Edge adjacency graph and neural network architecture for machining feature recognition

Recent advances in the field of artificial intelligence have yielded promising results with regard to the prospect of automating the task of machining feature recognition (MFR). Popular among the proposed methods for MFR are learning-based methods using deep neural networks, which have achieved impressive results. We propose the Edge Adjacency Graph Instance Segmentor (EAGIS), a learning-based MFR method comprising a graph neural network and a graph data structure representing the topological and geometric relationships of edges. Evaluations performed on the open-source synthetic MFInstSeg dataset show that EAGIS has comparable performance to existing learning-based methods, despite having a greatly reduced number of trainable parameters in its neural network architecture.