Scholarlink

<p>Defects in steel products significantly impact their quality, structural integrity, and durability. Aligned with the objectives of Industry 4.0, this study developed an automated system to accurately identify and classify defects on steel surfaces, enhancing overall manufacturing quality and efficiency. Traditional manual defect classification processes are inefficient and subjective, making them prone to human error and time-consuming. By addressing a real industrial challenge, this research focuses on deploying deep learning algorithms, specifically Convolutional Neural Networks (CNNs), to automate and improve quality control processes in the manufacturing sector, with strong potential for deployment in real-world industrial environments. The study’s primary objective is to create a robust steel defect classification system with several specific goals: enhancing the dataset through geometric transformations like rotation to improve model generalization, employing adaptive thresholding techniques for accurate defect localization, and implementing and evaluating various deep learning models such as basic CNN, Residual Network (ResNet), Adaptive Sampling Network (AsNet), and the enhanced Visual Geometry Group 19 (VGG19) architecture. Comprehensive evaluation using metrics such as accuracy, F1-score, and recall ensured a thorough assessment of the system’s effectiveness. Furthermore, the developed system demonstrates real-world applicability and efficiency, capable of classifying diverse steel defect scenarios, thereby improving manufacturing quality and reducing operational costs. Our customized VGG19 model, enhanced with additional layers, achieved superior performance achieving an accuracy of 99.87%. This study illustrates the practical application of deep learning in manufacturing, specifically using the VGG19 model to refine steel defect classification and quality control practices. The results promise substantial improvements in manufacturing efficiency and product quality assurance, highlighting the potential for future integration of this model into real-time manufacturing processes and further advancements in deep learning techniques.</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

SteelQC-VGG: a deep learning-driven quality control system for automated surface defect classification in steel manufacturing

  • 期刊论文

  • 发表日期：

    2025年12月5日

查看全文
  • Marouene Chaieb,
  • Heni Mayara,
  • Wajdi Hamdoun,
  • Maram Gabsi,
  • Ines Hilali Jaghdam

摘要

Defects in steel products significantly impact their quality, structural integrity, and durability. Aligned with the objectives of Industry 4.0, this study developed an automated system to accurately identify and classify defects on steel surfaces, enhancing overall manufacturing quality and efficiency. Traditional manual defect classification processes are inefficient and subjective, making them prone to human error and time-consuming. By addressing a real industrial challenge, this research focuses on deploying deep learning algorithms, specifically Convolutional Neural Networks (CNNs), to automate and improve quality control processes in the manufacturing sector, with strong potential for deployment in real-world industrial environments. The study’s primary objective is to create a robust steel defect classification system with several specific goals: enhancing the dataset through geometric transformations like rotation to improve model generalization, employing adaptive thresholding techniques for accurate defect localization, and implementing and evaluating various deep learning models such as basic CNN, Residual Network (ResNet), Adaptive Sampling Network (AsNet), and the enhanced Visual Geometry Group 19 (VGG19) architecture. Comprehensive evaluation using metrics such as accuracy, F1-score, and recall ensured a thorough assessment of the system’s effectiveness. Furthermore, the developed system demonstrates real-world applicability and efficiency, capable of classifying diverse steel defect scenarios, thereby improving manufacturing quality and reducing operational costs. Our customized VGG19 model, enhanced with additional layers, achieved superior performance achieving an accuracy of 99.87%. This study illustrates the practical application of deep learning in manufacturing, specifically using the VGG19 model to refine steel defect classification and quality control practices. The results promise substantial improvements in manufacturing efficiency and product quality assurance, highlighting the potential for future integration of this model into real-time manufacturing processes and further advancements in deep learning techniques.