Are there enough open source aero engineering projects to give the current ai approaches a remotely plausible amount of training data?