Scholarlink

<p>The intricate structure of braided carbon fiber reinforced polymer (CFRP) tubes presents significant challenges for an accurate and robust prediction of mechanical properties. To address this, an interpretable machine learning framework for predicting the axial compression energy absorption properties of CFRP tubes and analyzing the influence of structural parameters is proposed. A dataset is compiled from published studies, and four machine learning models are trained and compared. The eXtreme Gradient Boosting (XGBoost) emerges as the top-performing model and is further optimized using the Sparrow Search Algorithm (SSA). The predicted results of the optimized SSA-XGBoost model are interpreted via SHAP (SHapley Additive exPlanations). The relationship between the key input parameters and the output is quantitatively analyzed. The research results show that SSA-XGBoost demonstrates exceptional predictive accuracy across all four energy absorption metrics, with the root mean squared error (<i>RMSE</i>) of 1.9158, 1.3947, 1.9400, and 0.0464, and the coefficient of determination (<i>R</i>²) of 0.9860, 0.9685, 0.9897, and 0.8375. SHAP interpretation reveals that the wall thickness (<i>T</i>) is the most critical structural parameter, dominating two of the four energy absorption metrics with an approximately linear positive correlation. For one of the remaining metrics, <i>T</i> is the second most influential parameter, exhibiting an approximate quadratic relationship. The proposed framework not only achieves accurate and efficient prediction of CFRP tubes’ energy absorption but also quantitatively demonstrates the influence of structural parameters on the energy absorption capacity to a certain extent.</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

Mechanical properties prediction and feature analysis of CFRP tubes under axial compression based on interpretable artificial intelligence

  • 期刊论文

  • 发表日期：

    2025年7月31日

查看全文
  • Weimin Zhuang,
  • Bu Yang,
  • Hailun Zhang

摘要

The intricate structure of braided carbon fiber reinforced polymer (CFRP) tubes presents significant challenges for an accurate and robust prediction of mechanical properties. To address this, an interpretable machine learning framework for predicting the axial compression energy absorption properties of CFRP tubes and analyzing the influence of structural parameters is proposed. A dataset is compiled from published studies, and four machine learning models are trained and compared. The eXtreme Gradient Boosting (XGBoost) emerges as the top-performing model and is further optimized using the Sparrow Search Algorithm (SSA). The predicted results of the optimized SSA-XGBoost model are interpreted via SHAP (SHapley Additive exPlanations). The relationship between the key input parameters and the output is quantitatively analyzed. The research results show that SSA-XGBoost demonstrates exceptional predictive accuracy across all four energy absorption metrics, with the root mean squared error (RMSE) of 1.9158, 1.3947, 1.9400, and 0.0464, and the coefficient of determination (R²) of 0.9860, 0.9685, 0.9897, and 0.8375. SHAP interpretation reveals that the wall thickness (T) is the most critical structural parameter, dominating two of the four energy absorption metrics with an approximately linear positive correlation. For one of the remaining metrics, T is the second most influential parameter, exhibiting an approximate quadratic relationship. The proposed framework not only achieves accurate and efficient prediction of CFRP tubes’ energy absorption but also quantitatively demonstrates the influence of structural parameters on the energy absorption capacity to a certain extent.