A Semi Supervised Domain Generalized Fault Diagnosis Network Based on a Small Amount of Labeled Data

In recent years, deep learning has played an increasingly important role in the fault diagnosis of rotating machinery. However, its application in industrial fault diagnosis is limited by the need for a large number of labeled samples. Furthermore, actual fault conditions are often unseen, which further constrains the development and application of intelligent diagnosis. In this study, we propose a multi-domain semi-supervised and target domain generalization fault diagnosis network. This method uses labeled source domain vibration data to generate pseudo-labels for multiple unlabeled source domain vibration data through a semi-supervised multi-branch network. By fully leveraging existing data, it further performs knowledge transfer diagnosis for unseen target domains, achieving domain generalization with a minimal amount of labeled data. Experimental results on public bearing datasets and simulated data demonstrate that the proposed method performs well in semi-supervised domain generalization fault diagnosis with few labels, reducing dependence on labeled samples from the target domain and showing good potential for engineering applications.