Traffic loading impact on asphalt pavement performance: vehicle-tire-pavement interaction modeling and machine learning approach

Pavement responses are affected by the magnitude and frequency of dynamic loads generated by vehicles, which are significantly dependent on axle configuration, pavement roughness conditions, and vehicle speed. Random amplitudes of dynamic loads are generated by rough road surface due to development of pavement distresses after initial construction. Therefore, vehicle-tire-pavement interaction model is needed for analyzing dynamic pavement responses and pavement damage to moving loads and taking pavement roughness into consideration. The first objective of the research is to analyze dynamic responses of flexible pavement structure using an integrated vehicle-tire-pavement interaction approach. A full-truck model was adopted to estimate the dynamic tire forces considering pavement surface conditions, vehicle speeds, truck configuration, and axle type and loads. A modified method was proposed to derive frequency response functions under harmonic loads using the equivalent modulus of asphalt layer at the specific temperature and loading frequency. After that, the convolution integral method was used to simulate pavement responses under non-stationary loads with random amplitudes. The impulse response method was used to calculate pavement responses induced by dynamic loads considering vehicle-tire-pavement interaction. A methodology was proposed to incorporate the impact of dynamic loads on fatigue cracking development in the framework of M-E pavement design and analysis. In addition, dynamic responses of flexible pavements induced by wide-base tires considering pavement roughness condition were analyzed through the ratio of critical pavement responses between wide-base tire and dual-tire assembly, respectively, for the potential of fatigue cracking, near-surface cracking, and subgrade rutting. Long-term monitoring of in-service pavements is used to develop pavement performance models in pavement management system. The weigh-in-motion (WIM) data help to comprehensively understand vehicular loadings on pavement performance. Previous studies mainly used traditional statistical methods to quantify pavement damage due to vehicular loading. Because of the complexity of problem, the relationship between pavement performance and influential variables may not be apparent in traditional regression models. The second objective is to use machine learning approaches, including support vector regression and random survival forest models, to quantify the impact of traffic loading on pavement performance based on field data. Multi-variable nonlinear regression method and support vector regression method were applied and compared in terms of prediction accuracy and error. Random survival forest method was used to investigate the influence of traffic loading on pavement survival life. The variable importance technology was used to select the appropriate variables in the model and reduce prediction error. The proposed pavement performance models were further used to analyze pavement deterioration caused by overweight trucks with different truck traffic and axle load distributions.Ph.D.Includes bibliographical reference