Automation of Mission Operations & Testing with Digital Twins

How do you plan a one-off satellite mission from scratch? To be even more specific, how do you plan a one-off satellite mission from scratch and incorporate machine learning? Will your machine learning operate strictly on sensor data, or will it be used to interact with the spacecraft fault detection and correction mechanisms? How will you thoroughly test it - you are testing thoroughly, right?

With long lead times for hardware and a desire for shorter design lifecycles, it is difficult to maintain tight schedules without even considering the incorporation of machine learning. If machine learning is to be incorporated into a spacecraft bus first, you’ll need a working representation of that bus and traffic to even begin testing the algorithm. Depending on the implementation that could mean simulated ground control software with command ingest and telemetry output, power system, attitude control, time drivers, sensor data, and so on – in short, an entire digital twin.

The NASA Operational Simulator for Small Satellites (NOS3) provides an open source out of the box design reference mission. The system runs NASA’s core Flight Software (cFS) by default with ground system support for COSMOS, F’, YAMCS, and AIT. Currently it is being used to support the development of the On-Board Artificial Intelligence Research (OnAIR) plugin for adding ML plugins into cFS, Bump in the Wire for cybersecurity monitoring and research, as well as JPL’s SYNOPSIS for automated flagging of interesting sensor data for downlink.