期刊文献+

采用混合粗糙数据物理信息神经网络的扑翼气动性能预测方法

A Predicting Method on Aerodynamic Performance of Flapping Wing Using Hybrid Coarse Data Driven Physical Informed Neural Network
下载PDF
导出
摘要 为解决时空尺度上扑翼流动控制方程求解需要花费大量时间和计算资源的问题,基于强非线性曲线拟合能力的物理信息神经网络(PINN)深度学习方法,提出了一种混合粗糙数据驱动物理信息神经网络模型(HCDD-PINN),研究了模型对涉及非定常流动特征和动边界二维俯仰扑翼问题的训练和预测性能。通过使用相较于传统计算流体动力学方法(CFD)更为粗糙的数据驱动模型训练,将扑翼流动控制方程嵌入神经网络损失中,并施加初始条件和边界条件约束,采用一阶自适应矩优化算法(ADAM)和二阶拟牛顿法优化算法(L-BFGS-B),以前馈-反向传播方式最小化模型损失函数,从而提高模型预测控制方程数值解的准确性和可靠性。结果表明:与原始PINN模型相比,HCDD-PINN模型显著降低了流场的预测误差,能够准确地预测扑翼瞬时气动力和瞬时速度及压力场,训练时间缩短了75%。此外,训练完成的HCDD-PINN模型可以快速获得流场任意时刻的物理信息,而传统CFD方法则需要重新对流场进行计算。该研究为求解扑翼流动控制方程乃至流体非线性偏微分方程组(PDEs)提供了一种有效的替代方案。 The solution of the governing equations of the flapping wing problem on a spatiotemporal scale takes a great amount of time and computational resources.To solve this challenge,this paper proposes a hybrid coarse data-driven with physics-informed neural network model(HCDD-PINN).The model leverages its strong nonlinear curve fitting ability to investigate the training and predicting performance of the model for two-dimensional pitching flapping wing problems,which involves unsteady flow characteristics and motion boundaries.The model is trained using an order-of-magnitude grid,coarser than the traditional computational fluid dynamic(CFD)required.The architecture is devised to enforce the initial and boundary conditions and incorporate the governing equations into the loss of the neural network.The optimization process is conducted by ADAM and L-BFGS-B methods with a feed forward-back propagation manner,to improve the accuracy and reliability of predicting the time evolution of nonlinear PDEs solutions.The results show that the proposed HCDD-PINN framework exhibits improved stability and accuracy compared to the PINN models that lack internal data.It effectively reduces the prediction error of the flow field,and accurately predicts the instantaneous aerodynamic forces,velocity and pressure fields of flapping wings.Moreover,the training time has accelerated by approximately four times.Additionally,the interested variables of the flow field at any instant can be rapidly obtained by the trained HCDD-PINN model,which is superior to the traditional CFD method that usually needs to be re-run.This study provides an effective alternative for solving flapping wing governing equations and even nonlinear partial differential equations(PDEs).
作者 胡付佳 周逸伦 刘小民 HU Fujia;ZHOU Yilun;LIU Xiaomin(School of Energy and Power Engineering,Xi’an Jiaotong University,Xi’an 710049,China;Xi’an Shaangu Power Co.,Ltd.,Xi’an 710075,China)
出处 《西安交通大学学报》 EI CAS CSCD 北大核心 2023年第11期194-205,共12页 Journal of Xi'an Jiaotong University
基金 陕西省重点研发计划资助项目(2023-YBGY-280)。
关键词 物理信息神经网络 数据驱动 深度学习 扑翼 气动性能 physics-informed neural network data-driven deep learning flapping wing aerodynamic performance
  • 相关文献

参考文献5

二级参考文献30

共引文献24

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部