Skip to main content
QUICK REVIEW

[论文解读] Factorized Implicit Global Convolution for Automotive Computational Fluid Dynamics Prediction

Chris Choy, A. Kamenev|ArXiv.org|Feb 6, 2025
Aerodynamics and Fluid Dynamics Research被引用 3
一句话总结

本文提出 FIGConv,这是一个用于汽车 CFD 的平方复杂度神经网络,利用因子化隐式网格和二维重参数化来预测大型三维网格的阻力和每面压力,在 DrivAerNet 与 Ahmed body 数据集上超越了现有方法。

ABSTRACT

Computational Fluid Dynamics (CFD) is crucial for automotive design, requiring the analysis of large 3D point clouds to study how vehicle geometry affects pressure fields and drag forces. However, existing deep learning approaches for CFD struggle with the computational complexity of processing high-resolution 3D data. We propose Factorized Implicit Global Convolution (FIGConv), a novel architecture that efficiently solves CFD problems for very large 3D meshes with arbitrary input and output geometries. FIGConv achieves quadratic complexity $O(N^2)$, a significant improvement over existing 3D neural CFD models that require cubic complexity $O(N^3)$. Our approach combines Factorized Implicit Grids to approximate high-resolution domains, efficient global convolutions through 2D reparameterization, and a U-shaped architecture for effective information gathering and integration. We validate our approach on the industry-standard Ahmed body dataset and the large-scale DrivAerNet dataset. In DrivAerNet, our model achieves an $R^2$ value of 0.95 for drag prediction, outperforming the previous state-of-the-art by a significant margin. This represents a 40% improvement in relative mean squared error and a 70% improvement in absolute mean squared error over previous methods.

研究动机与目标

  • 证明可扩展的 CFD 仿真器有能力处理非常大的汽车网格的需求。
  • 提出一种新颖的体系结构(FIGConv),将三维卷积在高分辨率域上的复杂度从 O(N^3) 降低到 O(N^2)。
  • 引入因子化隐式网格以用更小的隐式网格表示高分辨率域。
  • 开发一种在因子化网格上通过二维重参数化实现全局卷积的机制(FIG 卷积)。
  • 在工业数据集上展示阻力预测和每面压力的最先进性能,同时保持快速推理。

提出的方法

  • 将因子化隐式网格定义为共同近似高分辨率三维域的多个低分辨率轴。
  • 应用 Factorized Implicit Convolution 对隐式网格并行执行全局卷积。
  • 使用二维重参数化在展平表示上高效实现大核三维卷积。
  • 通过学习型融合(从其他网格采样并添加至目标网格)在因子化网格之间融合信息。
  • 使用连续卷积概念和椭球邻域,从输入点云/网格初始化并持续对因子化网格进行卷积。
  • 采用一个 U 形编码器-解码器(FIGConvNet)来预测每面压力与总阻力,并通过跳跃连接进行信息聚合。
Figure 1 : FIGConvNet: ConvNet for drag prediction using FIG convolution blocks . The encoder and decoder consist of a set of FIG convolution blocks and we connect the encoder and decoder with skip connections. The output of the encoder is used for drag prediction and the output of the decoder is us
Figure 1 : FIGConvNet: ConvNet for drag prediction using FIG convolution blocks . The encoder and decoder consist of a set of FIG convolution blocks and we connect the encoder and decoder with skip connections. The output of the encoder is used for drag prediction and the output of the decoder is us

实验结果

研究问题

  • RQ1因子化隐式网格表示是否能够在极大汽车网格上实现高效、可扩展的三维 CFD 预测?
  • RQ2FIGConv 在 DrivAerNet 与 Ahmed body 数据集上是否能在阻力预测与每面压力方面达到有竞争力或更优的准确性?
  • RQ3所提出的二维重参数化是否能够在保持计算效率的同时有效使用大核卷积?
  • RQ4因子化网格秩与融合对预测精度和推理时间的影响是什么?

主要发现

Modelc_d Mean SE (↓)c_d Mean AE (↓)c_d Max AE (↓)c_d R^2 (↑)Time sec (↓)
PointNet++7.813E-56.755E-33.463E-20.8960.200
DeepGCN6.297E-56.091E-33.070E-20.9160.151
MeshGraphNet6.0E-56.08E-32.965E-20.9170.25
AssaNet5.433E-55.81E-32.39E-20.9270.11
PointNeXt4.577E-55.2E-32.41E-20.9390.239
PointBERT6.334E-56.204E-32.767E-20.9150.163
DrivAerNet DGCNN †8.0E-56.91E-38.80E-30.9010.52
FIGConvNet (Ours)3.225E-54.423E-32.134E-20.9570.051
  • FIGConvNet 在 DrivAerNet 上的阻力 R^2 为 0.957,且在阻力和每面压力预测方面优于基线。
  • 在 DrivAerNet 上,FIGConvNet 相较于之前的方法在绝对 MSE 上提升了 70%,在相对 MSE 上提升了 40%。
  • 对于 Ahmed body 数据集,FIGConvNet 在标准化压力误差方面达到 0.89%,模型体积较小(68.29 MB)。
  • 该方法推理速度更快,在单个 A100 GPU 上不同配置下,FIGConvNet 的运行时间为 0.051–0.061 秒。
  • 较大的因子化网格秩提升了压力准确性,但可能会降低阻力 R^2 并增加推理时间,体现了分辨率与性能之间的权衡。
  • 网格之间的融合在结果上显著提升,尤其随着网格秩的增加。
Figure 3 : Factorized Implicit Global Convolution 3D : The FIG convolution first creates a set of voxel grids that factorizes the domain. This allows representing a high resolution voxel grid domain implicitly that can be computationally prohibitive to save explicitly. Then, a set of global convolut
Figure 3 : Factorized Implicit Global Convolution 3D : The FIG convolution first creates a set of voxel grids that factorizes the domain. This allows representing a high resolution voxel grid domain implicitly that can be computationally prohibitive to save explicitly. Then, a set of global convolut

更好的研究,从现在开始

从论文设计到论文写作,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。