Scholarlink

<p>Friction stir welding (FSW) is a widely adopted method in industries such as aerospace, automotive, and marine for joining aluminum alloys, offering enhanced weld quality and mechanical properties. However, the optimization of FSW parameters remains a complex task due to the intricate relationships between process variables and weld characteristics. This study aims to develop a predictive model for estimating key FSW parameters, such as rotation speed, welding speed, and tilt angle, using machine learning techniques. An Auto-Encoder model is proposed and compared against traditional models including Long Short-Term Memory (LSTM), Support Vector Machines (SVM), and Gaussian Process Regression (GPR). The dataset includes FSW experiments on similar (AA7075) and dissimilar (AA5083 and AA6061) aluminum alloys. The Auto-Encoder model consistently outperforms other methods in predicting tensile strength, with a 5.679% average error for AA7075, demonstrating its superior accuracy and potential for industrial applications. The findings provide a robust framework for optimizing FSW processes, significantly reducing the trial-and-error approach traditionally used in parameter estimation.</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-based prediction of friction stir welding parameters for optimized tensile strength in aluminum alloys

  • 期刊论文

  • 发表日期：

    2025年12月16日

查看全文
  • Peerawat Luesak,
  • Worapot Sirirak,
  • Surasak Matitopanum,
  • Thanatkij Srichok,
  • Surajet Khonjun,
  • Somphop Chiaranai,
  • Ganokgarn Jirasirilerd,
  • Sarayut Gonwirat

摘要

Friction stir welding (FSW) is a widely adopted method in industries such as aerospace, automotive, and marine for joining aluminum alloys, offering enhanced weld quality and mechanical properties. However, the optimization of FSW parameters remains a complex task due to the intricate relationships between process variables and weld characteristics. This study aims to develop a predictive model for estimating key FSW parameters, such as rotation speed, welding speed, and tilt angle, using machine learning techniques. An Auto-Encoder model is proposed and compared against traditional models including Long Short-Term Memory (LSTM), Support Vector Machines (SVM), and Gaussian Process Regression (GPR). The dataset includes FSW experiments on similar (AA7075) and dissimilar (AA5083 and AA6061) aluminum alloys. The Auto-Encoder model consistently outperforms other methods in predicting tensile strength, with a 5.679% average error for AA7075, demonstrating its superior accuracy and potential for industrial applications. The findings provide a robust framework for optimizing FSW processes, significantly reducing the trial-and-error approach traditionally used in parameter estimation.

Graphical Abstract