Machining Feature Recognition Method Based on 3D Convolution Neural Network

The research of automatic recognition of machining features constitutes a crucial supporting pillar for intelligent design and manufacturing. Despite the availability of existing algorithms, there are prevalent problems including low recognition efficiency, low recognition accuracy, and limited applicability to hybrid intersection features. The recognition capability of neural networks has achieved significant success in various fields. The application potential of neural networks for machining feature recognition is currently being explored. In this paper, we propose a feedforward neural network and a novel 3D convolutional neural network framework based on semantic and voxelization methods to identify machining features from 3D CAD models. Sufficient experiments on the datasets of 24 machining characteristics reveal an excellent accuracy result. The proposed framework provides a reliable algorithmic basis for implementing real-time computer aided process planning (CAPP) systems.