Explainable Machine-Learning Model for Rapid Damage Assessment of CFST Columns after Close-In Explosion

Abstract

In the present study, the dynamic response and damage of concrete-filled steel tubular (CFST) columns under close-in explosion were numerically studied. An extensive parametric study was carried out to investigate the effects of column height, diameter, wall thickness, yield strength of steel, compressive strength of concrete, and axial load ratio on the residual midheight displacement (RMHD) and residual axial load-bearing capacity (RALBC). It was found that the RALBC is strongly correlated with the RMHD under different explosion scenarios. Three models were developed using Extreme Gradient Boosting (XGBoost) based on a database comprising 1,708 circular CFST column samples. These models aimed to predict the relationship between RMHD and RALBC, utilizing different combinations of input variables. Accurate prediction results can be obtained from all the models, and the selection of a model can be based on the availability of known input variables. The third prediction model, which does not require knowledge of the blast loading parameters and axial load ratio, which are usually difficult to obtain, can yield accurate results. Therefore, it can be used to quickly evaluate the RALBC of CFST columns. Finally, the prediction model was further interpreted locally and globally using the additive feature attribution method Shapley Additive Explanation (SHAP). Through the SHAP interpretation, the contribution of each input variable to the RALBC of CFST columns was analyzed. This provided valuable insights into the impact of individual variables on the prediction results.