Abstract: A computational fluid dynamics model was established to extract the pulsation pressure of the windows by taking the passenger car aerodynamic noise as the research object. The pulsation pressure was applied as excitation to the vehicle acoustic cavity model to simulate the noises at the drivers ears， and the simulation results were in good agreement with the experimental data. The intensity of Curle noise sources on car surface was taken as the optimization objective， and the sensitivity was identified by discrete adjoint method， then the rearview mirror， A-pillar section and hood were selected as the optimization areas. The Hamersley test design method was used to construct the sample space， and the free mesh deformation technique was used to parameterize the sample point model and the corresponding sound power value was calculated. The Kriging interpolation method was used to create the proxy model， and the multi-island genetic algorithm was used for global optimization. The optimization results show that the maximum sound power level of the window surfaces decrease by 2 dB， and the sound pressure level near the drivers ears drops by 0.7 dB（A）. 

Key words: aerodynamic noise, simulation, discrete adjoint method, proxy model

摘要： 以某乘用车气动噪声为研究对象建立了整车流体动力学模型，并用该模型提取车窗脉动压力，然后将该压力作为激励加载到车内声腔模型中对驾驶员耳旁噪声进行仿真分析，仿真结果与试验数据吻合。将车身表面Curle噪声源强度作为优化目标，采用离散伴随法进行灵敏度识别，进而确定后视镜、A柱截面、引擎盖为优化区域。采用哈默斯雷试验设计方法构建样本空间，利用网格自由变形技术参数化样本点模型，计算出对应的声功率值。运用Kriging插值法构建代理模型，使用多岛遗传算法对模型进行全局寻优。优化结果显示，与原模型相比，车窗表面声功率级最大值减小2 dB，驾驶员耳旁声压级下降0.7 dB（A）。

关键词: 气动噪声, 仿真, 离散伴随法, 代理模型