❄️ FrostNet – Deep Learning for Cryogenic Temperature Anomaly Detection

FrostNet is a time-series anomaly detection project focused on monitoring and analyzing cryogenic temperatures—typically temperatures below -150°C. This system is particularly relevant for applications in scientific research, industrial storage, space engineering, and superconductor technologies, where maintaining temperature stability is crucial.

Using a Long Short-Term Memory (LSTM) deep learning model, FrostNet is capable of learning patterns from historical data and identifying subtle anomalies that could signal faults, system failures, or unexpected behavior in a cryogenic environment.

---

🧠 Project Goals

• Generate synthetic cryogenic temperature data that simulates real-world sensor behavior.
• Preprocess and normalize the data for deep learning workflows.
• Train an LSTM-based neural network for time-series modeling and anomaly detection.
• Evaluate model performance using precision, recall, MAE, and RMSE.
• Visualize temperature fluctuations and detected anomalies to gain actionable insights.

---

📁 Folder & File Structure

FrostNet/
├── data/                               # Contains raw and preprocessed datasets
│   └── cryogenic\_synthetic.csv
├── outputs/                            # Trained LSTM model, evaluation logs, saved metrics
│   └── lstm\_model.pth
├── visualization/                      # Graphs, anomaly plots, prediction vs actual plots
│   └── anomaly\_visualization.png
├── 01\_data\_generation\_and\_loading.ipynb
├── 02\_data\_preprocessing\_and\_cleaning.ipynb
├── 03\_model\_building\_and\_training.ipynb
├── 04\_model\_evaluation\_and\_anomaly\_detection.ipynb
├── 05\_visualization\_and\_insights.ipynb
└── README.md

---

⚙️ Environment Setup

• Python version: 3.8
• Libraries used:
  • numpy
  • pandas
  • matplotlib
  • seaborn
  • scikit-learn
  • torch (PyTorch)
  • datetime
  • os, random (for synthetic data)

Note: You can install dependencies using the following command:

pip install -r requirements.txt

---

📓 Notebook Descriptions

1. 📄 01_data_generation_and_loading.ipynb

• Generates synthetic cryogenic temperature data using a random walk model with occasional spikes to simulate anomalies.
• Timestamps are generated at 1-minute intervals.
• Anomaly labels (1 for anomaly, 0 for normal) are inserted at random points.
• Output is saved as data/cryogenic_synthetic.csv.

2. 🧼 02_data_preprocessing_and_cleaning.ipynb

• Loads the generated data and inspects for missing values.
• Applies normalization to temperature readings using MinMaxScaler.
• Handles any data gaps or duplicates.
• Produces histograms and correlation plots to understand data distribution.

3. 🏗️ 03_model_building_and_training.ipynb

• Builds an LSTM model using PyTorch for sequence modeling.
• The model accepts a sliding window of temperature readings and learns to reconstruct sequences.
• Trained for 30 epochs using Mean Squared Error (MSE) loss.
• Saves the trained model to outputs/lstm_model.pth.

4. 📊 04_model_evaluation_and_anomaly_detection.ipynb

• Uses the trained LSTM model to generate predicted temperature sequences.

• Computes reconstruction error for each time point.

• Anomalies are detected using a fixed threshold on the reconstruction error.

• Evaluation Metrics:

  • Precision: 0.0120
  • Recall: 0.0600
  • MAE (Mean Absolute Error): 0.0587
  • RMSE (Root Mean Square Error): 0.2423

  

5. 🖼️ 05_visualization_and_insights.ipynb

• Plots original vs predicted temperature values.
• Highlights detected anomalies on time-series plots.
• Provides insights into model performance and failure cases.
• Charts are saved to the visualization/ folder for reporting and presentations.

---

📈 Key Features

• ✅ Fully self-contained and modular Jupyter notebooks
• ✅ End-to-end anomaly detection pipeline with synthetic data
• ✅ Deep learning using LSTM networks
• ✅ Supports visualization of anomalies for interpretability
• ✅ Logs and model outputs saved for reproducibility

---

📂 Dataset Details

File: data/cryogenic_synthetic.csv

Column Name	Description
Timestamp	1-minute interval datetime values
Temperature (°C)	Simulated cryogenic sensor readings
Anomaly	Binary label (1 = anomaly, 0 = normal condition)

The data mimics the behavior of real cryogenic systems with periodic anomalies and realistic noise patterns.

---

🚀 How to Run

1. Clone the repository:

git clone https://github.com/yourusername/FrostNet.git
cd FrostNet

2. Open the notebooks in Jupyter Notebook or JupyterLab.

3. Run notebooks in the following order:

   • 01_data_generation_and_loading.ipynb
   • 02_data_preprocessing_and_cleaning.ipynb
   • 03_model_building_and_training.ipynb
   • 04_model_evaluation_and_anomaly_detection.ipynb
   • 05_visualization_and_insights.ipynb

4. View graphs in the visualization/ folder and model metrics in outputs/.

---

💡 Future Enhancements

• Incorporate real-world cryogenic datasets.
• Use Transformer-based time-series models for comparison.
• Integrate with a dashboard (e.g., Streamlit) for live anomaly monitoring.
• Enable online learning for adapting to incoming data streams.

---

📜 License

This project is licensed under the MIT License. See the LICENSE file for more details.

---

👨‍💻 Author

• Name: [Isha Das]
• Email: [ishadas2006@gmail.com]

---