Fault diagnosis method for rotating machinery based on topology perception and dual-view classifier
Here,aiming at the problem of different data distributions under different operating conditions of rotating machinery and uneven sample categories due to scarce fault data to cause performance degradation of fault diagnosis model,a fault diagnosis method based on topology perception and dual-view classifier was proposed.This method could take a graph convolutional network(GCN)as diagnosis framework,and the proposed non-parametric topology perception module could adaptively update the graph data topology structure,constrain different domain data to obtain approximate message transmission paths,and effectively extract domain consistent fault features with GCN.Binary and multiple classifiers were used to construct a dual-view classifier,calculate the similarity between binary and multivariate outputs to reweight training data,and avoid biased training of model under category imbalance and weak recognition ability for minority class samples.Experiments were conducted using publicly available gear fault datasets of Xi'an Jiaotong University,MAFAULDA rotating machinery fault datasets and self-made sliding bearing fault simulation data.The results showed that the proposed method can effectively improve fault diagnosis performance under variable operating conditions and category imbalance.