Statistical Mixed Effects Models for Evaluation and Prediction of Accelerated Pavement Testing Results

Abstract

We elucidate the usefulness of the mixed-effects models in analyzing performance-type data resulting from accelerated pavement testing (APT) experiments. Our analysis is based on an APT test designed to evaluate the performance of materials and pavement structures using rut depth as the response variable. The model results indicate that both binder type and temperature are significant factors in rut depth development. By capturing the left-over variability associated with each test track unexplained by the experimental factors, the mixed effects models employed herein also allow inference to be made about unobserved phenomena such as failure probabilities and their confidence intervals, where failure is defined as the passage of a particular threshold such as a maximum rut depth. These failure probabilities provide estimates of the percentage of sections that would fail after a certain accumulated traffic load, which could help trigger maintenance actions in the field. This paper also provides insights regarding the duration of APT experiments conducted with a heavy vehicle simulator (HVS) from the perspective of stable parameter estimation, which may prove beneficial in optimizing loading time and testing focus in HVS experiments.