ISRO's PSLV rocket suffered two consecutive failures in May 2025 and January 2026, both caused by anomalies in the third stage (PS3) combustion chamber pressure drop. Combined mission loss exceeded ₹1000 Crore. The anomaly was visible in telemetry data but was detected too late to take action.
Orbit Shield ML is a machine learning system that monitors rocket telemetry sensor readings in real-time and detects anomalies before they cause mission failure.
- Trains on NASA CMAPSS turbofan engine dataset (rocket-equivalent sensor data)
- Uses an ensemble of two models:
- Isolation Forest — unsupervised outlier detection
- Autoencoder Neural Network — learns normal patterns, flags deviations
- Alerts when both models agree something is wrong
- Live dashboard shows anomaly scores across the full mission timeline
| Model | Accuracy | Anomaly Recall | Precision |
|---|---|---|---|
| Isolation Forest | 94% | 89% | 73% |
| Autoencoder | 91% | 82% | 58% |
| Ensemble (final) | 95% | 79% | 82% |
| Category | Technology |
|---|---|
| Language | Python 3.11 |
| Deep Learning | PyTorch |
| Machine Learning | Scikit-learn |
| Explainability | SHAP |
| Dashboard | Streamlit |
| Data Processing | Pandas, NumPy |
| Visualisation | Matplotlib, Seaborn |
| Version Control | Git, GitHub |
git clone https://github.com/Lohini06/orbit-shield-ml.git
cd orbit-shield-ml
pip install -r requirements.txt
streamlit run src/dashboard.pyorbit-shield-ml/ ├── data/ # NASA CMAPSS dataset ├── notebooks/ # Data exploration and model training ├── src/ # Dashboard code ├── models/ # Saved trained models └── outputs/ # Charts and results
This system directly addresses the gap identified by space analysts after the PSLV-C62 failure — the need for intelligent early warning systems on solid motor stages that cannot be shut down mid-flight.