Luminous APOGEE '26 Hackathon
"We don't just monitor failures; we predict the mechanical heartbeat of the grid."
In the world of industrial power, a single relay failure isn't just a technical glitch—it's a significant financial risk. A.R.M.S. is an autonomous ecosystem designed to bridge the gap between legacy hardware and intelligent maintenance.
By combining Deterministic Physics with Non-Linear Machine Learning, A.R.M.S. provides an Early Warning System that detects microscopic degradation in inverter relays months before they reach a breaking point.
| Platform | Access Link | Description |
|---|---|---|
| Mobile PWA | armsinv.vercel.app | Premium, mission-control interface for field engineers. |
| Industrial Dashboard | ThingsBoard Live | Primary industrial SCADA view for real-time fleet monitoring. |
The pipeline is optimized for zero-latency Edge Intelligence.
- The Pulse (Simulator): A physics-accurate Modbus TCP engine simulating inverter transients.
- The Nerve (Unified Bridge): An edge-computing gateway that polls raw registers and performs high-speed feature engineering.
- The Brain (Hybrid AI): An integrated RUL (Remaining Useful Life) engine that fuses Arrhenius physics with LSTM-based ML confidence.
- The Command (PWA/SCADA): Dual-stream visualization for both enterprise and mobile-first field inspection.
ARMS/
├── go.py # Unified system orchestrator
├── .env.local # Local environment configuration
│
├── backend/ # Core logic and system gateway
│ ├── inverter_sim.py # Modbus TCP hardware simulator
│ ├── modbus_mqtt_bridge.py # Edge-AI gateway & MQTT publisher
│ ├── rul_engine.py # Hybrid PIML inference module
│ └── requirements.txt # Python dependency list
│
├── pwa-dashboard/ # Next.js mobile-first application
│ ├── src/ # UI components and telemetry context
│ ├── public/ # Static assets and demo configurations
│ └── tailwind.config.js # Design system configuration
│
├── models/ # Pre-trained AI artifacts
│ ├── relay_rul_model.joblib # Trained inference model
│ └── relay_rul_scaler.joblib # Feature normalization scaler
│
├── docs/ # Technical setup & reference
│ └── THINGSBOARD_SETUP.md # Dashboard replication guide
│
├── assets/ # Visual assets & schematics
│ └── inverter_relay_schematic.png
│
└── utils/ # Connectivity & diagnostic tools
└── find_device.py # Network device discovery utility
- Physics-First Initialization: Models are accurate from Day 1. We use mechanical formulas to understand the baseline before the first data point is recorded.
- Arc Energy Mapping: The system calculates specific arcing energy per switch using inverter current and overload registers to identify invisible thermal fatigue.
- Deterministic Safety Guardrails: A secondary physics layer provides a "Veto" system, triggering critical alerts on mechanical limits even if the AI confidence is high.
# 1. Start the Pulse & Brain
python go.py
# 2. View the Future
# Access the PWA or ThingsBoard link via the Ecosystem Links above.Built with precision for APOGEE '26 by Arjun Bora & Arin Bhuyan.