Predicting concrete strength through packing density using machine learning models

Abstract

This study presents an innovative approach to predict concrete compressive strength using particle packing theories through machine learning techniques. The existing challenge in concrete engineering lies in the accurate estimation of concrete strength, a critical factor in construction. The adoption of particle packing theories, which hold great promise for enhancing concrete performance, has been limited due to the complexity and time-consuming nature of the required calculations. An approach encompassing particle packing models (JD Dewar Model, Compressible Packing Model, and Modified Toufar Model) with machine learning is the novelty of the work. These models optimize the packing density of aggregate proportions while minimizing the void ratio, essential for achieving desired compressive strength criteria. To train the model, a comprehensive dataset comprising 479 concrete mixtures, each associated with known compressive strength values relative to packing density, is utilized. A significant advancement in predicting concrete compressive strength is demonstrated by the results. The approach outperforms traditional empirical models, offering precise and reliable predictions based on packing density. Importantly, this innovation eliminates the need for time-consuming and costly trial-and-error procedures in concrete mix design. The strong performance of various models in predicting concrete strength using particle packing theories is underscored by the study, with R^2 values ranging from 0.664 to 0.999. By combining concepts of particle packing theories and machine learning, a more efficient and reliable method for predicting concrete compressive strength is achieved. This innovation has the potential to revolutionize concrete mix design, leading to more durable and cost-effective construction practices.