Bearing fault diagnosis based on prototypical network

Aiming at the problem that the accuracy of conventional algorithms is low in the case of few samples for bearing vibration signal fault diagnosis, this paper proposes a bearing fault diagnosis method based on prototypical network in few-shot and zero-shot scenarios. The method first uses the original vibration signals or spectrogram features as input; then uses the neural network model to extract the distinguishable features, and prototype center of each category is learned through prototypical network; finally, the classification of each sample is completed by the distance measurement method. The experimental results show that prototypical network method with scaled CQT features as input and convolutional neural network as encoder has excellent performance in few-shot and zero-shot bearing fault diagnosis.