Jingying Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jingying Wang has a total of 55 co-authors. Here are the top co-authors by number of shared publications:
Chunhian Lee
12 shared publicationsMingrui Li
11 shared publicationsShiyue Zhang
7 shared publicationsXue Song
6 shared publicationsFangzhou Han
4 shared publicationsYue Zhou
School of Aeronautic Science and Engineering, Beihang University, Beijing, China
4 shared publications
Latest Publications
- 1 October 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (10)
Artificial intelligence-driven airfoil aerodynamic design via deep surrogate modeling and evolutionary algorithm
Airfoil aerodynamic design relies on computationally expensive flow solvers, while aerodynamic performance optimization and target pressure distribution inverse design are treated separately. This study develops a framework combining low-dimensional geometric representation, deep surrogate modeling, and evolutionary search for aerodynamic prediction, constrained optimization, and inverse design. A density-guided principal component analysis strategy constructs the airfoil design space, while a fully connected autoencoder provides a compact representation of the surface pressure distribution. A multi-branch residual surrogate network jointly predicts lift coefficient, drag coefficient, and surface pressure coefficient distribution from airfoil geometry, Mach number, and angle of attack; mean absolute errors are 3.765 × 10−3, 1.178 × 10−4, and 1.399 × 10−2, respectively. The surrogate maintains high accuracy for combinations of unseen airfoil geometries and interpolated inflow conditions, demonstrating predictive coverage within the training parameter range. Each inference requires 0.024 ms, and 6 × 104 aerodynamic evaluations are completed in 20.5 s. Coupled with the covariance matrix adaptation evolution strategy, the surrogate supports two constrained design tasks. Aerodynamic-efficiency optimization improves the lift-to-drag ratio in all 20 cases without reducing lift or increasing drag. Target pressure distribution inverse design reconstructs the prescribed suction peak, low-pressure region, and pressure-recovery process. Independent high-fidelity computational fluid dynamics recalculations of 40 optimized airfoils show close agreement with surrogate predictions. Flow-field analyses reveal localized velocity and pressure changes for efficiency-oriented optimization, whereas inverse design produces broader upper-surface acceleration, increased suction, and modified pressure recovery. These results demonstrate an efficient framework linking airfoil geometry, aerodynamic performance, and distributed flow responses.
- 1 October 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (10)
Artificial intelligence-driven airfoil aerodynamic design via deep surrogate modeling and evolutionary algorithm
Airfoil aerodynamic design relies on computationally expensive flow solvers, while aerodynamic performance optimization and target pressure distribution inverse design are treated separately. This study develops a framework combining low-dimensional geometric representation, deep surrogate modeling, and evolutionary search for aerodynamic prediction, constrained optimization, and inverse design. A density-guided principal component analysis strategy constructs the airfoil design space, while a fully connected autoencoder provides a compact representation of the surface pressure distribution. A multi-branch residual surrogate network jointly predicts lift coefficient, drag coefficient, and surface pressure coefficient distribution from airfoil geometry, Mach number, and angle of attack; mean absolute errors are 3.765 × 10−3, 1.178 × 10−4, and 1.399 × 10−2, respectively. The surrogate maintains high accuracy for combinations of unseen airfoil geometries and interpolated inflow conditions, demonstrating predictive coverage within the training parameter range. Each inference requires 0.024 ms, and 6 × 104 aerodynamic evaluations are completed in 20.5 s. Coupled with the covariance matrix adaptation evolution strategy, the surrogate supports two constrained design tasks. Aerodynamic-efficiency optimization improves the lift-to-drag ratio in all 20 cases without reducing lift or increasing drag. Target pressure distribution inverse design reconstructs the prescribed suction peak, low-pressure region, and pressure-recovery process. Independent high-fidelity computational fluid dynamics recalculations of 40 optimized airfoils show close agreement with surrogate predictions. Flow-field analyses reveal localized velocity and pressure changes for efficiency-oriented optimization, whereas inverse design produces broader upper-surface acceleration, increased suction, and modified pressure recovery. These results demonstrate an efficient framework linking airfoil geometry, aerodynamic performance, and distributed flow responses.
- 1 October 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (10)
Effects of thermochemical nonequilibrium on shock interaction flow of V-shaped leading edge
This paper numerically investigates the effects of thermochemical nonequilibrium on the shock interactions and aerodynamic heating over the V-shaped leading edge. The numerical methodology is validated against experimental data from the JF22 hypervelocity wind tunnel. A comparative study is performed using the calorically perfect gas (CPG) model, as well as the one-temperature and two-temperature five-species chemical reaction models of air. The results demonstrate that thermochemical nonequilibrium significantly modifies the post-shock thermodynamic state through internal energy excitation and chemical reactions, thereby altering the shock-layer structure and detached-shock standoff distance. The different relaxation processes predicted by the one-temperature and two-temperature models lead to distinct shock interaction patterns, with the one-temperature model inducing a transition from Mach reflection to regular reflection from the same family. Compared with the CPG model, thermochemical nonequilibrium models predict enhanced wall heat flux due to the coupled effects of modified shock-layer states, conductive heat transfer, and species enthalpy transport. In particular, under the fully catalytic wall condition, atomic recombination releases chemical formation enthalpy, contributing to the species enthalpy diffusion term and the enhanced wall heat flux predicted by the thermochemical nonequilibrium models. These findings demonstrate that accurate thermochemical nonequilibrium modelling is essential for the reliable design and performance assessment of high-enthalpy scramjet inlets.
- 1 October 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (10)
Effects of thermochemical nonequilibrium on shock interaction flow of V-shaped leading edge
This paper numerically investigates the effects of thermochemical nonequilibrium on the shock interactions and aerodynamic heating over the V-shaped leading edge. The numerical methodology is validated against experimental data from the JF22 hypervelocity wind tunnel. A comparative study is performed using the calorically perfect gas (CPG) model, as well as the one-temperature and two-temperature five-species chemical reaction models of air. The results demonstrate that thermochemical nonequilibrium significantly modifies the post-shock thermodynamic state through internal energy excitation and chemical reactions, thereby altering the shock-layer structure and detached-shock standoff distance. The different relaxation processes predicted by the one-temperature and two-temperature models lead to distinct shock interaction patterns, with the one-temperature model inducing a transition from Mach reflection to regular reflection from the same family. Compared with the CPG model, thermochemical nonequilibrium models predict enhanced wall heat flux due to the coupled effects of modified shock-layer states, conductive heat transfer, and species enthalpy transport. In particular, under the fully catalytic wall condition, atomic recombination releases chemical formation enthalpy, contributing to the species enthalpy diffusion term and the enhanced wall heat flux predicted by the thermochemical nonequilibrium models. These findings demonstrate that accurate thermochemical nonequilibrium modelling is essential for the reliable design and performance assessment of high-enthalpy scramjet inlets.
- 1 October 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (10)
Airfoil pitch control in the wake of a backward-facing step via spatiotemporal deep reinforcement learning
Closed-loop lift control for an airfoil in a strongly nonuniform inflow is challenging, especially when only partial surface-pressure observations are available. This study investigates pitch control of a NACA (National Advisory Committee for Aeronautics) 0012 airfoil descending through the wake of a backward-facing step. A spatiotemporal deep-reinforcement-learning-based control framework is developed, in which the controller adopts sparse surface pressure and airfoil kinematics as feedback. The proposed MSALSTM policy, trained with proximal policy optimization, combines multi-head self-attention for spatial pressure encoding with long short-term memory for temporal inference of hidden flow states. Among the learned architectures, MSALSTM provides the most stable lift tracking and the smallest pitch oscillation. It also outperforms the fixed-angle and gain-scheduled proportional–integral–derivative baselines under the same descent condition. When a drag penalty is added to the reward function, the learned policy further increases the mean lift-to-drag ratio by 46%. Further analysis reveals that the attention module adaptively redistributes sensor emphasis according to the evolving flow state, while the recurrent memory organizes pressure-history information into structured hidden-state trajectories. These results demonstrate that spatiotemporal learning can reconstruct effective flow-state information from sparse wall-pressure data for high-Reynolds-number aerodynamic control.
- 1 October 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (10)
Airfoil pitch control in the wake of a backward-facing step via spatiotemporal deep reinforcement learning
Closed-loop lift control for an airfoil in a strongly nonuniform inflow is challenging, especially when only partial surface-pressure observations are available. This study investigates pitch control of a NACA (National Advisory Committee for Aeronautics) 0012 airfoil descending through the wake of a backward-facing step. A spatiotemporal deep-reinforcement-learning-based control framework is developed, in which the controller adopts sparse surface pressure and airfoil kinematics as feedback. The proposed MSALSTM policy, trained with proximal policy optimization, combines multi-head self-attention for spatial pressure encoding with long short-term memory for temporal inference of hidden flow states. Among the learned architectures, MSALSTM provides the most stable lift tracking and the smallest pitch oscillation. It also outperforms the fixed-angle and gain-scheduled proportional–integral–derivative baselines under the same descent condition. When a drag penalty is added to the reward function, the learned policy further increases the mean lift-to-drag ratio by 46%. Further analysis reveals that the attention module adaptively redistributes sensor emphasis according to the evolving flow state, while the recurrent memory organizes pressure-history information into structured hidden-state trajectories. These results demonstrate that spatiotemporal learning can reconstruct effective flow-state information from sparse wall-pressure data for high-Reynolds-number aerodynamic control.
- Open Access
- 1 July 2026
- Published by Frontiers Media SA in Frontiers in Chemistry
- Vol. 14, 1869326
State-to-state approach in hypersonic thermochemical nonequilibrium flow simulations: a review
Thermochemical nonequilibrium flows play a critical role in shaping the aerothermal environment of hypersonic vehicles and therefore strongly influence the analysis and design of thermal protection systems (TPS). Traditional multi-temperature models often fail to accurately represent non-Boltzmann population distributions and detailed vibration-dissociation coupling, whereas state-to-state (StS) models provide a more physically rigorous description by directly resolving transitions between internal energy levels. This review summarizes recent advances in StS modeling for hypersonic thermochemical nonequilibrium flows, including elementary rate-coefficient determination, benchmark applications in canonical nonequilibrium flows, and current extension strategies based on reduced-order modeling, coarse graining, and machine-learning-assisted acceleration. The reviewed literature shows that the reliability of StS predictions depends strongly on the availability, accuracy, and consistency of elementary kinetic data, while their broader engineering use remains limited by the enormous number of internal states and elementary processes, severe equation stiffness, incomplete kinetic databases, and insufficient experimental data for robust validation. Overall, this review clarifies the recent progress, key bottlenecks, and promising future directions of StS modeling in hypersonic nonequilibrium flows.
- 1 July 2026
- Published by Frontiers Media SA in Frontiers in Chemistry
- Vol. 14, 1869326
Open AccessState-to-state approach in hypersonic thermochemical nonequilibrium flow simulations: a review
Thermochemical nonequilibrium flows play a critical role in shaping the aerothermal environment of hypersonic vehicles and therefore strongly influence the analysis and design of thermal protection systems (TPS). Traditional multi-temperature models often fail to accurately represent non-Boltzmann population distributions and detailed vibration-dissociation coupling, whereas state-to-state (StS) models provide a more physically rigorous description by directly resolving transitions between internal energy levels. This review summarizes recent advances in StS modeling for hypersonic thermochemical nonequilibrium flows, including elementary rate-coefficient determination, benchmark applications in canonical nonequilibrium flows, and current extension strategies based on reduced-order modeling, coarse graining, and machine-learning-assisted acceleration. The reviewed literature shows that the reliability of StS predictions depends strongly on the availability, accuracy, and consistency of elementary kinetic data, while their broader engineering use remains limited by the enormous number of internal states and elementary processes, severe equation stiffness, incomplete kinetic databases, and insufficient experimental data for robust validation. Overall, this review clarifies the recent progress, key bottlenecks, and promising future directions of StS modeling in hypersonic nonequilibrium flows.
- 1 April 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (4)
Reinforcement learning of robust active flow control strategy with surrogate-based hybrid environment
This article introduces a training framework for active flow control (AFC) based on deep reinforcement learning (DRL) called PyFAC, which employs the Soft Actor–Critic algorithm as the DRL method. To reduce computational resource consumption and improve training efficiency, a Long Short-Term Memory (LSTM) model is incorporated as a surrogate environment for computational fluid dynamics (CFD). Additionally, a free-stream turbulence (FST) model is introduced to validate the robustness of the trained AFC agent under disturbing incoming flow conditions. Under the condition without incoming flow disturbance, the agent gains 3.16% extra lift compared with the baseline, which is not stimulated by AFC, and 28.3% improvement of control performance compared with ordinary control strategies without the involvement of DRL. It is observed that an agent trained in a plain condition without FST struggles to perform effectively in an FST environment due to the disturbances from the incoming flow. Conversely, introducing FST perturbations into the training environment allows for the development of a robust agent. Subsequently, the LSTM model is integrated as the CFD surrogate. The LSTM demonstrates high accuracy, with discrepancies compared to CFD remaining below 10−3. Agents trained in the LSTM surrogate environment producing control strategies comparable to those of the model-free agents and achieving nearly identical control performances. A comparative analysis of training times for LSTM-based and model-free DRL approaches is also conducted, revealing a 60% improvement in training efficiency with the LSTM-based method.
- 1 April 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (4)
Reinforcement learning of robust active flow control strategy with surrogate-based hybrid environment
This article introduces a training framework for active flow control (AFC) based on deep reinforcement learning (DRL) called PyFAC, which employs the Soft Actor–Critic algorithm as the DRL method. To reduce computational resource consumption and improve training efficiency, a Long Short-Term Memory (LSTM) model is incorporated as a surrogate environment for computational fluid dynamics (CFD). Additionally, a free-stream turbulence (FST) model is introduced to validate the robustness of the trained AFC agent under disturbing incoming flow conditions. Under the condition without incoming flow disturbance, the agent gains 3.16% extra lift compared with the baseline, which is not stimulated by AFC, and 28.3% improvement of control performance compared with ordinary control strategies without the involvement of DRL. It is observed that an agent trained in a plain condition without FST struggles to perform effectively in an FST environment due to the disturbances from the incoming flow. Conversely, introducing FST perturbations into the training environment allows for the development of a robust agent. Subsequently, the LSTM model is integrated as the CFD surrogate. The LSTM demonstrates high accuracy, with discrepancies compared to CFD remaining below 10−3. Agents trained in the LSTM surrogate environment producing control strategies comparable to those of the model-free agents and achieving nearly identical control performances. A comparative analysis of training times for LSTM-based and model-free DRL approaches is also conducted, revealing a 60% improvement in training efficiency with the LSTM-based method.
- Open Access
- 1 January 2026
- Published by Tech Science Press in Fluid Dynamics & Materials Processing
- Vol. 22 (1) , 1-10
Hypersonic Flow over V-Shaped Leading Edges: A Review of Shock Interactions and Aerodynamic Loads
For hypersonic air-breathing vehicles, the V-shaped leading edges (VSLEs) of supersonic combustion ramjet (scramjet) inlets experience complex shock interactions and intense aerodynamic loads. This paper provides a comprehensive review of flow characteristics at the crotch of VSLEs, with particular focus on the transition of shock interaction types and the variation of wall heat flux under different freestream Mach numbers and geometric configurations. The mechanisms governing shock transition, unsteady oscillations, hysteresis, and three-dimensional effects in VSLE flows are first examined. Subsequently, thermal protection strategies aimed at mitigating extreme heating loads are reviewed, emphasizing their relevance to practical engineering applications. Special attention is given to recent studies addressing thermochemical nonequilibrium effects on VSLE shock interactions, and the limitations of current research are critically assessed. Finally, perspectives for future investigations into hypersonic VSLE shock interactions are outlined, highlighting opportunities for advancing design and thermal management strategies.
- 1 January 2026
- Published by Tech Science Press in Fluid Dynamics & Materials Processing
- Vol. 22 (1) , 1-10
Open AccessHypersonic Flow over V-Shaped Leading Edges: A Review of Shock Interactions and Aerodynamic Loads
For hypersonic air-breathing vehicles, the V-shaped leading edges (VSLEs) of supersonic combustion ramjet (scramjet) inlets experience complex shock interactions and intense aerodynamic loads. This paper provides a comprehensive review of flow characteristics at the crotch of VSLEs, with particular focus on the transition of shock interaction types and the variation of wall heat flux under different freestream Mach numbers and geometric configurations. The mechanisms governing shock transition, unsteady oscillations, hysteresis, and three-dimensional effects in VSLE flows are first examined. Subsequently, thermal protection strategies aimed at mitigating extreme heating loads are reviewed, emphasizing their relevance to practical engineering applications. Special attention is given to recent studies addressing thermochemical nonequilibrium effects on VSLE shock interactions, and the limitations of current research are critically assessed. Finally, perspectives for future investigations into hypersonic VSLE shock interactions are outlined, highlighting opportunities for advancing design and thermal management strategies.
- 1 November 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (11)
Impact of differentiating factors on the hemodynamics of umbrella-shaped and fusiform-shaped inferior vena cava filters: A computational fluid dynamics simulation study
This work applies computational fluid dynamics to carry out a systematic investigation of two representative inferior vena cava filters (IVCFs): the umbrella-shaped Denali and the fusiform-shaped OptEase IVCFs. The study emphasizes how blood rheology, inferior vena cava (IVC) diameter, and flow rate influence the hemodynamic environment surrounding these devices. The simulations indicate that although the Newtonian assumption preserves the global flow pattern, it largely exaggerates the downstream stagnation region (roughly twice that predicted by non-Newtonian models) and reduces the overall velocity magnitude (centerline velocity drops by about 30%). It also weakens the acceleration inside the filter and underestimates both wall shear stress (WSS) and flow resistance. Such discrepancies could mislead clinical evaluation and structural safety assessments. Among the five non-Newtonian rheological models examined, all yield similar flow field distributions, but clear variations appear in WSS and resistance. The simplified cross model predicts the largest values, whereas the Carreau model provides intermediate results, thus offering a more representative choice. A smaller IVC diameter markedly lengthens the post-filter stagnation zone (at 15 mm, nearly twice that at 25 mm) and substantially increases filter-surface WSS, which may enhance thrombus trapping and dissolution yet simultaneously elevate resistance and fracture risk. Higher inflow rates exert effects comparable to a reduced vessel diameter, intensifying resistance and promoting local disturbances. Overall, these findings highlight that clinical application of IVCFs may incorporate imaging data to select filter type and geometry tailored to patient-specific IVC size and flow rate, thereby achieving effective embolus interception while limiting complications.
- 1 November 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (11)
Impact of differentiating factors on the hemodynamics of umbrella-shaped and fusiform-shaped inferior vena cava filters: A computational fluid dynamics simulation study
This work applies computational fluid dynamics to carry out a systematic investigation of two representative inferior vena cava filters (IVCFs): the umbrella-shaped Denali and the fusiform-shaped OptEase IVCFs. The study emphasizes how blood rheology, inferior vena cava (IVC) diameter, and flow rate influence the hemodynamic environment surrounding these devices. The simulations indicate that although the Newtonian assumption preserves the global flow pattern, it largely exaggerates the downstream stagnation region (roughly twice that predicted by non-Newtonian models) and reduces the overall velocity magnitude (centerline velocity drops by about 30%). It also weakens the acceleration inside the filter and underestimates both wall shear stress (WSS) and flow resistance. Such discrepancies could mislead clinical evaluation and structural safety assessments. Among the five non-Newtonian rheological models examined, all yield similar flow field distributions, but clear variations appear in WSS and resistance. The simplified cross model predicts the largest values, whereas the Carreau model provides intermediate results, thus offering a more representative choice. A smaller IVC diameter markedly lengthens the post-filter stagnation zone (at 15 mm, nearly twice that at 25 mm) and substantially increases filter-surface WSS, which may enhance thrombus trapping and dissolution yet simultaneously elevate resistance and fracture risk. Higher inflow rates exert effects comparable to a reduced vessel diameter, intensifying resistance and promoting local disturbances. Overall, these findings highlight that clinical application of IVCFs may incorporate imaging data to select filter type and geometry tailored to patient-specific IVC size and flow rate, thereby achieving effective embolus interception while limiting complications.
- 1 August 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (8)
Effects of coiling index and vessel diameter on umbilical artery flow: A computational fluid dynamics study
The hemodynamics of the umbilical artery (UA) are critical to fetal development, with the umbilical coiling index (UCI) and vessel diameter identified as key geometric factors influencing blood flow patterns and related pregnancy outcomes. In this study, a three-dimensional computational fluid dynamics (CFD) model based on the Carreau non-Newtonian viscosity of blood is applied, and validated by particle image velocimetry experiments, to investigate the independent and combined effects of UCI and vessel diameter on UA hemodynamics. The results reveal that the UCI increase promotes the uniformity of velocity distribution, significantly enhances inlet pressure (up to twice that of a normal UA) and peak wall shear stress (WSS), with the extension of high Q-criterion region. It potentially induces endothelial stress and impairs placental perfusion, thus increasing the risk of fetal growth restriction (FGR). However, the UCI decrease can accelerate blood flow, also destabilizing flow and possibly disrupting placental function. The vessel diameter reduction has a pronounced impact that remarkably increases flow velocity, pressure, high Q-criterion region size, and WSS (up to fivefold), which may exacerbate flow irregularities and elevate the risk of FGR. Enlarged diameters reduce velocity and may raise blood viscosity, with limited effects on pressure and WSS, potentially increasing the risk of thrombosis. When UCI and diameter vary simultaneously, the vessel diameter predominantly determines hemodynamic behavior, while the UCI only has a secondary influence. The secondary flow structures remain spatially consistent, showing robustness to geometric variations. This study quantitatively reveals how UCI and vessel diameter affect UA flow, offering insights into placental dysfunction and adverse outcomes.
- 1 August 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (8)
Effects of coiling index and vessel diameter on umbilical artery flow: A computational fluid dynamics study
The hemodynamics of the umbilical artery (UA) are critical to fetal development, with the umbilical coiling index (UCI) and vessel diameter identified as key geometric factors influencing blood flow patterns and related pregnancy outcomes. In this study, a three-dimensional computational fluid dynamics (CFD) model based on the Carreau non-Newtonian viscosity of blood is applied, and validated by particle image velocimetry experiments, to investigate the independent and combined effects of UCI and vessel diameter on UA hemodynamics. The results reveal that the UCI increase promotes the uniformity of velocity distribution, significantly enhances inlet pressure (up to twice that of a normal UA) and peak wall shear stress (WSS), with the extension of high Q-criterion region. It potentially induces endothelial stress and impairs placental perfusion, thus increasing the risk of fetal growth restriction (FGR). However, the UCI decrease can accelerate blood flow, also destabilizing flow and possibly disrupting placental function. The vessel diameter reduction has a pronounced impact that remarkably increases flow velocity, pressure, high Q-criterion region size, and WSS (up to fivefold), which may exacerbate flow irregularities and elevate the risk of FGR. Enlarged diameters reduce velocity and may raise blood viscosity, with limited effects on pressure and WSS, potentially increasing the risk of thrombosis. When UCI and diameter vary simultaneously, the vessel diameter predominantly determines hemodynamic behavior, while the UCI only has a secondary influence. The secondary flow structures remain spatially consistent, showing robustness to geometric variations. This study quantitatively reveals how UCI and vessel diameter affect UA flow, offering insights into placental dysfunction and adverse outcomes.
- 1 August 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (8)
Fast prediction of three-dimensional wing inviscid flow fields from sparse data based on an autoencoder and a deep attention network
Compared to traditional computational fluid dynamics methods, which involve high computational costs to solve the governing flow equations, deep learning has emerged as a promising alternative for rapid flow field prediction. This paper introduces a novel deep learning method for quickly reconstructing three-dimensional (3D) wing inviscid flow fields under sparse data conditions. Initially, an autoencoder model is utilized to extract the geometric features of the 3D wing. This model efficiently compresses the point cloud data of the wing into a six-dimensional latent vector, facilitating a compact representation of complex geometric information. Next, a residual-attention multilayer perceptron (RAMLP) model, which incorporates attention mechanisms and residual connections, is introduced to predict the flow field variables, including velocity components in the x, y, and z directions, as well as the pressure coefficient. The model takes geometric feature parameters, Mach number, angle of attack, and flow field coordinates as inputs, allowing it to generalize effectively across various incoming flow conditions. Finally, a comparative analysis is conducted to evaluate the performance of RAMLP against two commonly used aerodynamic surrogate models: multilayer perceptron (MLP) and multi-head perceptron (MHP), in the context of 3D wing inviscid flow field prediction. Although RAMLP has a more complex computational structure, it maintains efficient prediction times, processing a single condition in under one second. Furthermore, RAMLP exhibits superior prediction accuracy under sparse data conditions compared to both MLP and MHP models.
- 1 August 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (8)
Fast prediction of three-dimensional wing inviscid flow fields from sparse data based on an autoencoder and a deep attention network
Compared to traditional computational fluid dynamics methods, which involve high computational costs to solve the governing flow equations, deep learning has emerged as a promising alternative for rapid flow field prediction. This paper introduces a novel deep learning method for quickly reconstructing three-dimensional (3D) wing inviscid flow fields under sparse data conditions. Initially, an autoencoder model is utilized to extract the geometric features of the 3D wing. This model efficiently compresses the point cloud data of the wing into a six-dimensional latent vector, facilitating a compact representation of complex geometric information. Next, a residual-attention multilayer perceptron (RAMLP) model, which incorporates attention mechanisms and residual connections, is introduced to predict the flow field variables, including velocity components in the x, y, and z directions, as well as the pressure coefficient. The model takes geometric feature parameters, Mach number, angle of attack, and flow field coordinates as inputs, allowing it to generalize effectively across various incoming flow conditions. Finally, a comparative analysis is conducted to evaluate the performance of RAMLP against two commonly used aerodynamic surrogate models: multilayer perceptron (MLP) and multi-head perceptron (MHP), in the context of 3D wing inviscid flow field prediction. Although RAMLP has a more complex computational structure, it maintains efficient prediction times, processing a single condition in under one second. Furthermore, RAMLP exhibits superior prediction accuracy under sparse data conditions compared to both MLP and MHP models.