Research On Methods And Applications Of Reliability Information Fusion For Complex System With Small Sample Test

The reliability analysis and evaluation for aerospace products is difficult in engineering because of its complexity and small sample tests. In order to get accurate and credible estimation of reliability indexes, the analysis and evaluation should be based on the full use of information provided by all kinds of tests. Therefore, four kinds of systematic and engineering information fusion methods, named weighted-fusion-method, environmental-factor-fusion-method, reliability -growth-fusion-method and performance-degradation-fusion-method, are put forward, to exploit all relevant information, such as time-to-failure data and performance degradation data gained during tests in different periods and environments.In order to make full use of reliability information gained in the same environment for the same matrix products, such as simulation test data, component test data, expert information and history test information, four weighted fusion methods are proposed, which are respectively named correlation-function-method, sufficient-measure-method, maximum-entropy-and-moment -matching-method and fuzzy-logic-operator-method. The former three methods belong to linear weighted methods which are proposed for independent information sources, the last one belongs to nonlinear weighted methods raised for dependent information sources. These methods consider the belief degree of the multi-sources and avoid the subjectivity as much as possible. The key problem in system reliability analysis by using Bayes theory is how to express multi-source information in appropriate prior distribution. By using these weighted fusionmethods, this problem can be solved properly.In order to make full use of the reliability information with the form of zero-failure-data gained in different environments, three estimation methods for environment factor are proposed, which are named random-weighted-method, little-probability- method and athwart-moment -method. For solving the problem pf lacking samples in performance reliability estimation for liquid rocket engine, one factor, which can be estimated through the second maximum likelihood estimation method (ML-II) is put forward to convert ground information to space. The modified athwart-moment-method is raised to estimate environment factor of Weibull distribution with small failure scene test sample. Also, the little-probability-method can be used to estimate the factor when the scene test failure sample is small or even with zero failure.In order to make full use of information gained from changing matrixes, four reliability growth analysis methods are proposed, which are named middle-rank-method, modified-likelihood-function-method, conserve-analysis-method and multiple-failure-mode-method. The middle-rank-method can be used to analyze system reliability growth with small failure sample. The modified-likelihood-function-method is applied to analyze system reliability growth with zero failure data. The conserve-analysis-method is raised to solve the subjective problem in the reliability analysis for Weibull-distributed system and can ensure the highreliability demand for key sub-system in aerospace. The multiple-failure-mode-method, which can analyze different effects on reliability growth caused by different improving measures, is put forward on generalization of traditional reliability growth model.In order to exploit the reliability information hidden in the performance degradation data, the fusion methods for performance degradation information are studied. Firstly, the basic concepts and modeling process for performance reliability are proposed. Then, two models for degradation path are raised. One is linear-random-process, which is applied to linear degradation. The other is compound-Poisson-process, which considers the cumulative effect of degradation and suits different degradation path. For utilizing the performance degradation data gained in different environments, the fusion method based on degradation failure environment factor is raised. Meanwhile, the definition of the factor and the estimation methods for compound-Poisson-process are given. The reliability growth analysis method is put forward with the aim at using the performance degradation data gained in different periods of system development, which can reasonably analyze the effect caused by the improvement of design and manufacture technique. The practicability and efficiency of fusion methods above are validated through the reliability analysis of liquid rocket engine, human-aircraft temperature controlled bump and container of power system for SG-III.