🛠️ Predictive Maintenance System using XGBoost

Proactive equipment maintenance through intelligent Time-To-Failure predictions

Demo • Installation • Usage • API Reference • Contributing

---

🚀 Project Overview

Predictive Maintenance is a cutting-edge machine learning solution that transforms traditional reactive maintenance into a proactive, data-driven approach. Our system leverages advanced XGBoost Multi-Output Regression to predict equipment failures before they occur, potentially saving thousands in downtime costs.

🎯 Key Features

• Multi-Component Prediction: Simultaneously predicts failure times for 4 different machine components
• Short-Term Accuracy: Optimized for predictions under 10 weeks with ~91% accuracy
• Real-Time Interface: Flask-based web application for instant predictions
• Industrial Ready: Designed for real-world manufacturing environments
• Comprehensive Pipeline: End-to-end solution from data preprocessing to deployment

📈 Performance Metrics

Metric	Value
Test R²	~0.95
Test MAPE	~11.25%
Short-Term MAE	~0.91 weeks
Training Samples	200,000 (stratified)

---

📁 Project Structure

predictive-maintenance/
├── 📂 model_code/
│   ├── 🐍 Model_XGB.py          # Main training pipeline
│   └── 🔮 manual.py             # Manual prediction script
│
├── 📂 flask_code/
│   ├── 🌐 app.py                # Flask backend server
│   └── 📂 templates/
│       └── 🎨 index.html        # Web interface
│
├── 📊 TTF_Dataset_Weeks.csv     # Training dataset
├── 📋 requirements.txt          # Python dependencies
├── 📜 README.md                 # Project documentation
└── 🎯 .gitignore               # Git ignore rules

---

🛠️ Installation

Prerequisites

• Python 3.8 or higher
• pip package manager
• Git

Quick Start

1. Clone the Repository

   git clone https://github.com/yourusername/predictive-maintenance.git
   cd predictive-maintenance

2. Create Virtual Environment (Recommended)

   # Windows
   python -m venv venv
   venv\Scripts\activate
   
   # macOS/Linux
   python3 -m venv venv
   source venv/bin/activate

3. Install Dependencies

   pip install -r requirements.txt

   Or install manually:

   pip install xgboost pandas scikit-learn flask matplotlib seaborn joblib numpy

---

🚀 Usage

1. Train the Model

cd model_code
python Model_XGB.py

Output: Trained model files (model.pkl, scaler.pkl, encoders.pkl)

2. Manual Predictions

python manual.py

Provide machine parameters and get TTF predictions for all components.

3. Web Application

cd flask_code
python app.py

🌐 Access: Open your browser to http://127.0.0.1:5000

Features:

• Upload machine data via web interface
• Real-time TTF predictions
• Visual results display
• Export predictions as CSV

---

📊 Dataset Overview

Input Features

Category	Features
Sensor Data	volt, rotate, pressure, vibration
Machine Info	model, age, error_count
Maintenance	days_since_comp1_maint to days_since_comp4_maint

Target Variables

• ttf_comp1_weeks - Time to failure for Component 1
• ttf_comp2_weeks - Time to failure for Component 2
• ttf_comp3_weeks - Time to failure for Component 3
• ttf_comp4_weeks - Time to failure for Component 4

Data Preprocessing

• Stratified Sampling: 200,000 representative samples
• Feature Scaling: StandardScaler normalization
• Encoding: Label encoding for categorical variables
• Quality Assurance: Outlier detection and handling

---

🧠 Machine Learning Architecture

Model Pipeline

graph LR
    A[Raw Data] --> B[Preprocessing]
    B --> C[Feature Engineering]
    C --> D[XGBoost Multi-Output]
    D --> E[TTF Predictions]
    E --> F[Maintenance Scheduling]

Loading

Key Components

1. Data Preprocessing

   • Stratified sampling for balanced representation
   • Feature scaling and encoding
   • Missing value imputation

2. Model Architecture

   • XGBoostRegressor with MultiOutputRegressor wrapper
   • Custom hyperparameter tuning
   • Cross-validation for robust performance

3. Evaluation Strategy

   • Multiple metrics: RMSE, MAE, R², MAPE
   • Special focus on short-term predictions (<10 weeks)
   • Component-wise performance analysis

---

📊 Results & Performance

Overall Performance

Component 1: R² = 0.94, MAPE = 10.8%
Component 2: R² = 0.95, MAPE = 11.2%
Component 3: R² = 0.96, MAPE = 11.5%
Component 4: R² = 0.95, MAPE = 11.4%

Short-Term Accuracy (< 10 weeks)

• Mean Absolute Error: 0.91 weeks
• Critical for: Emergency maintenance scheduling
• Business Impact: Prevents unexpected downdowns

---

🔧 API Reference

Flask Endpoints

POST /predict

Predict TTF for machine components.

Request Body:

{
  "volt": 180.5,
  "rotate": 1200,
  "pressure": 85.2,
  "vibration": 45.1,
  "model": "Model_A",
  "age": 156,
  "error_count": 3,
  "days_since_comp1_maint": 45,
  "days_since_comp2_maint": 32,
  "days_since_comp3_maint": 67,
  "days_since_comp4_maint": 23
}

Response:

{
  "predictions": {
    "component_1": 8.5,
    "component_2": 12.3,
    "component_3": 6.1,
    "component_4": 15.7
  },
  "status": "success",
  "timestamp": "2024-01-20T10:30:00Z"
}

---

🔬 Model Interpretability

Feature Importance

The model considers these factors in order of importance:

1. Vibration levels - Primary indicator of component stress
2. Days since last maintenance - Critical for all components
3. Machine age - Long-term degradation factor
4. Error count - Operational stress indicator
5. Sensor readings - Real-time operational state

SHAP Analysis

🚧 Coming Soon: SHAP-based feature importance and prediction explanations

---

🚀 Deployment

Docker Deployment

FROM python:3.9-slim

WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt

COPY . .
EXPOSE 5000

CMD ["python", "flask_code/app.py"]

Production Considerations

• Scalability: Consider using Gunicorn for production
• Monitoring: Implement logging and performance monitoring
• Security: Add authentication and input validation
• Data Pipeline: Automate model retraining with new data

---

🛣️ Roadmap

Phase 1 (Current)

• XGBoost multi-output model
• Flask web interface
• Basic performance metrics

---

🤝 Contributing

We welcome contributions! Please see our Contributing Guidelines for details.

---

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

👥 Team

| Dharam Patel | Ayesha Patel | Vrunda Radadiya |

---

🙏 Acknowledgments

---

⭐ Star this repository if it helped you!

Report Bug · Request Feature · Documentation