Scholarlink

<p>The present research proposes a novel methodology for predicting and optimizing notch tensile strength (NTS) in friction-welded dissimilar stainless-steel joints (AISI 430 and AISI 304) using a hybrid experimental-machine learning (ML) approach. A Taguchi L32 orthogonal array design was employed to experimentally investigate the effects of friction force, forge force, and burn-off length. The optimal parameter combination (6 kN, 12 kN, 3&#xa0;mm) achieved an NTS of 669&#xa0;MPa. To reduce experimental load, ML regression models (k-NN, Decision Tree, Random Forest) were developed to predict NTS based on different input factors. Among them k-NN achieved the best performance (RMSE = 3.20, R<sup>2</sup> = 0.93) under a 90 − 10 train-test split. The proposed framework effectively predicts NTS using minimal experimental data and provides a scalable, cost-efficient tool for welding optimization. The findings hold substantial promise for industrial sectors like automotive and aerospace, where robust joint performance and process efficiency are critical. This is a pioneering work that applies ML for NTS prognosis in dissimilar stainless-steel joints, with implications for integration into smart manufacturing and digital twin environments.</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

Intelligent prediction of notch tensile strength in friction welded dissimilar stainless steel joints using machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月13日

查看全文
  • Jagadesh Kumar Jatavallabhula,
  • Vaddi Venkata Satyanarayana,
  • Gundeti Sreeram Reddy,
  • Ravinder Reddy Baridula,
  • Bridjesh Pappula

摘要

The present research proposes a novel methodology for predicting and optimizing notch tensile strength (NTS) in friction-welded dissimilar stainless-steel joints (AISI 430 and AISI 304) using a hybrid experimental-machine learning (ML) approach. A Taguchi L32 orthogonal array design was employed to experimentally investigate the effects of friction force, forge force, and burn-off length. The optimal parameter combination (6 kN, 12 kN, 3 mm) achieved an NTS of 669 MPa. To reduce experimental load, ML regression models (k-NN, Decision Tree, Random Forest) were developed to predict NTS based on different input factors. Among them k-NN achieved the best performance (RMSE = 3.20, R2 = 0.93) under a 90 − 10 train-test split. The proposed framework effectively predicts NTS using minimal experimental data and provides a scalable, cost-efficient tool for welding optimization. The findings hold substantial promise for industrial sectors like automotive and aerospace, where robust joint performance and process efficiency are critical. This is a pioneering work that applies ML for NTS prognosis in dissimilar stainless-steel joints, with implications for integration into smart manufacturing and digital twin environments.