Machine Learning for predicting the cross-sectional elastic buckling strengths of cold-formed steel members

The elastic buckling strength is a key parameter in predicting the design strength in cold-formed steel member design. Particularly, for cold-formed steel sections, there are several buckling modes that are typically categorized: local-plate, distortional, and/or global buckling. As the necessary first step in design, finding the elastic buckling solution generally necessitates numerical analysis for elastic buckling prediction due to the complexity nature of cross-sectional instabilities though some simplified analytical solutions are available. Even the most commonly used tool such as finite strip method (FSM) that has the least modeling effort has its own challenges and complexities in buckling prediction. The paper aims to explore Machine Learning (ML) for predicting the elastic buckling strength of cold-formed steel members - shifting away from the mechanic-based numerical analysis. The Artificial Neural Network (ANN), a subgroup of machine learning algorithms, will be employed for the model training. The ML model will start with the commonly used sections, lipped channel section (C), where the dataset includes those from standard industrial shapes and parametrically randomized shapes. Then, those data will be categorized into geometrical data and material data as inputs and elastic buckling strengths and half-wavelengths of local and distortional buckling as the outputs. The developed ML models’ efficiency and accuracy will be evaluated. Further extension of the ML models to more generalized sections will be further explored.