Abstract: To address the problems of large invalid space, high false collision rate, and high computational complexity in existing bounding box generation methods for collision detection in digital manufacturing of sheet metal bending, a GPU-CPU accelerated bounding box optimization algorithm for sheet metal bending is proposed. The algorithm exploits the geometric characteristics of extruded bodies in sheet metal processing environments, generates compact and efficient bounding box clusters through adaptive rasterization, parallel scan merging, and a parametric extrusion mechanism, and introduces a multi-resolution strategy and a hybrid acceleration framework to balance accuracy and computational efficiency. Experimental results show that, compared with the axis-aligned bounding box (AABB), oriented bounding box (OBB), convex hull, AlphaShape, and k-discrete oriented polytope (k-DOP) methods, the proposed algorithm significantly improves bounding box tightness and reduces computation time to the millisecond level.