Application of machine learning in asphalt and concrete material testing: A comprehensive review

• Published in: Građevinski materijali i konstrukcije Vol. 67; no. 4; pp. 183 - 200
• Main Authors: Khorshidi, Meisam, Dave, Eshan, Sias, Jo
• Format: Journal Article
• Language: English
• Published: Society for Materials and Structures testing of Serbia 01.01.2024
• Subjects: artificial neural networks (ann); ensemble methods; gaussian process regression (gpr); hybrid models; material performance prediction; pavement distress classification; predictive modeling; unsupervised learning
• ISSN: 2217-8139; 2335-0229; 2335-0229
• DOI: 10.5937/GRMK2400012K

This literature review explores the application of machine learning (ML) techniques in civil engineering material testing, with a focus on asphalt mixtures, concrete properties, and pavement system classification. The review provides a comprehensive comparison of various ML models, including Artificial Neural Networks (ANNs), Support Vector Machines (SVMs), Random Forest (RF), Gradient Boosting (GB), and Gaussian Process Regression (GPR), assessing their strengths and limitations in predicting material performance. Key findings indicate that ensemble methods, such as Gradient Boosting and XGBoost, consistently outperformed other models in terms of prediction accuracy and handling nonlinear relationships, although they require significant computational power. In contrast, simpler models like SVM and ANN demonstrated strong predictive capabilities with smaller datasets but were prone to overfitting and computational challenges. Additionally, unsupervised learning methods, such as K-means clustering and Principal Component Analysis (PCA), proved effective in classifying pavement conditions and detecting anomalies, with K-means offering simplicity and efficiency at the cost of sensitivity to initialization and cluster definitions. The review concludes by emphasizing the potential of hybrid and ensemble models to improve prediction accuracy and reduce computational costs, highlighting the need for further research to address data availability, model interpretability, and practical implementation challenges in real-world applications.