Adaptive cost-constrained optimization of concrete mixtures using machine learning-guided genetic algorithms

This study presents an adaptive framework for optimizing high-performance concrete mixtures by integrating machine learning with a genetic algorithm under cost constraints. An experimental dataset was used to train an XGBoost model, which accurately predicts 28-day compressive strength. The trained machine learning model was embedded as the objective function within the genetic algorithm to maximize compressive strength while incorporating cost limitations defined by user-provided unit prices and budget. Unlike most previous studies that treat cost and strength as two separate objectives in multi-objective formulations, this work introduces cost as a constraint and strength as the sole optimization objective, thereby simplifying the decision-making process. To bridge the gap between theory and practice, an Android application was developed. The application enables users to input real-time material prices and budget limits, which are transmitted to a server hosting the machine learning model and genetic algorithm. The server computes optimized mix proportions and returns both the predicted compressive strength and the optimal design to the user interface. The proposed adaptive optimization framework was shown to effectively adjust to market price fluctuations and varying budget scenarios, providing a practical and flexible solution for real-world applications. Furthermore, the single-objective formulation ensures a unique optimal solution, avoiding the complexity of selecting among multiple Pareto-optimal alternatives.