Machine learning techniques for recycled aggregate concrete strength prediction and its characteristics between the hardened features of concrete

• Published in: Arabian journal of geosciences 2021-11, Vol.14 (22), Article 2390
• Main Authors: Rizvon, Shamili Syed, Jayakumar, Karthikeyan
• Format: Article
• Language: English
• Subjects: Aggregates; Algorithms; Cement; Compressive strength; Concrete; Concrete aggregates; Concrete properties; Cylinders; Earth and Environmental Science; Earth science; Earth Sciences; Elasticity; Flexural strength; Learning algorithms; Machine learning; Mechanical properties; Microcracks; Microstructure; Mixtures; Modulus of elasticity; Original Paper; Plasticizers; Recycled materials; Scanning electron microscopy; Tensile strength; Voids; Water-cement ratio
• ISSN: 1866-7511; 1866-7538
• DOI: 10.1007/s12517-021-08674-z

This study aims to arrive at models to correlate the mechanical properties of recycled aggregate concrete (RAC). An experiment was performed on the recycled coarse aggregate (RCA) ratio of 0%, 25%, 50%, and 100% with 400kg/m 3 of cement content, varying water-to-cement ratio (w/c) 0.3, 0.4, 0.48, and super plasticizer (SP) dosage to produce 15 different mixes were investigated. Furthermore, the compressive strength (both cube f cu and cylinder f cyl ), modulus of elasticity (E c ), split tensile strength (f ct ), and flexural strength (f cr ) were tested and investigated. In view of the results, new models representing the impact of RCA were created. The results showed that in terms of replacement ratio, at 56d, the mix with 25% of RCA with water-to-cement ratio (w/c) 0.3 and super plasticizer (SP) 1.5%, recorded maximum strength of 59.86MPa, 4.81MPa, and 5.416MPa under cube compressive strength, split tensile, and flexural strength respectively. The proposed models can effectively predict the E c , f ct , f cr , f cyl , and f cu of RAC. Scanning electron microscope (SEM) was conducted to scrutinize the microstructure of selected mixes which shows comparatively low voids, micro-cracks, and pores. Also, machine learning techniques like multi-linear regression (MLR) and extreme gradient boosting (XGB) algorithms were utilized for the compressive strength prediction of concrete (CSC). Results indicated that XGB for cylinder compressive strength was found to be 2.7% greater than cube compressive strength and MLR for cylinder compressive strength was found to be 1.5% greater than cube compressive strength .