Evaluation of self-compacting rubberized concrete properties: Experimental and machine learning approach

• Published in: Structures (Oxford) Vol. 58; p. 105423
• Main Authors: Ofuyatan, Olatokunbo M., Muhit, Imrose B., Babafemi, Adewumi J., Osibanjo, Ibukunoluwa
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.12.2023
• Subjects: Durability tests; Machine learning; Mechanical properties; Powdered waste rubber tires; Regression models; Self-compacting rubberized concrete; Mechanical properties; Durability tests; Regression models; Powdered waste rubber tires; Machine learning; Self-compacting rubberized concrete
• ISSN: 2352-0124; 2352-0124
• DOI: 10.1016/j.istruc.2023.105423

Diverse negative impacts of waste tire disposal have created a menace to a cleaner environment worldwide. Global awareness on the use of unconventional materials in concrete necessitated the use of solid waste in concrete. Towards sustainable construction and building materials, in this study, powdered waste rubber tires (PWRT) were incorporated into self-compacting concrete as a partial substitute for fine aggregate. The suitability of the self-compacting rubberized concrete (SCRC) was assessed by conducting workability tests (slump flow, T50, and L-box), mechanical tests (compressive, splitting tensile, and flexural strength tests), microstructural analysis, and durability tests. The results showed that an increasing percentage of PWRT had an adverse effect on the workability and flowability of SCRC. Mechanical strength at 3, 7, 21, 28, 56, and 90 days exhibited a reduction with an increasing PWRT content. Furthermore, the microstructural analysis showed weaker adhesion at the interfacial transition zone in the SCRC. A correlation matrix with empirical relationships was also developed. The effect of acid attack on SCRC was measured by immersion in HCL and Na2SO4, and a poor resistance was noticed. Machine learning regression algorithms were employed to predict the SCRC mechanical properties, including linear, ridge, lasso, decision tree, random forest, extreme gradient boosting, and support vector. In addition, evaluation metrics with statistical checks were also used to assess the model's performance. Ridge regression appeared best suited for predicting the compressive strength, while random forest regression best estimates the tensile and flexural strength.