Satellite Maneuver Detection Web Service
The Satellite Maneuver Detection web service continuously analyzes Two-Line Element (TLE) data from space-track.org using machine learning algorithms to detect satellite maneuvers with high sensitivity — capable of identifying maneuvers as small as a few millimeters per second in velocity change.
For each space object, the service compares propagated and published orbital states and presents the result as a set of charts showing differences in velocity vectors and Keplerian elements: semimajor axis, eccentricity, inclination, RAAN, argument of perigee, mean anomaly. Each chart benchmarks the object against an ML-selected matching subset of a representative population of rocket bodies and debris and includes a maneuver-detection threshold calculated as the RMSE of the matching subset's differences multiplied by a user-defined sensitivity factor. Comparing orbital-element differences against statistical thresholds helps classify detected velocity changes.
Sample chart: Differences between Propagated and Published States of SENTINEL-3A (NORAD_CAT_ID: 41335) — Velocity Difference. The two peaks correspond to maneuvers with ΔV of approximately 9.43 mm/s (2026-07-22 08:25:37 UTC) and 6.05 mm/s (2026-08-06 07:00:37 UTC), per NASA/ILRS maneuver data. The maneuver detection threshold (red line, with LOWESS smoothing applied), calculated as the prediction (green line) plus three times RMSE, effectively separates these two maneuver events from data points impacted by regular propagation disturbances.
Further analysis of the differences in Keplerian elements (such as Semimajor Axis, Eccentricity, Inclination, RAAN, Argument of Perigee, and Mean Anomaly) for these two events demonstrates that the maneuvers were aimed at correcting the satellite's altitude. Other elements remained within statistical error margins, providing additional confirmation of maneuver activity.
The service also provides an interactive "Propagated vs. Published States of Space Objects" summary report showing maximum and average differences in orbital parameters for each object over a specified reporting period, with results sortable by these values. Sorting by maximum velocity difference identifies the strongest orbital-motion anomalies in general, while sorting by maximum inclination or RAAN change highlights the strongest candidates for orbital-plane maneuvers, enabling fast and efficient identification of potentially maneuvering objects.