Research on Flow Field Structure Prediction Method Driven by Physical Information

被引：0
|
作者：
Chen K. [1 ]
Ouyang H. [1 ]
Zhu Z. [1 ]
Hao J. [1 ]
Huang B. [1 ]
机构：
[1] School of Mechanical Engineering, Beijing Institute of Technology, Beijing
来源：
Beijing Ligong Daxue Xuebao/Transaction of Beijing Institute of Technology | 2023年 / 43卷 / 09期
关键词：
deep learning; flow field information prediction; physics-informed neural network (PINN); unsteady flow;
D O I：
10.15918/j.tbit1001-0645.2023.034
中图分类号：
学科分类号：
摘要：
Whether it is based on numerical simulation or physical experiments, the acquisition of high-precision flow field data is not only extremely limited, but also often accompanied by high costs. Existing methods cannot reconstruct a more refined flow field structure with limited data, which greatly restricts the design accuracy and design efficiency of related gas/hydrodynamic engineering problems. The proposed physics-informed neural network (PINN) framework can make the dilemma solved to a certain extent for the traditional data-driven neural network cannot deal with the sparse problem. In this paper, a sparse data-driven flow field reconstruction method was developed based on the PINN framework. Firstly, coupling physical information with the neural network and utilizing a small amount of data for training, the method was arranged to be able to output complete flow field data. Then, the prediction error mechanism was revealed, and the prediction ability of the method for different structural flow fields was discussed. The results show that, using only extremely limited flow field data, coupling NS equation, the physical information-driven neural network can achieve high-precision reconstruction of the entire flow field, and the vortex structure of the convective field can also be captured accurately. © 2023 Beijing Institute of Technology. All rights reserved.
引用
收藏
页码：950 / 961
页数：11
相关论文
共 25 条
  • [1] KRIZHEVSKY A, SUTSKEVER I, GEOFFREY E., ImageNet classification with deep convolutional neural networks[J], Communications of the ACM, 60, 6, (2017)
  • [2] MANNING C D, MIHAI S, JOHNET B, Et al., The Stanford CoreNLP Natural Language Processing Toolkit, Proceedings of 52nd Annual Meeting of the Association for Computational Linguistics, (2014)
  • [3] HINTON G, DENG L, YU D, Et al., Deep Neural Networks for Acoustic Modeling in Speech Recognition [J], IEEE Signal Processing Magazine, 29, 6, pp. 82-97, (2012)
  • [4] ZHANG Yang, ZHANG Tao, CHEN Jin, Et al., Research on network intrusion detection based on SMOTE algorithm and machine learning, Transactions of Beijing Institute of Technology, 39, 12, pp. 1258-1262, (2019)
  • [5] SHAO Shuai, WANG Meilin, CHEN Dongqing, Et al., Analysis of Android Application Component Exposure Vulnerability Based on Machine Learning, Transactions of Beijing Institute of Technology, 39, 9, pp. 974-977, (2019)
  • [6] BRENDEN M L, RUSLAN S, JOSHUA B., Human-level concept learning through probabilistic program induction[J], Science, 350, 6266, pp. 1332-1338, (2015)
  • [7] LING J, TEMPLETON J., Evaluation of machine learning algorithms for prediction of regions of high Reynolds averaged Navier Stokes uncertainty, Physics of Fluids, 27, 8, pp. 42032-42094, (2015)
  • [8] REN F, WANG C L, TANG H., Bluff body uses deep-reinforcement-learning trained active flow control to achieve hydrodynamic stealth, Physics of fluids, 33, 9, (2021)
  • [9] GAO C Q, ZHANG W W, KOU J Q, Et al., Active control of transonic buffet flow[J], Journal of fluid mechanics, 824, pp. 312-351, (2017)
  • [10] XIE T, GHIAASIAAN S M, Artificial neural network-based flow regime classification techniques for gas-liquid-fiber three-phase flows, Proceedings of ASME 2004 Heat Transfer/Fluids Engineering Summer Conference, (2004)
123