PRAMANA

AI Digital Twin for UAV Engine Health

A real-time digital twin and fault-diagnosis system for aero piston engines in MALE UAVs, built for DRDO's Smart India Hackathon 2026 problem statement.

> What I built

• A physics-based engine model streams live telemetry; an ML layer (autoencoder anomaly gate + fault classifier) catches faults before a threshold alarm would and localises them to a cylinder.
• Measured on held-out test installations: 94.5% recall on faulty runs and 98.6% top-1 diagnosis accuracy, with 15/15 fault types detected live.
• Fully offline AI assistant (Ollama, Whisper, Kokoro) explains the twin's state from real telemetry and never makes the land/continue decision itself.
• Owned the frontend end to end, including the live 3D engine visualization that judges called out as a highlight of the demo.
• Won the internal selection hackathon; the team now represents NMIMS Hyderabad at the main Smart India Hackathon 2026.

> Tech stack

• Python
• Physics Model (MVEM)
• Autoencoder + Classifier
• WebSocket
• React
• 3D Viz
• Ollama