Scholarlink

<p>Gas Metal Arc Welding (GMAW) is extensively used for structural steels, where the tensile strength of welded joints serves as a critical quality metric. However, predicting tensile strength from process parameters remains challenging due to nonlinear interactions among welding current, arc voltage, travel speed, and heat input. In this study, a curated dataset of 309 experimental records from published literature was analyzed to develop predictive models using machine learning. Three supervised algorithms-linear regression, artificial neural networks, and random forest-were implemented alongside an advanced gradient boosting method for comparison. Univariate regression analysis showed voltage and heat input as dominant individual predictors, and machine learning revealed that multivariate interactions govern tensile strength. Gradient boosting achieved the best performance with an <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\:{R}^{2}=0.991\)</EquationSource> </InlineEquation> and mean absolute percentage error below <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\:4\text{\%}\)</EquationSource> </InlineEquation>, outperforming both random forest ( <InlineEquation ID="IEq3"> <EquationSource Format="TEX">\(\:{R}^{2}=0.912\)</EquationSource> </InlineEquation> ) and neural networks ( <InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(\:{R}^{2}=0.883\)</EquationSource> </InlineEquation> ). These results highlight the capability of boosting-based models to capture nonlinearities and provide reliable predictive accuracy even with medium-sized datasets. The findings demonstrate that interpretable machine learning can serve as a robust framework for welding optimization, bridging the gap between empirical approaches and data-driven process design.</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 prediction of tensile strength in gas metal arc welding: gradient boosting, neural networks and random forests

  • 期刊论文

  • 发表日期：

    2025年12月29日

查看全文
  • Amir Reza Ansari Dezfoli,
  • Li-Shang Lin,
  • Sanaz Hadidchi,
  • Yi-Jen Huang

摘要

Gas Metal Arc Welding (GMAW) is extensively used for structural steels, where the tensile strength of welded joints serves as a critical quality metric. However, predicting tensile strength from process parameters remains challenging due to nonlinear interactions among welding current, arc voltage, travel speed, and heat input. In this study, a curated dataset of 309 experimental records from published literature was analyzed to develop predictive models using machine learning. Three supervised algorithms-linear regression, artificial neural networks, and random forest-were implemented alongside an advanced gradient boosting method for comparison. Univariate regression analysis showed voltage and heat input as dominant individual predictors, and machine learning revealed that multivariate interactions govern tensile strength. Gradient boosting achieved the best performance with an \(\:{R}^{2}=0.991\) and mean absolute percentage error below \(\:4\text{\%}\) , outperforming both random forest ( \(\:{R}^{2}=0.912\) ) and neural networks ( \(\:{R}^{2}=0.883\) ). These results highlight the capability of boosting-based models to capture nonlinearities and provide reliable predictive accuracy even with medium-sized datasets. The findings demonstrate that interpretable machine learning can serve as a robust framework for welding optimization, bridging the gap between empirical approaches and data-driven process design.