Industrial fault diagnosis based on few shot learning

In fault diagnosis, traditional machine learning methods usually require manual feature selection, which requires higher professional background and multiple attempts. Although deep learning avoids the selection of features, it needs a large amount of data. In industrial systems, it is usually difficult and expensive to obtain fault data, so it is not suitable for the fault diagnosis task with few data. In this paper, we use the idea of Few Shot Learning for fault diagnosis to avoid the difficulty of manual feature selection, and it is suitable for the fault diagnosis field with few samples. For the fault tasks to be diagnosed, this paper uses additional data to obtain prior knowledge. On the basis of the prototype network, we use one-dimensional convolutional neural network to extract the signal features of the fault signals, and complete the final classification with KNN. In order to alleviate the problem of few samples size, we also introduced the method of data enhancement to improve the fault diagnosis rate to a certain extent. The experimental results show that in the case of few samples, proposed few shot learning fault diagnosis method in this paper obtains good diagnosis results.