$\text{Anomaly Detection In Satellites Telemetry Data using Probabilistic Graphical Models}$

This repository implements many Probabilistic Graphical Models and Deep Learning Models, including DBNs, HMMs, GMMs, GNNs, VAEs, and iForest, for telemetry anomaly detection in spacecraft systems on the ESA-Mission1 Dataset.

$\text{●•Authors}$

$\text{Baimam Boukar Jean Jacques}$	$\text{Kipngeno Koech}$
Carnegie Mellon University Africa	Carnegie Mellon University Africa
bbaimamb@andrew.cmu.edu	bkoech@andrew.cmu.edu

●• $\text{Dataset Description}$

The analysis used telemetry data from the European Space Agency's ESA-Mission1. It has over 14 million records collected across several years. This continuous multivariate time series includes 87 mission-critical channels, annotated for anomalies and rare events through iterative manual and algorithmic refinement of flight control reports. The dataset targets two event categories

●• Anomalies $\to$ Unexpected behaviors or system failures
●• Nominal Events $\to$ unusual but expected operational patterns.

The data is divided into a training set spanning 14 years of operations and a test set covering a 6-month unpublished segment.

The dataset has 87 telemetry channels, 58 target channels monitored for anomalies, 18 auxiliary environmental variables, and 11 telecommand channels that are binary control commands, prefixed with telecommand_

●• $\text{Methodology}$

●• $\text{Reproduction Steps}$

●• $\text{Cite This Paper}$

@software{bbaimamb_bkoech_2025,
author = {Baimam Boukar Jean Jacques and Kipngeno Koech},
month = apr,
title = {{Causal Structure Analysis for Telemetry Anomaly Detection in Spacecraft Systems}},
url = {https://github.com/baimamboukar/causual-structure-discovery-spacecraft-telemetry},
version = {1.0},
year = {2025}
}