Abstract
Asphalt pavements experience various distresses in the field including rutting, cracking, and moisture damage. Performance decay models are often used in the asset management system to predict the future performance of pavement to program maintenance and rehabilitation treatments. The overall performance of the pavement is affected by the material properties and thickness of each layer, applied traffic, and environmental conditions. Rutting, cracking, and roughness increases with time. Performance deterioration can be difficult to predict since there are numerous factors that can affect the pavement condition. The first part of this study investigated and developed multiple types of artificial intelligence models to determine which would be best suited to utilize structural, environmental, and traffic data to predict pavement performance. This investigation determined that random forests regression was best suited for the data utilized in this study. Multiple random forests regression models were developed to predict various indicators of pavement performance such as the International Roughness Index (IRI), rutting, and cracking. Models were developed using a theoretical dataset generated using the Pavement ME software as well as a field dataset collected from the Long-Term Pavement Performance Program. The results demonstrated good correlations between all the theoretical and predicted performance indicators for the models developed using the theoretical dataset. Performance decay curves for pavement sections were also developed using the developed models, and the predicted performance decay curves were found to closely simulate the measured decay curves. The results for the models developed using the field dataset demonstrated good correlations between measured and predicted performance indicators for some of the investigated performance indicators. This study also developed random forests regression models using structure information, traffic data, and deflection data collected using the Falling Weight Deflectometers (FWD) and Traffic Speed Deflectometers (TSD). Models to predict IRI, rutting, and cracking were developed using theoretical datasets generated using the 3D Move software. The results demonstrated good correlations between all the theoretical and predicted performance indicators. Similar models were also developed to backcalculate the layer moduli of test sections pavements. The results demonstrated perfect correlations between theoretical and predicted layer moduli. The final aspect of this study was the development similar models to backcalculate layer moduli using a combination of field and theoretical data. The results from this model trained on a combination of field and theoretical data demonstrated good correlations between theoretical/measured and predicted layer moduli for some of the pavement layers.