🛡️ AI Based Industrial-IoT-Anomaly-Detector

A robust, AI-driven application designed to predict equipment failures before they occur. This project leverages machine learning (Random Forest) to analyze sensor data in real-time, helping industries optimize maintenance schedules, reduce downtime, and cut operational costs.

🔴 Live Demo: Click Here

📸 Dashboard Previews

1. Manual Diagnostics Mode

2. Automatic Fleet Monitoring Mode

🚀 Key Features

• Quadruple Operation Modes:
  • 📈 Historical Analytics (v4): NEW - Interactive Plotly dashboards charting massive historical timelines of sensor logs and offering direct CSV dataset exports.
  • 📡 Live IoT Sensor (v3): Connects directly to hardware (ESP32) via a local FastAPI backend. Ingests real-time physical telemetry.
  • 🏭 Simulation Diagnostics (v2): Real-time dashboard simulating a fleet of 9+ industrial machines. Tracks live telemetry, failure risks, and operational status simultaneously.
  • 🛠️ Manual Diagnostics (v1): Deep-dive analysis for single units. Input sensor parameters manually to stress-test specific scenarios and visualize failure probabilities.
• Intelligent Prediction Engine: Powered by a trained Random Forest Classifier (models/model.joblib), achieving consistently high accuracy (>98%) in detecting potential failures.
• Real-time Feature Engineering: Automatically calculates critical derived metrics like Temperature Difference and Mechanical Power on the fly.
• Professional UI/UX: Dark-mode enabled, responsive design built with Streamlit, featuring interactive metric cards, visual health indicators, and dynamic alerts.

🧠 Model & Data

The predictive model is trained on a synthetic dataset mirroring real-world industrial sensor data (based on the AI4I 2020 Predictive Maintenance Dataset).

Input Parameters (Sensors)

Feature	Unit	Description
Air Temperature	[K]	Ambient room temperature.
Process Temperature	[K]	Temperature of the manufacturing process.
Rotational Speed	[rpm]	Speed of the spindle/drive.
Torque	[Nm]	Torque applied during operation.
Tool Wear	[min]	Cumulative tool wear time.
Type	L/M/H	Quality variant (Low, Medium, High).

Output (Prediction)

• No Failure: ✅ System Healthy
• Failure: 🚨 Maintenance Required (Probability > 50%)

🛠️ Installation & Setup

Follow these steps to set up the project locally.

1. Clone the Repository

git clone https://github.com/ShivamMaurya14/Predictive-Maintenance-System.git
cd Predictive-Maintenance-System

2. Create a Virtual Environment (Optional but Recommended)

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

Ensure you have Python 3.8+ installed.

pip install -r requirements.txt

Note: If requirements.txt is missing, install the core libraries: pip install streamlit pandas numpy scikit-learn joblib

4. Training the Model (Optional)

The project comes with a pre-trained model (models/model.joblib). If you wish to retrain it:

python train.py

This will create a fresh model.joblib file in your models/ directory.

5. Run the Application

The application now features a decoupled microservice architecture (Backend + Frontend). You will need two terminal windows.

Terminal 1 (Backend API):

uvicorn api:app --reload

Terminal 2 (Frontend Dashboard):

streamlit run app.py

📂 Project Structure

├── api.py                 # FastAPI Backend Server (IoT Data Ingestion)
├── app.py                 # Main Streamlit Application (Frontend)
├── esp32_firmware/        # C++ firmware for the ESP32 microcontroller
├── train.py               # ML Model Training Script (Backend logic)
├── models/
│   └── model.joblib       # Serialized Random Forest Model
├── dataset/
│   └── ai4i2020.csv       # Training Dataset
├── nb/
│   └── ml-powered-maintenance-smarter-proactive.ipynb  # exploration
├── resources/             # Preview images & assets
├── requirements.txt       # Python Dependencies
├── Makefile               # Automation commands
├── LICENSE                # MIT License
└── README.md              # Project Documentation

🏭 Industry Applications

This predictive maintenance solution is versatile and can be adapted across various high-stakes industries:

• Manufacturing & Assembly Lines: Monitor robotic arms, conveyor motors, and CNC machines to prevent production bottlenecks.
• Energy & Utilities: Track turbine health in wind farms or pumps in hydroelectric plants to avoid catastrophic failures.
• Transportation & Logistics: Fleet management for heavy trucks or railway rolling stock, ensuring safety and reliability.
• Oil & Gas: Predictive analytics for drilling equipment and pipeline pressure valves in remote locations.
• Data Centers: Monitor cooling systems and power generators to maintain 99.9% uptime.

🔮 Future Scope & Roadmap

We are constantly evolving to meet Industry 4.0 standards. Here is our vision for the future:

🔗 Phase 1: Real-Time Integration

• IoT & MQTT Connectivity: Direct integration with industrial IoT standards (MQTT/OPC UA) to ingest live sensor streams from PLC controllers.
• Edge Deployment: Optimizing the model to run on edge devices (Raspberry Pi/Jetson Nano) for ultra-low latency predictions without cloud dependency.

💾 Phase 2: Enterprise Scalability

• Time-Series Database (InfluxDB): High-performance storage for historical sensor data to enable long-term trend analysis.
• API Gateway: Expose prediction results via a RESTful API for integration with existing ERP (SAP/Oracle) or MES systems.

🧠 Phase 3: Advanced Intelligence

• Deep Learning (LSTM/GRU): Experimenting with Recurrent Neural Networks to capture complex temporal dependencies in sensor patterns.
• Root Cause Analysis: Automatic identification of which sensor triggered the failure alert (e.g., "High vibrations caused bearing failure").
• Digital Twin: Creating a fully interactive 3D digital twin of the factory floor using Unity or Three.js linked to this backend.

🤝 Contributing

Contributions are welcome! Please fork the repository and submit a pull request for any enhancements, bug fixes, or new feature ideas.

---

Developed with ❤️ by Shivam Maurya