Universal domain adaptation for machinery fault diagnosis based on multi-scale dual attention network and entropy-based clustering

Recently, data-driven cross-domain fault diagnosis methods for rotating machinery have been successfully developed. However, most existing diagnostic methods assume that the label spaces of the source and target domains are the same. In practice, the relationship between the label space of the source domain and the target domain is unknown, that is, the universal domain adaptation (UDA) problem. Existing overall domain distribution alignment methods are less effective in facing UDA problems. Thus, this article proposes a deep learning-based UDA model. First, the proposed model combines multi-scale learning and dual attention block, which can improve the capability to extract effective features. Then, an entropy optimization strategy is introduced to promote target domain sample clustering without prior knowledge. Finally, the effectiveness of the proposed model is verified on a public dataset of rotating machinery. The results show that the proposed method outperforms six existing cross-domain fault diagnosis methods. This article proposed a universal domain adaptation diagnosis method to address the domain transfer problem in rotating machinery fault diagnosis. The proposed model achieved better performance than six representative domain adaptation-based diagnosis methods in comprehensive experiments under different operating conditions, proving the potential of the proposed model to solve universal domain adaptation fault diagnosis problems. image