AI-Driven Predictive Maintenance System for Industrial Machines

AI-powered machine failure prediction with 99.95% accuracy. Analyzes sensor data and tells you which machines need maintenance.

Quick Start

pip install -r requirements.txt
python run.py

Pick GUI mode (2), click "Train Models", wait a minute. Done.

Features

• XGBoost ML model - 99.95% accuracy on failure prediction
• 3 dashboards - Model performance, production monitoring, real-time animated
• WebSocket streaming - Live sensor data and predictions
• Multiple interfaces - CLI menu or web UI, your choice
• Centralized config - Change everything in one file

Usage

First Time

python run.py
# Choose GUI (2)
# Click "Train Models"
# Wait ~1 minute

This trains the models and generates dashboards. Only needed once.

Daily Use

python run.py
# Choose GUI (2)
# Click whatever you need

Models stay saved, no need to retrain.

Project Structure

APP/
├── run.py              # Start here
├── src/                # Source code
│   ├── config.py       # All settings
│   ├── web_dashboard.py    # FastAPI server + WebSocket
│   └── predictive_maintenance.py  # ML pipeline
├── dashboards/         # HTML dashboards (generated)
├── models/             # Trained models (generated)
├── data/               # Sensor data
└── results/            # Prediction CSVs

Dashboards

Model Performance - Accuracy metrics, confusion matrix, ROC curves

Production Monitoring - Risk distribution, high-risk machines, priorities

Real-Time Monitor - Live animated charts, sensor readings, failure probability

First two generate when you train. Third one is always there.

Configuration

Edit src/config.py:

SERVER = {"port": 8000}  # Change port
THRESHOLDS = {"high_risk_min": 0.7}  # Risk level cutoff
WEBSOCKET = {"update_interval": 1.0}  # Update speed

API Endpoints

When server is running:

GET  /                      Main UI
POST /api/train             Train models  
GET  /api/results           Prediction data
GET  /model_performance     Model dashboard
GET  /production_dashboard  Operations dashboard
GET  /realtime_dashboard    Live monitor
WS   /ws/realtime           WebSocket stream

Tech Stack

Backend: Python 3.11+, FastAPI, XGBoost, scikit-learn

Frontend: HTML/CSS/JS, Chart.js, Plotly

Data: Pandas, NumPy

Troubleshooting

ModuleNotFoundError: pip install -r requirements.txt

Port 8000 in use: Kill existing process or change port in config

Dashboard 404: Train models first - they're generated files

WebSocket won't connect: Models need to be trained

For Developers

Add route:

# In src/web_dashboard.py
@app.get("/api/custom")
async def custom():
    return {"data": "whatever"}

Modify ML pipeline: Edit src/predictive_maintenance.py

Deploy: Use Gunicorn:

gunicorn -w 4 -k uvicorn.workers.UvicornWorker src.web_dashboard:app

Performance

• Training: ~5 seconds
• Prediction: <10ms
• Memory: ~200MB
• WebSocket latency: ~50ms