Scholarlink

<p>Glass fiber reinforced plastic (GFRP) is widely used in engineering applications due to its excellent mechanical properties and resistance to environmental degradation. However, residual stresses induced by wind-sand erosion significantly affect its performance and durability. Current models for predicting these stresses are limited by their inability to capture the dynamic nature of erosion and lack interpretability. To address these gaps, this study proposes an interpretable machine learning framework that combines eight machine learning algorithms, including three single models (artificial neural network, support vector regression, and decision tree) and five ensemble models (Bagging, random forest, AdaBoost, gradient boosting, and extreme gradient boosting), with SHapley Additive exPlanations (SHAP) for enhanced model interpretability. A comprehensive dataset consisting of 625 experimental samples is used, with residual stress as the output parameter. The input features include key environmental and material-related parameters such as erosion angle (<i>θ</i><sub><i>erosion</i></sub>), erosion velocity (<i>v</i><sub><i>erosion</i></sub>), erosion time (<i>t</i><sub><i>erosion</i></sub>), sand flow rate (<i>Q</i><sub><i>sand</i></sub>), fiber modulus (<i>E</i><sub><i>f</i></sub>), thickness (<i>h</i>), and particle diameter (<i>d</i>). Additionally, engineered features such as the erosion factor and the Mechanical Stress Index (<i>MSI</i>) are incorporated to improve model performance. The results show that the introduction of <i>MSI</i> significantly improves model accuracy by approximately 10% and reduces computation time by 8%. Among the models, XGBoost outperforms others, achieving an R<sup>2</sup> value of 83.20% and demonstrating 44% faster computational efficiency than Gradient Boosting. SHAP analysis identifies <i>MSI</i> and erosion time (<i>t</i><sub><i>erosion</i></sub>) as the two most influential factors, accounting for about 80% of the predictive accuracy. The findings offer a robust predictive model for residual stresses in GFRP under wind-sand erosion and provide valuable insights into material degradation, thus improving its applicability in practical engineering scenarios. This study enhances the reliability and efficiency of footing design under harsh environmental conditions, contributing significantly to the field of geotechnical engineering.</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

Prediction of residual stresses in GFRP strips under wind-sand erosion by interpretable machine learning methods: feature engineering and SHAP analysis

  • 期刊论文

  • 发表日期：

    2025年4月15日

查看全文
  • Wenhao Ren,
  • Siha A,
  • Changdong Zhou,
  • Jiaxing Ma

摘要

Glass fiber reinforced plastic (GFRP) is widely used in engineering applications due to its excellent mechanical properties and resistance to environmental degradation. However, residual stresses induced by wind-sand erosion significantly affect its performance and durability. Current models for predicting these stresses are limited by their inability to capture the dynamic nature of erosion and lack interpretability. To address these gaps, this study proposes an interpretable machine learning framework that combines eight machine learning algorithms, including three single models (artificial neural network, support vector regression, and decision tree) and five ensemble models (Bagging, random forest, AdaBoost, gradient boosting, and extreme gradient boosting), with SHapley Additive exPlanations (SHAP) for enhanced model interpretability. A comprehensive dataset consisting of 625 experimental samples is used, with residual stress as the output parameter. The input features include key environmental and material-related parameters such as erosion angle (θerosion), erosion velocity (verosion), erosion time (terosion), sand flow rate (Qsand), fiber modulus (Ef), thickness (h), and particle diameter (d). Additionally, engineered features such as the erosion factor and the Mechanical Stress Index (MSI) are incorporated to improve model performance. The results show that the introduction of MSI significantly improves model accuracy by approximately 10% and reduces computation time by 8%. Among the models, XGBoost outperforms others, achieving an R2 value of 83.20% and demonstrating 44% faster computational efficiency than Gradient Boosting. SHAP analysis identifies MSI and erosion time (terosion) as the two most influential factors, accounting for about 80% of the predictive accuracy. The findings offer a robust predictive model for residual stresses in GFRP under wind-sand erosion and provide valuable insights into material degradation, thus improving its applicability in practical engineering scenarios. This study enhances the reliability and efficiency of footing design under harsh environmental conditions, contributing significantly to the field of geotechnical engineering.