Toward aerodynamic surrogate modeling based on β-variational autoencoders DOI

Víctor Francés-Belda,

Alberto Solera-Rico, Javier Nieto-Centenero

et al.

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(11)

Published: Nov. 1, 2024

Surrogate models that combine dimensionality reduction and regression techniques are essential to reduce the need for costly high-fidelity computational fluid dynamics data. New approaches using β-variational autoencoder (β-VAE) architectures have shown promise in obtaining high-quality low-dimensional representations of high-dimensional flow data while enabling physical interpretation their latent spaces. We propose a surrogate model based on space predict pressure distributions transonic wing given flight conditions: Mach number angle attack. The β-VAE model, enhanced with principal component analysis (PCA), maps space, showing direct correlation conditions. Regularization through β requires careful tuning improve overall performance, PCA preprocessing helps construct an effective improving training performance. Gaussian process is used variables from conditions, robust behavior independent β, decoder reconstructs field This pipeline provides insight into unexplored Furthermore, fine-tuning further refines reducing dependence enhancing accuracy. Structured significant improvements collectively create highly accurate efficient model. Our methodology demonstrates effectiveness β-VAEs aerodynamic modeling, offering rapid, cost-effective, reliable alternative prediction.

Language: Английский

Prediction of compressor blade cascade flow field based on Fourier neural operator DOI

Lixiang Jiang,

Xinlong Feng, Quanyong Xu

et al.

Aerospace Science and Technology, Journal Year: 2025, Volume and Issue: unknown, P. 110208 - 110208

Published: April 1, 2025

Language: Английский

Citations

0
Convolution-enhanced transformer architecture for inverse design of airfoils DOI
Pablo Figueroa, Semih Ölçmen

International Journal for Computational Methods in Engineering Science and Mechanics, Journal Year: 2025, Volume and Issue: unknown, P. 1 - 11

Published: May 9, 2025

Language: Английский

Citations

0
Learning dense gas-solids flows with physics-encoded neural network model DOI
Xiaolin Guo, Chenshu Hu,

Yuyang Dai

et al.

Chemical Engineering Journal, Journal Year: 2024, Volume and Issue: 485, P. 150072 - 150072

Published: Feb. 28, 2024

Language: Английский

Citations

3
A residual graph convolutional network for setting initial flow field in computational fluid dynamics simulations DOI Open Access
Xiaoyuan Zhang, Guopeng Sun, Peng Zhang

et al.

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(3)

Published: March 1, 2024

The computational cost of fluid dynamics (CFD) simulation is relatively high due to its complexity. To reduce the computing time required by CFD, researchers have proposed various methods, including efficient advancement correction methods for discrete control equations, multigrid reasonable initial field setting and parallel methods. Among these method can provide significant performance improvements, but there little work on it. Existing CFD industrial software typically uses inflow conditions flow or applies empirical which cause instability in calculation process make convergence difficult. With rapid development deep learning, are increasingly attempting replace simulations with neural networks achieved improvements. However, still face some challenges. First, they only predict regular grids. They cannot directly predictions irregular grids such as multi-block unstructured grids, so final be obtained through interpolation similar Second, although been claimed accuracy, a gap yet applied real scenarios. address issues, we propose Residual Graph Convolutional Network Initial Flow Field Setting (RGCN-IFS) simulations. This converts grid into graph structure an improved network field. In this way, any type grid. More importantly, does not simulations, it rather serves auxiliary role, providing appropriate fields calculations, improving efficiency while ensuring bridging accuracy between intelligent surrogate models

Language: Английский

Citations

3
A variable fidelity approach for predicting aerodynamic wall quantities of hypersonic vehicles using the ConvNeXt Encoder-Decoder framework DOI
Yuxin Yang, Shaobo Yao, Youtao Xue

et al.

Aerospace Science and Technology, Journal Year: 2024, Volume and Issue: unknown, P. 109605 - 109605

Published: Sept. 1, 2024

Language: Английский

Citations

3
Koopman neural operator approach to fast flow prediction of airfoil transonic buffet DOI
Deying Meng, Yiding Zhu, Jianchun Wang

et al.

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(7)

Published: July 1, 2024

Transonic buffet on airfoil is of great importance in the aerodynamic characteristics aircraft. In present work, a modified Koopman neural operator (KNO) applied to predict flow fields during transonic process OAT15A [ONERA (National Office for Aerospace Studies and Research) Aerospatiale Transport aircraft 15 Airfoil] airfoil. with different angles attack simulated by Reynolds averaged numerical simulation Menter's k−ω shear stress transport (SST) model at number Re=3×106. A prediction directly constructed between several previous time nodes that future node KNO. The predictions single sample multi samples are performed demonstrate accuracy efficiency sequence based iterative strategy achieved process. results indicate KNO can achieve fast accurate physical quantities buffet. Compared other deep learning models including Unet Fourier operator, has more advanced capability predicting higher less hardware requirements.

Language: Английский

Citations

1
Uncertainty involved drag divergence characteristic predicting method based on VAE DOI
Wei Liu, Hairun Xie, Jing Wang

et al.

Journal of Membrane Computing, Journal Year: 2024, Volume and Issue: 6(2), P. 53 - 66

Published: March 26, 2024

Language: Английский

Citations

0
LKFlowNet: A deep neural network based on large kernel convolution for fast and accurate nonlinear fluid-changing prediction DOI

Yan Liu,

Qingyang Zhang, Xinhai Chen

et al.

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(9)

Published: Sept. 1, 2024

The rapid development of artificial intelligence has promoted the emergence new flow field prediction methods. These methods address challenges posed by nonlinear problems and significantly reduce computational time cost compared to traditional numerical simulations. However, they often struggle capture dynamic sparse characteristics effectively. To bridge this gap, we introduce LKFlowNet, a large kernel convolutional neural network specifically designed for complex fields in fluid dynamics systems. LKFlowNet adopts multi-branch convolution computing architecture, which can skillfully handle changes. Drawing inspiration from dilated mechanism, developed RepDWConv block, re-parameterized depthwise that extends kernel's coverage. This enhancement improves model's ability long-range dependencies structural features dynamics. Additionally, customized physical loss function ensures accuracy consistency reconstruction. Comparative studies reveal outperforms existing architectures, providing more accurate physically consistent predictions variations such as velocity pressure fields. model demonstrates strong versatility scalability, accurately predicting various geometric configurations without modifying architecture. capability positions promising direction research, potentially revolutionizing combining high efficiency accuracy. Our results suggest could become an indispensable tool intelligent prediction, reshaping analysis processing

Language: Английский

Citations

0
CycleMLP++: An efficient and flexible modeling framework for subsonic airfoils DOI
Kuijun Zuo, Zhengyin Ye, Linyang Zhu

et al.

Expert Systems with Applications, Journal Year: 2024, Volume and Issue: unknown, P. 125455 - 125455

Published: Sept. 1, 2024

Language: Английский

Citations

0
Position query-guided cross-modal flow field prediction model of a transonic compressor cascade DOI
Liyue Wang, Haochen Zhang,

Xinyue Lan

et al.

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(10)

Published: Oct. 1, 2024

The gradient of flow parameters in a transonic compressor cascade field varies significantly, especially the region shock waves, which causes significant challenge to its high-precision prediction. In this study, position query-guided cross-modal prediction model (PGCM) is proposed effectively predict parameter distribution cascade. PGCM utilizes self-attention mechanism for global and deep geometric feature extraction configurations, contributes an in-depth understanding spatial relationships between coordinate points within field, accurately capturing analyzing structural complexity flow. addition, integrates cross-attention that establishes correlations different input sequences, enhances performance querying interpreting at specific coordinates. models are developed distributions geometries Mach numbers 0.78 0.93, respectively. validation results indicate performs significantly better than existing convolutional neural network vision transformer, pressure coefficient Cp distribution. adaptable variation conditions geometrical configurations efficiently accurate predicting This paper demonstrates promising potential conducting multi-modal information fusion enhance capability

Language: Английский

Citations

0