arXiv subjects

Shufan Zou

Publications and source records attributed to Shufan Zou.

7 recordsLinked to original sources

arXiv2024

Finite-difference-informed graph network for solving steady-state incompressible flows on block-structured grids

Advances in deep learning have enabled physics-informed neural networks to solve partial differential equations. Numerical differentiation using the finite-difference (FD) method is efficient in physics-constrained designs, even in parameterized settings. In traditional computational fluid dynamics(CFD), body-fitted block-structured grids are often employed for complex flow cases when obtaining FD solutions. However, convolution operators in convolutional neural networks for FD are typically limited to single-block grids. To address this issue, \blueText{graphs and graph networks are used} to learn flow representations across multi-block-structured grids. \blueText{A graph convolution-based FD method (GC-FDM) is proposed} to train graph networks in a label-free physics-constrained manner, enabling differentiable FD operations on unstructured graph outputs. To demonstrate model performance from single- to multi-block-structured grids, \blueText{the parameterized steady incompressible Navier-Stokes equations are solved} for a lid-driven cavity flow and the flows around single and double circular cylinder configurations. When compared to a CFD solver under various boundary conditions, the proposed method achieves a relative error in velocity field predictions on the order of $10^{-3}$. Furthermore, the proposed method reduces training costs by approximately 20\% compared to a physics-informed neural network. \blueText{To} further verify the effectiveness of GC-FDM in multi-block processing, \blueText{a 30P30N airfoil geometry is considered} and the \blueText{predicted} results are reasonable compared with those given by CFD. \blueText{Finally, the applicability of GC-FDM to three-dimensional (3D) case is tested using a 3D cavity geometry.

cs.LG↗

arXiv2024

A fully differentiable GNN-based PDE Solver: With Applications to Poisson and Navier-Stokes Equations

In this study, we present a novel computational framework that integrates the finite volume method with graph neural networks to address the challenges in Physics-Informed Neural Networks(PINNs). Our approach leverages the flexibility of graph neural networks to adapt to various types of two-dimensional unstructured grids, enhancing the model's applicability across different physical equations and boundary conditions. The core innovation lies in the development of an unsupervised training algorithm that utilizes GPU parallel computing to implement a fully differentiable finite volume method discretization process. This method includes differentiable integral and gradient reconstruction algorithms, enabling the model to directly solve partial-differential equations(PDEs) during training without the need for pre-computed data. Our results demonstrate the model's superior mesh generalization and its capability to handle multiple boundary conditions simultaneously, significantly boosting its generalization capabilities. The proposed method not only shows potential for extensive applications in CFD but also establishes a new paradigm for integrating traditional numerical methods with deep learning technologies, offering a robust platform for solving complex physical problems.

physics.flu-dyn↗

arXiv2024

Finite Volume Graph Network(FVGN): Predicting unsteady incompressible fluid dynamics with finite volume informed neural network

The rapid development of deep learning has significant implications for the advancement of Computational Fluid Dynamics (CFD). Currently, most pixel-grid-based deep learning methods for flow field prediction exhibit significantly reduced accuracy in predicting boundary layer flows and poor adaptability to geometric shapes. Although Graph Neural Network (GNN) models for unstructured grids based unsteady flow prediction have better geometric adaptability, these models suffer from error accumulation in long-term predictions of unsteady flows. More importantly, fully data-driven models often require extensive training time, greatly limiting the rapid update and iteration speed of deep learning models when facing more complex unsteady flows. Therefore, this paper aims to balance the demands for training overhead and prediction accuracy by integrating physical constraints based on the finite volume method into the loss function of the graph neural network. Additionally, it incorporates a twice-massage aggregation mechanism inspired by the extended stencil method to enhance the unsteady flow prediction accuracy and geometric shape generalization ability of the graph neural network model on unstructured grids. We focus particularly on the model's predictive accuracy within the boundary layer. Compared to fully data-driven methods, our model achieves better predictive accuracy and geometric shape generalization ability in a shorter training time.

physics.flu-dyn↗

arXiv2021

An explicit and non-iterative moving-least-squares immersed-boundary method with low boundary velocity error

In this work, based on the moving-least-squares immersed boundary method, we proposed a new technique to improve the calculation of the volume force representing the body boundary. For boundary with simple geometry, we theoretically analyse the error between the desired volume force at boundary and the actual force given by the original method. The ratio between the two forces is very close to a constant. Numerical experiments reveal that for complex geometry, this ratio exhibits very narrow distribution around certain value. A spatially uniform coefficient is then introduced to correct the force and fixed by the least-square method over all boundary markers. Such method is explicit and non-iterative, and can be easily implemented into the existing scheme. Several test cases have been simulated with stationary and moving boundaries. Our new method can reduce the residual boundary velocity to the level comparable to that given by the iterative method, but requires much less computing time. Moreover, the new method can be readily combined with the iterative method and further reduces the residual boundary velocity.

math.NA↗

arXiv2021

Realizing the ultimate scaling of the convection turbulence by spatially decoupling the thermal and viscous boundary layers

Turbulent convection plays a crucial role in many natural environments, ranging from Earth ocean, mantle and outer core, to various astrophysical systems. For such flows with extremely strong thermal driving, an ultimate scaling was proposed for the heat flux and velocity. Despite numerous experimental and numerical studies, a conclusive observation of the ultimate regime has not been reached yet. Here we show that the ultimate scaling can be perfectly realized once the thermal boundary layer is fully decoupled from the viscous boundary layer and locates inside the turbulent bulk. The heat flux can be greatly enhanced by one order of magnitude. Our results provide concrete evidences for the appearance of the ultimate state when the entire thermal boundary layer is embedded in the turbulent region, which is probably the case in many natural convection systems.

physics.flu-dyn↗

arXiv2020

Numerical validation and physical explanation of the universal force theory of three-dimensional steady viscous and compressible flow

In a recent paper, Liu et al. [``Lift and drag in three-dimensional steady viscous and compressible flow'', Phys. Fluids 29, 116105 (2017)] obtained a universal theory for the aerodynamic force on a body in three-dimensional steady flow, effective from incompressible all the way to supersonic regimes. In this theory, the total aerodynamic force can be determined solely with the vorticity distribution on a single wake plane locating in the steady linear far field. Despite the vital importance of this result, its validity and performance in practice has not been investigated yet. In this paper, we performed Reynolds-averaged Navier-Stokes simulations of subsonic, transonic, and supersonic flows over a three-dimensional wing. The aerodynamic forces obtained from the universal force theory are compared with that from the standard wall-stress integrals. The agreement between these two formulas confirms for the first time the validity of the theory in three-dimensional steady viscous and compressible flow. The good performance of the universal formula is mainly due to the fact that the turbulent viscosity in the wake is much larger than the molecular viscosity therein, which can reduce significantly the distance of the steady linear far field from the body. To further confirm the correctness of the theory, comparisons are made for the flow structures on the wake plane obtained from the analytical results and numerical simulations. The underlying physics relevant to the universality of the theory is explained by identifying different sources of vorticity in the wake.

physics.flu-dyn↗

arXiv2020

Risk assessment of airborne transmission of COVID-19 by asymptomatic individuals under different practical settings

The lack of quantitative risk assessment of airborne transmission of COVID-19 under practical settings leads to large uncertainties and inconsistencies in our preventive measures. Combining in situ measurements and numerical simulations, we quantify the exhaled particles from normal respiratory behaviors and their transport under elevator, small classroom and supermarket settings to evaluate the risk of inhaling potentially virus-containing particles. Our results show that the design of ventilation is critical for reducing the risk of particle encounters. Inappropriate design can significantly limit the efficiency of particle removal, create local hot spots with orders of magnitude higher risks, and enhance particle deposition causing surface contamination. Additionally, our measurements reveal the presence of substantial fraction of crystalline particles from normal breathing and its strong correlation with breathing depth.

physics.med-ph↗