Hits:

Affiliation of Author(s):能源与动力学院

Title of Paper:A Novel Feature Representation Method Based on Deep Neural Networks for Gear Fault Diagnosis

Journal:2017 PROGNOSTICS AND SYSTEM HEALTH MANAGEMENT CONFERENCE (PHM-HARBIN)

Key Words:fault diagnosis stacked autoencoder softmax regression visualization

Abstract:Intelligent diagnosis method has been a hot topic in the prognostics and fault diagnosis of rotating machinery. A deep neural network (DNN) method for gear fault diagnosis based on stacked autoencode (SAE) and softmax regression is presented in this work. SAE is first utilized to extract features from the frequency spectra of vibration signal. And then the learned features are used to train softmax regression classifier for identifying gear faults. The diagnosis results validate that the method is feasible and is able to acquire superior classification performance. With the increasing layers of DNN, the learned features of each hidden layer are becoming more and more robust for classification. In this paper, a novel view on the visualization of the learned features is displayed to evaluate the DNN-based fault diagnosis method.

ISSN No.:2166-5656

Translation or Not:no

Date of Publication:2017-01-01

Co-author:王金瑞,江星星,辛玉

Correspondence Author:lsz