Machine learning models for the ultimate strength of steel beams – influence of bending moment diagram

Abstract Artificial intelligence models using machine learning techniques are widely used in engineering to predict the mechanical behavior of structural members. Different machine learning (ML) algorithms such as artificial neural networks, random forests, and support vector regression were used to develop and train models in this study to predict the ultimate strength of steel beams, in particular that include the influence of the bending moment diagram on its lateral‐torsional buckling resistance. An extensive dataset was constructed using finite element analysis to obtain the ultimate strength of simply supported beams. A comparative study of different hyperparameters was carried out. The results show that the ML models outperform state‐of‐the‐art analytical models and that are able to capture the influence of bending moment diagrams. The limits of application of these ML models are explored, providing an overview of their potential use in designing real structures.