Tank Level Monitoring Using Thermal Video Processing Model

Le résumé fourni par la source

Abstract The conventional method of using liquid-level sensors like guided wave radar (GWR) for liquid level detection in oil tanks poses significant safety and environmental concerns. The process of opening the tank hatch for maintenance and calibration exposes workers to hazardous gasses and increases the risk of explosions. Additionally, the release of harmful substances during calibration contributes to environmental pollution. To address these challenges, we developed a non-intrusive approach using thermal video processing model as a safer and more environmentally friendly alternative to GWR. The model utilizes thermal videos captured by a thermal camera and employs advanced deep Learning image processing techniques to detect thermal differences between the full and empty parts of the tank. By accurately estimating the liquid level based on these thermal patterns, the non-intrusive approach eliminates the need for physical access to the tank, thereby mitigating safety risks and minimizing environmental impact. The performance of the thermal video processing model is evaluated by collecting data from multiple tanks and comparing the results with those obtained from guided wave radar. We evaluated the model in different edge cases including different weather conditions like snow, storm, rain, and also different temperatures. We also evaluated the model in shadow, camera shake, and blocking objects like tanker trucks, pipes or other tanks. The results of this work demonstrate that the thermal video processing model achieves comparable accuracy to guided wave radar for liquid level detection in oil tanks. The model successfully identifies and differentiates thermal patterns with more than 94% accuracy, enabling precise estimation of the liquid level with 98% accuracy. By eliminating the need to open the tank hatch, this method ensures worker safety by minimizing their exposure to hazardous gasses and reducing the risk of explosions and release of harmful substances to air. The deep learning video processing model performing tank level detection is very robust to edge cases including different temperature or weather conditions. It is also very resilient to shadow during the day and can detect levels even when less than 10% of the tank is visible. The findings demonstrate the potential of thermal image processing as a viable and efficient approach, offering a safer and more sustainable solution for the oil industry.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.