Machine Learning for Column Web Panel in Transverse Compression Component

Abstract

The component method enables efficient design of diverse steel joints by decomposing the joint into individual components characterized by their load resistance and initial stiffness. Machine Learning (ML) models exceed code‐based formulas in predictive accuracy; however, their limited interpretability raises concern about their use in reliability‐based structural design. This paper demonstrates the process of creating an ML design method for a connection component based on a large dataset created in IDEA StatiCa Connection. Data curation, model training, and validation are outlined, with major issues identified. Finally, the reliability of ML predictions is evaluated against experimental data in accordance with EN 1990, providing insights into the potential and limitations of ML in structural engineering design.