arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2304.14973physics.flu-dyn
收藏

结合深度神经网络与CFD仿真的进化多目标气动设计优化

Evolutionary Multi-Objective Aerodynamic Design Optimization Using CFD Simulation Incorporating Deep Neural Network

Yukito Tsunoda, Akira Oyama

AI总结:

提出将DNN流场推断与CFD仿真结合的进化多目标气动优化方法,并通过翼型优化验证其在96核下可将计算时间降至57.9%。

摘要:

本文提出一种进化多目标气动设计优化方法,该方法采用融入深度神经网络(DNN)的计算流体力学(CFD)仿真,以减少所需计算时间。在该方法中,DNN根据设计的网格数据推断流场,CFD仿真从推断出的流场开始,以更少的时间积分步数获得稳态流场。为证明所提方法的有效性,展示了一项多目标气动翼型设计优化。结果表明,在96核处理器条件下,设计优化的计算时间被压缩至57.9%。

英文摘要:

An evolutionary multi-objective aerodynamic design optimization method using the computational fluid dynamics (CFD) simulations incorporating deep neural network (DNN) to reduce the required computational time is proposed. In this approach, the DNN infers the flow field from the grid data of a design and the CFD simulation starts from the inferred flow field to obtain the steady-state flow field with a smaller number of time integration steps. To show the effectiveness of the proposed method, a multi-objective aerodynamic airfoil design optimization is demonstrated. The results indicate that the computational time for design optimization is suppressed to 57.9% under 96 cores processor conditions.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑