Experimental investigation and comparative machine learning prediction of the compressive strength of recycled aggregate concrete incorporated with fly ash, GGBS, and metakaolin

摘要

Recycled aggregates (RA) can provide a sustainable solution for replacing natural aggregates (NA) in the concrete mix. However, the stakeholders and inspection professionals lack confidence in predicting their compressive strength (CS) due to limited databases. Most of them solely focus on the concrete mix with natural aggregates only. Even though numerous researchers have proposed alternative mix designs for recycled aggregate concrete (RAC), utilizing RA is still not practicable. One of them is the lack of a simple and effective compressive strength prediction that uses RAC. This study focuses on the application of six different machine learning (ML) techniques: XG Boost, K-nearest neighbors (KNN), artificial neural network (ANN), support vector machine (SVM), linear regression, decision tree (DT), and random forest (RF), for predicting the CS of concrete mixed with RA. The input variables are weights of coarse RA, Portland cement, fly ash, ground granulated blast furnace slag, and metakaolin. The database is prepared by experimental testing of concrete cube specimens for 188 mixes in the concrete technology laboratory of IIT Bhubaneswar. For most of the mixes, coarse RA was the only coarse aggregate to get the compressive strength. It includes variations in water/binder from 0.25 to 0.75. It was observed that the addition of flyash, GGBS, and MK significantly impacted the CS at a later age. The ML model indicates that an accuracy of 0.95 was achieved on the current test database for predicting CS. Among all the machine-learning algorithms, XG Boost can be used for forecasting compressive strength since it provides excellent accuracy with minimal computation. This research can be used as a data-driven novel solution for developing concrete mixes to achieve a specified CS. However, this work employs only experimental data as a machine learning input, which can be improved further by including databases from the literature.