Buckling prediction of stainless steel circular hollow sections beam–columns using finite element analysis and machine learning

Abstract

This study presents a data-driven approach to predict the buckling resistance of stainless steel circular hollow section beam–columns by combining finite element analysis and machine learning. Due to the complex behavior of stainless steel and limited design guidelines, a validated finite element model was developed in Abaqus and used to perform a comprehensive parametric study, generating 955 simulations covering a broad range of geometric parameters, eccentricities, and five stainless steel grades. Two machine learning models including Artificial Neural Networks and Stacking were trained using features such as diameter, thickness, length, eccentricity, and material properties. The Artificial Neural Networks model outperformed the Stacking model, achieving a coefficient of determination of 0.998 and low prediction errors. SHapley Additive exPlanations, feature importance and sensitivity analysis identified diameter and thickness as the most critical factors. Comparisons with European stainless steel design code 1993-1-4 (2023) and the American stainless steel specification AISC 370 (2021) showed these standards are conservative, supporting the value of machine-learning-based predictions. A user-friendly web app was also developed to enable predictions, offering a practical and interpretable tool for enhancing structural design with machine learning.