Machine learning model for predicting structural response of RC columns subjected to blast loading

• Published in: International journal of impact engineering 2022-04, Vol.162, p.104145, Article 104145
• Main Authors: Almustafa, Monjee K., Nehdi, Moncef L.
• Format: Article
• Language: English
• Subjects: Algorithms; Blast loading; Blast loads; Blasting (explosive); Catastrophic collapse; Column; Concrete columns; Concrete structures; Economic impact; Economic models; Explosions; Far fields; Gradient boosting; Machine learning; Parameter identification; Performance measurement; Prediction; Random forests; Reinforced concrete; Robustness (mathematics); Structural response
• ISSN: 0734-743X; 1879-3509
• DOI: 10.1016/j.ijimpeng.2021.104145

•Novel ML model proposed for predicting behavior of RC columns under blast.•Large dataset for FRP RC columns under blast was compiled.•Statistical metrics indicate that developed model achieved superior accuracy.•Feature importance analyses agreed with experimental and numerical studies.•Considering simplicity, speed and accuracy, new model is strong contender. Considering the risk of exposure to blast and explosive loading, reinforced concrete structures are prone to experiencing partial or total progressive collapse initiated by column failures. Therefore, understanding and predicting the structural response of columns subjected to blast loading fosters proactive measures that could mitigate life and economic losses. The present study introduces a machine learning model to predict the maximum displacement of reinforced concrete columns exposed to blast loading using thirteen features pertaining to imperative column and blast properties. The dataset used in this study consists of 420 data examples retrieved from existing experimental, numerical and analytical studies in the open literature. The model was developed using ensemble tree-based algorithms and was validated through statistical performance metrics, numerous comparisons to existing methods, and feature importance analyses. Additionally, a critical analysis was conducted to assess the importance of features in both near-field and far-field blast exposures. The practical use of the proposed model, along with recommendations for model improvements were discussed. Overall, the usage of tree ensemble algorithms for the proposed model achieved very high prediction performance, resulting in MAE of 3.63 mm, MAPE of 13.31%, R2 of 97.4%, and VEcv of 96.83%, while displaying robust ability to identify correlations between influential parameters and the corresponding response.