Web crippling strength of cold-formed steel lipped channels under interior-two-flange loading

This study presents a novel application of explainable machine learning techniques to investigate the web crippling behaviour of Cold-Formed Steel (CFS) lipped channel beams under Interior-Two-Flange (ITF) loading condition. Three Machine Learning (ML) models, including Extreme Gradient Boosting (XGB), Random Forests (RF), and Artificial Neural Networks (ANNs) were developed to predict web crippling capacity under ITF loading. Considering a wide range of geometries for CFS lipped channels, a dataset of 400 instances was generated through validated numerical modelling to train and test ML models. All models demonstrated high reliability, exceeding the target reliability index of 2.5. Among these, XGB outperformed other models, achieving an R² score of 0.996 on the testing data. The predictive performance of the XGB model was further evaluated using 33 unseen experimental data points collected from the literature and compared with existing design formulas for web crippling capacity. XGB demonstrated superior accuracy, achieving the lowest error of 6.87 %. Shapley Additive Explanations (SHAP) were employed to elucidate the influence of geometric parameters on web crippling strength. The SHAP-based interpretations aligned with established domain knowledge, confirming the validity of the findings. This study presents a time-efficient, accurate, and interpretable approach to predicting web crippling strength, offering a valuable complementary approach for the design of CFS lipped channel sections.