A health indicator construction method of rolling bearing based on vibration image features and deep autoencoder network

Rolling bearing is a key component and weak link of rotating machinery, the construction of health indicator (HI) with good performance can provide support for the fault status assessment and fault prediction of rolling bearing. Traditional HI construction methods of rolling bearings are mainly limi...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:2023 5th International Conference on System Reliability and Safety Engineering (SRSE) s. 268 - 273
Hlavní autori: Duan, Yong, Cao, Xiangang, Zhao, Jiangbin, Zhang, Ruiyuan, Yang, Xin, Guo, Xingyu
Médium: Konferenčný príspevok..
Jazyk:English
Vydavateľské údaje: IEEE 20.10.2023
Predmet:
autoencoder
Employee welfare
Feature extraction
Gray-scale
health indicator
image features
Reliability
remaining useful life prediction
rolling bearing
Rolling bearings
Safety
Vibrations
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Rolling bearing is a key component and weak link of rotating machinery, the construction of health indicator (HI) with good performance can provide support for the fault status assessment and fault prediction of rolling bearing. Traditional HI construction methods of rolling bearings are mainly limited to one-dimensional feature extraction, and it is difficult to consider local and global features at the same time, which leads to poor nonlinear representation ability of HI. To solve the above problems, this paper proposes a rolling bearing health indicator construction method based on vibration image features and deep autoencoder network. Firstly, the one-dimensional vibration signal is reconstructed into a time-domain grayscale image, and the optimal image features are extracted and selected to reflect the vibration information in the two-dimensional image. Then, the optimal features are arranged and coupled to improve the information expression. Finally, a deep convolutional autoencoder network (DCAE) method with self-attention mechanism is proposed to construct the HI indicator unsupervised. The effectiveness of the method is verified by the XJTU-SY bearing dataset. The average comprehensive score of HI of different bearings is 0.8188, which is 0.2051 higher than that of other methods, and has better performance of remaining useful life prediction.
DOI:10.1109/SRSE59585.2023.10336116