Part CAD model reusing method based on point cloud classification network

Abstract

The invention discloses a part CAD (Computer-Aided Design) model reusing method based on a point cloud classification network. A retrieval problem of a three-dimensional part CAD model is converted into a classification problem of a three-dimensional point cloud model. A classification network and a retrieval network are established by taking a PointNet network as a framework, and a residual blockstructure of a ResNet network is introduced into the retrieval network, so that the recognition precision of high-similarity parts is improved. A three-dimensional part CAD model library for small-batch customized products is established, the three-dimensional part CAD model library is preprocessed to obtain a training data set, the part point cloud model is input into the classification networkand the retrieval network corresponding to the part category in sequence, and retrieving is performed to obtain the most similar part model. According to the method, three-dimensional space information of the part CAD model can be better captured, rapid and accurate recognition of the part CAD model is achieved, and then the processes of design, process planning, manufacturing, assembling and thelike of small-batch customized parts of an enterprise are accelerated.

Description

technical field [0001] The invention belongs to the field of intelligent information technology of advanced manufacturing technology, and in particular relates to a method for reusing part CAD models based on a point cloud classification network. Background technique [0002] At present, computer-aided three-dimensional modeling has become the main tool for manufacturing product design, resulting in a large number of three-dimensional CAD models. storage carrier. Therefore, effective retrieval and reuse of 3D computer-aided design models can greatly save time and cost in new product design, manufacturing, testing and assembly processes. With the emergence of massive CAD models, how to efficiently retrieve similar CAD models from the model library has become a key issue. [0003] Traditional CAD model retrieval techniques are usually accomplished through the following two steps: first, a 3D CAD model is represented by a descriptor, such as an information descriptor, a geome...

AI Technical Summary

Problems solved by technology

[0004] With the wide application of deep learning in the field of image recognition, 3D model retrieval methods based on multi-view image classification technology have received some attention. Such methods are often more sensitive to the local information of parts, and it is difficult to capture the overall 3D spatial characteristics. Small batches with small 3D spatial information and small size variations - Poor retrieval of customized parts

Recently, a method that takes 3D voxel as input and uses 3D-CNN to perform processing feature recognition, but because the 3D voxel model requires more storage space than the point cloud model, it cannot represent complex 3D part models, so this This method can only achieve simple classification of part models, but cannot solve the retrieval problem for similar parts

Images