Jonathan Kadan, SSC/SZGA; Amit Bala, Virginia Tech; Kevin Schroeder, Virginia Polytechnic Institute and State University; Jonathan Black, Virginia Polytechnic Institute and State University
Keywords: SSA, SDA, Maneuver Detection, Autonomy
Abstract:
The number of Resident Space Objects (RSO) is in the tens of thousands and this problem is being exacerbated by the rapidly accelerating rate of satellite launches. Coinciding with the launch of new RSO is the increasing maneuver capability of satellites. These more agile RSO can perform many types of maneuvers including station keeping and Rendezvous and Proximity Operations (RPO) at small timescales and at close distances to their targets. The increasing capabilities of RSO further emphasizes the need for efficient, autonomous sensor tasking and maneuver detection.
In order to maintain custody of this growing and maneuverable catalog, Space Domain Awareness (SDA) is trending away from human-in-the-loop satellite observation tasking and towards human-on-the-loop autonomous sensor tasking. These human-on-the-loop systems will need to identify, observe, and re-schedule observations of maneuvered satellites as quickly as possible to prevent loss of custody. For an autonomous system to react to changes in the SDA environment, accurate, real-time maneuver detection algorithms must be implemented.
This work focuses on the implementation of various maneuver detection techniques in an autonomous SDA simulation and evaluation of these algorithms for satellite maneuver detection. The standard practice in satellite maneuver detection is to evaluate the standard Normalized Innovations Squared (NIS) metric of an incoming observation. However, this is not the only option. Other potential choices for maneuver detection consist of variations of the standard NIS metric: sliding NIS and fading memory NIS.
From here, different sensor tasking solutions were implemented to determine a computationally efficient method of reward function augmentation that emphasized revisit of maneuvered RSO. To generate the data necessary for this analysis, scenarios with RSO in near Geosynchronous Earth Orbits (GEO) and maneuver events will be generated. The maneuvers modelled were representative of East/West and North/South station keeping maneuvers. Ground Based Electro-Optical (EO) and radar sensors, with their necessary constraints, were modeled. For example, the optical sensors modeled will implement necessary lighting constraints to determine RSO visibility.
An Unscented Kalman Filter (UKF) was used to consolidate observation data and produce RSO estimate ephemerides. A special perturbations dynamics model was used for RSO propagation. Sensor specifications were approximated with open-source documentation of the current Space Surveillance Network (SSN).
Comparison between control group scenarios without maneuvers and those with maneuvers were used to show how the augmented reward functions direct sensor tasking toward RSO w/ detected maneuvers while still maintaining catalog coverage. Maneuver detection failed at high process noise values and false positive maneuver detections were identified in low process noise simulations; the UKF was tuned accordingly.
This work was used to evaluate which maneuver detections algorithms perform best for GEO orbits and determined proper process noise levels for the UKF used to predicted measurement residuals for maneuver detections.
For the radar sensing scenarios, both Lambert Initial Orbit Determination (IOD) and Generalized Pseudo-Bayesian Estimator of Order 1 (GPB1) Multiple Model Adaptive Estimation (MMAE) were implemented to reduced RSO residuals to within nominal levels without having to severely modify the tasking solution.
Date of Conference: September 19-22, 2023
Track: Space Domain Awareness