Scholarlink

<p>Accurate prediction of the coefficient of friction (COF) in aluminum alloy (AA7075) composites is crucial for optimizing their performance in tribological applications. This study explores the application of six machine learning (ML) models—Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Least Absolute Shrinkage and Selection Operator (Lasso) Regression, Extra Trees Regressor, Kernel Ridge Regression (KRR), and Gaussian Process Regression (GPR)—to predict the coefficient of friction based on reinforcement type (silicon carbide (SiC) and titanium diboride (TiB₂)), weight percentage, and sliding distance. Feature importance analysis revealed sliding distance as the most influential factor (importance ≈ 1.0 in CatBoost), followed by reinforcement weight percentage (importance up to 0.6 in GPR and KRR). Pearson correlation analysis confirmed a strong positive correlation (r = 0.74) between sliding distance and the coefficient of friction. Among the models, Lasso Regression exhibited the best generalization performance with a test coefficient of determination (R²) of 0.9871, root mean square error (RMSE) of 0.0040, mean absolute error (MAE) of 0.0035, and mean absolute percentage error (MAPE) of 0.8754%. Gaussian Process Regression and Extreme Gradient Boosting also achieved high test R² values of 0.9809 and 0.9622, respectively, while Extra Trees showed the poorest generalization (test R² = 0.8439). Residual and relative error plots confirmed the stability of Lasso and Gaussian Process Regression predictions. This comparative analysis underscores the effectiveness of regularization-based and probabilistic models in coefficient of friction prediction, offering valuable insights for friction modeling in aluminum matrix composites.</p><p></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 friction forecasting in AA7075 composites through ensemble and probabilistic machine learning

  • 期刊论文

  • 发表日期：

    2025年10月24日

查看全文
  • A. Bhowmik,
  • Abeyram M. Nithin,
  • Raman Kumar,
  • K. Venkadeshwaran,
  • Abdulaziz Alhazaa,
  • Harjot Singh Gill,
  • Dhirendra Nath Thatoi,
  • Valentin Romanovski

摘要

Accurate prediction of the coefficient of friction (COF) in aluminum alloy (AA7075) composites is crucial for optimizing their performance in tribological applications. This study explores the application of six machine learning (ML) models—Extreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Least Absolute Shrinkage and Selection Operator (Lasso) Regression, Extra Trees Regressor, Kernel Ridge Regression (KRR), and Gaussian Process Regression (GPR)—to predict the coefficient of friction based on reinforcement type (silicon carbide (SiC) and titanium diboride (TiB₂)), weight percentage, and sliding distance. Feature importance analysis revealed sliding distance as the most influential factor (importance ≈ 1.0 in CatBoost), followed by reinforcement weight percentage (importance up to 0.6 in GPR and KRR). Pearson correlation analysis confirmed a strong positive correlation (r = 0.74) between sliding distance and the coefficient of friction. Among the models, Lasso Regression exhibited the best generalization performance with a test coefficient of determination (R²) of 0.9871, root mean square error (RMSE) of 0.0040, mean absolute error (MAE) of 0.0035, and mean absolute percentage error (MAPE) of 0.8754%. Gaussian Process Regression and Extreme Gradient Boosting also achieved high test R² values of 0.9809 and 0.9622, respectively, while Extra Trees showed the poorest generalization (test R² = 0.8439). Residual and relative error plots confirmed the stability of Lasso and Gaussian Process Regression predictions. This comparative analysis underscores the effectiveness of regularization-based and probabilistic models in coefficient of friction prediction, offering valuable insights for friction modeling in aluminum matrix composites.