The Research On Methods Of Safety Risk Modeling And Evaluation Based On Bayesian Networks

System safety risk modeling and evaluations are the major topics of safety researches. Based on the qualitative modeling and quantitative assessments of the risk event scenes, it assesses the level of system safety risk. System safety risk modeling and evaluations have great significance in guiding the design of system safety and controling the risk reasonably. With the increasing of the system complexity, safety risk analysis is facing new challenges.Bayesian Networks has been widely researched and applied in many fields because of its mathematical foundations and powerful modeling and analysis abilities. For the dynamic and complex factors of safety risk modeling and evaluations, such as the uncertainty of logical relationships, multiple states, failure correlations,maintenance factors and so on. We can solve the problems which exist in FT/ET by using Bayesian Networks. Based on comparative analysis of traditional methods and Bayesian Networks in safety risk modeling and evaluations, the main research contents are summarized as follows:(1) Firstly, this thesis introduces the basic theories of static and dynamic Bayesian Networks. And based on methods of safety risk modeling and evaluation by FT/ET, we summarize and research the methods of translating FT/ET into Bayesian Networks. And then Bayesian Networks modeling and evaluation methods are discussed against the problems like uncertain logic relationships and multiple states, etc.(2) This thesis makes comprehensive induction and classification for correlation questions in the process of safety risk modeling and evaluation. In order to solve the functionality-correlated failure in fault tree, the Probabilistic Dependency gate(PDEP) is introduced to extend and improve the traditional fault tree. But the safety risk modeling and evaluations based on Bayesian Networks of different correlation problems are made a special effort to study. And compared with traditional FT/ET analysis, the advantages of Bayesian Networks are discussed. By analyzing the case of highway tunnel fire safety event, the feasibility of Bayesian Networks safety risk modeling and evaluations of failure correlations is proved.(3) In the view of the maintenance factors that traditional FT/ET ignore in the system, we discuss the classifications and the concepts of repair policies for safety risk analysis, which are divided into four types: CR、SGR、SLR and SLR-min. By extending traditional fault tree, the Repair Box(RB) is introduced to make up the shortcomings that use traditional methods to model and assess repair factors. Then aiming at the shortcomings of the traditional methods, Bayesian Networks modeling and evaluation of the four different kinds of repair policies is researched. Through the case of CO2 removal system fault maintenance, the feasibility of Bayesian Networks for safety risk modeling and evaluations with repair factors is proved.