Automated detection and classification of defects in the outer surface of transportation pipelines

Abstract

This Dissertation explores the critical importance of defect detection in pipelines and advocates for the integration of artificial intelligence (AI) to enhance inspection capabilities. Traditional methods of pipeline inspection are prone to human error and lack scalability. By employing machine learning models, this study proposes a novel approach to pipeline inspection. The integration of AI offers numerous advantages, including increased efficiency, accuracy, and scalability in defect detection. Through rigorous experimentation and evaluation, this research demonstrates the effectiveness of AI-driven approaches in enhancing pipeline integrity management. Furthermore, the study emphasizes the sensitivity of AI-based defect detection systems and underscores the significance of feature engineering using deep learning techniques. By extracting rich features from pipeline images, the VGG architecture combined with machine learning models facilitates more robust and discriminative representations, enhancing the model's ability to detect subtle defects with high precision and recall. This dissertation contributes to the advancement of pipeline inspection practices, highlighting the potential of integrating advanced technologies to ensure safer, more reliable, and cost-effective maintenance strategies in the industrial sector.