Integrating particle packing approach with ML techniques to optimise the compressive strength of RCA based concrete mixes

Abstract

The present study focusses on predicting and optimizing the compressive strength of recycled concrete aggregate (RCA) based concrete mixes using machine learning algorithms. The issues with the gradation of irregular shape of recycled aggregates is leveraged using particle packing approach. Integration of the packing density into analysis of 466 data sets of RCA based concretes using ML techniques is the novelty of this work. Input variables water-to-binder ratio, binder content, natural fine aggregate, natural coarse aggregates, RCA percentage replacement (ranging from 0 % to 100 %), packing density, and output variable compressive strength (in MPa) are considered. Random Forest Regressor (RFR) algorithm demonstrated significant accuracy in predicting the RCA based concrete mixes. Validation through k-fold analysis reinforced robustness of the predictive model, revealing an average R2 of 0.99 with low predictive errors. Overall, ML-based regression analysis has come in handy to accurately predict the strength in RCA-based concrete mixes, offering a valuable alternative, avoiding, extensive laboratory experiments for strength optimization.