Triet Tran, Cornerstone Consulting & Services, LLC; Anthony N. Dills, L3Harris Corp
Keywords: AI/ML models, data association, maneuver classification, maneuver detection
Abstract:
The ability to operationally perform near real-time automatic detection and classification of resident space object (RSO) maneuvers is a highly desirable. The potential benefits include anomaly resolution and change detection that may reduce the number of Uncorrelated Tracks (UCTs) leading to improved space object catalog accuracy.
A challenging aspect of Space Domain Awareness (SDA) is detecting an RSO maneuver based solely on observations and current knowledge of the orbital states. Modern techniques have been developed by SDA researchers using statistical filtering techniques with various degrees of success. Artificial Intelligence/Machine Learning (AI/ML) techniques have seen significant growth in recent years and pose viable approaches within the space domains solution space to augment, supplement, and potentially replace traditional methods with varying degrees of performance.
The contribution of the present work is to demonstrate the feasibility of using AI/ML to perform data association and maneuver classification of the RSOs in the geosynchronous (GEO) orbit regime. Pattern of Life (POL) of the maneuvering RSO is the key to the success of the AI/ML model to recognize whether a set of optical measurements can be associated with an RSO candidate and whether a maneuver type can be classified for that RSO based on the same optical measurement set. This work presents an autoencoder (AE) to perform data association using historical maneuver data for a particular RSO as observed by a specific ground-based optical sensor. The same data is also applicable to identifying and training an optimized classification model by applying an open-source genetic programming technique known as Tree-Based Pipeline Optimization Tool (TPOT). The presented analysis used precision ephemerides publicly available from the National Oceanic and Atmospheric Administration (NOAA) and Federal Aviation Administration (FAA) for GEO earth orbiting satellites. The data used for assessing AI/ML candidate models are the simulated ground-based optical measurements on RSOs whose POL of maneuvers is derivable from publicly available ephemerides.
Results of this work point to an alternative approach of data association and maneuver classification. The results also demonstrate an approach for real-time association that could be implemented at an observing ground site to help reduce UCTs in collaboration with existing astrodynamics filtering techniques. The results also identify the most prominent features of the data that reduces the dimensionality within classification model. Finally, an approach to automatically label maneuvers in each data set of observations was developed and could provide a utility to the greater community.
The data association and the classification results each achieved better than 90% prediction accuracy based on validation during the training phase of the ML models. Future work will include model tuning, examining data diversity, assessing model prediction of real observation data, exploring the use of model ensemble to increase robustness, and addressing applicability to the more challenging aspect of data association in the area of Closely-Space Objects (CSOs).
Date of Conference: September 14-17, 2021
Track: Machine Learning for SSA Applications