🔧 Predictive Maintenance with NASA C-MAPSS Dataset

This project implements predictive maintenance using the NASA C-MAPSS Turbofan Engine Degradation Datasets (FD001–FD004). The goal is to predict the Remaining Useful Life (RUL) of engines before failure.

We benchmark multiple models, from classical ML to deep learning (LSTM, GRU, Transformer).

📂 Project Structure predictive-maintenance-cmapss/ │ ├── data/ │ ├── raw/ # Original C-MAPSS dataset (train/test/RUL txt files) │ └── processed/ # Preprocessed pickles (.pkl) generated by preprocessing │ ├── models/ # Saved trained models (.keras) ├── results/ # Model evaluation outputs (CSV, plots, reports) │ ├── final_report.pdf # Auto-generated summary report │ └── final_report.md # Markdown summary of results │ ├── notebooks/ # Jupyter notebooks for exploration │ ├── 01_exploration.ipynb │ ├── 02_preprocessing.ipynb │ ├── 03_baseline_models.ipynb │ ├── 04_deep_learning_models.ipynb │ ├── 05_other_dl_models.ipynb │ └── 07_transformer_fd001.ipynb │ ├── src/ # Training, evaluation, preprocessing scripts │ ├── 02_preprocessing.py │ ├── train.py │ ├── evaluate.py │ ├── aggregate_results.py │ └── utils.py │ ├── run_all.bat # Full automation: preprocess → train → evaluate → report ├── requirements.txt ├── .gitignore └── README.md

📊 Results

Performance is reported across all four subsets (FD001–FD004):

Model	Dataset	RMSE	MAE	R²
LSTM	FD001	63.61	50.67	-0.358
LSTM	FD002	68.73	54.47	-0.339
LSTM	FD003	87.25	63.52	-0.162
LSTM	FD004	102.05	76.52	-0.311
GRU	FD001	63.62	50.67	-0.358
GRU	FD002	68.37	54.17	-0.325
GRU	FD003	87.21	63.49	-0.161
GRU	FD004	102.11	76.57	-0.313
Transformer	FD001	47.59	36.94	0.240
Transformer	FD002	52.94	41.11	0.206
Transformer	FD003	69.67	53.07	0.259
Transformer	FD004	80.61	59.30	0.182

📄 See the final_report.pdf for full details with tables & charts.

📥 Dataset Setup

The dataset comes from NASA’s C-MAPSS Turbofan Engine Degradation Simulation.

🔗 Download here: NASA Prognostics Data Repository

After downloading:

Extract files into data/raw/

data/raw/ ├── train_FD001.txt ├── test_FD001.txt ├── RUL_FD001.txt ├── train_FD002.txt ├── ...

Run preprocessing:

python src/02_preprocessing.py

This will generate .pkl files under data/processed/.

⚙️ Installation

Clone repo

git clone https://github.com/aun151214/predictive-maintenance-cmapss.git cd predictive-maintenance-cmapss

Create environment

python -m venv .venv .venv\Scripts\activate # Windows

source .venv/bin/activate # Linux/Mac

Install dependencies

pip install -r requirements.txt

🚀 Usage 🔹 Train a Model python src/train.py --model lstm --dataset FD001 --epochs 100 --batch_size 64 python src/train.py --model gru --dataset FD002 --epochs 100 python src/train.py --model transformer --dataset FD004 --epochs 100

🔹 Evaluate a Model python src/evaluate.py --model lstm --dataset FD001

🔹 Run Full Pipeline (all models, all datasets, auto-report) .\run_all.bat

This will:

Preprocess datasets

Train & evaluate all models on FD001–FD004

Save trained models in models/

Save results in results/

Auto-generate final_report.pdf & final_report.md

📈 Key Insights

Transformer performed best overall on FD001–FD004 (positive R²).

LSTM and GRU underperformed on FD002–FD004 in this run, suggesting tuning/data augmentation needed.

Pipeline is fully automated → reproducible for any new dataset.

📜 License

This project is released under the MIT License.