A Bayesian duration estimation of crack initiation prediction in in-service pavements

Research has recently concentrated on modeling and predicting pavement distress and deterioration; this research has, almost exclusively, revolved around mechanistic-empirical models that place restrictions on estimated parameters compromising performance. Recent computational advances enable the estimation of complex and computationally cumbersome statistical models with two very attractive properties: i. they are based on explicit mechanistic models that stem directly from pavement engineering practice, and ii. estimation and interpretation is straightforward, transparent, and tractable. We address here the problem of pavement failure times on the basis of data collected from in-service pavements in 15 European countries using Bayesian stochastic duration models that account for both parameter uncertainty and model specification uncertainty. Results indicate that, as expected, construction, traffic and climatic factors affect pavement distress; further, the loglogistic functional form estimated via the Bayesian Inference we propose, describes distress initiation better than existing approaches.