中山大学航空航天学院,广东 广州 510006
聂滋森(1998年生),男;研究方向:计算流体力学;E-mail: niezsatg@gmail.com
姚清河(1980年生),男;研究方向:计算流体力学、并行算法、偏微分方程数值解;E-mail:yaoqhe@mail.sysu.edu.cn
纸质出版日期:2022-05-25,
网络出版日期:2021-07-19,
收稿日期:2020-07-21,
录用日期:2021-04-03
扫 描 看 全 文
聂滋森,陈辛阳,杨耿超等.基于U-Net的格子玻尔兹曼方法[J].中山大学学报(自然科学版),2022,61(03):101-109.
NIE Zisen,CHEN Xinyang,YANG Gengchao,et al.Lattice Boltzmann Method based on U-Net[J].Acta Scientiarum Naturalium Universitatis Sunyatseni,2022,61(03):101-109.
聂滋森,陈辛阳,杨耿超等.基于U-Net的格子玻尔兹曼方法[J].中山大学学报(自然科学版),2022,61(03):101-109. DOI: 10.13471/j.cnki.acta.snus.2020B085.
NIE Zisen,CHEN Xinyang,YANG Gengchao,et al.Lattice Boltzmann Method based on U-Net[J].Acta Scientiarum Naturalium Universitatis Sunyatseni,2022,61(03):101-109. DOI: 10.13471/j.cnki.acta.snus.2020B085.
格子玻尔兹曼方法(LBM)是一类广泛应用的介观尺度下的流体数值模拟方法。其缺陷在于,它对于计算资源的要求较高,一般情况下难以实现即时模拟。文章构造了一种新的基于U-Net的卷积神经网络(CNN,convolutional neural network)以对LBM进行加速,以一次卷积神经网络模型的运算代替原本需要进行多次的时间步迭代。对一系列层流绕柱流动的数值模拟进行试验,发现:该方法能够在保证计算精度较高的同时,相较于串行的LBM程序有约250倍加速,验证了该方法的有效性。
Lattice Boltzmann Method (LBM) is a kind of widely used mesoscopic fluid numerical simulation method. The drawback of LBM is the high computational cost, which causes difficulties in real-time simulation. In this work, we created a convolutional neural network (CNN) based on U-Net to accelerate LBM calculation. The purpose is to replace multiple LBM steps with one single operation of the CNN model. According to the result of our numerical experiment on a laminar flow around three obstacles in different geometries, the generated model can maintain the calculation at a high accuracy and accelerate the LBM calculation by over 250 times.
数据驱动模型LBM卷积神经网络神经网络结构代理模型
data-driven modelingLattice Boltzmann Methodconvolutional neural networkneural network architecturesurrogate model
BHATNAGAR P L, GROSS E P, KROOK M. A model for collision processes in gases. I. Small amplitude processes in charged and neutral one-component systems[J]. Physical Review ,1954,94(3): 511-525.
MOLEMAKER J, COHEN J M, PATEL S, et al. Low viscosity flow simulations for animation, eurographics/acm [C]//Siggraph Symposium on Computer Animation, 2008.
LENTINE M, WEN Z, FEDKI R. A novel algorithm for incompressible flow using only a coarse grid projection[J]. ACM Transactions on Graphics 29(4CD),2010, 114:1-9.
TREUILLE A, LEWIS A, POPOVIC Z. Model reduction for real-time fluids[J]. ACM Transactions on Graphics ,2006,25(3): 826-834.
LADICK L U, JEONG S H, SOLENTHALER B, et al. Data-driven fluid simulations using regression forests[J]. ACM Transactions on Graphics , 2015, 34(6):1-9.
SCHMIDHUBER J. Deep learning in neural networks: An overview[J]. Neural Netw, 2015, 61: 85-117.
YANG C, YANG X, XIAO X. Data-driven projection method in fluid simulation[J]. Comput Animat Virtual Worlds , 2016, 27(3/4): 415-424.
TOMPSON J, SCHLACHTER K, SPRECHMANN P, et al.Accelerating eulerian fluid simulation with convolutional networks [C]// Proceedings of the 34th International Conference on Machine Learning, 2017: 3424-3433.
GUO X, LI W, IORIO F. Convolutional neural networks for steady flow approximation, knowledge discovery and data mining [C]// Knowledge Discovery and Data Mining, 2016: 481-490.
HENNIGH O. Lat-Net: compressing lattice boltzmann flow simulations using deep neural networks[J]. arXiv:1705.09036.
RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[C]//International Conference on Medical Image Computing and Computer-assisted Intervention, 2015: 234-241.
HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition [C]// Computer Vision and Pattern Recognition, 2016: 770-778.
KINGMA D , BA J . Adam: A method for stochastic optimization[C]// The 3rd International Conference for Learning Representations , 2015.
NAIR V, HINTON G E. Rectified linear units improve restricted boltzmann machines vinod nair[C]// Proceedings of the 27th International Conference on Machine Learning. Haifa, Israel, 2010.
IOFFE S, SZEGEDY C. Batch normalization: Accelerating deep network training by reducing internal covariate shift[C]// In 32nd International Conference on Machine Learning, 2015: 448-456.
OBRECHT C, KUZNIK F, TOURANCHEAU B, et al. Multi-GPU implementation of the lattice Boltzmann method [J].Computers & Mathematics with Applications ,2013, 65(2): 252-261.
李承功. 流场的格子Boltzmann模拟及其GPU-CUDA并行计算 [D].大连: 大连理工大学, 2013.
0
浏览量
1
下载量
1
CSCD
关联资源
相关文章
相关作者
相关机构