公开DNS与求解器基线的三维湍流算例:GNN直接推进与辅助粗网格求解验证,均匀各向同性湍流受迫与衰减设置
周期域均匀各向同性湍流的DNS参考数据、受迫/衰减生成方法与统计物理约束
合并两组中所有针对周期域HIT的数据与生成方法文献。共同点:为“能否拿到全域、连续时间的三维速度场”提供直接证据与操作依据——衰减HIT的网格/谱初始化与时间推进细节、积分尺度与耗散系数的修正(低波数缺失带来的评价偏差)、物理空间确定性受迫方案(由静止场生成统计定常受迫HIT并给出强迫参数与统计的关联)、受迫HIT的两点两点时间闭合(给出闭合目标形式与可检验关系)、滤波后粒子追踪与滤波尺度处理;同时提供物理可信性判据:Kolmogorov 4/5律及其有限雷诺数检验、EDQNM类谱闭合、速度梯度不变量拓扑统计、强迫标量混合、可压缩HIT中的激波片与Euler旋转角不变量,以及最小雷诺数下收敛性研究以确定可用算例规模。这组回答“初始条件/外力/黏性/时间间隔/网格信息是否齐备、如何构造滤波场与闭合目标、如何用多次统计相似实现留出独立测试轨迹”。
- Theory and Modelling of Isotropic Turbulence: From Incompressible through Increasingly Compressible Flows(C. Cambon, 2024, Atmosphere)
- Examination of hypotheses in the Kolmogorov refined turbulence theory through high-resolution simulations. Part 1. Velocity field(Lian-Ping Wang, Shi-Yi Chen, J. Brasseur, J. Wyngaard, 1996, Journal of Fluid Mechanics)
- Evolution of the velocity gradient invariants in homogeneous isotropic turbulence with an inverse energy cascade(Feng Liu, Zhuoyu Wu, Pengfei Lv, Wei Yang, Yi Zhou, 2023, The Physics of Fluids)
- Characteristics of Linearly-Forced Scalar Mixing in Homogeneous , Isotropic Turbulence(Phares L. Carroll, Siddhartha Verma, G. Blanquart, 2012, No journal)
- A Numerical Investigation of the Minimum Ignition Energy Requirement for Forced Ignition of Turbulent Droplet-laden Mixtures(V. Papapostolou, G. Ozel Erol, C. Turquand d'Auzay, N. Chakraborty, 2020, Combustion Science and Technology)
- Effect of shocklets on the velocity gradients in highly compressible isotropic turbulence(Jianchun Wang, Yipeng Shi, Lian-Ping Wang, Zuoli Xiao, Xian-Tu He, Shi-Yi Chen, 2011, Physics of Fluids)
- On the Euler rotation angle and axis of a subgrid-scale stress model(B. Saeedi, B.-C. Wang, S. Bose, 2012, No journal)
- Correction of the turbulent integral length scale and its impact on turbulent dissipation law(Yu-Jie Wang, Jun-Sheng Yang, Jie Wang, Yan-Hong Zhou, Feng Liu, 2025, Acta Physica Sinica)
- Spatially explicit deterministic forcing scheme for forced homogeneous isotropic turbulence(Jian Fang, Le Fang, Yifan Xu, Zhi X. Chen, 2025, Journal of turbulence)
- Two-point, two-time closures applied to forced isotropic turbulence(W. McComb, A. Quinn, 2003, Physica A-statistical Mechanics and Its Applications)
- Filtered particle tracking in isotropic turbulence and stochastic modeling of subgrid-scale dispersion(J. Pozorski, S. Apte, 2009, International Journal of Multiphase Flow)
开源可运行求解器、谱/高阶离散基础设施与可微分模拟框架
合并两组中全部求解器与数值方法文献,用于回答“公开连续轨迹不足时能否自行生成小规模参考数据、代码依赖与CFD经验门槛有多高”。共同点:Xcompact3D(不可压/可压缩、自带强迫与算例)、谱方法经典教材(周期域Fourier谱离散、去混叠与时间推进格式的直接依据)、JAX-Shock与flow differentiable(GPU加速、可微、便于与学习模块端到端耦合)、Lethe与Nektar++(高阶/hp单元开源并行框架,含Taylor–Green涡等基准算例)、高阶LES与并行有限元单方程SGS实现、非线性Galerkin(粗网格/降阶推进思想)以及新一代LES求解框架。这组直接支撑“从同一初始场启动DNS参考与粗网格解”的实现路径与单算例算力估计。
- Xcompact3D: An open-source framework for solving turbulence problems on a Cartesian mesh(Paul Bartholomew, G. Deskos, R. Frantz, F. Schuch, E. Lamballais, S. Laizet, 2020, SoftwareX)
- Spectral methods for fluid dynamics(M. Hussaini, Thomas A. Zang, 1987, No journal)
- A differentiable, shock-capturing neural solver for compressible flow simulation(Bo Zhang, 2026, The Physics of Fluids)
- IC S THE NONLINEAR GALERKIN METHOD : A MULTI-SCALE METHOD APPLIED TO THE SIMULATION OF HOMOGENEOUS TURBULENT FLOWS(A. Debussche, T. Dubois, R. Temam, No journal)
- on: LEX v1.4: A New Large-Eddy Simulation Model in JAX with GPU Acceleration and Automatic Differentiation(No journal)
- Lethe: An open-source parallel high-order adaptative CFD solver for incompressible flows(B. Blais, Lucka Barbeau, Valérie Bibeau, Simon Gauvin, Toni El Geitani, Shahab Golshan, Raj Kamble, Ghazaleh Mirakhori, J. Chaouki, 2020, SoftwareX)
- Nektar++: An open-source spectral/hp element framework(C. Cantwell, D. Moxey, A. Comerford, A. Bolis, G. Rocco, G. Mengaldo, D. De Grazia, S. Yakovlev, Jean-Eloi W. Lombard, D. Ekelschot, Bastien E. Jordi, Hui Xu, Yumnah Mohamied, C. Eskilsson, B. Nelson, P. Vos, C. Biotto, R. Kirby, S. Sherwin, 2015, Computer Physics Communications)
- Applications and Recent Developments of the Open-Source Computational Fluid Dynamics High-Fidelity Spectral/Hp Element Framework Nektar++ for Turbomachinery Configurations(Guglielmo Vivarelli, J. Isler, F. Montomoli, C. Cantwell, S. Sherwin, Yuri Frey-Marioni, Raul Vázquez-Díaz, 2024, Volume 12C: Turbomachinery — Design Methods and CFD Modeling for Turbomachinery; Ducts, Noise, and Component Interactions)
- Development of High Order LES Solver for Heat Transfer Applications Based on the Open Source OpenFOAM Framework(L. Mangani, D. R. Launchbury, E. Casartelli, G. Romanelli, 2015, No journal)
- Parallel FEM LES with one-equation subgrid-scale model for incompressible flows(Xian-Xiang Li, Chun-Ho Liu, D. Leung, 2010, International Journal of Computational Fluid Dynamics)
- ΦFlow: Differentiable Simulations for PyTorch, TensorFlow and Jax(P. Holl, N. Thürey, 2024, International Conference on Machine Learning)
传统亚格子闭合模型的模型形式、动态过程与参数辨识基线
合并两组中偏“闭合模型本身”的文献,构成学习模型的对照基线。共同点:单方程LES及其后验测试、Langevin型LES闭合、动态确定系数的单方程SGS模型、二阶矩/湍流输运闭合的使用评估、三维小尺度参数化、模型参数与网格参数的辨识,以及亚格子应力/标量通量的模型形式、动态过程与反散射表征、尺度相似小尺度重构。这些给出“close-form 基线”的具体表达式与标定方式,是与数据驱动闭合做同条件对比时的必要底稿。
- a Posteriori Tests of One-Equation les Modeling of Rotating Turbulence(Hao-Yuan Lu, C. Rutland, Leslie M. Smith, 2008, International Journal of Modern Physics C)
- Fe b 20 06 A LES-Langevin model for turbulence(J. Laval, B. Dubrulle, 2008, No journal)
- One-Equation Subgrid Scale Model Using Dynamic Procedure for the Energy Production(T. Kajishima, Takayuki Nomachi, 2006, Journal of Applied Mechanics)
- Evaluation of the Use of Second Generation Wavelets in the Coherent Vortex Simulation Approach(D. Goldstein, O. Vasilyev, A. Wray, R. Rogallo, 2000, No journal)
- Parameterization of Small Scales of Three-Dimensional Isotropic Turbulence Utilizing Spectral Closures(J. Chollet, M. Lesieur, 1981, Journal of the Atmospheric Sciences)
- Identification of model and grid parameters for incompressible turbulent flows(Xiaoqin Zhang, 2007, No journal)
- Subgrid-scale scalar flux modelling based on optimal estimation theory and machine-learning procedures(Antoine Vollant, G. Balarac, C. Corre, 2017, Journal of Turbulence)
- Development of a new dynamic procedure for the Clark model of the subgrid-scale scalar flux using the concept of optimal estimator(Yannick Fabre, G. Balarac, 2011, Physics of Fluids)
- On the representation of backscatter in dynamic localization models(D. Carati, S. Ghosal, P. Moin, 1995, Physics of Fluids)
- Representing the Small Scales of Turbulence by Periodic Box Homogeneous Isotropic Turbulence Simulations(G. T. Zachariah, H. V. D. van den Akker, 2023, Flow Turbulence and Combustion)
滤波定义与a priori/a posteriori评价流程、自适应隐式LES及数据驱动闭合综述
这组聚焦“评价方法论”而非某个具体模型,回答“能否复现论文中的滤波、时间推进与评价条件”。共同点:从DNS显式滤波得到粗网格场做a priori检验、fDNS与DNS成对数据的模型评估、数据驱动非局域动态模型的可解释评估、无模型的自适应小波隐式LES(滤波即分辨率自适应)、以及数据驱动湍流闭合的系统综述与混合输运控制函数。它们提供误差度量口径、训练/检验切分方式和评价链条模板。
- A priori evaluation of large eddy simulation subgrid-scale scalar flux models in isotropic passive-scalar and anisotropic buoyancy-driven homogeneous turbulence(N. Ghaisas, S. Frankel, 2014, Journal of Turbulence)
- A DNS Evaluation of a Novel Multiple Mapping Conditioning Model with a Mixture Fraction-like Reference Variable(Zi-Sen Li, A. Wehrfritz, B. Savard, E. Hawkes, 2018, No journal)
- A Data-Driven Dynamic Nonlocal LES Model for Turbulent Flows(S. Seyedi, M. Zayernouri, 2021, No journal)
- Wavelet-based adaptive implicit LES/under-resolved DNS of forced isotropic turbulence(G. De Stefano, 2025, Theoretical and Computational Fluid Dynamics)
- Turbulence closure modeling with data–driven techniques: Influence of neural network hyperparameters and training procedure on generalizability(S. Taghizadeh, F. Witherden, Y. Hassan, S. Girimaji, 2021, No journal)
- Variational calculus in hybrid turbulence transport models with passive scalar(B. Chaouat, R. Schiestel, 2024, Journal of turbulence)
混合LES/RANS、合成湍流耦合、粗网格一致性与数据同化系数标定
保留为独立组:这组文献关注的不是闭合形式,而是“粗网格求解本身的不确定性与低成本替代”。共同点:URANS呈现类LES行为、混合LES/RANS及其与粗网格之间的界面与一致性问题、高雷诺数流动模拟的实现约束、基于数据同化的模型系数优化,以及LES与合成湍流生成耦合(在缺少现成入口/初始全场的条件下低成本构造算例的途径)。它们解释了为何学习闭合必须与求解器在线耦合验证,并指出网格分辨率、界面处理、系数标定与合成入口是需自行开发的环节。
- ON URANS SOLUTIONS WITH LES-LIKE BEHAVIOUR(P. Neittaanmäki, T. Rossi, K. Majava, O. Pironneau, W. Rodi, M. Strelets, 2004, No journal)
- ISSUES IN HYBRID LES-RANS AND COARSE GRID LES OF SEPARATED FLOWS(Von Terzi, No journal)
- High Reynolds number flow simulations using the localized dynamic subgrid-scale model(S. Menon, W.-W. Kim, 1996, No journal)
- Data assimilation-based coefficient optimization for the mixed subgrid-scale model in large-eddy simulation of turbulence(Yun-Peng Wang, Ze-Long Yuan, 2023, No journal)
- A Coupled LES-Synthetic Turbulence Method for Jet Noise Prediction(Joshua D. Blake, A. Sescu, D. Thompson, Y. Hattori, 2022, Aerospace)
机器学习/图神经网络亚格子与雷诺应力闭合模型的接入实例
合并两组中“以闭合量预测为目标、嵌入求解器”的文献,回答“学习模型如何插进粗网格求解流程”。共同点:多尺度GNN的SGS应力建模(HIT/混合层/槽道多雷诺数DNS滤波数据、Gaussian与box核、先验与后验测试)、各向异性GNN、保持可实现性的E(3)等变GNN与求解器在环微调、直接嵌入RANS的张量基神经网络闭合、面向S-A模型的ANN特征选择与泛化增强、以Burgers方程与backward-facing step为对象的NN-SGS先验/后验验证、以DNS/FDNS配对资料重构滤波丢失小尺度的ANN、以及进化神经网络做显式闭合的符号回归与无监督-有监督联合训练流水线。这组给出输入特征、滤波尺度、训练/验证集切分与在线耦合评价的可复现条件。
- Multi-scale graph neural network modeling of near-wall turbulence with robust flow generalizability(L. Hutin, Ye Wang, M. Tanahashi, 2026, The Physics of Fluids)
- The anisotropic graph neural network model with multiscale and nonlinear characteristic for turbulence simulation(Qiang Liu, Wei Zhu, Xi-Yu Jia, Feng Ma, Jun Wen, Yixiong Wu, Kuang-Qi Chen, Zhenhai Zhang, Shuang Wang, 2024, Computer Methods in Applied Mechanics and Engineering)
- Realisable E(3)-equivariant graph networks with near-wall reinforcement learning for turbulence closure(Fredrik Dubay Myklebust, Y. Li, 2026, Transport Phenomena)
- Data-driven turbulence closure using a tensor basis neural network without reliance on baseline models(Yao-hung Tsai, Yi Wang, Che-Jung Cheng, Chen-Yen Hung, Yi-En Chou, Kuan-Lin Li, Chien-Chou Tseng, R. Chern, Yi-Ju Chou, 2026, Journal of Fluid Mechanics)
- Generalization enhancement of artificial neural network for turbulence closure by feature selection(Linyang Zhu, Wei-Wei Zhang, Guo-Hua Tu, 2021, Advances in Aerodynamics)
- Development of subgrid-scale model for LES of Burgers turbulence with large filter size(A. G. Tabe Jamaat, B. Y. Hattori, 2022, The Physics of Fluids)
- Large eddy simulation of turbulent flow over a backward-facing step with a neural-network-based subgrid-scale model(Jonghwan Park, H. Choi, 2025, The Physics of Fluids)
- An artificial neural network model for recovering small-scale velocity in large-eddy simulation of isotropic turbulent flows(Jiangtao Tan, Guodong Jin, 2024, The Physics of Fluids)
- Unsupervised and supervised machine learning pipeline for super-resolution-based subgrid scale modelling in coarse-grid large-eddy simulations(Soju Maejima, S. Kawai, 2025, Journal of Fluid Mechanics)
- Implicit modeling of equivariant tensor basis with Euclidean turbulence closure neural network(Grzegorz Kaszuba, Tomasz Krakowski, Bartosz Ziegler, A. Jaszkiewicz, P. Sankowski, 2025, The Physics of Fluids)
- Evolutionary neural networks for learning turbulence closure models with explicit expressions(Hao-Chen Li, Yaomin Zhao, Fabian Waschkowski, R. Sandberg, 2024, The Physics of Fluids)
GNN/深度学习直接推进速度场、时空rollout与超分辨重建路线
保留并扩展为独立组,直接对应“GNN直接推进”的技术路线及其风险。共同点:GNN湍流建模与基于GNN的圆柱尾迹/明渠时空预测(含结构函数与Kolmogorov标度检验)、以Fourier投影保证无散、速度+涡量双项损失的三维HIT深度学习预测(跨Reλ外推、按子体积拼接、有限提前时间评估)、以受迫HIT DNS为数据的条件超分辨GAN(以SGS动能作条件)、CNN/并行上采样从粗网格恢复小尺度、LSTM/GRU时序速度场预测(误差累积)以及深度学习湍流模拟的综述与研究指南。这组用于确定算例规模、多步推进稳定性与评价指标(积分时间尺度、误差增长率)。
- Graph neural network based model of hydrodynamic closure laws in non-spherical particle-laden flows(L. Jbara, Aashish Goyal, A. Wachs, 2025, Computers & Fluids)
- Spatiotemporal prediction of bluff-body hydrodynamics using the graph neural network(Anjan Samanta, P. Gopmandal, Sankar Sarkar, 2025, The Physics of Fluids)
- Physics-constrained deep learning for prediction of three-dimensional homogeneous isotropic turbulence(N. Tsoy, Changhoon Lee, 2026, The Physics of Fluids)
- A conditional deep learning model for super-resolution reconstruction of small-scale turbulent structures in particle-Laden flows(H. Tofighian, J. Denev, N. Kornev, 2024, The Physics of Fluids)
- Generating small-scale turbulence in jet flows by training with a machine learning method(S. Choi, Yeulwoo Kim, Jin Hwan Hwang, 2026, The Physics of Fluids)
- Predicting high-resolution turbulence details in space and time(Kai-Yi Bai, C. Wang, M. Desbrun, Xiao-Pei Liu, 2021, ACM Transactions on Graphics)
- The Capability of Recurrent Neural Networks to Predict Turbulence Flow via Spatiotemporal Features(R. Hassanian, M. Riedel, Lahcen Bouhlali, 2022, 2022 IEEE 10th Jubilee International Conference on Computational Cybernetics and Cyber-Medical Systems (ICCC))
- Current and emerging deep-learning methods for the simulation of fluid dynamics(M. Lino, Stathi Fotiadis, A. Bharath, C. Cantwell, 2023, Proceedings of the Royal Society A)
- Guided Research Report Turbulence Modelling with JAX-SPH(MohammadAmin SaeidyPour, No journal)
基准算例的网格收敛检验与数值一致性/可重复性规范
保留为独立组,回答“如何判断算例与模型是否被真正复现”:三维基准算例的网格收敛性研究给出分辨率与统计量收敛的检验方式;数值一致的数据驱动建模摘要集强调训练/测试数据与求解器设置的一致性、可重复性与数据-代码-评价口径对齐。为独立测试轨迹划分、误差度量与结果可比性提供评价协议层面的支撑。
- Grid Convergence for Three Dimensional Benchmark Turbulent Flows(B. Diskin, W. Anderson, Mohagna J. Pandya, C. Rumsey, James L. Thomas, Yi Liu, H. Nishikawa, 2018, No journal)
- Session H Abstracts Numerically Consistent Data-Driven Subgrid-Scale Model for Large-Eddy Simulation(Michael Lawrence, Garcia Mentor, H. J. Bae, 2023, No journal)
将两套初始化分组按“功能角色”而非文献来源合并为8个并列板块,覆盖全部65篇文献:①HIT的DNS参考数据、受迫/衰减生成方法与统计物理约束(11篇)——回答全域连续时间三维速度场能否获得、滤波场与闭合目标如何构造、留出测试轨迹如何实现,并给出4/5律、谱闭合、不变量统计等物理可信性标尺;②开源可运行求解器与谱/高阶离散、可微分模拟框架(11篇)——回答现成轨迹不足时自行生成小规模参考数据的代码依赖与算力;③传统亚格子闭合模型形式、动态过程与参数辨识基线(10篇)——提供同条件对比的close-form基线;④滤波定义与a priori/a posteriori评价流程、自适应隐式LES及数据驱动闭合综述(6篇)——提供误差口径与评价链条模板;⑤混合LES/RANS、合成湍流耦合、粗网格一致性与数据同化标定(5篇)——揭示粗网格求解的不确定性来源与低成本替代;⑥ML/GNN闭合模型接入实例(11篇)——给出输入特征、滤波尺度、训练切分与求解器在环耦合方式;⑦GNN/深度学习直接推进与超分辨重建、时空rollout(9篇)——对应GNN直接推进路线及误差累积风险;⑧网格收敛检验与数值一致性规范(2篇)——支撑独立测试集划分与结果可重复性。整体链条为“数据—代码—基线—误差来源—学习闭合—直接推进—评价规范”,各板块相互并列、互不包含,可直接映射到调研报告的三类核查条目(数据可得性与闭合目标构造、求解器与基线复现、规模/算力/经验成本比较)。
合并两组中所有针对周期域HIT的数据与生成方法文献。共同点:为“能否拿到全域、连续时间的三维速度场”提供直接证据与操作依据——衰减HIT的网格/谱初始化与时间推进细节、积分尺度与耗散系数的修正(低波数缺失带来的评价偏差)、物理空间确定性受迫方案(由静止场生成统计定常受迫HIT并给出强迫参数与统计的关联)、受迫HIT的两点两点时间闭合(给出闭合目标形式与可检验关系)、滤波后粒子追踪与滤波尺度处理;同时提供物理可信性判据:Kolmogorov 4/5律及其有限雷诺数检验、EDQNM类谱闭合、速度梯度不变量拓扑统计、强迫标量混合、可压缩HIT中的激波片与Euler旋转角不变量,以及最小雷诺数下收敛性研究以确定可用算例规模。这组回答“初始条件/外力/黏性/时间间隔/网格信息是否齐备、如何构造滤波场与闭合目标、如何用多次统计相似实现留出独立测试轨迹”。相关文献:C. Cambon et. al, 2024 等 11 篇文献
合并两组中全部求解器与数值方法文献,用于回答“公开连续轨迹不足时能否自行生成小规模参考数据、代码依赖与CFD经验门槛有多高”。共同点:Xcompact3D(不可压/可压缩、自带强迫与算例)、谱方法经典教材(周期域Fourier谱离散、去混叠与时间推进格式的直接依据)、JAX-Shock与flow differentiable(GPU加速、可微、便于与学习模块端到端耦合)、Lethe与Nektar++(高阶/hp单元开源并行框架,含Taylor–Green涡等基准算例)、高阶LES与并行有限元单方程SGS实现、非线性Galerkin(粗网格/降阶推进思想)以及新一代LES求解框架。这组直接支撑“从同一初始场启动DNS参考与粗网格解”的实现路径与单算例算力估计。相关文献:Paul Bartholomew et. al, 2020 等 11 篇文献
合并两组中偏“闭合模型本身”的文献,构成学习模型的对照基线。共同点:单方程LES及其后验测试、Langevin型LES闭合、动态确定系数的单方程SGS模型、二阶矩/湍流输运闭合的使用评估、三维小尺度参数化、模型参数与网格参数的辨识,以及亚格子应力/标量通量的模型形式、动态过程与反散射表征、尺度相似小尺度重构。这些给出“close-form 基线”的具体表达式与标定方式,是与数据驱动闭合做同条件对比时的必要底稿。相关文献:Hao-Yuan Lu et. al, 2008 等 10 篇文献
这组聚焦“评价方法论”而非某个具体模型,回答“能否复现论文中的滤波、时间推进与评价条件”。共同点:从DNS显式滤波得到粗网格场做a priori检验、fDNS与DNS成对数据的模型评估、数据驱动非局域动态模型的可解释评估、无模型的自适应小波隐式LES(滤波即分辨率自适应)、以及数据驱动湍流闭合的系统综述与混合输运控制函数。它们提供误差度量口径、训练/检验切分方式和评价链条模板。相关文献:N. Ghaisas et. al, 2014 等 6 篇文献
保留为独立组:这组文献关注的不是闭合形式,而是“粗网格求解本身的不确定性与低成本替代”。共同点:URANS呈现类LES行为、混合LES/RANS及其与粗网格之间的界面与一致性问题、高雷诺数流动模拟的实现约束、基于数据同化的模型系数优化,以及LES与合成湍流生成耦合(在缺少现成入口/初始全场的条件下低成本构造算例的途径)。它们解释了为何学习闭合必须与求解器在线耦合验证,并指出网格分辨率、界面处理、系数标定与合成入口是需自行开发的环节。相关文献:P. Neittaanmäki et. al, 2004 等 5 篇文献
合并两组中“以闭合量预测为目标、嵌入求解器”的文献,回答“学习模型如何插进粗网格求解流程”。共同点:多尺度GNN的SGS应力建模(HIT/混合层/槽道多雷诺数DNS滤波数据、Gaussian与box核、先验与后验测试)、各向异性GNN、保持可实现性的E(3)等变GNN与求解器在环微调、直接嵌入RANS的张量基神经网络闭合、面向S-A模型的ANN特征选择与泛化增强、以Burgers方程与backward-facing step为对象的NN-SGS先验/后验验证、以DNS/FDNS配对资料重构滤波丢失小尺度的ANN、以及进化神经网络做显式闭合的符号回归与无监督-有监督联合训练流水线。这组给出输入特征、滤波尺度、训练/验证集切分与在线耦合评价的可复现条件。相关文献:L. Hutin et. al, 2026 等 11 篇文献
保留并扩展为独立组,直接对应“GNN直接推进”的技术路线及其风险。共同点:GNN湍流建模与基于GNN的圆柱尾迹/明渠时空预测(含结构函数与Kolmogorov标度检验)、以Fourier投影保证无散、速度+涡量双项损失的三维HIT深度学习预测(跨Reλ外推、按子体积拼接、有限提前时间评估)、以受迫HIT DNS为数据的条件超分辨GAN(以SGS动能作条件)、CNN/并行上采样从粗网格恢复小尺度、LSTM/GRU时序速度场预测(误差累积)以及深度学习湍流模拟的综述与研究指南。这组用于确定算例规模、多步推进稳定性与评价指标(积分时间尺度、误差增长率)。相关文献:L. Jbara et. al, 2025 等 9 篇文献
保留为独立组,回答“如何判断算例与模型是否被真正复现”:三维基准算例的网格收敛性研究给出分辨率与统计量收敛的检验方式;数值一致的数据驱动建模摘要集强调训练/测试数据与求解器设置的一致性、可重复性与数据-代码-评价口径对齐。为独立测试轨迹划分、误差度量与结果可比性提供评价协议层面的支撑。相关文献:B. Diskin et. al, 2018 等 2 篇文献
总计 65 篇相关文献
A new graph neural network (GNN) subgrid-scale (SGS) stress modeling framework based on multi-scale self-similarity input features is investigated as an improvement in generalizability over a previous GNN model developed by the authors' group. Training and a priori testing are conducted with an incompressible direct numerical simulation (DNS) database comprising homogeneous isotropic turbulence (HIT), turbulent mixing layer, and turbulent channel flow (CH) cases, each across five Reynolds numbers. DNS fields are filtered at two different scales using both Gaussian and box kernels to create a diverse large eddy simulation training dataset. The multi-scale Leonard stress tensor and spatial velocity fluctuations are used as inputs to capture scale interactions from a broader portion of the turbulent energy spectrum. Galilean invariance and scale equivariance are ensured by input definition and nondimensionalization, respectively. Randomized data augmentation is used to improve model robustness to changes in orientation. Spline-based graph-convolution layers enable non-local feature extraction while maintaining model compatibility with unstructured grids. A priori analysis of four GNN SGS models shows that graph convolution and multi-scale input features provide significant improvement in accuracy for incompressible flows at Reynolds numbers below, within and above the training range, with particular impact in the near-wall region. Model generalizability to compressible reactive flows is high except in a wall-bounded case. In a posteriori testing with incompressible turbulent CH, one multi-scale GNN was found to better match reference DNS profiles than conventional eddy-viscosity models, with computational overhead comparable to that of a dynamic k-equation model.
No abstract available
A deterministic forcing scheme, formulated in physical space, is proposed to generate turbulence from a quiescent initial flow field, by referencing the Taylor–Green vortex. The scheme is analytically defined, ensuring isotropy and divergence-free characteristics. A series of direct numerical simulations (DNS) of forced homogeneous isotropic turbulence across varying Reynolds numbers is performed to benchmark the forcing method. The resulting turbulence is fully developed and statistically isotropic. The forcing scheme is further applied to DNS of compressible turbulence at both low and high Mach numbers. At low Mach numbers, the simulations reach statistically steady states consistent with the incompressible turbulence, while at high Mach numbers, strong compressibility effects are present, characterised by significant dilation and the formation of shocklets. Additionally, correlations between turbulence statistics and forcing parameters are established, providing further guidance for the application of such a forcing scheme to generate isotropic turbulence.
Three-dimensional turbulence is inherently chaotic and multiscale, posing significant computational challenges for traditional simulation methods, such as direct numerical simulation (DNS), large eddy simulation, and Reynolds-averaged Navier–Stokes models. Although deep learning (DL) offers a promising data-driven alternative, most existing studies have focused on two-dimensional flows, limiting the physical fidelity and applicability to realistic turbulent processes. This article presents a physically constrained DL framework for predicting three-dimensional homogeneous isotropic turbulence. A neural network was augmented with a Fourier projection to enforce incompressibility and was trained using a two-term loss function that accounted for both the velocity and vorticity fields. The model was trained on DNS data at Reλ = 58, and its generalizability was evaluated on higher-Reynolds-number datasets at Reλ = 92, 160, and 433, where the larger domain was divided into subvolumes matching the training size, the flow fields of which were predicted separately, and then concatenated to reconstruct the full domain. Temporal sampling over several finite lead times enabled the forecasting capability of the model to be evaluated. The proposed approach accurately reproduces key turbulence statistics, resolves three-dimensional vortex structures, and maintains divergence-free velocity fields across prediction horizons up to a lead time of 0.3 Eulerian integral timescale. The results demonstrate that combining data-driven DL with physics-based constraints enables the efficient and reliable prediction of three-dimensional turbulence, balancing computational efficiency and physical fidelity.
An investigation of topological features of homogeneous isotropic turbulence with an inverse energy cascade is performed by using a direct numerical simulation (DNS). The inverse energy cascade is induced by reversing the velocity field (i.e., under the transformation from ui to -ui) of a freely decaying isotropic turbulence. In the backward energy transfer process, the joint probability density function (PDF) of Q and R exhibits a novel shape which seems to be symmetric with the well-known teardrop shape about the Q-axis ( Q and R are the second and third invariants of the velocity gradients tensor, respectively). The unexpected shape of the joint PDF of Q and R indicates that the turbulence with an inverse energy cascade is dominated by vortex compression and tubelike structures. The PDF of the intermediate eigenvalue of the strain-rate tensor in the inverse energy cascade is negatively skewed, which is opposite to the universal feature of the forward energy cascade. Nevertheless, the preferential alignment of vorticity with the intermediate eigenvector is rather robust, no matter the direction of the energy transfer is forward or backward. In addition, it is universal that the vorticity is mostly perpendicular to the eigenvector which corresponds to a strain-rate eigenvalue with the maximum absolute value among the three eigenvalues. Since the velocity gradient invariants are closely related to the local flow topology, the numerical results reported in this work are expected to shed light on the intrinsic dynamics and mechanisms of inverse energy cascade.
The wavelet-based adaptive implicit eddy-resolving simulation approach is demonstrated for linearly forced homogeneous isotropic turbulence at moderate Taylor-Reynolds number. The wavelet-filtered incompressible Navier-Stokes equations are solved using the adaptive multilevel wavelet collocation method for elliptic problems without employing any explicit modeling for the unclosed terms. Instead, the energy dissipation induced by the built-in low-pass filtering associated with the adaptive numerical scheme is effectively exploited, thus mimicking the effect of unresolved subgrid-scale coherent flow structures on the dynamics of the resolved ones. The results of various numerical simulations, performed at different spatial resolutions, and with different filtering strengths, prove both the feasibility and efficacy of the proposed no-model approach for wall-free turbulence, while addressing the corresponding limits of application.
No abstract available
ABSTRACT The effects of droplet diameter, overall (i.e. liquid and gaseous phases) equivalence ratio, and turbulence intensity on the variation of the Minimum Ignition Energy (MIE) for localized forced ignition of uniformly dispersed mono-sized n-heptane droplet-laden mixtures under homogeneous isotropic decaying turbulence have been analyzed based on Direct Numerical Simulation (DNS) data. The MIE was evaluated just for (i) obtaining thermal runway irrespective of the fate of the resulting flame kernel, and (ii) also for a successful self-sustained flame propagation without the assistance of an external energy source following the energy deposition by the ignitor. It has been found that the MIE requirement increases with increasing turbulence intensity and this trend for the MIE increase is especially significant for large values of turbulence intensity. The MIE requirement increases with increasing initial droplet diameter and with decreasing overall equivalence ratio. The MIE requirements for droplet-laden mixtures have been found to be greater than the corresponding value for homogeneous mixture with the same nominal values of initial turbulence intensity and equivalence ratio for the parameter range considered here. This behavior arises due to the deposited energy being partially utilized to supply the latent heat of evaporation and also due to the predominantly fuel-lean composition of the gaseous flammable mixture. This tendency of obtaining fuel-lean mixture strengthens with increasing (decreasing) initial droplet diameter (overall equivalence ratio). It has also been demonstrated that combustion takes place predominantly in the fuel-lean premixed mode although there is a finite probability of having non-premixed combustion in all cases. Moreover, there is a small probability of fuel-rich combustion occurring for small droplets, especially under fuel-rich overall equivalence ratios. The stochastic nature of the ignition event has been demonstrated by considering different realizations of statistically similar turbulent flow fields. The conditions giving rise to a successful thermal runaway/self-sustained flame propagation have been identified by a detailed analysis of the energy budget.
No abstract available
No abstract available
No abstract available
No abstract available
No abstract available
No abstract available
No abstract available
No abstract available
No abstract available
No abstract available
Abstract Accurate prediction of turbulent flows remains a critical challenge in computational fluid dynamics, particularly for complex, non-equilibrium environments where traditional Reynolds-averaged Navier–Stokes (RANS) models fail to capture deep anisotropy and nonlocal interactions. While data-driven turbulence modeling, such as tensor basis neural networks (TBNNs) offers a promising alternative, these approaches frequently struggle with physical realisability and the inherent nonlocality of turbulent transport. To address these limitations, this work introduces a realisability-preserving E(3)-equivariant graph neural network (E(3)-GNN) for Reynolds-stress anisotropy closure. By leveraging equivariant message passing and a spherical-tensor formulation, the architecture explicitly captures nonlocal neighborhood interactions while guaranteeing Galilean invariance. Furthermore, a stabilized positive semi-definite projection ensures that all predicted states permanently reside within the physically admissible Lumley triangle, completely eliminating the unphysical predictions that traditionally destabilize coupled solvers. Because accurate a priori stress predictions do not guarantee a posteriori stability during active RANS simulation, we additionally propose a solver-in-the-loop fine-tuning methodology using Group Relative Policy Optimization (GRPO). This step directly optimizes a scalar production-correction policy within the active RANS environment to capture dynamic energy propagation. Evaluated on the parameterised periodic hill benchmark family, with training and testing performed on distinct hill geometries, our E(3)-GNN achieves strict physical realisability by construction. Through GRPO fine-tuning, the fully coupled model successfully recovers the intense turbulent kinetic energy (k) peaks and mean velocity profiles in the separated shear layer that purely supervised models fail to capture, bridging the gap between offline regression accuracy and stable, coupled solver performance.
Improved turbulence models are necessary for achieving more accurate solutions in Reynolds-averaged Navier–Stokes (RANS) simulations. RANS is widely used in various engineering applications, and enhancing its accuracy is crucial for geometry design and control applications. With the increasing availability of high-fidelity datasets, machine learning (ML) techniques offer the opportunity for data-driven inference of RANS equation closure terms. By Pope's theoretical analysis, ML turbulence closure models for RANS simulations are often used alongside physics-informed preprocessing of equivariant tensor basis. In this work, we replace this established approach by designing an equivariant turbulence model with Euclidean Neural Networks (e3nn). We demonstrate that Pope's tensor basis is a special case of our model. We test the model on the periodic hills flow case dataset. Our approach significantly improves the prediction of anisotropic components of Reynolds stresses, leading to more accurate modeling of the flow field when integrated into RANS through a single injection.
No abstract available
No abstract available
Abstract We propose a novel machine-learning-based turbulence closure framework in which a tensor basis neural network (TBNN) is directly embedded into a Reynolds-averaged Navier–Stokes (RANS) formulation, eliminating reliance on traditional baseline turbulence models. The TBNN is trained to predict the Reynolds stress anisotropic tensor from local invariant inputs and geometry-informed features, including stream function and velocity potential. Its output is processed by a regression model that generates an optimised eddy viscosity field, which is then integrated into the RANS equations as a zero-equation turbulence closure. The framework is evaluated on three turbulent flows over complex geometries: a wavy-bottom channel, a smoothed step and a backward-facing step. Incorporating geometry-informed features significantly enhances model robustness, yielding numerically stable and convergent solutions across all cases. The predicted velocity fields and turbulence distributions closely match large eddy simulation (LES) data, confirming the accuracy of the proposed approach and demonstrating its ability to operate independently of conventional turbulence closures.
A predictive model has been developed based on experimental data obtained from open channel flow past two horizontal cylinders, one above the other, using a graph neural network framework. Upon investigating the error plots, it is evident that the decreasing mean square error indicates the model's capability to predict near accurate data. Furthermore, second- and third-order structure functions have been illustrated to validate Kolmogorov's 2/3 and 4/5 laws for both original and predicted data, which were found to be a little inconsistent near the two cylinders in the near-wake zone. To investigate this inconsistency further, scaling exponents from extended self-similarity have been calculated and plotted with the conventional Kolmogorov 1941 theory in order to visualize the deviation. It is found that strong anisotropic flow occurs in the vicinity of the rough bed and two cylinders, which resulted in substantial intermittency. This article bears a significant contribution to the understanding of turbulence intermittency and deep learning implementation in turbulent open channel flow.
Feature selection targets for selecting relevant and useful features, and is a vital challenge in turbulence modeling by machine learning methods. In this paper, a new posterior feature selection method based on validation dataset is proposed, which is an efficient and universal method for complex systems including turbulence. Different from the priori feature importance ranking of the filter method and the exhaustive search for feature subset of the wrapper method, the proposed method ranks the features according to the model performance on the validation dataset, and generates the feature subsets in the order of feature importance. Using the features from the proposed method, a black-box model is built by artificial neural network (ANN) to reproduce the behavior of Spalart-Allmaras (S-A) turbulence model for high Reynolds number (Re) airfoil flows in aeronautical engineering. The results show that compared with the model without feature selection, the generalization ability of the model after feature selection is significantly improved. To some extent, it is also demonstrated that although the feature importance can be reflected by the model parameters during the training process, artificial feature selection is still very necessary.
No abstract available
Developing physical closure models with explicit expressions based on a given dataset is essential to science and engineering. For such symbolic regression tasks, biology-inspired evolutionary algorithms are most widely used. However, typical evolutionary algorithms do not utilize any structural information inherent in training data, which limits their performance in finding accurate model structures and coefficients. By combining one evolutionary algorithm, gene expression programing (GEP), with an artificial neural network (ANN) for symbolic regression, we propose a novel evolutionary neural network method, in which candidate expressions are specifically designed so that they can be transformed between the GEP and ANN structures during training iterations. By combining the GEP's global searching and the ANN's gradient optimization capabilities, efficient and robust convergence to accurate models can be achieved. In addition, sparsity-enhancing strategies have been introduced to improve the interpretability of the trained models. The present method has been tested for finding different physical laws and then applied to turbulence modeling problems with different configurations, showing advantages compared to the existing GEP and ANN methods.
A neural-network-based large eddy simulation (LES) is conducted for turbulent flow over a backward-facing step. A fully connected neural network (NN) is trained using filtered direct numerical simulation (DNS) data at the Reynolds number of Reh=5100 based on the free-stream velocity and step height, to develop an NN-based subgrid-scale (SGS) model. Three sets of input variables are considered for training. NN1 is based solely on the filtered strain-rate tensor. NN2 employs the same tensor scaled by the eddy viscosity introduced in the Vreman model, which accounts for the local velocity gradients and grid anisotropy. NN3 further augments the NN2 input with the Leonard stress from the scale-similarity approach. The output of NNs is the SGS stress tensor. For a priori tests, all the NNs show better predictions of the SGS stress and SGS dissipation than the dynamic Smagorinsky model. A posteriori tests (i.e., LES) conducted at the trained Reynolds number and grid distribution show accurate predictions of the reattachment length, mean velocity, and root-mean-square velocity fluctuations. Furthermore, NN2 and NN3 demonstrate good performances on unseen coarser grid resolutions, but NN1 does not. Especially, NN2 maintains predictive capability even at a higher Reynolds number (Reh=28000), whereas NN3 requires ad hoc clipping on negative SGS dissipation for a stable LES at this Reynolds number.
Session H Abstracts Numerically Consistent Data-Driven Subgrid-Scale Model for Large-Eddy Simulation
No abstract available
No abstract available
Abstract This study proposes a machine-learning-based subgrid scale (SGS) model for very coarse-grid large-eddy simulations (vLES). An issue with SGS modelling for vLES is that, because the energy-containing eddies are not accurately resolved by the computational grid, the resolved turbulence deviates from the physically accurate turbulence. This limits the use of supervised machine-learning models commonly trained using pairs of direct numerical simulation (DNS) and filtered DNS data. The proposed methodology utilises both unsupervised learning (cycle-consistency generative adversarial network (GAN)) and supervised learning (conditional GAN) to construct a machine-learning pipeline. The unsupervised learning part of the proposed method first transforms the non-physical vLES flow field to resemble a physically accurate flow field. The second supervised learning part employs super-resolution of turbulence to predict the SGS stresses. The proposed pipeline is trained using a fully developed turbulent channel at the friction Reynolds number of approximately 1000. The a priori validation shows that the proposed unsupervised–supervised pipeline successfully learns to predict the accurate SGS stresses, while a typical supervised-only model shows significant discrepancies. In the a posteriori test, the proposed unsupervised–supervised-pipeline SGS model for vLES using a progressively coarse grid yields good agreement of the mean velocity and Reynolds shear stress with the reference data at both the trained Reynolds number 1000 and the untrained higher Reynolds number 2000, showing robustness against varying Reynolds numbers. A budget analysis of the Reynolds stresses reveals that the proposed unsupervised–supervised-pipeline SGS model predicts a significant amount of SGS backscatter, which results in the strengthened near-wall Reynolds shear stress and the accurate prediction of mean velocity.
A neural network (NN) with one hidden layer is implemented to establish a relationship between the resolved-scale flow field and the subgrid-scale (SGS) stress for large eddy simulation (LES) of the Burgers equation. Five sets of input are considered for the neural network by combining the velocity gradient and the filter size. The training datasets are obtained by filtering the direct numerical simulation (DNS) results of the Burgers equation with random forcing function. The number of modes is sufficiently large ( N = 65 536) to resolve extremely small scales of motion. In the a priori test, a correlation coefficient over 0.93 is achieved for the SGS stress between the NN models and the filtered DNS data. The results of the a posteriori test reveal that the obtained solutions are stable for all NN models without applying any stabilization techniques. However, not all NN models have a reasonable performance when embedded in the LES code. The applicability of the NN models to the Burgers equation with higher and lower viscosity is also investigated, and it is indicated that the most reliable NN models obtained in this paper can be applied to a set of parameters which are different from those used in training. The results of the SGS models constructed using the neural network are also compared with the existing models, and it is shown that the best obtained NN models outperform the Smagorinsky model and the gradient model, and are comparable to the dynamic Smagorinsky model. However, the NN models have an advantage over the dynamic Smagorinsky model in numerical cost.
No abstract available
No abstract available
No abstract available
Super-resolution reconstruction of turbulent flows using deep learning has gained significant attention, yet challenges remain in accurately capturing physical small-scale structures. This study introduces the Conditional Enhanced Super-Resolution Generative Adversarial Network (CESRGAN) for reconstructing high-resolution turbulent velocity fields from low-resolution inputs. CESRGAN consists of a conditional discriminator and a conditional generator, the latter being called CoGEN. CoGEN incorporates subgrid-scale (SGS) turbulence kinetic energy as conditional information, improving the recovery of small-scale turbulent structures with the desired level of energy. By being aware of SGS turbulence kinetic energy, CoGEN is relatively insensitive to the degree of detail in the input. As shown in the paper, its advantages become more pronounced when the model is applied to heavily filtered input. We evaluate the model using direct numerical simulation (DNS) data of forced homogeneous isotropic turbulence. The analysis of Q-criterion isosurfaces, energy spectra, and probability density functions shows that the proposed CoGEN reconstructs fine-scale vortical structures more precisely and captures turbulent intermittency better compared to the traditional generator. Particle-pair dispersion simulations validate the physical fidelity of CoGEN-reconstructed fields, closely matching DNS results across various Stokes numbers and filtering levels. This paper demonstrates how incorporating available physical information enhances super-resolution models for turbulent flows.
Small-scale motions in turbulent flows play a significant role in various small-scale processes, such as particle relative dispersion and collision, bubble or droplet deformation, and orientation dynamics of non-sphere particles. Recovering the small-scale flows that cannot be resolved in large eddy simulation (LES) is of great importance for such processes sensitive to the small-scale motions in turbulent flows. This study proposes a subgrid-scale model for recovering the small-scale turbulent velocity field based on the artificial neural network (ANN). The governing equations of small-scale turbulent velocity are linearized, and the pressure gradient and the nonlinear convection term are modeled with the aid of the ANN. Direct numerical simulation (DNS) and filtered direct numerical simulation (FDNS) provide the data required for training and validating the ANN. The large-scale velocity and velocity gradient tensor are selected as inputs for the ANN model. The linearized governing equations of small-scale turbulent velocity are numerically solved by coupling the large-scale flow field information. The results indicate that the model established by the ANN can accurately recover the small-scale velocity lost in FDNS due to filtering operation. With the ANN model, the flow fields at different Reynolds numbers agree well with the DNS results regarding velocity field statistics, flow field structures, turbulent energy spectra, and two-point, two-time Lagrangian correlation functions. This study demonstrates that the proposed ANN model can be applied to recovering the small-scale velocity field in the LES of isotropic turbulent flows at different Reynolds numbers.
No abstract available
No abstract available
No abstract available
Reynolds-Averaged-Navier-Stokes (RANS) modelling has had a significant amount of success within the industrial setting. In fact, such techniques are used on a daily basis to compute the flow field around a broad spectrum of geometries in very different conditions. To this end, many codes have now reached a good level of maturity in terms of capabilities and robustness. However, there is a growing need for more efficient designs and higher performance. Therefore, aspects that are either challenging or impossible to capture by means of standard second-order steady discretisations are starting to become of interest. The computational hardware improvements seen over the decades and the persistent development of high-order numerical techniques, are allowing Implicit Large Eddy Simulation (iLES) if not Direct Numerical Simulation (DNS). These would allow accurate capturing of unsteady flow phenomena that can be critical to properly understand aspects such as flow separation, reattachment and transition to turbulence in components relevant to the turbomachinery community. The advantage that high-order solvers, such as Nektar++, have over their standard second-order discretisation counterpart, relates to their ability to reach higher fidelity solutions with a saving in terms of number of degrees of freedom and therefore computational cost. In this paper, a detailed introduction to the characteristics and features of the open-source high-order Nektar++ Computational Fluid Dynamics (CFD) framework will be given. An initial discussion of the numerics will be provided, therefore explaining the main features of this solver, followed by a set of different capabilities that are of interest to the turbomachinery community, such as meshing, incompressible and compressible solvers and post-processing. Examples of successful deployment of these features to jet engine components will then be presented. To start, preliminary design components, such as flat plates with different loadings will be discussed. In these cases, the effects of adverse and favourable pressure gradients representative of those seen on modern engine intakes and fan blades, respectively, on boundary layers will be studied. Following, a description on low and high pressure turbine components will be provided with a discussion relating to the relevant experimental comparison. Finally, future developments and improvements to the code will be detailed, such as new schemes and interfaces to hardware enhancements, i.e. Graphical Processing Units.
Abstract High-order Computational Fluid Dynamics (CFD) methods have the potential to deliver higher accuracy at lower computational cost than conventional second-order methods. In this work, we introduce Lethe, an object-oriented, open-source, high-order (space and time) CFD software which leverages the well-established deal.II library. Lethe solves incompressible flow problems on 2D and 3D unstructured quad and hex meshes and allows dynamic mesh adaptation. It is parallel and it is adapted to the solution of large problems ( > 1 0 8 degrees of freedom) on distributed computer architectures. We illustrate some of its fundamental capacities through two benchmarks: a manufactured solution and the DNS of Taylor–Green vortices at R e = 1600 .
Abstract Xcompact3D is a Fortran 90–95 open-source framework designed for fast and accurate simulations of turbulent flows, targeting CPU-based supercomputers. It is an evolution of the flow solver Incompact3D which was initially designed in France in the mid-90’s for serial processors to solve the incompressible Navier–Stokes equations. Incompact3D was then ported to parallel High Performance Computing (HPC) systems in the early 2010’s. Very recently the capabilities of Incompact3D have been extended so that it can now tackle more flow regimes (from incompressible flows to compressible flows at low Mach numbers), resulting in the design of a new user-friendly framework called Xcompact3D . The present manuscript presents an overview of Xcompact3D with a particular focus on its functionalities, its ready-to-run simulations and a few case studies to demonstrate its impact.
Homogeneous isotropic turbulence (HIT) has been a useful theoretical concept for more than fifty years of theory, modelling, and calculations. Some exact results are revisited in incompressible HIT, with special emphasis on the 4/5 Kolmogorov law. The finite Reynolds number effect (FRN), which yields corrections to that law, is investigated, using both Kármán–Howarth-type equations and a statistical spectral closure of the Eddy-Damped Quasi-Normal Markovian (EDQNM)-type. This discussion offers an opportunity to give an extended review of such spectral closures, from weak turbulence, as in wave turbulence theory, to a strong one. Extensions of the 4/5 or 4/3 Kolmogorov/Monin laws to anisotropic cases, such as stably stratified and MHD turbulence, are briefly touched on. Before addressing more recent work on compressible isotropic turbulence, the simplest case of quasi-incompressible turbulence subjected to externally imposed isotropic compression or dilatation is presented. Rapid distortion theory is found to be a poor model in this isotropic case, in contrast with its relevance in strongly anisotropic flow cases. Accordingly, a fully nonlinear approach based on a rescaling of all fluctuating variables is used, in order to show its interplay with the linear operator. This opens the discussion on the cases of homogeneous incompressible turbulence, where RDT and nonlinear models are relevant, provided that anisotropy is accounted for. Finally, isotropic compressible flows of increasing complexity are considered. Recent studies using weak turbulence theory, modelling, and DNS are discussed. A final unpublished study involves interactions between the solenoidal mode, inherited from incompressible turbulence, and the acoustic and entropic modes, which are specific to the compressible problem. An approach to acoustic wave turbulence, with resonant triads, is revisited on this occasion.
Nektar++ is an open-source software framework designed to support the development of high-performance scalable solvers for partial differential equations using the spectral/hp element method. High-order methods are gaining prominence in several engineering and biomedical applications due to their improved accuracy over low-order techniques at reduced computational cost for a given number of degrees of freedom. However, their proliferation is often limited by their complexity, which makes these methods challenging to implement and use. Nektar++ is an initiative to overcome this limitation by encapsulating the mathematical complexities of the underlying method within an efficient C++ framework, making the techniques more accessible to the broader scientific and industrial communities. The software supports a variety of discretisation techniques and implementation strategies, supporting methods research as well as application-focused computation, and the multi-layered structure of the framework allows the user to embrace as much or as little of the complexity as they need. The libraries capture the mathematical constructs of spectral/hp element methods, while the associated collection of pre-written PDE solvers provides out-of-the-box application-level functionality and a template for users who wish to develop solutions for addressing questions in their own scientific domains. Program obtainable from: CPC Program Library, Queen's University, Belfast, N. Ireland No. of lines in distributed program, including test data, etc.: 1052456 No. of bytes in distributed program, including test data, etc.: 42851367 External routines: Boost, PFTW, MPI, BLAS, LAPACK and METIS (www.cs.umn.edu) Nature of problem: The Nektar++ framework is designed to enable the discretisation and solution of time-independent or time-dependent partial differential equations. Running time: The tests provided take a few minutes to run. Runtime in general depends on mesh size and total integration time.
Predicting the fine and intricate details of a turbulent flow field in both space and time from a coarse input remains a major challenge despite the availability of modern machine learning tools. In this paper, we present a simple and effective dictionary-based approach to spatio-temporal upsampling of fluid simulation. We demonstrate that our neural network approach can reproduce the visual complexity of turbulent flows from spatially and temporally coarse velocity fields even when using a generic training set. Moreover, since our method generates finer spatial and/or temporal details through embarrassingly-parallel upsampling of small local patches, it can efficiently predict high-resolution turbulence details across a variety of grid resolutions. As a consequence, our method offers a whole range of applications varying from fluid flow upsampling to fluid data compression. We demonstrate the efficiency and generalizability of our method for synthesizing turbulent flows on a series of complex examples, highlighting dramatically better results in spatio-temporal upsampling and flow data compression than existing methods as assessed by both qualitative and quantitative comparisons.
This study presents a deep learning (DL) neural network hybrid data-driven method that is able to predict turbulence flow velocity field. Recently many studies have reported the application of recurrent neural network (RNN) methods, particularly the Long short-term memory (LSTM) for sequential data. The airflow around the objects and wind speed are the most presented with different hybrid architecture. In some studies, the investigated data set in fluid dynamics were generated via known equations, and they have no random and chaotic behavior. Data series extracted from Computational Fluid Dynamics (CFD) have been used in many cases. This work aimed to determine a method with raw data that could be measured with devices in the airflow, wind tunnel, water flow in the river, wind speed and industry application to process in the DL model and predict the next time steps. This method suggests spatial-temporal data in time series, which matches the Lagrangian framework in fluid dynamics. Gated Recurrent Unit (GRU), the next generation of LSTM, has been employed to create a DL model and forecasting. Time series data source is from turbulence flow has been generated in a laboratory and extracted via 2D Lagrangian Particle Tracking (LPT). This data has been used for the training model and to validate the prediction in the suggested approach. The achievement via this method dictates a significant result and could be developed.
No abstract available
Understanding shock–solid interactions remains a central challenge in compressible fluid dynamics. We present JAX-Shock (just-in-time compiled, exact shock-capturing solver with automatic differentiation): a fully-differentiable, graphics processing unit-accelerated, high-order shock-capturing solver for efficient simulation of the compressible Navier–Stokes equations. Built entirely in JAX, the framework leverages automatic differentiation to enable gradient-based optimization, parameter inference, and end-to-end training of deep learning-augmented models. The solver integrates fifth-order weighted essentially non-oscillatory reconstruction with an Harten-Lax-van Leer-Contact flux to resolve shocks and discontinuities with high fidelity. To handle complex geometries, an immersed boundary method is implemented for the accurate representation of solid interfaces within the compressible flow field. In addition, we introduce a neural flux module trained to augment the numerical fluxes with data-driven corrections, significantly improving the accuracy and generalization. JAX-Shock also supports sequence–to–sequence learning for shock interaction prediction and reverse-mode inference to identify key physical parameters from data. Compared with purely data-driven approaches, JAX-Shock enhances generalization while preserving physical consistency. The framework establishes a flexible platform for differentiable physics, learning-based modeling, and inverse design in compressible flow regimes dominated by complex shock–solid interactions.
No abstract available
Over the last decade, deep learning (DL), a branch of machine learning, has experienced rapid progress. Powerful tools for tasks that have been traditionally complex to automate have been developed, such as image synthesis and natural language processing. In the context of simulating fluid dynamics, this has led to a series of novel DL methods for replacing or augmenting conventional numerical solvers. We broadly classify these methods into physics- and data-driven methods. Physics-driven methods, generally, tune a DL model to provide an analytical and differentiable solution to a given fluid dynamics problem by minimizing the residuals of the governing partial differential equations. Data-driven methods provide a fast and approximate solution to any fluid dynamics problem that shares some physical properties with the observations used when tuning the DL model’s parameters. Meanwhile, the symbiosis of numerical solvers and DL has led to promising results in turbulence modelling and accelerating iterative solvers. However, these methods present some challenges. Exclusively data-driven flow simulators often suffer from poor extrapolation, error accumulation in time-dependent simulations, as well as difficulties in training against turbulent flows. Substantial effort is, therefore, being invested into approaches that may improve the current state of the art.
No abstract available
No abstract available
No abstract available
No abstract available
Turbulent jet flows require significant computational resources at high Reynolds numbers when simulated using large-eddy simulation (LES). To address this challenge, a convolutional neural network (CNN) was used to reproduce small-scale fluctuations from large-scale velocity fields in turbulent jet flows. High-resolution LES data were used as ground truth, and the fast Fourier transform was applied to separate large-scale and subgrid-scale (SGS) components. The CNN was trained to predict SGS velocity fields using the large-scale components extracted from coarse-grid LES data. The performance of the model was evaluated under a 20% and 40% decrease in data resolution, and with a 50% increase in Reynolds number. The CNN-generated velocity fields reproduced energy spectra and turbulent kinetic energy, which closely agreed with those of the high-resolution LES. Although turbulence production and dissipation were less accurately predicted, due to complex energy transfer mechanisms, the CNN outputs demonstrated physical consistency in terms of mass and momentum conservation.
Large-eddy simulation (LES)-based jet noise predictions do not resolve the entire broadband noise spectra, often under-predicting high frequencies that correspond to un-resolved small-scale turbulence. The coupled LES-synthetic turbulence (CLST) model is presented which aims to model the missing high frequencies. The CLST method resolves large-scale turbulent fluctuations from coarse-grid large-eddy simulations (CLES) and models small-scale fluctuations generated by a synthetic eddy method (SEM). Noise is predicted using a formulation of the linearized Euler equations (LEE), where the acoustic waves are generated by source terms from the combined fluctuations of the CLES and the stochastic fields. Sweeping and straining of the synthetic eddies are accounted for by convecting eddies with the large turbulent scales from the CLES flow field. The near-field noise of a Mach 0.9 jet at a Reynolds number of 100,000 is predicted with LES. A high-order numerical algorithm, involving a dispersion relation preserving scheme for spatial discretization and an Adams–Bashforth scheme for time marching, is used for both LES and LEE solvers. Near-field noise spectra from the LES solver are compared to published results. Filtering is applied to the LES flow field to produce an under-resolved CLES flow field, and a comparison to the un-filtered LES spectra reveals the missing noise for this case. The CLST method recovers the filtered high-frequency content, agreeing well with the spectra from LES and showing promise at modeling the high-frequency range in the acoustic noise spectrum at a reasonable expense.
Non-zonal methods of hybrid Reynolds Averaged Navier–Stokes (RANS) equations-Large Eddy Simulation (LES) with subfilter transport equations enable to simulate turbulent flows out of spectral equilibrium on relatively coarse grid resolution. They can operate from RANS to LES depending on a control parameter linked to the grid step size. Variational analysis gives a useful framework to specify the functional dependence of this parameter to the grid size. This is the case for the partially integrated transport modelling method originally established in spectral space for homogeneous turbulence. The partitioning control function then monitors the ratio of the subfilter part to the total turbulent energy and the same process is developed in the present work for the thermal or transported scalar variance as well. So, we demonstrate here that this control function mechanism can be derived by variational calculus from a mathematical physics formalism developed both for first-order and second-order moment closures. We show that the result is entirely consistent with the spectral model derivation made in previous papers and that this approach can be transposed to almost any subfilter transport model developed in hybrid RANS-LES methodology in full generality. As a result, it will be evidenced also that resolved turbulent diffusion terms play a significant role in the acting mechanisms of turbulence and cannot be therefore neglected, even if these terms do not explicitly appear in LES subfilter closure because they are computed by the simulation itself and not modelled by the subfilter model. In addition, numerical simulations have been performed for illustrating the theoretical results of the variational analysis.
No abstract available
Turbulence modeling relies critically on accurate characterization of large-scale structures, with the integral length scale L serving as a key parameter for industrial applications ranging from combustion stability optimization to wind farm design and aerodynamic load prediction. However, Direct numerical simulation (DNS) of turbulence faces inherent limitations in resolving all wavenumbers within the lowwavenumber region of the turbulent kinetic energy spectrum due to finite computational domain sizes. This unresolved low-wavenumber deficiency leads to incomplete characterization of large-scale structures and introduces systematic deviations in key statistical quantities, particularly the integral length scale L and turbulence dissipation coefficient Cε. As turbulence evolves, the spectral peak wavenumber kp migrates toward lower wavenumbers, exacerbating the loss of large-scale information and causing computed statistics to diverge from physical reality. In this study, we perform high-fidelity DNS of homogeneous isotropic decaying turbulence in a periodic cubic domain of side length 4π with 3843 grid points. DNS cases are performed by using a standard pseudospectral solver and a fourth-order Runge-Kutta time integration scheme, with a semi-implicit treatment of the viscous term. The spatial resolution kmaxη = 1.65 ensures adequate resolution of dissipative scales (η is the Kolmogorov scale). Simulations start from a fully developed field initialized with a spectrum matching Comte-Bellot and Corrsin’s experimental data and evolve within a time interval where turbulence exhibits established isotropic decay characteristics. Existing correction models, predominantly based on equilibrium turbulence assumptions, fail to capture the non-equilibrium dynamics governed by large-scale structures. Based on a generalized von Kármán spectrum model, we use a correction framework to account for unresolved low-wavenumber contributions in homogeneous isotropic decaying turbulence. DNS data reveal that the uncorrected integral scale Lm significantly underestimates the true L, with errors escalating as kL/kp increases, where kL is the minimum resolvable wavenumber in the simulation domain. After correction, L exhibits a temporal evolution following the Saffmann-predicted power-law relationship $L \propto t^{2 / 5}$, contrasting sharply with the underestimated pre-correction values. Despite the spectral correction substantially increasing the spectral integral scale L, its value remains less than the physically derived integral scale Λ computed from the velocity correlation function, primarily due to the finite domain size limiting large-scale statistics and the moderate grid resolution, though higher-resolution simulations with the same domain show L converging towards Λ. Notably, the unmodified dissipation coefficient Cε remains constant, consistent with equilibrium turbulence assumptions, whereas the corrected Cε evolves according to the non-equilibrium scaling law $C_{\varepsilon} \sim R e_\lambda^{-1}$. Further analysis confirms that the ratio L/λ shifts from Kolmogorov’s $R e_\lambda^1$ dependence to a Reynolds-number-independent plateau after correction, fundamentally altering the turbulence dissipation paradigm. This transition from equilibrium to non-equilibrium dissipation behavior underscores the dominant role of large-scale structures in regulating energy cascade dynamics. Our results demonstrate that finite Reynolds numbers or strong initial-condition effects amplify the nonequilibrium characteristics of turbulence, preventing full-scale equilibrium. These findings reconcile long-standing theoretical discrepancies and provide a paradigm for modeling scale interactions in turbulence.
No abstract available
No abstract available
No abstract available
No abstract available
No abstract available