Novel hybrid machine learning approach for predicting structural response of RC beams under blast loading

Abstract

Explosions have inflicted colossal human and economic losses. However, existing response prediction methods for blast-loaded reinforced concrete (RC) structures are complex and suffer from several limitations. The novelty of this study consists of developing a hybrid machine learning model to predict the maximum displacement of RC beams under blast loading. The model was developed using a database compiled from published studies and considers influential features of RC beams and blast parameters. Hybrid gradient-boosted regression trees algorithm coupled with Henry gas solubility optimization algorithm was developed and validated using multiple statistical metrics, feature importance analyses, and comparisons to traditional methods. A failure mode and crack pattern classification model was also developed to predict the post-blast condition of RC beams. It was scrutinized using multi-class confusion matrix and feature importance analysis. The displacement prediction model achieved superior performance with MAE of 4.48 mm and R2 of 93.4%. The classification model attained adequate performance with binary-class accuracy of 93.1% and multi-class accuracy of 83.74%. Both models captured features’ importance, while strongly correlating to experimental studies. The model opens the door for creating an improved and integrated machine learning model for predicting the quantitative and qualitative behavior of RC beams under blast loading in future research.