An Application-oriented Perspective of Domain Generalization for Cross-Domain Fault Diagnosis

Traditional data-driven fault diagnosis methods generally assume that the training and testing distributions are the same, which does not hold in real-world industrial applications. To address domain shift problems, domain generalization-based fault diagnosis (DGFD) methods have been explored to achieve real-time cross-domain fault diagnosis. Some progress has been made in the area of DGFD for years. This paper presents an overview of recent advances in DGFD. First, we provide a formal definition of domain generalization and discuss several related learning paradigms used in intelligent fault diagnosis. Second, we define several major applications of domain generalization in intelligent fault diagnosis. Then, the motivations and challenges of these applications are discussed, and current solutions are summarized.