|
|
--- |
|
|
title: Engine Predictive Maintenance |
|
|
emoji: 🔧 |
|
|
colorFrom: blue |
|
|
colorTo: green |
|
|
sdk: streamlit |
|
|
sdk_version: 1.28.0 |
|
|
app_file: app.py |
|
|
pinned: false |
|
|
--- |
|
|
|
|
|
# Engine Predictive Maintenance |
|
|
|
|
|
Predict engine failures before they happen using machine learning. |
|
|
|
|
|
## 🎯 Features |
|
|
- **Real-time Monitoring**: Enter sensor readings manually for instant predictions |
|
|
- **Batch Processing**: Upload CSV files for bulk predictions |
|
|
- **Risk Assessment**: Get risk level (Low/Medium/High) with confidence scores |
|
|
- **Data Export**: Download predictions with timestamps for auditing |
|
|
|
|
|
## 📊 Model Performance |
|
|
- **Algorithm**: Random Forest with SMOTE (handles class imbalance) |
|
|
- **Accuracy**: 92%+ |
|
|
- **F2-Score**: 0.657 (optimized for recall - prioritizes catching failures) |
|
|
- **Training Data**: 1000+ engine sensor readings |
|
|
|
|
|
## 🚀 Quick Start |
|
|
|
|
|
### Manual Prediction |
|
|
1. Enter 6 sensor readings (Lube Oil Pressure, Temperature, RPM, etc.) |
|
|
2. Get instant failure probability and risk level |
|
|
3. Review model confidence metrics |
|
|
|
|
|
### Batch Prediction |
|
|
1. Upload a CSV file with multiple engine readings |
|
|
2. Get predictions for all rows |
|
|
3. Download results with timestamps |
|
|
|
|
|
## 📈 Sensor Inputs |
|
|
- Lube Oil Pressure |
|
|
- Lube Oil Temperature |
|
|
- Coolant Temperature |
|
|
- Engine RPM |
|
|
- Fuel Pressure |
|
|
- Coolant Pressure |
|
|
|
|
|
## ⚙️ Technical Details |
|
|
- **Model Format**: .joblib (scikit-learn native) |
|
|
- **Data Processing**: Standardized features, engineered ratios |
|
|
- **Inference**: <100ms per prediction |
|
|
- **Deployment**: Hugging Face Spaces (auto-scaling) |
|
|
|
|
|
## 📝 How It Works |
|
|
1. Input sensors are standardized using training data statistics |
|
|
2. Features are engineered (pressure ratios, temperature differences) |
|
|
3. Random Forest classifier produces failure probability |
|
|
4. F2-Score threshold (0.65) determines risk level classification |
|
|
|
|
|
## 🔒 Privacy & Security |
|
|
- All data processing happens in-browser |
|
|
- No data is stored or logged |
|
|
- Model weights are public and auditable |
|
|
- Predictions are real-time only |
|
|
|
|
|
## 🤝 Support |
|
|
For issues or feedback, visit: https://github.com/your-org/predictive-engine-maintenance |
|
|
|
|
|
--- |
|
|
*Built with scikit-learn, Streamlit, and Hugging Face Spaces* |
|
|
|
