Remaining Useful Life (RUL) prediction for Turbofan Engines

The project objective is to estimate Remaining Useful Life (RUL) with Machine Learning. RUL is defined as a remaining time that a component can function with an acceptable performance before it fails. The model should provide prediction uncertainties and be extensively tested with Prognostics metrics.

Turbofan Engines Degradation Simulation Data Set provided by NASA is being used in this project for Remaining Useful Life prediction.

See the notebooks descriptions:

1. 1-EDA.ipynb - contains Exploratory Data Analysis. The objective is to inspect the data, get general understanding of the datasets, check for patterns, understand what preprocessing should be done before the modeling.
2. 2-target-metrics-baseline.ipynb - define the target, discuss evaluation metrics, and build a baseline model.
3. 3-features_engineering.ipynb - features engineering pipelines.
4. 4-predict_rul_with_ML.ipynb - contains experiments with different algorithms, features sets and evaluation with prognostics metrics.

Setup

Environment setup:

python3.8 -m venv ./venv
source venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt
python -m ipykernel install --user --name=venv

Download the dataset:

wget -O CMAPSSData.zip  https://ti.arc.nasa.gov/c/6/
unzip -d CMAPSSData/ CMAPSSData.zip

References

1. A. Saxena, K. Goebel, D. Simon, and N. Eklund, "Damage Propagation Modeling for Aircraft Engine Run-to-Failure Simulation", in the Proceedings of the Ist International Conference on Prognostics and Health Management (PHM08), Denver CO, Oct 2008

2. Saxena, A.; Celaya, J.; Saha, B.; Saha, S.; Goebel, K. Metrics for Offline Evaluation of Prognostic Performance. Int. J. Progn. Health Manag. 2010

3. https://www.researchgate.net/publication/224358896_Recurrent_neural_networks_for_remaining_useful_life_estimation

4. Christ, M., Kempa-Liehr, A.W. and Feindt, M. (2016). Distributed and parallel time series feature extraction for industrial big data applications. ArXiv e-prints: 1610.07717 URL: http://adsabs.harvard.edu/abs/2016arXiv161007717C

5. Dempster A., Petitjean F. and Webb G., "ROCKET: Exceptionally fast and accurate time series classification using random convolutional kernels": https://arxiv.org/pdf/1910.13051.pdf