中文
This is a project for mechanical failure detection system. The reason why it is called sparrow may be because the eyesight of sparrow is particularly good, and they can detect small bugs. Besides, they are cute too. In the current version, we have tested some planetary gearbox vibration data in the lab, and there are 8 error types in total. And use Capsule Neural Networks for pattern recognition to detect error types in vibrations and it works really well.
- We get some vibration data in the lab
- And we have enough reasons to believe that the damage of the internal structure of the machine can be manifested in the vibration frequency. The frequency information of the signal can be obtained by using discrete Fourier transform or short-time Fourier transform.
- Obviously, we can turn a one-dimensional data into a graph and input it into the neural network
- Obviously, the capsule neural network proposed by Hinton has good performance. So we choose to use this schema to process the data
The curve of accuracy and loss in the training process is as follows
How to still have a high fault detection accuracy in a high-noise environment is a problem. Using wavelet transform might be a way.
git clone https://github.com/ecstayalive/Sparrow
git submodule --init --recursive
python main.py
@article{sabour2017dynamic,
title={Dynamic Routing Between Capsules},
author={Sabour, Sara and Frosst, Nicholas and Hinton, Geoffrey E},
journal={arXiv preprint arXiv:1710.09829},
year={2017}
}