Utilizing Machine Learning to Minimize Sample Height Errors in Marshall Asphalt Mixture Design Method

In this study, a Machine Learning (ML) method was utilized to predict whether the sample heights of the briquettes prepared during the hot mix asphalt design step (Marshall method) will be within the tolerances specified in ASTM D 6927 standard. Factors affecting the sample height were analyzed using a multilayer perceptron algorithm. In the analysis process, the sample heights of the road layers consisting of base, binder, wearing, and SMA wearing layers were estimated with an accuracy of about 90% or more, demonstrating the high accuracy of the model. As a result, there is high possibility of utilizing ML to prepare asphalt specimens within the required height range.