AbstractWe propose to create a first-of-its-kind dataset of on-orbit SmallSat telemetry data to be utilized as a benchmark and a data-driven testing ground for developing, applying and assuring safety-critical autonomous and resilient NASA software systems. The foundation of such a unique dataset will be based upon existing on-orbit telemetry data already obtained from the NASA IV&V flagship SmallSat, Simulation-To-Flight-1 (STF-1). By making use of cutting-edge Machine Learning data augmentation techniques, we will be capable of artificially expanding (i.e., more than 100×) the size of such a dataset. At the same time, we will be conserving the authentic data properties and patterns that make original on-orbit data so valuable for software assurance testing of NASA software systems such as cFS, NOS3, WFIRST, DeepSpace Gateway, SLS, etc. Our plan is to release the original and augmented datasets to the entire international Small Satellite community that is necessary for the advancement of research activities related to SmallSat flight software development, testing and maturation.
For example, suppose that the dataset discovers a previously unknown trend of temperature spikes that correlate to radio transmits when in direct sunlight. This information is expected to improve the spacecraft hardware modeling with more accurate flight software edge cases. Thus, this knowledge will help mission designers to advance current flight software capabilities is different ways, one of which will be that very short transmits are only performed when in sunlight, avoiding overheating components and thus increasing the SmallSat operational lifetime.
The STF-1 SmallSat mission, which is currently operating on-orbit, resulted in the development of a software simulation framework named the NASA Operational Simulator for Small Satellites (NOS3). The goal of NOS3 is to enhance small satellite software development, testing, and smallsat operator training by improving flight software assurance. With NOS3, the flight software executes as if it were operating in space. NOS3 provides the flight software with representative real-world simulated data inputs that it would expect during nominal on-orbit operations. For a full review, see the NASA IV&V ITC Team (2018) publication on the Journal of Small Satellites. STF-1’s flight software is the GSFC’s core Flight Software (cFS) and is integrated into NOS3 by default. The NASA-developed core Flight System (cFS; Wilmot 2005) is an open-source solution for spacecraft flight software, with flight heritage on numerous large and small NASA missions, including the Global Precipitation Measurement (GPM) and the Lunar Atmosphere Dust and Environment Explorer (LADEE). cFS is currently being utilized on WFIRST, DeepSpace Gateway (i.e., the Boots on the Moon, 2024 effort), Orion backup computer, and multiple SmallSat missions.
To date, the STF-1 SmallSat has downlinked 3.4 GB of data. In addition to science experiment data, this data is chiefly composed of spacecraft health and telemetry data, including, but not limited to, temperature, gyroscope rates, battery charge levels, on-board storage metrics, and information related to cFS housekeeping. In normal everyday computing, 3.4 GB is minimal; however, being able to downlink 3.4 GB (and counting) of data from an on-orbit spacecraft is a very challenging task, particularly when in-view passes are on average 7 minutes and only once or twice a day. Also, the engineering health and telemetry data is closely guarded by individual NASA missions and not normally distributed to the public.
Consequently, NASA on-orbit data is often very limited in terms of availability, quantity, and quality. This is a limitation despite the fact that such data are of paramount importance for developing and assuring safety-critical autonomous and resilient NASA software systems, as well as for applying reliable data analytics necessary to assess system risks. In this project we propose to take advantage of state of the art machine Leaning data augmentation approaches that will allow us to bridge the gap for the availability of critical data at an advance scale (at least 100× more than the original dataset size) opening-up to novel software assurance improvements.