Scholarlink

<p>The current transition from blast furnaces to electric arc furnaces (EAF) in the steel industry requires that contaminants, such as copper, be removed from the input stream already during the processing phase. In the case of copper, this is particularly important because copper does not accumulate in the slag phase or gas phase during steel production in an EAF but instead remains in the melt. The increased copper content leads to stresses at the grain boundaries after solidification and subsequently to failure due to fracture or cracking during forming processes. This contribution demonstrates how copper-containing particles in a&#xa0;shredder fraction can be detected using Convolutional Neural Networks (CNNs). For this purpose, 20 CNN architectures were compared based on their achievable prediction accuracy and inference latency. The most efficient architecture was selected based on these metrics, and the inference latency was further reduced through optimization. In the final step, the adapted architecture was applied in inline experiments to detect copper particles and control an ejection mechanism.</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

Klassifizierung von Metallschrott mittels Deep Learning Methoden

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年3月26日

查看全文
  • Gerald Koinig,
  • Melanie Neubauer,
  • Walter Martinelli,
  • Yves Radmann,
  • Nikolai Kuhn,
  • Thomas Fink,
  • Elmar Rückert,
  • Bojan Lorber,
  • Alexia Tischberger-Aldrian

摘要

The current transition from blast furnaces to electric arc furnaces (EAF) in the steel industry requires that contaminants, such as copper, be removed from the input stream already during the processing phase. In the case of copper, this is particularly important because copper does not accumulate in the slag phase or gas phase during steel production in an EAF but instead remains in the melt. The increased copper content leads to stresses at the grain boundaries after solidification and subsequently to failure due to fracture or cracking during forming processes. This contribution demonstrates how copper-containing particles in a shredder fraction can be detected using Convolutional Neural Networks (CNNs). For this purpose, 20 CNN architectures were compared based on their achievable prediction accuracy and inference latency. The most efficient architecture was selected based on these metrics, and the inference latency was further reduced through optimization. In the final step, the adapted architecture was applied in inline experiments to detect copper particles and control an ejection mechanism.