Scholarlink

<p>Corrosion inspection is a critical task for maintaining offshore production systems, which are exposed to aggressive marine environments and characterized by high operational and maintenance costs. Traditional visual inspection methods, although widely used, are inherently subjective, labor-intensive, and hazardous due to the need for on-site human intervention. This paper proposes an automated methodology for the detection and classification of atmospheric corrosion using conventional digital images, image processing algorithms, and supervised machine learning techniques. The methodology includes a standardized image acquisition protocol to enhance result consistency and reduce variability across inspection scenarios. Using a labeled dataset aligned with ASTM D610-1 standards, the approach achieved over 99% pixel-wise segmentation accuracy and 76% accuracy in corrosion type classification. The workflow enables automatic extraction of quantitative features such as corroded area percentage, spot size, and distribution, which are essential for supporting predictive maintenance strategies. The proposed system contributes to the advancement of smart maintenance by reducing subjectivity, improving repeatability, and providing a scalable alternative for corrosion monitoring in offshore oil and gas assets.</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

Automatic corrosion segmentation and classification using image processing and machine learning

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Marina Polonia Rios,
  • Paulo Ivson Netto Santos,
  • Deane Roehl

摘要

Corrosion inspection is a critical task for maintaining offshore production systems, which are exposed to aggressive marine environments and characterized by high operational and maintenance costs. Traditional visual inspection methods, although widely used, are inherently subjective, labor-intensive, and hazardous due to the need for on-site human intervention. This paper proposes an automated methodology for the detection and classification of atmospheric corrosion using conventional digital images, image processing algorithms, and supervised machine learning techniques. The methodology includes a standardized image acquisition protocol to enhance result consistency and reduce variability across inspection scenarios. Using a labeled dataset aligned with ASTM D610-1 standards, the approach achieved over 99% pixel-wise segmentation accuracy and 76% accuracy in corrosion type classification. The workflow enables automatic extraction of quantitative features such as corroded area percentage, spot size, and distribution, which are essential for supporting predictive maintenance strategies. The proposed system contributes to the advancement of smart maintenance by reducing subjectivity, improving repeatability, and providing a scalable alternative for corrosion monitoring in offshore oil and gas assets.