This is the repository for our publication Using Modular Neural Networks for Anomaly Detection in Cyber-Physical Systems.
We use Modular Neural Networks to model the inner dependencies of Cyber-Physical System (CPS) subsystems. Thereby, we can achieve a more robust detection of anomalies in CPS and a better allocation of their root-causes. For further information, we recommend you to read our [publication(#citation)].
We recommend Anaconda to install all requirements for our repository.
The requirements are saved in the venv.yml file.
For a quick installation run: mamba env create -f venv.yml
As empirical validation dataset, we used the robot-anomaly dataset of Grabaek et al. 2023.
You can access and download the dataset here.
Once you downloaded the dataset, save it in the ./data directory.
For replicating the results from our paper, run the main.py script from the ./code directory.
The script will run the reproducible hyperparameter search, as well as the subsequent replication studies, and evaluation studies.
By running the ./exp/exp_setup/evaluation.ipynb, you can calculate the metrics from the paper.
You can run your own studies by uncommenting the suitable codeblock in the main.py script.
When using this work, please cite:
@INPROCEEDINGS{Ehrhardt2024
author={Ehrhardt, Jonas and Overlöper, Phillip and Vranjes, Daniel and Steude, Henrik and Diedrich, Alexander and Niggemann, Oliver},
booktitle={2024 IEEE 29th International Conference on Emerging Technologies and Factory Automation (ETFA)},
title={Using Modular Neural Networks for Anomaly Detection in Cyber-Physical Systems},
year={2024},
pages={01-07},
keywords={Correlation;Knowledge based systems;Machine learning;Predictive models;Cyber-physical systems;Industrial robots;Data models;Anomaly detection;Manufacturing automation;Multi-layer neural network;Anomaly Detection;Modular Neural Networks;Cyber- Physical System;Industrial Robot},
doi={10.1109/ETFA61755.2024.10711115}}
Licensed under MIT license
