Enhancing Compressive strength prediction of Roller Compacted concrete using Machine learning techniques

Abstract

Roller compacted concrete (RCC) is a widely adopted material in pavement engineering and irrigation dam due to its fast, simple construction process and excellent cost-effectiveness compared to conventional concrete. However, accurately determining its compressive strength (CS) and optimizing its composition presents a challenge that conventional methods have not effectively addressed. This work proposes a machine learning approach to predict the CS of RCC with high speed and accuracy. The models considered are CatBoost, Artificial Neural Networks, Decision Tree, Random Forest, and Linear Regression. After fine-tuning the hyperparameters and conducting evaluations, the CatBoost model emerged as the best performer with a coefficient of determination of 0.983. Additionally, the proposed model can highlight the impact of input parameters on RCC's CS, providing valuable insights for CS measurement and RCC design in material engineering, thanks to Partial Dependence Plots (PDP) analysis. Specifically, the findings highlight the effects of cement content and age on the CS, which can help material engineers optimize the RCC composition for improved performance. Additionally, a user-friendly graphical user interface (GUI) has been developed to facilitate the use of the proposed ML model and streamline the optimization process of RCC mixtures.