Manufacturing Process Selection Using Graph Neural Networks: A Novel Approach to CAD Model Semantic Segmentation

Manufacturing process selection, as a core element in the manufacturing workflow, plays a crucial role in improving processing efficiency and manufacturing quality. However, due to the fragmented, experience-based, and often outdated nature of processing knowledge, current process selection methods heavily rely on human intervention. To address these challenges, this study proposes an innovative approach that reframes manufacturing process selection as a graph-based semantic segmentation task. Building on this concept, we introduce a graph neural network-based method for manufacturing process selection, capable of effectively identifying processing methods in CAD models. This method discretely samples the topological faces and edges of CAD models, storing them as node and edge attributes in an attributed adjacency graph. Finally, we design experiments to validate the effectiveness and accuracy of this method in identifying manufacturing processes. Experimental results demonstrate that representing 3D CAD models with attributed adjacency graphs effectively captures the geometric and topological relationships between different faces of the model. This method can accurately identify corresponding manufacturing processes at the topological face level of CAD models, offering a novel approach to automated process selection.