不确定健康状态驱动的预测性维修决策优化理论与方法研究

Current predictive maintenance technique suffers the problems of low precision in Health State Assessment (HSA), disconnection between HSA and Maintenance Decision (MD), as well as the weak compatibility of MD model regarding the different applications. Motivated to solve these problems and meet the requirement of integrated HSA/MD technique in predictive maintenance, the project innovates the traditional model-based/data-driven HSA and static population reliability based MD approaches. The main contents of this project include: 1) a novel HSA model which describes the relationships among system health degradation, observable fault symptoms and usable tests is constructed based on failure evolution-testability modeling mechanism and hierarchical modeling method; 2) a quantitative static/dynamic HSA reasoning model under imperfect tests is formulated based on Bayes rules, and an adaptive Lagrangian relaxation algorithm is proposed to solve its NP-hard problem; 3) the uncertainty of HSA and its quantitative generalization form are analyzed from the view of predictive maintenance, and a generalized maintenance decision optimization model taking the uncertainty assessment results into account is presented based on renewal-reward theory; 4) a case study of an aerospace mechanical and electrical servo-system to validate the approaches in the project is provided. The proposed theory and method enable the perfect integration between HSA and MD optimization, provide a new and operable way to equipment predictive maintenance, and have great significance to equipment maintenance development.

为解决目前预测性维修中健康状态评估准确性低、评估结果对维修决策支撑性弱、维修决策模型普适性差的问题，本项目瞄准预测性维修对健康状态评估与维修保障决策一体化技术的需求，革新传统的基于数据或物理模型的健康状态评估方法和基于总体静态可靠性特征的维修决策优化方法。基于故障演化可测性建模机制和分层建模思想描述系统健康状态与可测征兆、可用测试关联关系；基于贝叶斯推理理论构建不完美测试下的健康状态静/动态推理模型，针对模型的NP难特性，提出基于自适应拉格朗日的求解策略；根据预测性维修需求分析健康状态评估的不确定性及其泛化形式，基于更新过程理论建立不确定评估结果驱动的维修决策优化模型；以航天机电伺服系统为对象，验证所提模型和方法的有效性。项目的研究成果使得健康状态评估与维修决策优化无缝融合，为装备的预测性维修提供一条新的切实可行的技术途径，对实现装备维修保障模式的革新与发展具有重要意义。

为解决目前预测性维修中健康状态评估准确性低、评估结果对维修决策支撑性弱、维修决策模型普适性差的问题，本项目瞄准预测性维修对健康状态评估与维修保障决策一体化技术的需求，革新传统的基于数据或物理模型的健康状态评估方法和基于总体静态可靠性特征的维修决策优化方法。基于故障演化可测性建模机制和分层建模思想描述系统健康状态与可测征兆、可用测试关联关系；基于贝叶斯推理理论构建不完美测试下的健康状态静/动态推理模型，针对模型的NP难特性，提出基于自适应拉格朗日的求解策略；基于更新过程理论建立不确定评估结果驱动的维修决策优化模型。项目的研究成果使得健康状态评估与维修决策优化无缝融合，为装备的预测性维修提供一条新的切实可行的技术途径，对实现装备维修保障模式的革新与发展具有重要意义。