Research On System Reliability Analysis And Fault Diagnosis Methods Based On Bayesian Networks

Reliability is one of the important indicators in quality electrical products, it played a decisive role in products design, manufacturing, maintenance, using, testing, etc. How to deal with the influence of subjective factors, such as fuzziness, polymorphism and dynamic in analysis the reliability of mechanical and electrical products has an important guiding significance. For the mechanical and electrical products, the application of fault diagnosis technology also affects the effective using of the product. In the fault diagnosing, many factors affecting the process of fault diagnosis. The fault diagnosis decision sequential model which is more close to realistic situation will take various factors into consideration. The model will have great practical value for the maintenance and repairing of the mechanical and electrical products.First of all, variable function is used to represent fuzzy support radius of nodes’ fault states, a membership function with fuzzy support radius variable is constructed to describe the fault states, a failure probability calculation method of multi-state system based on Bayesian networks with variable fuzzy support radius is proposed. The variable is introduced to describe fuzzy support radius of membership function in the construction of membership function and based on it, the membership function is constructed. Introducing the established membership function with variable into Bayesian networks, the leaf node failure probability curve is obtained by analyzing leaf node failure probability of multi-state system, which is used by bucket elimination method. This method solves the fuzziness and uncertain of multi-states system effectively, and it has a much wider application.Secondly, in order to overcome limitation in processing fuzzy information and dynamic information of systems in the existing Bayesian networks reliability analysis methods, a new method of multi-state systems reliability analysis method based on Bayesian Networks which merged the dynamic and fuzziness of fault information is proposed, which expand the traditional Bayesian networks, and effectively solving the deficiencies which having the fuzziness and dynamic fault information of existing reliability analysis methods based on Bayesian network. The fuzzy set theory and component failure probability function of dynamic change are presented to Bayesian network, and building dynamic fuzzy subsets; The algorithm of fuzzy dynamic fault probability of leaf node fault state and fuzzy dynamic importance are proposed based on dynamic fuzzy subsets and Bayesian network characteristics; Getting the curve of fuzzy dynamic fault probability of leaf node fault state and fuzzy dynamic importance with the MATLAB software.Thirdly, for the fuzziness problem of interval-valued fuzzy subset boundary, the interval variable is taking into existing Bayesian network reliability analysis methods, a reliability analysis method based on an interval-valued triangular fuzzy Bayesian network is proposed in this paper. It expanded the application from a two-state system to a multi-state system, and assessed the reliability evaluation of the multi-state system. First, the interval-valued variables were introduced into a triangular fuzzy subset, and an interval-valued triangular fuzzy subset was built. Second, the algorithm of the defuzzified leaf node failure probability, interval-valued fuzzy posterior probability and interval-valued fuzzy importance were given based on the interval-valued triangular fuzzy subset and features of a Bayesian network. It was proved that the proposed method was feasible by comparing with T-S fuzzy importance analysis methods and fuzzy Bayesian network analysis methods. The results provide theoretical basis for further system fault diagnosis.Finally, the fuzzy multi-attribute problem of components’ fault information, a Bayesian network fault diagnosis model based on interval triangle fuzzy multi- attribute decision. In this model, Bayesian network causal dependencies are used to take fault multi-attribute hierarchies into Bayesian network structure, which reducing the complexity of multi-attribute hierarchy. In the condition that weight information is completely unknown, two kinds of gray correlation optimal models are constructed based on gray correlation analysis method. Sort meriting of faulty components, and giving the calculation method of interval triangle fuzzy multi- attribute decision based on gray correlation, it realized the fault diagnosis problem of system fuzzy multi-attribute.