Prediction of design and optimization of steel structures using machine learning and artificial neural networks

In the field of structural engineering, traditional design and optimization methodologies are being transformed by the integration of machine learning (ML) and deep learning (DL) techniques. The increasing complexity of structural systems driven by variations in geometric parameters such as eave height, span length, frame spacing, and support conditions, as well as the growing demand for cost-effective solutions and reduced design computation time, has prompted engineers to adopt innovative approaches, including predictive algorithms for steel section selection. A comprehensive dataset was generated using finite element modeling (FEM) to represent a wide range of structural configurations. These configurations were validated through manual calculations to ensure data accuracy for ML and DL training. The trained models analyze structural parameters to predict optimal section dimensions, addressing nuanced design requirements. Various ML algorithms, including Polynomial Regression, LightGBM, XGBoost, Random Forest, and artificial neural networks (ANN) were employed for predicting column and beam dimensions. Among them, Random Forest achieved the highest accuracy (Adjusted R² = 94.132%, MAE = 0.336), followed by ANN (Adjusted R² = 91.63%, MAE = 0.212). A graphical user interface (GUI) was also developed to bridge the gap between model predictions and practical implementation, enabling engineers to design cost-effective and resilient structures in compliance with AISC, ASCE, and ECP standards.