Progress and prospects of domain generalization mechanical fault diagnosis research

<p indent="0mm">Distribution shifts pose challenges for deploying intelligent fault diagnosis models in real-world engineering applications. To address these shifts, research on intelligent fault diagnosis using transfer learning has been actively pursued. Among the unsupervised transfer learning methods, domain adaptation and domain generalization are particularly noteworthy. Domain adaptation requires unlabeled target domain data, while domain generalization eliminates this need, offering greater flexibility and versatility. Since domain generalization was first applied to fault diagnosis in 2020, numerous diagnostic techniques have emerged, necessitating a thorough review. This paper first classifies existing domain generalization fault diagnosis studies in general by task type and further refines them based on methodological theories. Subsequently, it summarizes the core concepts of each theory and reviews related studies in fault diagnosis. Finally, promising future research directions are discussed.