Akademik Veri Yönetim Sistemi
Construction and Building Materials, cilt.484, 2025 (SCI-Expanded)
Hot Mix Asphalts (HMA) are subjected to various common failure modes including cracking and crack propagation and investigation of such distress are crucial in the design and maintenance of pavements. This paper aimed to conduct a comprehensive analysis on fracture resistance of various HMA mixtures containing reclaimed asphalt (RA) by using varying Deep Learning and ANFIS configurations. The experimental results on routinely collected mixtures from two asphalt mixing plants in Prague, Czechia were used to construct dataset. The properties of the asphalt mixtures including nominal maximum aggregate size (NMAS), RA content (RA%), bulk density (Gmb), maximum density (Gmm), air void content (Va%), bitumen content (BC%), and Stiffness Modulus (E) of the mixtures were considered as input variables. As response variables, the fracture toughness (KIC), fracture energy (GF) and flexibility index (FI) were utilized. The factor-response relationships and the effect of hyperparameters were investigated using varied feature selection techniques, and hyperband tunning techniques applied to configure model structures. R2, MAE, MAPE, MSE, RMSE and NMAE metrics were used, and the results were interpreted in terms of both engineering, statistical and data-driving perspective. Main findings indicated the It was found that the inclusion of the input variables “BC%, E, Gmm, Va%, NMAS” for KIC, “NMAS, Gmb, Gmm, BC%, E” for GF, and “E, BC%, Gmm, Va%, Gmb” for FI was appropriate for estimating the corresponding fracture parameters. RMSprop with lower learning rates provided improved accuracy. The configured hybrid models demonstrated efficiency in modeling fracture resistance, showing reduced error values compared to the base models, despite variations in RA% and mixture type.