Numerical investigation on seismic performance of drilled flange connection for steel frames

Compared to the welded square steel tubes, cold-formed square steel tubes offer the advantage of easier processing. However, this advantage limits the application of welding internal diaphragms in joints, thus encouraging the use of strengthened solutions for connecting concrete-filled steel tubular columns (CFST) and steel beams, such as end plates and side plates. Significant stress concentration occurs at these strengthened locations, leading to premature fractures in the welds. An innovative C-shaped end plate connection scheme (DFC), in which the end plates connect the I-shaped steel beam weakened by drilling holes in the flanges to the CFST, was proposed and tested in the previous study. Due to the complexity of DFC composition, the effects of various design parameters on seismic performance remain unclear. In this paper, a refined finite element model is developed, and verified by experimental results. Then, the effect of various design parameters on the seismic performance, such as ductility, energy dissipation and bending resistance, of DFC is investigated. Finally, to address the shortcomings of the traditional restoring force model (RFM), such as low prediction accuracy and poor generalization performance, two machine learning (ML) algorithms (RFM-MLP and RFM-XGBoost), combined with the specific restoring rule, are proposed to reconstruct the hysteresis curve. After applying grid search and Bayesian optimization methods to the hyperparameters, it is shown that the predictions of RFM-XGBoost are in particularly good agreement with the numerical results in the test set (R2 = 0.996). Furthermore, combining SHAP analysis with the proposed ML methods, the proposed ML methods are explained and can be used for preliminary judgments in practical design.