基于深度神经网络的含运动边界非定常流场预测方法研究  

Research on the Prediction Method of Unsteady Flow Field with Moving Boundary Based on Deep Neural Network

作　　者：韩仁坤 杜焦喜 刘子扬 李立 陈刚[1,2] Han Renkun;Du Jiaoxi;Liu Ziyang;Li Li;Chen Gang(State Key Laboratory for Strength and Vibration of Mechanical Structures,Xi’an Jiaotong University,Xi’an 710049,China;Shaanxi Key Laboratory for Environment and Control of Flight Vehicle,Xi’an Jiaotong University,Xi’an 710049,China;AVIC Aviation Computing Technology Research Institute,Xi’an 710049,China)

机构地区：[1]西安交通大学机械结构强度与振动国家重点实验室,陕西西安710049 [2]西安交通大学先进飞行器服役环境与控制陕西省重点实验室,陕西西安710049 [3]航空工业计算技术研究所,陕西西安710049

出　　处：《航空科学技术》2023年第12期37-42,共6页Aeronautical Science & Technology

基　　金：航空科学基金(20200014070001)。

摘　　要：针对飞行器设计过程中对流固耦合系统快速预测的需求,探索基于数据驱动的非定常流场建模策略,缩短流场演化求解耗时,从而加快流固耦合系统模拟速度。流固耦合系统中流场演化求解部分等价于含运动边界的非定常流场演化。本文提出了一种基于神经网络的流场预测模型,来学习并预测运动边界附近的非定常流场演化过程。此神经网络可以基于当前流场状态及边界运动信息预测出下一个时刻的流场状态。通过不同振动频率及振幅下的运动圆柱绕流问题,测试了本文提出的神经网络模型的预测精度及泛化能力。神经网络的预测结果和计算流体力学仿真结果中流场结构吻合度较高,且通过对预测的流场数据中边界上的压力积分得到的气动力也具有较高的精度。测试结果展示了该神经网络模型具有良好的预测性能,因此该方法可以用于快速、准确获得运动边界周围非定常流场状态。In order to meet the requirement of rapid prediction of fluid-structure interaction system in aircraft design,a data-driven unsteady flow field modeling strategy was explored to shorten the time spent on flow field evolution solution and accelerate the simulation speed of fluid-structure interaction system.The solution of flow field evolution in fluid-structure interaction system is partially equivalent to the evolution of unsteady flow field with moving boundary.This paper proposes a flow field prediction model based on neural networks to learn and predict the evolution of unsteady flow fields with moving boundaries.This neural network can predict the flow field at next timestep based on the current flow field and boundary motion information.The prediction accuracy and generalization ability of the proposed neural network model were tested by the flow around a moving cylinder under different vibration frequencies and amplitudes.The predicted flow fields of the neural network are in accordance with the computational fluid dynamics simulation results.The aerodynamic force obtained by integrating the pressure on the boundary of the predicted flow field data also has a high accuracy.The test results demonstrate that the good predictive performance of the neural network model,so this method can be used to quickly and accurately obtain the unsteady flow field state around the moving boundary.

关 键 词：运动边界 非定常 神经网络 深度学习 快速预测 

分 类 号：V211.3[航空宇航科学与技术—航空宇航推进理论与工程]