单点时间历史恢复壁湍流空间流场:可恢复性、验证与图网络学习方法
壁湍流近壁区大尺度组织、内外区耦合与相干结构的物理基础(时间历史信息来源与可恢复性机制)
合并三份初始分组中"内-外区相互作用模型与大尺度结构条带预测""壁湍流相干结构与时空关联的物理基础""近壁区大尺度结构与时空关联"三组。共同点是不提出重建算法,而是提供回答'单点时间历史能恢复什么、增益来自何种时空关联'的物理依据:对数区单点大尺度信号与内区/近壁特征的标定关系(IOIM/MSO)、跨雷诺数不变性与幅值/尺度调制、涡结构与条带的相干尺度、大尺度组织与内外层调和、粗糙壁模态速度场。它们是判断可恢复尺度与可迁移性的物理标尺;其中部分条目题录信息不足,须在证据表中标注待核实。
- ON THE RELATIONSHIP BETWEEN THE WALL PRESSURE AND VELOCITY FIELD IN A TURBULENT BOUNDARY LAYER(C. Wark, A. Naguib, William d, O. Juckenhoefel, 1998, No journal)
- Reynolds number invariance in the coherent structures of wall-bounded turbulence(An-Jia Ying, Zhi-Gang Li, Lin Fu, 2026, Journal of Fluid Mechanics)
- Abstract Submitted for the DFD16 Meeting of The American Physical Society Large Scale Organization of a Near Wall Turbulent Boundary Layer1(M. Stanislas, Raoul Florent Dekou Tiomajou, 2016, No journal)
- Vortical structures in the turbulent boundary layer: a possible route to a universal representation(M. Stanislas, L. Perret, J. Foucaut, 2008, Journal of Fluid Mechanics)
- Three-dimensional characterisation of macro-instabilities in a turbulent stirred tank flow and reconstruction from sparse measurements using machine learning methods(K. Mikhaylov, S. Rigopoulos, G. Papadakis, 2023, Chemical engineering research & design)
- Automatic identification and characterization of turbulent bursting from single-point records of the velocity field(R. Goldshmid, D. Liberzon, 2019, No journal)
- Stochastic modal velocity field in rough-wall turbulence(Roozbeh Ehsani, Michael Heisel, M. Puccioni, Jia-Rong Hong, Valerio Iungo, V. Voller, M. Guala, 2024, Journal of Fluid Mechanics)
- Large scale organization of a near wall turbulent boundary layer(R. Dekou, J. Foucaut, S. Roux, M. Stanislas, J. Delville, 2016, International Journal of Heat and Fluid Flow)
- Towards Reconciling the Large-Scale Structure of Turbulent Boundary Layers in the Atmosphere and Laboratory(N. Hutchins, K. Chauhan, I. Marusic, J. Monty, J. Klewicki, 2012, Boundary-layer Meteorology)
- Non-linear interactions in a boundary layer developing over an array of cubes using stochastic estimation(K. Blackman, L. Perret, 2016, Physics of Fluids)
测量链条与观测算子:热线空间分辨率/空间平均、TR-PIV传递函数、对流速度选择与高分辨率参考场
合并"壁湍流实验测量基础、热线与TR-PIV的分辨率/传递函数及高分辨率参考场""观测条件与测点布置(测量部分)""热线空间分辨率与对流假设"三组。共同点是刻画输入信号与参考场的物理可信边界:热线探针长度、端部导热与高展向长宽比传感的空间滤波如何畸变脉动方差与谱,TR-PIV激光片厚与有效分辨率的传递函数,Taylor冻结假设中当地/瞬时对流速度选择导致的剪切畸变,以及PIV–DNS对比、多平面同步测量与高分辨率实验场/间接测量时间分辨重建的获取方式。这决定'可恢复范围与尺度受何限制'及真实热线、PIV抽点、DNS抽点三类观测设置的差异。
- The spectral response of time-resolved PIV in a turbulent boundary layer(P. Manovski, W. Rowin, Henry Ng, P. Gulotta, Matteo Giacobello, C. D. de Silva, N. Hutchins, I. Marusic, 2025, Experiments in Fluids)
- A micro-hot-wire anemometry probe with elongated stubs for turbulent boundary layer measurements(Baptiste Baradel, Alain Giani, Fabien Méry, P. Combette, O. Léon, 2024, Experiments in Fluids)
- Experimental Investigation of a Turbulent Boundary Layer Using Simultaneous 3-D Flow Visualization and 2-D PIV(M. Melnick, B. Thurow, 2012, No journal)
- Turbulent boundary layers over rod- and cube-roughened walls(J. Lee, A. Seena, Seung Hyun Lee, H. Sung, 2012, Journal of Turbulence)
- Direct Numerical Simulation and PIV Measurement of Turbulent Boundary Layer over a Rod-Roughened Wall(Seung-Hyun Lee, J. Lee, H. Sung, 2009, No journal)
- Simultaneous orthogonal-plane particle image velocimetry measurements in a turbulent boundary layer(W. Hambleton, N. Hutchins, I. Marusic, 2006, Journal of Fluid Mechanics)
- Reconstruction of high-resolution turbulent flow fields from sparse measurement using the diffusion normalizing flows(Vijayamanikandan Vijayarangan, Harshavardhana A. Uranakara, H. G. Im, 2024, AIAA SCITECH 2024 Forum)
- Time-resolved flow reconstruction with indirect measurements using regression models and Kalman-filtered POD ROM(R. Leroux, L. Chatellier, L. David, 2017, Experiments in Fluids)
- Very high aspect ratio hot-wire based MEMS thermal sensor for near wall turbulent flow measurement with high sensitivity and low power consumption(C. Ghouila-Houri, J. Claudel, J. Gerbedoen, Q. Gallas, E. Garnier, A. Merlen, R. Viard, A. Talbi, P. Pernod, 2023, No journal)
- Hot-wire spatial resolution issues in adverse pressure gradient turbulent boundary layers(A. Dróżdż, R. Örlü, V. Sokolenko, P. Schlatter, W. Elsner, P. Niegodajew, 2024, Measurement)
- Evaluation of convective velocity choice in 3D velocity field reconstruction with Taylor’s hypothesis(S. Sheppard, J. Farnsworth, J. G. Brasseur, 2026, Experiments in Fluids)
- Analysis of hot-wire measurements accuracy in turbulent boundary layer(A. Dróżdż, W. Elsner, 2015, Open Engineering)
- Spatial averaging effects on the streamwise and wall-normal velocity measurements in a wall-bounded turbulence using a cross-wire probe(R. Baidya, Jimmy Philip, N. Hutchins, J. Monty, I. Marusic, 2019, Measurement science and technology)
线性随机估计与谱线性随机估计(LSE/SLSE):少量测点时间信号到条件平均场或模态系数的经典映射
合并"线性随机估计、谱线性随机估计与时空相关/因果信息流分析""线性随机估计、谱线性随机估计与条件采样(LSE/SLSE 本体)""经典线性随机估计(含谱域LSE)"三组中的 LSE/SLSE 主线。共同点是以两点时空相关(互谱、相关张量)为唯一信息来源,用线性/谱域估计核把少量测点(含单点单分量)信号映射到目标位置场或模态系数,并明确给出条件平均化、未解析随机成分与尺度依赖(低阶大尺度可恢复、小尺度仅统计一致)的结论,含涡量到场量的卷积反演、温度-速度耦合、可解释随机估计框架。是主要的方法谱系与基线来源。
- On combining linear stochastic estimation and proper orthogonal decomposition for flow reconstruction(B. Podvin, Sylvain Nguimatsia, J. Foucaut, C. Cuvier, Y. Fraigneau, 2018, Experiments in Fluids)
- An Extended View of the Inner-outer Interaction Model for Wall-bounded Turbulence Using Spectral Linear Stochastic Estimation(I. Marusic, W. Baars, N. Hutchins, 2015, Procedia Engineering)
- Stochastic estimation and flow sources associated with surface pressure events in a turbulent boundary layer(A. Naguib, C. Wark, O. Juckenhöfel, 2001, Physics of Fluids)
- Research on spectral estimation parameters for application of spectral proper orthogonal decomposition in train wake flows(Xiao-Bai Li, Guang Chen, Xi-Feng Liang, Dongrun Liu, Xiao-Hui Xiong, 2021, The Physics of Fluids)
- Spatial and spectral characteristics of information flux between turbulent boundary layers and porous media(Wenkang Wang, A. Lozano-Durán, R. Helmig, X. Chu, 2022, Journal of Fluid Mechanics)
- Spectral stochastic estimation of high-Reynolds-number wall-bounded turbulence for a refined inner-outer interaction model(W. Baars, N. Hutchins, I. Marusic, 2016, No journal)
- Linear-model-based study of the coupling between velocity and temperature fields in compressible turbulent channel flows(Cheng Cheng, Lin Fu (Associate Professor), 2023, Journal of Fluid Mechanics)
- Vortex-to-velocity reconstruction for wall-bounded turbulence via the field-based linear stochastic estimation(Cheng-Yue Wang, Qi Gao, Biao Wang, C. Pan, Jin-Jun Wang, 2021, Journal of Fluid Mechanics)
- LARGE SCALE STRUCTURE OF TURBULENT BOUNDARY LAYER SUBJECTED TO AN ADVERSE PRESSURE GRADIENT(Joung-Ho Lee, H. Sung, 2009, Proceeding of Sixth International Symposium on Turbulence and Shear Flow Phenomena)
- Use of Stochastic Estimation for the Wave-Turbulence Separation in the Water Column(P. Druault, B. Gaurier, G. Germain, 2026, Volume 7B: Ocean Renewable Energy)
- A stochastic estimation framework for interpretable force modelling in flapping-wing aerodynamics(Martín Navarro-González, Marco Raiola, 2025, Journal of Fluid Mechanics)
条件采样、条件平均相干结构与时空相关信息流分析(含单时刻与谱估计、二次与多方法对比)
由"条件采样与相干结构条件平均""线性随机估计与条件时空相关估计(方法比较部分)"合并而来,与上一组的区别在于:本组的目标是条件平均结构本身(壁面剪应力条件采样、能量传递结构、相干结构随机估计与 LSE 控制应用)以及方法间比较与信息流/因果分析,而非把 LSE 当作全场映射工具。它界定'条件平均流动结构估计'与'瞬时流场重建'的任务边界,说明为何非零时滞相关(对流相关)是时间信息增益的来源,并处理非唯一性与相位同步重建压力场等经典问题。
- Wall-Shear-Stress-Based Conditional Sampling Analysis of Coherent Structures in a Turbulent Boundary Layer(Sangjin Ryu, Ethan A. Davis, J. S. Park, Hai-Peng Zhang, J. Yoo, 2021, Journal of Fluids Engineering-transactions of The Asme)
- Energy transfer structures associated with large-scale motions in a turbulent boundary layer(Wenkang Wang, C. Pan, Jin-Jun Wang, 2020, Journal of Fluid Mechanics)
- Linear estimation and control of coherent structures in wall-bounded turbulence at Reτ = 2000(Stephan F. Oehler, S. J. Illingworth, 2018, No journal)
- Stochastic estimation of coherent structures in turbulent boundary layers(Y. Guezennec, W. Choi, 1989, Physics of Fluids)
- Comparison of stochastic estimation methods with conditional events optimization for the reconstruction of the flow around a supercritical airfoil in transonic conditions(A. Arnault, J. Dandois, J. Foucaut, 2016, Computers & Fluids)
- A comparison of single-time and spectral linear stochastic field pressure estimation in the flow over a backward-facing step(M. Hashemi, Joseph W. Hall, 2023, Progress in Canadian Mechanical Engineering. Volume 6)
- Streamwise evolution of coherent structures in wall-bounded turbulence under the influence of synthetic jets(Biao-Hui Li, Chao Liang, N. Jiang, Haiping Tian, 2026, The Physics of Fluids)
- Reconstruction of the Unsteady Pressure Field in a Low Speed Linear Cascade(F. Förster, D. Sims-Williams, G. Ingram, 2012, No journal)
时间历史、时间延迟嵌入与对流速度:时间维度被显式利用的途径及其窗口与因果性
合并"时间历史、时间延迟嵌入与对流速度"与"POD/DMD模态分解、降阶模型与时间延迟嵌入"中的延迟嵌入理论条目。共同点是显式建模时间:多时间延迟非线性随机估计、基于连续快照差分的延迟嵌入 DMD、Takens 延迟嵌入+流形学习的状态重构与相空间重构/分形维数判据、跨相位时间延迟估计测速、尺度依赖对流速度变异性与猝发过程调制。为回答'时间历史相对单时刻的增益来源''窗口长度与仅用过去观测的因果估计如何影响可恢复性'提供方法学与证据。
- Nonlinear multi-time-delay stochastic estimation: application to cavity flow and turbulent channel flow(D. Lasagna, Linda Fronges, Matteo Orazi, G. Iuso, 2015, AIAA Journal)
- Nonlinear multi-time-delay stochastic estimation : application to cavity shear layer and turbulent channel flow(Lasagna D. A., 2014, No journal)
- A snapshot difference and amplitude optimization based dynamic mode decomposition method for reconstruction and prediction of flow fields(Xuhui Luo, Yanhao Chen, Yan-Song He, Zhi-Fei Zhang, 2025, The Physics of Fluids)
- Manifold-embedded network dynamics for fluid flow prediction(Gui-Yong Zhang, Zi-Hao Wang, Chongbin Shi, Teizhi Sun, 2025, The Physics of Fluids)
- Investigating flow patterns and related dynamics in multi-instability turbulent plasmas using a three-point cross-phase time delay estimation velocimetry scheme(C. Brandt, S. C. Thakur, G. Tynan, 2016, Physics of Plasmas)
- Variability of convection velocities and structure inclination angles in wall-bounded turbulence(G. Cui, I. Jacobi, 2024, Journal of Fluid Mechanics)
- Bursting-process modulation in a turbulent boundary layer via the natural visibility graph method(Lijuan Shi, Meng Wang, Zhan-Qi Tang, N. Jiang, 2024, The Physics of Fluids)
- PHASE SPACE RECONSTRUCTION AND FRACTAL DIMENSION USING OF DELAY TIME AND EMBEDDING DIMENSION(Aalami Mohammad Taghi, M. Leila, 2013, No journal)
稀疏感知、可观测性准则与测点布置:恢复上限与缺失/多视场数据下的低维系数恢复
合并"观测条件与测点布置(设计部分)""稀疏感知、传感器布置与缺失/多视场数据下的模态系数恢复"两组。共同点是聚焦输入配置本身:用谱条件数/敏感性或伴随不确定性图选择测点、贪心算法优化稀疏传感器位置并以线性回归恢复 POD 时间系数以拼接多视场 PIV、面向估计的最优布置准则、以及在 PIV 缺失(互补与非互补)数据下用多尺度自编码器补全并与 gappy-POD 对比。恢复对象多为低维系数或缺失区域,指标为归一化估计误差,可直接支撑测点数量/位置/时间窗口影响的讨论。
- Spectral observability-based sparse sensing of cylinder-wake and turbulent channel flows(Jin Mi, Zi-Xiang Tong, Di-Ke Li, Jian-Qin Zhu, Yi-Min Zhang, 2026, The Physics of Fluids)
- Sensor placement for data assimilation of turbulence models using eigenspace perturbations(O. Bidar, S. R. Anderson, N. Qin, 2024, The Physics of Fluids)
- Enhancing PINN-Based Flow Reconstruction with Sparse and Decoupled Gradient-Weighted Sensor Placement(D. Gomes, A. E. Santo, J. Páscoa, 2025, Annual Conference of the IEEE Industrial Electronics Society)
- Data-driven unsteady entire flow field estimation using time-series multiple-field-of-view particle image velocimetry data(Haruka Kurahashi, T. Ikami, Hiroki Nagai, 2026, The Physics of Fluids)
- Effective Sensor Placements for the Estimation of Proper Orthogonal Decomposition Mode Coefficients in von Karman Vortex Street(Kelly Cohen, S. Siegel, Dave Wetlesen, J. Cameron, Aaron Sick, 2004, Journal of Vibration and Control)
- Reconstruction of missing flow field from imperfect turbulent flows by machine learning(Zhao-Hui Luo, Long-Yan Wang, Jian Xu, Zilu Wang, Meng Chen, Jian-Ping Yuan, A. C. Tan, 2023, The Physics of Fluids)
- Time-resolved reconstruction of flow field around a circular cylinder by recurrent neural networks based on non-time-resolved particle image velocimetry measurements(Xiao-Wei Jin, Shujin Laima, Wen-Li Chen, Hui Li, 2020, Experiments in Fluids)
POD/DMD/SPOD与降阶模型:模态截断、闭合建模与重构误差(低维状态表示层)
合并"POD/DMD/SPOD降阶模型与低维流形"(模型部分)与"POD/DMD模态分解、降阶模型"(不含延迟嵌入)两组。共同点是以 POD/DMD/SPOD 基与模态系数为中介,讨论模态截断误差、降阶 Galerkin 模型与稀疏回归涡黏闭合、参数化复杂网络降阶、SPOD 辨识尾流/叶轮模态、POD 与 DMD 相干结构对比及三维模态代理预测。是'模态系数或低维状态恢复'任务的表示基础与线性基线,也是判断未解析尺度如何处理的关键。
- A reduced order model based on adaptive proper orthogonal decomposition incorporated with modal coefficient learning for digital twin in process industry(Xiaoyang Zhu, Yangjian Ji, 2023, Journal of Manufacturing Processes)
- Aerodynamic interference and flow field mechanisms of three-dimensional tandem square cylinders based on proper orthogonal decomposition and high-order dynamic mode decomposition modeling(Kuiwen Yuan, Zeng-Shun Chen, Ye-Meng Xu, Li-Kai Zhang, Peng Hu, Si-Jia Li, 2025, The Physics of Fluids)
- Reduced-order complex network modeling of parametric fluid dynamic systems under partial observability(Zi-Hao Wang, Guiyong Zhang, Tie-Zhi Sun, Zhe Sun, Zhi-Fan Zhang, Bo Zhou, 2026, Nonlinear dynamics)
- Spectral proper orthogonal decomposition of time-resolved three-dimensional flow measurements in the turbulent wake of the Ahmed body(C.-W. Chen, S. Wang, S. Ghaemi, 2024, Journal of Fluid Mechanics)
- Data-driven identification of coherent structures in gas-solid system using proper orthogonal decomposition and dynamic mode decomposition(Dandan Li, Bidan Zhao, Jun-Wu Wang, 2022, The Physics of Fluids)
- Turbulent wake dynamics and flow characteristics over typical hilly terrain: A proper orthogonal decomposition and dynamic mode decomposition analysis(Wei-Cheng Hu, Qing-Shan Yang, Tong Zhou, B. Lu, Guo-Wei Qian, W. Shan, Yu Wang, 2025, The Physics of Fluids)
- The comparison of pump flow field via decomposition mode dynamic and proper orthogonal decomposition: An analysis of gappy proper orthogonal decomposition(Wenjie Peng, Jie-Yu Pei, Wen-Jie Wang, Shou-Qi Yuan, Jia Chen, Xing-Cheng Gan, Qi-Fan Deng, 2024, The Physics of Fluids)
- A sparse optimal closure for a reduced-order model of wall-bounded turbulence(Chua-Khoo Zhao, C. Chan, Yongyun Hwang, 2022, Journal of Fluid Mechanics)
- Proper orthogonal decomposition, dynamic mode decomposition, wavelet and cross wavelet analysis of a sloshing flow(T. Pagliaroli, F. Gambioli, F. Saltari, Jonathan Cooper, 2022, Journal of Fluids and Structures)
- Parsimonious Dynamic Mode Decomposition: A Robust and Automated Approach for Optimally Sparse Mode Selection in Complex Systems(Arpan Das, Pier Marzocca, O. Levinski, 2024, arXiv.org)
- Interpretation and prediction of the three-dimensional coherent structure and its dynamics of tornado-like vortex via delayed proper orthogonal decomposition(Lei Zhou, B. R. Noack, K. Tse, Xu-Hui He, 2025, The Physics of Fluids)
低维流形学习、隐空间与模态系数恢复、超分辨及神经算子/随机强迫的时空统计刻画
由"POD/DMD/SPOD降阶模型与低维流形、模态系数恢复"与"POD/降阶模态与低维状态恢复"中的学习型降阶条目合并。共同点是'先投影到低维子空间再重建/预测':可观测增广流形学习支持多源(DNS/实验)数据比较与超分辨验证、SuperOB 再正交化、非定常状态的降阶表示、4D-flow MRI 降阶建模、POD+DMD 多模态数据融合超分辨、潜空间维度影响、POD 模态最小二乘的激光雷达风场重建,以及以神经算子刻画随机强迫预测时空统计。兼作第五部分参考场可比性与潜空间容量影响的素材。
- Observable-augmented manifold learning for multi-source turbulent flow data(Kai Fukami, Kunihiko Taira, 2025, Journal of Fluid Mechanics)
- Wind field estimation for lidar-assisted control: a comparison of proper orthogonal decomposition and interpolation techniques(Esperanza Soto Sagredo, S. J. Andersen, Á. Hannesdóttir, J. Rinker, 2026, Wind Energy Science)
- SuperOB: Super-resolution flow reconstruction from sparse measurements via re-Orthogonalization of a global Basis(S. J. Andersen, Juan Pablo Murcia Leon, 2026, Journal of Physics, Conference Series)
- Reduced-order representation and reconstruction of non-stationary flow system using flow-time-history deep learning(Qing-Liang Zhan, Chun-Jin Bai, Xin Liu, Yao-Jun Ge, 2023, Acta Mechanica Sinica)
- Reduced-order modeling of 4D flow MRI and CFD in stenotic flow using Proper Orthogonal Decomposition (POD) and Dynamic Mode Decomposition (DMD)(Amirkhosro Kazemi, M. Stoddard, A. Amini, 2022, No journal)
- Towards multi‐modal data fusion for super‐resolution and denoising of 4D‐Flow MRI(I. Perez-Raya, Mojtaba F. Fathi, Ahmadreza Baghaie, R. Sacho, K. Koch, R. D'Souza, 2020, International Journal for Numerical Methods in Biomedical Engineering)
- Influence of latent space dimension on flow field reconstruction from limited number of sensors(Charly Witmeur, M. Runacres, J. Decuyper, 2025, International Instrumentation and Measurement Technology Conference)
- Time-resolved reconstruction of turbulent flows using linear stochastic estimation and sequential data assimilation(C. He, Ying-Zheng Liu, 2020, Physics of Fluids)
- Neural operator-based stochastic forcing for resolvent prediction of space–time turbulence statistics in channel flows(Chu-Tian Wu, Xin-Lei Zhang, Guo-Wei He, 2025, Journal of Fluid Mechanics)
N-S线性模型、卡尔曼滤波与resolvent状态估计:由壁面或单层测量恢复壁湍流大尺度结构
合并"N-S线性模型/贝叶斯/同化"(线性滤波部分)与"时间分辨状态估计、卡尔曼滤波与数据同化"(壁湍流部分)。共同点是把重建写成状态估计问题并用观测时间历史:基于 N-S 线性随机模型的卡尔曼滤波从单壁法向位置或壁面剪应力恢复大尺度结构并预测估计器性能、槽道流状态估计、resolvent-based 估计、POD+系统辨识+卡尔曼从标量/稀疏传感器恢复时间分辨场、局部测量反演全场。为'时间历史相对单时刻的增益'提供最优估计理论解释,并可讨论因果性与不确定性。
- Linear-model-based estimation in wall turbulence: improved stochastic forcing and eddy viscosity terms(Vikrant Gupta, Anagha Madhusudanan, M. Wan, S. J. Illingworth, M. Juniper, 2021, Journal of Fluid Mechanics)
- Linear estimation of coherent structures in wall-bounded turbulence at Reτ = 2000(Stephan F. Oehler, A. García-Gutiérrez, S. J. Illingworth, 2017, Journal of Physics: Conference Series)
- Estimation of turbulent channel flow at $Re_{\unicode[STIX]{x1D70F}}~=~100$ based on the wall measurement using a simple sequential approach(Takao Suzuki, Y. Hasegawa, 2017, Journal of Fluid Mechanics)
- Numerical implementation and evaluation of resolvent-based estimation for space–time energy spectra in turbulent channel flows(Bowen Yang, G. Jin, Ting Wu, Zi-Xuan Yang, G. He, 2020, Acta Mechanica Sinica)
- Flow Reconstruction Around a Surface-Mounted Prism from Sparse Velocity and/or Scalar Measurements Using a Combination of POD and a Data-Driven Estimator(Shengqi Lu, G. Papadakis, 2023, Flow Turbulence and Combustion)
- Recovering turbulent flow field from local quantity measurement: turbulence modeling using ensemble-Kalman-filter-based data assimilation(Zhiwen Deng, C. He, X. Wen, Ying-Zheng Liu, 2018, Journal of Vision)
- Estimation of time-resolved turbulent fields through correlation of non-time-resolved field measurements and time-resolved point measurements(S. Discetti, M. Raiola, A. Ianiro, 2018, Experimental Thermal and Fluid Science)
贝叶斯MAP、变分/非线性数据同化、模态同步与时间一致性约束(含风场4D-Var与超声测速时序)
由"N-S线性模型、贝叶斯MAP、4D-Var与模态同步的状态估计"与"数据同化"中的非滤波条目合并。共同点是以先验/协方差/动力学模型做反问题,把整段观测序列纳入:两点协方差先验下的贝叶斯最大后验全场重建、风电场尾流 4D-Var、Lagrangian 粒子记忆的 FlowFit 时序一致性约束、顺序变分同化中的模态同步与延迟坐标(直接对应'仅用过去观测的因果估计'与窗口长度问题)、多步条件高斯非线性同化跨层传递信息、超声测速时间分辨率提升。
- Bayesian based estimation of turbulent flow fields from lidar observations in a conventionally neutral atmospheric boundary layer(P. Bauweraerts, J. Meyers, 2020, Journal of Physics, Conference Series)
- Enhancing modal synchronisation in wall-bounded turbulence via sequential variational data assimilation(C. He, Peng Wang, Hyung Jin Sung, Wenwu Zhou, D. Peng, Ying-Zheng Liu, 2026, Journal of Fluid Mechanics)
- Turbulent flow field reconstruction in wind-farms using power measurements(A. Alreweny, S. Vandewalle, J. Meyers, 2024, Journal of Physics, Conference Series)
- Enforcing temporal consistency in physically constrained flow field reconstruction with FlowFit by use of virtual tracer particles(F. Ehlers, A. Schröder, S. Gesemann, 2020, Measurement science and technology)
- Time resolution improvement of ultrasonic velocity profiler for flow over cylinder using EPOD method with optimally placed time-resolved sensors(N. Tiwari, 2024, Experiments in Fluids)
- A Nonlinear Data Assimilation Algorithm with Closed-Form Approximations for Multi-Layer Flow Fields(Zhong-Rui Wang, Nan Chen, Di Qi, 2025, Monthly Weather Review)
物理约束神经网络(PINN/FVM-PINN)反演框架、稀疏含噪鲁棒性与求解效率(含壁湍流直接相关工作)
合并"物理约束神经网络稀疏含噪数据流场重建的方法与求解效率"与"PINN类流场反演"中的方法学条目。共同点是把 N-S/PDE 残差与稀疏离散监督结合,重点在稀疏度鲁棒性、压力场无测量推断、有限体积/弱形式降低高阶导数开销、域分解与压力基准不确定性、PIV 空间稀疏与测点高频信号的时空互补融合、结构与物理联合重建网络。其中从二维平面测量重建三维可实现壁湍流场(以 LSE 为基线并做物理约束消融)是与本调研最直接相关的核心文献。本组也是区分'物理残差小'与'真实场准确'证据的素材。
- Physics-informed deep learning for turbulent boundary layers: Spatio-temporal reconstruction and prediction(J. Zhu, Xiang-Rui Dong, Wu Zhou, Lin Ding, T. Cai, Yin-Lin Dong, 2026, Journal of Hydrodynamics)
- Reconstruction of the turbulent flow field with sparse measurements using physics-informed neural network(N. Chaurasia, Shubhankar Chakraborty, 2024, The Physics of Fluids)
- A sparse and decoupled supervision framework for multi-field reconstruction in turbulent flows using physics-informed neural networks(D. Gomes, A. Espírito Santo, J. Páscoa, 2025, The Physics of Fluids)
- Finite-Volume Physics-Informed U-Net for Flow Field Reconstruction With Sparse Data(Tong Zhu, De-Hao Liu, Yanglong Lu, 2024, Journal of Computing and Information Science in Engineering)
- Flow field reconstruction from sparse and noisy data based on finite volume physics-informed neural networks(Zhi-Yuan Geng, Hai-Long Liu, Rui Deng, Xiang Yu, Dapeng Jiang, Sheng-Ze Cai, 2026, The Physics of Fluids)
- Distributed physics-informed neural networks via domain decomposition for fast flow reconstruction(Yi-Xiao Qian, Jia-Xu Liu, Zewei Xia, Song Chen, Chao Xu, Sheng-Ze Cai, 2026, arXiv.org)
- Flow field reconstruction from sparse sensor measurements with physics-informed neural networks(Mohammad Hosseini, Y. Shiri, 2024, The Physics of Fluids)
- Temporal and spatial flow field reconstruction from low-resolution PIV data and pressure probes using physics-informed neural networks(Bo-Zhen Lai, Ying-Zheng Liu, X. Wen, 2024, Measurement science and technology)
- Physics-informed data-driven reconstruction of turbulent wall-bounded flows from planar measurements(G. S. Hora, P. Gentine, M. Momen, M. Giometto, 2024, The Physics of Fluids)
- PISA-Net: A Physics-Informed Structure-Aware Neural network for multiphysics field reconstruction in liquid cooling systems(Chao-Bo Zhang, Lizhe Jiang, Zhao Yang, Kun Yan, Jun Yan, 2025, Applied Thermal Engineering)
PINN在旋转、燃烧、多孔介质、血流与水下等场景的间接/积分观测算子建模与跨领域迁移条件
由"PINN在工业管网、血流与多孔介质等具体场景中的反演实践"与"物理约束反演与PINN重建"中的场景类条目合并。共同点是强调'观测算子'本身的建模:积分式流量计(线积分)约束的 T 型接头压力重建、超声/彩色多普勒低分辨率低信噪比输入的多帧耦合、4D-flow MRI 与 PIV 的血流动力学、RANS 约束的多孔-流体复合流动、旋转爆震与旋流燃烧、光场 PIV 超分辨(含成像退化与可靠性加权)、LDA 稀疏数据连续化、边界形状与速度联合反演、水下航行器与尾迹稀疏数据反演。提供'把单工况 PINN 反演作为教师场/伪标签'的已有实践与向壁湍流迁移的适用条件限制。
- Pressure Field Reconstruction in Turbulent Flows Using Integral Measurement-Based Physics Informed Neural Networks (PINNs)(D. Gomes, A. E. Santo, J. Páscoa, 2026, 2026 IEEE International Symposium on Artificial Intelligence for Instrumentation and Measurement (AI4IM))
- 3D velocity and pressure field reconstruction in the cardiac left ventricle via physics informed neural network from echocardiography guided by 3D color Doppler(H. Wong, W. Chan, Wenbin Mao, C. Yap, 2025, Computer methods and programs in biomedicine)
- Physics-Informed Neural Networks for the High-Resolution Reconstruction of Flow Measurement Indicators in Fluid Dynamics(Irena Radišić, Raffaele Tirotta, Alberto Zingaro, S. Pagani, L. Dede', 2026, arXiv.org)
- Physics-informed neural network for turbulent flow reconstruction in composite porous-fluid systems(Seohee Jang, M. Jadidi, S. Rezaeiravesh, Alistair Revell, Y. Mahmoudi, 2024, Machine Learning: Science and Technology)
- RF-PINNs: Reactive flow physics-informed neural networks for field reconstruction of laminar and turbulent flames using sparse data(Vikas Yadav, M. Casel, Abdulla Ghani, 2024, Journal of Computational Physics)
- Flow-field reconstruction in rotating detonation combustor based on physics-informed neural network(2023, The Physics of Fluids)
- Mixed convection physical-informed neural networks for flow reconstruction with sparse thermal data and partial physics in a lid-driven cavity(Zhuang Zhang, Quan-Jiang Li, Jing-Tao Wang, 2025, The Physics of Fluids)
- Reconstruction of 3D flow field around a building model in wind tunnel: a novel physics-informed neural network framework adopting dynamic prioritization self-adaptive loss balance strategy(En-Ze Rui, Zheng-Wei Chen, Y. Ni, Lei Yuan, Guangqing Zeng, 2023, Engineering Applications of Computational Fluid Mechanics)
- Simultaneous boundary shape estimation and velocity field de-noising in Magnetic Resonance Velocimetry using Physics-informed Neural Networks(Ushnish Sengupta, A. Kontogiannis, M. Juniper, 2021, arXiv.org)
- Inverse Navier-Stokes Field Reconstruction of High-Reynolds Urban Flows via Hard-Constraint Physics-Informed Neural Networks with Curriculum Learning(Md. Naiem Mobasshir Monim, Zul Ikram Musaddik Rayat, Mohammad Mamun, 2026, 2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence & Networking (QPAIN))
- Two-Stage Physics-Informed Neural Network for 4-D Wind Field Reconstruction From Single-Doppler Phased-Array Radar Observations(You-Cao Wu, Yun-Li Peng, Rongchi Liang, Lang Huang, Yifu Zhang, Jingyi Sun, P. Chan, Hai Li, Chun Shen, Jian-Bing Li, 2026, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing)
- Light-field-aware physics-informed learning for super-resolved volumetric flow reconstruction from sparse light field particle image velocimetry measurements(Xiao-Yu Zhu, Chuan-Long Xu, Zhi-Zhao Che, 2026, The Physics of Fluids)
- Reconstruction of a Continuous Flow Field from Discrete Experimental Data Points using Physics-Informed Neural Networks(M. E. G. Eck, Jakob G. R. von Saldern, P. Z. zur Nedden, A. Orchini, C. Paschereit, 2025, Flow Turbulence and Combustion)
- Flow field reconstruction of underwater vehicle based on physics-informed neural network(Wei Guo, Jun Ding, Cheng Cheng, Tianqi Wu, Xuesen Chu, 2025, No journal)
- Sparse data wind field wake reconstruction based on parametric adaptive physics-informed neural networks(Yu-Zhen Jin, Shixiong Ye, J. Cui, 2026, The Physics of Fluids)
- Physics-informed neural networks for blood flow inverse problems(Jeremías Garay, J. Dunstan, S. Uribe, F. S. Costabal, 2023, arXiv.org)
时均与统计量级稀疏数据反演、谱推断与统计一致性建模(任务边界及可支持结论)
合并"单点/一维数据的统计量与谱推断、统计一致性建模"与"物理约束反演"中的时均量条目。共同点在于恢复目标不是瞬时场而是统计量/时均量:由一维方向谱代数映射估算三维径向谱、随机强迫线性化 N-S 与凸优化在保持统计一致性前提下补全风电场尾流、从稀疏时均数据改进 URANS 建模与 RANS 闭合量反演、由时间平均速度/压力与温度场重建平均流。用于明确区分'统计量预测'与'瞬时流场重建',避免以统计量吻合充当瞬时场准确的证据。
- Mean flow reconstruction of unsteady flows using physics-informed neural networks(Lukasz Sliwinski, Georgios Rigas, 2023, Data-Centric Engineering)
- Reconstruction of the mean velocity, pressure, and turbulent viscosity fields from background-oriented schlieren measurements using a physics-informed neural network(Yulia K. Rudenko, N. Vinnichenko, A. V. Pushtaev, Y. Plaksina, A. V. Uvarov, 2025, The Physics of Fluids)
- Time-averaged flow field reconstruction based on a multifidelity model using physics-informed neural network (PINN) and nonlinear information fusion(En-Ze Rui, Guangqing Zeng, Yi-Qing Ni, Zheng-Wei Chen, Shuo Hao, 2023, International Journal of Numerical Methods for Heat & Fluid Flow)
- Flow temperature field reconstruction based on physics-informed neural network(Yu-Bao Yang, Yang Xu, Hong-Ping Wang, Yiping Liu, Jin-Jun Wang, 2025, The Physics of Fluids)
- Enhancing Unsteady Reynolds-Averaged Navier–Stokes Modelling from Sparse Data Through Sequential Data Assimilation and Machine Learning(Raphäel Villiers, V. Mons, D. Sipp, Eric Lamballais, M. Meldi, 2025, Flow Turbulence and Combustion)
- Algebraic estimation of the three-dimensional radial energy spectrum in quasi-homogeneous and wall-bounded turbulence(K. Kovács, G. Kristóf, 2026, The Physics of Fluids)
- Stochastic Dynamical Modeling of Wind Farm Turbulence(Aditya Bhatt, M. Rodrigues, F. Bernardoni, S. Leonardi, A. Zare, 2023, Energies)
纯数据驱动网络(CNN/自编码器/Transformer/算子)稀疏观测重建与传感器数量、噪声与部署泛化依赖
合并"CNN、Transformer与算子类监督学习重建"与"深度学习稀疏观测重建"中的非时序、非壁面条目。共同点是直接回归'测量→场'而不嵌入方程残差:最优稀疏传感器布点+CNN、卷积网络处理可压缩湍流、MLP+3D-CNN 由表面压力重建涡结构、CAE+Transformer 时序超分辨、物理偏置傅里叶-小波 Transformer 与算子/谱基线对比、多源约束条件潜扩散与对抗式神经算子、以及以'可部署性'(传感器布置、噪声、分布外泛化、物理一致性核查)为轴的综述性工作。为'学习方法相对经典估计的增益'提供基线谱系,但多为非壁湍流或降采样超分辨任务,需在证据表中区分任务设置。
- Theoretical Framework and Use of CNN Reconstruction With Optimal Sparse Sensor Placement in a Flow Field(D. Gomes, A. Espírito Santo, J. Páscoa, 2024, Volume 8: Fluids Engineering)
- Machine-Learning-Based Reconstruction of Turbulent Vortices From Sparse Pressure Sensors in a Pump Sump(Kai Fukami, B. An, M. Nohmi, M. Obuchi, K. Taira, 2022, Journal of Fluids Engineering)
- Reconstruction of time-resolved three-dimensional flow fields based on a multi-domain fusion transformer(De-Zhi Wang, Yuxuan Song, Mo Chen, Yiming Cao, Weihao Xia, Chang-Dong Yu, 2026, The Physics of Fluids)
- A Physics-Informed Fourier-Wavelet Transformer for Multiscale Computational Fluid Dynamics Surrogate Modeling(Somyajit Chakraborty, Mingze Pan, Xi-Zhong Chen, 2026, arXiv.org)
- Flow Field Reconstruction and Wind Pressure Estimation from Sparse Measurements using Physics-Informed Neural Networks: Application to Two-Dimensional Street Canyon Flow(Yi-Chen Wang, Bing-Chao Zhang, Chao-Yi Hu, Hong-Yuan Jia, Cheng-Hao Wei, H. Kikumoto, 2025, Building and Environment)
- High-fidelity flow-field reconstruction from sparse sensors using a hybrid graph neural network-transformer architecture(Yong-Qi Zhang, Jiang-Zhou Peng, Zhi-Qiao Wang, Wen-Xuan Yuan, Yu-Bai Li, Zhi-Fu Zhou, Hao-Ran Xie, Wei-Tao Wu, 2026, The Physics of Fluids)
- DeepTRNet: Time-resolved reconstruction of flow around a circular cylinder via spatiotemporal deep neural networks(Shujin Laima, Xuxi Zhou, Xiao-Wei Jin, Dong-Lai Gao, Hui Li, 2022, The Physics of Fluids)
- Convolutional neural networks for compressible turbulent flow reconstruction(F. Sofos, Dimitris Drikakis, I. Kokkinakis, S. Spottswood, 2023, The Physics of Fluids)
- Reconstructing flow fields from sparse measurements using a convolutional autoencoder integrated with an Informer model(Yu-Lu Liu, Jiakun Long, Bo-Fu Wang, Tienchong Chang, Xiang Qiu, 2025, Acta Mechanica Sinica)
- Deployable deep learning for flow-field reconstruction: sparse measurements, physical credibility, and generalization(Xiang-Jun Ren, Haibin Ma, Xiang Li, Aiying Ma, 2026, Frontiers of Mechanical Engineering)
壁面与单点信号驱动的深度学习重建及时空序列预测(时间分辨恢复、壁面反演缓冲层与湍流边界层预测)
合并"纯数据驱动神经网络与时间序列模型"与"深度学习稀疏观测重建"中的壁面/单点信号条目。共同点是输入为壁面量或单点/单条信号的历史,输出为近壁速度、涡量或未来场:非线性重建条带结构、时空耦合深度网络、跨多雷诺数的 PINN 与 LSTM 时空预测对比、从粗壁面测量深度学习反演湍流场、从壁面大尺度信号重构大尺度结构、光纤分布式传感+深度学习的分类与诊断、壁面信号预测缓冲层速度与聚合物应力。便于检验时间窗口长度、单时刻输入对照与训练数据量的实际贡献。
- Prediction of flow and polymeric stresses in a viscoelastic turbulent channel flow using convolutional neural networks(A. G. Balasubramanian, R. Vinuesa, O. Tammisola, 2025, Journal of Fluid Mechanics)
- Reconstructing channel turbulence from wall observations(By M. P. Encinar, A. Lozano-Durán, J. Jiménez, 2019, No journal)
- Nonlinear reconstruction of streaky structures in turbulent shear flow by experimental data-driven deep neural network(Xingyu Ma, Jiateng Pan, Jiacheng Wang, Nan Jiang, 2025, The Physics of Fluids)
- Spatiotemporal coupling deep neural network for time-resolved flow field reconstruction around a circular cylinder(Xiao-Wei Jin, Meng-Jie Xu, Ye-Fei Yang, Wen-Li Chen, 2024, The Physics of Fluids)
- Deep learning for turbulent boundary layers: spatio-temporal reconstruction and prediction(X. Dong, Wei Zhou, T. Cai, X. Cai, 2026, Proceedings of the 22nd International Symposium on the Application of Laser and Imaging Techniques to Fluid Mechanics)
- From coarse wall measurements to turbulent velocity fields with deep learning(A. Güemes, Hampus Tober, S. Discetti, A. Ianiro, B. Sirmaçek, Hossein Azizpour, R. Vinuesa, 2021, No journal)
- Data-driven reconstruction of large scales in atmospheric boundary layer flow with sparse measurements(Hao-Ran Du, Eric Savory, Laurent Perret, 2026, Experiments in Fluids)
- Integrated Fiber-Optic Sensing and Deep Learning for Supersonic Inlet Flow Reconstruction and Buzz Diagnosis(Yong-Kai Zhu, Jia-Wei Liu, Rui Li, Zi-Yuan Wang, Hai-Tao Wang, 2026, IEEE Sensors Journal)
壁面/多传感器观测驱动的瞬时随机场生成式重建与不确定性量化(条件生成、零样本适应与LSE基线对比)
合并"壁面/多传感器观测驱动的近壁瞬时场随机重建与不确定性量化"与生成式模型条目。共同点把任务定义为'由壁面或少量多传感器观测合成瞬时速度脉动场',强调病态本质:条件流匹配/条件潜扩散的零样本适应不同传感器配置、相对 LSE 与确定性 CNN 在未见实现与稀疏测量下的泛化比较、潜空间生成模型刻画湍流集合、基于物理的多传感器统计最优融合与预测不确定性显式量化。这是与本调研核心问题最贴近、且直接触及'瞬时重建 vs 条件平均'术语辨析的一组。
- Conditional flow matching for generative modelling of near-wall turbulence with quantified uncertainty(Xin-Yang Liu, Xian-Tao Fan, Jian-Xun Wang, 2026, Journal of Fluid Mechanics)
- Reconstruction of velocity fields from wall pressure measurements in a shock wave/turbulent boundary layer interaction(S. Piponniau, E. Collin, P. Dupont, J. Debiève, 2012, International Journal of Heat and Fluid Flow)
- Physics-based multi-sensor fusion for statistically optimal reconstruction of wall-bounded turbulence(Meng-Ying Wang, C. V. Krishna, M. Luhar, Maziar S. Hemati, 2018, Bulletin of the American Physical Society)
- Learning turbulent flows with generative models for super resolution and sparse flow reconstruction(Vivek Oommen, Siavash Khodakarami, Aniruddha Bora, Zhi-Cheng Wang, G. Karniadakis, 2026, Nature Communications)
- A Multi-Source Constrained Conditional Latent Diffusion Model for Reconstruction of Incomplete Turbulent Flow Fields(Xiu-Yan Liu, Cang-Jing Pei, Qi Qi, Da-Lei Song, Ting-Ting Guo, 2026, Flow Turbulence and Combustion)
图神经网络、时空图与图信号压缩:节点/边/观测掩码定义及图结构增益的实证评估
合并"图神经网络与时空图/消息传递流场重建"与"图神经网络、图信号压缩与时空图表达"两组。共同点是显式以图表示流场并做消息传递:可逆 GNN 由翼型表面压力反演全场、图注意力卷积网络在时变几何与极端稀疏输入(特征传播+有效性掩码)下重建并跨实验 PIV 测试、物理约束+流场消息(冲击检测)嵌入边的时空图、非结构网格低分辨率到未解析小尺度超分辨、从稀疏探针重建爆炸场并报告探针数量-误差关系与推理耗时、图小波压缩感知+尺度感知 GNN 与图采样/GNN 自编码器基线对比、以入流短时速度历史解码三维尾迹的注意力 GNN 代理。直接对应'是否已有空间边+时间边联合建模'与'图结构增益是否经消融验证';多面向非壁湍流,迁移条件须说明。
- Graph Neural Networks for Aerodynamic Flow Reconstruction from Sparse Sensing(Gregory Duthé, I. Abdallah, S. Barber, E. Chatzi, 2023, arXiv.org)
- Flow reconstruction in time-varying geometries using graph neural networks(B. Danciu, Vito A. Pagone, B. Böhm, Marius Schmidt, C. Frouzakis, 2024, arXiv.org)
- Physics-constrained and flow-field-message-informed graph neural network for solving unsteady compressible flows(Si-Ye Li, Zhen-Sheng Sun, Yujie Zhu, Chi Zhang, 2024, The Physics of Fluids)
- Super-resolution of turbulent reacting flows on complex meshes using graph neural networks(Priyabrat Dash, K. Aditya, C. Frouzakis, M. Bode, 2026, arXiv.org)
- Rapid and sparse reconstruction of high-speed steady-state and transient compressible flow fields using physics-informed graph neural networks(Jiang-Zhou Peng, Zhi-Qiao Wang, Xiao-Li Rong, Mei Mei, Ming-Yang Wang, Yong He, Wei-Tao Wu, 2024, The Physics of Fluids)
- Multilevel Graph Wavelet Compressed Sensing with Scale-Aware Neural Recovery(Amirhossein Nouranizadeh, Sarang Rajendra Patil, Alan John Varghese, V. Narayanan, Amit Chakraborty, Meng-Jia Xu, 2026, arXiv.org)
- Graph Neural Network Surrogates for Reconstructing Full-Field CFD Flow Fields from Inflow Data(Zhen-Guo Wang, Wen-Qi Fan, Dan-Yu Li, Ke Li, S.-L. Gan, Bo-Wen Yan, 2026, Proceedings of the 2026 2nd International Conference on Big Data, Communication Technology and Computer Applications)
跨领域单点时间序列反演空间场:任务形式化与方法结构参照,不构成流场可恢复性证据
由"跨领域单点时间序列反演空间场"组独立保留。这些工作形式上是'单点/少量观测→空间场',但对象为海洋次表层温度、室内与工程气流等非壁湍流场景;其价值在于提供任务形式化(单点时间序列所编码空间信息量的量化)、网络/插值策略与服务化部署对照。按调研要求必须标注迁移条件,只能作为结构背景,不能直接支撑壁湍流近壁可恢复性结论。
- Reconstructing subsurface temperature fields from single-point time series: a metric learning approach in the South China Sea(Lu-,-Hong-Feng-,-Li-Zheng-Bao-,-Guo-Zhong-Wen Hong, Meng-Yao Wang, Qing Xu, Zhong-Jie He, 2026, Frontiers in Marine Science)
- An airflow velocity field reconstruction method with sparse or incomplete data using physics-informed neural network(Gang Jing, Huan Wang, Xian-Ting Li, Gui-Jin Wang, Yingying Yang, 2024, Journal of Building Engineering)
- A practical approach to flow field reconstruction with sparse or incomplete data through physics informed neural network(Shengfeng Xu, Zhen-Xu Sun, Ren-Fang Huang, D. Guo, Guo-Wei Yang, Shengjun Ju, 2022, Acta Mechanica Sinica)
将三份各含52条文献的初始化分组整合为18个并列分组(共156条,去重后无重叠),按"物理与观测前提→经典估计理论→时间信息利用→稀疏感知与降维表示→状态估计与数据同化→物理约束学习→统计量级反演→纯数据驱动学习→生成式与不确定性→图结构表达→跨领域参照"的脉络排列,每一层对应用于回答调研的核心问题。①第一层(G1物理基础、G2测量与参考场)不提供算法,而是界定"可恢复尺度与时间信息从何而来"以及热线空间平均、TR-PIV传递函数、对流速度选择、参考场分辨率等观测上限,是区分真实热线、PIV抽点、DNS抽点、壁面传感器四类设置的前提。②第二层(G3 LSE/SLSE、G4条件采样与条件平均、G5时间历史与延迟嵌入)构成经典估计谱系:G3/G4本质输出条件平均或相干部分、不能给出未解析随机成分,据此可严格区分"条件平均流动结构估计"与"瞬时流场重建";G5单独承担"时间窗口长度、延迟嵌入、仅用过去观测的因果估计"议题。③第三层(G6 POD/DMD/SPOD、G7低维流形与模态系数/超分辨、G8稀疏感知与测点布置)处理降维表示与观测配置,为学习方法提供线性基线,并给出测点数量-位置-误差关系。④第四层(G9卡尔曼/resolvent状态估计、G10贝叶斯MAP/变分同化/模态同步)以最优估计与同化理论解释时间历史相对单时刻的增益,并可直接讨论因果性与不确定性阈值。⑤第五层(G11 PINN方法层含壁湍流直接相关工作、G12 PINN跨场景观测算子建模、G13时均统计量反演)支撑"物理残差小/与观测一致/真实场准确"三类证据的区分,以及PINN作教师场-伪标签的误差传播风险;G13独立存在以划定"统计量预测"与"瞬时场重建"的任务边界。⑥第六层(G14纯数据驱动网络、G15壁面与单点信号驱动重建及序列预测、G16生成式与不确定性量化)是衡量学习方法相对经典估计进展的主要素材,其中G16最贴近"壁面观测→瞬时脉动场"的核心问题并显式对比LSE基线。⑦第七层(G17 GNN/时空图)专门回答节点-边-掩码如何定义、是否联合空间边与时间边、图结构增益是否经消融与多基线验证,同时如实指出其多面向非壁湍流外流。⑧G18独立保留跨领域单点反演工作,仅作方法结构参照。分组刻意避免按"是否先进"排序,而按观测类型×方法范式×可支撑结论的强度划分,便于在报告中逐组比较观测条件、重建对象、可恢复性假设、时间信息的实际贡献(是否与单时刻输入对照、是否控制容量与数据量)、跨工况与跨雷诺数泛化验证,以及各性能数字的原文位置与归一化方式,避免跨任务错误比较。
合并三份初始分组中"内-外区相互作用模型与大尺度结构条带预测""壁湍流相干结构与时空关联的物理基础""近壁区大尺度结构与时空关联"三组。共同点是不提出重建算法,而是提供回答'单点时间历史能恢复什么、增益来自何种时空关联'的物理依据:对数区单点大尺度信号与内区/近壁特征的标定关系(IOIM/MSO)、跨雷诺数不变性与幅值/尺度调制、涡结构与条带的相干尺度、大尺度组织与内外层调和、粗糙壁模态速度场。它们是判断可恢复尺度与可迁移性的物理标尺;其中部分条目题录信息不足,须在证据表中标注待核实。相关文献:C. Wark et. al, 1998 等 10 篇文献
合并"壁湍流实验测量基础、热线与TR-PIV的分辨率/传递函数及高分辨率参考场""观测条件与测点布置(测量部分)""热线空间分辨率与对流假设"三组。共同点是刻画输入信号与参考场的物理可信边界:热线探针长度、端部导热与高展向长宽比传感的空间滤波如何畸变脉动方差与谱,TR-PIV激光片厚与有效分辨率的传递函数,Taylor冻结假设中当地/瞬时对流速度选择导致的剪切畸变,以及PIV–DNS对比、多平面同步测量与高分辨率实验场/间接测量时间分辨重建的获取方式。这决定'可恢复范围与尺度受何限制'及真实热线、PIV抽点、DNS抽点三类观测设置的差异。相关文献:P. Manovski et. al, 2025 等 13 篇文献
合并"线性随机估计、谱线性随机估计与时空相关/因果信息流分析""线性随机估计、谱线性随机估计与条件采样(LSE/SLSE 本体)""经典线性随机估计(含谱域LSE)"三组中的 LSE/SLSE 主线。共同点是以两点时空相关(互谱、相关张量)为唯一信息来源,用线性/谱域估计核把少量测点(含单点单分量)信号映射到目标位置场或模态系数,并明确给出条件平均化、未解析随机成分与尺度依赖(低阶大尺度可恢复、小尺度仅统计一致)的结论,含涡量到场量的卷积反演、温度-速度耦合、可解释随机估计框架。是主要的方法谱系与基线来源。相关文献:B. Podvin et. al, 2018 等 11 篇文献
由"条件采样与相干结构条件平均""线性随机估计与条件时空相关估计(方法比较部分)"合并而来,与上一组的区别在于:本组的目标是条件平均结构本身(壁面剪应力条件采样、能量传递结构、相干结构随机估计与 LSE 控制应用)以及方法间比较与信息流/因果分析,而非把 LSE 当作全场映射工具。它界定'条件平均流动结构估计'与'瞬时流场重建'的任务边界,说明为何非零时滞相关(对流相关)是时间信息增益的来源,并处理非唯一性与相位同步重建压力场等经典问题。相关文献:Sangjin Ryu et. al, 2021 等 8 篇文献
合并"时间历史、时间延迟嵌入与对流速度"与"POD/DMD模态分解、降阶模型与时间延迟嵌入"中的延迟嵌入理论条目。共同点是显式建模时间:多时间延迟非线性随机估计、基于连续快照差分的延迟嵌入 DMD、Takens 延迟嵌入+流形学习的状态重构与相空间重构/分形维数判据、跨相位时间延迟估计测速、尺度依赖对流速度变异性与猝发过程调制。为回答'时间历史相对单时刻的增益来源''窗口长度与仅用过去观测的因果估计如何影响可恢复性'提供方法学与证据。相关文献:D. Lasagna et. al, 2015 等 8 篇文献
合并"观测条件与测点布置(设计部分)""稀疏感知、传感器布置与缺失/多视场数据下的模态系数恢复"两组。共同点是聚焦输入配置本身:用谱条件数/敏感性或伴随不确定性图选择测点、贪心算法优化稀疏传感器位置并以线性回归恢复 POD 时间系数以拼接多视场 PIV、面向估计的最优布置准则、以及在 PIV 缺失(互补与非互补)数据下用多尺度自编码器补全并与 gappy-POD 对比。恢复对象多为低维系数或缺失区域,指标为归一化估计误差,可直接支撑测点数量/位置/时间窗口影响的讨论。相关文献:Jin Mi et. al, 2026 等 7 篇文献
合并"POD/DMD/SPOD降阶模型与低维流形"(模型部分)与"POD/DMD模态分解、降阶模型"(不含延迟嵌入)两组。共同点是以 POD/DMD/SPOD 基与模态系数为中介,讨论模态截断误差、降阶 Galerkin 模型与稀疏回归涡黏闭合、参数化复杂网络降阶、SPOD 辨识尾流/叶轮模态、POD 与 DMD 相干结构对比及三维模态代理预测。是'模态系数或低维状态恢复'任务的表示基础与线性基线,也是判断未解析尺度如何处理的关键。相关文献:Xiaoyang Zhu et. al, 2023 等 11 篇文献
由"POD/DMD/SPOD降阶模型与低维流形、模态系数恢复"与"POD/降阶模态与低维状态恢复"中的学习型降阶条目合并。共同点是'先投影到低维子空间再重建/预测':可观测增广流形学习支持多源(DNS/实验)数据比较与超分辨验证、SuperOB 再正交化、非定常状态的降阶表示、4D-flow MRI 降阶建模、POD+DMD 多模态数据融合超分辨、潜空间维度影响、POD 模态最小二乘的激光雷达风场重建,以及以神经算子刻画随机强迫预测时空统计。兼作第五部分参考场可比性与潜空间容量影响的素材。相关文献:Kai Fukami et. al, 2025 等 9 篇文献
合并"N-S线性模型/贝叶斯/同化"(线性滤波部分)与"时间分辨状态估计、卡尔曼滤波与数据同化"(壁湍流部分)。共同点是把重建写成状态估计问题并用观测时间历史:基于 N-S 线性随机模型的卡尔曼滤波从单壁法向位置或壁面剪应力恢复大尺度结构并预测估计器性能、槽道流状态估计、resolvent-based 估计、POD+系统辨识+卡尔曼从标量/稀疏传感器恢复时间分辨场、局部测量反演全场。为'时间历史相对单时刻的增益'提供最优估计理论解释,并可讨论因果性与不确定性。相关文献:Vikrant Gupta et. al, 2021 等 7 篇文献
由"N-S线性模型、贝叶斯MAP、4D-Var与模态同步的状态估计"与"数据同化"中的非滤波条目合并。共同点是以先验/协方差/动力学模型做反问题,把整段观测序列纳入:两点协方差先验下的贝叶斯最大后验全场重建、风电场尾流 4D-Var、Lagrangian 粒子记忆的 FlowFit 时序一致性约束、顺序变分同化中的模态同步与延迟坐标(直接对应'仅用过去观测的因果估计'与窗口长度问题)、多步条件高斯非线性同化跨层传递信息、超声测速时间分辨率提升。相关文献:P. Bauweraerts et. al, 2020 等 6 篇文献
合并"物理约束神经网络稀疏含噪数据流场重建的方法与求解效率"与"PINN类流场反演"中的方法学条目。共同点是把 N-S/PDE 残差与稀疏离散监督结合,重点在稀疏度鲁棒性、压力场无测量推断、有限体积/弱形式降低高阶导数开销、域分解与压力基准不确定性、PIV 空间稀疏与测点高频信号的时空互补融合、结构与物理联合重建网络。其中从二维平面测量重建三维可实现壁湍流场(以 LSE 为基线并做物理约束消融)是与本调研最直接相关的核心文献。本组也是区分'物理残差小'与'真实场准确'证据的素材。相关文献:J. Zhu et. al, 2026 等 10 篇文献
由"PINN在工业管网、血流与多孔介质等具体场景中的反演实践"与"物理约束反演与PINN重建"中的场景类条目合并。共同点是强调'观测算子'本身的建模:积分式流量计(线积分)约束的 T 型接头压力重建、超声/彩色多普勒低分辨率低信噪比输入的多帧耦合、4D-flow MRI 与 PIV 的血流动力学、RANS 约束的多孔-流体复合流动、旋转爆震与旋流燃烧、光场 PIV 超分辨(含成像退化与可靠性加权)、LDA 稀疏数据连续化、边界形状与速度联合反演、水下航行器与尾迹稀疏数据反演。提供'把单工况 PINN 反演作为教师场/伪标签'的已有实践与向壁湍流迁移的适用条件限制。相关文献:D. Gomes et. al, 2026 等 16 篇文献
合并"单点/一维数据的统计量与谱推断、统计一致性建模"与"物理约束反演"中的时均量条目。共同点在于恢复目标不是瞬时场而是统计量/时均量:由一维方向谱代数映射估算三维径向谱、随机强迫线性化 N-S 与凸优化在保持统计一致性前提下补全风电场尾流、从稀疏时均数据改进 URANS 建模与 RANS 闭合量反演、由时间平均速度/压力与温度场重建平均流。用于明确区分'统计量预测'与'瞬时流场重建',避免以统计量吻合充当瞬时场准确的证据。相关文献:Lukasz Sliwinski et. al, 2023 等 7 篇文献
合并"CNN、Transformer与算子类监督学习重建"与"深度学习稀疏观测重建"中的非时序、非壁面条目。共同点是直接回归'测量→场'而不嵌入方程残差:最优稀疏传感器布点+CNN、卷积网络处理可压缩湍流、MLP+3D-CNN 由表面压力重建涡结构、CAE+Transformer 时序超分辨、物理偏置傅里叶-小波 Transformer 与算子/谱基线对比、多源约束条件潜扩散与对抗式神经算子、以及以'可部署性'(传感器布置、噪声、分布外泛化、物理一致性核查)为轴的综述性工作。为'学习方法相对经典估计的增益'提供基线谱系,但多为非壁湍流或降采样超分辨任务,需在证据表中区分任务设置。相关文献:D. Gomes et. al, 2024 等 10 篇文献
合并"纯数据驱动神经网络与时间序列模型"与"深度学习稀疏观测重建"中的壁面/单点信号条目。共同点是输入为壁面量或单点/单条信号的历史,输出为近壁速度、涡量或未来场:非线性重建条带结构、时空耦合深度网络、跨多雷诺数的 PINN 与 LSTM 时空预测对比、从粗壁面测量深度学习反演湍流场、从壁面大尺度信号重构大尺度结构、光纤分布式传感+深度学习的分类与诊断、壁面信号预测缓冲层速度与聚合物应力。便于检验时间窗口长度、单时刻输入对照与训练数据量的实际贡献。相关文献:A. G. Balasubramanian et. al, 2025 等 8 篇文献
合并"壁面/多传感器观测驱动的近壁瞬时场随机重建与不确定性量化"与生成式模型条目。共同点把任务定义为'由壁面或少量多传感器观测合成瞬时速度脉动场',强调病态本质:条件流匹配/条件潜扩散的零样本适应不同传感器配置、相对 LSE 与确定性 CNN 在未见实现与稀疏测量下的泛化比较、潜空间生成模型刻画湍流集合、基于物理的多传感器统计最优融合与预测不确定性显式量化。这是与本调研核心问题最贴近、且直接触及'瞬时重建 vs 条件平均'术语辨析的一组。相关文献:Xin-Yang Liu et. al, 2026 等 5 篇文献
合并"图神经网络与时空图/消息传递流场重建"与"图神经网络、图信号压缩与时空图表达"两组。共同点是显式以图表示流场并做消息传递:可逆 GNN 由翼型表面压力反演全场、图注意力卷积网络在时变几何与极端稀疏输入(特征传播+有效性掩码)下重建并跨实验 PIV 测试、物理约束+流场消息(冲击检测)嵌入边的时空图、非结构网格低分辨率到未解析小尺度超分辨、从稀疏探针重建爆炸场并报告探针数量-误差关系与推理耗时、图小波压缩感知+尺度感知 GNN 与图采样/GNN 自编码器基线对比、以入流短时速度历史解码三维尾迹的注意力 GNN 代理。直接对应'是否已有空间边+时间边联合建模'与'图结构增益是否经消融验证';多面向非壁湍流,迁移条件须说明。相关文献:Gregory Duthé et. al, 2023 等 7 篇文献
由"跨领域单点时间序列反演空间场"组独立保留。这些工作形式上是'单点/少量观测→空间场',但对象为海洋次表层温度、室内与工程气流等非壁湍流场景;其价值在于提供任务形式化(单点时间序列所编码空间信息量的量化)、网络/插值策略与服务化部署对照。按调研要求必须标注迁移条件,只能作为结构背景,不能直接支撑壁湍流近壁可恢复性结论。相关文献:Lu- et. al, 2026 等 3 篇文献
总计 156 篇相关文献
This paper describes investigations into unsteady flow phenomena in a high turning linear cascade with a nominal steady flow. A fast response five hole probe was used to obtain time-accurate measurements. Since a single probe was used to traverse an entire measurement plane, the results were obtained as discrete points and were not synchronous. One method used to address this issue is to traverse the measurement plane point by point and record the signal of a reference probe simultaneously. Synchronization can be performed during post processing by analyzing the reference signal for a common periodic component and phase-shifting all signals with respect to the reference phase of that flow feature. This paper presents results from the application this unsteady reconstruction technique to the so-called Durham Cascade 28% downstream of the trailing edge in order to provide insight into the nature of the dominant unsteady mode. It was found that there was significant temporal variation in the strength of the passage vortex but the position of the vortex remains fixed.
No abstract available
The flow-field reconstruction of a rotating detonation combustor (RDC) is essential to understand the stability mechanism and performance of rotating detonation engines. This study embeds a reduced-order model of an RDC into a neural network (NN) to construct a physics-informed neural network (PINN) to achieve the full-dimensional high-resolution reconstruction of the combustor flow field based on partially observed data. Additionally, the unobserved physical fields are extrapolated through the NN-embedded physical model. The influence of the residual point sampling strategy and observation point spatial-temporal sampling resolution on the reconstruction results are studied. As a surrogate model of the RDC, the PINN fills the gap that traditional computational fluid dynamics methods have difficulty solving, such as inverse problems, and has engineering value for the flow-field reconstruction of RDCs.
Gas turbine combustors commonly feature swirling flows. The swirl is usually characterized by the swirler geometry. The development of swirl-stabilized burners includes the experimental assessment of the resulting flow field and the quantification of the swirl, e.g. through Laser Doppler Anemometry (LDA). LDA accurately acquires flow velocity components in a small probing volume. Nevertheless, the measurement quality represents a trade-off between invested measurement time and spatial resolution. In this work, the potential of Physics-Informed Neural Networks (PINNs) to assimilate flow fields from sparse LDA measurements is investigated. A novel burner is employed in which the swirl is fluidically adjustable from a non-swirled jet to a fully swirled flow through a secondary air flow injection. Data is acquired through LDA within spatial measurement grids at multiple axial distances from the swirler. Assuming symmetry, axial, tangential, and radial velocity components are determined. A PINN is subsequently trained with the acquired data, creating a continuous and differentiable flow field representation by evaluating RANS equations. By systematically reducing the training data while evaluating the physical validity of the reconstructed field, a minimum training data requirement is identified. As a result, for three operating conditions, the flow field is adequately characterized by a minimum of measurement points.
Accurate high-resolution flow field prediction based on limited experimental data is a complex task. This research introduces an innovative framework leveraging physics-informed neural network (PINN) to reconstruct high-resolution flow fields using sparse particle image velocimetry measurements for flow over a periodic hill and high-fidelity computational fluid dynamics data for flow over a curved backward-facing step. Model training utilized mean flow measurements, with increased measurement sparsity achieved through various curation strategies. The resulting flow field reconstruction demonstrated marginal error in both test cases, showcasing the ability of the framework to reconstruct the flow field with limited measurement data accurately. Additionally, the study successfully predicted flow fields under two different noise levels, closely aligning with experimental and high-fidelity results (experimental, direct numerical simulation, or large eddy simulation) for both cases. Hyperparameter tuning conducted on the periodic hill case has been applied to the curved backward-facing step case. This research underscores the potential of PINN as an emerging method for turbulent flow field prediction via data assimilation, offering reduced computational costs even with sparse, noisy measurements.
No abstract available
Processing techniques for particle-based optical flow measurement data such as 3D particle tracking velocimetry (PTV) or the novel dense Lagrangian particle tracking method ‘Shake-the-Box’ (STB) can provide time-series of velocity and acceleration information scattered in space. The following post-processing is key to the quality of space-filling velocity and pressure field reconstruction from the scattered particle data. In this work we describe a straight-forward extension of the recently developed data assimilation scheme FlowFit, which applies physical constraints from the Navier–Stokes equations in order to simultaneously determine velocity and pressure fields as solutions to an inverse problem. We propose the use of additional artificial Lagrangian tracers (virtual particles), which are advected between the flow fields at single time instants to achieve meaningful temporal coupling. This is the most natural way of a temporal constraint in the Lagrangian data framework. FlowFit’s core method is not altered in the current work, but rather its input in the form of Lagrangian tracks. This work shows that the introduction of such particle memory to the reconstruction process significantly improves the resulting flow fields. The method is validated in virtual experiments with two independent DNS test cases. Several contributions are revised to explain the improvements, including correlations of velocity and acceleration errors in the reconstructions and the flow field regularization within the inverse problem.
In the realm of experimental fluid mechanics, accurately reconstructing high-resolution flow fields is notably challenging due to often sparse and incomplete data across time and space domains. This is exacerbated by the limitations of current experimental tools and methods, which leave critical areas without measurable data. This research suggests a feasible solution to this problem by employing an inverse physics-informed neural network (PINN) to merge available sparse data with physical laws. The method's efficacy is demonstrated using flow around a cylinder as a case study, with three distinct training sets. One was the sparse velocity data from a domain, and the other two datasets were limited velocity data obtained from the domain boundaries and sensors around the cylinder wall. The coefficient of determination (R2) coefficient and mean squared error (RMSE) metrics, indicative of model performance, have been determined for the velocity components of all models. For the 28 sensors model, the R2 value stands at 0.996 with an associated RMSE of 0.0251 for the u component, while for the v component, the R2 value registers at 0.969, accompanied by an RMSE of 0.0169. The outcomes indicate that the method can successfully recreate the actual velocity field with considerable precision with more than 28 sensors around the cylinder, highlighting PINN's potential as an effective data assimilation technique for experimental fluid mechanics.
Physics-informed neural networks (PINNs) offer a promising approach to reconstructing flow variables from sparse data, yet most existing studies assume access to dense pointwise measurements, which contrast with the integral sensors commonly available in industrial infrastructures such as water distribution networks. This work investigates pressure field reconstruction in a turbulent T-junction using a PINN constrained exclusively by a small set of virtual integral flow-rate sensors, formulated as line integrals embedded in the loss function, together with the Reynolds-averaged Navier–Stokes equations and boundary conditions. This formulation explicitly aligns the sensing model with the type of integral flow meters found in real pipe networks, avoiding the unrealistic assumption of dense local instrumentation. The proposed framework accurately recovers the global pressure distribution and reproduces the high-gradient structure at the junction, achieving a maximum absolute error of 0.12, while keeping most of the domain within an error range of 0.02–0.06. In particular, the high-gradient triangular pressure signature at the junction is reconstructed with a shape and amplitude that closely match the CFD reference, despite the very limited supervision. This work improves on the peak errors reported in previous studies [1], [2], that used 2% pointwise-sensor supervision level. These results indicate that integral measurement-based PINNs can deliver competitive or superior accuracy while remaining closer to practical industrial monitoring constraints, making them a promising candidate for digital twin deployment in internal flows.
No abstract available
No abstract available
We present an estimation method combining Proper Orthogonal Decomposition (POD) and Linear Stochastic Estimation (LSE). The method is based on a direct mapping of the POD amplitudes from the measurement space to the state space. The method is tested in the turbulent boundary layer for a numerical simulation as well as for experimental data. The goal is to recover the full velocity field on a fine grid from coarse measurements of a single (longitudinal) velocity component. A significant fraction of the turbulent kinetic energy for each component is captured by the estimation. A scale-by-scale analysis shows that lower order modes corresponding to large scales are recovered accurately. Although exact reproduction is not possible at small scales, examination of the spatial and temporal content of the estimated field shows a good statistical agreement with the real field at all scales.
No abstract available
Abstract The role of large-scale motions (LSMs) in energy transfer is investigated by analysing wall-parallel velocity fields at low-to-moderate Reynolds number ($Re_{\tau }=1200\text {--}3500$), which are obtained via a two-dimensional (2-D) particle image velocimetry measurement with large field-of-view. Two types of energy flux, i.e. local interscale energy flux and in-plane spatial energy flux are inspected in detail. Targeting the energy transfer in large-scale regime, an anisotropic filter is designed based on the zero-crossing scale boundary in a 2-D energy transfer spectrum, across which the net energy flux is the maximum. This ‘optimal’ energy flux boundary separates the scale space into an energy donating large-scale part and an energy receiving small-scale one. The crossover energy flux, as well as the associated flow field structures, are studied by conditional statistics and linear stochastic estimation, in which the statistical spanwise symmetry is deliberately broken by designing special velocity gradient conditions for event probing. A strong connection between large-scale energy flux events and LSMs are found. Namely, forward scatter events have higher probability to reside on the wavy flank of low-momentum LSMs, if compared with the scenario of being clamped in the middle of two streamwise-aligned high- and low-momentum LSMs (Natrajan & Christensen, Phys. Fluids, vol. 18, issue 6, 2006, pp. 299–325). Meanwhile, pairs of positive and negative spatial transfer events tend to locate inside LSMs. It is thus argued that the meandering nature of LSMs, which forms the necessary velocity gradient, might play a determining role in the process of large-scale energy transfer. The spatial correlation between them is then schematized in a conceptual model, which explains most of the present observations.
Abstract It is generally believed that the temperature and the velocity fields are highly coupled in compressible wall-bounded turbulence. In the present study, we employ a linear model, i.e. the two-dimensional spectral linear stochastic estimation (SLSE), to study this coupling from the perspective of the multi-scale energy-containing eddies. Particular attention is paid to the relevant statistical characteristics of the temperature field. The connections of the two fields are found to be varied with the wall-normal position in the boundary layer. In a nutshell, their entanglement is strongest in the near-wall region, and only the extreme thermal events cannot be captured by SLSE. In the logarithmic region, only the scales that correspond to the attached eddies and the very large-scale motions (VLSMs) are firmly coupled. The near-wall footprints of the former are organized in an additive manner and fulfil the predictions of the celebrated attached-eddy model. In the outer region, the two fields are linearly coupled only at the scales corresponding to VLSMs. These findings are demonstrated to be insensitive to the Mach number effects and ascribed to the similarity between the momentum and the heat transfer in compressible wall turbulence. It is also shown that it is the Reynolds number rather than Mach number that acts as a key similarity parameter in constructing their coupling. The framework built in the present study may pave a way for investigating the multi-physics coupling in turbulence, and reinforcing our analysing and modelling capability to the interrelated problems.
Streamwise velocity fluctuations in the inner-region of wall-bounded turbulent flows can be predicted by the model of Marusic, Mathis & Hutchins (2010). Only a single large-scale velocity signal from an outer position in the logarithmic region is needed for the model, and all other parameters are determined from a once-off calibration experiment. Here we elucidate part of the model by investigating the scale-dependent coherence magnitude and phase throughout the boundary layer. The collection of coherent scales exhibits a shift with respect to the reference position that is shown to be independent of scale; thus the large-scales are non-dispersive. Because these scales comprise a strong coherence, their signature in the inner-region is predicted from an input signal acquired at the geometric center of the log-region. Previously this was achieved via single-time stochastic estimation. Here we leverage the inherent advantages of spectral linear stochastic estimation for the prediction of these large-scales.
Abstract The interaction between boundary layer turbulence and a porous layer is the cornerstone of interface engineering. In this study, the spatial and spectral-resolved transfer entropy is used to assess the asymmetry of the causal interaction next to the permeable wall. The analysis is based on pore-resolved direct numerical simulation of turbulent channel flow over cylinder arrays. The spatial map of transfer entropy reveals the information flux between the porous medium and arbitrary nearby positions, and paths connecting locations with maximum information transfer are identified. The paths in the ‘top-down’ and ‘bottom-up’ directions, respectively, lean upstream and downstream, demonstrating that the coupling process is directionally dependent. The scale dependence of transfer entropy is inspected with a surrogate data strategy. As wall permeability increases, the active scale range in causal interaction shifts from near-wall vortices to Kelvin–Helmholtz type eddies. In addition, linear stochastic estimation is used to determine the statistical velocity field for a local informative event. In an average sense, the interaction between a convecting sweep or ejection event and the up/down-welling motions at the pore unit is the core mechanism that contributes to the causal coupling. The statistical findings derived from the transfer entropy are then validated using a neural network-based remote sensing model.
No abstract available
No abstract available
No abstract available
No abstract available
No abstract available
No abstract available
No abstract available
No abstract available
The effects of adverse pressure gradients on turbulent structures were investigated by carrying out direct numerical simulations of turbulent boundary layers subjected to adverse pressure gradients. The equilibrium adverse pressure gradient flows were established by using a power law free-stream distribution U∞ ~ x.. Estimates of the conditionally averaged velocity fields associated with the spanwise vortical motion were obtained by using linear stochastic estimation. These results confirm that the outer region of the adverse pressure gradient boundary layer is populated with streamwise-aligned vortex organizations. INTRODUCTION Turbulent boundary layers (TBLs) are subjected to adverse pressure gradients (APGs) in numerous engineering applications, such as diffusers, turbine blades and the trailing edges of airfoils. The upper limit of the efficiency of such devices is almost always determined by the APG, so the behaviour of APG flow is of practical importance. Understanding the coherent structures in TBLs is essential to understanding boundary layer turbulence. Since organized motions play a crucial role in the production of wall turbulence, the study of coherent structures contributes to the accuracy of turbulence models and advances in flow control strategies (Robinson, 1991). Therefore elucidating the behaviour of the coherent structures in APG TBLs provides insight into the physics of APG flow and improves its modelling. Many important features of APG TBLs are quite well understood. In general, as the magnitude of an APG increases, the mean velocity profile develops a large wake region and the turbulent kinetic energy decreases in the near-wall region (Nagano et al. 1993). In an experimental study of an APG TBL, Skåre & Krogstad (1994) found that the statistics in the outer layer are significantly affected by a strong APG. However, little is known about the effect of APG on coherent structures in TBLs. The main objective of this study was to examine statistically the effects of APGs on the coherent structures of TBLs. In the present study, we simulated a spatially developing turbulent boundary layer with adverse pressure gradient using direct numerical simulation in order to elucidate the effect on the outer layer turbulent structures of adverse pressure gradient. First the numerical method is described briefly and some results are compared with the flows with zero pressure gradient. NUMERICAL METHODS Direct numerical simulations are performed to investigate the physics of spatially developing turbulent boundary layer flows subjected to adverse pressure gradient. A schematic diagram of the computational domain is displayed in Fig. 1. Time-dependent zero pressure gradient turbulent inflow data are provided at the inlet based on the method of Lund et al. (1998). A convective boundary condition at the exit has the form (∂u/∂t) + c(∂u/∂x) = 0, where c is the local bulk velocity. The no-slip condition is imposed at the solid wall. Periodic boundary conditions are applied in the spanwise direction. Townsend (1961) and Mellor & Gibson (1966) showed that an approximate equilibrium flow is obtained when the variation of freestream has the form of a power-law relation in the streamwise direction. A free-stream velocity U∞(x) along the upper boundary of the computational domain is prescribed as,
No abstract available
No abstract available
Dynamic mode decomposition (DMD) is extensively used in fluid dynamics to identify dominant flow structures and facilitate flow reconstruction and prediction. Conventional DMD, however, builds temporal snapshot matrices based exclusively on state variables at individual time instants, which often inadequately captures the behavior of localized or complex flows. To overcome this limitation, this study proposes an enhanced DMD technique that incorporates time-delay embedding by using difference values between state variables at consecutive time steps as the building blocks of the snapshot matrix. This approach not only retains instantaneous flow field information but also emphasizes temporal change characteristics, thereby improving the analytical capability of DMD. Moreover, a modal amplitude optimization procedure is integrated to increase the precision of reconstruction and prediction. The proposed method is validated on three test cases with increasing complexity and diverse backgrounds. Results show that the modified method significantly enhances the robustness and reconstruction accuracy of DMD, indicating its improved reliability for practical fluid dynamics applications.
No abstract available
In nonlinear fluid dynamics, identifying and predicting system behavior often requires dimensionality reduction to a low-dimensional manifold that preserves essential geometric and topological properties. This study combines Takens' delay embedding theorem for phase space reconstruction with manifold learning techniques to achieve an invariant representation of flow field dynamics. Additionally, complex network theory is employed to abstract system dynamics into a network structure, capturing community evolution over time. Using Isometric Mapping for nonlinear dimensionality reduction minimizes reconstruction error, facilitating accurate decoding and simplification of system complexity. Clustering algorithms further categorize flow field states into interpretable second-order states. For modeling and prediction, a Markov model-based random walk method is applied, demonstrating high-precision temporal predictions for first-order flow field states and steady-state predictions for second-order states. Unlike traditional symbolic regression, this approach leverages complex network mapping to establish probability transition matrices instead of explicit system equations, ensuring robustness across higher dimensions without sacrificing accuracy. The method proves efficient, unsupervised, and adaptable, promising advancements in nonlinear fluid system modeling and prediction.
No abstract available
Abstract Reconstructing near-wall turbulence from wall-based measurements is a critical yet inherently ill-posed problem in wall-bounded flows, where limited sensing and spatially heterogeneous flow–wall coupling challenge deterministic estimation strategies. To address this, we introduce a novel generative modelling framework based on conditional flow matching for synthesising instantaneous velocity fluctuation fields from wall observations, with explicit quantification of predictive uncertainty. Our method integrates continuous-time flow matching with a probabilistic forward operator trained using stochastic weight-averaging Gaussian, enabling zero-shot conditional generation without model re-training. We demonstrate that the proposed approach not only recovers physically realistic, statistically consistent turbulence structures across the near-wall region but also effectively adapts to various sensor configurations, including sparse, incomplete and low-resolution wall measurements. The model achieves robust uncertainty-aware reconstruction, preserving flow intermittency and structure even under significantly degraded observability. Compared with classical linear stochastic estimation and deterministic convolutional neural network methods, our stochastic generative learning framework exhibits superior generalisation for unseen realisations under same flow conditions and resilience under measurement sparsity with quantified uncertainty. This work establishes a robust semi-supervised generative modelling paradigm for data-consistent flow reconstruction and lays the foundation for uncertainty-aware, sensor-driven modelling of wall-bounded turbulence.
Approximating the three-dimensional radial energy spectrum based on one-dimensional directional data is challenging, as experiments and reduced datasets provide limited spatial information. Additionally, spectral analysis is often restricted directionally by rapid spatial variations in flow properties. Classical relations, such as Batchelor's, formally connect one-dimensional and three-dimensional spectra, but their practical application requires numerical differentiation, which heavily amplifies noise in discrete spectral data. In addition, Batchelor's formula relies on the assumption of isotropy. To avoid these limitations, this study assesses a simple algebraic mapping for radial spectrum estimation. The parametrization and validation of the proposed function are based on direct numerical simulation data of Kolmogorov flow. The approximate radial spectrum is compared against the exact spectrum computed from the full three-dimensional velocity field. In addition, the resulting turbulent kinetic energy and dissipation values are also examined. The results show that the approximate spectrum reproduces the energy-containing and intermediate scales well, preserving the total turbulent kinetic energy. Larger discrepancies appear at high wavenumbers, indicating reduced accuracy near the smallest resolved scales. The conversion is extended to wall-bounded flows by introducing a local mapping function dependent on the nondimensional wall distance. Furthermore, the practical utility of the approximation is demonstrated by fitting the data to Pope's model spectrum to extract the integral length scale across the channel height. Overall, the proposed approach provides a simple, robust, and differentiation-free method for radial spectrum approximation.
Abstract This study explores partial synchronisation in turbulent channel flows using sequential variational data assimilation with sparse observations, emphasising the roles of model and observation uncertainties. Unlike previous work that focused on synchronisation using direct numerical simulation, this study considers synchronisation under imperfect models and noisy data. In the first part, a synchronisation map is constructed, revealing invariance with respect to variations in the predictive model, Reynolds number and mesh resolution. Full synchronisation emerges above a critical level of equivalent observation density. At lower observation densities, modal synchronisation is observed, where the energies of dominant modes evolve independently of initial conditions. As data become sparser, the system transitions to a non-synchronisation regime, with assimilated flows exhibiting minimal correlation with the observations. The second part of this study uses the master flow interpolated from down-sampled sparse observations. The delay-coordinate strategy is introduced to enhance the modal synchronisation. Results indicate that the optimal $\sigma$ lies near the threshold between modal synchronisation and non-synchronisation. This demonstrates that the modal synchronisation serves as a critical prerequisite for leveraging historical information in data assimilation to improve the accuracy of turbulence reconstruction. These findings extend the scope of synchronisation theory and provide valuable guidance for advancing data assimilation methodologies.
Abstract Estimation of instantaneous and statistical properties of wall-bounded turbulence across different Reynolds numbers has significant applications in scientific research and engineering, where the key challenge is to identify the Reynolds number invariance underlying the varying flow properties as Reynolds number changes. In this study, we develop a refined inner–outer interaction model (IOIM) that isolates the footprints of large-scale structures populating within a prescribed wall-normal range $y_{R}^+ \in [ y_{R1}^+ , y_{R2}^+ )$ . When the reference layers are located within $ 80 \leqslant y_{R1}^+ \lt y_{R2}^+ \leqslant 0.3\textit{Re}_\tau$ , the energy spectral densities of the isolated large-scale structures, including the wall-attached and wall-detached eddies, are found to be independent of Reynolds numbers with direct numerical simulations spanning $ \textit{Re}_\tau \sim O(10^2)$ – $O(10^3)$ . The spectral energy densities of the near-wall small-scale motions also show little Reynolds number dependence. The increase in near-wall turbulent kinetic energy with Reynolds number is explained from the identified Reynolds number invariance of coherent structures. Meanwhile, certain amplitude and scale modulation effects caused by large-scale structures are found to make the distribution patterns of fluctuation amplitudes and scales of near-wall structures more susceptible to Reynolds number influence. The findings of this study provide insights for the developments of predictive models for wall-bounded turbulence across different Reynolds numbers.
In the present study, a set of physics-informed and data-driven approaches are examined towards the development of an accurate reduced-order model for a turbulent plane Couette flow. Based on the utilisation of the proper orthogonal decomposition (POD), a particular focus is given to the development of a reduced-order model where the number of POD modes are not large enough to cover the full dynamics of the given turbulent state, the situation directly relevant to the reduced-order modelling for turbulent flows. Starting from the conventional Galerkin projection approach ignoring the truncation error, three approaches enhanced by both physics and data are examined: (1) sparse regression of the POD-Galerkin dynamics; (2) Galerkin projection with an empirical eddy-viscosity model; (3) Galerkin projection with an optimal eddy viscosity obtained from a newly proposed sparse regression – an idea applying the sparse identification of nonlinear dynamics framework to an eddy-viscosity model. The sparse regression of the POD-Galerkin dynamics does not perform well, as the number of POD modes for the given chaotic dynamics appears to be too small. While the unsatisfactory performance of the Galerkin projection model with an empirical eddy viscosity is observed, the newly proposed approach, which combines the concept of an optimal eddy-viscosity closure with a sparse regression, more accurately approximates the chaotic dynamics than the other reduced-order models considered. This is demonstrated with the mean and time scales of the POD mode amplitudes as well as the first- and second-order turbulence statistics.
To reveal the regulatory mechanisms of synthetic jets on coherent structures in a flat-plate turbulent boundary layer at a low friction Reynolds number of 489, single-slot and double-slot synthetic jet configurations were designed. Particle image velocimetry measurements were coupled with three analytical methods, including two-point correlation, space-time correlation, and linear stochastic estimation, to systematically investigate the streamwise evolution of coherent structures and associated drag reduction influences. Experimental results indicate that the double-slot jets achieve an average drag reduction rate of 21.76%, obviously outperforming the single-slot jets, which attain 14.45%. The peak drag reduction rate reaches 25.60% for the double-slot configuration, compared to 20.18% for the single-slot counterpart, while showing weaker streamwise attenuation in control efficacy. Synthetic jets suppress turbulent bursting and interlayer momentum exchange by lifting near-wall low-momentum fluids, compressing the spatiotemporal scales of coherent structures, reducing their migration velocity, and augmenting sweep velocity. The secondary disturbance of the double-slot jets extends the streamwise control range and enhances the disruption of large-scale hairpin vortex packets. This study offers robust experimental and theoretical support for the development of low-power-consumption drag reduction technologies for underwater vehicles.
Abstract Representing complex flows by evolving vortex structures is an important principle in many investigations of wall-bounded turbulence. The practice of this principle benefits from the bi-directional transformation between the velocity field and the corresponding vortex field. While the velocity-to-vortex transformation could be implemented by various vortex identification criteria, few efforts have been devoted to the inverse process. This work develops a linear reconstruction method, which allows an effective reconstruction for the velocity field of wall turbulence based on a given vortex field. The vortex field is defined as a vector field by combining the swirl strength and the real eigenvector of the velocity gradient tensor. The reconstructed velocity fields are calculated by convolution operations on the vortex fields, with the kernel functions derived by the field-based linear stochastic estimation. The method can effectively recover the turbulent motions in a large scale range, showing clear advantages over the Biot–Savart formula in the near-wall region. The method is also employed to investigate the inducing effects of vortices at different heights. The wall-bounding effect on the induced motions is observed from the contribution spectra of vortices. The higher-order moments of the reconstructed streamwise velocity component present larger deviations from the original data, which is discussed and explained reasonably. At last, the vortex fields filtered by prescribed thresholds are employed to reconstruct the velocity fields. It is found that the strongest vortex components occupying 5 % of the total volume can reasonably recover the main flow features including both the near-wall streaks and the large-scale motions.
For wall-bounded flows, the model of Marusic et al. [Science 329, 193 (2010)] allows one to predict the statistics of the streamwise fluctuating velocity in the inner region, from a measured input signal in the logarithmic region. Normally, a user-defined large-scale portion of the input forms the large-scale content in the prediction by scaling its amplitude, as well as temporally shifting the signal to account for the physical inclination of these scales. Incoherent smaller scales are then fused to the prediction via universally expressed fluctuations that are subject to an amplitude modulation. Here we present a refined version of the model using spectral linear stochastic estimation, which eliminates a user-defined scale separation of the input. Now, an empirically derived transfer kernel comprises an implicit filtering via a scale-dependent gain and phase; this kernel captures the coherent portion in the prediction. An additional refinement of the model embodies a relative shift between the stochastically estimated scales in the prediction and the modulation envelope of the universal small scales. Predictions over a three-decade span of Reynolds numbers, Re-tau similar to O(10(3)) to O(10(6)), highlight promising applications of the refined model to high-Reynolds-number flows, in which coherent scales become the primary contributor to the fluctuating energy.
The estimation problem for a fully-developed turbulent channel flow at Reτ = 2000 is considered. Specifically, a Kalman filter is designed using a Navier–Stokes-based linear model. The estimator uses time-resolved velocity measurements at a single wall-normal location (provided by DNS) to estimate the time-resolved velocity field at other wall-normal locations. The estimator is able to reproduce the largest scales with reasonable accuracy for a range of wavenumber pairs, measurement locations and estimation locations. Importantly, the linear model is also able to predict with reasonable accuracy the performance that will be achieved by the estimator when applied to the DNS. A more practical estimation scheme using the shear stress at the wall as measurement is also considered. The estimator is still able to estimate the largest scales with reasonable accuracy, although the estimator’s performance is reduced.
No abstract available
The spatial averaging effects due to a cross-wire probe on the measured turbulence statistics in a wall-bounded flow are investigated using a combined approach of direct numerical simulation data, theoretical methods and experiments. In particular, the wire length (l), spacing () and angle () of a cross-wire probe configured to measure the streamwise and wall-normal velocities are systematically varied to isolate effects of each parameter. The measured streamwise velocity from a cross-wire probe is found to be an average of the filtered velocities sensed by the two wires. Thus, in general, an increase in the sensor dimensions when normalised by viscous units leads to an attenuated variance for the streamwise velocity (), resulting from a larger contribution to the spatial averaging process from poorly correlated velocities. In contrast, the variance for the wall-normal velocity () can be amplified, and this is shown to be the result of an additional contributing term (compared to ) due to differences in the filtered wire-normal velocity between the two wires. This additional term leads to a spurious wall-normal velocity signal, resulting in an amplified variance recorded by the cross-wire probe. Compared to the streamwise and wall-normal velocity variances, the Reynolds shear stress () perhaps surprisingly shows less variation when l, and are varied. The robustness of Reynolds shear stress to the finite sensor size is due to two effects: (i) Reynolds shear stress is devoid of energetic contributions from the near-isotropic fine scales unlike the and statistics, hence cross-wire probe dimensions are typically sufficiently small in terms of viscous unit to adequately capture the statistics for a range of l and investigated; (ii) the dependency arises due to cross terms between the filtered velocities from two wires, however, it turns out that these terms cancel one another in the case of Reynolds shear stress, but not for the and statistics. We note that this does not, however, suggest that is easier to measure accurately than the normal stresses; on the contrary, in a companion paper (Baidya et al 2019 Meas. Sci. Technol. 30 085301) we show that measurements are more prone to errors due to uncertainty in probe geometry and calibration procedure.
No abstract available
We present an approach to sensor placement for turbulent mean flow data assimilation in the context of Reynolds-averaged Navier–Stokes (RANS) simulations. It entails generating a spatial uncertainty map through the eigenspace perturbations (ESPs) of the baseline turbulence model (e.g., the k−ω shear stress transport model) to quantify the epistemic structural errors in the model. A novel greedy search algorithm is proposed to place sensors targeting regions of highest uncertainty in the spatial uncertainty map generated from ESP. The algorithm is computationally efficient (e.g., computational cost negligible compared to a RANS solution) and is both easy to implement and tune. It involves two hyper-parameters (a constraint to avoid sensor clustering and the number of sensors) which we investigate in-depth. A variational (adjoint-based) data assimilation approach is used for flow reconstruction. The proposed strategy was tested on three two-dimensional wall-bounded flows (Reynolds number ranging 5.6×103–9.36×105) involving flow separation and reattachment. For the wall-mounted hump case, we found that data assimilation using 33 sensors with our proposed sensor placement algorithm reduced the average velocity prediction error by 60% vs 38% with a simple uniform placement of sensors. Furthermore, we found that we could achieve 61% error reduction using our algorithm with only three sensors. Notably, in all tested cases, the error reduction using our method for sensor placement was close in accuracy to the instances where the entire flow field data were used for flow reconstruction, which involved two to three orders of magnitude more data points than the placed sensors.
This study explores the implementation of physics-informed neural networks (PINNs) to analyze turbulent flow in composite porous-fluid systems. These systems are composed of a fluid-saturated porous medium and an adjacent fluid, where the flow properties are exchanged across the porous-fluid interface. The segregated PINN model employs a novel approach combining supervised learning and enforces fidelity to flow physics through penalization by the Reynolds-averaged Navier-Stokes (RANS) equations. Two cases were simulated for this purpose: solid block, i.e. porous media with zero porosity, and porous block with a defined porosity. The effect of providing internal training data on the accuracy of the PINN predictions for prominent flow features, including flow leakage, channeling effect and wake recirculation was investigated. Additionally, L2 norm error, which evaluates the prediction accuracy for flow variables was studied. Furthermore, PINN training time in both cases with internal training data was considered in this study. Results showed that the PINN model predictions with second-order internal training data achieved high accuracy for the prominent flow features compared to the RANS data, within a 20% L2 norm error of second-order statistics in the solid block case. In addition, for the porous block case, providing training data at the porous-fluid interface showed errors of 18.04% and 19.94% for second-order statistics, representing an increase in prediction accuracy by 7% compared to without interface training data. The study elucidates the impact of the internal training data distribution on the PINN training in complex turbulent flow dynamics, underscoring the necessity of turbulent second-order statistics variables in PINN training and an additional velocity gradient treatment to enhance PINN prediction.
Explosion flow fields are characterized by shock waves with varying intensity and position (i.e., explosive loads), which are the primary causes of structural damage. Accurate and rapid prediction of explosive loads is crucial for structural blast-resistant design and daily security management. While existing empirical models and numerical simulation methods can capture the propagation characteristics of explosive shock waves, high-precision simulation requires a massive computational workload, which is insufficient to meet the fast computational demands of various explosive scenarios. To address this contradiction, this study constructed a sparse reconstruction model for two-dimensional explosion fields based on machine learning algorithms. The model utilizes sparse observational data to establish a mapping relationship to the distribution of the entire flow field. The model is built by a physics-informed graph neural network (PIGN). The graph neural network is employed to associate node features, while the physical network is utilized to control model convergence, aiming to enhance model performance. Using the constructed dataset, the PIGN model was tested. Performance and generalization capabilities of the model were assessed by comparing its results with numerical simulation. This evaluation analyzed the relative error distribution and error statistical results of the reconstructed flow field. The results indicate that the PIGN model can effectively reconstruct explosion fields, with an average error in the reconstructed flow field below 4%. Furthermore, when the number of probe points reaches 10, the average error of the flow field reconstructed by the model is close to 6%. This model not only provides a highly reliable distribution of explosion overpressure and pressure-time variations but also, with a well-trained model, accomplishes flow field reconstruction within 1 ms. It offers a novel approach for achieving rapid and reasonable prediction of explosion fields or two-dimensional compressible flow fields.
High-fidelity computational fluid dynamics (CFD) simulations are widely used to analyse three-dimensional turbulent wake flows in large-scale engineering domains, but remain too expensive for iterative design optimisation and real-time decision making. This paper proposes an attention-based graph neural network (GNN) surrogate that learns a direct mapping from short inflow velocity histories to three-dimensional (3D) wake fields on an unstructured monitoring grid. The model operates on CFD sampling points connected by a radius-based graph, and combines a temporal inflow encoder with a graph message-passing decoder to reconstruct all three velocity components in a downstream region of interest. A proof-of-concept study is conducted on a high-fidelity CFD dataset emulating a canonical inflow–wake configuration representative of array-type bluff-body flows, where both inflow and interior data are extracted from the same solver. Numerical results show that the surrogate reproduces the main velocity deficit and its downstream recovery with sub-m/s errors, explaining a non-negligible portion of the variance in the dominant streamwise component while capturing large-scale lateral and vertical spreading patterns. Once trained, the GNN surrogate reduces the cost of full-field prediction from hours of CFD computation to milliseconds of neural-network inference, highlighting its potential as a bridge between mesoscale inflow information (e.g. NWP- or lidar-based reconstructions) and fast volumetric flow-field prediction in complex engineering environments.
This paper investigates deep learning methods in the framework of convolutional neural networks for reconstructing compressible turbulent flow fields. The aim is to develop methods capable of up-scaling coarse turbulent data into fine-resolution images. The method is based on a parallel computational framework that accepts five image sets of various resolutions, trained to correspond to the respective fine resolution. The network architecture mainly consists of convolutional layers, constructing an encoder/decoder network. Based on the U-Net scheme, three different implementations are presented, with residual and skip connections. The methods are implemented in a supersonic shock-boundary-layer interaction problem. The results suggest that simple networks perform better when trained on limited data, and this can be a practical and fast solution when dealing with turbulent flow data, where the computational burden is most of the time difficult to decrease. In such a way, a coarse simulation grid can be upscaled to a fine grid.
With the increasing use of deep neural networks as surrogate models for accelerating computational simulations in mechanics, the application of artificial intelligence in computational fluid dynamics has seen renewed interest in recent years. However, the application of deep neural networks for flow simulations has mainly concentrated on relatively simple cases of incompressible flows. The strongly discontinuous structures that appear in compressible flows dominated by convection, such as shock waves, introduce significant challenges when approximating the nonlinear solutions or governing equations. In this work, we propose a novel physics-constrained, flow-field-message-informed (FFMI) graph neural network for spatiotemporal flow simulations of compressible flows involving strong discontinuities. To enhance the nonlinear approximation capability of strong discontinuities, a shock detector method is leveraged to extract the local flow-field messages. These messages are embedded into the graph representation to resolve the discontinuous solutions accurately. A new FFMI sample-and-aggregate-based message-passing layer, which aggregates the edge-weighted attributes with node features on different hop layers, is then developed to diffuse and process the flow-field messages. Furthermore, an end-to-end paradigm is established within the encoder–decoder framework to transform the extracted information from the flow field into latent knowledge about the underlying fluid mechanics. Finally, a variety of one- and two-dimensional cases involving strong shock waves are considered to demonstrate the effectiveness and generalizability of the proposed FFMI graph neural network.
4D‐Flow magnetic resonance imaging (MRI) has enabled in vivo time‐resolved measurement of three‐dimensional blood flow velocities in the human vascular system. However, its clinical use has been hampered by two main issues, namely, low spatio‐temporal resolution and acquisition noise. While patient‐specific computational fluid dynamics (CFD) simulations can address the resolution and noise issues, its fidelity is impacted by accuracy of estimation of boundary conditions, model parameters, vascular geometry, and flow model assumptions. In this paper a scheme to address limitations of both modalities through data‐fusion is presented. The solutions of the patient‐specific CFD simulation are characterized using proper orthogonal decomposition (POD). Next, a process of projecting the 4D‐Flow MRI data onto the POD basis and projection coefficient mapping using generalized dynamic mode decomposition (DMD) enables simultaneous super‐resolution and denoising of 4D‐Flow MRI. The method has been tested using numerical phantoms derived from patient‐specific aneurysmal geometries and applied to in vivo 4D‐Flow MRI data.
The turbulent wake dynamics over hilly terrain with typical topographic shapes and slopes under different oncoming flow conditions are numerically investigated using large-eddy simulations, followed by proper orthogonal decomposition (POD) and dynamic mode decomposition (DMD) analysis. First, the effects of terrain shape, slope and inflow turbulence on the statistical and spectral characteristics of wake turbulence over hilly terrain are clarified. Then, the dominant flow patterns in the topographic wake are identified by POD and DMD, respectively. Finally, the turbulent wake flow fields over typical hilly terrain are reconstructed using DMD. The results show that the peak spectral frequencies at half topography elevation decrease as the terrain shape transitions from three-dimensional to two-dimensional but increase with decreasing terrain slope and oncoming turbulence intensity. From POD and DMD analyses, it is demonstrated that the wake dynamics over steep terrain under turbulent inflow conditions are predominantly governed by the separated shear layer shed from the hillside and hilltop on the horizontal and vertical planes, respectively. Some discrepancies are observed in the higher-order modes extracted from POD and DMD, indicating strong interactions among multi-frequency eddy motions in the topographic wake. Moreover, the wake fields over typical hilly terrain under different incoming flow conditions can be effectively reconstructed using only 30% of DMD modes with reasonable accuracy. The reconstruction errors are primarily concentrated within the separated shear layer due to the strong flow nonlinearity in this region.
This work proposes a novel reduced-order model that integrates proper orthogonal decomposition (POD) and high-order dynamic mode decomposition (HODMD) to explore the dominant aerodynamic modes and interference mechanisms of three-dimensional finite-length tandem square cylinders. By leveraging POD to extract spatially dominant mode characteristics and integrating HODMD's amplitude-dominant modes to identify those significantly contributing to interference effects, results demonstrate that the POD-HODMD model can be effectively applied to the study of interference effects in three-dimensional finite-length tandem square cylinders, providing a robust method for reduced-order analysis of complex flow field data. The interference effects significantly influence the aerodynamic characteristics of the structures within the vortex shedding frequency bandwidth, and various structural spacings modify the positions of shear layer separation and reattachment, as well as their combined effects with the inter-structural flow field on the pressure distribution patterns on each surface of the structures. The interference effects on the flow field patterns vary across different frequency ranges. Low-frequency modes reshape gap vortex distributions and velocity orientations at structural ends, while high-frequency modes modify gap vortex distributions at the mid-span plane and wake continuity of the disturbed structure. The vortex shedding mode causes separation and wake vortices to become fuller and larger in scale. This technology not only provides a systematic method for analyzing aerodynamic and flow field characteristics for tandem structures but also deepens insight into the intricate dance of fluid–structure interactions within such configurations.
Abstract. This study presents and evaluates three wind field reconstruction methods for real-time inflow characterization, with potential applications in lidar-assisted wind turbine control. The first method applies a least-squares fit of proper orthogonal decomposition (POD) modes to lidar measurements (POD-LSQ). The second uses inverse distance weighting (IDW) interpolation across the rotor plane. The third, POD-IDW, applies the POD-LSQ fit to the interpolated field. The methods are tested under semi-realistic conditions derived from large-eddy simulations (LESs), using a hub-mounted lidar sensor implemented in HAWC2 on the DTU 10 MW reference turbine. Measurements are extracted under varying inflow conditions. A rotor-effective wind speed estimate, combined with the known vertical shear profile from LES, serves as the baseline for comparison. Reconstruction performance is quantified using a global mean absolute error, evaluated across combinations of scan count, POD mode number, and lidar beam angle. Optimal parameters are selected based on the minimum error. To assess physical accuracy, reconstructions are compared against true wind speeds, evaluating the effects of probe volume averaging, multi-distance measurement selection, cross-contamination, and other sources of error. For optimal inputs, POD-IDW achieves the highest accuracy, reducing error by 45.5 % compared with the baseline estimation, at 5.4 times the computational cost. IDW performs similarly (44.9 %) with optimal inputs, while POD-LSQ achieves a 39.4 % reduction with minimal overhead (7 %). Spectral analysis shows that volume averaging and scanning strategies introduce low-pass filtering that attenuates high-frequency turbulence, while preserving low-frequency content more accurately than the baseline. Reconstruction quality strongly depends on the number and spatial distribution of lidar measurements and the number of retained POD modes. Although demonstrated under idealized conditions, the methods show strong potential for real-time applications. Future work should integrate these reconstructions with flow-aware controllers to evaluate fatigue load reduction, particularly at tower level.
To meet the real-time requirements of digital twin and enhance computational efficiency, this paper explores the application of modal decomposition techniques, proper orthogonal decomposition (POD) and dynamic mode decomposition (DMD), in the field of pump fluid dynamics instead of traditional numerical simulation, which is limited by the complexity of the Navier–Stokes equations. Before decomposition, transient flow field of pump impeller and volute needs to be calculated under 0.6Q and 1.0Q, with significant differences. The results show that the first five modes capture most of the flow field's energy, with errors remaining below 10−2 even at 30 modes. However, the flow at off-design conditions is more unstable, with higher reconstruction errors using POD and more fragmented flow features in the DMD analysis. More importantly, to address potential incomplete data, gappy-POD was used to reconstruct data from single and multiple snapshots with varying levels of data loss. For single snapshots, reconstruction error is minimally affected by data sparsity, with errors below 0.0005 at 20 modes. For multiple snapshots, reconstruction accuracy is more sensitive to the iteration count and sparsity level, with a negative correlation between the sparsity level and the number of modes. Especially, when the sparsity level is less than or equal to 50%, the error does not significantly decrease after 10 iterations. At an 80% sparsity level, the iteration count significantly impacts the data repairing, with the first ten modes being more beneficial for repairing in missing data after 15 iterations.
No abstract available
This study proposes a three-dimensional mode-based surrogate framework to predict the tornado-like vortex (TLV) derived from the fuzzy neural network and delayed proper orthogonal decomposition method. First, near-break-down TLV is simulated via large-eddy simulation, and its mean, fluctuating and statistical flow feature is analyzed. Then, three-dimensional spatiotemporal features of coherent structure are extracted and interpreted. Next, the capability of the proposed framework to predict the future state of an unsteady chaotic TLV flow field is systematically evaluated, including the spatiotemporal variation of velocity, pressure, and vorticities as well as flow statistics. Finally, parametric analysis is also conducted to investigate the influence of three key parameters [i.e., Fuzzy rules of the state network or output network (K1 or K2), time delayed embedding number (d)] contained in the framework and the step number of forward prediction (K) on the predicted accuracy. Results show that for near-break-down TLV, vortex wandering effect largely affects its dynamical feature, and its three-dimensional characteristics are distinct, exhibiting the essence of the swirling jet flow. 3D mode-based surrogate model can correctly predict the tornado-like vortex with a relative error of less than 2% for the radial, tangential, and vertical velocity component. It is found that fuzzy rules and time-delayed embedding number has great effect on prediction accuracy. Thus, to achieve optimal predicting effect, it is suggested that d is taken as 8, K1, and K2 are taken as 18, and when making multi-step predictions, the largest K should not exceed 7.
No abstract available
Spatiotemporal coherent structures are critical in quantifying the hydrodynamics of dense gas-solid flows. In this study, two data-driven methods, proper orthogonal decomposition (POD) and dynamic mode decomposition (DMD), are applied to identify and characterize the dominant spatiotemporal coherent structures in a bubbling fluidized bed. It is found that (i) with the same number of modes (or coherent structures), POD captures more defined energy than DMD; (ii) the main coherent structure of POD is symmetric and confirms the existence of bubble-emulsion two-phase structure; (iii) the coherent structures with a frequency of zero Hz in DMD analysis can construct the mean flow field more reasonably than POD; and (iv) POD reconstructs the transient flow fields more accurately with the same number of modes. The present study offers insights into the coherent structures in gas-solid systems.
Abstract This study investigated the turbulent wake flow behind a flat-back Ahmed body using a combination of time-resolved tomographic particle image velocimetry measurements and spectral proper orthogonal decomposition (SPOD). The experiments were conducted at a Reynolds number of ReH = 10 000, which is defined as U∞H/ν, where U∞ represents the free-stream velocity, H is the height of the body and ν represents the viscosity. The SPOD analysis revealed four distinct flow motions present in the wake of the Ahmed body, each occupying a specific range of Strouhal number, StH. Here, StH is defined as f × H/U∞, where f is the frequency of the motion. At the lowest resolved StH of 0.007, the system exhibited a bi-stability mode, in which the wake switched between asymmetric states consisting of a tilted toroidal vortex and a streamwise vortex. In the next StH range of 0.014 to 0.123, the flow demonstrated swinging/flapping motions, characterized by small spanwise and vertical movements of the wake barycentre. These movements were attributed to the tilting of the toroidal vortex. The third category included the vortex-shedding motions and consisted of quasi-streamwise vortices. These vortices advected in the downstream direction, causing oblique displacements of the barycentre in the streamwise-vertical plane. The peak energy of the vortex shedding was observed at StH = 0.164. Finally, shear layer instabilities induced small vertical and spanwise velocity fluctuations along the shear layers at a high StH of 1.147.
No abstract available
No abstract available
No abstract available
Fluid dynamics of the heart chamber can provide critical biological cues for understanding cardiac health and disease and have the potential for supporting diagnosis and prognosis. However, directly acquiring fluid dynamics information from clinical imaging remains challenging, as they are often noisy and have limited resolution, preventing accurate detailed fluid dynamics analysis. Image-based flow simulations offer high detail but are typically difficult to align with clinical velocity measurements, and as a result, may not accurately depict true fluid dynamics. Inverse-computing velocity fields from images via intra-ventricular flow mapping (VFM) has been reported, but it can become inaccurate when faced with missing or noisy measurement data, which is common with modalities such as ultrasound. Here, we propose a physics-informed neural network (PINN) framework that can accurately reconstruct detailed 3D flow fields of the cardiac left ventricle within a localized time window, using supervision from color Doppler measurements, despite their low resolution and signal-to-noise ratio. This framework couples PINN solvers at consecutive time frames with discrete temporal numerical differentiation and is thus named the "Coupled Sequential Frame PINN" or CSF-PINN. We used image-based flow simulations of fetal and adult hearts to generate synthetic color Doppler velocity data at different spatial and temporal resolution for testing the framework. Results show that CSF-PINN can accurately predict high levels of fluid dynamics details, including flow patterns, intraventricular pressure gradients, vorticity structures, and energy losses. CSF-PINN outperforms vanilla PINN in both accuracy and computational efficiency, however, its accuracy is more limited for velocity-gradient-dependent parameters, such as vorticity and wall shear stress (WSS) magnitude. CSF-PINN's accuracy is maintained even when color Doppler velocity data are spatially and temporally sparse and noisy, and when complex motions of the mitral valve are modelled. These are scenarios in which previous methodologies, including image-based flow simulations and VFM, have struggled. Additionally, we propose a scheme for advancing fluid dynamics predictions to subsequent time windows by using training from the previous time window to initialize networks for the subsequent window, further minimizing errors.
No abstract available
Flow temperature fields are more challenging to measure than the velocity fields. Accurate reconstruction of temperature distributions from the velocity data is crucial in heat transfer problems, especially for experimental data analysis. We apply a physics-informed neural network (PINN) to accurately infer temperature fields at arbitrary space-time points within the measurement domain, using only the velocity data and initial temperature field. The PINN minimizes a composite loss function that incorporates residuals from both the Navier–Stokes and energy equations. We validate this approach using three representative two-dimensional heat transfer cases: steady flow around a heated cylinder, synthetic jet impinging on a heated wall, and Rayleigh–Bénard convection. Recognizing that the energy and momentum equations are decoupled under incompressible conditions, we introduce an alternative network architecture in which the temperature is predicted via a dedicated subnetwork. This modification improves reconstruction accuracy by approximately 10%.
: Efficient thermal management in liquid cooling systems relies heavily on the 16 accurate reconstruction of temperature and velocity fields. However, obtaining full-17 field information under sparse sensor deployment remains a critical challenge. To 18 address this issue, this study proposes a Physics-Informed Structure-Aware Network 19 (PISA-Net) for adaptive and high-fidelity reconstruction of coupled thermal-fluid fields 20 in liquid-cooled environments with limited measurements. The proposed framework 21 integrates sparse temperature and velocity data with geometric information of heat 22 sources and flow channels, enabling structure-aware representation of varying thermal 23 configurations. A physics-informed loss term, derived from the steady-state energy 24 conservation equation, is incorporated to enforce physical consistency during training. 25 This hybrid learning strategy effectively combines data-driven approximation with 26 physical constraints, improving both predictive accuracy and generalizability. 27 Numerical validation on a representative cold plate configuration demonstrates that 28 PISA-Net achieves a normalized mean absolute error of 0.98% for temperature and 29 velocity field reconstruction using only eight sensor measurements. In addition, the 30
A data assimilation technique for turbulent flows, based on the use of the physics-informed neural network (PINN), is presented. Mean temperature fields, measured with the nonintrusive background-oriented schlieren (BOS) method, and sparse velocity measurements are used as input data. The velocity, pressure, turbulent viscosity, and thermal conductivity fields are reconstructed using PINN. PINN searches for the flow field, which approximately satisfies the governing equations while remaining close to the available experimental data. Reconstruction is performed for the free and impinging subsonic axisymmetric turbulent jets of hot air. The results are compared with conventional Reynolds-averaged Navier–Stokes simulations. The feasibility of taking into account the spatial variation of the turbulent Prandtl number is demonstrated. However, due to a large number of unknown parameters, accurate reconstruction of the turbulent Prandtl number field requires velocity data measured with high spatial resolution.
Aiming at the technical challenge of reconstructing high-resolution flow fields from sparse flow field data, research on flow field reconstruction based on the physics-informed neural network (PINN) for underwater vehicles has been carried out. To begin with, the Lattice Boltzmann Method (LBM) was used to construct the benchmark flow field data (Timesequenced velocity and pressure fields in three-dimensional space) of the SUBOFF model, and the sparse flow field dataset is generated by spatial down-sampling of the benchmark flow field data. Subsequently, the fully data-driven algorithm and the PINN deep learning algorithm for SUBOFF flow reconstruction (with spatio-temporal coordinate points as inputs and velocity and pressure fields as outputs) were developed, the fully data-driven model and the PINN were obtained by training with the sparse flow data set. Eventually, only the spatiotemporal coordinate points from the benchmark flow field data were input into the two models, and the complete flow field was reconstructed using model inference inversion, and the results were compared and analyzed with the benchmark flow field data generated by the LBM, to validate the correctness of the reconstructed flow field for the two models. The results demonstrate that the PINN model is significantly more accurate than the fully data-driven model in reconstructing the flow field from sparse flow field data. The study provides a new flow field reconstruction solution that could contribute to the optimal design of the hydrodynamic performance of underwater vehicles and similar structures.
While physics-informed neural networks (PINNs) provide a robust framework for integrating observational data with physical laws, their application to Doppler radar wind field reconstruction is frequently hindered by nonconvex optimization and peak velocity underestimation. To address these limitations, this article proposes a two-stage PINN (TS-PINN) framework incorporating proxy background initialization and physics-informed regularization for four-dimensional (4-D) wind field reconstruction from single-Doppler radar observations. The first stage employs an orthogonal basis function expansion retrieval method to generate a proxy background wind field for pretraining, effectively guiding the optimization trajectory away from spurious local minima. In the second stage, the network is fine-tuned against radial observations and the anelastic Navier–Stokes equations to resolve complex nonlinear dynamics and achieve physics-based field completion within the sampled spatiotemporal domain. Simulations of a mesoscale typhoon and a fine-scale microburst demonstrate that TS-PINN accurately reconstructs 4-D velocity fields, achieving reductions of 30.36% and 39.98% in the root mean square error of wind speed and direction, respectively, compared to traditional three-dimensional variational method (3D-VAR) and basic PINN baselines. Field experiments further validate that TS-PINN successfully retrieves full wind vectors from single-radar data, showing good agreement with triple-radar variational retrieval products within observable regions.
Physics-informed neural networks (PINNs) have recently emerged and attracted extensive attention as an alternative approach to computational fluid dynamics (CFD) methods, which can provide competitive solutions to a variety of forward and inverse fluid problems. In this study, we reconstruct a 3D wind field around a building model in wind tunnel test with a Reynolds number of 2.4 × 104 by formulating a novel PINN framework, which is the first exploration of PINNs for building wind engineering problems. To surmount the hurdle in multi-objective optimization for PINN training, a dynamic prioritization (dp) self-adaptive loss balance strategy is proposed (termed dpPINN), which adaptively reconciles the loss terms of distinct scales to facilitate convergence in PINN training. A zero-equation turbulence model and the wind velocity data collected in near-wall regions are embedded in dpPINN training. Comparison results indicate that dpPINN predictions show good consistency with observation data, which is superior to two current PINN paradigms in prediction accuracy. Furthermore, the influence of neural network configurations, turbulence models, and the layout arrangements of training points on the dpPINN prediction is investigated. It is demonstrated that the dpPINN could be a powerful auxiliary means for airflow simulation and reconstruction in wind engineering applications.
In this paper, we present an innovative approach using physics-informed neural networks to reconstruct high-frequency, full-field flows, including the pressure field, by integrating sparse, noisy, low-temporal-resolution particle image velocimetry (PIV) data with high-temporal-resolution pressure probe data. This method effectively leverages the spatial richness of PIV data and the temporal abundance of pressure probe data, offering a complementary spatial and temporal data fusion. The incorporation of physical laws via equation constraints enables the neural network to accurately learn and predict complex fluid dynamics, achieving comprehensive flow field reconstructions. Rigorous testing across various flow types—periodic, non-periodic, and complex—against critical parameters like sampling interval (SI), number of grid points per snapshot (NGPIS), and number of pressure probes (NPP) has demonstrated remarkable accuracy. The results show reconstruction errors for velocity components (u, v) and pressure (p) below 5% with sufficient data, and around 10% for v and p, and below 5% for u in data-limited scenarios. A case study with SI = 30, NGPIS = 2000, NPP = 5 underscores the enhanced robustness and accuracy of random sampling, especially under various noisy conditions. Thus, this approach shows significant potential for temporal and spatial reconstruction of flow fields.
Flow field reconstruction from sparse and noisy data remains a challenging task in fluid mechanics due to low efficiency and high hardware demands. This study introduces the finite volume method into the physics-informed neural network (PINN) method, proposing a finite volume physics-informed neural network (FVPINN) method under weak-form constraints. The method transforms volume integrals to surface integrals via the divergence theorem and employs control volume discretization, thereby reducing reliance on automatic differentiation for high-order derivatives. The experimental results show that FVPINNs achieve comparable prediction accuracy to Vanilla PINNs in the two-dimensional (2D) steady Cavity flow and Kovasznay flow problems, under both noiseless and noisy conditions, with average training time reduced by 44.91% and 43.94%, respectively. In the unsteady 2D cylinder flow scenarios, FVPINNs exhibit slightly higher sensitivity to noise, particularly in velocity field reconstruction, yet still maintain predictive performance comparable to Vanilla PINNs across all noise levels. Notably, FVPINNs achieve a significant reduction of 45.46% in average training time, highlighting their superior computational efficiency in transient flow modeling. Overall, FVPINNs significantly improve computational efficiency while maintaining prediction accuracy, showing strong robustness and potential for broader application.
Reconstructing physically consistent flow fields from sparse, indirect observations is a fundamental challenge in scientific machine learning. In urban environments, wind velocity measurements are scarce, whereas low-cost scalar pollutant sensors are increasingly ubiquitous. This paper presents a physics-informed deep learning framework that reconstructs full wind velocity fields using only sparse scalar concentration measurements, effectively converting pollution sensors into “virtual velocity sensors.” The framework introduces two key architectural innovations: (i) hard enforcement of boundary conditions via a distance-function ansatz, which embeds no-slip constraints directly into the network to eliminate boundary leakage; and (ii) a Reynolds number curriculum learning strategy (homotopy continuation) that mitigates spectral bias and stabilizes optimization in convection-dominated regimes. Without relying on explicit turbulence models, the approach captures dominant flow structures using a geometry-governed Effective Reynolds number formulation. Validation against the CODASC wind tunnel benchmark demonstrates accurate reconstruction of the primary vortex topology and leeward pollutant accumulation, confirming the method's robust generalization to real physical systems.
Real-time acquisition of wind turbine wake data supports monitoring and layout optimization of smart wind farms, attracting significant attention in the field of wind energy. However, existing measurement technologies, such as light detection and ranging (LiDAR), have notable limitations as they can only collect sparse data points. Previous PINN-based wake reconstruction studies typically treat viscosity or diffusivity as a global constant. To overcome this limitation, we propose a parameterized adaptive physics-informed neural network (PA-PINN) that learns a spatiotemporally adaptive diffusivity λ(x,y,t) through a lightweight sub-network. This design enables the model to adjust the local diffusion intensity according to the wake shear level while strictly satisfying the Navier–Stokes equations. The reference velocity field data are derived from the EllipSys3D Large-Eddy Simulation (LES) under neutral atmospheric conditions. The improved model has the following advantages: (i) it captures two-dimensional information, including the streamwise velocity (U) and lateral velocity (V); (ii) it achieves high-precision reconstruction of the flow field in unknown areas with minimal data; (iii) it employs an adaptive optimization strategy based on the flow field gradient to accelerate convergence speed and reduce error. The experimental results demonstrate that the model achieves high-precision wake reconstruction using only 1% sparse data and can stably reconstruct the overall flow field structure under substantial noise interference, showcasing strong robustness and versatility. Furthermore, the spatiotemporal evolution experiment further verifies the model's advantages in capturing complex flow behaviors, offering an accurate and reliable solution for wake reconstruction and wind farm layout optimization, with broad application prospects in wind energy and other fluid mechanics systems.
Light field particle image velocimetry (LF-PIV) enables single-camera volumetric flow measurement, but its performance is limited by low spatial resolution and large depth-wise velocity uncertainty. This study proposes a LF-aware physics-informed neural network (LF-PINN) for super-resolved flow reconstructions from sparse LF-PIV measurements. Building on standard PINNs, which enforce governing equations during learning, the proposed method further incorporates LF imaging characteristics into the physics-informed framework to improve the reconstruction accuracy. This is achieved by introducing a degradation model and a reliability-weighted learning strategy that account for the spatial averaging effect and the anisotropic velocity uncertainty inherent in LF-PIV measurements. Numerical tests on forced isotropic turbulence demonstrate that LF-PINN effectively corrects depth-wise velocity errors, reducing the relative error of the depth-wise velocity component to below 10% when initial measurement errors reach 20%. Incorporating pressure information further enhances flow reconstruction accuracy, reducing the root mean square divergence of the flow field by 22%. Compared with standard PINN formulations and a proper orthogonal decomposition-based interpolation method, LF-PINN achieves superior performance in both spatial and temporal super-resolution reconstruction from sparse measurements. Fine-scale vortical structures and turbulence statistics are recovered with higher fidelity. Experimental measurements of the wake behind a circular cylinder further show that LF-PINN produces clearer wake structures and velocity profiles that are in better consistence with planar-PIV references. These results demonstrate that embedding LF imaging characteristics into physics-informed learning provides an effective route toward super-resolved and physically consistent volumetric flow reconstruction.
No abstract available
No abstract available
High-resolution flow field measurements often face a trade-off between spatial resolution and field of view (FOV) size, particularly in particle image velocimetry (PIV). To address this issue, we propose a data-driven method for reconstructing temporally consistent entire flow fields from time-series multi-FOV data acquired independently using a single camera. The method uses reduced-order modeling and sparse sensing to estimate time-varying coefficients of the proper orthogonal decomposition modes by minimizing discrepancies in overlapping regions between adjacent FOVs. The sparse sensor placement is optimized using a greedy algorithm, and the reconstruction problem is solved as a linear regression problem. The proposed method was validated using both simulated benchmark data and experimental PIV data obtained around a flat-plate airfoil. Results demonstrate that the proposed method accurately reconstructs unsteady flow structures, including vortex shedding and separation phenomena, with a normalized estimation error of approximately 0.17 when sufficient sensors are deployed. Moreover, the sensor distributions, which are concentrated in highly varying regions, indicate effective feature capture. This proposed method enables high-resolution observation of unsteady aerodynamic phenomena with minimal hardware.
Subsurface ocean observations remain severely limited in spatial coverage due to the high cost and operational difficulty of in-situ deployment. Although moored buoys and profiling floats enable continuous, minute-level sampling at fixed locations, the temporal evolution information they record is largely underutilized in existing frameworks—where point measurements are typically reduced to daily means or treated as static interpolation nodes. This study investigates a fundamental question: can single-point time series encode sufficient spatial information to reconstruct the surrounding temperature field? To address this, we propose a metric learning-based framework that learns to align temporal features with spatial patterns in a shared latent space, enabling spatial inversion from single-point time series. The framework is validated using the 100 m depth temperature field in the northern South China Sea, derived from a high-resolution reanalysis product. Results show that temporal and spatial features exhibit no significant correlation when optimized solely for reconstruction; after metric learning-based joint adjustment, they form clear cluster structures in the latent space. The reconstructed spatial fields capture the dominant modes of oceanic features, primarily internal wave-related spatial patterns, though fine-scale details remain limited. Quantitative evaluation yields an RMSE of 0.12 °C and an R² of 0.21, providing the first quantified evidence for the feasibility of the “time-depth for spatial-breadth” strategy. These findings suggest that single-point temporal dynamics encode meaningful spatial information, offering a new perspective for sparse-observation ocean state estimation.
— The present work aims to study and compare two Linear Stochastic Estimation (LSE) methods, namely conventional single-time LSE and a much less explored technique, Spectral Linear Stochastic Estimation (SLSE). To assess the accuracy of each technique the estimated unsteady pressure fields are compared directly to the actual instantaneous pressure field using the results of a long run-time Large-Eddy Simulation of the flow over a backward-facing step. Wall pressure fluctuations were obtained through virtual multi-point measurements placed on the wall and were used to estimate the time-resolved field pressure. Conventional single-time LSE significantly underpredicted the field pressure fluctuations. The rms errors in the instantaneous estimate were lowest in the vicinity of the reference probes due to the single-point technique not incorporating correlations at non-zero time-lags associated with mean convection. The estimated pressure field was improved significantly using spectral estimation achieving a 400 to 500% improvement in rms error between the estimated and actual pressure in the shear-layer region.
No abstract available
In this paper, we consider the reconstruction of 3D turbulent flow fields from a time series of lidar data in a conventionally neutral atmospheric boundary layer (CNBL). For the reconstruction we use the maximum a posteriori estimate of the flow field. This corresponds to an optimization problem, with a cost function that has two contributions; a first term originating from the prior belief on the probability of having a certain turbulent flow field without any observations. Flow field fluctuations are assumed normally distributed and thus statistically fully determined by the mean and two-point covariance of the velocity field. The second term, is related to the likelyhood of the observations, influenced by model and measurement uncertainties. The two-point covariance is computed and found to be significantly altered by the Coriolis force, breaking up longer streamwise velocity streaks and veering spanwise structures by ∼ 45° with respect to the mean flow direction. For the reconstruction, we consider two different scanning modes, a plan position indicator (PPI) mode and a trajectory which is based on a Lissajous curve. For the PPI scanning mode we find that the mean squared error of the reconstructed velocity field is around 10% of the background variance in the scanning plane, and quickly increases outside this region. The Lissajous curve on the other hand attains an average error of 40% over the scanning region, which spans almost the whole BL height.
No abstract available
Complexities of flow patterns in the azimuthal cross-section of a cylindrical magnetized helicon plasma and the corresponding plasma dynamics are investigated by means of a novel scheme for time delay estimation velocimetry. The advantage of this introduced method is the capability of calculating the time-averaged 2D velocity fields of propagating wave-like structures and patterns in complex spatiotemporal data. It is able to distinguish and visualize the details of simultaneously present superimposed entangled dynamics and it can be applied to fluid-like systems exhibiting frequently repeating patterns (e.g., waves in plasmas, waves in fluids, dynamics in planetary atmospheres, etc.). The velocity calculations are based on time delay estimation obtained from cross-phase analysis of time series. Each velocity vector is unambiguously calculated from three time series measured at three different non-collinear spatial points. This method, when applied to fast imaging, has been crucial to understand the rich ...
No abstract available
We propose a spatiotemporal coupling deep neural network approach for time-resolved reconstruction of the velocity field around a circular cylinder. The neural network leverages two distinct data types: (1) non-time-resolved velocity field around the cylinder, consisting of fixed frequency sampling and variable frequency sampling velocity field, and (2) the time-resolved surface pressure sequence around the cylinder. The deep neural network comprises two sub-networks: a convolutional autoencoder (CAE) for nonlinear mode extraction and a Transformer for sequence-to-sequence learning. We refer to this architecture as CTNet (CAE-Transformer Network). The encoder in the CAE maps non-time-resolved velocity field to a latent vector, enabling the extraction of nonlinear modal coefficients. An appropriate time window length for the surface pressure sequence is then selected to establish a Transformer sequence learning model, using the chosen sequence as input to predict the corresponding nonlinear modal coefficients. Once the Transformer is well trained, the time-resolved nonlinear modal coefficients of velocity field can be achieved. Along with the well-trained decoder in the CAE, the time-resolved velocity field can be reconstructed from the output of the Transformer. We verify the performance of CTNet by a simulated dataset at a representative Reynolds number of 3900. The results show a relative reconstruction error of just 6.3% for the time-resolved velocity field, demonstrating high reliability in the reconstruction. We further compare the reconstructed velocity field obtained with and without the utilization of variable frequency sampling velocity field. Notably, the inclusion of variable frequency sampling velocity field significantly improves the reconstruction quality.
Computational fluid dynamics (CFD) plays an important role in the research of naval and ocean engineering. In recent years, the introduction of deep learning technology has brought new opportunities to fluid mechanics, which not only reduces the consumption of computational resources and time costs but also improves the accuracy of numerical simulations and experimental measurements. Nevertheless, recent works have attempted to incorporate temporal information into super-resolution reconstruction, most mainstream methods still focus on single-frame spatial reconstruction, failing to fully leverage the and spatiotemporal correlations of fluid flow. To address this issue, we propose a time-resolved three-dimensional flow-field reconstruction framework, called spatiotemporal feature interaction-based super-resolution transformer framework, based on a multi-domain fusion transformer. First, a multi-domain fusion transformer is designed to fuse the multi-scale features of the flow field by capturing different spatial and frequency-domain characteristics. Second, we employ a spatiotemporal feature interaction module to extract the inherent spatiotemporal relationships between consecutive flow fields. This allows the model to reconstruct a series of N high-resolution flow fields when fed with a continuous sequence of N low-resolution flow fields. In addition, corresponding turbulence super-resolution datasets are constructed for model training and testing, covering different types of fluid flows. Finally, to further verify the generalization performance of the model, a flow case around the National Advisory Committee for Aeronautics 0012 and 4412 hydrofoil are established based on CFD simulations. Experimental results show that the proposed time-resolved three-dimensional flow-field reconstruction framework not only achieves high-precision results in reconstructing various fluid fields but also obtains the optimal super-resolution reconstruction performance in the hydrofoil flow case, demonstrating excellent flow-field reconstruction capability and strong generalization.
We propose spatiotemporal deep neural networks for the time-resolved reconstruction of the velocity field around a circular cylinder (DeepTRNet) based only on two flow data types: the non-time-resolved wake velocity field and sparse time-resolved velocity measurements at specific discrete points. The DeepTRNet consists of two operations, i.e., compact spatial representations extraction and sequential learning. We use the convolutional autoencoder (CAE) in DeepTRNet to extract compact spatial representations embedded in the non-time-resolved velocity field. The nonlinear CAE modes and corresponding CAE coefficients are thus obtained. Based on the nonlinear correlation analysis of the velocity field, we use the bidirectional recurrent neural networks (RNN) with the gated recurrent unit for mapping the sparse time-resolved velocity measurements to the CAE coefficients via sequential learning. The early stopping technique is used to train the DeepTRNet to avoid overfitting. With the well-trained DeepTRNet, we can reconstruct the time-resolved velocity field around the circular cylinder. The DeepTRNet is verified on the simulated datasets at two representative Reynolds numbers, 200 and 500, and the experimental dataset at Reynolds number 3.3 × 104 with the steady jet at the rear stagnation point of the cylinder. We systematically compare the DeepTRNet method and the RNN-proper orthogonal decomposition (POD) approach. The DeepTRNet can obtain the accurate time-resolved velocity field depending on the two data types mentioned above. The DeepTRNet method outperforms the RNN-POD method in the reconstruction accuracy, especially for the reconstruction of small-scale flow structures. In addition, we get the reliable velocity field even for the high-frequency components.
No abstract available
No abstract available
This work proposes a novel approach that combines an optimal sparse sensor placement algorithm with a Convolutional Neural Network (CNN) to accurately reconstruct computational fluid dynamics (CFD) simulations. Fluid mechanics is a field with an ever-increasing amount of high-quality data, and machine learning offers new opportunities to extract essential flow features and enable real-time predictions and control. The key challenge addressed herein is to determine the optimal placement of a minimal number of sensors to capture the complexity of fluid flows while enabling fast CNN-based reconstruction. The proposed methodology was applied to a lid-driven cavity flow dataset comprising 100 steady laminar simulations of the vertical velocity component. The sparse sensor placement algorithm was used to determine the optimal sensor positions, which were then used to transform the original CFD dataset to feed into the CNN. The results demonstrate the approach’s effectiveness, achieving an accuracy of 95.50% using only 1000 sensors out of a possible set of 16384 and a single Proper Orthogonal Decomposition (POD) mode. This work has the potential to significantly reduce the computational burden associated with CFD simulations within Digital Twins, enabling real-time predictions and control for various fluid flow applications.
A data-driven algorithm is proposed for flow reconstruction from sparse velocity and/or scalar measurements. The algorithm is applied to the flow around a two-dimensional, wall-mounted, square prism. To reduce the problem dimensionality, snapshots of flow and scalar fields are processed to derive POD modes and their time coefficients. Then a system identification algorithm is employed to build a reduced order, linear, dynamical system for the flow and scalar dynamics. Optimal estimation theory is subsequently applied to derive a Kalman estimator to predict the time coefficients of the POD modes from sparse measurements. Analysis of the flow and scalar spectra demonstrate that the flow field leaves its footprint on the scalar, thus extracting velocity from scalar concentration measurements is meaningful. The results show that remarkably good reconstruction of the flow statistics (Reynolds stresses) and instantaneous flow patterns can be obtained using a very small number of sensors (even a single scalar sensor yields very satisfactory results for the case considered). The Kalman estimator derived at one condition is able to reconstruct with acceptable accuracy the flow fields at two nearby off-design conditions. Further work is needed to assess the performance of the algorithm in more complex, three-dimensional, flows.
The development of micro-hot-wire anemometry probes for turbulence studies requires mitigating spatial filtering, end-conduction effects and probe intrusivity. Keeping these factors in mind, this work analytically and experimentally investigates the relevance of a micro-hot-wire probe design featuring elongated stubs, inspired by Wollaston-wire probes but fabricated using modern micro-fabrication techniques. The resulting probes are shown to be relatively easy to manufacture and capable of providing satisfactory velocity measurements in a zero-pressure-gradient turbulent boundary layer at Reτ≈1150\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text {Re}_{\tau }\approx 1150$$\end{document} with δ=18mm\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta ={18}{\hbox {mm}}$$\end{document}. Different probes were tested, all featuring a micro-wire length sufficiently small to alleviate spatial filtering of near-wall small-scale turbulent structures. The investigation focuses on assessing end-conduction effects and probe intrusivity, with the latter still observable close to the wall for such micro-probes.
This study presents the application of time-resolved particle image velocimetry (TR-PIV) to measure the mean and fluctuating velocity components in a turbulent boundary layer (TBL) over an axisymmetric body of revolution. A narrow wall-normal strip of the flow was captured using a synchronised high-speed laser and camera at a recording frequency of up to 80 kHz. The resulting streamwise and wall-normal velocity TR-PIV data were validated against hot-wire anemometry measurements and direct numerical simulations (DNS) of a flat plate under matched flow conditions. The mean flow results showed good agreement between all methods, while the expected attenuation due to the spatial averaging was found in the TR-PIV turbulence statistics closer to the wall. A key outcome of this study is the establishment of an effective laser sheet thickness for the TR-PIV using DNS as a reference. This study fills a gap in understanding the spectral response and limitations of TR-PIV in such complex flows, particularly how spatial resolution and noise influence the accuracy of turbulence measurements. The TR-PIV streamwise velocity energy spectra were compared with DNS data that were spatially filtered to match the resolution of the TR-PIV and hot-wire. A transfer function was derived to determine cut-off wavelengths as a function of wall-normal distance. The cut-off wavelengths enabled the quantification of the resolvable turbulence scales within the TBL, revealing that the spatial resolution is a limiting factor for TR-PIV. The methodology was applied to locations with zero and favourable pressure gradients, providing insights into how pressure gradients influence spectral content and the limitations of TR-PIV in capturing the full range of turbulence scales. The outlined methodology is applicable more broadly and can be used to enhance the accuracy of experimental techniques in future boundary layer investigations.
This study proposes a new method for detecting bursting events based on a natural visibility graph and reveals the amplitude modulation of the large-scale to small-scale bursting processes. Through hot-wire measurements of the turbulent boundary layer, velocity signal sequences with different normal heights are obtained. First, a natural visibility graph method is used to map the velocity signal into a visibility network to obtain a time series of the network degree centrality. Next, a statistical analysis of the degree centrality shows that it can reflect hidden structural characteristics. Moreover, the degree centrality signals have a comparable spatial/temporal scale to the Taylor microscale, which indicates that the degree centrality could provide a potential method to characterize the instantaneous wavenumber/frequency information of small-scales. Then, a new method for detecting bursting events is provided based on the degree centrality values. Characterizations of small-scale bursting events are investigated, providing results for the amplitude modulation between large-scale and small-scale bursting processes. These results show that, in the near-wall region, an increase in amplitudes of a small-scale bursting process is induced for intervals of positive large-scale fluctuations and vice versa for negative large-scale fluctuations. The outer zone shows opposite. The phenomenon of temporal shifting of the amplitude modulation is then investigated and shows that the high-speed large-scales lag behind the small-scale bursting process in the near-wall region, while the outer region shows the opposite. The distance of the small-scale bursting processes leading to the large-scales in the near-wall space is log-linear with the wall-normal height.
No abstract available
Coherent structures are critical for controlling turbulent boundary layers due to their roles in momentum and heat transfer in the flow. Turbulent coherent structures can be detected by measuring wall shear stresses that are footprints of coherent structures. In this study, wall shear stress fluctuations were measured simultaneously in a zero pressure gradient turbulent boundary layer using two house-made wall shear stress probes aligned in the spanwise direction. The wall shear stress probe consisted of two hot-wires on the wall aligned in a V-shaped configuration for measuring streamwise and spanwise shear stresses, and their performance was validated in comparison with a direct numerical simulation result. Relationships between measured wall shear stress fluctuations and streamwise velocity fluctuations were analyzed using conditional sampling techniques. The peak detection method and the variable-interval time-averaging (VITA) method showed that quasi-streamwise vortices were inclined toward the streamwise direction. When events were simultaneously detected by the two probes, stronger fluctuations in streamwise velocity were detected, which suggests that stronger coherent structures were detected. In contrast to the former two methods, the hibernating event detection method detects events with lower wall shear stress fluctuations. The ensemble-averaged mean velocity profile of hibernating events was shifted upward compared to the law of the wall, which suggests low drag status of the coherent structures related with hibernating events. These methods suggest significant correlations between wall shear stress fluctuations and coherent structures, which could motivate flow control strategies to fully exploit these correlations.
No abstract available
Abstract This paper discusses the issue of measuring velocity fluctuations of turbulent boundary layer using hot-wire probes. The study highlights the problem of spatial resolution, which is essential when measuring small-scales in wall-bounded flows. Additionally, attention was paid to the inconsistency in streamwise fluctuation measurements using single- and X-wire probes. To clarify this problem, the energy spectra using wavelet transformation were calculated. The analysis was performed for turbulent boundary layer flow, which was characterized by Reynolds number based on the friction velocity equal Reτ≈ 1000.
Abstract In this work we propose a neural operator-based coloured-in-time forcing model to predict space–time characteristics of large-scale turbulent structures in channel flows. The resolvent-based method has emerged as a powerful tool to capture dominant dynamics and associated spatial structures of turbulent flows. However, the method faces the difficulty in modelling the coloured-in-time nonlinear forcing, which often leads to large predictive discrepancies in the frequency spectra of velocity fluctuations. Although the eddy viscosity has been introduced to enhance the resolvent-based method by partially accounting for the forcing colour, it is still not able to accurately capture the decay rate of the time-correlation function. Also, the uncertainty in the modelled eddy viscosity can significantly limit the predictive reliability of the method. In view of these difficulties, we propose using the neural operator based on the DeepONet architecture to model the stochastic forcing as a function of mean velocity and eddy viscosity. Specifically, the DeepONet-based model is constructed to map an arbitrary eddy-viscosity profile and corresponding mean velocity to stochastic forcing spectra based on the direct numerical simulation data at $Re_\tau =180$ . Furthermore, the learned forcing model is integrated with the resolvent operator, which enables predicting the space–time flow statistics based on the eddy viscosity and mean velocity from the Reynolds-averaged Navier–Stokes (RANS) method. Our results show that the proposed forcing model can accurately predict the frequency spectra of velocity in channel flows at different characteristic scales. Moreover, the model remains robust across different RANS-provided eddy viscosities and generalises well to $Re_\tau =550$ .
Abstract Stochastically generated instantaneous velocity profiles are used to reproduce the outer region of rough-wall turbulent boundary layers in a range of Reynolds numbers extending from the wind tunnel to field conditions. Each profile consists in a sequence of steps, defined by the modal velocities and representing uniform momentum zones (UMZs), separated by velocity jumps representing the internal shear layers. Height-dependent UMZ is described by a minimal set of attributes: thickness, mid-height elevation, and streamwise (modal) and vertical velocities. These are informed by experimental observations and reproducing the statistical behaviour of rough-wall turbulence and attached eddy scaling, consistent with the corresponding experimental datasets. Sets of independently generated profiles are reorganized in the streamwise direction to form a spatially consistent modal velocity field, starting from any randomly selected profile. The operation allows one to stretch or compress the velocity field in space, increases the size of the domain and adjusts the size of the largest emerging structures to the Reynolds number of the simulated flow. By imposing the autocorrelation function of the modal velocity field to be anchored on the experimental measurements, we obtain a physically based spatial resolution, which is employed in the computation of the velocity spectrum, and second-order structure functions. The results reproduce the Kolmogorov inertial range extending from the UMZ and their attached-eddy vertical organization to the very-large-scale motions (VLSMs) introduced with the reordering process. The dynamic role of VLSM is confirmed in the $-u^{\prime }w^{\prime }$ co-spectra and in their vertical derivative, representing a scale-dependent pressure gradient contribution.
Abstract We use Navier–Stokes-based linear models for wall-bounded turbulent flows to estimate large-scale fluctuations at different wall-normal locations from their measurements at a single wall-normal location. In these models, we replace the nonlinear term by a combination of a stochastic forcing term and an eddy dissipation term. The stochastic forcing term plays a role in energy production by the large scales, and the eddy dissipation term plays a role in energy dissipation by the small scales. Based on the results in channel flow, we find that the models can estimate large-scale fluctuations with reasonable accuracy only when the stochastic forcing and eddy dissipation terms vary with wall distance and with the length scale of the fluctuations to be estimated. The dependence on the wall distance ensures that energy production and energy dissipation are not concentrated close to the wall but are evenly distributed across the near-wall and logarithmic regions. The dependence on the length scale of the fluctuations ensures that lower wavelength fluctuations are not excessively damped by the eddy dissipation term and hence that the dominant scales shift towards lower wavelengths towards the wall. This highlights that, on the one hand, energy extraction in wall turbulence is predominantly linear and thus physics-based linear models give reasonably accurate results. On the other hand, the absence of linearly unstable modes in wall turbulence means that the nonlinear term still plays an essential role in energy extraction and thus the modelled terms should include the observed wall distance and length scale dependencies of the nonlinear term.
Abstract The scale-dependent variability of convective velocities and structure inclination angles in wall-bounded turbulence was studied experimentally via space–time energy spectrum measurements. We found that the variability of convection velocities for large-scale motions (LSMs) decreased inversely with streamwise wavenumbers, and that the variability trend was not fully explained by earlier applications of Kraichnan's ‘random-sweeping’ model of turbulence that assumed perfect scale separation. By analytically extending the random-sweeping model to allow for a dominant large scale in the random-sweeping signal that can interact with other LSMs, we showed how scale interactions can explain the variability trend in convection velocities for LSMs. The variability in convection velocities was also shown to correlate with the scale-dependent inclination angles of coherent structures that were obtained via cross-spectral analysis. Large-scale motions tended to exhibit shallower inclination to the wall with increasing convection velocity, while small-scale motions and those far from the wall exhibited the reverse behaviour. We proposed that these two opposite relationships between inclination angle and convection velocity can be explained in terms of a balance between opposing effects of the mean shear and the coherent structure geometry. Descriptions and models of convection velocity variability effects are useful both for modelling turbulence spectra and explaining the geometry of coherent structures.
Low-fidelity engineering wake models are often combined with linear superposition laws to predict wake velocities across wind farms under steady atmospheric conditions. While convenient for wind farm planning and long-term performance evaluation, such models are unable to capture the time-varying nature of the waked velocity field, as they are agnostic to the complex aerodynamic interactions among wind turbines and the effects of atmospheric boundary layer turbulence. To account for such effects while remaining amenable to conventional system-theoretic tools for flow estimation and control, we propose a new class of data-enhanced physics-based models for the dynamics of wind farm flow fluctuations. Our approach relies on the predictive capability of the stochastically forced linearized Navier–Stokes equations around static base flow profiles provided by conventional engineering wake models. We identify the stochastic forcing into the linearized dynamics via convex optimization to ensure statistical consistency with higher-fidelity models or experimental measurements while preserving model parsimony. We demonstrate the utility of our approach in completing the statistical signature of wake turbulence in accordance with large-eddy simulations of turbulent flow over a cascade of yawed wind turbines. Our numerical experiments provide insight into the significance of spatially distributed field measurements in recovering the statistical signature of wind farm turbulence and training stochastic linear models for short-term wind forecasting.
No abstract available
No abstract available
Physics-informed data-driven reconstruction of turbulent wall-bounded flows from planar measurements
Obtaining accurate and dense three-dimensional estimates of turbulent wall-bounded flows is notoriously challenging, and this limitation negatively impacts geophysical and engineering applications, such as weather forecasting, climate predictions, air quality monitoring, and flow control. This study introduces a physics-informed variational autoencoder model that reconstructs realizable three-dimensional turbulent velocity fields from two-dimensional planar measurements thereof. Physics knowledge is introduced as soft and hard constraints in the loss term and network architecture, respectively, to enhance model robustness and leverage inductive biases alongside observational ones. The performance of the proposed framework is examined in a turbulent open-channel flow application at friction Reynolds number Reτ=250. The model excels in precisely reconstructing the dynamic flow patterns at any given time and location, including turbulent coherent structures, while also providing accurate time- and spatially-averaged flow statistics. The model outperforms state-of-the-art classical approaches for flow reconstruction such as the linear stochastic estimation method. Physical constraints provide a modest but discernible improvement in the prediction of small-scale flow structures and maintain better consistency with the fundamental equations governing the system when compared to a purely data-driven approach.
Abstract Neural network models have been employed to predict the instantaneous flow close to the wall in a viscoelastic turbulent channel flow. Numerical simulation data at the wall are used to predict the instantaneous velocity fluctuations and polymeric-stress fluctuations at three different wall-normal positions in the buffer region. Such an ability of non-intrusive predictions has not been previously investigated in non-Newtonian turbulence. Our comparative analysis with reference simulation data shows that velocity fluctuations are predicted reasonably well from wall measurements in viscoelastic turbulence. The network models exhibit relatively improved accuracy in predicting quantities of interest during the hibernation intervals, facilitating a deeper understanding of the underlying physics during low-drag events. This method could be used in flow control or when only wall information is available from experiments (for example, in opaque fluids). More importantly, only velocity and pressure information can be measured experimentally, while polymeric elongation and orientation cannot be directly measured despite their importance for turbulent dynamics. We therefore study the possibility to reconstruct the polymeric-stress fields from velocity or pressure measurements in viscoelastic turbulent flows. The neural network models demonstrate a reasonably good accuracy in predicting polymeric shear stress and the trace of the polymeric stress at a given wall-normal location. The results are promising, but also underline that a lack of small scales in the input velocity fields can alter the rate of energy transfer from flow to polymers, affecting the prediction of the polymeric-stress fluctuations.
No abstract available
Obtaining reliable flow data is essential for the fluid mechanics analysis and control, and various measurement techniques have been proposed to achieve this goal. However, imperfect data can occur in experimental scenarios, particularly in the particle image velocimetry technique, resulting in insufficient flow data for accurate analysis. To address this issue, a novel machine learning-based multi-scale autoencoder (MS-AE) framework is proposed to reconstruct missing flow fields from imperfect turbulent flows. The framework includes two missing flow reconstruction strategies: complementary flow reconstruction and non-complementary flow reconstruction. The former requires two independent measurements of complementary paired flow fields, posing challenges for real-world implementation, whereas the latter requires only a single measurement, offering greater flexibility. A benchmark case study of channel flow with ordinary missing configuration is used to assess the performance of the MS-AE framework. The results demonstrate that the MS-AE framework outperforms the traditional fused proper orthogonal decomposition method in reconstructing missing turbulent flow, irrespective of the availability of complementary paired faulty flow fields. Furthermore, the robustness of the proposed MS-AE approach is assessed by exploring its sensitivity to various factors, such as latent size, overlap proportion, reconstruction efficiency, and suitability for multiscale turbulent flow structures. The new method has the potential to contribute to more effective flow control in the future, thanks to its characteristic that eliminates the requirement for complementary flow fields.
The efficacy of reconstructing three-dimensional (3D) volumes of time-evolving three-component velocity from planar experimental measurements is explored within strong shear-distorting turbulent flows using Taylor’s frozen turbulence hypothesis. In flows with a strong mean shear-rate the instantaneous turbulence structure is spuriously distorted with the classical Taylor’s hypothesis method, where the local ensemble average velocity is used as the convective velocity to reconstruct 3D volumes. In the current study, two additional models for the convective velocity that extend the classical Taylor’s hypothesis approach are explored with varied levels of mean shear-rate and turbulence intensity in order to reconstruct 4D velocity fields for time-resolved analysis of turbulence structure. The classical Taylor’s frozen turbulence hypothesis is compared with local time average and instantaneous convective velocities, with and without a Poisson solver step to maintain continuity. Direct numerical simulation data of a turbulent channel flow from the Johns Hopkins Turbulence Database are used to assess the accuracy of the three methods while varying mean shear-rate and turbulence intensity independently. The three methods are also applied to time-resolved sPIV measurements on transverse planes within the near-wall surface layer of a canonical flat-plate turbulent boundary layer to assess the statistical means of the reconstructions. The instantaneous convective velocity method is generally found to be most accurate at reconstructing instantaneous velocity fields, although systematic biases are observed in mean statistics. In flows with lower turbulence intensities, the local time average convective velocity is comparable for significantly lower computation and implementation costs.
No abstract available
No abstract available
In this work, we present a nonlinear flow field reconstruction approach to streaks in turbulent shear flow in wind tunnel tests using experimental data-driven deep neural network. A spanwise row of submerged vortex generators is deployed on a planar backward-facing step, and therefore, they induce stationary high- and low-speed streaks in the separated shear flow downstream of the step. Using the discrete velocity field measurements, we train a deep neural network for modeling the nonlinear relations between the vortex generator parameters and the resulting velocity fields. With an optimized hidden layer design, the experimental data-driven model provides a high-resolution nonlinearly interpolated flow field of the streaky structures and demonstrates promising generalization for flow field prediction.
No abstract available
No abstract available
Abstract Unsteady aerodynamic forces in flapping wings arise from complex, nonlinear flow structures that challenge predictive modelling. In this work, we introduce a data-driven framework that links experimentally observed flow structures to sectional pressure loads on physical grounds. The methodology combines proper orthogonal decomposition and quadratic stochastic estimation (QSE) to model and interpret these forces using phase-resolved velocity fields from particle image velocimetry measurements. The velocity data are decomposed in a wing-fixed frame to isolate dominant flow features, and pressure fields are reconstructed by solving the Poisson equation for incompressible flows. The relationship between velocity and pressure modes is captured through QSE, which accounts for nonlinear interactions and higher-order dynamics. We introduce an uncertainty-based convergence criterion to ensure model robustness. Applied to a flapping airfoil, the method predicts normal and axial forces with less than 6 % average error using only two velocity modes. The resulting model reveals an interpretable underlying mechanism: linear terms in the QSE model the circulatory force linked to the formation of vortices on the wing, while quadratic terms capture the nonlinear component due to added-mass effects and flow–vorticity interactions. This data-driven yet physically grounded approach offers a compact tool for modelling the unsteady aerodynamics in flapping systems with potential to generalise to other problems.
No abstract available
In this paper, one aims to support for the characterization of tidal turbine farm sites where complex interactions occur between surface waves and large scale flow structures that develop within the water column. Some specific experiments are carried out to study the interaction between a periodic water surface wave (propagating either with or against the current) and some large scale wake flow structures developing behind an obstacle. Using simultaneous measurements of the velocity field across the water column and the surface wave signal, the Linear Stochastic Estimation is implemented to separate the wave and turbulent parts of the raw data by exploiting the statistical relationship between velocity and surface wave motions. With this new wave-turbulence separation approach, we demonstrate how wave propagation affects the flow throughout the spectral domain and contribute to an increase of the turbulent kinetic energy. The wave impact on the flow structure distortion is also highlighted. All these effects are accentuated when waves propagate against the current. These findings have important implications for the characterization of the tidal turbine farm sites, highlighting the importance of accounting for wave and turbulence effects simultaneously rather than examining them separately.
Abstract This study seeks a low-rank representation of turbulent flow data obtained from multiple sources. To uncover such a representation, we consider finding a finite-dimensional manifold that captures underlying turbulent flow structures and characteristics. While nonlinear machine-learning techniques can be considered to seek a low-order manifold from flow field data, there exists an infinite number of transformations between data-driven low-order representations, causing difficulty in understanding turbulent flows on a manifold. Finding a manifold that captures turbulence characteristics becomes further challenging when considering multi-source data together due to the presence of inherent noise or uncertainties and the difference in the spatiotemporal length scale resolved in flow snapshots, which depends on approaches in collecting data. With an example of numerical and experimental data sets of transitional turbulent boundary layers, this study considers an observable-augmented nonlinear autoencoder-based compression, enabling data-driven feature extraction with prior knowledge of turbulence. We show that it is possible to find a low-rank subspace that not only captures structural features of flows across the Reynolds number but also distinguishes the data source. Along with machine-learning-based super-resolution, we further argue that the present manifold can be used to validate the outcome of modern data-driven techniques when training and evaluating across data sets collected through different techniques. The current approach could serve as a foundation for a range of analyses including reduced-complexity modelling and state estimation with multi-source turbulent flow data.
State estimation in multi-layer turbulent flow fields with only a single layer of partial observation remains a challenging yet practically important task. Applications include inferring the state of the deep ocean by exploiting surface observations. Directly implementing an ensemble Kalman filter based on the full forecast model is usually expensive. One widely used method in practice projects the information of the observed layer to other layers via linear regression. However, large errors appear when nonlinearity in the highly turbulent flow field becomes dominant. In this paper, we develop a multi-step nonlinear data assimilation method that involves the sequential application of nonlinear assimilation steps across layers. Unlike traditional linear regression approaches, a conditional Gaussian nonlinear system is adopted as the approximate forecast model to characterize the nonlinear dependence between adjacent layers. At each step, samples drawn from the posterior of the current layer are treated as pseudo-observations for the next layer. Each sample is assimilated using analytic formulae for the posterior mean and covariance. The resulting Gaussian posteriors are then aggregated into a Gaussian mixture. Therefore, the method can capture strongly turbulent features, particularly intermittency and extreme events, and more accurately quantify the inherent uncertainty. Applications to the two-layer quasi-geostrophic system with Lagrangian data assimilation demonstrate that the multi-step method outperforms the one-step method, particularly as the tracer number and ensemble size increase. Results also show that the multi-step CGDA is particularly effective for assimilating frequent, high-accuracy observations, which are scenarios where traditional EnKF methods may suffer from catastrophic filter divergence.
Turbulence remains a complex aspect of fluid mechanics, where accurately reconstructing high-resolution turbulent flow fields is particularly challenging due to frequently sparse and incomplete experimental data. This study leverages Physics-Informed Neural Networks (PINNs) to reconstruct turbulent flow fields from non-uniform, discrete supervisory data, improving adaptability across varying flow conditions and reducing reliance on direct Reynolds stress measurements—often unavailable in practice. Our novel approach enables indirect learning of Reynolds stress tensors from collected velocity and pressure data, streamlining the computational process bypassing complex turbulence model equations. Specifically, the framework excels in navigating complex dynamics within critical engineering structures, such as T-junctions and sudden expansions, typical in water distribution networks. Extensive testing across data scenarios—with sparsity levels of 100%, 10%, 5%, and 2%—demonstrates the model's robustness, maintaining high accuracy in flow prediction and turbulence characterization. The model notably attained L2 errors of 0.0413 and 0.0434 for pressure and velocity reconstruction in critical areas of the 5% dataset. For a sparsity of 2%, the L2 errors were 0.0837 and 0.0769. These results highlight the potential of our PINN framework to complement traditional turbulence modeling approaches by enabling real-time flow monitoring and predictive analytics within digital twin systems, particularly in scenarios where reduced computational cost is essential and full-resolution Reynolds-Averaged Navier–Stokes or Direct Numerical Simulations are impractical.
Reconstructing flow fields from sparse observations constitutes a fundamental challenge in aerodynamic analysis, since existing methods often fail to reconcile accuracy with computational efficiency under severe data constraints. We introduce a Graph Transformer Reconstruction Network (GTREN), a novel model designed to infer latent data patterns from extremely sparse measurements and to reconstruct high-fidelity, full-field flow solutions. The approach represents discrete grid points as nodes within a graph data structure, enabling a principled encoding of spatial relationships. We integrate a Transformer-style attention mechanism into the graph network's message-passing operations, allowing the model to selectively emphasize salient neighbor interactions while attenuating irrelevant signals; this facilitates efficient and stable information propagation. After training, GTREN uncovers implicit correlations among sparse sensor measurements and successfully generalizes these relationships to reconstruct the entire flow field. Results indicate that, with only ten measurement points, the model accurately reconstructs both pressure and velocity distributions across the field. When compared to Computational Fluid Dynamics simulations, the mean squared error (MSE) of the reconstruction is as low as 0.05%. By contrast, a conventional graph neural network yields an MSE of 18.31%. We also provide a systematic analysis of how both the number and spatial arrangement of measurement points affect reconstruction accuracy. Our analysis reveals that, rather than merely increasing sensor count, strategically locating sensors within the core vortex-shedding region of the wake substantially improves reconstruction accuracy—offering practical guidance for flexible sensor deployment. Overall, GTREN achieves high-precision full-field reconstruction with very few pressure sensors, enabling real-time sparse-sensing flow-field monitoring.
Purpose Current methods for flow field reconstruction mainly rely on data-driven algorithms which require an immense amount of experimental or field-measured data. Physics-informed neural network (PINN), which was proposed to encode physical laws into neural networks, is a less data-demanding approach for flow field reconstruction. However, when the fluid physics is complex, it is tricky to obtain accurate solutions under the PINN framework. This study aims to propose a physics-based data-driven approach for time-averaged flow field reconstruction which can overcome the hurdles of the above methods. Design/methodology/approach A multifidelity strategy leveraging PINN and a nonlinear information fusion (NIF) algorithm is proposed. Plentiful low-fidelity data are generated from the predictions of a PINN which is constructed purely using Reynold-averaged Navier–Stokes equations, while sparse high-fidelity data are obtained by field or experimental measurements. The NIF algorithm is performed to elicit a multifidelity model, which blends the nonlinear cross-correlation information between low- and high-fidelity data. Findings Two experimental cases are used to verify the capability and efficacy of the proposed strategy through comparison with other widely used strategies. It is revealed that the missing flow information within the whole computational domain can be favorably recovered by the proposed multifidelity strategy with use of sparse measurement/experimental data. The elicited multifidelity model inherits the underlying physics inherent in low-fidelity PINN predictions and rectifies the low-fidelity predictions over the whole computational domain. The proposed strategy is much superior to other contrastive strategies in terms of the accuracy of reconstruction. Originality/value In this study, a physics-informed data-driven strategy for time-averaged flow field reconstruction is proposed which extends the applicability of the PINN framework. In addition, embedding physical laws when training the multifidelity model leads to less data demand for model development compared to purely data-driven methods for flow field reconstruction.
No abstract available
No abstract available
Abstract Data assimilation of flow measurements is an essential tool for extracting information in fluid dynamics problems. Recent works have shown that the physics-informed neural networks (PINNs) enable the reconstruction of unsteady fluid flows, governed by the Navier–Stokes equations, if the network is given enough flow measurements that are appropriately distributed in time and space. In many practical applications, however, experimental measurements involve only time-averaged quantities or their higher order statistics which are governed by the under-determined Reynolds-averaged Navier–Stokes (RANS) equations. In this study, we perform PINN-based reconstruction of time-averaged quantities of an unsteady flow from sparse velocity data. The applied technique leverages the time-averaged velocity data to infer unknown closure quantities (curl of unsteady RANS forcing), as well as to interpolate the fields from sparse measurements. Furthermore, the method’s capabilities are extended further to the assimilation of Reynolds stresses where PINNs successfully interpolate the data to complete the velocity as well as the stresses fields and gain insight into the pressure field of the investigated flow.
No abstract available
Monitoring the internal flow stability of supersonic inlets is critical for flight safety but faces dual challenges: acquiring high-fidelity data under harsh environments and inferring global flow states from sparse measurements. To address these challenges, this article proposes a “sensor-to-algorithm” closed-loop framework for precise shock wave monitoring and buzz diagnosis. At the sensing level, a temperature-decoupled fiber Bragg grating (FBG) pressure sensor array is designed to overcome cross sensitivity, ensuring high-fidelity and interference-immune data acquisition in supersonic flows. At the algorithm level, advanced deep learning models are employed to process the sparse optical signals. Specifically, a mask-guided convolutional neural network (MG-CNN) reconstructs global flow field visualizations for terminal shock localization, while a time–frequency-aware CNN (TF-CNN) identifies complex dynamic buzz regimes. Experimental results demonstrate that this integrated framework achieves a shock localization error of 3.8 mm and a 99.04% accuracy in buzz diagnosis, providing a highly reliable solution for supersonic inlet health monitoring.
Accurate reconstruction of velocity and pressure fields in fluid flows is essential for monitoring and control in industrial systems. Physics-Informed Neural Networks (PINNs) have emerged as powerful tools to infer hidden flow quantities by embedding physical laws directly into the deep learning process. However, reconstructing such fields remains particularly challenging due to their indirect nature and high sensitivity to sensor placement. This work introduces a gradient-weighted probabilistic sensor placement strategy within a sparse and decoupled supervision framework for PINNs. Instead of uniformly or deterministically selecting measurement points, the method assigns sampling probabilities based on the local gradient magnitudes of each target variable. This enables strategic allocation of sensors in physically informative regions while maintaining variable-specific sparse supervision. The approach is tested on a two-dimensional turbulent T-junction case, where it significantly improves reconstruction accuracy, particularly for the pressure field. Compared to the baseline strategy used in [1], the $\mathcal{L}_2$ error for pressure is reduced from 0.278 to 0.142, for transverse velocity from 0.144 to 0.094, and for streamwise velocity from 0.084 to 0.046. These results represent reductions of approximately 45–50% in the most challenging variables, highlighting the effectiveness of gradient-informed sensor strategies in enhancing PINN-based flow reconstruction.
This study introduces a physics-informed neural network (PINN) framework for the spatio-temporal reconstruction and prediction of turbulent boundary layer flow fields over a broad range of Reynolds numbers. By incorporating the governing Navier-Stokes (N-S) equations into the loss function, the proposed model effectively integrates physical constraints with sparse flow field data, enabling high-fidelity spatial reconstruction, resolution optimization, and temporal prediction of velocity and pressure fields. The method is validated using simulated shock-induced turbulent boundary layer data and experimental turbulent channel flow data. Results show that the PINN model accurately reconstructs flow fields with 60% missing data, infers pressure fields without pressure measurements, and provides reliable short-term temporal predictions. The framework exhibits strong generalizability and robustness, highlighting its potential for complex turbulent flow analysis in both computational and experimental fluid dynamics.
Fluid dynamics is governed by partial differential equations (PDEs) which are solved numerically. The limitations of traditional methods in data assimilation hinder their effective engagement with experiments. Physics-informed neural network (PINN) has emerged as a hybrid data-physics-driven model for convective problems. However, the approach suffers from low accuracy and poor efficiency due to the way of incorporating PDEs. In this work, a novel convolutional neural network framework integrating finite volume method (FVM) is developed to address the challenge. The interface variables of the grid are predicted by the neural network for the first time, rather than a complex procedure in FVM. The physical law is then learned by minimizing the residual of the discretized conservative form of PDEs. A comparison between this model and the original PINN regarding prediction accuracy demonstrates the superiority of embedding PDEs through FVM. The effects of sampling strategies and quantities are studied. The result confirms the model’s capability to utilize sparsely available internal measurement data. Furthermore, the model performs well even in scenarios where partial initial and boundary conditions are absent.
Getting access to the state of turbulent flow from limited sensor measurements in engineering systems is a major challenge. Development of technologies to accurately estimate the state of the flow is now possible with the use of machine learning. We present a supervised machine learning technique to reconstruct turbulent vortical structures in a pump sump from sparse surface pressure measurements. For the current flow reconstruction technique, a combination of multi-layer perceptron and three-dimensional convolutional neural network is utilized. This technique provides accurate flow estimation from only a few sensor measurements, identifying the presence of adverse vortices. The dependence of the model performance on the amount of training data, the number of input sensors, and the noise levels are investigated. The present machine learning-based flow estimator supports safe operations of pumps and can be extended to a broad range of applications for industrial fluid-based systems.
Spatio-temporal prediction of turbulent flow fields remains a challenging task due to complex multi-scale dynamics and high-dimensional data. Recent advances in deep learning offer promising data-driven approaches, yet their comparative performance under physical constraints warrants further investigation. In this study, a physics-informed neural network (PINN) framework and a long short term memory (LSTM) network are implemented and systematically compared for the spatio-temporal prediction of turbulent boundary layer flows over a wide range of Reynolds numbers. Both models are trained and tested using simulated data from a shock-induced turbulent boundary layer and experimental data from particle image velocimetry (PIV) measurements of turbulent channel flow. Prediction accuracy is evaluated across multiple time horizons, with particular attention to short-term and extended temporal extrapolation. Results indicate that the PINN framework achieves higher physical fidelity and robustness in reconstructing velocity and pressure fields from sparse or incomplete data, especially in scenarios requiring alignment with fundamental flow physics. The comparative analysis highlights the complementary strengths and limitations of physics-embedded and purely data-driven deep learning approaches for turbulent flow prediction, providing guidance for method selection based on prediction horizon, data availability, and physical accuracy requirements.
Sparse sensor placement is central to reconstructing unsteady flow fields from limited measurements. This study develops a spectral-conditioning-based forward sequential sensor placement (FSSP) strategy for sparse reconstruction in a spectral proper orthogonal decomposition (SPOD) basis. The method is motivated by the frequency-domain reconstruction formulation and a perturbation-bound interpretation of the sampled observation matrices. To obtain a single physical sensor layout over the resolved spectrum, the same row-selection matrix is applied to all frequency-dependent SPOD bases. Sensor locations are then selected sequentially to reduce the summed squared condition numbers of the sampled observation matrices. The method is assessed using two flows with distinct spectral characteristics: a narrowband laminar cylinder wake and a broadband turbulent channel-flow slice. The comparisons include a proper orthogonal decomposition (POD) layout obtained by column-pivoted orthogonal–triangular (QR) factorization (POD-QR), a regularized multi-frequency SPOD-D-optimal sensor layout (SPOD-D-opt), and random sensor layouts, all evaluated using a common retained SPOD reconstruction operator to isolate the effect of sensor placement. The results show that layout sensitivity is limited once the dominant structures of the narrowband wake are sufficiently sampled, whereas the role of frequency-resolved conditioning is more evident in the broadband channel flow. In the latter case, SPOD-FSSP generally provides better-conditioned observation matrices together with improvements in global and spectral reconstruction metrics. These findings support spectral conditioning as a practical and physically interpretable criterion for sparse sensing of unsteady flows.
No abstract available
No abstract available
The SuperOB algorithm is introduced for reconstructing flow fields from sparse measurements by re-orthogonalizing a subset of points from a global basis derived via Proper Orthogonal Decomposition (POD). SuperOB is tested on Large Eddy Simulation data of highly transient wind speed ramp and compared against reconstruction using POD. SuperOB accurately captures temporal dynamics and spatial structures using only a few measurement points. SuperOB is computationally efficient and requires neither machine learning nor interpolation. It can infer three-dimensional flow characteristics from one-dimensional measurements, making it suitable for reconstructing dynamic inflows for wind turbine performance validation, load assessment, or real-time control.
High-resolution flow fields are essential for understanding, predicting, and controlling fluid systems, yet they remain difficult to obtain from either experiments or high-fidelity simulations. Deep-learning-based flow-field reconstruction has therefore become an attractive route for recovering dense flow information from sparse sensors, low-resolution simulations, or incomplete measurements. Existing studies have demonstrated impressive reconstruction accuracy on canonical benchmark cases, but benchmark accuracy alone does not ensure engineering usability. This Mini Review reframes deep-learning-based flow-field reconstruction from a deployability-oriented perspective. Here, deployability denotes the usability of a reconstruction method under realistic engineering conditions, including physically constrained sensor placement, noisy or asynchronous measurements, operating states outside the training distribution, and the requirement that reconstructed outputs satisfy physical-consistency checks. Rather than organizing the literature only by network architecture, this review examines three coupled axes: sparse measurement realism, physical credibility, and generalization. Finally, we discuss operator learning, spectral bias, benchmark reproducibility, and propose a minimum pre-deployment validation checklist covering sensor robustness, measurement quality, physical residuals, spectral consistency, generalization, and derived engineering quantities.
Neural operators are promising surrogates for dynamical systems but when trained with standard L2 losses they tend to oversmooth fine-scale turbulent structures. Here, we show that combining operator learning with generative modeling overcomes this limitation. We consider three practical turbulent-flow challenges where conventional neural operators fail: spatio-temporal super-resolution, forecasting, and sparse flow reconstruction. For Schlieren jet super-resolution, an adversarially trained neural operator (adv-NO) reduces the energy-spectrum error by 15 × while preserving sharp gradients at neural operator-like inference cost. For 3D homogeneous isotropic turbulence, adv-NO trained on only 160 timesteps from a single trajectory forecasts accurately for five eddy-turnover times and offers 114 × wall-clock speed-up at inference than the baseline diffusion-based forecasters, enabling near-real-time rollouts. For reconstructing cylinder wake flows from highly sparse Particle Tracking Velocimetry-like inputs, a conditional generative model infers full 3D velocity and pressure fields with correct phase alignment and statistics. These advances enable accurate reconstruction and forecasting at low compute cost, bringing near-real-time analysis and control within reach in experimental and computational fluid mechanics.
No abstract available
No abstract available
Knowledge of the turbulent wind field within the atmospheric boundary layer is of great importance in maximizing wind-farm power generation and reducing structural loading. In this context, we propose a 4D-Var algorithm that integrates wind-turbine power measurements, a large-eddy simulation model, and background information to reconstruct the wind-farm’s inflow field. Following the collection of measurements, the 4D-Var problem is efficiently solved using the L-BFGS optimization algorithm. The results show fast convergence and good tracking of both reference power and wind-speed measurements across the turbines.
Accurate reconstruction of convective heat transfer mechanisms remains a significant challenge in thermal engineering, particularly for ill-posed inverse problems with incomplete boundary conditions. Conventional numerical methods, including finite element method and finite volume approaches, often struggle with solution uniqueness and numerical stability under these ill-posed conditions. In contrast, physics-informed neural networks offer a promising alternative by integrating sparse data with physical conservation laws. In this study, we proposed a novel mixed convection physics-informed neural network (MC-PINN) framework for flow field reconstruction in mixed convection heat transfer problems, utilizing sparse temperature measurements and partial physical constraints. To enhance its ability to capture multi-scale flow features, the MC-PINN framework incorporates Fourier feature embeddings and an augmented Lagrangian method, which dynamically adjusts the loss function weights during training. Bayesian inference is employed for hyperparameter optimization, enhancing model accuracy in handling complex flows. Numerical experiments demonstrate that the proposed MC-PINN can accurately reconstruct the velocity field with a coefficient of determination R2 > 99% using only 0.1% sparse thermal measurements, and it exhibits strong robustness to data noise. Compared to the vanilla PINN, the MC-PINN demonstrates a 57%–58.4% improvement in training efficiency. The extended model generalizes well across varying Richardson numbers, highlighting its effectiveness for ill-posed mixed convection problems and its promise in advancing data-physics fusion in computational fluid dynamics.
No abstract available
This paper introduces the Parsimonious Dynamic Mode Decomposition (parsDMD), a novel algorithm designed to automatically select an optimally sparse subset of dynamic modes for both spatiotemporal and purely temporal data. By incorporating time-delay embedding and leveraging Orthogonal Matching Pursuit (OMP), parsDMD ensures robustness against noise and effectively handles complex, nonlinear dynamics. The algorithm is validated on a diverse range of datasets, including standing wave signals, identifying hidden dynamics, fluid dynamics simulations (flow past a cylinder and transonic buffet), and atmospheric sea-surface temperature (SST) data. ParsDMD addresses a significant limitation of the traditional sparsity-promoting DMD (spDMD), which requires manual tuning of sparsity parameters through a rigorous trial-and-error process to balance between single-mode and all-mode solutions. In contrast, parsDMD autonomously determines the optimally sparse subset of modes without user intervention, while maintaining minimal computational complexity. Comparative analyses demonstrate that parsDMD consistently outperforms spDMD by providing more accurate mode identification and effective reconstruction in noisy environments. These advantages render parsDMD an effective tool for real-time diagnostics, forecasting, and reduced-order model construction across various disciplines.
Scientific machine learning methods such as neural operators and physics-informed neural networks have advanced engineering applications and inverse problems, but their training typically requires large volumes of simulated data. This makes data preparation and model training expensive. We propose Graph Wavelet Compressed Sensing (GWCS), a learning-based framework for offline compression of graph signals by representing them as sparse, interpretable wavelet-domain representations using the spectral graph wavelet transform. The framework combines a nonparametric multilevel importance sampler, which retains high-energy wavelet coefficients within each scale for a given compression ratio, with a scale-aware graph neural network that reconstructs the signal from the sparse coefficients. We evaluate the proposed framework on synthetic approximately band-limited graph signals over random graphs and four PDE simulation datasets over meshes, which include Turbulent Radiative Layer, Viscoelastic Instability, Kolmogorov Flow, and Dynamic Stall. We compare against graph signal sampling methods and graph autoencoder baselines. Results demonstrate that the framework achieves high reconstruction fidelity and substantial data compression compared to existing benchmarks.
State-of-the-art deep learning models have been extensively utilized to reconstruct small-scale structures from coarse-grained data in turbulent flows. However, their application has predominantly been restricted to structured uniform meshes, limiting their applicability to data associated with complex geometries that are typically simulated on structured non-uniform or unstructured meshes. Machine learning (ML) models based on graph neural networks (GNNs), known for their ability to process unstructured data, offer a promising alternative. In this study, we leverage the inherent flexibility of GNNs featuring message passing layers to develop a methodology for reconstructing unresolved small-scale structures from low-resolution data on complex meshes. The accuracy of the proposed approach is demonstrated using two cases: a reacting channel flow on a structured non-uniform mesh, and a reacting hydrogen fueled internal combustion (IC) engine featuring an unstructured mesh. Evaluation of results based on visual agreement, statistical metrics, and cumulative error reduction indicates the effectiveness of the method in accurately reconstructing fine-scale features. Overall, this study provides a pathway for integrating data-driven small-scale reconstruction and subgrid-scale modeling to enhance the accuracy of coarse-grained simulations on complex meshes.
The paper presents a Graph Attention Convolutional Network (GACN) for flow reconstruction from very sparse data in time-varying geometries. The model incorporates a feature propagation algorithm as a preprocessing step to handle extremely sparse inputs, leveraging information from neighboring nodes to initialize missing features. In addition, a binary indicator is introduced as a validity mask to distinguish between the original and propagated data points, enabling more effective learning from sparse inputs. Trained on a unique data set of Direct Numerical Simulations (DNS) of a motored engine at a technically relevant operating condition, the GACN shows robust performance across different resolutions and domain sizes and can effectively handle unstructured data and variable input sizes. The model is tested on previously unseen DNS data as well as on an experimental data set from Particle Image Velocimetry (PIV) measurements that were not considered during training. A comparative analysis shows that the GACN consistently outperforms both a conventional Convolutional Neural Network (CNN) and cubic interpolation methods on the DNS and PIV test sets by achieving lower reconstruction errors and better capturing fine-scale turbulent structures. In particular, the GACN effectively reconstructs flow fields from domains up to 14 times larger than those observed during training, with the performance advantage increasing for larger domains.
Sensing the fluid flow around an arbitrary geometry entails extrapolating from the physical quantities perceived at its surface in order to reconstruct the features of the surrounding fluid. This is a challenging inverse problem, yet one that if solved could have a significant impact on many engineering applications. The exploitation of such an inverse logic has gained interest in recent years with the advent of widely available cheap but capable MEMS-based sensors. When combined with novel data-driven methods, these sensors may allow for flow reconstruction around immersed structures, benefiting applications such as unmanned airborne/underwater vehicle path planning or control and structural health monitoring of wind turbine blades. In this work, we train deep reversible Graph Neural Networks (GNNs) to perform flow sensing (flow reconstruction) around two-dimensional aerodynamic shapes: airfoils. Motivated by recent work, which has shown that GNNs can be powerful alternatives to mesh-based forward physics simulators, we implement a Message-Passing Neural Network to simultaneously reconstruct both the pressure and velocity fields surrounding simulated airfoils based on their surface pressure distributions, whilst additionally gathering useful farfield properties in the form of context vectors. We generate a unique dataset of Computational Fluid Dynamics simulations by simulating random, yet meaningful combinations of input boundary conditions and airfoil shapes. We show that despite the challenges associated with reconstructing the flow around arbitrary airfoil geometries in high Reynolds turbulent inflow conditions, our framework is able to generalize well to unseen cases.
Accurate, spatially resolved flow field measurements are essential for the reliable assessment of hemodynamic quantities in cardiovascular research and clinical practice. Experimental techniques, such as 4D flow MRI, PIV, or Doppler ultrasound, often yield data that are sparse, noisy, or under-resolved, particularly near vessel walls and in regions of complex flow. This limits the fidelity of distributed or derived hemodynamic indicators such as the wall shear stress and the clinical utility of such measurements. To address these challenges, we propose a physics-informed neural network (PINN) framework that integrates the incompressible Navier-Stokes equations with velocity measurements coming from experimental flow field data. By embedding physical laws into data, PINN enhances the reconstruction of velocity fields, enables the estimation of unmeasured quantities such as pressure and wall shear stress, and improves the spatial resolution of hemodynamic indicators. We show the effectiveness of our approach using both in silico and experimental data. First, we apply our method to the FDA nozzle benchmark, leveraging both control particle image velocimetry (PIV) measurements and computational fluid dynamics (CFD) simulations. Next, we apply our method to the more complex case of blood flow in an aneurysm model, exploiting in vitro 4D flow MRI data. In both cases, the synergy between data-driven learning and physics-based regularization yields results that align more closely with ground truth observations than standard CFD or pure data-driven approaches. Our findings highlight the potential of PINNs to improve the fidelity of under-resolved flow field measurements and yield spatially resolved hemodynamic indicators.
Physics-informed surrogate models can accelerate computational fluid dynamics simulations. However, many existing methods reproduce global flow patterns more reliably than localized multiscale structures. This study presents a physics-informed Fourier-wavelet transformer for next-step velocity-field reconstruction in real-world flow benchmarks. The proposed formulation combines hybrid Fourier-wavelet spectral encoding with physics-biased self-attention based on partial differential equation residual diagnostics. It also uses self-supervised pretraining through Masked Physics Prediction and Equation Consistency Prediction. The experiments are conducted on two real benchmark cases: cylinder-wake flow and fluid-structure interaction. All approaches are evaluated under a shared local protocol and compared with spectral, transformer-based, operator-learning, and physics-informed neural-network baselines. On the cylinder-wake benchmark, the proposed model achieves the best aggregate accuracy, with an all-channel normalized mean-squared error of 0.05875 and an all-channel Pearson correlation coefficient of 0.97019. On the fluid-structure-interaction benchmark, it gives the lowest all-channel normalized mean-squared error of $2.70 \times 10^{-4}$, compared with $4.02 \times 10^{-4}$ for the strongest baseline. Component-wise field comparisons and scale-separated diagnostics further show stronger recovery of localized wake structures, including near-body, wake-core, and far-wake features. The results demonstrate improved real-world flow reconstruction while maintaining a practical accuracy-cost tradeoff.
Physics-Informed Neural Networks (PINNs) offer a powerful paradigm for flow reconstruction, seamlessly integrating sparse velocity measurements with the governing Navier-Stokes equations to recover complete velocity and latent pressure fields. However, scaling such models to large spatiotemporal domains is hindered by computational bottlenecks and optimization instabilities. In this work, we propose a robust distributed PINNs framework designed for efficient flow reconstruction via spatiotemporal domain decomposition. A critical challenge in such distributed solvers is pressure indeterminacy, where independent sub-networks drift into inconsistent local pressure baselines. We address this issue through a reference anchor normalization strategy coupled with decoupled asymmetric weighting. By enforcing a unidirectional information flow from designated master ranks where the anchor point lies to neighboring ranks, our approach eliminates gauge freedom and guarantees global pressure uniqueness while preserving temporal continuity. Furthermore, to mitigate the Python interpreter overhead associated with computing high-order physics residuals, we implement a high-performance training pipeline accelerated by CUDA graphs and JIT compilation. Extensive validation on complex flow benchmarks demonstrates that our method achieves near-linear strong scaling and high-fidelity reconstruction, establishing a scalable and physically rigorous pathway for flow reconstruction and understanding of complex hydrodynamics.
Physics-informed neural networks (PINNs) have emerged as a powerful tool for solving inverse problems, especially in cases where no complete information about the system is known and scatter measurements are available. This is especially useful in hemodynamics since the boundary information is often difficult to model, and high-quality blood flow measurements are generally hard to obtain. In this work, we use the PINNs methodology for estimating reduced-order model parameters and the full velocity field from scatter 2D noisy measurements in the ascending aorta. The results show stable and accurate parameter estimations when using the method with simulated data, while the velocity reconstruction shows dependence on the measurement quality and the flow pattern complexity. The method allows for solving clinical-relevant inverse problems in hemodynamics and complex coupled physical systems.
Magnetic resonance velocimetry (MRV) is a non-invasive experimental technique widely used in medicine and engineering to measure the velocity field of a fluid. These measurements are dense but have a low signal-to-noise ratio (SNR). The measurements can be de-noised by imposing physical constraints on the flow, which are encapsulated in governing equations for mass and momentum. Previous studies have required the shape of the boundary (for example, a blood vessel) to be known a priori. This, however, requires a set of additional measurements, which can be expensive to obtain. In this paper, we present a physics-informed neural network that instead uses the noisy MRV data alone to simultaneously infer the most likely boundary shape and de-noised velocity field. We achieve this by training an auxiliary neural network that takes the value 1.0 within the inferred domain of the governing PDE and 0.0 outside. This network is used to weight the PDE residual term in the loss function accordingly and implicitly learns the geometry of the system. We test our algorithm by assimilating both synthetic and real MRV measurements for flows that can be well modeled by the Poisson and Stokes equations. We find that we are able to reconstruct very noisy (SNR = 2.5) MRV signals and recover the ground truth with low reconstruction errors of 3.7 - 7.5%. The simplicity and flexibility of our physics-informed neural network approach can readily scale to assimilating MRV data with complex 3D geometries, time-varying 4D data, or unknown parameters in the physical model.