An improved system for efficient shape optimization of vehicle aerodynamics with “noisy” computations
An improved system for efficient shape optimization of vehicle aerodynamics with “noisy” computations
复制标题
DOI：
10.1007/s00158-022-03323-9
复制
发表时间：
2022-08
影响因子：
3.9
通讯作者：
Qingyu Wang;T. Nakashima;Chenguang Lai;Xinru Du;Taiga Kanehira;Y. Konishi;Hiroyuki Okuizumi;Hidemi Mutsuda
中科院分区：
文献类型：
--
作者：
Qingyu Wang;T. Nakashima;Chenguang Lai;Xinru Du;Taiga Kanehira;Y. Konishi;Hiroyuki Okuizumi;Hidemi Mutsuda
为了基于产生不可避免的噪声的计算流体动力学（CFD）模拟有效地实现车辆的空气动力学目标的切实改进，提出了一种改进的系统。该系统被称为基于回归克里格再插值（RKri）的有效全局优化（EGO）与伪期望改进（PEI）准则（RKri-EGO-PEI），用于直接滤除CFD模拟产生的噪声，保持代理模型的平滑趋势，并以并行方式进行点填充。为了保证优化过程中调整超参数的RKri和寻找一个适当的质量的解决方案的PEI功能，优化器的性能优势的并行EGO算法称为EGO-PEI进行了全面的研究。然后，选择最佳的作为RKri-EGO-PEI系统的优化器。为了确认所提出的系统的性能，RKri-EGO-PEI竞争与普通的克里格为基础的EGO-PEI（OK-EGO-PEI）和基于RKri-EGO（RKri-EGO）系统上的一个现实世界的车辆空气动力学优化问题。研究结果表明，以探索-利用为中心目标的优化器不仅能在适当的填充点数内提高EGO-PEI算法的收敛性，而且能以较少的迭代次数保证EGO-PEI算法的收敛性.此外，RKri-EGO-PEI搜索更低的阻力系数（Cd）的车辆模型具有更快的速度和更小的挂钟时间成本比OK-EGO-PEI和RKri-EGO下的优化问题与“噪声”计算。
To efficiently achieve tangible improvements in the aerodynamic objectives of a vehicle based on a computational fluid dynamics (CFD) simulation that produces unavoidable noise, an improved system is proposed. This system, called regression kriging with re-interpolation (RKri)-based efficient global optimization (EGO) with a pseudo expected improvement (PEI) criterion (RKri-EGO-PEI), is used to directly filter out the noise produced by the CFD simulation, maintain a smooth trend of the surrogate model, and conduct point infills in a parallel manner. To guarantee optimization processes for tuning the hyper-parameters of RKri and searching for a solution of appropriate quality to the PEI function, the performance advantages of the optimizers on a parallel EGO algorithm called EGO-PEI are comprehensively investigated. Then, the best is chosen as the optimizer for the RKri-EGO-PEI system. To confirm the performance of the proposed system, RKri-EGO-PEI competes with ordinary kriging-based EGO-PEI (OK-EGO-PEI) and RKri-based EGO (RKri-EGO) systems on a real-world optimization problem of vehicle aerodynamics. The results of the investigation show that the performance of the optimizer with a higher central goal of exploration–exploitation can not only promote a higher-level convergence of the EGO-PEI algorithm within an appropriate number of point infills, but also ensure the same convergence level of the EGO-PEI algorithm as that using other optimizers, with fewer iterations. In addition, RKri-EGO-PEI searches for a lower drag coefficient (Cd) of the vehicle model with a faster speed and smaller wall-clock time cost than those of OK-EGO-PEI and RKri-EGO under an optimization problem with “noisy” computations.