基于藤蔓树降维剪枝的多部件系统剩余寿命预测

Residual life prediction of multi-component system based on vine tree reduced-dimensional reduction pruning

随着多部件系统日益复杂化和智能化,系统中多个部件间协同工作,其退化的相互影响往往不可忽略。因此,在考虑多部件系统部件间相互影响的基础上,提出了一种构造藤蔓树降维剪枝二元Copula函数来对高维变量间相关结构进行建模的剩余寿命预测方法。首先通过Lévy过程来建模多部件系统的退化模型,利用Akaike信息准则进行Copula函数的最优遴选。其次通过Copula函数来表征高维多变量下两两部件间的双向随机相关性,采用藤蔓树降维剪枝的方法来对高维变量间相关结构进行降维,并进行多部件系统部件的实时剩余寿命预测建模;接着考虑单个部件异质性的影响下,利用贝叶斯参数估计和极大似然估计结合的方法来实现对随机模型以及随机超参数的动态更新。最后通过商用模块化航空推进系统仿真涡轮发动机模型来验证所提出模型的合理性和有效性。

With increase in complexity and intelligence of multi-component system,multi-component in system work synergistically,mutual effects of their degradation can’t be ignored.Here,based on considering mutual effects among components in multi-component system,a residual life prediction method was proposed to model correlated structures among high-dimensional variables using a bivariate Copula function constructed with vine tree reduced-dimensional reduction pruning.Firstly,the degradation model of a multi-component system was modeled using Lévy process,and Akaike information criterion was used to choose the optimal Copula function.Secondly,Copula function was used to characterize bidirectional random correlation between each two components under high-dimensional multi-variable.The vine tree reduced-dimensional reduction pruning method was used to reduce dimensions for correlated structures among high-dimensional variables,and perform modelling for real-time residual life prediction of components of multi-component system.Next,considering effects of individual component heterogeneity,a combination of Bayesian parametric estimation and maximum likelihood estimation was used to dynamically update random models and random hyperparameters.Finally,the rationality and effectiveness of the proposed model were verified with commercial modular aero-propulsion system simulation turbine engine model.