🔍 OTDR Fiber Fault Detection & Localization Dashboard

Interactive Streamlit dashboard for detecting, localizing, and characterizing fiber faults from OTDR traces using pre-trained ML models. Designed with a modern dark theme, professional analysis views, and one-click exports.

✨ Highlights

• Auto-loads all required models and scaler from a local models/ folder
• Multiple data input modes: CSV upload, single-sample form, or auto-generated sample
• Binary fault detection + detailed analysis (class, position, reflectance, loss)
• Interactive Plotly charts, including predicted fault highlight and optional 3D view
• Research-style report with actionable insights and recommended steps
• Quick CSV export of predictions and results

📦 Repository Contents

• otdr_dashboardv2.py – Streamlit app
• models/ – Place your pre-trained models here (see below)
• README.md – You are here

🧠 Required Models (Auto-loaded)

Place the following files in a local models/ directory (relative to the project root):

Purpose	Filename	Type
Feature Scaler	scaler.pkl	Pickle (sklearn StandardScaler)
Binary Classification	binary_model.h5	Keras/TensorFlow
Fault Class Detection	multiclass_model.h5	Keras/TensorFlow
Position Localization	position_model.pkl	Pickle
Reflectance Analysis	reflectance_model.pkl	Pickle
Loss Analysis	loss_model.pkl	Pickle

If any file is missing, the app will show a clear sidebar error (it won’t crash). You can still use parts of the pipeline that have their models available.

✅ Prerequisites

• Python 3.9–3.11
• Recommended: TensorFlow 2.15+ (for .h5 model compatibility)

Install the required packages:

pip install streamlit pandas numpy plotly joblib scikit-learn tensorflow

If you prefer a virtual environment:

python -m venv .venv
. .venv/bin/activate   # on macOS/Linux
.venv\Scripts\activate # on Windows
pip install streamlit pandas numpy plotly joblib scikit-learn tensorflow

🚀 Run the App

Windows PowerShell:

cd "C:\Users\abhay\OneDrive\Desktop\IBMfinal"
py -m streamlit run otdr_dashboardv2.py

Or generic Python:

streamlit run otdr_dashboardv2.py

📥 Data Format (CSV)

Required columns:

• SNR
• P1 … P30 (30 OTDR trace points)

Optional columns (if available in your dataset):

• Class (0–7)
• Position (0.00–0.30 normalized)
• Reflectance, Loss (normalized)

Example (header):

SNR,P1,P2,...,P30,Class,Position,Reflectance,Loss

🖥️ Using the Dashboard

1. Models are auto-loaded from models/ at startup.
2. Choose an input method:
   • Upload CSV dataset
   • Enter a single sample via form
   • Generate a sample OTDR trace
3. Run Binary Fault Detection.
4. If a fault is detected, run Detailed Analysis to get:
   • Fault class, predicted position, reflectance, loss
   • Interactive trace highlighting the predicted fault location
   • Summary cards and research-style report
5. Export results as CSV via quick-download or the Export tab.

📊 Visualizations

• OTDR Trace with Predicted Fault Position (interactive line+markers)
• Residual/Error Distribution (histogram)
• Model Performance (accuracy, precision, recall, F1) as metrics + bar chart
• Optional 3D OTDR trace (position vs value vs time index)

Prediction-to-position mapping:

• Model output for position is normalized (0.00–0.30 in 0.01 steps)
• Mapped as: pX where X = int(prediction * 100)
  • Example: 0.06 → p6

🧪 Notes on Models & Compatibility

• Keras .h5 models are loaded with compile=False; ensure TensorFlow 2.15+ if possible.
• If you trained with normalization, ensure your scaler.pkl is the same one used during training.
• If loading fails, the app will display model-specific error messages in the sidebar.

🧭 Project Structure

IBMfinal/
├─ otdr_dashboardv2.py
├─ models/
│  ├─ scaler.pkl
│  ├─ binary_model.h5
│  ├─ multiclass_model.h5
│  ├─ position_model.pkl
│  ├─ reflectance_model.pkl
│  └─ loss_model.pkl
└─ README.md

🆘 Troubleshooting

• Error: “Required model file '…' not found”
  • Ensure the file exists in models/ with the exact filename.
• TensorFlow/Keras load errors
  • Re-save the model with your current TF/Keras version.
  • Consider exporting in SavedModel format if you have access to training code.
• Poor predictions
  • Verify dataset column names and scales (SNR, P1–P30).
  • Ensure the same scaler used in training is provided (scaler.pkl).

🤝 Contributing

Issues and PRs are welcome! If you add support for new model formats or better visualizations, please include clear instructions and sample files.

📄 License

MIT License (or your preferred license). Update this section accordingly.

---

Built with Streamlit, Plotly, TensorFlow, and scikit-learn. Designed for practical OTDR fault analysis with clean, modern UX.