Modeling of recycled coarse aggregates concrete characteristics via hybrid improved Grey Wolf Optimizer deep neural network and multi-objective Grey Wolf Optimization

Abstract

This study presents a thorough methodology for forecasting and enhancing the essential performance characteristics of concrete with Recycled Coarse Aggregates (RCA), including water content (Wt), density, and Ultrasonic Pulse Velocity (UPV). The attributes are represented as functions of cement content (300-400 kg/m�), RCA percentage (0-100 %), and slump (5-12 � 1 cm). An experimental database was created using a three-factor, three-level design, followed by exploratory analysis using correlation matrices and three-dimensional response surface plots to identify significant input-output associations. Various modeling techniques, such as Response Surface Methodology (RSM), Decision Tree (DT), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and a Deep Neural Network (DNN) enhanced by the Improved Grey Wolf Optimizer (IGWO), were assessed. The accuracy of the model was meticulously evaluated by 4-fold cross-validation, Taylor diagrams, and radar plots, demonstrating that the DNN-IGWO hybrid surpassed all other models, achieving the minimal prediction errors and the maximal correlation coefficient (R2) values for all target responses. The validated models were later employed as surrogates inside a multi-objective optimization framework employing the Multi-Objective Grey Wolf Optimizer (MOGWO). The optimization produced well-distributed Pareto fronts, emphasizing the intrinsic trade-off between density and UPV. The resultant "knee" solutions identify specific concrete compositions that optimize UPV (3.6-3.7) while ensuring a high density (?2400 kg/m�) and moderate water content. This study provides a scientifically robust, data-driven paradigm for the sustainable design of concrete mixtures, facilitating the appropriate incorporation of recycled aggregates while enhancing material qualities.