An improved FGT-based MCMC adaptive importance sampling method

The efficiency of reliability simulation has long been a research hotspot. Crude Monte Carlo method is too time-consuming to analyze systems with long life and high reliability. In order to improve computing efficiency and save computing time, this paper presents an improved FGT-based MCMC adaptive importance sampling method. This novel approach firstly generates training samples from failure region by means of Markov Chain method. Then the traditional adaptive importance sampling is improved with modified Fast Gauss Transform which can effectively enhance the computing speed of kernel destiny estimator. Finally, samples obtained from improved adaptive sampling importance can calculate failure probability rapidly. A case study of Y-tube demonstrates the feasibility and usability of the proposed method.