Our new paper presents a physics-based machine learning model for predicting asphalt test results, combining a data-driven approach with physical constraints to improve reliability and interpretability.
The main contribution is the methodology itself: a framework that leverages existing experimental data while remaining consistent with known material behavior.
Although we demonstrate the approach using the Marshall test, this choice was driven by data availability, and the methodology is intended to be extended to other performance-based asphalt tests soon.
This is the first paper to come out of the AsphaltMine.org initiative, setting the foundation for future work in data-enabled asphalt materials evaluation, so make sure to check the paper and the try the Marshall prediction tool.