Reliability analysis of damage prediction in asphalt pavement sections using FlexPAVETM program and S-VECD fatigue model

Abstract

ABSTRACT With mechanistic and mechanistic-empirical modelling of pavement performance there are a considerable number of uncertainties that are inherent in the prediction process. Many of these sources of uncertainty are challenging to quantify, but one that is technically possible to evaluate is the effect of uncertainty in the characterisation and modelling of the mechanical properties of the asphalt concrete. In this study, the Bayesian inference-based Markov Chain Monte Carlo (MCMC) was used to estimate how uncertainties from the simplified viscoelastic continuum damage (S-VECD) model parameters propagate to pavement performance simulation errors in the FlexPAVETM program. Performance predictions were evaluated based on the propagation of three different material parameter uncertainties; (1) modulus, (2) damage characteristic curves, and (3) failure criterion. These factors were systematically varied according to chosen reliability levels and using data from real-world experimental data on different asphalt mixtures. High-performance computing resources were then used to predict pavement performance on the thousands of different combinations of material property uncertainty, and a simplified prediction algorithm was developed to more easily estimate the uncertainty. The results show that the simplified uncertainty prediction model could estimate the reliability of the damage and cracking predictions from FlexPAVETM within 1.2% and 4.8%, respectively.