Smart Industrial Machine Monitoring & Predictive Maintenance 🏭

An Industry 4.0 predictive maintenance system that uses Sensor Fusion, Edge DSP, Unsupervised AI (Isolation Forest), and 3D Digital Twin Visualization to detect machinery failures before they occur — without needing historical failure data.

---

The Problem

In modern manufacturing, unplanned machinery downtime costs billions annually. Traditional IoT systems stream raw, high-frequency sensor data to the cloud — causing bandwidth exhaustion, high latency, and expensive cloud compute. And because industrial motors rarely fail, acquiring labelled failure datasets for supervised AI is nearly impossible.

---

The Solution

Three-layer architecture:

1. Edge DSP on ESP32 — processes raw vibration data locally using FFT, transmitting only Dominant Frequency and Magnitude. Reduces network payload by over 90%.
2. AI Gateway — a Python backend runs an Unsupervised Isolation Forest model. It learns the machine's healthy baseline and calculates an anomaly score from reconstruction error — no failure labels needed.
3. 3D Digital Twin — a React dashboard with live 3D mesh rendering dynamically reflects the real-time health status of each machine.

---

Key Features

• Bare-Metal I2C — bypasses standard libraries to prevent microcontroller panics during heavy physical vibration
• Hardware-Accelerated FFT — converts time-domain acceleration into frequency-domain data (Hz) directly on the ESP32
• Unsupervised Anomaly Detection — Isolation Forest detects subtle shifts in vibration frequencies and temperature correlations
• 3D Digital Twin HUD — React Three Fiber with conditional mesh rendering: Green = Healthy, Yellow = Warning, Red = Critical
• Real-time Telemetry — live Recharts graphs for motor temperature, vibration magnitude, dominant frequency, and anomaly score
• Multi-machine Support — monitors 6 machines simultaneously (Compressor, Motor Drive, Pump, Conveyor, Robot Arm, Turbine)
• FastAPI Backend — SSE-based streaming API for real-time data push to the frontend
• CSV Data Logger — PySerial logger captures and stores all sensor readings with timestamps

---

Tech Stack

Layer	Tech
Hardware	ESP32-WROOM · MPU6050 · DS18B20 · DHT22
Firmware	C++ / Arduino Core · arduinoFFT · OneWire
AI & Backend	Python 3 · Scikit-learn (Isolation Forest) · FastAPI · PySerial · Pandas · NumPy
Frontend	React.js · Vite · Tailwind CSS · Recharts · React Three Fiber (WebGL 3D)

---

Hardware Wiring

Sensor	ESP32 Pin	Protocol	Notes
MPU6050	GPIO 21 (SDA), GPIO 22 (SCL)	I2C	Must be rigidly mounted to motor
DS18B20	GPIO 4	1-Wire	Requires 4.7kΩ pull-up resistor
DHT22	GPIO 5	Digital	Measures ambient temperature & humidity

Do not power a heavy-vibration DC motor directly from ESP32 pins — use an external isolated power supply to prevent brownouts.

---

Project Structure

Smart-Industrial-Monitoring/
├── sketch_mar17b.ino       # ESP32 firmware — FFT + sensor fusion
├── main.py                 # FastAPI server + SSE streaming + machine state
├── ai_engine.py            # Isolation Forest anomaly detection engine
├── logger.py               # PySerial CSV data logger
├── machine_data_log.csv    # Recorded sensor telemetry dataset
├── App.jsx                 # React 3D Digital Twin dashboard
├── requirements.txt        # Python dependencies
└── *.glb                   # 3D model for Digital Twin visualization

---

Team & Contributions

Kushagra Agarwal — Project Lead

• Designed the overall system architecture (Edge DSP → AI Gateway → Digital Twin)
• Wrote the ESP32 firmware — bare-metal I2C, hardware FFT, sensor fusion across MPU6050, DS18B20, and DHT22
• Built and trained the Unsupervised Isolation Forest AI engine (ai_engine.py)
• Led hardware setup, wiring, and physical testing on live motor hardware

Ankit Notnani — Contributor

• Assisted in building the React frontend and 3D Digital Twin dashboard (App.jsx) — component structure, telemetry graph integration with Recharts, and real-time state updates
• Helped develop the FastAPI backend (main.py) — API route structure, CORS setup, and SSE streaming endpoint for live data push to the frontend
• Assisted with the CSV data pipeline — integrating the PySerial logger output with the backend data processing flow

Project guided by Dr. Ayush Agrawal.

---

Getting Started

Backend

# Install dependencies
pip install -r requirements.txt

# Start the data logger (connect ESP32 first)
python logger.py

# Start the API server
python main.py

Frontend

npm install
npm run dev

Open http://localhost:5173

---

How It Works — Data Flow

ESP32 (FFT) → Serial → logger.py → CSV → main.py (FastAPI) → SSE → React Dashboard
                                              ↓
                                      ai_engine.py (Anomaly Score)
                                              ↓
                                      3D Digital Twin (Health Status)

---

Future Scope

• LSTM Autoencoder for deeper temporal anomaly detection
• Real-time hardware API integration (replace CSV polling)
• Cloud deployment with multi-plant dashboard
• Mobile alerts for critical anomaly events