基于三维深度学习的汽车气动性能实时预测

Real-time prediction of vehicle aerodynamic performance based on 3D deep learning

为适应现代汽车快速设计的需求,采用基于三维深度学习算法的汽车气动参数实时预测,计算汽车的空气阻力系数。利用Rhinoceros软件对包含多种车型的汽车模型库进行T样条曲面重构,制作汽车外形的三维点云数据集;分别利用FLUENT和CFX对模型逐个进行不同风速工况下的仿真分析,得到相应的空气阻力系数,并建立三维深度学习的训练和测试数据集;采用PointNet深度学习框架训练并计算各模型的空气阻力系数。训练集的对比结果表明,采用深度学习方法快速预测汽车气动性能可得到基本满意的效果。

To meet the needs of modern vehicle rapid design, the real-time prediction of automobile aerodynamic parameters based on the 3 D deep learning algorithm is used to calculate the vehicle air resistance coefficient. Rhinoceros software is used to reconstruct the T-spline surface of the vehicle model library containing various models, and the 3 D point cloud data set of vehicle shape is made;the vehicle model under different wind speed conditions is simulated and analyzed one by one using fluent and CFX, and the corresponding air resistance coefficient is obtained, and the training and test data set of 3 D deep learning is established;the air resistance coefficient of each model is trained and calculated using the PointNet deep learning framework. The comparison results of training sets show that the deep learning method can quickly predict the aerodynamic performance of vehicles, and the effect is basically satisfactory.