Aerodynamic data fusion method based on physics-informed neural network

LIU Xia; FENG Wenhui; LIAN Feng; ZHANG Shuaiyu; ZHANG Guanghua; KONG Yinan; HAN Chongzhao

doi:10.7638/kqdlxxb-2023.0060

English Version Journal Login

您当前的位置：

首页 >

文章列表页 >

Aerodynamic data fusion method based on physics-informed neural network

更新时间：2026-05-28

- Aerodynamic data fusion method based on physics-informed neural network
- Vol. 41, Issue 8, Pages: 87-96(2023)
- 作者机构：
  
  1. 西安交通大学电子与信息学部自动化科学与工程学院,西安,710049
  2. 中国空气动力研究与发展中心,绵阳,621000
- 作者简介：
- 基金信息：
- DOI：10.7638/kqdlxxb-2023.0060
  CLC： TP181;V211.3
- Published：2023
- 稿件说明：
移动端阅览
刘霞, 冯文晖, 连峰, 等. 基于物理信息神经网络的气动数据融合方法[J]. 空气动力学学报, 2023, 41(8): 87−96.

LIU Xia, FENG Wenhui, LIAN Feng, et al. Aerodynamic data fusion method based on physics-informed neural network[J]. 2023, 41(8): 87-96.
刘霞, 冯文晖, 连峰, 等. 基于物理信息神经网络的气动数据融合方法[J]. 空气动力学学报, 2023, 41(8): 87−96. DOI： 10.7638/kqdlxxb-2023.0060

LIU Xia, FENG Wenhui, LIAN Feng, et al. Aerodynamic data fusion method based on physics-informed neural network[J]. 2023, 41(8): 87-96. DOI： 10.7638/kqdlxxb-2023.0060.

摘要

为了解决训练传统深度神经网络对大数据的依赖问题，气动数据中包含的物理结构信息需要被充分利用。物理信息神经网络（physics-informed neural network

PINN）是一种非监督的学习算法，采用深度神经网络直接逼近流场偏微分方程的解，因此适用于气动数据的建模。然而训练PINN时，损失函数反映的是抽样点处神经网络所拟合的偏微分方程值的偏差，对于复杂的非线性偏微分方程，这一偏差不能准确反映神经网络所拟合的函数与微分方程解函数的偏差，而且用神经网络拟合初始条件和边界条件时，不可避免存在拟合误差，误差随空间和时间累计，这使PINN的建模精度相比传统的模型没有优势。为了解决这些问题，本文把PINN与流场的计算流体动力学（Computational Fluid Dynamics

CFD）仿真结果进行融合，在流场抽样点处的损失函数中增加了PINN在该点的输出与流场在该点的CFD值偏差，从而提高了神经网络的建模精度。根据CFD仿真时使用的模型，融合方式采用瞬时模式或时均模式。测试结果表明该方法能够有效提高PINN的建模精度。

Abstract

To solve the issue that conventional deep neural networks(DNN) needs prohibitively vast amount of data

the structure information contained in aerodynamic data needs to be fully utilized. Physics-informed neural network (PINN) is an unsupervised learning algorithm that uses deep neural networks to directly approximate solutions to partial differential equations (PDEs) of the flow fields

thus suitable for modeling aerodynamic problems. However

during the training process of PINN

the loss function only depicts the deviation of PDE at the training sample points

and for complex nonlinear partial differential equations

this deviation cannot accurately reflect the error between the function fitted by neural networks and the solution of PDEs. Moreover

errors inevitably exist in the boundary and initial conditions when fitted with neural networks

which accumulate in space and time

making the modeling accuracy of PINN less favorable compared to traditional models. In order to address the above issue

the present study integrates PINN with computational fluid dynamics (CFD) simulation results of the flow fields

which adds the deviation between PINN output and CFD value in the loss function at sampling points of the flow field

thus improving the modeling accuracy of the neural network. According to the model used in CFD simulations

the fusion mode can adopt instantaneous mode or time-averaged mode. Experimental results suggest that the proposed method can effectively improve the modeling accuracy of PINN.

关键词

Keywords

references

DENG C, CHEN G, WANG W Z, et al. Aerodynamic data fusion algorithms based on aerodynamic model and uncertainly[J]. Acta Aerodynamica Sinica, 2022, 40(4): 117-123. (in Chinese)

WANG X, NING C J, WANG W Z, et al. Intelligent fusion method of multi-source aerodynamic data for flight tests[J]. Acta Aerodynamica Sinica, 2023, 41(2): 12-20. (in Chinese)

MOHAMMADI-AMIN M, ENTEZARI M M, ALIKHANI A. An efficient surrogate-based framework for aerodynamic database development of manned reentry vehicles[J]. Advances in Space Research, 2018, 62(5): 997-1014. DOI: 10.1016/j.asr.2018.06.022

CARY A W, CHAWNER J, DUQUE E P, et al. CFD vision 2030 road map: progress and perspectives[C]// AIAA Aviation 2021 Forum, Virtual Event. Reston, Virginia: AIAA, 2021: AIAA2021-2726. doi: 10.2514/6.2021-2726

PERDIKARIS P, RAISSI M, DAMIANOU A, et al. Nonlinear information fusion algorithms for data-efficient multi-fidelity modelling[J]. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 2017, 473(2198): 20160751. DOI: 10.1098/rspa.2016.0751

BENGIO Y, COURVILLE A, VINCENT P. Representation learning: a review and new perspectives[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 35(8): 1798-1828. DOI: 10.1109/TPAMI.2013.50

QIAN W Q, ZHAO T, HUANG Y, et al. Modeling and data mining of engineering airfoil aerodynamic characteristics[J]. Acta Aerodynamica Sinica, 2021, 39(6): 175-183. (in Chinese)

ZHANG J, ZHANG G B, CHENG Y Q, et al. A multi-task learning method for large discrepant aerodynamic data[J]. Acta Aerodynamica Sinica, 2022, 40(6): 64-72. (in Chinese)

HU L W, ZHANG J, XIANG Y, et al. Neural networks-based aerodynamic data modeling: a comprehensive review[J]. IEEE Access, 2020, 8: 90805-90823. DOI: 10.1109/ACCESS.2020.2993562

CUOMO S, DI COLA V S, GIAMPAOLO F, et al. Scientific machine learning through physics-informed neural networks: where we are and what’s next[J]. Journal of Scientific Computing, 2022, 92(3): 88. DOI: 10.1007/s10915-022-01939-z

RAISSI M, PERDIKARIS P, KARNIADAKIS G E. Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations[J]. Journal of Computational Physics, 2019, 378: 686-707. DOI: 10.1016/j.jcp.2018.10.045

RAISSI M, YAZDANI A, KARNIADAKIS G E. Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations[J]. Science, 367( 6481): 1026–1030. doi: 10.1126/science.aaw4741

CHENG C, ZHANG G T. Deep learning method based on physics informed neural network with resnet block for solving fluid flow problems[J]. Water, 2021, 13(4): 423. DOI: 10.3390/w13040423

JAGTAP A D, KAWAGUCHI K, KARNIADAKIS G E. Adaptive activation functions accelerate convergence in deep and physics-informed neural networks[J]. Journal of Computational Physics, 2020, 404: 109136. DOI: 10.1016/j.jcp.2019.109136

ZHU Y H, ZABARAS N, KOUTSOURELAKIS P S, et al. Physics-constrained deep learning for high-dimensional surrogate modeling and uncertainty quantification without labeled data[J]. Journal of Computational Physics, 2019, 394: 56-81. DOI: 10.1016/j.jcp.2019.05.024

HE Q Z, BARAJAS-SOLANO D, TARTAKOVSKY G, et al. Physics-informed neural networks for multiphysics data assimilation with application to subsurface transport[J]. Advances in Water Resources, 2020, 141: 103610. DOI: 10.1016/j.advwatres.2020.103610

GAO H, SUN L N, WANG J X. PhyGeoNet: physics-informed geometry-adaptive convolutional neural networks for solving parameterized steady-state PDEs on irregular domain[J]. Journal of Computational Physics, 2021, 428: 110079. DOI: 10.1016/j.jcp.2020.110079

KHARAZMI E, ZHANG Z Q, KARNIADAKIS G E M. hp -VPINNs: Variational physics-i nformed neural networks with domain decomposition[J ] . Computer Methods in Applied Mechanics and Engineering, 2021, 374: 113547. DOI: 10.1016/j.cma.2020.113547

YANG L, MENG X H, KARNIADAKIS G E. B-PINNs: Bayesian physics-informed neural networks for forward and inverse PDE problems with noisy data[J]. Journal of Computational Physics, 2021, 425: 109913. DOI: 10.1016/j.jcp.2020.109913

JAGTAP A D, KHARAZMI E, KARNIADAKIS G E. Conservative physics-informed neural networks on discrete domains for conservation laws: applications to forward and inverse problems[J]. Computer Methods in Applied Mechanics and Engineering, 2020, 365: 113028. DOI: 10.1016/j.cma.2020.113028

ALIAKBARI M, MAHMOUDI M, VADASZ P, et al. Predicting high-fidelity multiphysics data from low-fidelity fluid flow and transport solvers using physics-informed neural networks[J]. International Journal of Heat and Fluid Flow, 2022, 96: 109002. DOI: 10.1016/j.ijheatfluidflow.2022.109002

SIRIGNANO J, SPILIOPOULOS K. DGM: a deep learning algorithm for solving partial differential equations[J]. Journal of Computational Physics, 2018, 375: 1339-1364. DOI: 10.1016/j.jcp.2018.08.029

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Aerodynamic force measurement algorithm based on U-Net in maglev flight wind tunnel

Rapid design method of rocket aerodynamic shape based on conditional diffusion model

Intelligent prediction method of flow field and aerodynamic characteristics for two-dimensional blunt body combined sections in steady wind

Reconstruction of flow fields around large-span flexible PV array based on physics-informed deep learning

Deep neural network-based reduced-order modeling of high-speed airfoil flow field

Related Author

Deng Guoyao

Hu Zhe

Xu Dejun

Zhang Jun

Ni Zhangsong

Li Chuan

Chen Jiandong

He Lei

Related Institution

Chengdu Fluid Dynamics Innovation Center, Chengdu

State Key Laboratory of Aerodynamics, Mianyang

Computational Aerodynamics Institute of China Aerodynamics Research and Development Center, Mianyang

College of Computer Science, Sichuan University, Chengdu

School of Civil Engineering, Chongqing University, Chongqing

AI问答

⁰