life estimation in rolling bearings utilizing data-driven probabilistic E-support vectors regression[J]. IEEE Transactions on Reliability, 2013, 62 (2) : 821 –832.(2) [8] Caesarendraa W, Widodo A, Yang B S. Application of relevance vector machine and logistic regression for machine degradation assessment[J]. Mechanical Systems and Signal Processing, 2010, 24 (4) : 1161 –1171.(2) [9] 申中杰, 陈雪峰, 何正嘉, 等. 基于相对特征和多变量支持向量机的滚动轴承剩余寿命[J]. 机械工程学报, 2013, 49 (2) : 183 –189.  ( Shen Zhong-jie, Chen Xue-feng, He Zheng-jia, et al. Remaining life predictions of rolling bearing based on relative features and multivariable support vector machine[J]. Journal of Mechanical Engineering, 2013, 49 (2) : 183 –189. ) (2) [10] Lin C J,Weng R C.Simple probabilistic predictions for support vector regression[R].Taipei:National Taiwan University,2004. (3)

©️2019 CSDN 皮肤主题: 大白 设计师: CSDN官方博客