nhorro/spacecraft-anomaly-detection

Experiments with algorithms to identify anomalies in multivariate series, targeted for spacecraft applications.

Anomaly detection for spacecraft/satellite applications

This project contains experiments with algorithms to identify anomalies in multivariate series with a focus on spacecraft applications.

Each experiment is a jupyter notebook that generates one or more models to process multivariate series.

Dependencies are contained in a conda environemnt.

conda env create -f anomalies-env.yml
conda activate anomalies-env.yml

Project organization:

./
    data                 Contains data, or URIs to download.
    models               Trained models are exported here.
    notebooks            Experiments.
    README.md            This file.
    anomalies-env.yml    Conda environment for Python3/Jupyter.    

References and useful links

General anomaly detection:

• Deep Learning for Anomaly Detection: A Review
• ADRepository: Real-world anomaly detection datasets
• Deep distance-based outlier detection (KDD18)
• Unsupervised Representation Learning by Predicting Random Distances
• DevNet: An End-to-end Anomaly Score Learning Network
• OCAN: One-Class Adversarial Nets for Fraud Detection
• Anomaly Detection for Multivariate Time Series through Modeling Temporal Dependence of Stochastic Variables
• Beginning Anomaly Detection Using Python-Based Deep Learning by Sridhar Alla and Suman Adari (Apress, 2019)

Aerospace spacecraft/satellite telemetry specific:

• Polaris RECOMMENDED
• Satellite-Telemetry-Anomaly-Detection
• ESA ACF - Anomaly Detection in Satellite Telemetry
• satellite-telemetry-project
• LASP Public lecture: Machine Learning at LASP

Embedded bus (CAN, etc.) oriented:

• CANBus Traffic Anomaly Detection with LSTM and Autoencoders

Datasets

• Open Source Global Network of Satellite Database