Bayesian Degradation Modeling in Accelerated Pavement Testing with Estimated Transformation Parameter for the Response

Abstract

We discuss Bayesian degradation models that were developed for flexible pavements based on accelerated pavement testing with the heavy vehicle simulator. The models are fitted to data from the Florida Department of Transportation, where rutting performance of three binder types was tested under three temperature settings. The analysis utilizes Bayesian linear mixed-effects models for longitudinal degradation data where the parameter estimates and their posterior marginal distributions are obtained via a Markov chain Monte Carlo (MCMC) technique. The linearity in this model is achieved by utilizing a covariate-dependent Box-Cox transformation of the response variable, where the transformation parameter is estimated as part of the modeling procedure. The paper illustrates the various forms of useful inference that can easily be obtained via the output from the MCMC chains and provides insights regarding the accelerated test experiment at hand. As expected, the results suggest that rut depth development is affected both by the binder type, as well as the test temperature. What is more, the conditional inference made possible by the Bayesian approach utilized here clearly demonstrates the dependence of the inference for the covariate effects on the value of the Box-Cox transformation parameter. Hence the transformation of the response variable is an important step in model building that has to be carefully considered.