Real-Time Predictive Maintenance using Multi-Modal Intelligence.
EdgeSense AI is a complete, end-to-end hardware-to-UI machine learning platform designed to predict industrial machine failure before it happens. By simultaneously monitoring acoustic telemetry and 3-axis vibration metrics directly from the edge, this system streams multi-modal intelligence to a high-performance visual dashboard via WebSockets.
- Multi-Modal ML Intelligence: Combines an Acoustic CNN with a Vibration LSTM to generate highly-confident predictive maintenance scores and isolate anomalous machinery behavior.
- Real-time Telemetry Dashboard: A stunning, ultra-modern Next.js application that renders real-time streaming signals at sub-10ms latency using Live WebSockets.
- Optimized Edge Inference: Uses
onnxruntimefor high-speed prediction on resource-constrained environments (successfully deployed on Render Free Tier). - Actionable Alerts & CSV Export: Generates critical fault warnings mapped to specific bearing/motor defects and allows an instant snapshot export of fleet health via CSV.
- Dockerized Deployment: Ready-to-go Docker configuration for local development and cloud-native scaling.
- Custom Training Pipeline: A full end-to-end Python pipeline to normalize, train, and export PyTorch models into performant ONNX artifacts.
The project is decoupled into clear, standalone domains:
/next_frontend— A React + Next.js (+ Tailwind CSS / Framer Motion) futuristic visualization dashboard. Deployed on Vercel./backend— A FastAPI Python server handling WebSocket streams, ONNX inference, and SQLite history. Deployed on Render./train— ML Pipeline for data preparation and PyTorch-to-ONNX training./firmware— Embedded C++ (PlatformIO) code for ESP32 devices attached to physical sensors.
The application is currently live and can be accessed via the following links:
- Frontend (Dashboard): https://edge-sense-ai-real-time-predictive.vercel.app/
- Backend (API & WebSockets): https://edgesense-api.onrender.com
- Interactive API Documentation: https://edgesense-api.onrender.com/docs
The easiest way to run the entire stack locally is via Docker.
docker-compose up --build- Frontend:
http://localhost:3000 - Backend:
http://localhost:8000
To connect physical ESP32 hardware to the live backend, update the serverUrl in firmware/src/main.cpp:
const char* serverUrl = "https://edgesense-api.onrender.com/api/predict";- Place your raw matrices or
.npzfiles insidetrain/data/. - Run the master PowerShell script:
./run_training.ps1 - The script will train the models, convert them to ONNX, and update the backend models automatically.
Built with ❤️ for the 2026 Engineering Hackathon. Let's eliminate unplanned downtime forever.