A comparative study of Monte Carlo Simulation and M5Tree method on reliability analysis of truss structure

Abstract

Accurate failure probability is essential in structural reliability analysis. Plenty of methods utilized to calculate the failure probability are proposed. Monte Carlo Simulation (MCS) is one of the typical methods for solving this kind of engineering problems. In recent years, M5Tree algorithm has begun to be applied in this domain and shown great potential. The paper expounds the principles and the truss-reliability analysis process of both MCS and M5Tree through two truss examples. While evaluating the reliability using MCS, the random variables from uncertain aspects are simulated to gain the failure probability directly. In the M5Tree+MCS method, input datasets are used to generate the performance function to obtain the probability. The results indicate that MCS has great operability and practicability in dealing with general engineering problems while its computational efficiency is unsatisfactory compared to M5Tree algorithm +MCS. Overfitting in M5Tree algorithm always happens and affects the analysis on testing data, a defect of M5Tree. The accuracy of failure probability and computational efficiency are the main criterions for judging.