Surrogate-Assisted Material-Efficient Design Optimization of Reinforced Concrete Frames Using XGBoost and NSGA-II

Abstract

This study proposes a surrogate-assisted multi-objective optimization framework for reinforced concrete (RC) frame design, aiming to improve material efficiency, maintain code-compliant structural safety, and support digitally enabled decision-making in sustainable construction. The framework combines Latin hypercube sampling, automated structural analysis, member-level verification based on TCVN 5574:2018, machine-learning-based surrogate modeling, and NSGA-II optimization within a unified workflow. A mechanics-consistent dataset was established from a representative four-story RC frame by systematically varying geometric, sectional, and material parameters over a practical design domain. Two surrogate model families, XGBoost and artificial neural networks (ANN), were developed to predict three governing utilization-ratio responses, namely beam flexure, beam shear, and column eccentric compression. Both models achieved high predictive accuracy on unseen data, while XGBoost showed more stable generalization and stronger interpretability across the three targets. The selected surrogate was then integrated with NSGA-II to construct Pareto-optimal solutions that quantify the trade-off between total material weight and structural safety reserve. The results demonstrate that the proposed framework can significantly reduce repeated trial-and-error analyses while preserving explicit code awareness and structural meaning. The main contribution of the work lies in linking mechanics-based data generation, code-governing response decomposition, and surrogate-assisted Pareto optimization into a practical design framework for low-waste and digitally enhanced RC-frame engineering.