Fault diagnosis method for gearboxes based on deep convolutional binary decomposition network
LIN Huibin
XIAN Xianzhao
HE Guolin
Abstract：[Objective]To address the issue of harmonic interference affecting local fault feature in gearbox fault diagnosis,a harmonic separation and impact feature extraction method based on a deep convolutional binary decomposition network(DCBDN)was proposed.[Methods]Firstly,by improving the feature transmission and output patterns of the stacked autoencoder network,a separation constraint for harmonic components was introduced to achieve harmonic separation and impact fault feature extraction during the network's feature propagation process.Subsequently,a binary output network training approach grounded in a fault mechanism model was developed for the proposed network.A simulated dataset was constructed based on the fault mechanism model,and the parameters of both harmonic and impact feature extractors within the model were dynamically updated via effect compensation,thereby completing network training.[Results]Simulation and test analyses demonstrate that compared with existing convolutional autoencoder models and fast spectral kurtosis methods,the proposed method effectively separates coupled harmonic and fault impact components,exhibiting superior anti-interference capability and enhanced local fault feature extraction performance.
Keywords：Deep convolutional binary decomposition networkGearboxImpact feature extractionHarmonic separation
Publication Date：2025-09-15
Online Publishing Date：2026-09-12(First online date of this platform, not the publication date of the document)
Pages：10( 119-127,135 )