Real-Time Pipeline Leak Detection Using YOLO Model with Generative AI for Intelligent Response Automation

Abstract

Pipeline leakage presents a significant operational, environmental, and safety challenge in the oil and gas industry, with undetected failures leading to catastrophic consequences, including product loss, environmental contamination, and infrastructure damage. Traditional detection methods, such as pressure and flow monitoring or sensor networks, are limited by latency, false alarms, and the need for complex hardware integration. Recent advances in computer vision have demonstrated the potential for automated visual leak detection using deep learning models, particularly real-time object detectors that identify leak signatures in image data from monitoring cameras. This work proposed an AI-based pipeline monitoring system that integrates real-time leak detection with severity estimation, risk scoring, and structured corrective guidance. A YOLO11 model, trained on a curated pipeline image dataset, identifies leak regions and calculates bounding-box-derived severity and risk scores. Detected leaks are analysed using a Vision Language model (Gemini 2.5 flash) to generate actionable, professionally formatted recommendations, delivered via text and audio. Experimental evaluation demonstrates that the proposed system provides accurate visual leak detection and practical decision support, offering a scalable, low-cost alternative to conventional sensing-heavy systems and establishing a framework for enhanced AI-assisted pipeline monitoring.