Work

Improved prediction of resilient modulus of unbound aggregates using artificial neural networks

Year: 2026
Type: article
Abstract: Aggregate base course is the primary load-bearing layer beneath paved surfaces, while the subbase improves load distribution and provides separation above the subgrade. Because aggregate quality varies widely in type, gradation, shape, texture, angularity, and durability, understanding how these properties influence pavement performance is essential. The resilient modulus (MR) is a key mechanistic parameter describing the recoverable response of unbound aggregates under repeated loading, yet laboratory MR testing is costly and time-consuming. This study developed multilinear regression (MLR) models to relate aggregate properties to a consistent set of MR test results using Minnesota Department of Transportation’s Aggregate Index Property database, compiled from the Aggregate Source Information System and Laboratory Information Management System. Recognizing the significant influence of particle shape characteristics beyond gradation alone, the study also incorporated imaging-based indices of shape, texture, and angularity into an artificial neural network (ANN) model. The ANN was used to evaluate how a comprehensive set of aggregate properties affects the accuracy of predicted MR model parameters. The ANN outperformed the enhanced MLR model in predicting the k1, k2, and k3 parameters, capturing complex nonlinear relationships. These findings highlight the potential of ANN-based surrogate models to reduce reliance on extensive laboratory testing and improve pavement material characterization.
Language: English
Open Access status: diamond
OpenAlex ID: W7204197960
FWCI:
Cites: 7
Cited by:
Related to: 10