A multi-task learning for cavitation detection and cavitation intensity recognition of valve acoustic signals

[Project Page] || [arXiv]

Requirements

• Python 3.8.11
• torch 1.9.1
• torchvision 0.10.1

Note: our model is trained on NVIDIA GPU (A100).

Code execution

• train.py is the entry point to the code.
• main.py is the main function of our model.
• network.py is the network structure of 1-D double hierarchical residual block (1-D DHRB) and 1-D double hierarchical residual network (1-D DHRN).
• opts.py is all the necessary parameters for our method (e.g. learning rate and data loading path and so on).
• Execute train.py

Note that, for the current version. test.py is nor required as the code calls the test function every iteration from within to visualize the performance difference between the baseline and the our method.

• Download trained models from here.
• Download datasets from here.

Updates

[01.03.2022] We submit preprinted versions on the arXiv.

[08.03.2022] We upload all source codes for our method. And we add a link to our paper on the arXiv.

[19.04.2022] Our paper was accepted by the [Engineering Applications of Artificial Intelligence (EAAI)]. The Impact Factor of this journal is 6.212, ranking it 7 out of 91 in Engineering, Multidisciplinary.

[10.05.2022] Our paper is now available online. And the link is here.

For any queries, please feel free to contact YuSha et al. through yusha20211001@gmail.com

Citation

If you find our work useful in your research, please consider citing:

@article{SHA2022104904,
title = {A multi-task learning for cavitation detection and cavitation intensity recognition of valve acoustic signals},
author = {Yu Sha and Johannes Faber and Shuiping Gou and Bo Liu and Wei Li and Stefan Schramm and Horst Stoecker and Thomas Steckenreiter and Domagoj Vnucec and Nadine Wetzstein and Andreas Widl and Kai Zhou},
journal = {Engineering Applications of Artificial Intelligence},
volume = {113},
pages = {104904},
year = {2022},
}
'''