Scholarlink

<p>The growing demand for lightweight, high-strength materials in the aerospace and automotive industries has brought aluminium alloys such as AA2024-T351 into the spotlight due to their outstanding strength-to-weight ratio. However, achieving reliable joints in such alloys remains a significant challenge, particularly when using conventional welding techniques. To address this, the present study focuses on optimizing process parameters in Direct Drive Friction Welding (DDFW) for AA2024-T351. A novel drilled-to-bossed geometry, inspired by the traditional mortise-and-tenon joint, is introduced to enhance mechanical interlocking and improve weld integrity. Experimental trials were systematically designed using an L<sub>18</sub> orthogonal array to evaluate both tensile and torsional strength. The experiments were conducted on a customized engine lathe, with friction pressure, forging pressure, spindle speed, faying surface geometry, and friction time selected as the controllable process parameters. In parallel, advanced deep learning models, including an ensemble residual network, a compact attention network, and an adaptive multiscale network, were implemented to predict ultimate tensile strength based on the welding conditions. The results revealed that the optimal parameter combination of 30&#xa0;MPa friction pressure, 70&#xa0;MPa forging pressure, 2200&#xa0;rpm spindle speed, drilled-to-bossed geometry, and 4&#xa0;min of friction time yielded exceptional mechanical performance, achieving a tensile strength of 538&#xa0;MPa and torsional strength of 325&#xa0;MPa. The drilled-to-bossed configuration demonstrated significantly higher joint strength compared to conventional flat-to-flat joints. Moreover, among the deep learning models, the ensemble residual network achieved the highest predictive accuracy with an R² value of approximately 0.81, effectively capturing the complex relationship between process parameters and weld strength.</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

Enhancing weld performance of AA2024-T351 using drilled-to-bossed geometry and deep learning prediction in direct drive friction welding

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月10日

查看全文
  • Eyob Messele Sefene,
  • Alemayehu Admas Abay,
  • Akshansh Mishra,
  • Assefa Asmare Tsegaw,
  • William Gomez,
  • Addisalem Desalegn Sisay,
  • Ajay Kumar

摘要

The growing demand for lightweight, high-strength materials in the aerospace and automotive industries has brought aluminium alloys such as AA2024-T351 into the spotlight due to their outstanding strength-to-weight ratio. However, achieving reliable joints in such alloys remains a significant challenge, particularly when using conventional welding techniques. To address this, the present study focuses on optimizing process parameters in Direct Drive Friction Welding (DDFW) for AA2024-T351. A novel drilled-to-bossed geometry, inspired by the traditional mortise-and-tenon joint, is introduced to enhance mechanical interlocking and improve weld integrity. Experimental trials were systematically designed using an L18 orthogonal array to evaluate both tensile and torsional strength. The experiments were conducted on a customized engine lathe, with friction pressure, forging pressure, spindle speed, faying surface geometry, and friction time selected as the controllable process parameters. In parallel, advanced deep learning models, including an ensemble residual network, a compact attention network, and an adaptive multiscale network, were implemented to predict ultimate tensile strength based on the welding conditions. The results revealed that the optimal parameter combination of 30 MPa friction pressure, 70 MPa forging pressure, 2200 rpm spindle speed, drilled-to-bossed geometry, and 4 min of friction time yielded exceptional mechanical performance, achieving a tensile strength of 538 MPa and torsional strength of 325 MPa. The drilled-to-bossed configuration demonstrated significantly higher joint strength compared to conventional flat-to-flat joints. Moreover, among the deep learning models, the ensemble residual network achieved the highest predictive accuracy with an R² value of approximately 0.81, effectively capturing the complex relationship between process parameters and weld strength.