基于退化建模的剩余寿命预测

发布时间：2018-12-16 05:22

【摘要】：在实际工程应用中,系统或设备可能会发生各种各样的故障,需要进行视情维修(Condition Based Maintenance,CBM)。而剩余寿命预测(Remaining Life,RL)是视情维修过程中的核心问题之一;另一方面,在废旧产品回收利用和再制造过程中通过剩余寿命预测才能判断回收部件是否值得再制造,对再制造部件也需要预测其剩余寿命从而确定其价值。因此设备或系统的剩余寿命预测具有重大的研究意义和实用价值。首先,本文综述了系统或设备剩余寿命预测的两种方法:基于系统或设备损伤机理的模型方法和基于数据的统计分析方法,针对数据处理的建模过程,介绍了基于维纳过程,马尔科夫链,泊松过程的方法;然后综述了相应的参数估计方法:极大似然估计方法(MLE),期望最大化算法(EM)和贝叶斯方法。其次,针对冲击载荷对降级系统的影响分别建立了三种模型:静态冲击退化模型,累积冲击退化模型,极端冲击退化模型。并推导出这三种模型的剩余寿命概率密度函数,数值试验结果表明考虑冲击载荷进行剩余寿命预测提高了预测的精度。接着针对具有工作状态和存储状态的降级系统,采用极大似然估计法和EM算法估计了模型中的未知参数,并使用蒙特卡罗方法预测剩余寿命,从而推导出剩余寿命分布函数。最后对相关数据进行对比和分析,验证了所提方法的有效性。最后对本文所提出的方法进行了总结和展望。由于缺少相应的实证实验,很难获得足够的相关数据和定量信息,因此本文提出的两种方法的实际应用效果还需要进一步验证。另外利用连续时间马尔科夫链近似描述系统工作状态和存储状态的降级过程条件过于苛刻,在这方面还需要进一步研究和发展。
[Abstract]:In practical engineering applications, the system or equipment may occur a variety of failures, the need for situational maintenance (Condition Based Maintenance,CBM). Residual life prediction (Remaining Life,RL) is one of the core problems in the process of situational maintenance. On the other hand, in the process of recycling and remanufacturing of waste products, the residual life prediction can be used to determine whether the recovered parts are worth remanufacturing or not, and it is also necessary to predict the residual life of the remanufactured parts to determine their value. Therefore, the residual life prediction of equipment or system has great significance and practical value. Firstly, this paper summarizes two methods for predicting the residual life of system or equipment: the model method based on the damage mechanism of system or equipment and the statistical analysis method based on data. In view of the modeling process of data processing, the Wiener process is introduced. Markov chain, Poisson process method; Then the corresponding parameter estimation methods are summarized: maximum likelihood estimation (MLE), expectation maximization algorithm (EM) and Bayesian method (Bayesian method). Secondly, three models are established for the impact of impact load on the degradation system: static impact degradation model, cumulative impact degradation model and extreme impact degradation model. The probability density function of residual life of these three models is derived. The numerical results show that the prediction accuracy of residual life is improved by considering the impact load. Then, for the degraded system with working state and storage state, the unknown parameters in the model are estimated by maximum likelihood estimation and EM algorithm, and the residual life is predicted by Monte Carlo method, and the residual life distribution function is deduced. Finally, the validity of the proposed method is verified by comparing and analyzing the relevant data. Finally, the methods proposed in this paper are summarized and prospected. Due to the lack of corresponding empirical experiments, it is difficult to obtain enough relevant data and quantitative information. Therefore, the practical application results of the two methods proposed in this paper need to be further verified. In addition, it is too harsh to use continuous time Markov chain to approximate describe the working state and storage state of the system, which needs further research and development.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：O213.2

【参考文献】