Predicting For Remaining Life Based On Degradation Modeling

In practical engineering applications, systems or equipment may have a variety of failures, thus Condition Based Maintenance(CBM) is required. Residual life(RL) is one of the core problem in the process of CBM; On the other hand, in the process of recycling and remanufacturing of used products, The RL prediction can be made to determine whether the recovery component is worth to be manufactured, and the RL of the remanufacturing component also need to predict to determine its value.Therefore, the prediction of the RL of the equipment or system is of great significance and practical value.Firstly, this paper summarizes two methods of the RL prediction of equipment or system: the model method based on system or equipment damage mechanism and the statistical analysis method based on data. In view of the process of data processing, this paper introduces the method of Wiener process, Markov chain, Poisson process; and then summarizes the corresponding parameters estimation methods: Maximum Likelihood Estimation method(MLE), Expectation Maximization Algorithm(EM) and Bayesian method.Secondly, three kinds of models are built for the shock load on the degradation system: the static shock degradation model, the cumulative shock degradation model and the extreme shock degradation model. The residual life probability density function of the three models is derived, and the numerical test results show that the prediction accuracy of the remaining life prediction is improved by considering the impact load.Then according to the degradation systems with work and store state, the MLE method and EM algorithm are used to estimate the unknown parameters in the model, the systems residual life is predicted by using Monte Carlo method, and then the RL distribution function is derived. Finally, the relevant data are compared and analyzed, and the effectiveness of the proposed method is verified.Finally, this paper summarizes and prospects the methods of remaining life prediction. Due to the lack of relevant practical experiments, it is difficult to obtain sufficient relevant data and quantitative information, so the practical application effectiveness of the two methods proposed in this paper is also needed to further verify. In addition, the use of the continuous time Markov chain approximation to describe the condition of downgrade process system state and storage state is too harsh, which still need further research and development in this area.