Bearing fault diagnosis method based on improved Siamese neural network with small sample

Fault diagnosis of rolling bearings is very important for monitoring the health of rotating machinery. However, in actual industrial production, owing to the constraints of conditions and costs, only a small number of bearing fault samples can be obtained, which leads to an unsatisfactory effect of traditional fault diagnosis models based on data-driven methods. Therefore, this study proposes a small-sample bearing fault diagnosis method based on an improved Siamese neural network (ISNN). This method adds a classification branch to the standard Siamese network and replaces the common Euclidean distance measurement with a network measurement. The model includes three networks: a feature extraction network, a relationship measurement network, and a fault classification network. First, the fault samples were input into the same feature extraction network in pairs, and a long and short-term memory (LSTM) network and convolutional neural network (CNN) were used to map the bearing signal data to the low-dimensional feature space. Then, the extracted sample features were measured for similarity by the relationship measurement network; at the same time, the features were input into the classification network to complete the bearing fault recognition. When the number of training samples was particularly small (training set A, 10 samples), the accuracy of 1D CNN, Prototype net and Siamese net were 49.8%, 60.2% and 58.6% respectively, while the accuracy of the proposed ISNN method was 84.1%. For the 100-sample case of training set D, the accuracy of 1D CNN was improved to 93.4%, which was still higher than that of prototype and Siam network, while the accuracy of ISNN method reached 98.1%.The experimental results show that the method in this study achieved higher fault diagnosis accuracy and better generalization in the case of small samples.