Computers & Fluids最新文献

{"title":"Invariant control strategies for active flow control using graph neural networks","authors":"Marius Kurz ,&nbsp;Rohan Kaushik ,&nbsp;Marcel Blind ,&nbsp;Patrick Kopper ,&nbsp;Anna Schwarz ,&nbsp;Felix Rodach ,&nbsp;Andrea Beck","doi":"10.1016/j.compfluid.2025.106854","DOIUrl":"10.1016/j.compfluid.2025.106854","url":null,"abstract":"<div><div>Reinforcement learning (RL) has recently gained traction for active flow control tasks, with initial applications exploring drag mitigation via flow field augmentation around a two-dimensional cylinder. RL has since been extended to more complex turbulent flows and has shown significant potential in learning complex control strategies. However, such applications remain computationally challenging owing to its sample inefficiency and associated simulation costs. This fact is worsened by the lack of generalization capabilities of these trained policy networks, often being implicitly tied to the input configurations of their training conditions. In this work, we propose the use of graph neural networks (GNNs) to address this particular limitation, effectively increasing the range of applicability and getting more <em>value</em> out of the upfront RL training cost. GNNs can naturally process unstructured, three-dimensional flow data, preserving spatial relationships without the constraints of a Cartesian grid. Additionally, they incorporate rotational, reflectional, and permutation invariance into the learned control policies, thus improving generalization and thereby removing the shortcomings of commonly used convolutional neural networks (CNNs) or multilayer perceptron (MLP) architectures. To demonstrate the effectiveness of this approach, we revisit the well-established two-dimensional cylinder benchmark problem for active flow control. The RL training is implemented using Relexi, a high-performance RL framework, with flow simulations conducted in parallel using the high-order discontinuous Galerkin framework FLEXI. Our results show that GNN-based control policies achieve comparable performance to existing methods while benefiting from improved generalization properties. This work establishes GNNs as a promising architecture for RL-based flow control and highlights the capabilities of Relexi and FLEXI for large-scale RL applications in fluid dynamics.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"303 ","pages":"Article 106854"},"PeriodicalIF":3.0,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145218983","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A novel interface reconstruction method based on B-Spline parametric surfaces: Application to free-falling natural particle in thermo-buoyant flows","authors":"Lorenzo Vallisa ,&nbsp;Delphine Laboureur ,&nbsp;Maria Teresa Scelzo ,&nbsp;Silvania Lopes ,&nbsp;Michel De Paepe","doi":"10.1016/j.compfluid.2025.106843","DOIUrl":"10.1016/j.compfluid.2025.106843","url":null,"abstract":"<div><div>This paper provides an innovative formulation based on B-Spline surfaces to explicitly represent liquid–solid interfaces in multiphase problems. B-Spline surfaces constitute a robust framework for representing complex geometries. Their ability to provide continuous access to surface information across the entire parametric domain offers a substantial advantage in managing boundary point distributions, particularly when compared to the traditional connectivity matrix employed in hybrid front-tracking methods and immersed boundary approaches. This work begins by introducing a definition for a closed parametric surface that fully encompasses the boundary of a solid object. Building on this foundation, it presents a parsing algorithm designed for systematic organization. The algorithm processes an unordered set of boundary points, representing the rigid body’s boundary, and arranges them into a structure aligned with the defined closed parametric surface. Leveraging the advanced geometric capabilities of B-Splines, we introduce a novel procedure for the computation of the indicator function, whose computational advantages over established methods such as the Poisson equation is then proved in a benchmark test. The presented geometrical tools are then integrated into a direct-forcing immersed boundary method. The solver is enhanced with a <em>extension field</em> technique, to address numerical instabilities caused by spurious oscillations arising near the immersed boundary. Additionally, it is extended to incorporate the effects of thermobuoyant flow dynamics. The solver is rigorously validated through both isothermal and non-isothermal test cases and subsequently applied to investigate the comparative free-falling behavior of a spherical object and a natural particle with an equivalent spherical volume. The results demonstrate that, despite the morphology of the natural particle exhibits only a slight deviation from that of a spherical object, its asymmetries significantly influence its trajectory.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"303 ","pages":"Article 106843"},"PeriodicalIF":3.0,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145263970","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Applying a compact porous media model to numerically derive resistance coefficients for lattice structures","authors":"Kyle Pittman ,&nbsp;Jacob Riglin ,&nbsp;Jay Chen ,&nbsp;Cesar Dominguez ,&nbsp;Marvin Davis ,&nbsp;Rami Batrice","doi":"10.1016/j.compfluid.2025.106798","DOIUrl":"10.1016/j.compfluid.2025.106798","url":null,"abstract":"<div><div>Additive Manufacturing allows for exploring various geometries to achieve specific engineering criteria. Lattices are one geometry with unique properties, including being periodically repeating structures which allow flow through them to be represented as a porous media according to Darcy-Forchheimer equations. These equation’s coefficients are generally experimentally derived, but this work demonstrates the ability to numerically derive them with CFD. Simulations were performed using three-dimensional stead state Reynolds-averaged Navier-Stokes with a k-ω Shear Stress Transport turbulence model using Ansys Fluent. Three lattice geometries were investigated and drag coefficients were derived. The method was validated against externally published data for similar geometries demonstrating strong agreement, and grid convergence for all simulations was calculated with a Grid Convergence Index method. Wall roughness is demonstrated to have a non-negligible impact on results and roughness values are considered for the primary focus Octahedral geometry where both smooth wall and rough wall coefficients were derived. The porosity coefficients for the Octahedral geometry at 1.0 [m/s] were found to be 2.89×10⁶ and 2.90×10⁶ [1/(Pa*m*s)] for the permeability coefficients, 6.37×10¹ and 5.44×10¹ [m²/kg] for the inertial resistance coefficients, and with a max pressure drop of 5116.7 [Pa] and 4429.5 [Pa] for the smooth walls and rough walls, respectively. The derived numerical method enables rapid exploration and optimization of new lattice designs for diverse engineering applications.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"303 ","pages":"Article 106798"},"PeriodicalIF":3.0,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145264408","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An Eulerian–Lagrangian approach to simulate turbidity currents","authors":"Boyang Chen ,&nbsp;Zhen Liu ,&nbsp;Bruño Fraga","doi":"10.1016/j.compfluid.2025.106832","DOIUrl":"10.1016/j.compfluid.2025.106832","url":null,"abstract":"<div><div>We present an Eulerian–Lagrangian four-way coupled algorithm within a large-eddy simulation framework to simulate turbidity currents. Our approach preserves the particle-driven, dispersed nature of such currents with significantly reduced reliance on semi-empirical parametrisation. We are capable of reproducing key processes such as the entrainment of fluid within the particle-laden current and the settlement and re-suspension of solid particles. Particle interactions are handled using a soft-sphere collision model. Our results are successfully validated versus experimental results. We investigate a lock-exchange setup in a numerical flume, comparing predicted front velocities and deposition profiles with experimental measurements. Furthermore, we analyse the differences between particle-driven and gravity-driven currents (simulated via an Eulerian–Eulerian approach), focusing on propagation velocity, transition to turbulence and the generation of coherent structures in the shear layer.</div><div>We use our model to examine the evolution and driving mechanisms of turbidity currents. We describe in detail how Kelvin–Helmholtz singularities evolve into a well-defined current head and sediment trail, while also accounting for the mechanical effect of lifting the sliding gate that initially separates the laden and unladen liquids. The dynamics of particle settling and re-suspension and their correlation with bed shear stress and the current’s reach are predicted, showing good agreement with experimental data. The local friction velocity at the flume’s bed peaks at the current head during early development and is later redistributed via a re-suspension mechanism linked to instantaneous turbulent structures. Finally, an energy budget analysis reveals that turbulent kinetic energy dissipation is primarily due to the settling of solid particles.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"303 ","pages":"Article 106832"},"PeriodicalIF":3.0,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145264458","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Comparison of entropy stable collocation high-order DG methods for compressible turbulent flows","authors":"Anna Schwarz ,&nbsp;Daniel Kempf ,&nbsp;Jens Keim ,&nbsp;Patrick Kopper ,&nbsp;Christian Rohde ,&nbsp;Andrea Beck","doi":"10.1016/j.compfluid.2025.106874","DOIUrl":"10.1016/j.compfluid.2025.106874","url":null,"abstract":"<div><div>High-order methods are well-suited for the numerical simulation of complex compressible turbulent flows, but require additional stabilization techniques to capture instabilities arising from the underlying non-linear hyperbolic equations. This paper provides a detailed comparison of the effectiveness of entropy stable discontinuous Galerkin methods for the stabilization of compressible (wall-bounded) turbulent flows. For this investigation, an entropy stable discontinuous Galerkin spectral element method is applied on Gauss–Legendre and Gauss–Lobatto nodes. In the compressible regime, an additional stabilization technique for shock capturing based on a convex blending of a low-order finite volume with the high-order discontinuous Galerkin operator is utilized. The present investigation provides a systematic study from convergence tests, to the Taylor–Green vortex and finally to a more intricate turbulent wall-bounded 3D diffuser flow, encompassing both weakly compressible and compressible regimes. The comparison demonstrates that the DGSEM on Gauss–Lobatto nodes is generally less accurate for an equal amount of degrees of freedom. Conversely, it is faster than the DGSEM on Gauss–Legendre nodes due to a less severe time step restriction and simpler numerical operator. A performance comparison reveals that the DGSEM on Gauss–Lobatto nodes generally outperforms the DGSEM on Gauss nodes for under-resolved turbulence in the subsonic regime on a periodic domain. Conversely, the opposite effect can be observed for wall-bounded flows as well as the supersonic regime, the latter depending of course on the chosen shock-capturing scheme. To the author’s knowledge, this is the first time for which a comparison of entropy stable DGSEM on Gauss–Lobatto and Gauss–Legendre has been performed for compressible, wall-bounded turbulent flows with separation.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"303 ","pages":"Article 106874"},"PeriodicalIF":3.0,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145359265","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Physics appropriate interface capturing reconstruction approach for viscous compressible multicomponent flows","authors":"Amareshwara Sainadh Chamarthi","doi":"10.1016/j.compfluid.2025.106858","DOIUrl":"10.1016/j.compfluid.2025.106858","url":null,"abstract":"<div><div>The paper proposes a physically consistent numerical discretization approach for simulating viscous compressible multicomponent flows. It has two main contributions. First, a contact discontinuity (and material interface) detector is developed. In those regions of contact discontinuities, the THINC (Tangent of Hyperbola for INterface Capturing) approach is used for reconstructing appropriate variables (phasic densities). For other flow regions, the variables are reconstructed using the Monotonicity-preserving (MP) scheme (or Weighted essentially non-oscillatory scheme (WENO)). For reconstruction in the characteristic space, the THINC approach is used only for the contact (or entropy) wave and volume fractions. For the reconstruction of primitive variables, the THINC approach is used for phasic densities and volume fractions only, offering an effective solution for reducing dissipation errors near contact discontinuities. The numerical results of the benchmark tests show that the proposed method captured the material interface sharply compared to existing methods. The second contribution is the development of an algorithm that uses a central reconstruction scheme for the tangential velocities, as they are continuous across material interfaces in viscous flows. In this regard, the Ducros sensor (a shock detector that cannot detect material interfaces) is employed to compute the tangential velocities using a central scheme across material interfaces. Using the central scheme does not produce any oscillations at the material interface. The proposed approach is thoroughly validated with several benchmark test cases for compressible multicomponent flows, highlighting its advantages. The physics appropriate approach also shown to prevent spurious vortices, despite being formally second-order accurate for nonlinear problems, on a coarser mesh than a genuinely high-order accurate method.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"303 ","pages":"Article 106858"},"PeriodicalIF":3.0,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145264410","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Obtaining converged flow solutions using quantum annealing","authors":"Takahito Asaga ,&nbsp;Yuichi Kuya","doi":"10.1016/j.compfluid.2025.106856","DOIUrl":"10.1016/j.compfluid.2025.106856","url":null,"abstract":"<div><div>This paper proposes numerical methods to obtain converged flow solutions by quantum annealing. The proposed quantum annealing methods are developed for lattice gas automata (LGA) and finite difference methods (FDMs). The quadratic unconstrained binary optimization (QUBO) model for LGA consists of the cost functions for the steady-state flow condition, collision law condition, boundary condition, and flow field condition. In contrast, the QUBO model for FDMs is built directly from the discretized governing equations expressed in a binary form. In the numerical experiments of channel flows, both proposed methods successfully extract the converged velocity profiles from a large number of flow state combinations by quantum annealing. The obtained solutions closely match those obtained by the conventional or analytical approach. Since, due to the difference in characteristics between LGA and FDMs, FDMs can reduce the scale of combinatorial optimization problems more efficiently than LGA, the proposed FDM-based method obtains more accurate solutions than the proposed LGA-based method.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"303 ","pages":"Article 106856"},"PeriodicalIF":3.0,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145325681","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A kinetic scheme based on positivity preservation with exact shock capture","authors":"Shashi Shekhar Roy ,&nbsp;S.V. Raghurama Rao","doi":"10.1016/j.compfluid.2025.106857","DOIUrl":"10.1016/j.compfluid.2025.106857","url":null,"abstract":"<div><div>In this paper, we present a kinetic model with flexible velocities that satisfy positivity preservation conditions for the Euler equations. Our 1D kinetic model consists of two velocities and employs both the asymmetrical and symmetrical models. Switching between the two models is governed by our formulation of kinetic relative entropy, together with an additional criterion that ensures a robust and accurate scheme yielding entropic results. In 2D, we introduce a novel three-velocity kinetic model, defined to ensure a locally one-dimensional formulation for the resulting macroscopic normal flux. For first order accuracy, we also obtain a limit on the time step that ensures positivity preservation. The resulting numerical scheme captures grid-aligned steady shocks exactly. Several benchmark compressible flow test cases are solved in 1D and 2D to demonstrate the efficacy of the proposed solver.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"303 ","pages":"Article 106857"},"PeriodicalIF":3.0,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145325684","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Transfer function of low-pass filters on unstructured grids","authors":"Hiroyuki Asada,&nbsp;Kanako Maruyama,&nbsp;Soshi Kawai","doi":"10.1016/j.compfluid.2025.106873","DOIUrl":"10.1016/j.compfluid.2025.106873","url":null,"abstract":"<div><div>Low-pass filters designed on unstructured grids are investigated in terms of a transfer function in the wavenumber space. The transfer functions on unstructured grids are derived, and the properties of the low-pass filters for removing high-wavenumber components and inducing phase errors are investigated through the derived transfer functions. The transfer function reveals that the low-pass filters on unstructured grids can achieve the property that higher-wavenumber components are removed more by adjusting a filter coefficient to a small value, whereas large filter coefficients induce unfavorable amplifications of high-wavenumber components. The presence of phase errors induced by the low-pass filters on triangle unstructured cells is also found by the transfer function. Furthermore, the transfer function shows that the numerical methods for evaluating the gradients that appear in the filter formulation affect the characteristics of the low-pass filters and that the simplest central scheme can have an advantage in terms of retaining numerical robustness by removing high-wavenumber components compared to the edge-normal augmentation scheme. The numerical experiments of inviscid Taylor–Green vortex and shock-vortex interaction are also conducted with the low-pass filter coupled with the non-dissipative kinetic energy and entropy preserving (KEEP) scheme on unstructured grids, demonstrating the validity of the present transfer function of the low-pass filter.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"303 ","pages":"Article 106873"},"PeriodicalIF":3.0,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145325679","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Convergence acceleration algorithms for non-overlapping domain decomposition in near-wall turbulence modeling","authors":"N. Smirnova ,&nbsp;S. Utyuzhnikov ,&nbsp;V. Titarev ,&nbsp;M. Petrov","doi":"10.1016/j.compfluid.2025.106855","DOIUrl":"10.1016/j.compfluid.2025.106855","url":null,"abstract":"<div><div>In turbulence modeling the resolution of near-wall boundary layer requires most of computing time. The near-wall domain decomposition (NDD) approach proved to be efficient in tackling this problem. It represents a trade-off between computing time and accuracy. In this method, the computational domain is divided into two non-overlapping subdomains: the inner layer and region outside. The interface boundary conditions of Robin type are set by transferring the boundary conditions from the wall to the interface boundary. In contrast to the exact NDD, in the approximate NDD a simplified system of equations is solved in the near-wall subdomain. In the current paper a variant of the exact NDD is proposed, that uses an operator corresponding to the approximate NDD approach as a preconditioner. To improve the efficiency of NDD methods the GMRES method is applied. The efficacy of NDD algorithms are compared against for the low-Reynolds-number model.</div></div>","PeriodicalId":287,"journal":{"name":"Computers & Fluids","volume":"303 ","pages":"Article 106855"},"PeriodicalIF":3.0,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145263969","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}