Few-shot Learning Fault Diagnosis Method Based on Local Descriptor
Due to its advantages in feature description,deep learning has been applied to many fields in recent years.However,in the field of fault diagnosis,it is difficult to obtain a large number of fault samples to train models.To solve this problem,this paper proposes a bearing fault diagnosis method based on local descriptors according to localized feature descriptions,and uses the improved Deep Nearest Neighbor Neural Network(DN4)for fault diagnosis.Firstly,the vibration signal is transformed into a two-dimensional time-frequency image using a short-time Fourier transform.The ResNet-12 network is employed to extract the backbone network features,and Efficient Multi-scale Attention(EMA)is implemented to enhance the extraction of fault features.The local descriptor of bearing faults is obtained.Finally,the fault category is obtained using the K-nearest neighbor algorithm.To ascertain the efficacy of the proposed methodology,tests of bearing fault diagnosis were conducted with varying numbers of samples and encompassing diverse working conditions.The results of the test demonstrate that the proposed method exhibits a favorable effect on fault identification and generalization performance based on few-shot learning,as well as possessing certain engineering application value.
万方数据
维普
