机械产品多变量相依退化建模方法研究综述

Review of Multivariate Dependent Degeneration Modeling Methods for Mechanical Products

退化建模是开展机械产品可靠性评估与剩余寿命预测、指导产品使用维护管理决策优化的关键技术。随着机械产品日趋结构复杂化与功能多样化方向发展,多变量退化数据由于能够更全面地反映产品健康状态,其建模方法在近几年受到研究者们的广泛关注,特别是针对多变量退化过程中普遍存在的随机相依性,构建诸多实用有效的多变量相依退化模型,提高了复杂机械产品寿命与可靠性的评估精度。然而,目前还没有形成针对该主题的系统性综述。为促进多变量相依退化建模方法的发展与应用,从研究现状、发展趋势及应用研究等方面,对现有的多变量相依退化建模方法进行回顾和总结。首先,介绍三类常用多变量相依退化建模方法的原理及其优缺点;接着,探讨考虑多源随机性、外部协变量以及时变随机相依性等因素影响下的多变量相依退化模型扩展研究;随后,讨论多变量相依退化模型在加速试验方案设计、剩余寿命预测和维修策略优化中的应用研究;最后,针对未来值得深入研究的方向进行归纳和总结。

Degradation modeling has been recognized as a critical approach for mechanical products in conducting reliability assessments and remaining useful life(RUL)predictions,as well as guiding the optimization of operation and maintenance actions.With the increasing complexity in structure and diversification in the function of modern mechanical products,more interest and attention have been paid to the modeling of multivariate degradation data as it can comprehensively reflect the health status of products.Especially,focusing on the stochastic dependence that is widespread in multivariate degradation processes,many practical and effective multivariate dependent degradation(MVDD)models have been proposed,based on which the accuracy of reliability assessment and RUL prediction of products is greatly enhanced.However,there is no systematic review focusing specifically on this topic.Therefore,the research status,recent advances,and applications of existing MVDD modeling methods are reviewed to facilitate their developments and applications.First,three commonly used MVDD modeling methods are introduced in detail,including their principles,advantages,and disadvantages.Then,extended studies of MVDD models involving the effects of multi-source variability,external covariates,and time-varying stochastic dependences are investigated.In addition,applications of MVDD models for accelerated test design,RUL prediction,and maintenance strategy optimization are discussed.Finally,several possible directions for further research are summarized.