Stack Denoising Autoencoder and State-Space Model Based Bearing RUL Prediction Method

Rolling element bearing is a critical component in a machinery, so its remaining useful life (RUL) prediction becomes a research hotspot in recent years. However, the RUL prediction has two difficulties: 1) effective health indicator (HI) is hard to be constructed; 2) degradation initial time is dif...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:2022 International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD) s. 1 - 6
Hlavní autori: Yang, Lei, Xu, Cuiwentong, Duan, Rongkai, Liao, Yuhe
Médium: Konferenčný príspevok..
Jazyk:English
Vydavateľské údaje: IEEE 30.11.2022
Predmet:
Analytical models
Degradation
Noise reduction
Paris-Erdogan model
particle filter
Predictive models
remaining useful life
Sensitivity
Smoothing methods
stack denoising autoencoder
state-space model
Uncertainty
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Rolling element bearing is a critical component in a machinery, so its remaining useful life (RUL) prediction becomes a research hotspot in recent years. However, the RUL prediction has two difficulties: 1) effective health indicator (HI) is hard to be constructed; 2) degradation initial time is difficult to be identified accurately. To solve the problems, a RUL prediction method based on stack denoising autoencoder (SDA) and non-overlapping sliding window (NOSW) threshold method is proposed in this work. The HI is constructed by the SDA from 19 time-domain features, which balances the sensitivity and robustness of different features. A novel NOSW threshold method is used to identify the degradation initial time and divide the life cycle into normal operating stage and degradation stage. A state-space model based on the Paris-Erdogan model is established and its noise intensity is estimated by a smoothing estimation method. The particle filtering is employed to track the degradation path and quantify the uncertainty of RUL prediction. In order to verify the effectiveness of the proposed method, two real bearing datasets of XJTU-SY are utilized to experiment. The results demonstrate the proposed method can track the degradation path well and predict the RUL accurately.
DOI:10.1109/ICSMD57530.2022.10058208