Research On Reliability Analysis Approach Of Complex System Based On Stochastic Computing

Reliability of complex systems is the key point to restrict their application and development due to the increasing of system complexity.Complex systems have the characteristics of fuzzy,multi-state,multi-phase,multi-mission,dynamic maintainability.The methods to evaluate the reliability of systems are encountering some technical problems caused by complex systems.The reliability analysis and design technology of complex systems plays an important role on preventing major accidents,improving reliable operation ability of important equipment or facilities.Reliability evaluation of complex systems becomes one of hot issues in the field of reliability engineering,and it meets the demand of our country’s important strategic.Hence,related theories and technical of complex systems reliability analysis urgently need to be solved.This thesis focuses on the reliability analysis and optimization of complex systems,due to the limitations of existing reliability analysis methods,such as high computational complexity,state space explosion and reliability analysis unconsidering dynamic of system.These complex systems include fuzzy systems,multi-state systems with dependent components,multi-state phased-mission systems with imperfect fault coverage,multi-state weighted k-out-of-n systems.The innovative work and research contents of this thesis are as follows: 1.This thesis constructed a hybrid stochastic computational(HSC)model of fuzzy systems,composed of expert evaluation unit and stochastic computing unit.In expert evaluation unit,the failure possibility of a basic event described by linguistic value is transformed to a triangular fuzzy number.In stochastic computational unit,the fuzzy number is replaced by its expected value,and then the value is encoded in stochastic non-Bernoulli sequence to calculate the reliability of fuzzy system.The theory validation and case studies indicate that,HSC approach can reduce the computational complexity and improve the convergence speed,the boundary value and failure probability distribution of system reliability can be calculated by the proposed approach.The computational efficiency and depth of reliability analysis are effectively increased compared to fuzzy expected value approach(FEVA),fuzzy arithmetic approach(FAO)and hybrid fuzzy Monte Carlo(HFMC)simulation approach.2.The dynamic universal generating function(DUGF)and multi-state Monte Carlo(MMC)simulation are main methods to evaluate the reliability of multi-state systems.However,the computational complexity of a DUGF method is exponential in the number of component states.MMC simulation encounters problems of slow convergence and long running time.This thesis proposed extended stochastic computational(ESC)models for the reliability evaluation of multi-state systems with dependent components.The performances and corresponding distribution probabilities of multi-state components are encoded in multiple-valued Bernoulli sequences.Hence,the number of component states is not a crucial factor to influence the computational complexities of ESC compared to DUGF.The simulation results indicate that the ESC approach is more efficient and more accurate than MMC simulation.The calculation accuracy of ESC is slightly lower than that of DUGF;however,the computational efficiency is significantly higher than that of DUGF.3.This thesis focuses on the reliability analysis method of multi-state phased-mission system with three different imperfect fault coverage conditions(listed as Element Level Coverage,Fault Level Coverage,and Performance dependent Coverage),based on the stochastic multiple-valued(SMV)approach.The simulation results indicate that the computational complexity of SMV approach is significantly lower compared to the UGF method.SMV approach is suitable for the analysis of complex multi-state phase mission systems with imperfect fault coverage.4.This thesis provided stochastic multiple-valued(SMV)for analyzing the reliability static multi-state k-out-of-n systems and dynamic stochastic multipe-valued(DSMV)models for dynamic multi-state k-out-of-n systems,then validating the models by the theoretical proof.The efficient of proposed models is verified by several cases studies.The results indicate that the run time of the SMV/DSMV approach increases linearly with the length of the sequences and the number of components,while the run time required by UGF,fuzzy universal generating function(FUGF)and extended universal generating function(EUGF)rapidly increases with the number of components and the number of components’ states.The SMV/DSMV approach is more efficient than the UGF 、FUGF and EUGF methods for a complex system with a larger number of component states.5.The system is supposed to achieving higher reliability with the increase of the number of spare components or redundancy,however,the increase of redundancy will inevitably lead to the increase of system cost.Here,the input parameters of system reliability optimization design are cost and process parameters.The cost parameters include the cost of system components with redundant costs,system failure cost and operation cost.The process parameters include mission time,failure probability of components.The relationships between the total cost and system parameters are thoroughly discussed.The method to calculate total cost is revised in this thesis.An evaluation standard of R_per_Cost which indicates the reliability value obtained for one unit of total cost is also proposed in this thesis.A number of benchmarks are presented for the cost effective analysis.As indicated by the analysis in this thesis,the system reliability design scheme can be optimized under the condition of the most efficient utilization of the total cost.