Application of Machine-Learning Models on Analysis and Prediction of the Fatigue Life of Pavement Quality Concrete Made with Recycled Concrete Aggregate

Abstract

Abstract Fatigue-life estimation of pavement quality concrete (PQC) used in concrete pavement is important in light of estimation of the design life of a pavement subjected to repeated applications of wheel load. The present study investigates the flexural fatigue performance of PQC mixes of M40 and M50 grade, made with pretreated recycled concrete aggregate (PRCA) that completely replaced the natural coarse aggregate (NCA). Initially, the static flexural strength tests followed by flexural fatigue tests were conducted on PQC specimens of each grade cured for 90 days. The fatigue life of each PQC mix was evaluated at various stress levels and loading frequencies. The fatigue-life data were then analyzed with a two-parameter Weibull distribution. The sensitivity analysis was performed to understand the influence of the stress level and loading frequency on the fatigue life of PQC samples. Six machine-learning (ML) models were then used to predict fatigue life, out of which extreme gradient boosting showed the highest predictive accuracy in terms of predicting fatigue-life data of different mixes. This study demonstrated the limitations of current approaches in fatigue-life prediction and the ML models’ capability to achieve superior accuracy. The microstructural study of PQC mixes showed improved bonding of PRCA with cement matrix, which supported the hypothesis that the proposed treatment method in this study enhanced the flexural strength and fatigue life of all mixes, offering RCA as a sustainable alternative for NCA in concrete pavements.