Predicting asphalt mixture compaction quality using ground penetrating radar and machine learning

Reliable evaluation of asphalt mixture compaction quality is critical for ensuring pavement performance and durability. This study proposes a novel non-destructive compaction prediction framework that integrates Ground Penetrating Radar (GPR) with machine learning (ML) algorithms to assess asphalt compaction quality. The proposed framework establishes a complete and systematic prediction pipeline linking compaction parameters, dielectric response, and compaction state, in which compaction parameters, including temperature, number of roller passes, and mixture gradation, are first mapped to GPR-derived dielectric constants and subsequently converted into the corresponding compaction degree. Laboratory experiments involving four asphalt mixture types (AC-13, AC-20, ATB-25, OGFC-13) demonstrated strong correlations between dielectric constants and compaction indices, with correlation coefficients ranging from 0.69 to 0.85. Based on this quantitative relationship, four machine learning models, namely multilayer perceptron (MLP), random forest (RF), support vector machine (SVM), and Gaussian process regression (GausPR), were developed to predict dielectric constants from compaction parameters. Among them, the SVM model achieved the highest predictive accuracy, with correlation coefficients of 0.996 for dense-graded mixtures and 0.995 for open-graded mixtures. Full-scale slab compaction experiments further confirmed the reliability and spatial adaptability of the proposed framework, with average prediction errors remaining around 1.0 percent for dense-graded mixtures and exhibiting strong spatial consistency. The results demonstrate that the proposed GPR and ML framework enables non-destructive, high-resolution, and continuous real-time monitoring of asphalt compaction quality, providing practical value for intelligent construction and quality control in asphalt pavement engineering.