- 1 - Comparison of different turbulence models to compute wall affected room airflows Comparison of different turbulence models to compute wall affected room airflows Ch. Heschl 1 , W. Sanz 2 , P. Klanatsky 1 and F. Madou 1 Fachhochschulstudiengnge Burgenland GmbH 1 Steinamangerstrae 21, A-7423 Pinkafeld TU Graz, Institute for Thermal Turbomachinery and Machine Dynamics 2 Inffeldgasse 25, A-8010 Graz Abstract It is well-known that the turbulence anisotropy has a remarkable influence on the flow distribution of three dimensional wall jets. So it can be expected that accurate simulations of room airflows with air supplies mounted just below the ceiling demands a higher level turbulence closure. In this paper detailed three dimensional PIV measurement results of the velocity distribution in a symmetrical model room with air inflow at an angle =15 towards the wall are presented. The experimental data are compared with the velocity distributions computed by different turbulence models. The potential of the improvement of the simulation by using different nonlinear eddy viscosity, pressure strain and hybrid Reynolds stress models is discussed. Key Words: v2f turbulence model, nonlinear v2f turbulence model, cubic pressure strain model, wall reflection effect, two component closure, TCL, hybrid RSM turbulence model, wall affected room airflows, UDF, PIV
Introduction One of the most used air supply systems are inlet grilles mounted just below the ceiling. An accurate prediction of their flow characteristics in ventilated rooms requires that the governing influence of Coanda- effect and anisotropic Reynolds stress induced secondary motion can be exactly modelled. In the last years both flow phenomena were discussed by different authors in separate investigations. For example, the Coanda-effect was discussed by Lai and Lu (1992 and 1996), Newman (1961) and Nasr and Lai (1998). The studies included CFD Forum 2005 Bad Nauheim, Deutschland - 2 - Comparison of different turbulence models to compute wall affected room airflows investigations about the mean velocity and the turbulence field for two dimensional offset jets. Three dimensional effects affected by anisotropic Reynolds stresses were not investigated. But this phenomenon was investigated for a three dimensional wall jet by Craft and Launder (2001), Lbcke et al. (2003) and Abrahamsson (1997). The room airflow induced by a wall jet without an offset mounted below the ceiling was discussed by Schlin and Nielsen (2004). The authors concluded that the spreading rate parallel to the surface is too small when the linear eddy viscosity concept is used to compute the turbulence quantities. It is also well known that three dimensional effects arise when the air inlet system is mounted with an offset from the ceiling and the side walls. For example this flow phenomenon was investigated by Heschl et al. (2002a and 2002b). The principle used in their experimental and numerical setup is shown in Figure 1.
Figure 1: Investigated full scale test room, inlet (purple) and outlet (yellow) In order to evaluate the thermal comfort and the indoor air quality with CFD simulations all primary and secondary flow mechanism must be reproduced. In this paper detailed three dimensional PIV measurement data of the velocity distribution in a symmetrical room are presented. The experimental data is compared with the velocity and turbulence distribution computed with many different turbulence models as presented by Heschl et al. (2005). Experimental setup The test bench for the experimental investigations consists of an air supply system with mass flow measurement facilities, a scale down model made of plexiglass, an air-conditioned test room and a PIV measurement system. The PIV system was assembled for two dimensional measuring. To record a larger measurement field two cameras were used. The geometrical data of the plexiglass model is described in the appendix. The experimental arrangement is also shown in the appendix (Figure 14). The test bench was equipped with a fan unit with frequency control, which takes the air over the orifice plate, rectifier, flow measurement unit, temperature sensors, seeding mixing chamber into the plexiglass test model. To get a better illumination the plexiglass model was implemented in a black coated test room. To minimize the thermal influences the laser energy supply and the computer equipment were placed outside of the test room. To ensure isothermal conditions the test room was additionally ventilated. The inlet air temperature, the outlet air temperature and the surface temperature of the surrounding walls of the plexiglass model were monitored by 10 Pt100 temperature sensors over the whole measurement time. Also the CFD Forum 2005 Bad Nauheim, Deutschland - 3 - Comparison of different turbulence models to compute wall affected room airflows ambient pressure and the inlet mass flow were monitored. The used plexiglass model and the traverse system are shown in Figure 2.
Figure 2: Plexiglass model The light sheet optic and the two cameras were mounted on a two dimensional traverse system, so that only one calibration for the PIV-measurements was necessary. The measured planes were defined in such way that the whole flow area in the x-y plane could be measured by changing the lightsheet and camera positions. The field of view of one camera was 155 x 200 mm, so that the complete height of the model room was measured in every position. In the z-direction the x and y velocity components were also measured by traversing the lightsheet and cameras in steps of 20 mm. The interrogation areas were defined with 32x32 pixels and an overlapping of 25%, so that in the x-y planes every 3.6 mm one velocity vector was measured and in the z-direction every 20mm. The inlet was defined by an air supply duct which consists of evenly distributed 222 holes with a diameter of 2mm. The supply duct was fed from both end sides over a flow rectifier (see Fig. 3). The geometrical data of the air supply duct are described in the appendix. The mean inlet velocity was 14.84 m/s (air change rate 84) and the inlet angle was 15 towards the ceiling wall.
Figure 3: Seeding mixing chamber and air supply duct In order to eliminate the fluctuations due to the turbulence the measuring data were averaged over a measuring time of 30 seconds with a sample rate of 4 Hz. For the accurate determination of the density during the measurement the temperatures, the relative humidity and the ambient pressure were logged. Numerical setup All computational results presented in this paper were obtained with the commercial CFD code FLUENT 6.1.21 by using the turbulence models described in Heschl et al. (2005). For the momentum equation and the transport equation of turbulence parameters the second order upwind discretization was applied.
CFD Forum 2005 Bad Nauheim, Deutschland - 4 - Comparison of different turbulence models to compute wall affected room airflows Table 1: Used turbulence models (cp. Heschl et al. 2005) turbulence model designation short cut remark Standard k model ske standard model in Fluent k model with enhanced wall treatment ske-et standard model in Fluent Reynolds stress model with linear pressure strain model and wall reflection term IP standard model in Fluent Reynolds stress model with linear pressure strain model, wall reflection term and enhanced wall treatment IP-et standard model in Fluent Reynolds stress model with quadratic pressure strain model SSG standard model in Fluent v2f model v2f in Fluent implemented via UDF v2f model modified by Davidson v2f-dav in Fluent implemented via UDF nonlinear v2f model by Pettersson Reif (1999) v2f-pet in Fluent implemented via UDF nonlinear v2f model by Pettersson Reif (1999), the v2f turbulence is modified by Davidson et al. (2003) v2f-pede in Fluent implemented via UDF nonlinear k model by Ehrhard (1999) ske-ehr in Fluent implemented via UDF nonlinear k model by Lbcke (2003) ske-lue in Fluent implemented via UDF Reynolds stress model with cubic pressure strain model CUBIC in Fluent implemented via UDF Reynolds stress model with two component closure pressure strain model TCL in Fluent implemented via UDF Reynolds stress model with linear pressure strain model, standard wall function and wall reflection effect model by Craft and Launder (1992) IP-cl in Fluent implemented via UDF Hybrid Reynolds stress model with linear pressure strain model and wall reflection term IP-hyb in Fluent implemented via UDF Hybrid Reynolds stress model with quadratic pressure strain model SSG-hyb in Fluent implemented via UDF Reynolds stress model with quadratic pressure strain model and and wall reflection effect model by Craft and Launder (1992) SSG-cl in Fluent implemented via UDF
The exact determination of the flow situation in the model room requires an accurate reproduction of the boundary conditions. One of the key parameters is the inlet flow. Because the inlet consists of 222 holes a simplification of the inlet geometry is necessary. In the present study the holes were replaced by an inlet slot of the same area. The mean inlet velocity distribution was determined using the CFD Forum 2005 Bad Nauheim, Deutschland - 5 - Comparison of different turbulence models to compute wall affected room airflows procedure described by Heschl et al. (2004) for the flow distribution in air supply ducts. For the computations two different grids a fine and a coarse grid were used. The coarse grid was designed with a 30 + y for turbulence models with standard wall functions. The first grid line had a wall distance of 1.5mm, and a grading ratio of 1 and a total number of 8 lines inside the boundary layer region (see Figure 4). The number of grid cells for the whole computational domain was about 600000.
Figure 4: Coarse grid (15 inlet angle) The fine grid was designed for the low-Re turbulence models (e.g. v2f turbulence model) with a 1 + y . In this case the first grid line had a wall distance of 0.25mm, a grading ratio of 1 and a total number of 30 lines inside the boundary layer region (see Figure 5). The number of grid cells for the whole computational domain was about 2000000. A hybrid mesh was used for the discretization of the computational domain. The inlet region was meshed with hexahedron cells and the main flow domain with tetrahedron cells. The used meshes at the inlet region are shown in the Figure 5 and Figure 4. Because a round inlet supply duct is used the inlet regions were meshed with coppered triangles. To achieve a reasonable skewness parameter it was necessary to mesh the boundary layer of the inlet regions differently from the rest of the computation domain.
Figure 5: Fine grid (15 inlet angle) At the end of the outlet channel a pressure outlet boundary condition was used. All presented results showed good convergence behaviour. But the nonlinear eddy viscosity turbulence models did not achieve the residual criteria for the continuity of 0.1%. No stationary solution were achieved for the coarse grid using the realizable (rke), rng, ske-ehr, ske-lue and the hybrid Reynolds stress turbulence models. The Reynolds stress models ske-lue and IP-et CFD Forum 2005 Bad Nauheim, Deutschland - 6 - Comparison of different turbulence models to compute wall affected room airflows did also not achieve stationary solutions for the fine grid. Results and discussion The incoming air forms a jet in the room and the ambient air is entrained by turbulent mixing processes (Fleischhacker, G. and Schneider, W. 1980). Because of the ceiling not enough air is entrained from above. In order to meet the continuity and momentum balance low pressure arises forcing the jet towards the ceiling (Coanda- effect). Due to this attachment a three- dimensional wall jet forms which is to a large extent reflected by the opposite wall. There is an interaction between the wall jet on the ceiling and the back flow air by a turbulent mixing process (entrainment). The distance s (see Figure 1) allows a side entrainment of the air which induces cross flows. On the one hand the entrainment occurs due to the turbulent mixing and on the other hand due to pressure differences. For the demonstration of this effect Figure 6 presents the velocity vectors computed by the SSG model (coarse grid) in a yz- plane at x/H=2.0).
Figure 6: Velocity vectors in the yz- plane x/ H=2.0 The solid line represents the isoline for the x-component of velocity Ux=0. It can be interpreted as a fictitious jet boundary. In this case the jet did not reach the floor. The air is forced by the entrainment process to recirculate near floor level and induce thereby secondary motion. This process is mainly influenced by turbulent mixing process which can well predicted by the linear eddy viscosity concept. But the secondary flow induced by the turbulent mixing process will be superimposed by the secondary motion caused by the stress induced axial vorticity. This axial vorticity is generated by the anisotropic Reynolds normal stresses and can only in contrast to the entrainment process be modelled with nonlinear eddy viscosity turbulence models and pressure strain models which ensure the wall reflection effect (cp. Heschl et al., 2005). The influence of the entrainment and normal shear stress induced secondary motion can be identified by the comparison of numerical results predicted with the linear eddy viscosity turbulence model and the IP-hyp Reynolds stress model. The IP-hyb model determines only the turbulent shear stresses with the Boussinesq assumption, whereas the turbulent normal stresses use transport equations (see Heschl et al. 2005). So the difference between both calculations is caused by the way the normal stresses are modelled. The so induced axial vorticity generates an additional secondary motion near the wall. For the investigated plexiglass model the computed contours of velocity magnitude are shown in Figure 7 and Figure 8. The inlet angle was defined with 15. The coarse grid with the standard k CFD Forum 2005 Bad Nauheim, Deutschland - 7 - Comparison of different turbulence models to compute wall affected room airflows turbulence model and with the RSM with hybrid IP pressure strain model was used to compute the velocity distribution.
Figure 8: Computed velocity magnitude (used turbulence model: I P-hyb) In Figure 7 and Figure 8 the differences in the velocity distributions between the linear eddy viscosity and the RSM turbulence model with the IP-hyb pressure strain model are obvious. Corresponding to the three dimensional wall jets the anisotropic turbulent normal stresses have an important influence on the flow prediction. Especially the penetration of the wall jet into the occupied zone is deeper predicted by the ske turbulence model. But in contrast to the investigation from Schlin and Nielsen (2004) the differences are not so significant. This can be explained by the different geometric dimensions of the used model as summarized in Table 2. Table 2: Confrontation of the geometrical parameters Parameter Schlin and Nielsen (2004) Presented investigation 0.0 15.0 b/B 0.2 0.8325 H M /H 1.0 0.9025 L/H 12.0 3.0 B/H 4.7 2.0 H 128mm 400mm The investigations by Schlin and Nielsen (2004) were based on a scale down model with a small inlet area directly mounted below the ceiling. Therefore the inlet air forms directly a three dimensional wall jet and a lateral distribution will be not inhibited by the side walls. In contrast the present study investigates an air supply duct which is mounted just below the ceiling with an inlet angle =15 . An essential difference can be expected by the ratio b/B. Because in this study the inlet air will be supplied nearly over the whole room width, the lateral spreading is inhibited by the side walls, so that the influence of the near wall redistribution effect will be reduced. To get further information about the computed flow distribution near the occupied zone the velocity magnitude in the y-z plane at x/H=2.0 is compared with experimental data in the Figure 9 to Figure 10. Also the secondary motion in the y-z CFD Forum 2005 Bad Nauheim, Deutschland - 8 - Comparison of different turbulence models to compute wall affected room airflows plane is demonstrated with velocity vectors in the Figure 11. All linear eddy viscosity models predicted too small lateral spreading rates and no correct secondary motions. Among them the linear v2f turbulence model provides the best result. Also the backflow behaviour near the floor and the side walls cannot be predicted correctly. Especially the deep penetration of the jet leads to too high velocities in the occupied zone and therefore to significant deviations. The nonlinear models proposed by Lbcke et al. (2003) and Ehrhard (1999) predicted similar results as the linear eddy viscosity models. In the comparison between the different pressure strain models the SSG model provides the best results. Also the predicted secondary motion agrees well with the experimental data, so that this model is the most promising one for further applications. The use of the SSG model with the wall reflection term by Craft and Launder (1992) shows a minor improvement. Though the cubic and the TCL models are higher order pressure strain models as the SSG model, they cannot achieve better results. The hybrid models deliver also better results as the linear and nonlinear eddy viscosity models. The additional computational effort for the two transport equations seems justified.
CFD Forum 2005 Bad Nauheim, Deutschland - 9 - Comparison of different turbulence models to compute wall affected room airflows Figure 9: Velocity magnitude in the y-z plane - x/ H=2.0 (inlet angle 15, coarse grid):
Experimental data
ske model
rke model
rng model
ske-ehr model
ske-lue model
IP model
IP-cl model
SSG model
SSG-hyb model
SSG-cl model
CUBI C model
0.0 m/ s 0.5 m/ s 1.0 m/ s 1.5 m/ s 1.8 m/ s CFD Forum 2005 Bad Nauheim, Deutschland - 10 - Comparison of different turbulence models to compute wall affected room airflows Figure 10: Velocity magnitude in the y-z plane - x/ H=2.0 (inlet angle 15, fine grid):
Experimental data
ske-et model
ske-lue model
IP-et model
SSG-hyb model TCL model CUBI C model
v2f model
v2f-pet model
0.0 m/ s 0.5 m/ s 1.0 m/ s 1.5 m/ s 1.8 m/ s CFD Forum 2005 Bad Nauheim, Deutschland - 11 - Comparison of different turbulence models to compute wall affected room airflows Figure 11: Principle y and z velocity vectors in the y-z plane - x/ H=2.0 (inlet angle 15, fine and coarser grid):
Experimental data
ske - coarse mesh
IP coarse mesh
I P-hyb coarse m.
SSG coarse mesh
SSG-hyb coarse m.
ske-ehr coarse m.
ske-lueb - coarse m.
ske-ewt fine mesh
v2f-lin. fine mesh
v2f-nonlin. - fine m.
TCL - fine mesh
CFD Forum 2005 Bad Nauheim, Deutschland - 12 - Comparison of different turbulence models to compute wall affected room airflows Summary Linear eddy viscosity turbulence models are widely used to predict air flows in ventilated rooms. In many air supply configurations an important interaction between the air flow and the surrounding walls arises. Especially the anisotropic Reynolds stresses near walls can produce secondary motions which obviously influence the room airflow behaviour. Hence, an extensive investigation about the accuracy of standard and higher order turbulence models was performed in order to test their capability in predicting this flow phenomenon. Several linear as well as nonlinear eddy viscosity models and higher order pressure strain as well as hybrid pressure strain models were implemented into Fluent via compiled User Defined Functions (UDF). A complex hybrid mesh was used to test the numerical stabilities. For the used grid the implemented models were stable but all nonlinear eddy viscosity models had poor convergence behaviour. Comparing the investigated linear and nonlinear turbulence models the v2f model agreed best with PIV measurements. If Reynolds stress turbulence models are used, the SSG pressure strain model provides the best results. The hybrid RSM models achieve similar good results with less computational effort, so that they are a promising alternative. The quality of the flow prediction of the SSG pressure strain model could not be improved using the cubic and two component closure models. It can be concluded, that the turbulence models available in Fluent are a good compromise between stability and accuracy. Acknowledge The experimental and numerical investigations were partly accomplished by the impulse program FHplus funded by the BMVIT and BMBWK. References Abrahamsson H. (1997): On Turbulent Wall J ets. PhD thesis, Department of Thermo and Fluid Dynamics, Chalmers University of Technology, Gteburg Sweden. ISBN 91-7197-491-1 Heschl Ch., Sanz W., Kelz A. (2004): Validation of determined pressure loss and discharge coefficient of air supply ducts. FLUENT Anwenderkonferenz 2004, 29.-30. September 2004 in Bingen Craft T.J ., Launder B.E., (2001): On the spreading mechanism of the three-dimensional turbulent wall jet. J ournal of Fluid Mechanics, Vol. 435, Pages 305-326. Craft T.J ., Launder B.E., (1992): New wall-reflection model applied to the turbulent impinging. jet. AIAA J ournal, Vol. 30, Pages 2970-2972. Davidson L., Nielsen P.V. and Sveningsson A. (2003): Modification of the v2f model for computing the flow in a 3D wall jet. Turbulence, Heat and Mass Transfer, Vol. 4, Pages 577-584 CFD Forum 2005 Bad Nauheim, Deutschland - 13 - Comparison of different turbulence models to compute wall affected room airflows Ehrhard J . (1999): Untersuchung linearer und nichtlinearer Wirbelviskosittsmodelle zur Berechnung turbulenter Strmungen um Gebude. VDI Verlag Dsseldorf, Fortschrittsbericht VDI Reihe 7 Nr. 367ISBN 3-18-336707-6 Fleischhacker G. and Schneider W. (1980): Experimentelle und theoretische Untersuchungen ber den Einfluss der Schwerkraft auf isotherme turbulente Freistrahlen. Gesundheitsingenieur 101-5, 129-140. Heschl Ch., Fesharaki M., Geyer J . (2002b): Calculation of turbulent flow for building ventilation by means of jet profiles with regard to wall influences. EPIC 2002 AIVC Conference 23-26 October, Lyon France, ISBN 2-86834-118-7 Heschl Ch., Fesharaki M., Steinkellner M. (2002a): Validation der CFD-Analyse von Raumluftstrmungen mittels Messungen. Fluent Anwender Konferenz, 25.-26. September in Frankenthal, Deutschland Heschl Ch., Sanz W., Klanatsky P. (2005): Implementation and comparison of different turbulence models for three dimensional wall jets with Fluent. Fluent CFD Forum 2005, 19. Oktober in Bad Nauheim, Deutschland Lai, J .C.S. & Lu, D. (1992): An inclined two-dimensional wall jet. Proc. 5th Asian Congress of FluidMechanics, 1, pp. 195-98. Lai, J .C.S. & Lu, D. (1996): Effect of wall inclination on the mean flow and turbulence characteristics in a two-dimensional wall jet. Int. J . Heat & Fluid Flow 17(4), pp. 377-385 Lbcke, H.M., Rung Th., Thiele F. (2003): Prediction of the spreading mechanism of 3D turbulent wall jets with explicit Reynolds-stress closure, International J ournal of Heat and Fluid Flow, Vol. 24, Pages 434-443 Nasr, A., and Lai, J .C.S.(1998): A turbulent plane offset jet with small offset ratio. Experiments in Fluids, 24,pp. 47-57. Newman, B.G. (1961): The deflection of plane jet by adjacent boundaries - Coanda effect, in Boundary Layer and Flow Control (ed. G.V. Lachmann), Pergamon Press 1: pp. 232 265 Pettersson Reif B.A. (1999): A nonlinear constitutive relationship for the v2f model. Center for Turbulence Research, Annual Research Briefs, Pages 267-276 Pope S.B. (1975): A more general eddy-viscosity hypothesis. J . Fluid Mech. Vol. 72; Pages 331- 340 Schlin A., Nielsen P.V. (2004): Impact of turbulence anisotropy near walls in room airflow. Indoor Air 2004, Vol. 14, Pages 159-168 CFD Forum 2005 Bad Nauheim, Deutschland - 14 - Comparison of different turbulence models to compute wall affected room airflows Appendix
Figure 12: Geometrical dimension of the used plexiglass model (quantities in mm, the thickness of the plexiglass model is 6mm) CFD Forum 2005 Bad Nauheim, Deutschland - 15 - Comparison of different turbulence models to compute wall affected room airflows
Figure 13: Geometrical dimension of the used air supply duct
CFD Forum 2005 Bad Nauheim, Deutschland - 16 - Comparison of different turbulence models to compute wall affected room airflows f l o w m e a s u r e m e n t A / D - c o n v e r t e r T o r i f i c e i n s t a l l a t i o n p r i m a r y r e c t i f i e r A / D - c o n v e r t e r S e e d i n g
g e n e r a t o r M i x i n g
c h a m b e r T T s e c o n d a r y r e c t i f i e r m o i s t u r e s e n s o r p T a b s T s e c o n d a r y r e c t i f i e r I n l e t
p i p e T e s t
M o d e l T T T (Experiment 2, 3 and 4) L a s e r
s u p p ly
1 P I V
P C T e s t
r o o m L a s e r
s u p p ly
2 P I V
P C L a s e r
s u p p ly
2 L a s e r
s u p p ly
1 (Experiment 1)
Figure 14: Experimental arrangement (in the present study the laser supply and the PIV-Workstation were outside the test room)