Scholarlink

<p>The Aluminum 7075 alloy is renowned for its exceptional strength and is extensively used in aerospace and defense applications. Despite its advantageous properties, welding this alloy presents significant challenges, primarily due to the reduction in mechanical strength at the joint. To address these limitations, this study utilizes friction stir welding (FSW)—a solid-state welding technique that significantly reduces heat-related distortions and metallurgical defects typically associated with conventional fusion welding processes. The research investigates the impact of various FSW process parameters on the mechanical performance of the welded joints. Mechanical behavior is evaluated using tensile tests and Charpy impact tests, providing comprehensive insight into strength and impact toughness under different welding conditions. The experimental data collected from these tests are further analyzed using machine learning (ML) techniques to predict key mechanical properties, which often involve complex, non-linear relationships among multiple input variables. To enhance the prediction accuracy, several ensemble learning algorithms are employed, including Gradient Boosting, Adaptive Boosting (AdaBoost), Extreme Gradient Boosting (XGBoost), and Stacking. These ensemble models incorporate diverse base learners such as Nearest Neighbor Regressor (NNR), Decision Tree Regressor, Random Forest Regressor, and Support Vector Regressor (SVR). The performance of each model is rigorously evaluated using a range of error metrics, including Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE) and R<sup>2</sup>. The experimental results clearly indicate that the Stacking model with Support Vector Regressor as the base learner delivers superior prediction accuracy for impact energy and Gradient Boosting with Decision Tre Regressor outperforms for elongation. In contrast, for Ultimate Tensile Strength (UTS), Stacking with Nearest Neighbor Regressor demonstrates the best performance. Additionally, Explainable Artificial Intelligence techniques, such as SHAP is employed to find out feature importance. In addition, tornado plots are used to determine the relative influence of each input parameter on the output variables—namely energy absorption, tensile strength, and elongation—providing valuable insights for process optimization.</p> Graphical Abstract <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

Machine learning approach for prediction of mechanical properties of friction stir welded AA7075 aluminium alloy

  • 期刊论文

  • 发表日期：

    2025年10月7日

查看全文
  • Rohit Ahuja,
  • Ratnesh Kumar Raj Singh

摘要

The Aluminum 7075 alloy is renowned for its exceptional strength and is extensively used in aerospace and defense applications. Despite its advantageous properties, welding this alloy presents significant challenges, primarily due to the reduction in mechanical strength at the joint. To address these limitations, this study utilizes friction stir welding (FSW)—a solid-state welding technique that significantly reduces heat-related distortions and metallurgical defects typically associated with conventional fusion welding processes. The research investigates the impact of various FSW process parameters on the mechanical performance of the welded joints. Mechanical behavior is evaluated using tensile tests and Charpy impact tests, providing comprehensive insight into strength and impact toughness under different welding conditions. The experimental data collected from these tests are further analyzed using machine learning (ML) techniques to predict key mechanical properties, which often involve complex, non-linear relationships among multiple input variables. To enhance the prediction accuracy, several ensemble learning algorithms are employed, including Gradient Boosting, Adaptive Boosting (AdaBoost), Extreme Gradient Boosting (XGBoost), and Stacking. These ensemble models incorporate diverse base learners such as Nearest Neighbor Regressor (NNR), Decision Tree Regressor, Random Forest Regressor, and Support Vector Regressor (SVR). The performance of each model is rigorously evaluated using a range of error metrics, including Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE) and R2. The experimental results clearly indicate that the Stacking model with Support Vector Regressor as the base learner delivers superior prediction accuracy for impact energy and Gradient Boosting with Decision Tre Regressor outperforms for elongation. In contrast, for Ultimate Tensile Strength (UTS), Stacking with Nearest Neighbor Regressor demonstrates the best performance. Additionally, Explainable Artificial Intelligence techniques, such as SHAP is employed to find out feature importance. In addition, tornado plots are used to determine the relative influence of each input parameter on the output variables—namely energy absorption, tensile strength, and elongation—providing valuable insights for process optimization.

Graphical Abstract