Turbofan RUL Prediction - NASA CMAPSS

End-to-end predictive maintenance system for turbofan engines using the NASA CMAPSS dataset. The model predicts the Remaining Useful Life (RUL) of a turbofan engine high-pressure compressor (number of operational cycles remaining before failure) in order to improve maintenance decisions.

---

Results

Model	RMSE (test)	MAE (test)	R² (test)
Baseline	39.92	35.26	0.00
Linear Regression	19.00	15.39	0.77
Random Forest	15.90	10.92	0.84
RF Tuned	15.51	10.71	0.85
XGBoost	16.98	11.48	0.82
XGBoost Tuned	15.53	11.18	0.85

NASA Benchmark (Random Forest Tuned - full train) = RMSE = 17.34 · MAE = 12.10 · R² = 0.806 · NASA Score = 907.2

---

Project Structure

turbofan-rul-prediction/
│
├── data/
│   ├── raw/                  # Original NASA CMAPSS files
│   └── processed/            # Cleaned and feature-engineered datasets
│
├── notebooks/
│   ├── 01_data_exploration.ipynb       # EDA, sensor visualization, variance analysis
│   ├── 02_feature_engineering.ipynb    # RUL computation, feature creation, correlation filtering
│   ├── 03_model.ipynb                  # Model training, tuning, comparison, NASA benchmark
│   └── 04_business_conclusion.ipynb    # NASA scoring function, operational risk
│
├── outputs/
│   ├── models/               # Saved pipelines (.joblib)
│   └── plots/                # All generated figures (.png)
│
├── src/
│   ├── config.py             # Paths and variables
│   ├── preprocessing.py      # RUL computation, feature engineering, filtering
│   └── model.py              # Training utilities, evaluation, MLflow experiment tracking
│
├── requirements.txt
└── README.md

---

Quickstart

Prerequisites

git clone https://github.com/elouanXP/turbofan-rul-prediction
cd turbofan-rul-prediction
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt

Data

Download the CMAPSS dataset and place CMAPSSdata.zip in data/raw/.

Run notebooks

Execute notebooks in order from the notebooks/ directory:

01_data_exploration.ipynb
02_feature_engineering.ipynb
03_model.ipynb
04_business_conclusion.ipynb

MLflow UI

mlflow ui

Open http://127.0.0.1:5000 to explore all tracked experiments and compare runs.

---

Technical Stack

Category	Tools
Data manipulation	pandas, numpy
Machine learning	scikit-learn, XGBoost
Experiment tracking	MLflow
Visualization	matplotlib, seaborn
Model persistence	joblib

---

Author

elouanXP