Postblast damage assessment of concrete-filled double-skin tube columns by intelligence-based techniques.

This paper aims to use machine learning (ML) based on a regression algorithm to predict the damage degree of a concrete-filled double-skin steel tubular column (CFDST) after explosion. Three classical single learning algorithms and five excellent ensemble learning algorithms were used to train 2476 sets of data. Five regression indicators were used to evaluate and compare the effectiveness and accuracy of the ML model. The results show that the integrated CatBoost model is superior to the other models, and the model performances are ranked as CatBoost > LightGBM > XGBoost > GBDT > MLP > RF > SVR > DT. Shapley Additive exPlanations (SHAP) is used to explain the contribution of each feature parameter to CatBoost prediction results. Finally, the CatBoost model is used to predict the collected test data, and R2 = 0.989, MSE = 2.98 × 10−6, RMSE = 0.00172, MAE = 0.00151, and MAPE = 7.4% are obtained. This shows that the model has high prediction accuracy and can accurately predict the damage degree after CFDST explosion. • An extensive database of CFDST columns subjected to the close-in explosion is established. • Three classical single learning algorithms and five excellent integrated learning algorithms are used to predict the damage degree of concrete-filled double-skin steel tubular column (CFDST) columns under close-in explosion loads. • After model training and evaluation, the CatBoost model has the best accuracy among the eight models. • The global and local interpretation of the CatBoost model is carried out through the Shapley Additive exPlanations(SHAP), and the influence of different input parameters on the model is visualized. • In the actual damage assessment, compared with the problem that the traditional P-I curve regression formula can only obtain the damage degree of the CFDST column, the CatBoost model can obtain accurate damage conditions. [ABSTRACT FROM AUTHOR]