| Space systems are complicated in nature with behavior of complex configuration and function,phased-mission,multi states,dynamic interaction between the components and continuous parameter variables,all of which are time dependent criteria that can influence the potential accident scenario evolvation and make the system safety risk modeling more difficulty.The conventional combinatorial model based on Event Tree/ Fault Tree(ET/FT)model can quantify static relationships between logical variables with limited capacity: the static model can't deal with multi-states,failures' dependencies and dynamic interaction,etc.,and the problem is frequently encountered in dynamic Fault tree model and is referred to as the state space explosion when mapped into a time-consuming Markov Chain.A Bayesian Network(BN)is a graphical inference technique and has become popular as a robust alternative to most classical methods benefiting from the conscious tradeoff between modeling advantages and model sizes,Its unique bidirectional inference mechanism which can be used either to predict the probability or to update the probability as well as diagnostic for risk analysis.Therefore,it is important to develop a new advanced method that can address the above mentioned problems,and the primary objective of this study is to propose a safety risk model based on Bayesian Networks for space systems with characteristics as complex configuration,phased dependency and dynamic interaction.Based on the translation and quantification of ET/FT model,this paper study the extend modeling method based on Bayesian Networks include: the multi-state and statistically dependent Event Tree auto mapping procedure,the time-dependent modeling method in event tree for phased mission and dynamic interaction extend modeling method.The details of the contents are as follow.1)The procedures for mapping and extending ET/FT model to Bayesian network are described systemically.After the graphical and numerical mapping rules of Fault Tree have been reviewed,the assumption terms,series multiple and the integration implicitly of the probabilistic common cause failure are extended by Bayesian networks.Due to the lack of deterministic logic constraint,the mathematical description of multi-state and statistically dependent Event Tree and the reduction algorithm of arcs result from complete conversion based on the judgment of conditional independence and merger of event sequence are presented.2)To solve the problem that the time-dependency make the ET/FT based risk analysis more difficult when ETs are typically used to portray progression of phase mission over time,the simple success/fails logic and non-sequential logic are discussed respectively,and a three-level hierarchical model has been developed and illustrated for respectively representing the entire mission states,the reliability of support subsystem and the component states.Furthermore,the methods of module time-dependency and entire model integration are presented for the complex computing problem of combinatorial model due to the shared basic events.3)In order to model space system dynamic evolution with interaction between components and continuous parameter variables such as pressure,temperature,liquid level and attitude,the discrete time and discrete states dynamic Bayesian networks,along with the determining methods of conditional probabilities of discrete states components with a view to random failure,failure on demand and repair etc.and interval values of continuous variables are presented.4)Three real subsystems(propellant distribution module,auxiliary power unit and level control cooling unit)of the spacecraft are cited to show the usefulness and suitability of the Bayesian Networks for the safety risk modeling in reference to space system with types of characteristics such as complex configuration and function,multi-states,cascade effect,common cause failure,time-dependent between pivotal event,and dynamic interaction. |
PDF Full Text Request