Scholarlink

<p>Friction stir welding (FSW) is a solid-state joining process widely used for aluminum alloys due to its unique ability to produce high-quality joints with minimal defects. However, predicting the mechanical properties of friction stir welded joints remains challenging because of the complex nonlinear interactions among process parameters. This study presents an explainable machine learning framework for predicting the ultimate tensile strength (UTS) of friction stir welded AA6061-T651 aluminum alloy. A Random Forest regression model was developed using experimental data consisting of four key process parameters: rotational speed, welding speed, tool tilt angle, and axial load. Data preprocessing included normalization and train-test splitting, followed by hyperparameter optimization using GridSearchCV. The optimized model achieved a coefficient of determination (R2) of approximately 0.84 with a root mean square error (RMSE) of about 6&#xa0;MPa. To interpret the machine learning model’s feature importance, SHAP explainability, partial dependence plots, permutation importance, and learning curve analysis were performed. The results indicate that rotational speed and axial load exert the strongest influence on joint strength, while welding speed and tilt angle play secondary roles within the investigated parameter space. The explainable artificial intelligence framework provided both accurate prediction and physical insight into the FSW process, enabling data-driven optimization of welding parameters and reducing the need for extensive experimental trials. The proposed approach demonstrates the potential of applying machine learning and explainable AI techniques to intelligent manufacturing applications.</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 modelling of friction stir welded AA6061-T651 tensile strength using random forest and SHAP analysis

  • 期刊论文

  • 发表日期：

    2026年6月18日

查看全文
  • Ifeanyi Uchegbulam,
  • Samuel Chimenem Opuwill,
  • Meekness Nnoka,
  • Tonye Alaso Jack

摘要

Friction stir welding (FSW) is a solid-state joining process widely used for aluminum alloys due to its unique ability to produce high-quality joints with minimal defects. However, predicting the mechanical properties of friction stir welded joints remains challenging because of the complex nonlinear interactions among process parameters. This study presents an explainable machine learning framework for predicting the ultimate tensile strength (UTS) of friction stir welded AA6061-T651 aluminum alloy. A Random Forest regression model was developed using experimental data consisting of four key process parameters: rotational speed, welding speed, tool tilt angle, and axial load. Data preprocessing included normalization and train-test splitting, followed by hyperparameter optimization using GridSearchCV. The optimized model achieved a coefficient of determination (R2) of approximately 0.84 with a root mean square error (RMSE) of about 6 MPa. To interpret the machine learning model’s feature importance, SHAP explainability, partial dependence plots, permutation importance, and learning curve analysis were performed. The results indicate that rotational speed and axial load exert the strongest influence on joint strength, while welding speed and tilt angle play secondary roles within the investigated parameter space. The explainable artificial intelligence framework provided both accurate prediction and physical insight into the FSW process, enabling data-driven optimization of welding parameters and reducing the need for extensive experimental trials. The proposed approach demonstrates the potential of applying machine learning and explainable AI techniques to intelligent manufacturing applications.