Key Technology Of Reliability Growth Test Of Small Sample Complex System

The reliability growth testing aims to make the potential failure modes in system design,manufacture,and operation surfaced.After root cause analysis,corrective actions are incorporated to reduce or eliminate system failures.Through continuous iteration of test,analysis,and fix,the effectiveness of corrective measures is verified in subsequent tests and the system reliability is improved gradually.Reliability growth testing is an important engineering method to ensure the reliability of complex systems.The research on reliability growth testing is mainly divided into two areas: reliability growth planning and reliability growth assessment.The former aims to plan test resources before the start of the actual testing.It builds a reliability growth curve and provides a reference for tracking,monitoring,and adjusting the reliability growth testing.The latter tries to evaluate current system reliability,extrapolate or project future system reliability based on test data.This dissertation focuses on the system reliability growth that consists of reliability growth planning and reliability growth assessment.First,considering the influence of uncertainty on reliability growth planning,a robust criterion for reliability growth evaluation is proposed.After a framework for multi-source prior information fusion is established for Bayesian reliability growth assessment with small samples,the reliability growth assessment approach handling missing data is derived where the missing data may be caused by factors such as instrument failure and observation mechanism.Finally,the model uncertainty of reliability growth model is analyzed in that the model uncertainty can affect the decision making.The main contents and innovation results of this dissertation are as follows:(1)A robustness criterion for reliability growth planning is proposed.The factors influencing the reliability growth planning were analyzed.The information gap theory was adopted to characterize the uncertain factors and inform the uncertainty problem decision-making.A robustness evaluation index of reliability growth planning was constructed and the calculation model for the robustness was derived.Finally,the method is demonstrated through a case study and the result is compared with the worst-case analysis to verify the feasibility of the proposed method.(2)A multi-source information fusion framework for reliability growth assessment is proposed.The different sources of prior information for Bayesian reliability growth assessment were analyzed.The evidence theory was adopted to establish the framework for multi-source information fusion.Based on the fused information,prior distributions were constructed in terms of equivalent moments of the Belief function and the Plausibility function.Posterior distributions of unknown parameters were inferred from different prior combinations.The case study shows that the epistemic uncertainty of the prior information was propagated to the posterior information,and the fusion of the prior information improves the accuracy of the inference results than that from single source information.(3)An approach for reliability growth assessment with missing data is established.Based on the characteristics of order statistics,the reliability growth assessment of left censored and interval censored data of the power law process were studied using the expectation maximization algorithm.Both the analytical and stochastic solutions for left censored and interval censored data were derived.The analytical functions for both the two types of censored data were established and the Monte Carlo Expectation Maximization algorithm was used to infer the stochastic solutions.Case studies demonstrate and verify the effectiveness of the proposed method.(4)A method to quantify the model uncertainty of reliability growth models is proposed.Reliability growth models not only evaluate the current system reliability but also predict the system reliability in the future or the next stage.Data-driven reliability growth modeling is essentially an approximation of the reliability growth process.This approximation leads to model uncertainty.In practice,a statistically tested model is considered to be able to characterize the actual reliability growth process,but statistical testing usually only answers whether the model can characterize actual data,and does not quantify the uncertainty of the model itself.A framework for model uncertainty quantification of reliability growth models is established.It provides a new method for the evaluation of the reliability growth model and informs the reliability growth prediction about more comprehensive uncertainty information.Based on the Bayesian method,this dissertation establishes a model uncertainty quantitative framework for the reliability growth model.Case studies demonstrate and verify the effectiveness of this method.