Scholarlink

<p>Reheating furnaces are essential in steel manufacturing, ensuring steel reaches the optimal temperature for hot-rolling. Burners within these furnaces produce flames to maintain the necessary thermal conditions. However, inconsistent burner performance can result in irregular or extreme flames, compromising steel quality and production safety. Traditionally, flame monitoring has relied on human supervision, which is inefficient and prone to errors. To overcome these limitations, we propose a computer vision-based system for automated flame monitoring and anomaly detection. The system analyzes the video stream from a thermal camera that continuously monitors the furnace interior. Our methodology involves three steps: (1) detecting flames and furnace keypoints using a deep learning model, (2) quantifying flames across burner regions with traditional computer vision techniques, and (3) identifying anomalies using an interpretable machine learning model. Validation with real-world data from a large steel manufacturing facility demonstrates that the system achieves an F1 score above 80% in detecting anomalies across various burner zones. To support operators, the results are presented in a dashboard that provides both real-time and historical insights into furnace performance. This enables timely anomaly detection and intervention, ensuring safe, efficient, and high-quality steel production.</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

Flame monitoring and anomaly detection in steel reheating furnaces based on thermal video using a hybrid AI computer vision system

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月25日

查看全文
  • Jelle Vanhaeverbeke,
  • Steven Verstockt,
  • Sofie Van Hoecke

摘要

Reheating furnaces are essential in steel manufacturing, ensuring steel reaches the optimal temperature for hot-rolling. Burners within these furnaces produce flames to maintain the necessary thermal conditions. However, inconsistent burner performance can result in irregular or extreme flames, compromising steel quality and production safety. Traditionally, flame monitoring has relied on human supervision, which is inefficient and prone to errors. To overcome these limitations, we propose a computer vision-based system for automated flame monitoring and anomaly detection. The system analyzes the video stream from a thermal camera that continuously monitors the furnace interior. Our methodology involves three steps: (1) detecting flames and furnace keypoints using a deep learning model, (2) quantifying flames across burner regions with traditional computer vision techniques, and (3) identifying anomalies using an interpretable machine learning model. Validation with real-world data from a large steel manufacturing facility demonstrates that the system achieves an F1 score above 80% in detecting anomalies across various burner zones. To support operators, the results are presented in a dashboard that provides both real-time and historical insights into furnace performance. This enables timely anomaly detection and intervention, ensuring safe, efficient, and high-quality steel production.