A Deep Learning-based Approach for Two-Phase Flow Pattern Classification

This repository contains the implementations used in the preparation of the article "A Deep Learning-based Approach for Two-Phase Flow Pattern Classification Using Void Fraction Time Series Analysis." The repository includes scripts for preprocessing, model training, evaluation, and visualization of results.

---

📁 Content

Repository Structure

.
├── README.md                     # Documentation file (this file)
├── data                          # Preprocessed datasets and features/labels
│   ├── gmm                       # GMM processed data
│   ├── window_4000_overlap_1000_hzdr_norm.npy
│   └── window_4000_overlap_1000_tud_norm.npy
├── metrics                       # Evaluation metrics (e.g., confusion matrices)
│   ├── cross_dataset
│   └── single_dataset
├── models                        # Saved models and artifacts
│   ├── cross_dataset
│   └── single_dataset
├── notebooks                     # Jupyter Notebooks for experiments
│   ├── SVM-cross_dataset.ipynb
│   ├── SVM-single_dataset.ipynb
│   ├── results.ipynb
│   ├── tsai_models-cross_dataset.ipynb
│   └── tsai_models-single_dataset.ipynb
└── requirements.txt              # Python dependencies

1. Data

• Description: Contains preprocessed data used for training and evaluation.
• Structure:
  • gmm: Processed GMM data.
  • window_*_hzdr_norm.npy and window_*_tud_norm.npy: Preprocessed feature and label data.

2. Metrics

• Description: Contains evaluation results, including JSON files for metrics and confusion matrices.
• Structure:
  • cross_dataset: Metrics from cross-dataset evaluations.
  • single_dataset: Metrics from single-dataset evaluations.

3. Models

• Description: Stores saved models and their associated artifacts.
• Structure:
  • cross_dataset: Models trained for cross-dataset evaluations.
  • single_dataset: Models trained for single-dataset evaluations.

4. Notebooks

• Description: Contains Jupyter Notebooks for experiments and results visualization.
• Files:
  • SVM-cross_dataset.ipynb: SVM cross-dataset evaluation.
  • SVM-single_dataset.ipynb: SVM single-dataset evaluation.
  • results.ipynb: Summary and visualization of results.
  • tsai_models-cross_dataset.ipynb: Deep learning cross-dataset evaluation.
  • tsai_models-single_dataset.ipynb: Deep learning single-dataset evaluation.

Key Information

• HZDR Data Shape: (1474, 4003)
• TUD Data Shape: (15141, 4003)
  • The first dimension represents the number of samples.
  • The second dimension consists of:
    • 4000 time-series samples.
    • 3 elements for the label, formatted using one-hot encoding.

Due to restrictions, the HZDR and TUD datasets are referenced in the code but are not included in the repository.

---

⚙️ Setup

Tested on

• Python Version: 3.12.8
• Platform: WSL2 - Ubuntu 24.04

Installation Steps

1. Clone the repository:

   git clone https://github.com/ambrosioj/two-phase-time-series-deep-learning.git
   cd two-phase-time-series-deep-learning

2. Install the required Python packages:

   pip install -r requirements.txt

3. Set up a Jupyter Notebook tool to execute the provided .ipynb files.

---

🔐 Ignored Files

The .gitignore file is configured to ignore:

• All .npy files except those matching *_cm.npy (e.g., confusion matrices).
• Temporary and environment-related files.
• IDE and OS-generated files (e.g., .vscode, .DS_Store).

---

🔄 Acknowledgments

This repository serves as a comprehensive resource for exploring deep learning methods applied to two-phase flow pattern classification.

For any questions or clarifications, please contact the repository maintainer.

Jefferson dos Santos Ambrosio - [email protected]