Rapid Discovery of Evasive Satellite Behaviors
Award last edited on: 7/7/2023

Sponsored Program
Awarding Agency
Total Award Amount
Award Phase
Solicitation Topic Code
Principal Investigator
Christopher Bowman

Company Information

Data Fusion & Neural Networks LLC (AKA: Data Fusion & Neural Networks~DF&NN))

17150 West 95th Place
Arvada, CO 80007
   (720) 872-2145

Research Institution

University of Texas - Austin

Phase I

Contract Number: FA8750-18-C-0120
Start Date: 4/16/2018    Completed: 4/16/2019
Phase I year
Phase I Amount
The DF&NN team has significant experience in delivering Space Domain Awareness tools. These tools will be applied to 2 years of GEO ephemeris data of active satellites at DF&NN. The RDESB prototype will reduce the risk in rapidly discovering the behavioral patterns of potentially evasive and/or ambiguous active resident space objects. RDESB will detect non-Keplerian behavior in ephemeris data for 2015 and then learn these normal behaviors with ANOM. ANOM application on the 2016 ephemeris will detect abnormal maneuvers, cross-tags, etc. which are tracked within Abnormality Detection Classification Viewer. The Smoking Gun tool will be extended to find temporal relationship correlations amongst these abnormal events. The ClassCat GUI will be extended to enable the user to create the Ontology-based Knowledge Graph for classification and relationship ontology labels of the abnormal signature detections to include confidences of alternative labels. The marked normal maneuvers (e.g., E/W, N/S, etc.) and relationship behaviors will be learned by sparse categorization NNs which automatically select the significant variables necessary. These sparse NNs flag 2016 abnormal and missed maneuvers and relationship precursors to create sensor tasking for increased sensor updates during times of predicted maneuvers and after these abnormality detections to achieve and maintain custody of UCTs.

Phase II

Contract Number: FA8750-19-C-1018
Start Date: 9/6/2019    Completed: 9/6/2021
Phase II year
Phase II Amount
The problem addressed in this effort is to automatically learn historical ephemeris space catalog time, position, and velocity entity track update error uncertainties (i.e., without track error covariances) and to automatically (e.g., without expert event labeling) produce: - unmodeled non-gravitational space catalog update flags - abnormal unmodeled catalog update flags with abnormality scores - categorization of similar normal unmodeled from historical behavior in new catalog unmodeled update behavior - discovered temporal relationships in historical unmodeled satellite behaviors (e.g., station keeping maneuver temporal gaps with temporal uncertainties) - missed maneuver flags with confidences and other normal unmodeled behavior relationships predicted from discovered historical temporal relationships We will improve the RDESB prototype Ephemeris Abnormal Catalog Update (eACU) which is already detecting abnormal APES ephemeris track updates on a Thought Cloud virtual machine in Bldg. 423 at Kirtland AFB. The second on-line data source is the Nebula Data Lake at the National Space Defense Center (NSDC). The Phase 2 RDESB goal is to provide threat indications and warnings left of significant unknown events in the NSDC following the same operationalization path DF&NN has for ACU implementation under an AFRL installation program.