STRESS-ORIENTED TOOLPATH PLANNING FOR FUSED FILAMENT FABRICATION USING A GRAPH NEURAL NETWORK SURROGATE MODEL

Abstract

This paper presents a novel, time-efficient toolpath generation algorithm for Fused Filament Fabrication (FFF) additive manufacturing that aligns the printing direction with the maximum principal stress orientations to enhance the mechanical strength of printed parts. A Graph Neural Network (GNN)-based surrogate model is developed to replace conventional Finite Element Analysis (FEA) for stress computation on unstructured meshes. The model directly predicts both the magnitude and direction of the maximum principal stresses in approximately 10 ms. Based on stress predictions, a dense infill toolpath generation algorithm is proposed using a Depth-First Search (DFS) framework. An improved hexagonal grid generation method is developed to determine the grid orientation according to the global dominant principal stress direction and can adaptively adjust grid spacing to preserve geometric accuracy along part contours. An improved DFS algorithm is then employed to generate preliminary toolpaths using a simplified printing criterion that permits a ±30° deviation between the printing direction and the principal stress direction. Breakpoints of the printing path are then eliminated by connecting adjacent paths, and the final toolpath is optimized using the nearest-neighbor path connection strategy. Using the cantilever beam dataset as a case study, the proposed algorithm can generate a complete printing path within 2∼3 seconds and improve the compressive strength of Polylactic Acid (PLA) parts by more than 35%. This algorithm can be applied to all extrusion- and deposition based additive manufacturing processes, including composite 3D printing and metal directed energy deposition.