Space Force funds experiment on use of AI to predict satellite failures

by Sandra Erwin

Last week I wrote about DARPA CASTLE using AI to have the network learn when it was getting attacked. And i mused how cool it would be if we applied it to satellites. Turns out DARPA was thinking the same way!

RS21 a data science startup has been awarded a contract to research the use of AI to predict satellite failures. “The AI-powered monitoring system performs fault detection to predict satellite malfunctions. It uses real-time satellite telemetry data and anomaly messages to predict a lead time before satellites fail,”.

Ok, so its not 100% the same of the CASTLE cybersecurity platform.

BUT, say this AI works to characterize and learn the fault detection and real vehicle anomolies. Then, how hard of it is it to jump to learn about when it is being attacked? (real question, my background is in space and mechanical engineering, I have no real clue how hard writing AI programs are)

Here is my pitch to DARPA though, to a satellite operator one symptom (say spotty telemetry on the down link) can have many causes (rain, bad ground or satellite antenna pointing, failures in the communication subsystem units, etc) AI can completely help with this. Trending of unit data, end of life predictions, and other data it uses can better inform the community of what a true root cause is. So applying it to also look for something that is nefarious should also be included.