Why does fault diagnosis need deep learning?


Fault diagnosis entry-level players ask a question, vibration signal analysis directly do spectrum analysis is good, why need artificial intelligence? – amaze2 answer – knowhttps://www.zhihu.com/question/332473558/answer/1349385215

For simple mechanical equipment, such as one-stage parallel gearbox, the signal is relatively simple, direct spectrum analysis can meet the demand.

However, if it is a complex equipment, such as multi-stage planetary gear transmission, strong environmental noise, and the fault is in the early stage and relatively weak, the failure frequency may not be found on the spectrum.

What should I do?

Deep learning provides an idea.

By the way, a deep learning method for fault diagnosis under strong noise, residual shrinkage network, is recommended.

The residual shrinkage network uses a soft threshold function in its interior, which is similar to wavelet threshold denoising. In the deep learning model, the noise information is automatically eliminated to obtain more accurate fault features.

Why does fault diagnosis need deep learning?

M. Zhao, S. Zhong, X. Fu, B. Tang, M. Pecht, Deep residual shrinkage networks for fault diagnosis, IEEE Transactions on Industrial Informatics, vol. 16, no. 7, pp. 4681-4690, 2020.