Scholarlink

<p>Ultrasonic metal welding (USMW) is increasingly employed for joining electrical conductors, especially in the context of lightweighting strategies that involve replacing copper with aluminum. This makes the technique highly relevant across sectors such as automotive and aerospace. However, USMW still suffers from limited process transparency. In current industrial practice, weld quality is primarily verified through selective destructive testing. As a result, it is feasible neither to inspect every joint nor to avoid false classifications, leading to both unnecessary rejects and undetected defective welds. This contribution presents a machine learning (ML)-based approach to enable real-time process monitoring using data directly obtained from the welding system. By leveraging signal characteristics captured during welding, the system not only distinguishes between “OK” and “NOK” welds but also enables a detailed categorization of defect types. In addition, a predictive model for pull-out force was integrated, allowing quantitative assessment of weld integrity without physical testing. Validation results show a classification accuracy of 99.9% and a mean absolute error for regression of 75 N, demonstrating the method’s potential to enhance process reliability while significantly reducing both scrap and test effort. The approach lays the groundwork for data-driven quality assurance in USMW and supports the implementation of robust inline monitoring.</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

Development of a process monitoring method for ultrasonic metal welding of automotive wires based only on machine sensor data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月27日

查看全文
  • Andreas Gester,
  • Anja Tetzner,
  • Guntram Wagner,
  • Peter Gluchowski,
  • Melanie Becker,
  • Morten Deutsch,
  • David Leoka

摘要

Ultrasonic metal welding (USMW) is increasingly employed for joining electrical conductors, especially in the context of lightweighting strategies that involve replacing copper with aluminum. This makes the technique highly relevant across sectors such as automotive and aerospace. However, USMW still suffers from limited process transparency. In current industrial practice, weld quality is primarily verified through selective destructive testing. As a result, it is feasible neither to inspect every joint nor to avoid false classifications, leading to both unnecessary rejects and undetected defective welds. This contribution presents a machine learning (ML)-based approach to enable real-time process monitoring using data directly obtained from the welding system. By leveraging signal characteristics captured during welding, the system not only distinguishes between “OK” and “NOK” welds but also enables a detailed categorization of defect types. In addition, a predictive model for pull-out force was integrated, allowing quantitative assessment of weld integrity without physical testing. Validation results show a classification accuracy of 99.9% and a mean absolute error for regression of 75 N, demonstrating the method’s potential to enhance process reliability while significantly reducing both scrap and test effort. The approach lays the groundwork for data-driven quality assurance in USMW and supports the implementation of robust inline monitoring.