Research On Fault Diagnosis Of Gearbox Under Variable Conditions With Small Sample

Aiming at the problem of gearbox fault diagnosis which is a small sample problem,an improved CVAE method is proposed to increase the number of fault samples.Compared with the traditional CVAE,the improved CVAE adds a class separator to the network structure.This class separator can ensure that the network generates specific types of fault samples and reduce the difference between the fault characteristics output by the class separator network layer.So that the reconstruction loss of the generated data is further reduced.Experiments show that this method can not only increase the number of fault samples,but also increase the diversity of fault samples.On the premise of accumulating certain historical data,the fault diagnosis model trained based on the data enhancement method will reduce the diagnosis accuracy of the model under variable working conditions,and a deep fault diagnosis network based on difference features is proposed.The difference feature is selected by the maximum mean error,and then migration methods such as the maximum error of the multi-core mean and the domain category classifier are used to reduce the difference between the difference features under different working conditions,thereby improving the fault diagnosis accuracy of the model under variable working conditions.Experiments prove the effectiveness and superiority of this method.For only a few fault samples,of data enhancement and other methods to improve the accuracy of fault diagnosis is not obvious,a lightweight fault diagnosis network based on feature metrics is proposed.The feature extraction network in the fault diagnosis model extracts the fault features of the fault data.The feature matching network learns a measurement standard to measure the similarity between different fault features,and sends the known fault samples and unknown fault samples into the fault diagnosis model through the features.For fault diagnosis,the experiment proves that the method can achieve ideal fault diagnosis accuracy under several fault samples.Aiming at the current situation of different fault diagnosis problems with different numbers of fault samples,a fault diagnosis method is proposed,which can achieve better fault diagnosis accuracy under the condition of different numbers of fault samples.The fault diagnosis method in the above research is not a fault diagnosis model customized for a certain mechanical gearbox,but for similar rotating machinery gearboxes,the method researched in this article can be used for fault diagnosis,so it has strong practical application value.