- 発表日 2022/07/01
- Computer-Aided Design and Applications, 19(4), pp.677-693,(2022)（英文ジャーナル)
- DOIコード 10.14733/cadaps.2022.677-693
Recognition of Free-form Features for Finite Element Meshing using Deep Learning
Finite element (FE) models used in large-scale vehicle simulations are usually composed of high-quality FE meshes that comply with in-house meshing specifications. The meshing patterns, location, and resolution are strictly defined for specific free-form features, such as ribs and bosses, in the specifications. However, finding such free-form features on the given computer-aided design (CAD) models requires a great deal of manual operation. This study proposes a deep-learning (DL) approach to recognize free-form features on CAD models for the automatic generation of FE models to address this issue. This approach allows training a deep neural network on point clouds with reasonable recognition accuracy using a dataset with a large variety of free-form feature shapes represented by the point cloud. The dataset is generated from parametric CAD modeling. It classifies the types of the free-form feature shapes on an input product model and label local feature areas on them. The proposed free-form feature recognition method was experimentally verified using two types of complicated product models. First, models where multiple free-form feature shapes are located independently. Second, where multiple feature shapes are smoothly connected and interacted. The labeling verification showed excellent automatic identification of the feature locations and local feature areas on free-form feature shapes. These results suggest that the proposed DL approach for free-form feature recognition effectively generates FE models.
関連記事
