Structural Reliability Analysis (SRA) Based On Improved Latin Hypercube Important Sampling

The accuracy of numerical simulation methods for structural reliability analysis(SRA), such as Monte Carlo method and important sampling method, is highly dependent on the quality of samples. So it is crucial to improve the representativeness of samples for the numerical simulation of SRA. Latin Hypercube sampling(LHS) method is widely used for its almost full coverage of the definition domain, but samples from traditional LHS distribute with a considerable degree of randomness, and dimension and number of samples will directly affect the accuracy and efficiency of the LHS method of reliability analysis.LH importance sampling method for reliability analysis is firstly improved in this paper,and then applied in the sensitivity analysis. Numerical examples show that with respect to the Monte Carlo method, the accuracy and efficiency of LH importance sampling method for sensitivity analysis is greatly improved. Upon this, the improved optimal LH importance sampling method based on ESE(enhanced stochastic evolutionary) algorithm is proposed, which avoids some unstable problem caused by the randomness of variables and bad solution due to the dimension of problem, and decrease the sampling size with the same precision of traditional SRA method. Numerical results show that better uniformity and representativeness of samples is obtained by the improved LH importance sampling method, which improves the accuracy and efficiency of SRA.