Article
Computer Science, Artificial Intelligence
Cihat Duru, Hande Alemdar, Ozgur Ugras Baran
Summary: In this study, a new approach based on neural networks is proposed for estimating the pressure field around an airfoil, achieving a high accuracy of 88% for unseen airfoil shapes and significant speed-up compared to CFD simulations. The model allows for performance analysis of different airfoils and the effect of shock, showing promise in accurately capturing flow patterns.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Mechanics
Justin Sirignano, Jonathan F. MacArt
Summary: Near-wall flow simulation is a challenging task in aerodynamics modelling, with traditional methods often yielding inaccurate results. However, a deep learning closure model for large-eddy simulation (LES) has been developed, utilizing untrained neural networks and adjoint partial differential equation optimization. The DL-LES models exceed the performance of standard LES models and achieve accurate predictions on a relatively coarse mesh.
JOURNAL OF FLUID MECHANICS
(2023)
Article
Computer Science, Interdisciplinary Applications
Rafael Diez Sanhueza, Stephan H. H. J. Smit, Jurriaan W. R. Peeters, Rene Pecnik
Summary: This paper presents a machine learning methodology to improve the predictions of traditional RANS turbulence models in channel flows subject to strong variations in their thermophysical properties. The methodology utilizes efficient optimization routines, symbolic algebra solvers, and a neural network architecture with logarithmic and hyperbolic tangent neurons. A weighted relaxation factor methodology and L2 regularization are introduced to correct the machine learning predictions and mitigate over-fitting. The results demonstrate the effectiveness of the developed machine learning methodology in improving RANS turbulence models without prior modeling assumptions.
COMPUTERS & FLUIDS
(2023)
Article
Thermodynamics
Xiao He, Jianheng Tan, Georgios Rigas, Mehdi Vahdati
Summary: This paper presents two methods to improve the explainability of machine learning models in the context of turbulence model development. The methods include reducing model complexity and explaining the correlation between inputs and outputs. The study focuses on using machine learning to improve the prediction accuracy of a specific turbulence model in transonic bump flows. The results show that these methods can provide valuable insights into the causal links between input features and the model outputs.
INTERNATIONAL JOURNAL OF HEAT AND FLUID FLOW
(2022)
Article
Thermodynamics
Chongyang Yan, Haoran Li, Yufei Zhang, Haixin Chen
Summary: This paper implements a data-driven Reynolds-averaged turbulence modeling approach using field inversion and machine learning to modify the Spalart-Allmaras model. The results show that the augmented model can reproduce the quantity of interest with relatively high accuracy and has a certain extent of generalization ability in similar flow conditions.
INTERNATIONAL JOURNAL OF HEAT AND FLUID FLOW
(2022)
Article
Mechanics
Zhicheng Wang, Dixia Fan, Xiaomo Jiang, Michael S. Triantafyllou, George Em Karniadakis
Summary: This paper demonstrates how to accelerate the computationally taxing process of deep reinforcement learning for active control of bluff body flows at high Reynolds number using transfer learning. The results show that transfer learning greatly reduces training episodes and improves stability compared to training from scratch. The wake flow at high Reynolds number is analyzed, revealing an asymmetry in the hydrodynamic forces on the two rotating control cylinders for the first time.
JOURNAL OF FLUID MECHANICS
(2023)
Article
Engineering, Aerospace
Chutian Wu, Shizhao Wang, Xin-Lei Zhang, Guowei He
Summary: Model-consistent training is popular for data-driven turbulence modeling because it improves model generalizability and reduces data requirements. However, it lacks interpretability for the causal relationship between model inputs and outputs.
AEROSPACE SCIENCE AND TECHNOLOGY
(2023)
Article
Engineering, Aerospace
XiangLin Shan, YiLang Liu, WenBo Cao, XuXiang Sun, WeiWei Zhang
Summary: This paper presents a turbulence modeling approach for separated flows using data assimilation technique and deep neural network (DNN). By optimizing the parameters of the Spalart-Allmaras (SA) turbulence model and embedding the DNN model within a RANS solver, the accuracy of simulations for turbulent attached and separated flows is significantly improved. This approach does not rely on traditional turbulence models during the simulation process and achieves a mean relative error reduction of over 57% for lift coefficient calculations.
Article
Engineering, Aerospace
Kuijun Zuo, Shuhui Bu, Weiwei Zhang, Jiawei Hu, Zhengyin Ye, Xianxu Yuan
Summary: In this study, a data-driven method based on convolutional neural network and multi-head perceptron is used to predict the flow field around airfoils. The experimental results show that the multi-head perceptron can achieve better prediction results for sparse flow field compared to the multi-layer perceptron.
AEROSPACE SCIENCE AND TECHNOLOGY
(2022)
Article
Thermodynamics
Weishuo Liu, Jian Fang, Stefano Rolfo, Charles Moulinec, David R. Emerson
Summary: Machine-learning techniques offer a new perspective for constructing turbulence models for RANS simulations. An iterative ML-RANS computational framework is proposed, ensuring built-in reproducibility. The discussion around closure term and suitable target variables, as well as the study of multi-valued problem for establishing a proper regression system, contribute to the effectiveness of the ML model.
INTERNATIONAL JOURNAL OF HEAT AND FLUID FLOW
(2021)
Article
Computer Science, Interdisciplinary Applications
Didier Lucor, Atul Agrawal, Anne Sergent
Summary: PINNs show promise as candidates for full fluid flow PDE modeling, but challenges remain in sustaining turbulence. By minimizing composite loss functions, surrogate modeling using PINNs for turbulent natural convection flows can reduce the need for large training datasets.
JOURNAL OF COMPUTATIONAL PHYSICS
(2022)
Article
Thermodynamics
Rafael Diez Sanhueza, Ido Akkerman, Jurriaan W. R. Peeters
Summary: This study investigates the use of convolutional neural networks to predict the skin friction values and Nusselt numbers of rough surfaces. The results show that machine learning can accurately predict these values, outperforming existing correlations in the literature. The stability and sensitivity of the deep learning results are also examined.
INTERNATIONAL JOURNAL OF HEAT AND FLUID FLOW
(2023)
Article
Engineering, Aerospace
Yao Li, Jin-ping Li, Fan-zhi Zeng, Mao Sun, Chao Yan
Summary: In this paper, a Bayesian uncertainty quantification analysis of turbulence model parameters is conducted to improve the shear stress transport model's performance for transonic flow simulations. The sensitivity analysis shows that the pressure coefficients are mainly influenced by four parameters: a1, kappa, /3*, and /31. The posterior uncertainty analysis reveals opposite trends for the predicted shock wave positions in the examples of RAE2822 and ONERA M6.
AEROSPACE SCIENCE AND TECHNOLOGY
(2023)
Article
Meteorology & Atmospheric Sciences
Yu Cheng, Marco G. Giometto, Pit Kauffmann, Ling Lin, Chen Cao, Cody Zupnick, Harold Li, Qi Li, Yu Huang, Ryan Abernathey, Pierre Gentine
Summary: In large-eddy simulations, subgrid-scale processes are parameterized as a function of filtered grid-scale variables. This paper applies supervised deep neural networks (DNNs) to learn subgrid stresses and achieves higher correlation compared to traditional models, with applicability to different resolutions and stability conditions.
JOURNAL OF ADVANCES IN MODELING EARTH SYSTEMS
(2022)
Article
Engineering, Aerospace
Yanjun Chen, Shengye Wang, Wei Liu
Summary: This paper builds upon innovative ideas in data-driven turbulence modeling and reconstructs two models for turbulence transition prediction. The results demonstrate improved accuracy and generalization abilities compared to previous models, highlighting the potential of machine learning as a supplementary approach in turbulence transition modeling.
Article
Computer Science, Interdisciplinary Applications
Jiaqing Kou, Saumitra Joshi, Aurelio Hurtado-de-Mendoza, Kunal Puri, Charles Hirsch, Esteban Ferrer
Summary: In this work, the numerical advantages of the high-order Flux Reconstruction (FR) method and the simplicity of the mesh generation of the Immersed Boundary Method (IBM) are combined for steady and unsteady problems over moving geometries using the volume penalization (penalty-IBM) method. The efficiency of the approach in handling moving geometries is demonstrated through various numerical test cases, showcasing the potential of this method in industrial design processes.
JOURNAL OF COMPUTATIONAL PHYSICS
(2022)
Article
Mechanics
Yilang Liu, Wenbo Cao, Weiwei Zhang, Zhenhua Xia
Summary: This research investigates the effects of different coupling modes on the convergence and stability between RANS equations and turbulence models in machine learning. The results show that the frozen coupling mode used commonly in machine learning turbulence models may lead to divergence and instability, while the mutual coupling mode can maintain good convergence and stability.
Article
Engineering, Mechanical
Linyang Zhu, Xuxiang Sun, Yilang Liu, Weiwei Zhang
Summary: This study applies artificial intelligence techniques and data-driven machine learning methods to model turbulence in transonic wing flows. By constructing a fully connected deep neural network model, the researchers achieve successful modeling results and demonstrate good generalization capabilities in test cases. This research is significant for improving turbulence modeling techniques.
ACTA MECHANICA SINICA
(2022)
Article
Engineering, Aerospace
Xu Wang, Jiaqing Kou, Weiwei Zhang, Zhitao Liu
Summary: This paper proposes a machine learning framework based on multifidelity methods to improve the accuracy and efficiency of unsteady aerodynamic prediction of aircraft at high angles of attack. Wind-tunnel tests are conducted to verify the prediction accuracy of the method in various parameter ranges, and a comparison is made with a method using high-fidelity data only.
Article
Engineering, Aerospace
Luo Fuqing, Gao Chuanqiang, Lyu Zhen, Zhang Weiwei, Xu Qiannan
Summary: This study establishes an unsteady aerodynamic model using the system identification method, and couples a tuned mass damper (TMD) with the aeroelastic system to construct a flutter suppression model. By analyzing the parameters of the TMD, it is found that the participation of TMD changes the flutter instability mode of the original system. The proposed reduced-order model (ROM) provides a fast stability analysis method for flutter suppression with TMD, as well as guidance for TMD design and optimization.
PROCEEDINGS OF THE INSTITUTION OF MECHANICAL ENGINEERS PART G-JOURNAL OF AEROSPACE ENGINEERING
(2023)
Article
Engineering, Mechanical
Kai Ren, Chuanqiang Gao, Fangqi Zhou, Weiwei Zhang
Summary: The smart skin system effectively eliminates the fluctuating load of transonic buffet by measuring the airfoil lift coefficient and using data-driven, model-free adaptive control, without affecting the aerodynamic performance in the non-buffeting state.
Article
Computer Science, Artificial Intelligence
Jiaqing Kou, Laura Botero-Bolivar, Roman Ballano, Oscar Marino, Leandro de Santana, Eusebio Valero, Esteban Ferrer
Summary: This study presents a framework for optimizing the airfoil shape of wind turbine blades to reduce trailing edge noise. The framework uses Amiet's theory, the TNO-Blake model, and XFOIL simulations to evaluate noise and boundary layer parameters. Particle swarm optimization is employed to find the optimized airfoil configuration, while traditional shape optimization techniques are compared to machine learning methods using a variational autoencoder. The autoencoder-based optimized airfoil reduces overall sound pressure level by 3% (1.75 dBA) and improves aerodynamic properties compared to the baseline NACA0012 airfoil.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Mechanics
Kuijun Zuo, Zhengyin Ye, Weiwei Zhang, Xianxu Yuan, Linyang Zhu
Summary: In this article, a novel data-driven deep attention network (DAN) is proposed for reconstructing incompressible steady flow fields around airfoils. The traditional method of obtaining aerodynamic parameters through solving Navier-Stokes equations is time-consuming. The DAN utilizes a transformer encoder to extract geometric representations of the airfoils and then predicts the flow fields using a multilayer perceptron. Experimental results demonstrate that the proposed DAN improves model interpretability and achieves good prediction accuracy and generalization capability for different airfoils and flow-field states.
Article
Mechanics
Wenbo Cao, Yilang Liu, Xianglin Shan, Chuanqiang Gao, Weiwei Zhang
Summary: Iterative steady-state solvers are widely used in computational fluid dynamics. In this study, an online dimension reduction optimization method is proposed to enhance the convergence of the traditional iterative method for obtaining steady-state solutions of unstable problems. The method combines proper orthogonal decomposition (POD) and optimization to iteratively improve the solution until convergence. The proposed method demonstrates high efficiency, robustness, and easy implementation in various iterative solvers.
Article
Mechanics
Qiao Zhang, Chuanqiang Gao, Fangqi Zhou, Dangguo Yang, Weiwei Zhang
Summary: This study uses the Delayed-Detached Eddy Simulation and Discrete Frequency Response method to analyze the flow field and sound propagation law in different transonic buffeting states. It is found that low-frequency and small-amplitude shock oscillation in light buffeting states do not trigger large separated flow, while deep buffeting states produce high-frequency and large-amplitude shock oscillations resulting in large separated bubbles. Collisions between upstream traveling waves and shock wave oscillations increase the frequency and sound pressure levels of the shock waves. The main sound sources in this process are shock oscillations and the von Karman mode.
Article
Mechanics
Zhen Lyu, H. D. Lim, Weiwei Zhang
Summary: This paper presents a peculiar nodal-shaped oscillation in vortex-induced vibration (VIV), which is characterized by a cycle of divergence and decay. Wind tunnel tests and flow analysis reveal that the dynamics of this oscillation are governed by the coupling and competition between the wake mode and the structure mode.
Article
Mechanics
Chang-Zhe Chen, Zao-Jian Zou, Lu Zou, Ming Zou, Jia-Qing Kou
Summary: A novel reduced-order model based on higher order dynamic mode decomposition (HODMD) is proposed for the time series prediction of ship course-keeping motion in waves. The approach is validated using data from course-keeping tests of an ONR tumblehome ship model. The HODMD approach shows better prediction accuracy compared to standard DMD, and the effects of tunable parameters on accuracy and computational times are analyzed.
Article
Mechanics
Wengang Chen, Jiaqing Kou, Wenkai Yang
Summary: In this letter, the authors propose a method to accelerate the solution of unsteady adjoint equations using dynamic mode decomposition (DMD). By approximating the pseudo-time marching as an infinite-dimensional linear dynamical system and analyzing the adjoint vectors using DMD, the efficiency of solving unsteady adjoint equations is significantly improved.
THEORETICAL AND APPLIED MECHANICS LETTERS
(2023)
Article
Engineering, Mechanical
Linyang Zhu, Weiwei Zhang, Guohua Tu
Summary: This paper proposes a new posterior feature selection method based on a validation dataset, which is efficient and universal for complex systems, including turbulence modeling. The method ranks features according to model performance on the validation dataset and generates feature subsets based on feature importance. Experimental results show significant improvement in the model's generalization ability after feature selection.
ADVANCES IN AERODYNAMICS
(2022)