Bearing remaining useful life prediction based on deep autoencoder and deep neural networks

•We propose a novel eigenvector based on time–frequency-wavelet joint features.•We present an autoencoder joint features, which retain the DNN scale.•We present a new deep learning prediction framework for bearing RUL. Bearings play a crucial part in reliable operation of rotating machinery in manuf...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Journal of manufacturing systems Jg. 48; S. 71 - 77
Hauptverfasser: Ren, Lei, Sun, Yaqiang, Cui, Jin, Zhang, Lin
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Ltd 01.07.2018
Schlagworte:
Deep learning
Industrial big data
Prognostics
Remaining useful life prediction
Smart manufacturing
99-00
Deep learning
Remaining useful life prediction
00-01
Industrial big data
Prognostics
Smart manufacturing
ISSN:0278-6125, 1878-6642
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:•We propose a novel eigenvector based on time–frequency-wavelet joint features.•We present an autoencoder joint features, which retain the DNN scale.•We present a new deep learning prediction framework for bearing RUL. Bearings play a crucial part in reliable operation of rotating machinery in manufacturing systems. There is a growing demand for smart prognostics of bearing remaining useful life (RUL). The data driven approach for bearing RUL prediction has shown promising potential to support smart prognostics. The recent advances in deep learning and industrial big data provide new solutions for data driven bearing RUL prediction yet still face significant challenges, e.g. optimal feature selection and efficient feature compression. This paper proposes a new deep learning based prediction framework for bearing RUL by using deep autoencoder and deep neural networks (DNN). A novel eigenvector based on time–frequency-wavelet joint features is proposed to effectively represent bearing degradation process. A deep autoencoder based joint features compression and computing method is presented to retain effective information without increasing the scale of DNN. The experiment results showed that the proposed method can achieve better efficiency in bearing RUL prediction.
ISSN:0278-6125
1878-6642
DOI:10.1016/j.jmsy.2018.04.008