Prediction model of the failure mode of beam-column joints using machine learning methods

Abstract

Beam-column joints are important components that control the performance of reinforced concrete frame structures under seismic loads. Beam-column joint failures may induce partial or overall collapses of the structures. Brittle failure is joint shear failure before beam yielding, and ductile failure is joint shear failure after beam yielding or beam yielding without joint failure. In this study, based on the collected 580 test data of interior beam-column joints, nine features were constructed to reflect the characteristics of the joints’ design parameters. Twelve machine learning methods are applied to predict failure modes of beam-column joints. After comparing the prediction performance and comparing the results from four design codes, the prediction model provided by the XGBoost algorithm is recommended in this study for its excellent classification results of the failure modes of beam-column joints. Moreover, the SHAP method was used to explain the features’ effects in the prediction models. Accordingly, the interior beam-column joints’ failure mode can be accurately predicted, and the suggestions for changing the failure mode from brittle failure to ductility failure can be offered for the beam-column joints.