TY  - JOUR
AV  - none
KW  - Asphalt; Binders; Forecasting; Mean square error; Rheology; Silica; Surface properties
KW  -  Complex modulus; Composite modified asphalts; Machine learning models; Machine-learning; Modified asphalt binders; Nano Silica; Phase angles; Response-surface methodology; Rheological property; Silica waste
KW  -  Machine learning
ID  - scholars17981
TI  - Computational modelling for predicting rheological properties of composite modified asphalt binders
N2  - The complicated viscoelastic characteristics of asphalt binders make it a challenging task to precisely predict their rheological behavior. This study aims to investigate and compare the suitability of response surface methodology (RSM) and machine learning (ML) modeling approaches in predicting the complex modulus (G*), phase angle (δ), and rutting parameter (G*/sinδ) of Nano Silica (NS) and/or waste denim fiber (WDF) modified asphalt binders before and after short-term aging. To achieve this, an experimental scheme was designed for RSM and ML modeling with three variables including NS contents (0â??6), WDF contents (0â??6), and testing temperature (40â??76 °C) as the inputs, and provided the G*, δ and G*/sinδ before and after short-term aging as the outputs. A wide range of ML algorithms was evaluated to determine the optimum ML model that can be used to accurately predict the rheological properties of NS/WDF-modified asphalt binders. RSM analysis results indicated that the G*, δ, and G*/sinδ of NS/WDF composite asphalt are significantly affected by the NS, WDF, and test temperatures. The RSM-developed models showed coefficient of determination (R2) values exceeding 0.97 for all responses, indicating adequate agreement between experimental results and models developed by RSM. From ML algorithms optimization and among all evaluated ML models, it was found that Gaussian process regression (GPR) exhibited the highest R2 with a value of (0.99) and the lowest Root Mean Square Error (RMSE) with a value of approximately 1. The performance evaluation of the GPR model for predicting all responses showed a very small difference between the predicted and experimental results, highlighting the prediction accuracy of the developed ML models. © 2023 The Authors
N1  - cited By 0
Y1  - 2023///
VL  - 19
UR  - https://www.scopus.com/inward/record.uri?eid=2-s2.0-85176446159&doi=10.1016%2fj.cscm.2023.e02651&partnerID=40&md5=decaa3a3bcd0d437382348aed91cd5d2
A1  - Al-Sabaeei, A.M.
A1  - Alhussian, H.
A1  - Abdulkadir, S.J.
A1  - Sutanto, M.
A1  - Alrashydah, E.
A1  - Mabrouk, G.
A1  - Bilema, M.
A1  - Milad, A.
A1  - Abdulrahman, H.
JF  - Case Studies in Construction Materials
ER  -