Journal of Graphics ›› 2021, Vol. 42 ›› Issue (1): 117-123.DOI: 10.11996/JG.j.2095-302X.2021010117
• Digital Design and Manufacture • Previous Articles Next Articles
Online:
Published:
Supported by:
Abstract: In view of the current situation that the traditional methods of manufacturability analysis based on knowledge and rules are not flexible and the existing methods of manufacturability analysis based on deep learning are unable to give the specific reasons for the non-manufacturability of parts, a method of manufacturability analysis based on deep learning was proposed. Firstly, a large number of CAD models with manufacturability category labels were constructed through digital modeling technology, and the point cloud was extracted to build the data set needed for deep learning. Then, based on the PointNet network, a deep learning network for hole feature manufacturability analysis was built, and the network training and parameter adjusting process were completed. Then, compared with the 3D-convolutional neural networks (3D-CNN), the deep learning network constructed in this paper exhibits better robustness and lower time complexity. Finally, the manufacturability analysis of hole feature in a sample part was carried out to identify the non-manufacturable hole feature, and the reason of non-manufacturability was explained. The experimental results show that the method can not only ensure high recognition accuracy, but also identify the reason why the feature cannot be manufactured, which is of greater application value.
Key words: manufacturability analysis, digitization modeling, deep learning, hole features, PointNet 
CLC Number:
TP 391 
ZHANG Hang, ZHANG Shu-sheng, YANG Lei . Deep learning based manufacturability analysis approach for hole features [J]. Journal of Graphics, 2021, 42(1): 117-123.
0 / / Recommend
Add to citation manager EndNote|Ris|BibTeX
URL: http://www.txxb.com.cn/EN/10.11996/JG.j.2095-302X.2021010117
http://www.txxb.com.cn/EN/Y2021/V42/I1/117