Abstract: The braking system had the characteristics of diverse working conditions, functional levels and complex operation relationship between internal structures, and was often affected by wear, vibration, fatigue, impact, over-maintenance and other factors, which caused the problems of multiple fault states and multiple failure modes between normal operation and complete failure of the system and components in the actual operation process. Through the reliability analysis method based on the super ellipsoid Bayesian network, the multi-state problem of the Bayesian network describing the fault states of components and events was studied. Meanwhile, combing with the super ellipsoid Bayesian network model to constrain the interval variables, the super ellipsoid Bayesian network model was constructed, which was introduced into the reliability analysis of motor car breaking, and compared with the analysis results of interval Bayesian network. The results show that compared with the interval Bayesian network, the analysis results of the super ellipsoid Bayesian network model have smaller probability interval range and higher accuracy, which not only solves the conservative problem of interval calculation results of interval Bayesian network, but also improves the accuracy of reliability analysis of braking system. Then, by solving the important parameters such as sensitivity and posteriori probability of the braking system, the weak links of the proposed system and the high-risk events affecting the reliability of the braking system are found out, which provides theoretical guidance for the formulation of the maintenance strategy and technical transformation.

Key words: vehicle engineering; super ellipsoid of Bayesian networks; polymorphism; sensitivity; posterior probability

摘要： 制动系统工作条件多样、功能层次、内部结构间运行关系复杂等特性，经常受到磨损、振动、疲劳、冲击、以及过度维修等多种因素影响，在实际运行过程中，会导致的系统及部件在正常工作和完全失效两状态之间存在着多故障状态、多失效模式问题。通过基于超椭球贝叶斯网络的可靠性分析方法，研究了贝叶斯网络描述部件及事件的故障状态的多态问题，同时结合超椭球模型对区间变量进行约束，构建超椭球贝叶斯网络模型，引入动车制动可靠性分析中，并与区间贝叶斯网络分析结果进行对比。结果表明：超椭球贝叶斯网络模型分析结果较区间贝叶斯网络的概率区间范围更小，精度更高，既解决了区间贝叶斯网络区间计算结果的保守问题，又提高了制动系统可靠性分析的准确性，然后通过求解制动系统的灵敏度、后验概率等重要参数，找出了系统的薄弱环节和影响制动系统可靠性的高风险事件，为检修策略的制定及技术改造提供理论指导。

关键词: 车辆工程；超椭球模型贝叶斯；多态性；灵敏度；后验概率