航空学报 > 2015, Vol. 36 Issue (2): 555-563   doi: 10.7527/S1000-6893.2014.0257

基于半随机滤波-期望最大化算法的剩余寿命在线预测

冯磊1, 王宏力1, 司小胜1, 杨晓君1, 王标标2   

  1. 1. 第二炮兵工程大学, 西安 710025;
    2. 96275部队, 洛阳 471003
  • 收稿日期:2013-12-30 修回日期:2014-09-15 出版日期:2015-02-15 发布日期:2014-09-19
  • 通讯作者: 杨晓君 Tel.: 029-84741447 E-mail: yxj029@163.com E-mail:yxj029@163.com
  • 作者简介:冯磊 男,博士,讲师。主要研究方向:故障诊断、设备寿命预测与健康管理。E-mail:fengl1983@126.com;王宏力 男,博士,教授,博士生导师。主要研究方向:故障诊断、设备最优化管理、精确制导。Tel:029-84741361 E-mail:whl741361@sohu.com;杨晓君 男,博士,讲师。主要研究方向:设备寿命预测、数字信号处理、目标定位与跟踪。Tel:029-84741447 E-mail:yxj029@163.com
  • 基金资助:

    国家自然科学基金(61174030,61304240,61374126,61473094);中国博士后科学基金(2014M552589)

Real-time residual life prediction based on semi-stochastic filter and expectation maximization algorithm

FENG Lei1, WANG Hongli1, SI Xiaosheng1, YANG Xiaojun1, WANG Biaobiao2   

  1. 1. The Second Artillery Engineering College, Xi'an 710025, China;
    2. Unit 96275, Luoyang 471003, China
  • Received:2013-12-30 Revised:2014-09-15 Online:2015-02-15 Published:2014-09-19
  • Supported by:

    National Natural Science Foundation of China (61174030, 61304240, 61374126, 61473094); China Postdoctoral Science Foundation (2014M552589)

摘要:

剩余寿命(RL)预测是设备预测维护的关键环节。准确在线预测能够为维护策略的实时安排提供更加精确的技术支持,有效避免失效的发生。工程实际中,反映设备退化过程的性能指标往往不能直接监测,为解决隐含退化过程的剩余寿命在线预测问题,提出一种基于半随机滤波-期望最大化(EM)算法的预测方法。首先以剩余寿命为隐含状态,构建状态空间模型描述直接监测数据与设备剩余寿命间的随机关系。为实现单个设备剩余寿命的在线预测,依据到当前时刻为止的监测数据,采用扩展卡尔曼滤波(EKF)与期望最大化算法相互协作的方法实时估计与更新模型未知参数和剩余寿命分布。最后,将该方法用于惯性测量组合(IMU)剩余寿命在线预测问题,实验结果表明该方法能够提高预测的准确性并减少预测的不确定性。

关键词: 剩余寿命, 预测, 半随机滤波, 期望最大化, 扩展卡尔曼滤波

Abstract:

The prediction of residual life (RL) is the key of the predictive maintenance for engineering equipment. Accurate and real-time prediction can provide more effective decision support to the subsequent maintenance schedule and avoid the failure effectively. In engineering practice, the performance index reflecting the degradation process of the equipment is generally not observed directly. To tackle the residual life problem under hidden degradation, a prediction method based on semi-stochastic and expectation maximization (EM) algorithm is proposed in this paper. First, the residual life is taken as the hidden state and the prediction model is constructed by building the stochastic relationship between the residual life and monitoring data. Secondly, based on the monitoring data up to the current time, a collaborative method by the extended Kalman filter (EKF) and expectation maximization algorithm is presented to achieve a real-time estimation and updating of the residual life distribution and unknown model parameters. Finally, the proposed method is validated by the application to the inertial measurement unit (IMU) and the results indicate that the method can improve the accuracy and reduce the uncertainty of the estimated residual life.

Key words: residual life, prediction, semi-stochastic filter, expectation maximization, extended Kalman filter

中图分类号: