Adaptive mesh refinement and load balancing based on multi-level block-structured Cartesian mesh

We developed a framework for a distributed-memory parallel computer that enables dynamic data management for adaptive mesh refinement and load balancing. We employed simple data structure of the building cube method (BCM) where a computational domain is divided into multi-level cubic domains and each cube has the same number of grid points inside, realising a multi-level block-structured Cartesian mesh. Solution adaptive mesh refinement, which works efficiently with the help of the dynamic load balancing, was implemented by dividing cubes based on mesh refinement criteria. The framework was investigated with the Laplace equation in terms of adaptive mesh refinement, load balancing and the parallel efficiency. It was then applied to the incompressible Navier–Stokes equations to simulate a turbulent flow around a sphere. We considered wall-adaptive cube refinement where a non-dimensional wall distance y⁺ near the sphere is used for a criterion of mesh refinement. The result showed the load imbalance due to y⁺ adaptive mesh refinement was corrected by the present approach. To utilise the BCM framework more effectively, we also tested a cube-wise algorithm switching where an explicit and implicit time integration schemes are switched depending on the local Courant-Friedrichs-Lewy (CFL) condition in each cube.     

Solution adaptive grids applied to low Reynolds number flow

A numerical study has been undertaken to investigate the use of a solution adaptive grid for flow around a cylinder in the laminar flow regime. The main purpose of this work is twofold. The first aim is to investigate the suitability of a grid adaptation algorithm and the reduction in mesh size that can be obtained. Secondly, the uniform asymmetric flow structures are ideal to validate the mesh structures due to mesh refinement and consequently the selected refinement criteria. The refinement variable used in this work is a product of the rate of strain and the mesh cell size, and contains two variables C_m and C_str which determine the order of each term. By altering the order of either one of these terms the refinement behaviour can be modified. Copyright © 2003 John Wiley & Sons, Ltd.     

Handling contacts in an Eulerian frame: a finite element approach for fluid structures with contacts

ABSTRACT

Eulerian variational formulations for deformable solids, with or without fluids around them, end up, after implicit time discretisation, as large non-linear systems for the velocities in the moving domains. Handling moving domains and moving boundaries requires careful meshing procedures; on the other hand, the detection of contact is particularly simple with a distance function. Then at every time step, a variational inequality can be used to update the velocities. This article gives new implementation details and two new complex simulations: a very soft bouncing ball in an axisymmetric flow and a disk hit by a club.     

Parallel adaptive high-order CFD simulations characterising SOFIA cavity acoustics

ABSTRACT

This paper presents large-scale parallel computational fluid dynamics simulations for the Stratospheric Observatory for Infrared Astronomy (SOFIA). SOFIA is an airborne, 2.5-m infrared telescope mounted in an open cavity in the aft fuselage of a Boeing 747SP. These simulations focus on how the unsteady flow field inside and over the cavity interferes with the optical path and mounting structure of the telescope. A temporally fourth-order accurate Runge–Kutta, and a spatially fifth-order accurate WENO-5Z scheme were used to perform implicit large eddy simulations. An immersed boundary method provides automated gridding for complex geometries and natural coupling to a block-structured Cartesian adaptive mesh refinement framework. Strong scaling studies using NASA's Pleiades supercomputer with up to 32 k CPU cores and 4 billion computational cells show excellent scaling. Dynamic load balancing based on execution time on individual adaptive mesh refinement (AMR) blocks addresses irregular numerical cost associated with blocks containing boundaries. Limits to scaling beyond 32 k cores are identified, and targeted code optimisations are discussed.     

Adaptive mesh refinement for two-phase slug flows with an a priori indicator

The purpose of this paper is to present a novel adaptive mesh refinement AMR technique for computing unstable one dimensional two-phase flows in pipelines. In multiphase flows, the prediction and localisation of inter-facial waves, slugs and instabilities related to flow conditions under study require high levels of accuracy. This is more apparent in systems at industrial scales, where flow lines possess highly distorted regions and irregular topologies.Uniform fine meshes for these long devices are costly and in general situations the optimum space discretisation could not be determined a priori.Adaptive mesh refinement AMR procedure provides a remedy to this problem by refining the mesh locally, allowing to capture regions where sharp discontinuities and steep gradients are present. With appropriate algorithm and data organisation, AMR helps to reduce CPU time and speeds up simulations of flows in long pipes. The effectiveness of AMR methods relies on estimators that determine where refinement is required. We show in this work that for transient flows combining gradient-based error estimator with Kelvin–Helmholtz stability condition can improve the acceleration of computation and locate regions where refinements are required. The Kelvin–Helmholtz is a local condition and is an a priori indicator for the refinement.     

A novel Cartesian cut cell solver for incompressible viscous flow in irregular domains

A Cartesian cut cell solver with solution‐based adaptive mesh refinement is developed for simulating viscous, incompressible flows with arbitrary complex geometries. The cut cells are automatically generated using Volume CAD (VCAD), a framework for storing geometric and material attribute data. Unlike earlier cut cell methods, this solver organizes the cutting patterns into only six categories and further subdivides the resulting pentagon into two quadrilaterals, such that mesh data can be stored by uniform data structure and the post‐processing of flow data can be handled conveniently. A novel method is proposed to treat minuscule cut cells without the process of cell merging. A collocated finite volume method, which can be used even when multiple cell shapes and orthogonal and non‐orthogonal grids exist in the decomposition, is employed to discretize the Navier–Stokes equations. A modified SIMPLE‐based smoothing pressure correction scheme is applied in this cut cell method to suppress checkerboard pressure oscillations caused by collocated arrangement. The solver is first used to simulate a channel flow to demonstrate its calculation accuracy expressed with L₁ and L_∞ norm errors and then the method is utilized to solve three benchmark problems of flow and heat transfer within irregular domains to verify its feasibility, efficiency, accuracy and potential in engineering applications. Copyright © 2010 John Wiley & Sons, Ltd.     

A unified mesh refinement method with applications to porous media flow

A unified algorithm is presented for the refinement of finite element meshes consisting of tensor product Lagrange elements in any number of space dimensions. The method leads to repeatedly refined n-irregular grids with associated constraint equations. Through an object-oriented implementation existing solvers can be extended to handle mesh refinements without modifying the implementation of the finite element equations. Various versions of the refinement procedure are investigated in a porous media flow problem involving singularities around wells. A domain decomposition-type finite element method is also proposed based on the refinement technique. This method is applied to flow in heterogeneous porous media. © 1998 John Wiley & Sons, Ltd.     

The use of dynamic grid adaptation algorithms for the modelling of flow around a circular cylinder in sub‐critical flow regime

In the present study a dynamic grid adaptation (DGA) algorithm is utilized for predicting flow around a circular cylinder in sub‐critical flow regime at a Reynolds number of 1.4×10⁵. The reason for adopting a DGA algorithm is the unsteadiness of the flow field which makes a conventional mesh inefficient. The concept being adopted is to concentrate mesh refinement in regions with high gradients and high turbulent viscosity, while in the region further downstream where the flow is fully developed a coarser mesh will develop and turbulence is modelled with the large eddy simulation (LES) turbulence model. The aim of the study is to present an appropriate variable for mesh refinement, which accomplishes a high rate of mesh refinement in the region with high gradients. The new variable is a product of the local mesh cell size and the rate of strain and includes two additional variables to allow control over the refinement behaviour. The results are compared with experimental data at the corresponding Reynolds number and also with numerical results obtained with conventional mesh. It is demonstrated that DGA algorithms can give results of a very high quality for a mesh that is significantly smaller than for a conventional mesh. Copyright © 2003 John Wiley & Sons, Ltd.     

Segmented multigrid domain decomposition procedure for incompressible viscous flows

The application of grid stretching or grid adaptation is generally required in order to optimize the distribution of nodal points for fluid-dynamic simulation. This is necessitated by the presence of disjoint high gradient zones, that represent boundary or free shear layers, reversed flow or vortical flow regions, triple deck structures, etc. A domain decomposition method can be used in conjunction with an adaptive multigrid algorithm to provide an effective methodology for the development of optimal grids. In the present study, the Navier-Stokes (NS) equations are approximated with a reduced Navier-Stokes (RNS) system, that represents the lowest-order terms in an asymptotic Re expansion. This system allows for simplified boundary conditions, more generality in the location of the outflow boundary, and ensures mass conservation in all subdomain grid interfaces, as well as at the outflow boundary. The higher-order (NS) diffusion terms are included through a deferred corrector, in selected subdomains, when necessary. Adaptivity in the direction of refinement is achieved by grid splitting or domain decomposition in each level of the multigrid procedure. Normalized truncation error estimates of key derivatives are used to determine the boundaries of these subdomains. The refinement is optimized in two co-ordinate directions independently. Multidirectional adaptivity eliminates the need for grid stretching so that uniform grids are specified in each subdomain. The overall grid consists of multiple domains with different meshes and is, therefore, heavily graded. Results and computational efficiency are discussed for the laminar flow over a finite length plate and for the laminar internal flow in a backward-facing step channel.     

Simplex space-time meshes in compressible flow simulations

Employing simplex space-time meshes enlarges the scope of compressible flow simulations. The simultaneous discretization of space and time with simplex elements extends the flexibility of unstructured meshes from space to time. In this work, we adapt a finite element formulation for compressible flows to simplex space-time meshes. The method obtained allows, for example, flow simulation on spatial domains that change topology with time. We demonstrate this with the two-dimensional simulation of compressible flow in a valve that fully closes and opens again. Furthermore, simplex space-time meshes facilitate local temporal refinement. A three-dimensional transient simulation of blow-by past piston rings is run in parallel on 120 cores. The timings point out savings of computation time gained from local temporal refinement in four-dimensional space-time meshes.     

A new general-purpose least-squares finite element model for steady incompressible low-viscosity laminar flow using isoparametric C1-continuous Hermite elements

This paper deals with a critical evaluation of various finite element models for low-viscosity laminar incompressible flow in geometrically complex domains. These models use Galerkin weighted residuals UVP, continuous penalty, discrete penalty and least-squares procedures. The model evaluations are based on the use of appropriate tensor product Lagrange and simplex quadratic triangular elements and a newly developed isoparametric Hermite element. All of the described models produce very accurate results for horizontal flows. In vertical flow domains, however, two different cases can be recognized. Downward flows, i.e. when the gravitational force is in the direction of the flow, usually do not present any special problem. In contrast, laminar flow of low-viscosity Newtonian fluids where the gravitational force is acting in the direction opposite to the flow presents a difficult case. We show that only by using the least-squares method in conjunction with C¹-continuous Hermite elements can this type of laminar flow be modelled accurately. The problem of smooth isoparametric mapping of C¹ Hermite elements, which is necessary in dealing with geometrically complicated domains, is tackled by means of an auxiliary optimization procedure. We conclude that the least-squares method in combination with isoparmetric Hermite elements offers a new general-purpose modelling technique which can accurately simulate all types of low-viscosity incompressible laminar flow in complex domains.     

Boundary element analysis of three-dimensional mixing flow of Newtonian and viscoelastic fluids

The boundary element method (BEM) is implemented for the simulation of three-dimensional transient flows of typical relevance to mixing. Creeping Newtonian and viscoelastic fluids of the Maxwell type are examined. A boundary-only formulation in the time domain is proposed for linear viscoelastic flows. Special emphasis is placed on cavity flows involving simple- and multiple-connected moving domains. The BEM becomes particularly suited in multiple-connected flows, where part of the boundary (stirrer or rotor) is moving, and the remaining outer part (cavity or barrel) is at rest. In this case, conventional methods, such as the finite element method (FEM), generally require remeshing or mesh refinement of the three-dimensional fluid volume as the flow evolves and the domain of computation changes with time. The BEM is shown to be much easier to implement since the kinematics of the elements bounding the fluid is known (imposed). It is found that, for simple cavity flow induced by a rotating vane at constant angular velocity, the tractions at the vane tip and cavity face exhibit non-linear periodic dynamical behavior with time for fluids obeying linear constitutive equations. © 1998 John Wiley & Sons, Ltd.     

Adaptive mesh refinement method-based large eddy simulation for the flow over circular cylinder at ReD = 3900

With the development of computational power, large eddy simulation (LES) method is increasingly used in simulating complex flow. However, there still exist many factors affecting the LES quality and appropriate mesh resolution is among one of them. This work aims to develop an automatic procedure to refine the LES mesh by combining adaptive mesh refinement (AMR) and LES quality criteria. An LES refinement criterion is developed by estimating the proper grid length scale which meets the accuracy requirement of LES method. With this criterion, the baseline mesh is automatically refined with the AMR method. In this work, an efficient one-shot refinement strategy is also proposed to reduce the overall simulation time. Current AMR-based LES method is verified with the typical LES test case about the flow past circular cylinder at Re _D = 3900. Results show that the automatically refined mesh provides systematically better agreement with experimental results and with current method the balance between accuracy and computational expense for LES can be obtained.     

Automatic generation of unstructured grids with Delaunay triangulation and its application

This paper is consisted of two parts. In the first part, a method is described which generates two-dimensional triangle mesh using the Delaunay triangulation criterion. An automatic algorithm was proposed which combines several advantages of the existing methods. Local mesh refinement can also be easily performed with this method. Examples of generated grids were presented for several convex, non-convex and multi-connected domains to demonstrate the effectiveness and feasibility of the proposed method. In the second part, the turbulent heat transfer in an annular space finned by wave-like longitudinal fins was numerical simulated. The proposed technique was adopted to generate the grid in the cross-section. The standard K-ɛ model in conjuction with wall function method was used to simulate the fluid flow and heat transfer in the complex geometry. The discretization of the governing equations was described. The computational results were compared with the authors' test data and the agreement was reasonably good. Received on 9 July 1998     

Two-dimensional numerical simulations of multi-headed detonations in oxygen-aluminum mixtures using an adaptive mesh refinement

Abstract. Two-dimensional numerical simulations of detonations in two-phase lean mixtures of aluminum particles and pure oxygen have been performed. The computational procedure adopts an adaptive mesh refinement methodology in order to increase spatial resolution in the most interesting parts of the flow field. A one-step heterogeneous reaction describes the evaporation and combustion of aluminum. Depending on the gas-phase temperature, the combustion product is aluminum oxide or aluminum monoxide. The results show that the heterogeneous detonations resemble gaseous single-phase ones although the scale of the phenomena is very different. The detonation of aluminum dust evolves into the 2-headed mode of propagation with the characteristic detonation cell width equal to cm. For aluminum dust the cellular structure is much finer. The detonation initially propagates in the 11-headed mode with the characteristic cell width equal to cm and evolves into the 8.5-headed mode with the characteristic cell size $\lambda_{\rm cell}$ equal to cm. Received 7 May 2001 / Accepted 25 March 2002 Published online 23 January 2003 Correspondence to: K. Benkiewicz (e-mail: kbenk@cow.me.aoyama.ac.jp)     

High order parallelisation of an unstructured grid,discontinuous-Galerkin finite element solver for the Boltzmann–BGK equation

ABSTRACT

This paper outlines the implementation and performance of a parallelisation approach involving partitioning of both physical space and velocity space domains for finite element solution of the Boltzmann-BGK equation. The numerical solver is based on a discontinuous Taylor–Galerkin approach. To the authors' knowledge this is the first time a ‘high order’ parallelisation, or `phase space parallelisation', approach has been attempted in conjunction with a numerical solver of this type. Restrictions on scalability have been overcome with the implementation detailed in this paper. The developed algorithm has major advantages over continuum solvers in applications where strong discontinuities prevail and/or in rarefied flow applications where the Knudsen number is large. Previous work by the authors has outlined the range of applications that this solver is capable of tackling. The paper demonstrates that the high order parallelisation implemented is significantly more effective than previous implementations at exploiting High Performance Computing architectures.     

Assessment of scale-adaptive turbulence models for volute-type centrifugal pumps at part load operation

Unsteady three-dimensional (3D) computational fluid dynamics (CFD) simulations are conducted with the open-source software OpenFOAM to assess the scale-adaptive k-ω-SST-SAS turbulence model (SAS) on a radial, volute-type centrifugal pump at part load operation which is characterized by high unsteadiness and flow separation. SAS results are compared to spatially high resolved and ensemble-averaged flow measurement data in terms of flow angle and turbulence intensity (TI) in the rotor–stator interaction region. Differences to simulation results obtained with the statistical state-of-the-art k-ω-SST turbulence model (SST) are highlighted. The flow angle is predicted with a reasonable agreement to measurement data by both, SST and SAS models. In the highly transient flow of strong rotor–stator interaction near the volute tongue, SST results show a significant overprediction of measured TI while the SAS model yields a considerably better agreement to measurement data even with a typical URANS grid resolution. A grid refinement does not further improve the agreement to measurement data. An in-depth analysis of the SAS model on separated flow, i.e., periodic hill test case, together with a large eddy simulation (LES) reference solution is performed and reveals that with successive grid refinement, in contrast to LES, the SAS model in its present form of Egorov and Menter (2008) does not resolve a successively larger portion of the turbulence spectrum, and the modeled part is not successively reduced. For that purpose, a re-calibration or even a re-formulation of the scale-adaption source term in the transport equation of the turbulent dissipation rate may be indispensable, which will be the subject of future studies.     

Parallel adaptive refinement for unsteady flow calculations on 3D unstructured grids

A parallel adaptive refinement algorithm for three‐dimensional unstructured grids is presented. The algorithm is based on an hierarchical h‐refinement/derefinement scheme for tetrahedral elements.The algorithm has been fully parallelized for shared‐memory platforms via a domain decomposition of the mesh at the algebraic level. The effectiveness of the procedure is demonstrated with applications which involve unsteady compressible fluid flow. A parallel speedup study of the algorithm also is included. Published in 2004 by John Wiley & Sons, Ltd.     

Feature‐based adaptive mesh refinement for wingtip vortices

Feature‐based solution‐adaptive mesh refinement is an attractive strategy when it is known a priori that the resolution of certain key features is critical to achieving the objectives of a simulation. In this paper, we apply vortex characterization techniques, which are typically employed to visualize vortices, to identify regions of the computational domain for mesh refinement. We investigate different refinement strategies that are facilitated by these vortex characterization techniques to simulate the flow past a wing in a wind tunnel. Our results, which we compare with experimental data, indicate that it is necessary to refine the region within and near the vortex extent surface to obtain an accurate prediction. Application of the identified mesh refinement strategy also produced observed improvement in the results predicted for a spinning missile with deflected canards. Copyright © 2010 John Wiley & Sons, Ltd.     

Study on flow past two spheres in tandem arrangement using a local mesh refinement virtual boundary method

A local mesh refinement virtual boundary method based on a uniform grid is designed to study the transition between the flow patterns of two spheres in tandem arrangement for Re=250. For a small gap (L/D=1.5), the flow field is axisymmetric. As the spacing ratio increases to 2.0, the pressure gradient induces the circumferential fluid motion and a plane‐symmetric flow is constructed through a regular bifurcation. For L/D?2.5, the vortices are periodically shed from the right sphere, but the planar symmetry remains. The case for L/D=3.0 is picked up to give a detail investigation for the unsteady flow. The shedding frequency of vortical structure from the upper side of the right sphere is found to be double of the frequency of the lower side. With the flow spectra of various gaps given, the underlying competitive mechanism between the two shedding frequencies is studied and a critical spacing gap is revealed. Copyright © 2005 John Wiley & Sons, Ltd.