Scholarlink

In industrial manufacturing, ensuring the quality and performance of welded components is essential. This is especially true for friction stir welding (FSW) of aluminum alloys such as AA5083 and AA6061, where defect detection is essential for maintaining high standards. In this context, the application of Convolutional Autoencoders (CAEs) offers a promising solution for identifying defects that vary in appearance and location. This work addresses this need by evaluating three distinct CAE models (InceptionCAE, ResNetCAE, and MVTecCAE) applied to a dataset of FSW images. The performance of these models has been assessed using multiple pre-processing techniques, including noise removal, contrast enhancement, and brightness adjustment, to enhance image quality and improve defect detection accuracy. The implementation of these models produced successful results, demonstrating a viable approach to enhance the efficiency and reliability of defect detection processes in the FSW of AA5083 and AA6061 alloys.

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

Friction Stir Welding of Aluminum Alloys Defect Detection Based on Convolutional Autoencoders

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Paula Arcano-Bea,
  • Marta-María Álvarez-Crespo,
  • Francisco Zayas-Gato,
  • Pablo Fariñas,
  • José Luis Calvo-Rolle

摘要

In industrial manufacturing, ensuring the quality and performance of welded components is essential. This is especially true for friction stir welding (FSW) of aluminum alloys such as AA5083 and AA6061, where defect detection is essential for maintaining high standards. In this context, the application of Convolutional Autoencoders (CAEs) offers a promising solution for identifying defects that vary in appearance and location. This work addresses this need by evaluating three distinct CAE models (InceptionCAE, ResNetCAE, and MVTecCAE) applied to a dataset of FSW images. The performance of these models has been assessed using multiple pre-processing techniques, including noise removal, contrast enhancement, and brightness adjustment, to enhance image quality and improve defect detection accuracy. The implementation of these models produced successful results, demonstrating a viable approach to enhance the efficiency and reliability of defect detection processes in the FSW of AA5083 and AA6061 alloys.