知识与数据融合的可靠性定量模型建模方法

可靠性定量设计的关键是建立可靠性定量模型。现有的可靠性定量模型建模方法主要基于设计人员对产品对象故障规律的知识,包括故障模式、环境扰动、故障机理等。但知识固有的有限性和不完整性必然会给可靠性定量模型带来模型误差和输入参数的不确定性。针对这个问题,提出了基于贝叶斯理论融合知识和数据的可靠性定量模型建模方法,量化并更新模型误差和输入参数的不确定性。为此,首先说明了知识与数据融合的可靠性定量模型建模工作,建立了知识与数据融合的可靠性定量模型建模框架;接着阐述了基于贝叶斯理论的知识与数据融合原理;然后介绍了基于贝叶斯理论融合知识与数据的通用方法,并分别针对性能波动数据和性能退化数据2种常见数据类型进一步详细讨论了各自适用的贝叶斯融合方法;最后通过机载轴向柱塞泵的案例验证了前述方法的可行性和有效性。 

Integrated method of knowledge and data for quantitative reliability modeling

Keys of quantitative reliability design lie in the establishment of the quantitative reliability model. Current modeling methods mainly rely on design staff's knowledge on product failure rules, including failure modes, environmental disturbances, failure mechanisms, etc. However, the inherent finiteness and imperfection of knowledge are bound to bring both model error and input uncertainties to the quantitative reliability model. To address this problem, we proposed a knowledge-and-data integrated Bayesian modeling method to develop the quantitative reliability model, quantifying the model error and input uncertainties. First of all, the tasks of the knowledge-and-data integrated modeling of the quantitative reliability model were explained, and the corresponding framework was established. Then the principle of Bayesian integration of knowledge and data was clarified. After that, the general method of Bayesian integration of knowledge and data was proposed, and two specific Bayesian integration methods for both performance fluctuation and degradation data were addressed respectively. Finally, the effectiveness and feasibility of the proposed method were illustrated by a case of an airborne axial piston pump.