دانشگاه آزاد اسلامی واحد تهران جنوب

صفحه اصلی / نخستین همایش هوش مصنوعی و پژوهش های نوظهور: همگرایی انسان و سیستم های هوشمند

Advanced Pipe Identification: A Deep Learning Approach in Petroleum Engineering

نویسندگان :

Mahdi Mohammad Ali Ebrahim¹ Seyyed Shahab Tabatabaee Moradi² Alireza Behinrad³

1- دانشگاه صنعتی سهند 2- دانشگاه صنعتی سهند 3- دانشگاه صنعتی سهند

کلمات کلیدی :

Pipe Identification،Deep Learning،Object Detection،Petroleum Engineering

چکیده :

This study created YOLOv12, a novel object detection framework applied to petroleum engineering for automated pipe recognition and classification from industrial datasets. Leveraging YOLOv12's robust real-time processing performance, the research foucsed on amplifying an automated and trustworthy approach in pipeline and asset management. The developed general model had a mean Average Precision (mAP) of 98.70%, with Average Precision scores across all IoU ranges (0.5-0.95) indicating precision at 99.8%, and recall at 99.6%, illustrating a very high accuracy and robustness level incorporating a very imbalanced data set. Model training stabilized between 20, and 30 epochs, suggesting good efficiency with terminal convergence without overfitting. Focal loss provided good effectiveness in classes; meanwhile, data augmentation allowed for a more generalizable model. Performance-wise, the model offers practical and efficient solutions for scalable and verifiable automated inventory management, lowering operational costs. The work was significant, it is a first step into applying state of the art deep learning to niche industrial applications. In particular, the challenges of class imbalance, and model generalization were addressed to begin the foundation development for future petroleum engineering research on specific applications.