Scholarlink

<p>Micrographs play a crucial role in evaluating, developing, and enhancing coatings by providing detailed insights into their microscopic structures. This paper presents a novel methodology for generating synthetic images from micrographs of copper coatings on stainless steel AISI 304 disks and validates their utility in data augmentation for the identification of electrodeposition time. We employ two convolutional neural networks –one to encode micrograph textures and another to generate synthetic images– resulting in realistic synthetic micrographs. The similarity between these synthetic and real images is assessed using contrast and energy property distributions via gray-level co-occurrence matrices and the Kolmogorov–Smirnov test. The resulting <i>p</i>-values indicate a significant similarity, confirming the method’s suitability for data augmentation. By expanding the original dataset 16 times and utilizing the AlexNet deep learning architecture, we demonstrate, across various experimental scenarios, the network’s enhanced ability to identify the electrodeposition time, a critical parameter for coating consistency and durability, used to generate the coated material. The synthetic images enable the network to learn subtle visual variations in the copper coating that are often challenging for the human eye to detect.The ability to generate high-fidelity synthetic images not only enriches training datasets, but also accelerates the development of robust, AI-driven quality control systems, allowing for more efficient and precise manufacturing practices. The proposed methodology has implications for quality control and the optimization of electrodeposition processes.</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

Deep learning-based identification of electrodeposition time: a case study using synthetic images of copper coatings

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年5月10日

查看全文
  • Carlos Acuña,
  • David Chávez,
  • Carmen Velázquez ,
  • David Velazco,
  • Gregorio Vargas,
  • Mario Castelán

摘要

Micrographs play a crucial role in evaluating, developing, and enhancing coatings by providing detailed insights into their microscopic structures. This paper presents a novel methodology for generating synthetic images from micrographs of copper coatings on stainless steel AISI 304 disks and validates their utility in data augmentation for the identification of electrodeposition time. We employ two convolutional neural networks –one to encode micrograph textures and another to generate synthetic images– resulting in realistic synthetic micrographs. The similarity between these synthetic and real images is assessed using contrast and energy property distributions via gray-level co-occurrence matrices and the Kolmogorov–Smirnov test. The resulting p-values indicate a significant similarity, confirming the method’s suitability for data augmentation. By expanding the original dataset 16 times and utilizing the AlexNet deep learning architecture, we demonstrate, across various experimental scenarios, the network’s enhanced ability to identify the electrodeposition time, a critical parameter for coating consistency and durability, used to generate the coated material. The synthetic images enable the network to learn subtle visual variations in the copper coating that are often challenging for the human eye to detect.The ability to generate high-fidelity synthetic images not only enriches training datasets, but also accelerates the development of robust, AI-driven quality control systems, allowing for more efficient and precise manufacturing practices. The proposed methodology has implications for quality control and the optimization of electrodeposition processes.