Scholarlink

<p>This study evaluates advanced machine learning (ML) and deep learning (DL) models for predicting the tensile shear and bending strength of friction stir spot welding joints involving Al 5754 and Al 6111 alloys. ML techniques include Linear Regression, Decision Tree, Random Forest (RF), K-Nearest Neighbors, Support Vector Regression, and XGBoost, while DL models comprise Recurrent Neural Network (RNN) and Backpropagation Neural Network (BPNN). The models were assessed for discrepancies between experimental and predicted results, with the best-performing model identified using <i>R</i>-squared (<i>R</i><sup>2</sup>), Root-Mean-Square Error, Mean Square Error, and Mean Absolute Error. The data preprocessing phase included feature scaling and an 85:15 train-test split. Key input process parameters included spindle speed, dwell time, plunge depth, and tool pin profile. The results demonstrate that XGBoost yielded the highest predictive accuracy, achieving an <i>R</i><sup>2</sup> score of 99.99% for both tensile shear and bending strength, while RF offered a strong balance between accuracy and robustness. Other ML models struggled with the dataset’s complexity, resulting in lower performance. Among DL approaches, the BPNN outperformed the RNN, achieving approximately 99.8% accuracy by effectively capturing complex data patterns.</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

Performance Evaluation of Friction Stir Spot Welding of Al 5754 and Al 6111 using Machine Learning Approaches

  • 期刊论文

  • 发表日期：

    2025年8月13日

查看全文
  • R. Keshavamurthy,
  • B. E. Naveena,
  • P. N. Vikram Gowda,
  • T. V. Shrivathsa

摘要

This study evaluates advanced machine learning (ML) and deep learning (DL) models for predicting the tensile shear and bending strength of friction stir spot welding joints involving Al 5754 and Al 6111 alloys. ML techniques include Linear Regression, Decision Tree, Random Forest (RF), K-Nearest Neighbors, Support Vector Regression, and XGBoost, while DL models comprise Recurrent Neural Network (RNN) and Backpropagation Neural Network (BPNN). The models were assessed for discrepancies between experimental and predicted results, with the best-performing model identified using R-squared (R2), Root-Mean-Square Error, Mean Square Error, and Mean Absolute Error. The data preprocessing phase included feature scaling and an 85:15 train-test split. Key input process parameters included spindle speed, dwell time, plunge depth, and tool pin profile. The results demonstrate that XGBoost yielded the highest predictive accuracy, achieving an R2 score of 99.99% for both tensile shear and bending strength, while RF offered a strong balance between accuracy and robustness. Other ML models struggled with the dataset’s complexity, resulting in lower performance. Among DL approaches, the BPNN outperformed the RNN, achieving approximately 99.8% accuracy by effectively capturing complex data patterns.