基于支持向量机和卡尔曼滤波的机械零件剩余寿命预测模型研究  

Research on Remaining Useful Life Predictive Model of Machine Parts Based on SVM and Kalman Filter

作　　者：于震梁[1] 孙志礼[1] 曹汝男 王鹏 YU Zhen-liang;SUN Zhi-li;CAO Ru-nan;WANG Peng(School of Mechanical Engineering and Automation, Northeastern University, Shenyang 110819, Liaoning, China)

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：《兵工学报》2018年第5期991-997,共7页Acta Armamentarii

基　　金：国家自然科学基金项目(51775097)

摘　　要：现有机械零件剩余寿命预测模型在建模过程中,无法同时采用已有数据库数据及被预测产品实时退化数据,为了弥补其不足,提出一种支持向量机(SVM)和非线性卡尔曼滤波相结合的机械零件剩余寿命预测模型。根据现有全寿命试验数据训练所得的SVM回归模型,建立非线性卡尔曼滤波状态更新方程,依据机械零件退化特征构造时间更新方程,设定初始剩余寿命值及其方差,通过逐步迭代计算各时刻剩余寿命估计值及一定置信水平的置信区间。该计算模型能够充分利用现有零件与同类零件全寿命试验数据和被预测零件的实时状态退化数据,实现剩余寿命预测。以某型号滚动轴承为例,验证了所提出剩余寿命预测模型的精度、稳定性及工程应用价值。A new remaining useful life （RUL） predictive model for machine parts is proposed, which combines support vector machine （SVM） and non-linear Kalman filter. The proposed model is expected to fix the awkward situation in that the degradation data from both database and the predicted part cannot be used at the same time by most of the exiting RUL models. The SVM regression model trained by data from full-life tests is treated as the status update equation of non-linear Kalman filter. The time update e- quation is constructed according to the degradation characteristics of machine parts. RUL estimations and corresponding confidence intervals of point-in-time are computed iteratively after setting the initial RUL value and its variance. The proposed model makes the best use of data from both full-life tests of the same or similar parts and current part during its degradation. The accuracy of RUL estimation, the stability and practical values of the proposed model are illustrated by analyzing a certain type of antifriction bearings.

关 键 词：机械零件 剩余寿命 支持向量机 非线性卡尔曼滤波 置信区间 

分 类 号：TB114.37[理学—概率论与数理统计]