Remaining-useful-life-prediction-by-TaFCN-in-CMAPSS-dataset

Trend attention fully convolutional network for remaining useful life estimation in the turbofan engine PHM of CMAPSS dataset. Signal selection, Attention mechanism, and Interpretability of deep learning are explored.

Paper: Trend attention fully convolutional network for remaining useful life estimation

The website of the paper：https://www.sciencedirect.com/science/article/pii/S0951832022002356

Easy to reproduce all details of our paper successfully

The downloaded compressed package can reproduce all the details of our article, including almost all the experimental tables and figures. To make code easy to run successfully, we debug the files carefully. Generally speaking, if environments are satisfied, you can directly run all the xxx.py files inside after decompressing the compressed package without changing any code.

(1) Download and rename Remaining-useful-life-prediction-by-LM-and-TaFCN-main.zip to TaFCN.zip (Rename to avoid errors caused by long directories)

(2) Unzip TaFCN.zip, then unzip fd.rar in it

(3) Run any xxx.py directly

Paper of Code and Citation

(1) To better understand our code, please read our paper.

Paper: Trend attention fully convolutional network for remaining useful life estimation

The website of the paper：https://www.sciencedirect.com/science/article/pii/S0951832022002356

(2) Please cite this paper and the original source of the dataset when using the code for academic purposes.

GB/T 7714:

Fan L, Chai Y, Chen X. Trend attention fully convolutional network for remaining useful life estimation[J]. Reliability Engineering & System Safety, 2022: 108590.

BibTex:

@article{fan2022trend, title={Trend attention fully convolutional network for remaining useful life estimation}, author={Fan, Linchuan and Chai, Yi and Chen, Xiaolong}, journal={Reliability Engineering & System Safety}, pages={108590}, year={2022}, publisher={Elsevier} }

Relationship between Code and Paper

(1) Section 2.2. Loss boundary to mapping ability

:code\signal selection

(2) Section 2.3. Trend attention fully convolutional network

:code\main(grid_FD_multi_channel_one_FCN_RUL_TaNet_attention_1out_all_train_for_test).py

(3) Section 4. Interpretability

:code\figure\interpretability_analysis.py

(4) Fig. 6. Attention analysis of TaNet.

:code\figure\interpretability_TaNet_analysis.py

(5) Fig. 7. Accumulated prediction error over RUL.

:code\figure\prediction_error.py

(6) Fig. 5. Wilcoxon signed rank test comparison of eight combinations

:code\figure\heatmap_p.py and code\table\wilcxon.py

Environment and Acknowledgement:

(1) Environment:

tensorflow-gpu 1.15.0

keras 2.2.4

scipy 1.5.2

pandas 1.0.5

numpy 1.19.1

(2) Acknowledgement: Thanks for the following references sincerely.

github：https://github.com/Vardoom/PredictiveMaintenanceNASA/blob/master/preprocess.ipynb

github：https://github.com/schwxd/LSTM-Keras-CMAPSS

github：https://github.com/cauchyturing/UCR_Time_Series_Classification_Deep_Learning_Baseline