Abstract: The functional structure of various industrial systems is becoming more and more complex,and there is a complex stochastic dependence among the components,which is an important factor that cannot be ignored in the modeling of the remaining useful life prediction of the system.For the multi-component system with stochastic dependence,a nonparametric kernel density prediction method for real-time residual life of multi-component system based on Copula theory was proposed on the basis of studying the stochastic dependence characteristics among components and its effect on continuous degradation state.The degradation distribution of each component was obtained by kernel density estimation;then the Copula function was used to characterize the stochastic dependence of component degradation,and the Akaike information criterion was used to optimize the Copula function;then a real-time recursive prediction model of component residual life considering the influence of continuous degradation stochastic dependence was established.The validity and accuracy of the model were verified by the experiment on a gearbox test-bed.

Key words: remaining useful life prediction, multi-component system, stochastic dependence, Copula function, kernel density estimation

摘要： 各类工业系统的功能结构日趋复杂,部件间存在复杂的随机相关性是进行系统剩余寿命预测建模中不可忽视的重要因素。针对多部件系统,在研究部件间随机相关性特征及其对连续退化状态影响的基础上,提出基于Copula理论的多部件系统实时剩余寿命非参数核密度预测方法。首先通过核密度估计得到各部件的退化分布函数;并采用Copula函数表征部件退化的随机相关性,利用赤池信息准则进行Copula函数优选;然后建立考虑存在连续退化随机相关性影响条件下可实时递推的部件剩余寿命预测模型;最后通过齿轮箱试验台进行试验,验证了所提模型的有效性和准确性。

关键词: 剩余寿命预测, 多部件系统, 随机相关性, Copula函数, 核密度估计