Introduction

Implementation for AI-driven method for exhaustive hazard scenario generation in chemical process systems

install

conda env create -f environment.yml
conda activate graphdecoder

Dataset

This dataset was generated based on rule-based logic. The input and output are built independently using different scripts. You can access the processed dataset and pre-trained model weights via the following Google Drive link: https://drive.google.com/drive/folders/1-l79eFNjTlllNm4z0sQ9jAv7sgOAJ9_p?usp=sharing

Input

• data_forming.ipynb: Generates node features for Process Flow Diagrams (PFDs).
  These features serve as the structural input for downstream tasks.

Output

• Hazard scenarios are created using yEd Graph Editor and saved in total.graphml.
• json_generate.ipynb: Reads total.graphml and converts the scenarios into structured JSON format, saved as graph_relations_labeled_structured.json.

Integration

• data_preprocess.ipynb: Combines the input features and output scenarios into a unified binary file graph_data.bin.
  This file serves as the dataset for model training, validation, and testing.

Usage

Running with VS Code (Using launch.json)

This project includes a pre-configured .vscode/launch.json file that defines several run/debug configurations for Visual Studio Code.

To use these:

1. Click the Run and Debug icon on the sidebar (or press Ctrl+Shift+D).
2. In the dropdown at the top of the panel, choose one of the available configurations, such as:
   • Run Generate_case_study Module
   • Run Main (Train)
3. Click the ▶ Run button or press F5.

Acknowledgments

The authors thankfully acknowledge the financial support provided by the Mary Kay O’Connor Process Safety Centre at Texas A&M University.