AAU-Net
-
We provide the improved code for paper "Adversarial Algorithm Unrolling Network for Interpretable Mechanical Anomaly Detection".
-
Compared with the original Code[https://github.com/Botao-An/AAU_Net], Zhi Cao[cyfbxl1@gmail.com] improved the loss functions of AAU-Net and solved the problem of instability during training. The referred paper is "Wasserstein Auto-Encoders"
-
The code is originally debugged on the computer with the following configuration.
Hardware Software Software Intel Core i7-10700KF CUP Anaconda 4.9.2 CUDA 11.7 RAM 32GB Python 3.8 cuDNN 8.4.0 NVIDIA GeForce RTX 3080 GPU PyTorch 1.8.1 TorchVision 0.9.1 -
We give a demo of simulation for you to test the performance of AAU-Net initially.
-
To run the model, you should firstly prepare the dataset in a standard format as the simulated one. More details can be found in file pre and post processing.ipynb. Then you can run AAU-Net by file train.py. To check and visulize the results, you can use pre and post processing.ipynb again.
-
If you want do research based this code, please cite the following papers. For any questions you can contact e-mail: cyfbxl1@gmail.com or Albert_An@foxmail.com.
@article{an2023interpretable, title={Interpretable Neural Network via Algorithm Unrolling for Mechanical Fault Diagnosis}, author={An, Botao and Wang, Shibin and Zhao, Zhibin and Qin, Fuhua and Yan, Ruqiang and Chen, Xuefeng}, journal={IEEE Transactions on Instrumentation and Measurement}, volume={71}, pages={1--11}, year={2023}, publisher={IEEE} } @article{an2022adversarial, title={Adversarial Algorithm Unrolling Network for Interpretable Mechanical Anomaly Detection}, author={An, Botao and Wang, Shibin and Qin, Fuhua and Zhao, Zhibin and Yan, Ruqiang and Chen, Xuefeng}, journal={IEEE Transactions on Neural Networks and Learning Systems}, volume={}, pages={}, year={2022}, publisher={IEEE} }