Choosing a suitable sample size in descriptive sampling

Abstract

Descriptive sampling (DS) is an alternative to crude Monte Carlo sampling (CMCS) in finding solutions to structural reliability problems. It is known to be an effective sampling method in approximating the distribution of a random variable because it uses the deterministic selection of sample values and their random permutation,. However, because this method is difficult to apply to complex simulations, the sample size is occasionally determined without thorough consideration. Input sample variability may cause the sample size to change between runs, leading to poor simulation results. This paper proposes a numerical method for choosing a suitable sample size for use in DS. Using this method, one can estimate a more accurate probability of failure in a reliability problem while running a minimal number of simulations. The method is then applied to several examples and compared with CMCS and conventional DS to validate its usefulness and efficiency.

An improved radial basis function network for structural reliability analysis

“…X 1 to X 4 have means 120 and standard deviations 12, X 5 has mean 50 and standard deviation 15, and X 6 has mean 40 and standard deviation 12. The exact failure probability is 0.0123 [18,19].…”