故障预测和系统健康管理的贝叶斯推断

Prognostics and system health management (PHM) is a modern reliability study to improve the safety and performance of components and systems, and thus there has been a growing interest in a variety of fields including electronics, smart grid, nuclear plant, power industry, aerospace and military application, fleet industrial maintenance, and public health management. PHM is a systematic approach for failure prevention by monitoring the health/status of products and systems, predicting failure progression, and mitigating operating risks through repair or replacement. Prognostics can yield an advance warning of impending failure in a system, thereby helping in making maintenance decisions and executing preventive actions prior to failure occurrence to extend system life. Remaining useful life estimation, defined as the length from current time to the end of the useful life, is one of the vital indexes in PHM, and has been commonly used in reliability studies with applications. While the application of health/status monitoring is established, degradation data that describes quality characteristics over time is collected and model-based assessments based on degradation data are commonly used for remaining useful life estimation. Therefore, reliability assessment has been one of the important research topics in PHM, and there are many unsolved problems that are statistically challenging and important in engineering viewpoint. The project is concerned with some fundamental issues of statistical analysis for remaining useful life estimation for PHM. Due to sophisticated modeling of high reliable components and systems, classical approach may fail or become difficult to make inference on remaining useful life. Bayesian approach has become an alternative for analyzing the reliability data because of its computational and methodological advantages. Moreover, the prior information combined with the data produces more precise estimates than the data alone. Therefore, the applicability of Bayesian methods in reliability study has increased in recent years. The main objective of this project is to apply the Bayesian inference in complicated models in PHM where classical approaches are difficult to implicate.

故障预测和系统健康管理（PHM）是一项现代的可靠性研究，旨在提升部件和系统的安全性和性能，因此它在各个领域引起了大众日益浓厚的兴趣。PHM 是一套系统性故障防止方法，它通过监视产品和系统的健康来防止故障的发生。故障预测可对系统即将发生的故障进行预警，从而在故障出现之前做出预防作业，以延长系统寿命。在建立了健康监视应用时，在退化数据的基础上采用基于模型的评估方法对剩余使用寿命进行估计。本项目涉及针对PHM 的剩余使用寿命进行统计建模和分析。由于高可靠性元件和系统模型上的复杂性，传统方法包括的最大似然估计可能无法或难以对剩余使用寿命做出推断。贝叶斯方法以其在计算和方法上的优势，已成为分析可靠性数据的一项替代方法。此外，与数据结合的先验信息可给出比仅采用数据更为精确的估计。本项目的主要目的在于在PHM中对传统方法难以实现的复杂模型进行贝叶斯推断

复杂工程系统应实现高安全性、高可靠性和低寿命周期费用的目标。预测与系统健康管理则是工业界实现这些目标的全新理念和技术。从状态监测到状态预测,从维修到健康管理是一个质的飞跃。因此，研究基于状态预测的复杂工程系统健康管理理论和方法就成为现实的必然要求。基于预测的系统健康管理是指在系统状态预测基础上的系统生产计划,维修和备件管理。目前在复杂工程系统层次上，还没有完善的健康评估和预测方法。利用系统状态预测信息进行一体化复杂工程系统健康管理的理论和方法也很少见到。我国是一个制造大国,同时也是一个工程装备使用大国。如何将基于预测的 系统健康管理的概念应用于我国现代工业装备管理具有重大的理论和现实意义。因此，本项目将重点研究如下两个关键问题1)复杂工程系统健康评估和预测; 2)基于预测的复杂工程系统健康管理。课题组开展了可充电电池元件在 ARBIN BT2000 中的退化实验，收集了多组退化数据。此外，课题组对轴承退化数据进行建模和分析。通过本项目的深入研究，从实用和新颖的角度建立可充电电池系统和轴承健康管理和预测的数据分析融合的关键技术，实现数据分析平台，满足多种实际具体的应用需求，推动健康管理和预测技术在移动互联网和云计算时代走向实际应用。