<aside> 🦅

Predictive, physics-calibrated, edge-deployable digital twin preventing in-flight piston engine failures with real-time diagnostics, RUL forecasting, and TreeSHAP explainability.

🌐 Live Web App: sih26054-male-uav-digital-twin.onrender.com

📦 GitHub: Aircraft-Engine-Flight-Deck

</aside>

🚨 1. The Aviation Crisis

"A warning light tells an operator what has already broken; a digital twin tells them what is about to break, why it is failing, and how much safe flight time remains."

In general aviation and long-endurance unmanned aviation, in-flight propulsion failure is the single largest cause of catastrophic airframe losses and forced crash landings.

<aside> 📉

65%+ of forced off-field landings stem directly from uncontained piston engine degradation — thermal runaway, cooling loss, oil starvation, or injector clogging.

Existing cockpits rely on static threshold "idiot-lights" that only trigger after catastrophic failure has begun, leaving operators a 15–30 second panic window.

</aside>

⚖️ The Paradigm Shift

Feature ❌ Traditional Warning Lights ✅ AeroTwin Digital Twin
Monitoring Reactive threshold exceedance Continuous physics-informed tracking
Detection Window Instant of catastrophic failure 10–15 minutes early warning
Operator Response Time 15–30 seconds (emergency crash landing) 5–15 minutes (orderly airport diversion)
Diagnostic Context Generic "ENGINE FAULT" light Pinpointed root cause + TreeSHAP explainability
Action Plan Panic decision making Automated descent profile + Squawk 7700 gating

⚙️ 2. The Solution: 3-Layer Physics-Informed Architecture

AeroTwin fuses first-principles thermodynamics with high-speed machine learning and explainable AI across 5 processing layers:

[ Aircraft Engine / Sensor Harness ]  ──►  14 Channels @ 1 Hz
                    │
                    ▼
[ Layer 1: Data Quality Guard ]        ──►  Outlier rejection, drift detection, rolling median imputation
                    │
                    ▼
[ Layer 2: MAVLink v2 & RF Link ]      ──►  Msg ID 50001 (56-byte payload, CRC-16), 900 MHz RFD900x
                    │
                    ▼
[ Layer 3: Edge Compute (RPi 4) ]      ──►  Isolation Forest anomaly detector (5.75 ms / ~23 ms Pi 4)
                    │
                    ▼
[ Layer 4: Ground AI Diagnostics ]     ──►  XGBoost Classifier (93.68% F1) + RUL Regressor (63s MAE) + TreeSHAP
                    │
                    ▼
[ Layer 5: Flight Deck Dashboard ]     ──►  Real-Time Telemetry, Digital RUL Timer & Squawk 7700

🔬 Supported In-Flight Failure Modes