To solve the low efficiency and insufficient accuracy of trajectory planning in industrial robot blade grinding, this paper proposes a layered trajectory planning method for local precision grinding of blade surfaces. A scanner is utilized to capture the local allowance deviation between the actual and theoretical blade models, so as to construct a three-dimensional allowance distribution map. In view of the non-uniform allowance distribution, the DBSCAN clustering algorithm is applied for regional segmentation, converting inefficient global machining into efficient local machining. The local residual allowance is stratified into successive processing layers, and the Alpha Shapes algorithm is adopted to extract the outer contour of the point cloud for each layer. The planar grinding trajectory is planned and projected onto the blade surface, and discretized with a preset grinding depth to generate the final machining trajectory. Simulation and experimental results demonstrate that the proposed local-layered method effectively eliminates idle strokes and repetitive grinding, improves grinding efficiency, and achieves accurate and efficient correction of blade local errors while ensuring machining precision. This method provides a novel technical approach for the intelligent grinding of complex curved parts.