An end-to-end MLOps project for predicting Remaining Useful Life (RUL) of industrial equipment using the NASA C-MAPSS Turbofan Engine Degradation Dataset. This project demonstrates production-grade ML engineering practices including CI/CD, experiment tracking, model serving, containerization, and monitoring.
Predictive maintenance uses machine learning to predict when equipment will fail, enabling proactive maintenance scheduling. This project:
- Predicts RUL (Remaining Useful Life) of turbofan engines
- Trains multiple models (Random Forest, Gradient Boosting, LSTM, etc.)
- Tracks experiments with MLflow
- Serves predictions via REST API
- Monitors performance through Streamlit dashboard
- Automates CI/CD with GitHub Actions
- ⬇️ Reduce unplanned downtime by 30-50%
- 💰 Lower maintenance costs through optimized scheduling
- 📈 Extend equipment lifespan with timely interventions
┌─────────────────────────────────────────────────────────────────────────────┐
│ PREDICTIVE MAINTENANCE SYSTEM │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌───────────┐ │
│ │ Data │───▶│ Data │───▶│ Data │───▶│ Model │ │
│ │ Ingestion │ │ Validation │ │Transformation│ │ Training │ │
│ └──────────────┘ └──────────────┘ └──────────────┘ └─────┬─────┘ │
│ │ │ │ │ │
│ │ │ │ ▼ │
│ │ │ │ ┌───────────┐ │
│ │ │ │ │ Model │ │
│ │ │ │ │Evaluation │ │
│ │ │ │ └─────┬─────┘ │
│ │ │ │ │ │
│ ┌──────▼───────────────────▼───────────────────▼──────────────────▼─────┐ │
│ │ MLflow Tracking Server │ │
│ │ (Experiments, Parameters, Metrics, Artifacts) │ │
│ └───────────────────────────────┬───────────────────────────────────────┘ │
│ │ │
│ ┌───────────────────────────────▼───────────────────────────────────────┐ │
│ │ Model Registry │ │
│ │ (Versioning, Staging, Production) │ │
│ └───────────────────────────────┬───────────────────────────────────────┘ │
│ │ │
│ ┌────────────────────────┼────────────────────────┐ │
│ ▼ ▼ ▼ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ FastAPI │ │ Streamlit │ │ Batch │ │
│ │ REST API │ │ Dashboard │ │ Prediction │ │
│ └──────┬──────┘ └──────┬──────┘ └──────┬──────┘ │
│ │ │ │ │
│ ┌──────▼───────────────────────▼────────────────────────▼───────────────┐ │
│ │ Prometheus + Grafana │ │
│ │ (Monitoring & Alerting) │ │
│ └───────────────────────────────────────────────────────────────────────┘ │
│ │
└──────────────────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────────────────┐
│ INFRASTRUCTURE │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ Docker │ │ GitHub │ │ DVC │ │ MongoDB │ │
│ │ Compose │ │ Actions │ │ (Data) │ │ (Storage) │ │
│ └─────────────┘ └─────────────┘ └─────────────┘ └─────────────┘ │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
| Category | Technologies |
|---|---|
| ML/DL | scikit-learn, TensorFlow/Keras, LSTM |
| MLOps | MLflow, DVC, Docker, GitHub Actions |
| API | FastAPI, Uvicorn, Pydantic |
| Data | Pandas, NumPy, MongoDB |
| Visualization | Streamlit, Plotly, Matplotlib |
| Testing | pytest, pytest-cov, hypothesis |
| Code Quality | Black, isort, flake8, mypy, pre-commit |
| Monitoring | Prometheus, Grafana |
- ✅ Automated data ingestion from multiple sources
- ✅ Data validation with quality checks and anomaly detection
- ✅ Feature engineering (lag features, rolling statistics)
- ✅ Multiple model training (RF, GB, Linear, Ridge, Lasso, SVR, LSTM)
- ✅ Hyperparameter tuning with GridSearchCV
- ✅ Model evaluation with comprehensive metrics (RMSE, MAE, R², MAPE)
- ✅ Experiment tracking with MLflow
- ✅ Model registry for versioning and staging
- ✅ Data versioning with DVC
- ✅ CI/CD pipeline with GitHub Actions
- ✅ Containerization with Docker & Docker Compose
- ✅ Pre-commit hooks for code quality
- ✅ REST API with FastAPI for real-time predictions
- ✅ Batch prediction pipeline for large datasets
- ✅ Monitoring dashboard with Streamlit
- ✅ Health checks and API documentation (Swagger/OpenAPI)
- ✅ Risk level classification (Critical, High, Medium, Low)
predictive-maintenance/
├── .github/
│ └── workflows/
│ ├── main.yml # CI/CD pipeline
│ └── model-training.yml # Scheduled training
├── api/
│ ├── __init__.py
│ ├── main.py # FastAPI application
│ └── schemas.py # Pydantic models
├── config/
│ ├── config.yaml # Main configuration
│ └── schema.yaml # Data schema
├── dashboard/
│ └── app.py # Streamlit dashboard
├── data/
│ ├── raw/ # Raw data
│ ├── validated/ # Validated data
│ ├── transformed/ # Processed features
│ └── predictions/ # Batch predictions
├── monitoring/
│ ├── prometheus.yml # Prometheus config
│ └── grafana/ # Grafana dashboards
├── notebooks/
│ └── eda.ipynb # Exploratory analysis
├── src/
│ ├── components/
│ │ ├── data_ingestion.py
│ │ ├── data_validation.py
│ │ ├── data_transformation.py
│ │ ├── model_trainer.py
│ │ ├── model_evaluation.py
│ │ └── batch_prediction.py
│ ├── pipelines/
│ │ └── training_pipeline.py
│ ├── utils/
│ │ ├── logger.py
│ │ └── model_utils.py
│ ├── constants/
│ │ └── __init__.py
│ └── mlflow_tracking.py
├── tests/
│ ├── unit/
│ │ ├── test_data_validation.py
│ │ ├── test_model_evaluation.py
│ │ └── test_api.py
│ ├── integration/
│ │ └── test_pipeline.py
│ └── conftest.py # Pytest fixtures
├── artifacts/
│ ├── models/ # Trained models
│ ├── logs/ # Application logs
│ └── reports/ # Evaluation reports
├── .dvc/ # DVC configuration
├── .pre-commit-config.yaml # Pre-commit hooks
├── docker-compose.yml # Docker services
├── Dockerfile # Multi-stage Dockerfile
├── requirements.txt # Dependencies
├── setup.py # Package setup
├── pyproject.toml # Build configuration
├── pytest.ini # Pytest configuration
└── README.md # This file
1. Data Ingestion → Load raw sensor data from source
2. Data Validation → Validate schema, types, and ranges
3. Transformation → Feature engineering & scaling
4. Model Training → Train multiple models
5. Model Evaluation → Compare and select best model
6. Model Registry → Version and stage models
| Model | Type | Use Case |
|---|---|---|
| Random Forest | Ensemble | Baseline, robust |
| Gradient Boosting | Ensemble | High accuracy |
| Linear Regression | Linear | Interpretable |
| Ridge/Lasso | Linear | Regularized |
| SVR | Kernel | Non-linear |
| LSTM | Deep Learning | Sequence modeling |
| Method | Endpoint | Description |
|---|---|---|
| GET | /health |
Health check |
| GET | /models |
List available models |
| POST | /predict |
Single/batch prediction |
| POST | /predict/batch |
File-based batch prediction |
| POST | /models/reload |
Reload models |
The Streamlit dashboard provides:
- Overview: Key metrics, model comparison
- Model Performance: Detailed metrics, visualizations
- Predictions: Interactive prediction interface
- Data Explorer: Feature distributions, correlations
- System Health: API status, resource usage
| Model | RMSE | MAE | R² |
|---|---|---|---|
| Random Forest | 18.5 | 12.3 | 0.87 |
| Gradient Boosting | 17.2 | 11.8 | 0.89 |
| LSTM | 15.8 | 10.5 | 0.91 |