基于动态贝叶斯网络推理分析的复杂多部件系统视情维修决策研究

In order to provide solutions to condition-based maintenance (CBM) optimization problems for multi-component systems, this research aims to develop a novel modeling approach for complex multi-component systems and the corresponding optimization theory for dynamic CBM optimization problems, using inferring analysis with dynamic Bayesian network. Examples based on real maintenance problems are supposed to demonstrate the research results..To free the limitations set by existing analytical approaches in modeling the reparable system for CBM problems, a modeling approach based on system physics and using dynamic Bayesian network is proposed, which is capable of describing the system degradation, the dependencies between components and the maintenance actions. Using this modeling approach, a quantization approach based on inferring analysis with dynamic Bayesian network is developed to provide system status information from both time and space perspectives in order to reveal maintenance effects on the system. Finally, for dynamic CBM policy, a distributed intelligent optimization algorithm is proposed to improve the optimization efficiency, with the help of intelligent optimization theory and graph decomposition theory. Research will utilize real cases instead of numerical analysis. .This research is going to establish a link between CBM lab research and the real maintenance problems, thus a real progress of CBM for multi-component systems in both theory and applications can be expected.

针对复杂多部件系统的视情维修决策问题，本项目基于动态贝叶斯网络及其推理分析，建立一套面向复杂可修系统的维修模型建模理论，同时根据动态复杂视情维修策略优化的需求提出一套相应的优化决策方法，最后通过对真实系统维修问题的研究验证研究成果。.从系统运行机理出发，建立可修系统性能退化特性、部件间依赖关系和维修行为的动态贝叶斯网络模型表征体系，解除现有解析方法对系统的简化和约束条件。提出基于动态贝叶斯网络推理分析的维修效果时空量化方法，为策略优化提供维修前后系统状态在时空域的准确变化信息。提出多层次约束下的多部件视情维修策略，并基于智能优化理论和图模型分解理论提出分布式智能优化技术，提高动态策略的优化效率。区别于现有研究中的抽象案例，采用与实际系统对应的仿真案例和真实案例开展研究和验证研究成果。.项目研究能够让研究成果与实际需求形成直接映射，切实有效地推进视情维修理论的发展和其在实际问题中的应用。

针对复杂多部件系统的视情维修决策问题，本项目从能量分析角度，提出了一套描述系统性能退化特性与部件间依赖关系的统一表征体系，并构建了该体系的建模方法及其贝叶斯网络模型结构。该方法将各部件的退化过程映射至能量域，解除了现有解析方法对系统的简化和约束条件，同时可以通过代数运算方式将处于不同物理场的部件联合，简化了部件个体建模及部件间依赖关系的描述。.基于上述多部件系统模型，从不同维修层次需求出发，提出了一种系统级、现场可更换单元级与车间可更换单元级三层结合的多层级约束视情维修策略；同时，引入了诊断策略成本对维修的影响，对诊断策略与维修策略进行了联合优化，平衡了监测与维修的成本。.为优化维修策略，提出了一种基于半马尔科夫决策过程的优化算法，解决了具有复杂连续状态退化过程的单部件视情维修阈值优化问题；在此基础上，提出了一种基于量子遗传算法的智能优化算法，解决了多部件系统各部件维修阈值的联合优化问题，提高了策略的优化效率。.采用三个实际工程系统为研究案例，验证了系统建模、维修策略及优化算法的有效性。