1369093 - Asphalt Concrete Crack and Rut Depth Prediction Using Extensive Database.

The Illinois Flexibility Index Test (I-FIT) and the Hamburg Wheel Tracking Test (HWTT) have been adopted for Asphalt concrete (AC) balanced mix design (BMD). A model to predict the flexibility index (FI) and rut depth would improve mix design efficiency. This study proposes two deep learning models to predict I-FIT FI and HWTT rut depth. A database of 19,138 I-FIT datasets and 7,602 HWTT datasets were collected from the Illinois Department of Transportation (IDOT) (after data preprocessing was conducted). Two models were developed to predict FI and rut depth based on AC material properties. To account for AC intrinsic variability, both models adopt Monte Carlo Dropout. Monte Carlo dropout allows the computation of a distribution of predicted FI and rut depth which accounts for material variability. Monte Carlos allows to analyze the entire distribution of predictions to obtain the average as “best estimate” and the variability can be measured as CV. The model results indicate that the distribution of predictions was within a coefficient of variation (CV) of 10 to 30%, typical for I-FIT and HWTT data. Finally, Shapley Additive Explanations (SHAP) were used to calculate the contribution of each input feature to the final model prediction. In addition, SHAP provides the relationship between the input and the parameter to be predicted.