Open AccessHydro-generators fault diagnosis with short-time-wavelet-entropy and variational auto-encoder
Author(s) -
Ryad Zemouri,
Simon Bernier,
Olivier Kokoko,
Arezki Merkhouf
Publication year - 2021
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/1207/1/012009
Subject(s) - wavelet , stator , entropy (arrow of time) , computer science , artificial intelligence , magnetic flux , discrete wavelet transform , pattern recognition (psychology) , wavelet transform , algorithm , mathematics , engineering , physics , magnetic field , electrical engineering , quantum mechanics
The prognosis and health management (PHM) of hydroelectric plants are full of difficulties caused by the complexity of the hydro-generators where each machine is different and almost unique. At industrial level, several tools are used to monitor the generator condition. Among these tools, the measurement of magnetic stray flux is one which is gaining interest. This measurement is generally based on an inductive sensor and mainly mounted near the stator. The main advantages of the magnetic stray flux are the non-invasive nature and the simplicity of its implementation. In this work, the discrete wavelet transform (DWT) is used to decompose the stray flux signal. Short-Time-Wavelet-Entropy (STWE) is then applied to extract the features from the sub-bands. Finally, a variational auto-encoder (VAE) is used in an unsupervised learning process to structure the STWE signatures of more than 400 stray flux measurement collected on real hydroelectric plants. The obtained results show that the VAE has well captured the features from the wavelet entropy (WE) signatures. An analysis of the resulting latent space shows a strong correlation between a given trajectory in the reduced space and an increase of the WE.