Automatic Recognition Of Machining Features Based On 3D Point Cloud Data

CAD/CAM integration is an important part of computer integrated manufacturing technology.Automatic identification of machining features is the core technology to realize CAD/CAM integration.Data-driven deep learning is widely used in various pattern recognition.Convolutional neural network methods based on 3D point cloud data have better robustness than traditional symbol-based reasoning methods,which provides a new approach to automatically identify the machining features of the solid boundary model or 3D point cloud model.The paper analyzes the definition and classification of machining features in different backgrounds,and compares and analyzes the machining feature recognition methods generated under different definitions.The main development directions of feature recognition are summarized after a survey of existing methods about machining feature recognition.The method and theory of feature recognition of convolutional neural networks(CNN)are investigated.The characteristics of point cloud data are analyzed,and then the machining feature recognition technology and basic principles of CNN-based machining feature recognition are expounded.Then the method and system scheme of CNN machining feature recognition based on 3D point cloud are proposed.According to the design requirements analysis of the system scheme,the input data sample collection method based on the unified rule 3D point cloud network model is proposed.The method is mainly divided into three steps: The CAD model is transformed into a point cloud data model,and the point cloud model is resampled into a unified number of point cloud data sets and converted into a data format that can be understood by machine learning.24 large-scale point cloud data required for CNN network training of machining features are constructed by modifying the CAD model size parameters and geometric position.A new CNN network architecture based on 3D point cloud data machining feature recognition was designed based on PointNet,and the feasibility analysis of the network model architecture was carried out according to the characteristics of point cloud data.The training program for 3D point cloud data is designed in this paper.A more stable CNN machining feature recognizer is constructed by programming using Python language and training using a large number of 3D point cloud data sample on the framework of TensorFlow machine learning system.The machining features are classified and predicted to identify different machining features by extracting input feature of the point cloud data of the boundary model.Finally,the automated recognition experiments were performed on 24 individual machining features by the verification data set of the system,and the higher recognition accuracy and recognition efficiency were obtained.And then the validity and feasibility of the method were verified,which lays a foundation for the subsequent CAD/CAPP system integration.