A hybrid learning framework for manufacturing feature recognition using graph neural networks

Abstract

Manufacturing feature recognition is a critical issue in intelligent manufacturing, which can extract valuable geometric information from solid models for humans not to be subservient to machines and automation. Features can bridge the information gap between computer-aided design (CAD), computer-aided process planning (CAPP), computer-aided engineering (CAE), and computer-aided manufacturing (CAM). Based on the concept of feature, the whole process of digital seamless connection from design to manufacture can be implemented. However, the existing methods have not solved the feature recognition problem well. These methods have some limitations, such as lack of learning ability/low learning efficiency, lack of expandability, low accuracy, and so on. To improve the performance of the feature recognition, a hybrid learning framework based on Graph neural network (GNN) termed DeepFeature is proposed. First, a method that can extract features from CAD solid model is used. Then, a scheme for the construction and storage of feature datasets is developed. The feature samples in the dataset are sufficient, and the representation of each feature is different. Next, DeepFeature models are constructed, and the models are trained and driven by the samples in feature datasets. Finally, the features of the parts are classified based on rules and GNN models. The interacting features with multiple base planes are decomposed into several isolated features for feature classification. The experimental results show that the proposed hybrid learning framework not only has high feature recognition accuracy but also has good robustness in handling interacting features.