基于 CFD 与机器学习的 Sedan 车辋高和倾角的多目标气动优化
核实验
2026-04-21 v3 仪器与探测器
摘要
本研究利用 CFD 分析研究了 Audi A4 轿车的气动性能。开发了在 SolidWorks 中构建的 3D 模型，与 DrivAer Notchback 风洞数据验证，仅有 3.25% 的阻力系数 (Cd) 偏差。车身高度变化于 1.336 到 1.536 米，倾角变化于 0 到 5 度，涉及四个雷诺数。梯度提升方法被证实是最准确的预测模型（Cd 的 R 平方为 0.97，Cl 的 R 平方为 0.96），优于随机森林和 LightGBM。进行了在平衡、聚焦阻力和聚焦下力条件下的差分进化优化。雷诺数对最优位置的影响最小，因此仅报告一个雷诺数下的详细结果，其他雷诺数显示类似趋势。基线几何形态的 Cd 为 0.313，Cl 为 0.0288。平衡优化实现 Cd = 0.287，Cl = -0.0826。最小阻力条件达到 Cd = 0.285，伴随轻微正升力 (Cl = 0.0142)，而最大下力优化达到 Cl = -0.1084，伴随 6.71% 的阻力惩罚 (Cd = 0.334)。在车身高度为 1.341 到 1.365 米、倾角为 0.158 到 4.610 度的范围内，找到了近最优解，表明气动性能稳健。机器学习预测进一步通过 CFD 验证，Cd 误差较小。
引用
@article{arxiv.2509.02916,
title = {Initial results of the TRIUMF ultracold advanced neutron source},
author = {B. Algohi and D. Anthony and L. Barrón-Palos and M. Bossé and M. P. Bradley and A. Brossard and T. Bui and J. Chak and R. Chiba and C. Davis and R. de Vries and K. Drury and B. Franke and D. Fujimoto and R. Fujitani and M. Gericke and P. Giampa and C. Gibson and R. Golub and K. Hatanaka and T. Hepworth and T. Higuchi and G. Ichikawa and I. Ide and S. Imajo and A. Jaison and B. Jamieson and M. Katotoka and S. Kawasaki and M. Kitaguchi and W. Klassen and E. Korkmaz and E. Korobkina and F. Kuchler and M. Lavvaf and T. Lindner and N. Lo and S. Longo and K. W. Madison and Y. Makida and J. Malcolm and J. Mammei and R. Mammei and Z. Mao and C. Marshall and J. W. Martin and R. Matsumiya and M. McCrea and E. Miller and M. Miller and K. Mishima and T. Mohammadi and T. Momose and M. Nalbandian and T. Okamura and S. Pankratz and R. Patni and R. Picker and K. Qiao and W. D. Ramsay and W. Rathnakela and T. Reimer and D. Salazar and J. Sato and W. Schreyer and T. Shima and H. M. Shimizu and S. Sidhu and S. Stargardter and R. Stutters and P. Switzer and I. Tanihata and Tushar and S. Vanbergen and W. T. H. van Oers and N. Yazdandoost and Q. Ye and A. Zahra and M. Zhao},
journal= {arXiv preprint arXiv:2509.02916},
year = {2026}
}