Experimental and analytical evaluation of crumb rubber concrete with multi-blended industrial waste aggregate and SCMs: Strength, durability, microstructure, and ML prediction

Abstract

Rubberized concrete offers notable environmental benefits through waste tire utilization; however, its application is constrained by comparatively low mechanical strength. The addition of suitable industrial waste materials may offer a viable strategy to offset the strength reduction and improve the structural performance of rubberized concrete. Consequently, researchers are investigating supplemental cementitious materials (SCMs) as alternatives to conventional binders and industrial waste materials as substitutes for fine aggregates in the construction industry. This study proposes a framework to improve the properties of blended crumb rubber concrete (BCRC) incorporating industrial by-products such as copper slag (CS) and steel slag (SS) as fine aggregates, along with silica fume (SF) and rice husk ash (RHA) as SCMs. Four types of BCRC mixes were investigated for fresh concrete properties, mechanical properties through compressive, split tensile, and flexural strengths, as well as durability indicators like alkali–silica reaction (ASR) and rapid chloride penetration test (RCPT). Out of all mixes, BCRC4 (20 %CR+20 %SF+10 %RHA+20 %CS+10 %SS) shows the 54 % improved strength and durability properties that resulted from balanced use of SCMs and industrial waste materials. However, predicting the compressive strength of BCRC remains complex due to its heterogeneous composition and age-dependent behavior. To address this, seven widely used machine-learning techniques were employed to estimate compressive strength at advanced curing ages, including linear regression, decision tree, random forest, extreme gradient boosting (XGBoost), support vector machine (SVM), and Lloyd’s algorithm (K-means clustering). Model performance was evaluated using regression analysis and statistical indicators. Among the models, linear regression demonstrated the highest predictive accuracy, achieving an R² value of 0.94 with the lowest mean squared error (0.39) and mean absolute error (0.48). Furthermore, SHAP (SHapley Additive exPlanations) analysis was conducted for the optimal BCRC4 mix to interpret the contribution of individual input variables, providing valuable insights into the influence of alternative fine aggregates and SCMs on compressive strength development with curing age.