Efficient Prediction of Asphalt Concrete Mix Design Based on Performance Tests

Abstract: Superpave asphalt concrete (AC) mix design procedure remains empirical, time and effort consuming, and requires material resources. To add performance tests to the Superpave design, Illinois adopts the balanced mix design (BMD), compromising Illinois Flexibility Index Test (I-FIT) and the Hamburg Wheel Tracking Test (HWTT). To ensure efficient, cost effective, and sustainable AC mixes, a self-learning system is needed to optimize AC mix volumetric design that BMD requirements. This study presents a future-ready machine learning (ML)-based framework comprised of a multi-objective genetic algorithm for predicting and optimizing binder content and aggregate gradation. A dataset, compiled by the Illinois Department of Transportation (IDOT) from actual pavement construction projects, was used. The study trained and tested models using two primary datasets: Dataset A, containing 18,562 I-FIT data samples from 2,201 AC mix designs (2016-2020), and Dataset B, comprising 8,117 HWTT data samples from 3,772 AC mix designs (2010-2020). The tool framework operates in two steps: Step 1 predicts binder content and aggregate gradation based on target mix type, flexibility index (FI), and rut depth (RD), utilizing an Extreme Gradient Boosting (XGB) model. The mix type variables include Nominal Maximum Aggregate Size (NMAS), pavement layer, number of gyrations, PG grade, recycled asphalt pavement (RAP) content, recycled asphalt shingles (RAS) content, design voids in mineral aggregate (VMA), and blend bulk specific gravity (Gsb). Step 2 uses the predictions from Step 1 as input for further optimization using a nondominated sorting genetic algorithm (NSGA-II), which fine-tunes the mix design to meet the desired FI and RD values, while adjusting RAP, RAS, blend Gsb, and design VMA. Numerous AC mix cases were used to validate the tool accuracy, efficiency, and effectiveness. To enhance its predictive capabilities across various materials and environmental conditions, the tool application beyond Illinois mixes should be explored.