A large-scale three-dimensional flow field-oriented computing method, system and medium
By constructing a training dataset and modifying the input and output of a point cloud neural network, combined with a sampling method without replacement, the problems of grid size and accuracy in 3D flow field prediction in existing technologies are solved, and efficient and reliable prediction of complex 3D flow fields is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CALCULATION AERODYNAMICS INST CHINA AERODYNAMICS RES & DEV CENT
- Filing Date
- 2026-03-23
- Publication Date
- 2026-05-26
AI Technical Summary
Existing intelligent CFD methods are difficult to directly transfer to 3D scenes, cannot handle 3D flow fields with complex shapes, have limited mesh size, and have uncontrollable prediction errors, which cannot meet the engineering accuracy requirements of the aerospace field.
A training dataset is constructed, the input and output dimensions and loss function of the point cloud neural network are modified, and the point cloud neural network is trained using a sampling-without-replacement method. This method is applicable to 3D spatial meshes of any size and type, and is combined with CFD software for flow field prediction.
It enables efficient prediction of three-dimensional flow fields of arbitrary type and scale, improves prediction accuracy and reliability, and meets the engineering needs of the aerospace field.
Smart Images

Figure CN121902706B_ABST