This repository contains the constructed datasets as well as trained models for the paper: "I came, I saw, I hacked: Automated Generation of Process-independent Attacks for Industrial Control Systems"
Here, we demonstrate fingerprinting of various processes and sectors of an Industrial Control System (ICS) to gain automatically contruct attack vector that has intelligence to perform meaningful damage
Part 1: ICS Sector fingerprinting using HMI images
Collected Images
In the current version of the dataset, we classify an HMI image to belong to one of the three sectors:
- Chemical
- Energy
- Water and wastewater
The constructed dataset (each containing the sectors):
Training images: /HMI Images/Images/Train.zip
Test images: /HMI Images/Images/Test.zip
Text extracted from images
The raw data extracted from the images using OCR: /HMI Images/HMI_text/TextResults_(Test/Train).txt
The translated data from the images: /HMI Images/HMI_text/TranslatedText_(Test/Train).txt
The cleaned data from the images: /HMI Images/HMI_text/TransTextClean_(Test/Train).txt
Models
Final trained model using VGG16 architecture to be applied on images: /HMI Images/Models/NDSS_VGG16_078.7z
Final trained model using Multinomial Naive Bayes to be applied on cleaned text: /HMI Images/Models/MNB.sav
Training Scripts
All the training scripts used to train the architectures discussed in the paper /HMI Images/Training Scripts/
Sample Evaluation script
A sample script to utilize the models to predict accuracy: Evaluation_final_script.py
Please note, there might be a difference in test accuracy of ~2%. We encourage advancement of other architectures and models to train the constructed dataset.
E. Sarkar, H. Benkraouda, M. Maniatakos
"I came, I saw, I hacked: Automated Generation of Process-independent Attacks for Industrial Control Systems"
ACM Asia Conference on Computer and Communications Security (AsiaCCS 2020)