Performance characterization and relationship modeling of in-situ asphalt mortar and mixture using data mining

This study characterizes asphalt pavement mixture performance using asphalt mortar-scale indices, bridging the gap between mortar and mixture properties. Performance indices were obtained for both mortar and mixture scales, including low-temperature crack resistance, intermediate-temperature durability, and high-temperature rutting resistance. Unsupervised learning of principal component analysis (PCA), and cluster analysis were employed to explore relationships among mortar indices and reduce dimensionality. The results showed that low-temperature mortar principal components were the Glover-Rowe (G-R) parameter and the 0.1 Hz modulus. Intermediate-temperature components included G⁎sinδ, the whole-temperature aging index (WAI), and slope, while high - temperature components were G⁎/sinδ, %R1.6, and J nr 12.8. Cluster analysis effectively distinguished mortar types and service ages. Artificial neural network (ANN) and boosted tree models quantified relationships between mortar and mixture. Findings indicated that in-situ equivalent single-axle loads (ESAL) primarily influenced low fracture energy (Gf), fatigue life, and complex creep stiffness modulus (CCSM). Mortar performance indices had significant effects. For example, the low-temperature Gf of the mixture was 47.69% influenced by the 0.1 Hz modulus, G-R parameter, and 1 Hz modulus. Intermediate-temperature strength and crack resistance were influenced by the fatigue factor and WAI, while fatigue life was influenced by at least 24.91% by WAI and slope. CCSM was 21.13% influenced by mortar indices, including unrecovered creep compliance (J nr ) and recovery rate (%R). The average strain rate (CASR) was at least 41.47% influenced by %R. In summary, mixture performance requires mortar performance factors and in-situ factors like ESAL, to fully characterize pavement mixture. • Cross-scale relationships between mortar and mixture performance are quantified. • Unsupervised learning reduces dimensions and clusters mortar performance indices. • ANN and decision-tree models show mortar performance drives mixture behavior.