Gas Leak Detection Using Acoustics and Artificial Intelligence

Gas leak detection on a production site is a major challenge for the safety and health of workers, for environmental considerations and from an economic point of view. In addition, flammable gas leaks are a safety risk because if ignited, they can cause serious fires or explosions. For these reasons, Acoem Metravib in collaboration with TotalEnergies One Tech R&D Safety has developed for the past four years a system called AGLED for the early detection, localization and classification of such leaks exploiting acoustics and artificial intelligence driven by physics. Numerous tests have been conducted on a theater representative of gas production facilities created by TotalEnergies in Lacq (France) to build a robust learning database of leaks varying in flowrates, exhaust diameters and also types (hole, nozzle, flange...). Moreover, to limit the number of false alarms, a relearning strategy has been implemented to handle unexpected disturbances (wildlife, human activities, meteorological events...). The presented paper describes the global architecture of the system from noise acquisition to the gas leak probability and coordinates. It gives a more in-depth look at the relearning algorithm and its performance in various environments. Finally, thanks to a complementary collaboration with Air Liquide, an example of test campaign in a real industrial environment is presented with an emphasis on the improvement obtained through relearning.