Deep Learning Approach for Pitting Corrosion Detection in Gas Pipelines

Area of Science:

• Computer Vision
• Materials Science
• Artificial Intelligence

Background:

• Pitting corrosion in gas pipelines poses significant risks.
• Manual inspection for corrosion is labor-intensive and costly.
• Advanced detection methods are crucial for infrastructure integrity.

Purpose of the Study:

• To develop an automated computer vision methodology for detecting pitting corrosion in gas pipelines.
• To create an efficient and accurate deep learning model for corrosion classification.
• To reduce the reliance on manual pipeline inspection.

Main Methods:

• Curated a large dataset of 576,000 pipeline images.
• Designed and optimized a custom convolutional neural network (CNN) for binary classification.
• Utilized deep learning for image analysis and corrosion detection.

Main Results:

• Achieved a high classification accuracy of 98.44% for detecting pitting corrosion.
• The custom CNN model demonstrated superior performance compared to contemporary classifiers.
• The methodology effectively distinguishes between corroded and non-corroded pipeline images.

Conclusions:

• The proposed computer vision and deep learning approach offers an effective solution for automated pitting corrosion detection in gas pipelines.
• This method significantly streamlines inspection processes, reducing time and costs.
• The high accuracy of the CNN model validates its potential for real-world applications in pipeline integrity management.