Research On Bearing Fault Diagnosis Technology Based On Local And Global Deep Neural Network

In the current industrial information age,it is a major challenge to effectively obtain equipment status information to ensure its efficient safe production and fault maintenance.In the field of fault diagnosis,global information features and local information features respectively represent the overall structure of the data and the neighborhood relationship,and the local and global information of the data structure should be considered.The single feature extracted by traditional machine learning algorithms is highly subjective and has a low degree of discrimination.Aiming at the problem that traditional machine learning algorithms cannot coordinate the local and global fault feature information of bearings,this paper proposes a local global deep neural network algorithm.New type of bearing fault diagnosis method.As a compromise,this algorithm combines the local feature extractor with the global feature extractor to diagnose the bearing fault.First,the original vibration spectrum signal of the bearing was expressed through double sparse expression to reduce noise and dimensionality of the data while eliminating the phenomenon of data imbalance and reducing the redundant attributes of the data.Second,the local feature extraction layer uses the improved convolutional deep belief network to effectively extract the local discrimination of the data.The information effectively identifies the fault category,and then the global feature extraction layer uses the kernel principal component analysis to extract the global discriminative information of the data,and finally the extracted comprehensive features were classified into different faults using the soft-max classifier.Based on the characteristics of open source,the bearing vibration signal data of Western Reserve University was selected as samples for fault classification to verify the effectiveness of the method and compared with existing methods.Observation and diagnosis results show that the proposed fault diagnosis method can take into account the partial and the global fault feature information effectively classifies bearing faults.The selection of laboratory bearing vibration signal data proves that the model has strong generalization ability.