Prediction of rheological properties and ageing performance of recycled plastic modified bitumen using Machine learning models

Abstract

Recycled plastics can be used to improve the rheological properties of neat bitumen. The rheological properties of recycled plastic modified bitumen (RPMB) depend on multiple factors, including the modified bitumen composition and ageing conditions. In this study, four ensemble machine learning (EML) models were developed based on experimental data records, aimed at predicting the rheological properties of RPMB in terms of complex shear modulus and phase angle at both unaged and short-term aged conditions. The performance of four EML models was evaluated using various metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), coefficient of determination (R2) and Objective Function Value (OBJ). The results indicated that CatBoost exhibited the highest accuracy, achieving a coefficient of determination of 0.98 for complex shear modulus prediction and 0.93 for phase angle prediction. Additionally, SHapley Additive exPlanations (SHAP) analysis, individual conditional expectation and partial dependence plots were utilised to identify the magnitude and impact of inputs variables on complex shear modulus and phase angle, as well as analyse the relationship between input and output variables. The findings highlighted that the temperature and frequency of the dynamic shear rheometer test, the amount of polymer and the penetration of the base bitumen are key factors influencing the rheological properties of RPMBs. The proposed approach provides useful information to decision-makers for effectively evaluating the composition of bitumen mixtures and the rheological properties of RPMBs based on specific project conditions.