Scholarlink

<p>Welding is fundamental to the manufacturing industry and is critical for the structural integrity and durability of manufactured goods. The integration of automation and intelligent technologies has rendered intelligent welding techniques essential for increasing the quality and efficiency of welding operations. Machine vision, which is integral to these systems, mimics human visual perception to enable real-time monitoring and quality control in welding. This review explores the role of machine vision in welding, focusing on the contributions of traditional machine learning and deep learning to process monitoring. Traditional methods, such as SVMs and ANNs, are effective in scenarios with limited data and simple feature extraction. These methods excel in tasks where physical features have clear significance but are hindered by the need for complex feature engineering and large annotated datasets. In contrast, deep learning models such as CNNs, GANs, and YOLO can extract complex features from extensive datasets, reducing the need for domain expertise and excelling in large-scale data scenarios. CNNs are particularly adept at image recognition in welding because of their precision and robustness; GANs excel in data augmentation and preprocessing; and YOLO models are recognized for their rapid and accurate real-time object detection capabilities, which are vital for dynamic welding processes. The integration of deep learning into welding processes is anticipated to further automate and intelligentize the industry, with potential advancements in end-to-end weld penetration prediction and multioutput monitoring of high-speed laser welding. As computational resources grow and algorithms evolve, deep learning is expected to enhance quality control, reduce reliance on skilled labor, and optimize manufacturing efficiency. This shift toward smart manufacturing, propelled by deep learning, is set to revolutionize the welding industry by enabling real-time monitoring, defect detection, and predictive maintenance, ensuring the highest standards of quality and consistency in welding products.</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

From classical algorithms to deep learning: a review of machine vision for monitoring welding dynamics

  • 期刊论文

  • 发表日期：

    2025年10月7日

查看全文
  • Lingxiao Song,
  • Peilei Zhang,
  • Kefan Chen,
  • Zufa Li,
  • Hua Yan,
  • Yuze Huang

摘要

Welding is fundamental to the manufacturing industry and is critical for the structural integrity and durability of manufactured goods. The integration of automation and intelligent technologies has rendered intelligent welding techniques essential for increasing the quality and efficiency of welding operations. Machine vision, which is integral to these systems, mimics human visual perception to enable real-time monitoring and quality control in welding. This review explores the role of machine vision in welding, focusing on the contributions of traditional machine learning and deep learning to process monitoring. Traditional methods, such as SVMs and ANNs, are effective in scenarios with limited data and simple feature extraction. These methods excel in tasks where physical features have clear significance but are hindered by the need for complex feature engineering and large annotated datasets. In contrast, deep learning models such as CNNs, GANs, and YOLO can extract complex features from extensive datasets, reducing the need for domain expertise and excelling in large-scale data scenarios. CNNs are particularly adept at image recognition in welding because of their precision and robustness; GANs excel in data augmentation and preprocessing; and YOLO models are recognized for their rapid and accurate real-time object detection capabilities, which are vital for dynamic welding processes. The integration of deep learning into welding processes is anticipated to further automate and intelligentize the industry, with potential advancements in end-to-end weld penetration prediction and multioutput monitoring of high-speed laser welding. As computational resources grow and algorithms evolve, deep learning is expected to enhance quality control, reduce reliance on skilled labor, and optimize manufacturing efficiency. This shift toward smart manufacturing, propelled by deep learning, is set to revolutionize the welding industry by enabling real-time monitoring, defect detection, and predictive maintenance, ensuring the highest standards of quality and consistency in welding products.