Data-driven gear remaining useful life (RUL) prediction often degrades under unseen working conditions. Although single domain generalization (SDG) addresses this problem using only single-source data, existing methods mainly rely on static augmentation and overlook the temporal evolution of mechanical degradation. To address this issue, a degradation manifold feature enhancement-based SDG framework (DMFE-SDG) is proposed. The core of this framework is the degradation manifold feature enhancement (DMFE) method. The DMFE module combines degradation process evolution (DPE) with liquid parameter update (LPU) to simulate condition-induced degradation variations and progressively evolve features along the temporal dimension. This generates smooth virtual degradation trajectories and expands the source feature distribution toward potential unseen domains. Experiments on a gear run-to-failure test bench demonstrate improved RUL prediction and generalization under unseen working conditions.

gear,Remaining useful life (RUL)