Advancing buffet onset prediction: a deep learning approach with enhanced interpretability for aerodynamic engineering DOI Creative Commons
Jing Wang, Liu We, Hairun Xie

et al.

Complex & Intelligent Systems, Journal Year: 2024, Volume and Issue: 11(1)

Published: Nov. 8, 2024

The interaction between the shock wave and boundary layer of transonic wings can trigger periodic self-excited oscillations, resulting in buffet. Buffet severely restricts flight envelope civil aircraft is directly related to their aerodynamic performance safety. Developing efficient reliable techniques for buffet onset prediction crucial advancement aircraft. In this study, utilizing a comprehensive database supercritical airfoils generated through numerical simulations, convolutional neural network (CNN) model firstly developed perform classification based on flow fields. After that, employing explainable machine learning techniques, including Gradient-weighted Class Activation Mapping (Grad-CAM), random forest algorithms, statistical analysis, research investigates correlations supervised CNN features key physical characteristics with separation region, wave, leading edge suction peak, post-shock loading. Finally, metric established good generalization accuracy, providing valuable guidance engineering design

Language: Английский

Fast simulation of airfoil flow field via deep neural network DOI
Kuijun Zuo, Zhengyin Ye, Shuhui Bu

et al.

Aerospace Science and Technology, Journal Year: 2024, Volume and Issue: 150, P. 109207 - 109207

Published: May 10, 2024

Language: Английский

Citations

12
Simulation of supersonic axisymmetric base flow with a data-driven turbulence model DOI
Seoyeon Heo, Yeji Yun, Minjae Jeong

et al.

Aerospace Science and Technology, Journal Year: 2024, Volume and Issue: 147, P. 109014 - 109014

Published: Feb. 29, 2024

Language: Английский

Citations

5
A tensor basis neural network-based turbulence model for transonic axial compressor flows DOI
Ziqi Ji, Gang Du

Aerospace Science and Technology, Journal Year: 2024, Volume and Issue: 149, P. 109155 - 109155

Published: April 23, 2024

Language: Английский

Citations

5
Additive-feature-attribution methods: A review on explainable artificial intelligence for fluid dynamics and heat transfer DOI Creative Commons
A. Cremades, Sergio Hoyas, Ricardo Vinuesa

et al.

International Journal of Heat and Fluid Flow, Journal Year: 2024, Volume and Issue: 112, P. 109662 - 109662

Published: Dec. 9, 2024

Language: Английский

Citations

4
Augmentation of predictive turbulence modeling applied to low pressure turbines using adaptive field inversion DOI

Lifeng Gou,

Jian Ye, Zhengping Zou

et al.

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(1)

Published: Jan. 1, 2025

Based on the framework of field inversion and machine learning, an adaptive modification for Reynolds-Averaged Navier–Stokes-based turbulence models is proposed simulation low-pressure turbine cascades involving flow separation. This method adjusts results by modifying source terms correspondingly at different spatial locations. First, specific regions are obtained Gaussian mixture adaptively according to baseline distribution correction term inferred ensemble-based with effective utilization high-fidelity data. Then a corrective model form quantities calculated established Gradient Boosting Decision Tree used T106 cascade cases. The demonstrate that modified model, reduced deficiency predicting load can be obtained. also predict more accurate separation onset damping eddy viscosity in separated region case out training set. With added solely region, computational cost compared full-field inversion, possibly applied simulating three-dimensional considering rotation effects.

Language: Английский

Citations

0
Interpreting tensor basis neural networks with symbolic transcendental Reynolds stress models for transonic axial compressor flows DOI
Ziqi Ji, He Lu, Penghao Duan

et al.

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(2)

Published: Feb. 1, 2025

Transonic axial compressor flows exhibit complex turbulence structures that pose significant challenges for traditional models. In recent years, neural network-based models have demonstrated promising results in simulating these intricate flows. However, often lack interpretability, a crucial aspect of understanding the underlying physical mechanisms. Symbolic regression, capable training highly interpretable models, offers potential solution to elucidate mechanisms underpinning this study, we employ evolutionary symbolic regression interpret tensor basis networks (TBNNs) and develop explicit transcendental Reynolds stress (ETRSM) transonic Our are trained on inputs outputs pre-trained TBNN. We introduce method independently predicts coefficients each basis, significantly reducing computational costs enhancing rationality prediction process. six models: three algebraic. Through rigorous fluid dynamics (CFD) simulations, demonstrate an exceptional ability TBNN, while algebraic show limited success. The ETRSM, characterized by high interpretability transferability, effectively interprets TBNN achieves comparable accuracy TBNN-based compressors. These underscore industry-level CFD problems highlight importance incorporating additional features such Furthermore, separates individual coefficients, costs.

Language: Английский

Citations

0
An artificial neural network-based quadratic constitutive Reynolds stress model for separated turbulent flows using data-augmented field inversion and machine learning DOI
Tianchi Gong, Yan Wang

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(3)

Published: March 1, 2025

Reynolds-averaged turbulence models have become one of the most important and popular techniques for practical engineering applications in aeronautics astronautics. However, poor performance prediction flow separations restricts its application ranges due to traditional linearity equilibrium hypotheses that constitute equation Reynolds stress modeling. In this study, an artificial neural network-based quadratic constitutive (ANN-QCR) model is proposed simulating turbulent flows with by using field inversion machine learning technique (FIML) high-fidelity experimental data. particular, decomposed into linear non-linear parts, respectively. The former evaluated Spalart–Allmaras a correction factor imposed on production term account non-equilibrium effect, while latter self-calibrated factor. These factors are predicted network (ANN) depending local features. unified framework FIML updates weights ANN-QCR directly gradient-based discrete adjoint method, ensuring consistency between training. data-augmented well validated through several separated induced adverse pressure gradients, shock wave boundary interfaces, higher angles attack, numbers (Re). With optimization target at lift coefficients, established also improves predictive other quantities, such as drag coefficients distributions. addition, captures development separation bubbles better increase angle attack. Benefiting from compatibility convergence forward simulation, generalization capability present successfully various numerical simulations problems across wide range attack good accuracy.

Language: Английский

Citations

0
Data augmented prediction of Reynolds stresses for flows around an axisymmetric body of revolution DOI
Yi Liu, Shizhao Wang, Xinlei Zhang

et al.

Ocean Engineering, Journal Year: 2024, Volume and Issue: 296, P. 116717 - 116717

Published: Jan. 18, 2024

Language: Английский

Citations

3
Ensemble variational method with adaptive covariance inflation for learning neural network-based turbulence models DOI Open Access
Qingyong Luo, Xinlei Zhang, Guowei He

et al.

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

Published: March 1, 2024

This work introduces an ensemble variational method with adaptive covariance inflation for learning nonlinear eddy viscosity turbulence models where the Reynolds stress anisotropy is represented tensor-basis neural networks. The ensemble-based has emerged as important alternative to data-driven modeling due its merit of non-derivativeness. However, training accuracy can be affected by linearization assumption and sample collapse issue. Given these difficulties, we introduce hybrid method, which inherits merits in non-derivativeness analysis. Moreover, a scheme proposed based on convergence states alleviate detrimental effects collapse. capability model tested flows square duct, over periodic hills, around S809 airfoil, increasing complexity data from direct observation sparse indirect observation. Our results show that learn relatively accurate network-based scenarios small size variances, compared Kalman method. It highlights superiority practical applications, since sizes reduce computational costs, variance ensure robustness avoiding nonphysical samples stresses.

Language: Английский

Citations

2
Supersonic combustion flow field reconstruction based on multi-view domain adaptation generative network in scramjet combustor DOI
Mingming Guo,

Erda Chen,

Ye Tian

et al.

Engineering Applications of Artificial Intelligence, Journal Year: 2024, Volume and Issue: 136, P. 108981 - 108981

Published: July 16, 2024

Language: Английский

Citations

2