关于数据预处理的问题

qinenergy / DA-diagnosis

Code for our paper "Domain adaptive transfer learning for fault diagnosis." 2019 Prognostics and System Health Management Conference (PHM-Paris). IEEE, 2019.

MIT License

20 stars 1 forks source link

关于数据预处理的问题 #2

Open echo-che opened 2 years ago

echo-che commented 2 years ago

您好您在代码中给出的源域和目标域数据经过归一化处理吗为什么我的数据归一化处理后的准确率和您的相差10%个点

qinenergy commented 2 years ago

Hi the data is not normalized. Depending on your normalization technique, information might get lost. It also may affect the hyper-parameter needed for the domain adaptation.

echo-che commented 2 years ago

您好我有一个疑问就是必须给数据做FFT处理吗为什么不做fft的准确率很低（大概只有0.34）做fft的好处是什么呢

qinenergy commented 2 years ago

The current network architecture is designed to take in FFT data. Most bearing faults can be identified easier in the frequency space, this is also what human experts do in real life. There are papers which target directly deal with the raw data, but it is out of the scope of this paper.