Scholarlink

<p>This study presents a physics-informed machine learning framework to predict the peak displacement response of unreinforced masonry (URM) walls subjected to blast-induced ground vibrations (BIGV). The proposed methodology integrates Monte Carlo–based data augmentation with ensemble learning algorithms, supported by validated finite element simulations. Field data from 35 blast events in Indian open-cast coal mines were used to derive Peak Particle Velocity (PPV) and dominant frequency, which served as key input features. These blast inputs were simulated in ABAQUS to compute the corresponding peak displacements of three masonry walls with aspect ratios of 1.0, 1.5, and 2.0. To improve model generalization and robustness, 500 synthetic blast cases were generated using Monte Carlo sampling, expanding the dataset for supervised learning. Among the models tested, Random Forest yielded the best predictive accuracy (R² = 0.928; RMSE = 0.74&#xa0;mm), outperforming CatBoost, Gradient Boosting, and conventional regression techniques. PPV emerged as the most influential predictor, followed by wall geometry; frequency had negligible impact in the studied range. Model performance was assessed using 5-fold cross-validation, and interpretability was enhanced through feature importance and SHAP analysis. Prediction intervals were also constructed to communicate model uncertainty. The findings emphasize the significance of PPV control and structural geometry in mitigating blast-induced damage.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Ensemble learning with data augmentation for predicting brick masonry wall response under blast-induced ground vibrations

  • 期刊论文

  • 发表日期：

    2025年8月25日

查看全文
  • Monika Tewari,
  • Bibhuti Bhusan Mandal,
  • Anup Tiwari

摘要

This study presents a physics-informed machine learning framework to predict the peak displacement response of unreinforced masonry (URM) walls subjected to blast-induced ground vibrations (BIGV). The proposed methodology integrates Monte Carlo–based data augmentation with ensemble learning algorithms, supported by validated finite element simulations. Field data from 35 blast events in Indian open-cast coal mines were used to derive Peak Particle Velocity (PPV) and dominant frequency, which served as key input features. These blast inputs were simulated in ABAQUS to compute the corresponding peak displacements of three masonry walls with aspect ratios of 1.0, 1.5, and 2.0. To improve model generalization and robustness, 500 synthetic blast cases were generated using Monte Carlo sampling, expanding the dataset for supervised learning. Among the models tested, Random Forest yielded the best predictive accuracy (R² = 0.928; RMSE = 0.74 mm), outperforming CatBoost, Gradient Boosting, and conventional regression techniques. PPV emerged as the most influential predictor, followed by wall geometry; frequency had negligible impact in the studied range. Model performance was assessed using 5-fold cross-validation, and interpretability was enhanced through feature importance and SHAP analysis. Prediction intervals were also constructed to communicate model uncertainty. The findings emphasize the significance of PPV control and structural geometry in mitigating blast-induced damage.