Sie sind auf Seite 1von 8

International Journal of Mechanical Sciences 78 (2014) 19 26

Contents lists available at ScienceDirect

International Journal of Mechanical Sciences


journal homepage: www.elsevier.com/locate/ijmecsci

Comparison of effects on technical variances of computational uid dynamics (CFD) software based on nite element and nite volume methods
Woowon Jeong a,n, Jaehoon Seong b
a b

Department of Mechanical Engineering, Myongji University, San 38-2, Namdong, Cheoingu, Yongin, Gyeonggi-do 449-728, South Korea Department of Mechanical Engineering, California State Polytechnic University, 3801 West Temple Ave, Pomona, CA 91768, USA

art ic l e i nf o
Article history: Received 19 March 2013 Received in revised form 16 October 2013 Accepted 18 October 2013 Keywords: Finite element method (FEM) Finite volume method (FVM) Computational uid dynamics (CFD) Flow pattern

a b s t r a c t
Recently, many studies for ow characteristics have used commercial computational uid dynamics (CFD) software to simulate ow phenomena in various elds. The aim of this study is to contribute to the discussion on the efciency of nite element (FE) and nite volume (FV) methods, which have mainly used as CFD solvers for situations with the same number of mesh elements and geometries. We compared the accuracy of the results and the ow patterns among three different commercial codes Ansys CFX, Fluent and ADINA, which are used in many elds. We calculated three-dimensional incompressible laminar ow problems with simple pipe and bifurcation models. Our results showed that FEM CFD codes (ADINA) were more inuenced by the mesh type and quality compared to FVM CFD codes (Ansys CFX, Fluent). The computation time was found to be signicantly shorter for the FVM compared to the FEM codes. & 2013 Elsevier Ltd. All rights reserved.

1. Introduction Computational uid dynamics (CFD) is one of the branches of uid mechanics predicting uid ow, heat transfer, mass transfer, chemical reactions, and related phenomena by solving the mathematical equations which govern these processes using a numerical process. Claude-Louis Navier and George Gabriel Stokes introduced viscous transport into the Euler equations, which resulted in the NavierStokes equation based on current CFD. Richardson [1] developed the rst numerical weather prediction system when he divided physical space into grid cells and used the nite difference approximations of Bjerknes's primitive differential equations. The earliest numerical solution for ow past a cylinder was carried out by Thom [2]. Thus, CFD was developed from the pioneering efforts by Richardson [1], Thom [2], Courant et al. [3], Southwell [4], and von Neumann [5], who in their endeavors to procure insight into uid motion produced the development of powerful numerical techniques that can describe all types of uid ow [6]. The theoretical division of NASA contributed many numerical methods, and Spalding and his colleagues developed many codes and algorithms [7]. Commercial CFD codes began to open the market place from the early 1980s. During the last 30 years, a market for commercial CFD software

Corresponding author. Tel.: 82 70 86354985; fax: 82 31 3306957. E-mail addresses: wjeong76@gmail.com, woowon40@naver.com (W. Jeong).

grew quickly, and commercial CFD software is used in almost all engineering elds [8]. CFD is based on three principle numerical approaches the Finite Difference Method (FDM), Finite Element Method (FEM) and Finite Volume Method (FVM). Finite difference (FD) discretization is the oldest method and is based on the application of polynomial, Legendre polynomial, Fourier and Taylor series expansions to represent the differential equations [9]. This scheme motivated the use of an integral form of partial differential equations (PDEs) and subsequently, the development of the nite element and nite volume techniques. Current CFD mainly uses the FEM and FVM more than the FDM, which has trouble handling complicated geometries. Finite element (FE) discretization divides up the region into a number of smaller regions (nite elements) for the computational domain is based on a piecewise approximation of the solution. The PDEs that are solved are typically obtained by restating the conservation equation in a weak formulation [10,11]. This solving process was established by the Galerkin method. Finite volume (FV) discretization is based on an integral form of the PDE to be solved, with the values of the conserved variables averaged across the volume. The PDE is written in a form which can be solved for a given nite volume (or cell). The computational domain is discretized into nite volumes, and then for every volume the governing equations are solved [10,12]. Fig. 1 represents the structure mesh for the two discretization methods. The FE and FV schemes were applied and compared by many previous researchers in the eld of engineering uid dynamics. These

0020-7403/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ijmecsci.2013.10.017

20

W. Jeong, J. Seong / International Journal of Mechanical Sciences 78 (2014) 19 26

Fig. 1. A representation of a structured mesh for the two discretization methods analyzed by (a) the nite element method, and (b) the nite volume method.

schemes are still being analyzed by many researchers. In previous studies comparing the FE and FV schemes, the results of these methods have had slight differences, as FVM provides discrete solutions, while FEM provides a continuous (up to a point) solution. However, all of the FEM and FVM results compared favorably with the experimental data [1215]. Fluid ow has a lot of characteristics which are difcult to measure experimentally, e.g. the ow in a complex geometry and turbulent ow. However, the rapid growth of computer hardware and software technology during the last couple of decades makes it possible to be able to use it as a supplement for experimentally impossible and complex problems, by using a computational uid dynamics (CFD) program. The generally used commercial codes based on FEM and FVM, such as ANSYS/FLUENT, ANSYS CFX, ADINA, ANSYS/FLOTRAN, STAR-CD, COMSOL, FLOW3D, FIDAP and CFD-ACE, are widely used for simulating engineering uid ow due to accuracy, robustness and convenience. Ansys CFX and Ansys Fluent, one of famous commercial CFD software packages, is based on a nite volume method approach. These codes differ mainly in the way they integrate uid ow equations and in their equation solution strategies. Ansys CFX solver uses cell-vertex nite volumes to discretize the domain. In vertex based schemes, the ow variables are stored at the vertices of the mesh elements [16]. In contrast, the Ansys Fluent solver uses cell-centered nite volumes. In cell centered schemes, the ow variables are stored at the centers of the mesh elements [8]. Ansys CFX focuses on one approach to solve the governing equations of motion (the coupled algebraic multigrid), whereas Ansys Fluent offers several solution approaches (density-based as well as segregated and coupled pressure-based methods). These programs are the most popular commercial packages available for most engineering elds. ADINA is a commercial code that can analyze uid ow related problems using nite element/nite volume codes in one single program. In calculating the uid ow, ADINA is supported by three schemes a nodal based Finite element method (FEM) and Flowcondition-based-interpolation (FCBI), and a cell based FCBI-C. FCBI was created to complement the problems of normal FEM. Therefore, ADINA is called FEM software that mainly uses a nodal based FCBI. The FCBI scheme is a nite element method, since the domain of the problem is considered as an assemblage of discrete nite elements connected at nodal points on the element boundaries, and the velocity and the pressure are interpolated within each element [1722]. This scheme also considered as a nite volume

method, since the weak form of NavierStokes equations are satised over the control volume. Hence the FCBI nite element solution satises mass conservation and momentum conservation for the control volumes (in contrast, the traditional nite element methods do not satisfy mass and momentum conservations) [23]. The main reason the FCBI scheme was proposed as a hybrid of the nite element and nite volume methods, not merely a nite volume method, is that the interpolation functions are not dened in the nite volume methods. The FCBI-C scheme specicity is that all the solution variables are dened at the center of the element, and the coupling between the velocity and the pressure is treated iteratively. The most prominent feature of this program is that solid models created in ADINA can be coupled with any uid model for analyzing the uid structure interaction. Currently, many researchers are using this approach to analyze the motion of the solid on a ow using ADINA software. Commercial CFD codes are used widely in various elds, such as aerospace engineering, automotive engineering, construction, energy, healthcare, chemical engineering, MEMS, and so on. However, only a few published studies have compared commercial CFD codes based on FEM/FVM. OCallaghan et al. [24] compared FVM (FLUENT), FEM (ADINA) and the theoretical predictions of blood ow through an idealized femoral artery. Their results showed that FVM and FEM were qualitatively similar but quantitatively different from the theoretical solutions, with FVM providing a better approximation to theory. Geller et al. [25] analyzed the accuracy and computational efciency of a research simulation code based on the FEM for twodimensional incompressible laminar ow problems with complex geometries, compared to the commercial code CFX based on the FVM. They observed lower relative errors in the simulation with the FEM of temperature and drag coefcient for the transient case. Ludwig and Dziak [26] compared the two commercial codes FEM (COMSOL) and FVM (FLUENT) for the simulation of isothermal laminar ow of liquid falling lm. They showed that the relative error of the FEM was signicantly higher, and the FVM signicantly reduces the time of the calculations comparing to FEM. Molina-Aiz et al. [27] simulated natural ventilation in greenhouses for the purpose of the comparison of FEM software (ANSYS/FLOTRAN) and FVM software (ANSYS/FLUENT). They observed that on average, the FEM required twice as much computing time per cell and per step compared to FVM, and that the amount of required memory storage was approximately 10 times greater for the FEM. They showed that the calculation time of the Fluent application was signicantly reduced and the operational memory similarly reduced. Even though these programs are used to simulate the engineering ow, studies comparing the results among commercial programs are rare.

W. Jeong, J. Seong / International Journal of Mechanical Sciences 78 (2014) 19 26

21

The alternation of ow patterns in a complex geometry could affect the uid dynamic factors, such as wall shear stress, velocity, pressure and ow rate. The aim of this study is to compare the alternation of the ow patterns and ow rates in commercial CFD

codes based on FEM (ADINA 8.6) and FVM (ANSYS CFX 11 and Fluent 6.3) under the same mesh and boundary conditions, using simple pipe and bifurcation models.

2. Methods 2.1. Model geometries In order to compare ow rates and ow patterns in the simulations of the three commercial CFD software packages, we modeled two models simple pipe and bifurcation (Y-shaped) models (Fig. 2). There are a variety of commercial 3D computer aided design (CAD) tools to generate model geometries SolidWorks, ProEngineer, CATIA, etc. These software tools enable one to save their work in many widely used le formats IGES, STL, Parasolid, and STEP. Therefore, we constructed all the models by the commercial 3D CAD software, SolidWorks. The length of the simple pipe used was 150 mm in accordance with the equation of the fully developed ow (Le/D 0.06Re, Le: length, D: diameter, Re: Reynolds number) at laminar, incompressible ow [28]. In the bifurcation model, the diameters of the inlet, right and left exit were 7 mm, 6.2 mm and 5.2 mm, respectively. The bifurcation angle was 771. The total length from an inlet to an outlet was 89 mm. 2.2. Mesh generation Mesh generation is the process of dividing the computational domain. The kinds of mesh are hexahedrons, tetrahedrons,

Fig. 2. (a) Geometry of the pipe and (b) the Y-shaped model.

Fig. 3. Computational meshes (a) hexahedron mesh of the pipe, (b) tetrahedron mesh of the pipe, (c) mixed mesh (hexaheron and tetrahedron mesh) of the bifurcation model, and (d) tetrahedron mesh of the bifurcation model.

22

W. Jeong, J. Seong / International Journal of Mechanical Sciences 78 (2014) 19 26

polyhedrons, prisms and pyramids. Hexahedron mesh has good qualities, but it has limitations in representing complex geometries. Tetrahedron mesh can easily represent complex geometries, but accuracy is lacking comparing to hexahedron mesh. Recently, the accuracy of tetrahedron mesh, however, has improved, and many researchers use a lot for representing complex geometries. In order to compare the effect on the mesh type, the mesh of a simple pipe model was composed with hexahedron and tetrahedron mesh by a commercial mesh generate program ICEM-CFD and Gambit, which can convert the mesh le to the ADINA, Ansys CFX and Ansys Fluent formats. The mesh of a bifurcation model was composed with tetrahedron and mixed mesh (tetrahedron and hexahedron). The composed mesh was imported into commercial CFD codes for analysis. Fig. 3 showed two different mesh models for pipe and bifurcation models. 2.3. Governing equation The numerical uid mechanical study was based on the continuity and momentum equations. The continuity and momentum equations were expressed as: ! U v 0 t ! !! ! ! v U v v p U g F t 1 2

Umax: maximum velocity, Uave: average velocity, r: radius (0 r r r R). The reference pressure of the outlet boundary conditions is applied to the atmospheric pressure in all models. The boundary conditions of ADINA, Ansys CFX and Fluent were the same for all the simulation models. ADINA and Fluent can both apply a 2-dimensional axisymmetric boundary condition, whereas Ansys CFX cannot apply this. Therefore, we solved for a 3-dimensional ow for all models. In addition, the ow within complex geometries is difcult to represent using a 2-dimensional ow. The pipe model was composed of a 64,000 hexahedron element mesh and a 577,000 tetrahedron element mesh. The bifurcation model was composed of a 78,900 mixed element mesh and a 372,000 tetrahedron element mesh. Further renement of the meshes does not affect the results of the numerical calculations. Computations were performed using a desktop computer equipped with a 2.26 GHz CPU and 2 GB of RAM, and a Dell Precision 670 computer equipped with a 3.8 GHz CPU and 4 GB of RAM.

3. Results We simulated the ow elds to observe the effect on the numerical scheme of the FEM and FVM under the same mesh for a simple pipe and Y-shape as shown in Fig. 3. The ow rates and ow patterns were calculated for a steady ow simulation. In order to compare the result between the FEM and FVM CFD codes, we used the same post-processing program (Ensight). 3.1. Pipe model In order to compare with the theory of a fully developed ow, we did a calculation to apply the uniform velocity prole at an inlet for models composed with a tetrahedron mesh and a hexahedron mesh. We observed that the maximum velocity at the center of a pipe depends on the distance in all CFD software packages. However, CFD codes of the FEM were not fully developed in the tetrahedron mesh compared with the hexahedron mesh and other solutions, as shown in Fig. 4. The CFD codes of the FVM agreed well with the mathematical calculations in the tetrahedron and hexahedron mesh (Table 3), whereas the FEM codes signicantly decreased by about 15% compared with the mathematical calculations in the tetrahedron mesh (Table 2). For ADINA FCBI-C, the number of maximum iterations of the velocity pressuretemperature (VPT) loop and the velocitypressure loop within the VPT loop had to be increased to obtain the approximate value when compared with the theoretical value. The ow rate of an inlet and an outlet on applied velocity at an inlet showed that hexahedron mesh was little bit higher than math, whereas tetrahedron mesh was little bit lower than math. The result values of a cross-sectional plane, such as average values and ow rate, could be inuenced by numbers of meshes, overall mesh quality, and mesh density near the wall. However, the measured ow rate showed less than 5% error among all CFD software packages in this study (Tables 69). A fully developed velocity condition agreed well in all codes regardless of mesh type. To measure the accurate

! where p is the static pressure, is the stress tensor, and g is the ! gravitational body force, F is the external body forces. The stress tensor is given by

2 ! ! v v U v I 3 !T

where is the molecular viscosity, I is the unit tensor, and the second term on the right hand side is the effect of volume dilation, which is zero for an incompressible ow. 2.4. Numerical methods We used the material property of an incompressible, homogeneous and Newtonian uid with a viscosity of 0.0035 Pa s and a density of 1060 kg/m3, and the ow was assumed to be laminar. Continuity and NavierStokes equations are solved for the uid. No-slip boundary conditions are imposed at the walls for all models. We solved steady state problems for all models. Table 1 summarizes the boundary conditions. We applied a uniform velocity prole and fully developed velocity prole at the inlet boundary in a pipe model. The mean velocity of an inlet boundary was 0.16 m/s under steady state conditions. The Reynolds number was set to 297 with a ow rate of 6.15 mL/s. An inlet boundary of the bifurcation model applied fully developed velocity prole. The equation for the fully developed velocity prole was expressed as follows [29]:   r 2  U max 2U ave 1 4 R
Table 1 The boundary conditions of CFD simulation. Ansys CFX Space Fluid type Flow type Discretization method Inlet boundary condition

Ansys Fluent

ADINA FCBI-C

ADINA FCBI

3-Dimensional ow Incompressible Laminar FVM (cell-vertex) FVM (cell-centered) Velocity (uniform/fully developed)

FVM (cell-centered)

FEM (nodal-based)

W. Jeong, J. Seong / International Journal of Mechanical Sciences 78 (2014) 19 26

23

Table 6 Comparison volume ow rate of the inlet and outlet with mathematical volume ow rate in hexahedron mesh of the pipe under uniform velocity condition. Dist (mm) Inlet/Math (%) Outlet/Inlet (%) Ansys CFX 0.7 0.2 Fluent 3.4 2.8 ADINA FCBI-C 3.0 0.8 ADINA FCBI 3.0 0.1

Math: 6.15 mL/s (Q AV where A: area, V: velocity) at the inlet

Table 7 Comparison volume ow rate of the inlet and outlet with mathematical volume ow rate in tetrahedron mesh of the pipe under uniform velocity condition. Dist (mm) Inlet/Math (%) Outlet/Inlet (%) Ansys CFX 0.2 0.9 Fluent 1.2 1.0 ADINA FCBI-C 0.2 5.2 ADINA FCBI 0.2 1.5

Math: 6.15 mL/s (Q AV where A: area, V: velocity) at the inlet. Fig. 4. Velocity Contour of uniform ow in a pipe. (A-1) Ansys CFX with hexahedron mesh, (A-2) Ansys CFX with tetrahedron mesh, (B-1) Ansys Fluent with hexahedron mesh, (B-2) Ansys Fluent with tetrahedron mesh, (C-1) ADINA FCBI-C with hexahedron mesh, (C-2) ADINA FCBI-C with tetrahedron mesh, (D-1) ADINA FCBI with hexahedron mesh, and (D-2) ADINA FCBI with tetrahedron mesh.

Table 8 Comparison volume ow rate of the inlet and outlet with mathematical volume ow rate in hexahedron mesh of the pipe under fully developed velocity condition. Dist (mm) Ansys CFX 0.4 1.3 Fluent 0.5 0.5 ADINA FCBI-C 1.4 0.0 ADINA FCBI 2.0 0.0

Table 2 The relative errors of between maximum velocity of the outlet and average velocity of the inlet under uniform velocity condition in tetrahedron mesh of the pipe. Velocity (m/s) Uout Uout/2Uave (%) Ansys CFX 0.311 2.7 Fluent 0.300 6.2 ADINA FCBI-C 0.290 9.4 ADINA-FCBI 0.271 15.3

Inlet/Math (%) Outlet/Inlet (%)

Math: 6.15 mL/s (Q AV where A: area, V: velocity) at the inlet.

Table 9 Comparison volume ow rate of the inlet and outlet with mathematical volume ow rate in tetrahedron mesh of the pipe under fully developed velocity condition. Dist (mm) Inlet/Math (%) Outlet/Inlet (%) Ansys CFX 0.5 0.2 Fluent 0.9 0.2 ADINA FCBI-C 2.6 0.6 ADINA FCBI 0.5 0.0

Uout: maximum velocity of the outlet, Uave: averaged velocity of the inlet

Table 3 The relative errors of between maximum velocity of the outlet and average velocity of the inlet under uniform velocity condition in hexahedron mesh of the pipe. Velocity (m/s) Uout Uout/2Uave (%) Ansys CFX 0.322 0.6 Fluent 0.311 2.8 ADINA FCBI-C 0.308 3.8 ADINA FCBI 0.314 1.7

Math: 6.15 mL/s (Q AV where A: area, V: velocity) at the inlet.

Uout: maximum velocity of the outlet, Uave: averaged velocity of the inlet

Table 4 The relative errors of between maximum velocity of the outlet and average velocity of the inlet under fully developed velocity condition in tetrahedron mesh of the pipe. Velocity (m/s) Uout Uout/2Uave (%) Ansys CFX 0.313 2.0 Fluent 0.302 5.6 ADINA FCBI-C 0.297 7.3 ADINA FCBI 0.265 17.0

Uout: maximum velocity of the outlet, Uave: averaged velocity of the inlet

Table 5 The relative errors of between maximum velocity of the outlet and average velocity of the inlet under fully developed velocity condition in hexahedron mesh of the pipe. Velocity (m/s) Uout Uout/2Uave (%) Ansys CFX 0.328 2.5 Fluent 0.318 0.7 ADINA FCBI-C 0.320 0.1 ADINA FCBI 0.331 3.5

Fig. 5. Velocity contour of a fully developed ow in a pipe. (A-1) Ansys CFX with hexahedron mesh, (A-2) Ansys CFX with tetrahedron mesh, (B-1) Ansys Fluent with hexahedron mesh, (B-2) Ansys Fluent with tetrahedron mesh, (C-1) ADINA FCBI-C with hexahedron mesh, (C-2) ADINA FCBI-C with tetrahedron mesh, (D-1) ADINA FCBI with hexahedron mesh, and (D-2) ADINA FCBI with tetrahedron mesh.

Uout: maximum velocity of the outlet, Uave: averaged velocity of the inlet

averaged value the mesh needs to be made dense near the wall in the ADINA CFD code, because ADINA does not interpolate between the wall node and the next node at the cross-sectional area. When we applied the fully developed velocity, a center maximum

24

W. Jeong, J. Seong / International Journal of Mechanical Sciences 78 (2014) 19 26

velocity decreased rapidly more than 17% in the FEM codes comparing with the FVM codes in the tetrahedron mesh (Table 4), whereas a center maximum velocity in the hexahedron mesh was delivered well to an outlet in all models (Table 5). In case of uniform velocity condition, a center maximum velocity in the FVM codes was developed well, whereas a center maximum velocity of the FEM codes with tetrahedron mesh decreased signicantly, as shown Fig. 4. In case of fully developed velocity prole, a center maximum velocity should be delivered to an
Table 10 Comparison to calculation time on commercial CFD codes with tetrahedron mesh in a pipe. Ansys CFX Time 28 min Fluent 30 min ADINA FCBI-C 42 min 34 s ADINA FCBI 2 h 40 min

outlet. Fig. 5 showed that a center maximum velocity in hexahedron mesh was delivered well in all codes, whereas a center maximum velocity of FEM codes with tetrahedron mesh were decreased signicantly. The calculation time was about 5 times greater for FEM compared to FVM (Tables 10 and 11).

3.2. Bifurcation model We simulated a steady ow to observe the ow pattern within a complex geometry. We observed that volume ow rate at the main branch and daughter branch was similar irrespective of the CFD codes and mesh types. Fig. 6(a) shows that the FEM codes did not transfer the maximum velocity well to the bifurcation in the tetrahedron mesh. However, the FVM codes were able to transfer the maximum velocity with the tetrahedron mesh and mixed mesh (hexahedron mesh and tetrahedron mesh). The ow pattern near the bifurcation showed that the FEM codes were not able to determine the recirculation ow in the tetrahedron mesh, whereas FVM codes were able to determine the recirculation ow, as shown in Fig. 6(a). The simulations using the mixed mesh were able to determine the recirculation ow for all tools (Fig. 6(a)). When we observed the ow pattern at the cross-sectional area near the bifurcation, mixed mesh showed ow vorticity in the results from all the methods. However, the FEM codes with the tetrahedron mesh were signicantly weak as shown in Fig. 6(b and c).

Table 11 Comparison to calculation time on commercial CFD codes with hexahedron mesh in a pipe. Ansys CFX Time 5 min 43 s Fluent 3 min 43 s ADINA FCBI-C 6 min 56 s ADINA FCBI 2 h 36 min 24 s

Fig. 6. (a) Velocity contour of the Y-shaped model, (b) Secondary ow of the cross-sectional area in the left daughter branch, (c) Secondary ow of the cross-sectional area in the right daughter branch. (A-1) Ansys CFX with mixed mesh, (A-2) Ansys CFX with tetrahedron mesh, (B-1) Ansys Fluent with mixed mesh, (B-2) Ansys Fluent with tetrahedron mesh, (C-1) ADINA FCBI-C with mixed mesh, (C-2) ADINA FCBI-C with tetrahedron mesh, (D-1) ADINA FCBI with mixed mesh, and (D-2) ADINA FCBI with tetrahedron mesh.

W. Jeong, J. Seong / International Journal of Mechanical Sciences 78 (2014) 19 26

25

4. Discussions The ow pattern in the pipe is changed depending on the length and bending angle of the pipe, and various branch shapes. Various ow patterns, such as recirculation ows, retrograde ows, secondary ows and vortex ows, affect the pipe due to various uid dynamic factors. Fluid dynamic factors, such as pressure, wall shear stress, the ow velocity and particle residence time, could do a lot of damage to various parts, e.g. to pumps, pipes, microchannels and blood vessels. However, there are many difculties in measuring these uid dynamic factors experimentally. Recently, uid dynamic factors have been calculated in various elds due to the development of much commercial CFD software. For the accurate calculation of uid ow, various mesh types, such as hexahedron, tetrahedron, prism and polygon, are used. The results of the hexahedron mesh are similar to the results of the theory, compared to the results from the other mesh types [30]. However, the hexahedron mesh is difcult to use for representing for complex geometries. Thus, the tetrahedron mesh, which is able to easily represent complex geometries, is used a lot. Sometimes tetrahedron meshes, however, produce less accuracy [31]. Our results showed that the FVM codes were most unaffected by the mesh types, whereas the FEM codes agreed well with hexahedron mesh, but these one with tetrahedron mesh could not represent for a fully developed ow. In addition, the complex ow pattern near bifurcation could not be represented. Previous studies using a FEM tools developed by user obtained good results, whereas, studies using a commercial FEM code were usually bad comparing to a commercial FVM code. Even though the FEM procedure in ADINACFD is not commonly used as much, but it has been occasionally used for educational purpose. However, ow analyses of 2D geometry agreed well in all of the FEM and the FVM codes. Most companies of commercial CFD codes is still upgrading and developing their CFD codes to solve the realistic phenomena of the uid ow. Local ow patterns, such as recirculation ows, retrograde ows and secondary ows, play an important role in the ow dynamic effects at a bifurcation or a diffuser. The ow pattern of bifurcation models have been studied by many previous researchers in various elds [3235]. Studies of the effects on ow patterns in various engineering elds using commercial CFD software have validated the use of this software, and they are still being studied. CFX, Fluent and ADINA are used a lot in various elds. In particular, ADINA is used a lot to calculate the effect of the blood ow on the elasticity of the blood vessel wall [3638], because this software can calculate uid structure interactions (FSIs). CFX and Fluent was used to calculate the ow characteristics of the diffuser and bifurcation models, and it was validated [3941]. This study did not mention for turbulent ow, but previous studies investigated the turbulent ow in diffuser and bifurcation models using commercial CFD codes [4246]. FEM code developed by user showed good results [13,47,48]. A commercial FEM code (ADINA) calculated stenosis and bifurcation models [36,38,49,50]. The study of stenosis model used ne mesh and explored the ow pattern, but the study of bifurcation model did not explore the ow characteristics. Our results have observed that FEM codes with the tetrahedron mesh had trouble representing the ow pattern. There was no trouble in using the hexahedron mesh. Simple geometry can compose easily as ne mesh. However, 3dimensional complex geometry is difcult to compose as ne mesh. As for the result of this study, FEM is supposed to be important to make ne mesh comparing to FVM. The errors reported here are hence at least partly due to the postprocessing used in Ensight, for example if the post-processing of ADINA is used, the mass ow rates are frequently much more accurate.

In previous studies, the computing time in an FEM was observed to be generally greater than in FVM [13,14,27]. Accordingly, in our study, the computing time consumed was signicantly greater in FEM. This study used a computer of low specication. Recently, the development of operating system of a computer upgraded the specication of a computer. The specication of a computer plays a signicant role in the calculation of geometry with numerous meshes it can affect the results, especially FCBI solution. However, the specication of the computer employed in this study does not affect the accuracy of the results because this study uses a very simple geometry with few meshes. Most recent studies use many meshes, sometimes over millions of meshes. In such cases, the specication of a computer should be considered in CFD solution, especially FCBI solution. Moreover, each CFD software package has a different default convergence tolerance. Alteration of the convergence tolerance values can affect the calculation time and the result. Convergence tolerance values are normally set to default values, but modication of these values depends heavily on the decision of the CFD user. This computational study has several limitations. The chief limitation of this study is that it uses specic commercial codes and versions (ADINA, Ansys CFX, Fluent) that are based on FEM and FVM. These codes are the most widely used codes in the industry. ADINA is popular in the biouid eld because it supports uid structure interaction analysis [5153]. Ansys CFX and Fluent are in use in various engineering elds. Ansys CFX and Fluent have been compared with user-developed FEM codes or commercial FEM codes [13,27], but these studies were rare. ADINA has never been compared. Therefore, ADINA needs to be compared with Ansys CFX and Fluent, which are the typical commercial FVM codes, because many engineers use these codes. Besides, the version of commercial CFD codes is upgraded every year. However, important ow analysis algorithm barely incur changes, while incidental functions, such as turbulent model, some function and graphic tool are commonly upgraded. The results obtained from the current version (ADINA 8.9, Ansys CFX 14, Fluent 14) are similar to those obtained from their previous version (ADINA 8.6, Ansys CFX 11, Fluent 6.3). Simple ow (laminar and steady ow) is another limitation of this computational study. This study did not calculate n for turbulent and unsteady ow. Turbulent ow is a challenging eld to analyze in a CFD. Comparison between CFD codes for turbulent ow is difcult because CFD has many turbulence models and near-wall modeling, such as Reynolds-averaged NavierStokes (RANS) based models, large eddy simulation (LES), detached eddy simulation (DES), direct numerical simulation (DNS), and other hybrid models. However, simple uid ow pattern, such as recirculation ow and secondary ow, can be represented using simple geometry, such as a bifurcation model. The ow pattern of unsteady ow is also similar to that of steady ow. Because ow patterns of these CFD codes (ADINA, Ansys CFX and Fluent) using simple geometry and ow have never been compared, the results of ows calculated by these CFD codes need to be examined. Examining the turbulent ow for various turbulent models should be considered in future studies. The version of a commercial code is another limitation of this study. The version of commercial codes is upgraded every year. Therefore, commercial codes are still under development and complementing in their company. However, the fundamental algorithm of a code undergoes minimal changes. Despite these limitations, the results of this study may provide the characteristic of these commercial codes to engineers.

5. Conclusions In this study, we compared commercial CFD software based on the nite element (FE) and nite volume (FV) techniques. Although the governing equations and boundary conditions were

26

W. Jeong, J. Seong / International Journal of Mechanical Sciences 78 (2014) 19 26 [26] Ludwig W, Dziak J. CFD modelling of a laminar lm ow. Chem Process Eng 2009;30:41730. [27] Molina-Aiz FD, Fatnassi H, Boulard T, Roy JC, Valera DL. Comparison of nite element and nite volume methods for simulation of natural ventilation in greenhouses. Comput Electron Agric 2010;72:6986. [28] White FM. Fluid mechanics. 6th ed.. New York: McGraw-Hill; 2008. [29] Oosthuizen PH, Naylor D. An introduction to convective heat transfer analysis, WCB/. New York: McGraw Hill; 1999. [30] Wieners C. Conforming discretizations on tetrahedrons, pyramids, prisms and hexahedrons, Preprint. University of Stuttgart; 1998. [31] Biswas R, Strawn RC. Tetrahedral and hexahedral mesh adaptation for CFD problems. Appl Numer Math 1998;26:13551. [32] Balashazy I, Hofmann W, Heistracher T. Computation of local enhancement factors for the quantication of particle deposition patterns in airway bifurcations. J Aerosol Sci 1999;30:185203. [33] Lee YK, Shih C, Tabeling P, Ho CM. Experimental study and nonlinear dynamic analysis of time-periodic micro chaotic mixers. J Fluid Mech 2007;575: 42548. [34] Nandakumar K, Masliyah JH, Law H. Bifurcation in steady laminar mixed convection ow in horizontal ducts. J Fluid Mech 1985;152:14561. [35] Zhao SZ, Xu XY, Hughes AD, Thom SA, Stanton AV, Ariff B, et al. Blood ow and vessel mechanics in a physiologically realistic model of a human carotid arterial bifurcation. J Biomech 2000;33:97584. [36] Kaazempur-Mofrad MR, Isasi AG, Younis HF, Chan RC, Hinton DP, Sukhova G, et al. Characterization of the atherosclerotic carotid bifurcation using MRI, nite element modeling, and histology. Ann Biomed Eng 2004;32:93246. [37] Rissland P, Alemu Y, Einav S, Ricotta J, Bluestein D. Abdominal aortic aneurysm risk of rupture: patient-specic FSI simulations using anisotropic model. J Biomech Eng 2009;131:031001. [38] Tang D, Yang C, Ku DN. A 3-D thin-wall model with uidstructure interactions for blood ow in carotid arteries with symmetric and asymmetric stenoses. Comput Struct 1999;72:35777. [39] Liu H, Li P, Lew JV. CFD study on ow distribution uniformity in fuel distributors having multiple structural bifurcations of ow channels. Int J Hydrogen Energy 2010;35:918698. [40] Long Q, Xu XY, Ariff B, Thom SA, Hughes AD, Stanton AV. Reconstruction of blood ow patterns in a human carotid bifurcation: a combined CFD and MRI study. J Magn Reson Imaging 2000;11:299311. [41] Triesch O, Bohnet M. Measurement and CFD prediction of velocity and concentration proles in a decelerated gassolids ow. Powder Technol 2001;115:10113. [42] Abraham JP, Sparrow EM, Tong JCK, Bettenhausen DW. Internal ows which transist from turbulent through intermittent to laminar. Int J Therm Sci 2010;49:25663. [43] Banks J, Bressloff N. Turbulence modeling in three-dimensional stenosed arterial bifurcations. J Biomech Eng 2007;129:4050. [44] Iaccarino G. Predictions of a turbulent separated ow using commercial CFD codes. Trans ASME J Fluids Eng 2001;123:81928. [45] Marshall I, Zhao S, Papathanasopoulou P, Hoskins P, Xu XY. MRI and CFD studies of pulsatile ow in healthy and stenosed carotid bifurcation models. J Biomech 2004;37:67988. [46] Varghese SS, Frankel SH. Numerical modeling of pulsatile turbulent ow in stenotic vessels. Trans ASME J Fluids Eng 2003;125:44560. [47] Gullman-Strand J, Trnblom O, Lindgren B, Amberg G, Johansson AV. Numerical and experimental study of separated ow in a plane asymmetric diffuser. Int J Heat Fluid Flow 2004;25:45160. [48] Liu X, Guang X, Wang S. FEM calculation and experimental study of the 3-D ow in vane diffuser. J Aer P 1999;14:1314. [49] Tang D, Teng Z, Canton G, Yang C, Ferguson M, Huang X, et al. Sites of rupture in human atherosclerotic carotid plaques are associated with high structural stresses: an in vivo MRI-based 3D uid-structure interaction study. Stroke 2009;40:325863. [50] Younis HF, Kaazempur-Mofrad MR, Chan RC, Isasi AG, Hinton DP, Chau AH, et al. Hemodynamics and wall mechanics in human carotid bifurcation and its consequences for atherogenesis: investigation of inter-individual variation. Biomech Model Mechanobiol 2004;3:1732. [51] Borghi A, Wood NB, Mohiaddin RH, Xu XY. Computational analysis of ow and stress patterns in patient specic thoracic aortic aneurysm models. PatientSpecic Comput Model 2012:13359. [52] Liu A, Nickerson A, Troyer A, Yin X, Cary R, Thornburg K, et al. Quantifying blood ow and wall shear stresses in the outow tract of chick embryonic hearts. Comput Struct 2011;89:85567. [53] Moosavi M-H, Fatouraee N, Katoozian H, Pashaei A, Camara O, Frangi AF. Numerical simulation of blood ow in the left ventricle and aortic sinus using magnetic resonance imaging and computational uid dynamics. Comput Methods Biomech Biomed Eng 2012:110.

the same in all CFD codes, the results of the interpolation schemes of each CFD code were varied slightly. FVM software (CFX and Fluent) showed similar results regardless of the mesh type hexahedron or tetrahedron meshes. However, the most commonly used FCBI based on FEM and one of ADINA uid modules was inuenced by the mesh type used, whereas FCBI-C based on FVM showed similar results for different mesh types. The calculation time for the same number of meshes was approximately 5 times greater for the FEM-based module as compared to the FVM based module. Consequently, commercial FEM CFD codes would need to consider the mesh type, quality, and number of mesh elements unlike the FVM CFD codes. References
[1] Richardson L. The approximate arithmetical solution by nite differences of physical problems involving differential equations, with an application to the stresses in a masonry dam. Philos Trans R Soc A 1911;210:30757. [2] Thom A. The ow past circular cylinders at low speeds. Proc R Soc A 1933;141:65169. [3] Courant R, Friedrichs K, Lewy H. ber die partiellen Differenzengleichungen der mathematischen Physik. Math Ann 1928;100:3274. [4] Southwell RV. Relaxation methods in engineering science. London, UK: Oxford University Press; 1940. [5] Von Neumann J, Richtmyer R. A method for the numerical calculation of hydrodynamic shocks. J Appl Phys 1950;21:2327. [6] Shang J. Three decades of accomplishments in computational uid dynamics. Prog Aerosp Sci 2004;40:17397. [7] Runchal AK. Brian spalding: CFD & reality. Int J Heat Mass Transfer 2008;10. [8] Fluent I. FLUENT user's guide. Fluent Inc; 2003. [9] Peiro J, Sherwin S. Finite difference, nite element and nite volume methods for partial differential equations. Handbook of materials modeling. 241546. [10] Ferziger JH, Peric M. Computational methods for uid dynamics. 3rd ed., New York: Springer; 2002. [11] Kumar A. CFD modeling of gasliquidsolid uidized bed. Rourkela, India: B. Tech, NIT; 2009. [12] Ahmad N, Rappaz J, Desbiolles JL, Jalanti T, Rappaz M, Combeau H, et al. Numerical simulation of macrosegregation: a comparison between nite volume method and nite element method predictions and a confrontation with experiments. Metall Mater Trans A 1998;29:61730. [13] Gohil T, McGregor RHP, Szczerba D, Burckhardt K, Muralidhar K, Szkely G. Simulation of oscillatory ow in an aortic bifurcation using FVM and FEM: a comparative study of implementation strategies. Int J Numer Methods Fluids 2010;66:103767. [14] Lukcov-Medvidov M, Teschke U. Comparison study of some nite volume and nite element methods for the shallow water equations with bottom topography and friction terms. ZAMM J Appl Math Mech/Zeitschrift fr Angewandte Mathematik und Mechanik 2006;86:87491. [15] E Onate, SR Idelsohn. A comparison between nite element and nite volume methods in CFD. In: Proceedings of the European computational uid dynamics conference'92; 1992. p. 93100. [16] A.C.F.X., Ansys CFX-Solver Theory Guide, Release, 11; 2006. [17] Bathe KJ, Ledezma GA. Benchmark problems for incompressible uid ows with structural interactions. Comput Struct 2007;85:62844. [18] Bathe KJ, Pontaza JP. A ow-condition-based interpolation mixed nite element procedure for higher Reynolds number uid ows. Math Models Methods Appl Sci 2002;12:52540. [19] Bathe KJ, Zhang H. A ow-condition-based interpolation nite element procedure for incompressible uid ows. Comput Struct 2002;80:126777. [20] Kohno H, Bathe KJ. Insight into the ow condition based interpolation nite element approach: solution of steady state advection-diffusion problems. Int J Numer Methods Eng 2005;63:197217. [21] Kohno H, Bathe KJ. A nine node quadrilateral FCBI element for incompressible uid ows. Commun Numer Methods Eng 2006;22:91731. [22] Kohno H, Bathe KJ. A ow condition based interpolation nite element procedure for triangular grids. Int J Numer Methods Fluids 2006;51:67399. [23] Banijamali B, Bathe KJ. The CIP method embedded in nite element discretizations of incompressible ows. Int J Numer Methods Eng 2007;71:6680. [24] S OCallaghan, M Walsh, T McGloughlin. Comparison of nite volume, nite element and theoretical predictions of blood ow through an idealized femoral artery. In: Proceedings of the 2003 Summer bioengineering conference; 2003. p. 2529. [25] Geller S, Krafczyk M, Tolke J, Turek S, Hron J. Benchmark computations based on lattice-Boltzmann, nite element and nite volume methods for laminar ows. Comput Fluids 2006;35:88897.

Das könnte Ihnen auch gefallen