Accurately diagnosing rotating equipment faults under varying working conditions while simultaneously identifying previously unseen fault types presents a critical challenge known as Open-Set Domain Generalization (OSDG), limiting the practical deployment of many diagnostic systems. This paper addresses this gap by proposing a novel OSDG fault diagnosis method with a dualistic meta-learning strategy. The core of the proposed method lies in a meta-learning process that simultaneously optimizes for generalization across working domains and fault categories using gradient matching. This training process, applied to informative time-frequency features extracted via the Wigner-Ville Distribution (WVD), yields highly discriminative feature representations. These features not only improve classification accuracy for known faults under unseen conditions but are also structured to enable the effective detection of emerging unknown fault categories using auxiliary binary classifiers. Finally, evaluation experiments on diverse OSDG tasks, including comparisons with state-of-the-art methods, demonstrate the effectiveness of the proposed method in accurately identifying both known and novel faults under new working conditions.