Multiple fault diagnosis for rolling bearings method employing CEEMD-GCN based on horizontal visibility graph

Abstract

Multiple faults often occur in the operation of rotating machinery transmission systems. The fault signals of multiple bearings interfere with each other, which makes feature extraction and diagnosis of complex compound fault signals difficult. Because the graph convolution networks (GCN) can effectively map the structural information from complex data and its model has a certain generalization ability, this paper proposes a multiple fault diagnosis method for rolling bearings employing complete ensemble empirical mode decomposition (CEEMD) and a GCN (CEEMD-GCN) based on a horizontal visibility graph (HVG). Firstly, in order to highlight the effective feature information in the multiple fault signal and reduce noise interference, multiple indicators of correlation and kurtosis are used to reconstruct the decomposed signals through CEEMD; secondly, the reconstructed signals are constructed as an HVG, and the HVG maps the time series signal to the graphic structure data, reflecting the local geometric characteristics of the vibration signal through the horizontal visibility relationship; finally, taking the signal samples obtained by the HVG algorithm as the input data of the model, the GCN model is trained to realize the diagnosis of multiple faults. The experimental results show that the presented methodology is superior to other methods and exhibits generalization ability for multiple fault diagnosis.