GraSTNet: Graph Neural Networks and Set Transformer for Machining Feature Recognition

The integration of automated machining feature identification within the context of 3D-CAD models presents a bridge between Computer-Aided Design (CAD) and Computer-Aided Manufacturing (CAM) systems. A new of machining feature recognition algorithm is presented to eliminate manual tasks prone to errors, consequently leading to enhancements in productivity. The choice of a deep learning-based method over a rule-based approach is grounded in its scalability, adaptability, and generalization capabilities. A hybrid approach with graph neural network and set transformers (GraSTNet) is followed for localization and classification tasks of machining features. A study between state-of-the-art approaches and the 3D representation techniques are presented and the outcomes are discussed by analyzing their performances on benchmark datasets. To ensure real-world applicability, GraSTNet's performance extends beyond cubic parts to rotational components, while the number of trainable parameters compared to existing state-of-the-art methods are reduced, signifying a balance between efficiency and model size. The results of this study hold promise for the advancement of smart manufacturing practices and the evolution of interdisciplinary collaboration between CAD and CAM systems.