An integrated machine learning and genetic algorithm approach for properties prediction of fly ash-based steel fiber-reinforced concrete

Abstract

Enhancement of concrete strength is critically important for increasing construction materials’ lifespan and sustainability. Traditionally, concrete mixture optimization methods—especially those used for fly ash and steel fiber concretes—normally fail to accurately predict the strength due to the high degree of complexity and non-linearity involved in the interaction of their components. These limitations are overcome in this study, which uses advanced artificial intelligence techniques—The Multilayer Perceptron (MLP) Neural Networks, Gradient Boosting Machines (GBM), and Convolutional Neural Networks (CNN) to optimize concrete mixtures for improved strength. Among these, the MLP neural network was selected for this work because of its ability to model highly complex, nonlinear relationships and hence will be able to capture the intricate interactions among fly ash, steel fibers, and other additives. For this reason, Gradient Boosting Machine was chosen for its robustness against overfitting and high accuracy in handling linearity or nonlinearity in an optimization problem. Traditionally, CNN has been applied to image processing, but in this work, it had been uniquely adapted to include the spatial distribution of concrete mix components, hence giving a new dimension in strength prediction. In this study, every method was used with a comprehensive data set and the input variables were taken as the percentages of fly ash and steel fibers, the water-cement ratio, aggregate size distribution, and curing delays. The accuracies of prediction for the proposed models were improved significantly, with the Mean Absolute Error (MAE) for compressive strength by the MLP model and an R² value of 0.90–0.95 by the GBM model. It is interpreted from CNN that there could be a potential reduction in prediction error by 10–15% compared to traditional methods. The work provides a robust framework for concrete strength optimization with substantial improvements in the reliability and performance of concrete materials used in construction.