| Tartalmi leírás: |
Toolpath synthesis in Computer-Aided Manufacturing (CAM) has long been constrained by rigid geometric heuristics and predefined templates (e.g., trochoidal or zig-zag patterns). Although these approximations ensure coverage, they inherently limit the physical optimality of the cutting process, often failing to maintain constant Cutter-Workpiece Engagement (CWE) in complex geometries. This paper proposes a Neural-Differential Framework that transitions path planning from discrete template matching to continuous, physics-aware synthesis. By employing a neural surrogate to approximate the material removal dynamics, an end-to-end differentiable pipeline is constructed that optimizes spline parameters (control points and knot vectors) via gradient-based optimization algorithms. This approach unlocks global path optimization, allowing the tool trajectory to emerge as a solution to physical constraints rather than geometric rules. Furthermore, the framework leverages the massive parallelism of modern GPU architectures to perform high-fidelity physics evaluation at scale, a capability that traditional CPU-bound geometric kernels lack.
|
