Group maintenance management is pivotal to ensure operational safety and performance of multi-component plants attributed to its capacity to share maintenance resources/time. Most group maintenance models, however, are globally/partially static following pre-specified maintenance sequences, with limited focus on the adaptability of group partition procedure. To fill this gap, we devise an innovative global-dynamic condition-based group maintenance policy. In contrast to existing methods, it allows for (a) postponement of component maintenance upon inspection to facilitate flexible resource allocation, and (b) automatic refinement of group maintenance structures to promote adaptivity. The proposed policy is shown to establish a global renewal mechanism for maintenance group partition over an infinite time horizon, which constitutes a dynamic union of both scheduled maintenance and opportunistic maintenance to mitigate downtime. A heuristic grouping algorithm is developed to realize efficient maintenance group planning, which verifies model effectiveness via numerical experiments.