Abstract
文章来源:正航仪器
浏览次数:
发布时间:2014-10-08
Abstract
For existed problems in structure reliability and corresponding parameter sensitivity analysis, the thesis developed a series of reliability methods combining with new theory, such as the radial-based importance sampling method (RBIS), integration order descending method (IODM), Latin hypercube sampling method (LHS), etc. The details are listed as follows possess high precision and efficiency in the application.
(1) For the convergence of the importance sampling based reliability sensitivity analysis and Monte Carlo based reliability sensitivity analysis, the variance and variation coefficient of the reliability sensitivity assessment are derived in this contribution. Furthermore, the interval with the given confidence of the reliability sensitivity assessment based on importance sampling is computed approximately. The illustrations show that the importance sampling based reliability sensitivity is more efficient than Monte Carlo simulation based reliability sensitivity at the acceptable precision.
(2) For the estimation of the reliability sensitivity, an improved importance sampling method is presented. The improved method needs to search the most probable point (MPP) in the failure region firstly.And then, by use of this property of the failure region, the presented method only calculates the performance function values of the importance samples located outside the hyper-sphere to complete the estimation of the reliability sensitivity. Therefore, the presented method is more efficient than the traditional method. For the system with single failure mode and the system with multiple failure modes in series, the formulae of the variance and the variation coefficient of the improved method are derived in addition.
(3) The aforementioned improved importance sampling method needs to search the most probable point (MPP) in the failure region, which is very difficult for many complicated structures. In order to solve the problem, the adaptive radial-based importance sampling (ARBIS) method is presented. Using information provided by the required samples for the reliability sensitivity estimation, the optimal radials of the ARBIS based methods can be determined by gradual iteration. Therefore the robustness and the accuracy of the ARBIS based methods are improved greatly. Furthermore, the ARBIS-based Direct Method (DM) and the ARBIS-based Transformation Method (TM) are established for the reliability sensitivity analysis with correlative normal variables by firstly transforming the correlative normal variables into independent ones.
(4) In order to improve the efficiency and precision of the reliability sensitivity analysis, two methods based on the Integration Order Descending Method (IODM) are presented for analyzing reliability sensitivity of structure. Furthermore, for the system with single failure mode and the system with multiple failure modes, the formulae of the two reliability sensitivity methods are derived.
(5) Latin hypercube sampling (LHS) and updated Latin hypercube sampling (ULHS) by statistical correlation reducing equation are employed to analyze structural reliability sensitivity and its variance, with single failure mode and with multiple failure modes. The reliability sensitivity estimates based on the LHS and the ULHS are more robust than that on the Monte Carlo simulation in case of small sampling size. The variance of the sensitivity estimation based on ULHS can be reduced more than that based on LHS in small sampling size.
(6) While the structure includes fuzzy variables, for the symmetric trapezium membership function, the Max-Min method and the Equivalent-Area method are adapted to transform it to the Gaussian one equivalently. For the symmetric parabolic membership function, the Improved Max-Min method and the Improved Equivalent-Area method are presented to transform it to the Gaussian one equivalently. And for the symmetric Cauchy membership function, the Equivalent-Area method is adapted to transform it to the Gaussian one equivalently. Then the fuzzy random reliability and reliability sensitivity problem can be transformed into the random ones, which can be analyzed by the line sampling method. After that the chain rule is employed to obtain the sensitivity of the fuzzy random failure probability with respect to the distribution parameters of the equivalent Gaussian membership function.
(7) Based on the extended reliability, the adaptive kernel density approximation method and the polynomial approximation method are adapted to estimate the probability density function. After that two methods are extended into the estimation of the global reliability sensitivity. Both numerical and engineering examples are adopted to compare the established two methods with the finite mixture density estimation based global reliability sensitivity method and the maximum entropy based global reliability sensitivity method. The results of the illustrations show that the maximum entropy and the polynomial approximation based on global reliability sensitivity method are more robust than the others.
(8) For the statistical analysis of the small sample fatigue life, the sampling distribution of the population standard deviation is simulated by the Bootstrap method, then the confidence interval of the population standard deviation is estimated by the percentile method with rectification, and the confidence interval of the fatigue scatter factor is estimated especially. For the actual fatigue life test data of aeronautical material on 140 steel-alloy test pieces and 295 aluminum-alloy test pieces, the confidence interval of the fatigue scatter factor is estimated by the above method, and the law of the confidence interval with the fatigue test stress is studied also, which provides an alternative method for analyzing fatigue life test data in the engineering practice.
Keywords: Failure Probability; Reliability; Reliability Sensitivity; Variance; Coefficient of Variance; Confidence Interval; Correlative Variables; Monte Carlo; Important Sampling; Radial-Based Importance Sampling (RBIS); Adaptive Radial-Based Importance Sampling (ARBIS); Integration Order Descending Method (IODM); Latin Hypercube Sampling (LHS); Fuzzy Random Reliability Sensitivity; Extended Reliability; Global Reliability Sensitivity; Bootstrap Method.