ABSTRACT
Asphalt pavement should be represented as a series system of limit state functions associated with the international roughness index, rut depth, alligator cracking, and transverse cracking. Traditional regression-based prediction models are too simplified to account for the relationship between pavement performance and the operating conditions associated with climate, traffic, pavement structure and property parameters. In this study, a deep learning model based on bidirectional long short-term memory neural networks is trained using the long-term pavement performance database to learn the nonlinear and complex relationship between four performance indicators and their associated parameters. Based on multiple time-variant limit-state functions incorporating the uncertainties associated with these parameters, deep learning model prediction, and international roughness index measurement, Monte Carlo simulation is conducted to estimate the system reliability of the asphalt pavement. In an illustrative example, the effects of different parameters on the life-cycle system reliability are investigated based on two pavement sections.
