Application of approximate Bayesian computation for estimation of modified weibull distribution parameters for natural fiber strength with high uncertainty

摘要

Despite the unique advantages of natural fibers as a reinforcement in polymer composites, they have high natural variability in their mechanical properties, resulting in significant uncertainties in the properties of natural fiber composites. This study aims to propose a multilevel framework based on the Approximate Bayesian Computation (ABC) to analyze the uncertainty of fitting the Weibull distribution to the strength data of date palm fibers. Two computationally efficient algorithms of the ABC, namely the Metropolis-Hasting as a family of Markov Chain Monte Carlo and the Sequential Monte Carlo (SMC), are employed for estimating the highest density interval of the fitting parameters of the modified 3-parameter Weibull distribution, and their performances are evaluated. Moreover, appropriate probability distributions that best fit the estimated parameters are determined based on the goodness of fit to describe their characteristics. It is found that the SMC algorithm leads to a higher scatter in the posterior predictive distribution of the fitting parameters. The results suggest that the uncertainty of the fitting parameters should be considered to have a reliable model for the probability of natural fiber failure. [GRAPHICS] .