Intelligent Fault Diagnosis of Rotating Machines Based on Wavelet Time-Frequency Diagram and Optimized Stacked Denoising Auto-Encoder

When a stacked denoising auto-encoder (SDAE) manually sets several parameters, the gradient of neuron weight becomes dispersed, reducing the ability to retrieve sensitive fault feature information from a bearing vibration signal under multiple working conditions and strong noise. A bearing health mo...

Full description

Saved in:
Bibliographic Details
Published in:IEEE sensors journal 2022-09, Vol.22 (17), p.17139-17150
Main Authors: Jia, Ning, Cheng, Yao, Liu, Yunyang, Tian, Youyuan
Format: Article
Language:English
Subjects:
Coders
Continuous wavelet transform
Continuous wavelet transforms
Diagnostic systems
Fault diagnosis
Feature extraction
Mathematical models
Noise monitoring
Noise reduction
Optimization
Parameters
Rotating machinery
Rotating machines
Search algorithms
sparrow search algorithm
stacked denoising auto-encoder
the wavelet time-frequency diagram
Time-frequency analysis
Vibrations
Wavelet transforms
Weight reduction
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:When a stacked denoising auto-encoder (SDAE) manually sets several parameters, the gradient of neuron weight becomes dispersed, reducing the ability to retrieve sensitive fault feature information from a bearing vibration signal under multiple working conditions and strong noise. A bearing health monitoring and defect diagnostic model based on variational mode decomposition (VMD) combined with continuous wavelet transform (CWT) and SDAE optimized by sparrow search algorithm (SSA) is presented to tackle this problem. The wavelet time-frequency diagram is obtained by VMD and CWT, which maps the fault characteristic information to different local positions in time and scale, and then the wavelet time-frequency diagrams are input into the SDAE for in-depth training. To achieve the ideal structure of SDAE and increase the feature extraction capabilities of SDAE for weak signals, SSA is utilized for the global combination and adaptive selection of several SDAE parameters. The bearing failure diagnostic model based on VMD-CWT-SSA-SDAE outperforms BPNN, SVM, the traditional SDAE, GA-SDAE, PSO-SDAE, and SSA-DBN in diagnosis accuracy, generalization performance, and anti-noise performance when tested on various data sets.
ISSN:1530-437X
1558-1748
DOI:10.1109/JSEN.2022.3193943