Compressive strength of concrete with recycled aggregate; a machine learning-based evaluation

Abstract

In the recent decades, researchers have shown an increased interest in the reusing recycled concrete aggregate generated by buildings’ demolitions because of the undeniable economic and environmental benefits. The main objective of this research is therefore to evaluate the influence of replacing natural fine and coarse aggregate on the 3-, 7- and 28-day compressive strength of concrete using machine learning-based approaches. To this aim, the paper is presented in two parts. Firstly, linear, nonlinear regression and Random Forest methods are used to propose prediction models for estimating compressive strength of concrete with recycled aggregate. Then, the results are compared to those of Artificial Neural Network and M5P models available in the literature in terms of performance metrics parameters and Taylor diagram. Consequently, the most accurate model is introduced and used in the second part for assessing the effect of using recycled aggregates on the concrete compressive strength. To this end, three 10-set databases are generated by replacing natural fine, coarse and both of them with recycled fine, coarse and both of them, respectively, by the portions of 0–90% with the step of 10%. The compressive strength predicted by the ML-based model are evaluated in three different testing ages (3, 7 and 28 days). The results proved the high accuracy of the proposed prediction models and Random Forest model which was developed in this study was the most accurate model. Moreover, it was concluded that incorporating recycled aggregate could decrease compressive strength insignificantly. Finally, reduction in long-term strength of the concrete containing recycled fine aggregate is less than those containing coarse aggregate or both fine and coarse aggregate.

Keywords

• Recycled aggregate concrete
• Compressive strength
• Machine learning
• ANN
• Regression
• Random forest