Air leak material identification in pressurized space vehicles using a Convolutional Neural Network

Pressurized space vehicles of all types are at risk of depressurization due to leaking air. Leaks may be caused by micro-meteor and orbital debris (MMOD) impact or structural aging and failure overtime. This paper addresses the issue of leak type detection by analyzing airborne ultrasonic waves using a Convolutional Neural Network. Depending on the vessel material, size of the leak, and pressure gradient, different waveforms are produced. Once a large number of samples have been recorded, the resulting data is used for the training of a Convolutional Neural Network for leak classification.