Sustainable Polyurethane-Based Polymer Concrete: Mechanical and Non-destructive Properties with Machine Learning Technique

Abstract

Polyurethane-based polymer concrete (PUC) has become a popular material for pavement repair. However, its compressive strength (f_c) is essential to achieve effective repair work. This study predicted the compressive strength and evaluated the non-destructive test (NDT) properties of the PUC mixtures, prepared by mixing aggregate-to-polyurethane (PU) at 80/20, 85/15, and 90/10 ratios by weight. The experimental datasets from mechanical and NDT tests were utilized to train machine learning (ML) models, including multilinear regression (MLR), artificial neural network (ANN), support vector machine (SVM), Gaussian regression process (GPR), and stepwise regression (SWR) models for estimating the f_c. Moreover, scanning electron microscopy (SEM) was employed to evaluate the microstructure of PUC. Feature selection tools were used to explore optimal input variables for estimating the (f_c) of the PUC samples. The PUC-10 specimen revealed a maximum ultrasonic pulse velocity (UPV) value of 3.05 km/h. The microstructure analysis shows micro-voids with crack propagation between the aggregate and PU binder in the specimen containing 10% PU after testing. All the developed models showed high prediction accuracy. In addition, SVM outperformed other models at the training phase with R² values of 0.9614, and ANN demonstrated the highest performance at the testing phase with R² values of 0.9502.