COMPARISON OF THREE k-e TURBULENCE MODELS FOR PREDICTING VENTILATION AIR JETS

Q. Liu, S. J. Hoff, G. M. Maxwell, D. S. Bundy
ABSTRACT. Published data on unconfined plane-wall jets and plane-free jets were reviewed and used to assess the accuracy of numerical simulations, A plane-free jet was numerically simulated using the standard k-e model and four nonuniform grid patterns (70 x32,100 x52,120 x 60, and 120 x 74), The solution for a plane-free jet with adequate grid resolution was in good agreement with the published data, A plane-wall jet was numerically simulated usingfivedifferent grids (70 x 32, 100 x 52, 120 x 60, 120 x 74, and 120 x 92) and three k-e turbulence models (the standard k-e model. Lam and Bremhorst low Reynolds number model, and Lam and Bremhorst low Reynolds number model with wall functions). The simulations predicted velocity decay and velocity profile well, but overpredicted the jet spread and entrainment ratio by 20 to 40%, indicating the need for a better turbulence model for wall jet predictions. Keywords. Ventilation, Numerical, Grid pattern.

J

et-ventilated structures are widely used in both agricultural and nonagricultural systems. Numerous research has been done on ventilation air flows involving jets as well as on jets themselves. Because jets have been studied extensively, they provide good test cases to estimate the effects of using different turbulence models and grids. Turbulence modeling is an important part of numerical simulation. Currently, most of the ventilation simulations use one of three k-e models (k is turbulent kinetic energy, e is the dissipation rate of turbulent kinetic energy): the standard k-e model (Launder and Spalding, 1974), Lam and Bremhorst (1981) low Reynolds number model, or the Lam and Bremhorst low Reynolds number with wall functions (Chen et al., 1990). Aldiough most of the studies reported agreement with the corresponding experimental results using the three models, detailed comparison of the performances of the three models is not available in a ventilation setting. Figure 1 is a sketch of a plane-free jet and a plane-wall jet. A plane-free jet is a two-dimensional jet fliat is not attached to any solid surface. If a two-dimensional jet is attached to a solid surface, the jet is a plane-wall jet. For ventilation applications, the jet characteristics of interest include the velocity profile, velocity decay (or peak velocity), jet spread, and entrainment ratio. The velocity

profile is the velocity profile at a given x location (fig. 1). The velocity decay is the peak velocity u^ as a function of X. The jet spread is the location of yo.5 as a function of x. The entrainment ratio is the flow rate at location x divided by the inlet flow rate, Qx/QoThe objective of this study was to compare the performances of three k-e turbulence models in predicting the jet characteristics of interest to ventilation. Available experimental data were used to assess the performance of numerical simulation and the impacts of different grids and turbulence models. Since inadequate grid resolution can be a major source of error (Thangam and Speziale, 1992), five different grid patterns were used to ensure adequate grid resolution and to evaluate the impact of using different grids in the numerical simulation of jets. Published data on unconfined isothermal air jets were summarized in terms of velocity profile, jet spread, velocity decay, and entrainment ratio for both plane-free and plane-wall jets. The three models used in this study were selected from available turbulence models (Liu et al..

Article was submitted for publication in April 1995; reviewed and approved for publication by the Structures and Environment Div. of ASAE in September 1995. Journal Paper No. J-16264 of Iowa Agriculture and Home Economics Experiment Station, Ames, Iowa. Project No. 3140. The authors are Qianbao Liu, ASAE Member Engineer, Assistant Professor, Dept. of Agricultural Engineering, University of Puerto Rico, Mayaguez, Puerto Rico; Steven J. Hoff, ASAE Member Engineer, Associate Professor, Dwaine S. Bundy, ASAE Member Engineer, Professor, Agriculmral and Biosystems Engineering Dept., and Gregory M. Maxwell, Associate Professor, Mechanical Engineering Dept., Iowa State University, Ames. Corresponding author: Steven J. Hoff, 105 Davidson Hall, Agricultural and Biosystems Engineering, Iowa State University, Ames, lA 50011; telephone: 515-294-6180; e-mail: <hoffer@abel .ae.iastate.edu>.

..
b

Y

i
M).9

AUn/2 ^Un

X

t

!
Figure 1-A plane-free jet and a plane-wall j e t

Transactions of the ASAE VOL. 39(2):689-698 © 1996 American Society of Agriculmral Engineers 0001-2351 / 96 / 3902-0689

689

1963. jet spread. 1990). The suggested division line is at x/b of about 100 (ASHRAE. Sigalla. and other properties. the division line between zones 3 and 4 may need clarification. Tuve (1953) found that the wall-jet spread was slightly less than that of the corresponding free jet.0 X 10^. LITERATURE REVIEW EXPERIMENTAL AND ANALYTICAL STUDIES ON PLANE JETS Unconfined air jets can be divided into four zones (ASHRAE. ASHRAE (1989) (6) Walker (1977) The same as equation 5 (7) 3 L « 0.073(x+12b) (3) (4) (5) Awbi (1991). Other studies quantified the difference as about 30% less than the corresponding free jet (Launder and Rodi.erf(0. (1973) finding on inlet momentum flux.. Narasimha et al. Plane jets have been studied in terms of velocity profile. (1992) reaffirmed the Narasimha et al. jet spread.05 m and the inlet velocity was 6 m/s and was assumed uniform across the inlet. which is also referred to as the self-similar zone (zone 3). (1973) showed that inlet momentum flux JQ (JQ = ^o^b) rather than both UQ and b could be used to determine "most (if not all) mean flow parameters" in the fully developed region (x/b > 30). 1961. Rajaratnam (1976) Launder and Rodi (1981). The jets studied were a plane-free jet and a plane-wall jet. 4. Using experimental data from 17 studies. The velocity decay of a wall jet was found to be equivalent to that of a free jet having twice the inlet width of the wall jet (Tuve. Myers et al. Schwarz and Cosart. ASHRAE (1989) Velocity profile (wall jets) J L . 1958.. Walker. and shear stress (Koestel et al. The same was true for the velocity profile of a wall jet except for the region close to the wall (fig. Refer to nomenclature for all variables used in the table. The jets were isothermal and unconfined (except confinement due to a soUd wall that form a wall jet for the plane-wall jet).. Rajaratnam. and velocity decay (Rajaratnam.). The entrainment ratio of the plane-wall jet was also about 30% less than the corresponding free jet due to the reduced jet spread. jet spread. Launder and Rodi (1981) reviewed experimental studies on plane-wall jets. 1981. 1989)—a short zone in which the maximum velocity of the jet remains practically unchanged (zone 1). Schwarz and Cosart (1961) Rajaratnam (1976) "o Velocity decay (wall jets) Entrainment ratio (free jets) V X Rajaratnam (1976). 5b). 1976). 1953. However. NUMERICAL SIMULATIONS A number of numerical simulations on jets have been conducted.. Fifteen studies on plane-wall jets were compared for velocity decay. The inlet height b (half of the inlet height for the plane-free jet) was 0. Setrak (1988) used the standard k-e model (Launder and Spalding. Many experiments measured jets at x/b over 100 (Launder and Rodi. However. 1989). the jet spread and entrainment ratio calculations for wall jets were assumed to be equivalent to the corresponding free jet (ASHRAE. 1977. and 5 3 L « 0. Table 1. I.1x yo.693Ti2) m Sources (1) Rajaratnam (1976). 1992). 1970.5=0. Tuve. 3b and fig. the entrainment ratio was usually calculated using the velocity profile. Because of the difficulty in measuring flow rate. 1976). 1989).40A/^ Qo Entrainment ratio (wall jets) V b (8) Equations 2.5 = 0. 1981. 1950. momentum loss along the jet.68n)] m (2) Rajaratnam (1976) Jet spread (free jets) Jet spread (wall jets) Velocity decay (free jets) yo.1995a) because they were conmionly used in ventilation simulations. Zone 3 is the longest and the most important in engineering applications among the four zones. Wygnanski et al. Wygnanski et al. The results cited here are for the self-similar portion of the jet (zone 3) only. 1969. The Reynolds number based on inlet height and mean velocity was 2. 1989. The results reviewed in tiiis study are for zone 3 only. 1953. 1973). In their experimental study. turbulence intensity. Wilson et al. Summary of jet characteristics Equation and Number Velocity profile (free jets) -y--exp(-0. Narasimha et al. Table 1 is a list of results reported in the literature and the results were used to assess the accuracy of the numerical simulations. a zone of fully established turbulent flow. a transition zone (zone 2). and a terminal zone (zone 4). The results showed that zone 3 behavior continued at x/b > 100 and as high as x/b =1000.48TI1/7[1 . Walker (1977). Walker.28A/^+12 Qo vb 690 TRANSACTIONS OF THE ASAE . turbulence properties.. Black et al. ASHRAE. Studies on both plane-free jets and plane-wall jets were reviewed and summarized by Rajaratnam (1976) and Abramovich (1963). 1977. although the velocity expressions were different (table 1. 1974) and SIMPLE (Patankar.. Albright.

and Baker and Kelso (1990). possibly a result of the same problem. The use of Reynolds stress models for engineering were limited because of the limited gains by using the models and the greater CPU time requirement (Bradshaw. Fang and Grot. They pointed out that this discrepancy was due to the fact that C^ was set as a constant in the model. 1987). Another problem of the standard k-e model is the poor accuracy when predicting complex flows. Most of the numerical simulations in ventilation applications use the SIMPLE(R) algorithm (Patankar. 1980): ^fpUi<t>--r. 1993.. 39(2):689-698 (1981). turbulent kinetic energy. 1994. By evaluating the standard k-e model using direct numerical simulation. The Lam and Bremhorst low Reynolds number model compared favorably with a number of turbulence models (Liu et al. and dissipation for turbulent Table 2.. and a Reynolds stress model and found that the standard k-e model was not as accurate as the other two models. Jones and Whittle (1992). Awbi.^|^)=S. Most of the new models reviewed have not been tested for ventilation simulation. momentum. Maghirang and Manbeck.09 was used for other simulations. (9) where (|) = scalar variable FA gf^ = effective diffusion coefficient for (]) SA = source term for < [ ) The governing equations include continuity. Gan and Awbi. However. the problem of numerical diffusion may dominate the transport process without adequate grid resolution. In more recent studies. an algebraic model. Chen and Jiang. Patankar. 1991). Choi et al. (1994). The importance of grid resolution was addressed by Baker and Kelso (1990) and Thangam and Speziale (1992). 1988). 1987). Launder and Rodi (1983) concluded that the wall jet is a significantly more complex flow than a simply strained boundary layer flow for numerical simulation. As reviewed by Liu et al... 32 X 32. Jet separation and reattachment are commonly encountered in ventilation air flows. 1993. Rodi and Spalding (1970) and Ng and Spalding (1972) also simulated wall jets in an earlier search for better turbulence models. Bergstrom (1994) used an algebraic stress model with a low Reynolds number model in a two region manner. It was found that numerical simulations using the standard k-e model predicted a jet spread of about 30% greater than experimental results for a plane-wall jet in stagnant surroundings. In a review of experimental and numerical studies on wall jets.09 to 0. two problems are commonly reported using the standard k-e model in engineering applications. was adjusted from 0. Choi et al. Hoff (1990) found that his numerical simulation might not converge if all the grid points were not in a fully turbulent region. 1994. The standard k-e model is widely used to predict ventilation air flows (Worley and Manbeck. (1992). One is that all the grid points have to be located in the fully turbulent region and wall functions are needed to bridge the solid surface and the first grid point adjacent to the solid surface (Launder and Spalding. Choi et al. Mixing length models needed to adjust the constants from flow to flow in wall jet prediction. Patel et al..05 for wall jet simulation while 0. Using a grid of 200 x 100. Cazalbou and Bradshaw (1993) also showed that a number of constants (including C^) in the standard k-e model were different from region to region in wall-bounded flows (boundary layer flow and channel flow). The predicted results were comparable with the experimental results. two-equation models including the standard k-£ model. Simulation of a confined wall jet was part of the study done by Choi et al. c^. Thangam and Speziale (1992) showed that the reported errors in predicting flow over a back-facing step were mainly due to inadequate grid resolution. (1994) compared the performances of the standard k-e model. Summary of the partial differential equations Equation Continuity Momentum 1 Ui ^<|>.. Grid patterns of 12 x 12. 1989. Baker et al. Murakami et al. (1990) used the Lam and Bremhorst low Reynolds number model with wall functions to predict water and air flows in a model and found that this combination was better than the standard k-e model. One constant in the standard k-e model. 1991.eff 0 ^i + ^At 0 Turbulence energy Dissipation k e ^i+[^lt/ak] \i+[li^/o^] P(G-e) (pe/k)(CiG-C2e) 691 .1980) to simulate a wall jet and the effects of obstacles on the wall jet. An empirical entrainment ratio was embedded in the boundary condition at the free boundary of the wall jet. Numerical studies of wall jets using three types of turbulence models were reviewed—models using mixing length hypothesis. Wilcox. 1992. (1988) using the standard k-8 model and a modified TEACH program. and the Reynolds stress models. 1991. Because of these problems. 1991. 1990. 1995. Knappmiller and Kirkpatrick. Because of this problem. and Hoff (1990). Chen et al. 1980). (1995a). Reynolds stress models were better than the two equation models and the models using mixing length hypothesis for predicting wall jets. NUMERICAL SOLUTION GOVERNING EQUATIONS The governing equations for steady state flows can be written in the general form of (White. Detailed discussions of this algorithm and its application for ventilation simulation were given by Awbi (1991). the mixing length model by itself was not popular in engineering applications (Nallasamy. 1974). Hoff (1990) used a low Reynolds number model proposed by Lam and Bremhorst VOL. 1995a. The combination of Lam Bremhorst model with wall functions were also used in Chen et al. Murakami et al. 1984).. 22 x 22. many new turbulence models have been proposed and some of them showed significant improvement over the standard k-e model. especially flow with separation (White. 1990. and 42 x 42 were tested. As pointed out by Baker and Kelso (1990). Patel et al. (1988) found that a fine grid of 42 x 42 did not converge but that coarser grids converged.

A denser grid was not attempted. 1 was the coarsest grid which had one grid point inside the inlet. The terms <[). not all the grid points in grids 4 and 5 were in a Table 4.2 (Wilcox. especially for free shear flows.08 1. Plane-wall jet simulations using grids 4 and 5 with the STD model did not converge. For plane-wall jet only. 1974): Ht-C^k2/e G' Table 3.. It uses damping functions to extend its use to the low Reynolds number region and no wall functions are needed. 5 Plane-free Jet STD LBW LB X X X X Plane-wall Jet STD LBW LB X X X X X X X X* X X X* X X Convergence not achieved.kinetic energy. The LB model requires a dense grid in the solid surface region and has relatively poor convergence behavior (Patel et al. However. C^^ = 0.3. 1984. 1990). 1988). 1974): f^. The STD model has been used successfully in numerous applications. 1984). However. The damping functions used also need improvement (Patel et al.5.e x p (. 1 Grid No.0165Ry)]2/l + 2Q=5\ region where y*" > 11.8 to 9 for the grid points adjacent to the solid surface with all of them located in the region where y+ < 11. Cj = 1. 4 15 120x74 1. 5 30 120x92 1. only the STD model was used because the three models are essentially the same for free shear flows. Combination of grids and turbulence models used Grid No. C.1 15 48 Grid No. the low Reynolds number model proposed by Lam and Bremhorst (1990). Ok . The main difference between grids was the number of grid points in the jet region.1 23 77 Grid No.08 y progression factor 1. As shown in table 3. The LBW model differs from the LB model in that wall functions are used in the solid surface 692 Five different grids were selected (table 3). and the Lam and Bremhorst low Reynolds number model with wall functions (Chen et al. 2 5 100x52 1. Oe . referred to as the STD. the grid is not likely to be as dense as required by the LB model. however. 1990). Grid 3 was the densest grid used for the STD model. The governing equations for each model are listed in table 2 with other equations. Grid 5 was the densest grid used with 30 grid points located inside the inlet.1. 1974). Under this condition.. It had y+ values ranging from 0. 3 8 120x60 1. The LB model is intended to overcome the problems that the STD model has. referred to as the LBW.1.44f. For plane-free jets. using grid 5 may not be dense enough for the LB model (Lam and Bremhorst. 2 Grid No.1. 1993.5 is questionable. referred to as the LB. which requires y+ > 11. Compared with other low Reynolds number model studies using grid points located with y " * " values less than 0.5 for all the grid points.. as reported in tiie literature. List of grids used Grid No. However. Chen and Patel.5 (y*" calculated as k^^^ymin/^).fi-f2-l For LB and LBW models (Lam and Bremhorst.1 1.. 1981).1 7 26 Grid No. it requires all the grid points to be in the fully turbulent flow region and it is not accurate for complex flows. 1981): f^= [l . . Grid No. 1 Points inside inlet 1 Mesh* 70x32 X progression factor 1.08 1. Variables used in table 2 are (Launder and Spalding.09f^.. TRANSACTIONS OF THE ASAE .1 0.This combination allows the model to be used in low Reynolds number regions and does not require a dense grid in the solid surface region because wall functions are used.09 1.0. All three models were used in this study for the planewall jet. F^^eff' ^^^ ^i^ ^^ summarized in table 2 for each equation. SELECTION OF GRIDS f2=l-exp(-R^^ ^ ve R„ ymin"" ^^ shortest distance to any solid surface TURBULENCE MODELS Three turbulence models were used: the standard k-e model (Launder and Spalding. Chen et al.1 Minimum y+t 111 1134 Maximum y+f Grid No. The y+ values for grid points adjacent to the solid surface (for plane-wall jets only) are also shown in table 3. the LBW model is the same as the LB model. 3 Grid No.0. the grid used was much denser than grids commonly used for ventilation simulation. 4 Grid No. Table 4 lists the combinations of grids and turbulence models considered.8 9 \3y 3x1 U/ lay/ Number of points in x direction by the number of grids in y direction. the performance of the LBW model near the solid surface region where y+ < 11.92f2 For STD model (Launder and Spalding.

the momentum at DE should be less than or equal to the momentum at the adjacent upstream section* au/ay = 0. wall jet.2204-0.9962-0. ae/ay . Lam and Bremhorst.0.4IL (Djilali et al. so the STD model solution was used as the initial guess for the LB and LBW models. The parameters k.2k3/2 u . dk/ay = 0. de/dx . RESULTS AND DISCUSSION The predicted results of velocity profile.fully turbulent region.3203x2 0. k . The y direction size was arbitrary yet large enough that little impact was expected on the prediction. 1980.. uniform conditions were assumed across the inlet. and entrainment ratio are presented. The problem was traced to the exponential terms in the Lam and Bremhorst low Reynolds number model.5) u = 0. LB model and LBW model (y+ < 11. Central difference scheme was used to discretize the source terms. dz/dx = 0. DjilaH et al. Grids 1 and 2 were not considered for the LB model because the LB model requires a dense grid in the solid surface region. TREATMENT OF THE EXPONENTIAL TERMS |B- DOMAIN DF INTEREST During the initial stage of this study. 1989. dk/Bx = 0. 2) were selected as symmetry lines. 1994. Wilcox. ae/ay = o wall functions (Launder and Spalding. the left (except the inlet) and bottom boundary (AB and AE in fig. 1995b).9429x+0. A computation domain of 250b in the x direction and 150b in the y direction was used in this study. 1989.5 5. DEC 3000 workstations from Digital Equipment Corporation were used for the calculation.5972X+0.1582x2-0. ak/ay = o.01471x3 0. Chieng and Launder. DETAILS OF THE NUMERICAL SCHEME o<x< 1 l<x<3.8146-0. etc.2vk/ymin2 (Chieng and Launder.1009-0. Boundary conditions AB AE BC DE CD. 1980) u . however. The guessed velocity field to start the iteration process may result in divergence using either the LBW or LB models.0. The 250b was selected to ensure that the size of the domain was long enough to cover the suggested self-similar region of up to 100b (ASHRAE. e. Numerous discussions related to these parameters are available (Chen and Patel. 1981.9 3. 1988. 1974.0. V . dv/dx . required about three times the CPU time that the STD model required to complete one iteration. the inlet condition Ikble 5. skin friction coefficient. It was found that the STD model was less likely to diverge. The impact of that assumption on numerical simulation was not known.0 x lO"-^.004744x2-2. and e . Patel et al. L = 2b was used in the study. As mentioned earlier.5) CD.5<x<8 x>8 This approximation has a relative error of < 2% in the range of 0 < x < 8 and an absolute error of < 0. Bw/dx = 0.0..0. The flow rate at any given axial location x was needed for entrainment ratio calculation. BOUNDARY CONDITIONS The computation domain is shown in figure 2.0.021xl0-V 0 (10) A X 1 jet> wall boundary (wall jet> or centerllne of j e t ifree symmetry line exit Figure 2-The computation domain. wall jet STD model and LBW model (y-». velocity decay. Choi et al. The polynomials are: e-^« /0. are not summarized here because these are not of primary interest for ventilation studies. Chen et al. Use of the approximation improved the speed of the LB and the LBW models to a rate comparable of the STD model. The e value at the inlet was calculated as 8 = (C^^^^^/^) / 0. 1989). The flow rate was calculated by integrating the velocity profile from the center of the jet (the wall in the case of wall jet) to the point where the air velocity component in the x direction was zero. V = 0.0. e = 0. It was suggested that the inlet conditions might be important to achieve similitude (Liu et al.325xlO-V 0. The SIMPLER (Patankar. V = 0. The convergence criterion was that the residual of the continuity equation should be < 2. Uniform condition was assumed at the inlet for all the grids used to ensure comparability between grids.. Each polynomial was a least square fit of data points of the exponential function for the section. 1988). plane-free jet CD. To solve this problem. dkldx = 0.09677x+0.0. The division line between each section was drawn by observing the e~^ curve and minimizing the curvature in each section.0 au/ay .9<x<5.. k . the exponential term e~^ was divided into five sections and each section was approximated by a polynomial. 1993). The boundary conditions are listed in table 5. V .. 1984. if used as proposed with the exponential terms. VOL.0038 in the range of 0 < x < oo.03759x+0.005uo2. 1980) method was used to numerically solve the partial differential equations. 1990) This condition was imposed to eliminate the possibility of unrealistic inflow at the exit.0 u = uo.> 11. To minimize the complication of a solid surface. it was noticed that the LB and the LBW models. Gan and Awbi.01447x2-7. jet spread. 39(2):689-698 693 .

(5) o Grid 1(70x32) A Grid 2 (100x52) ^ Grid 3 (120x60) . 3. and 4 showed less than 5% difference between each other in the predicted results.5 (b) (b) Figure 3-Results of the plane-free jet simulation . The entrainment ratio showed a slightly lower value than the experimental result (eq. PLANE-FREE JET 30 24 ^ Eqn.was generally considered to have little impact on the selfsimilar portion (zone 3) of the jet (ASHRAE. The results are shown in figures 3 and 4. The effects using the STD and the LB model were similar 1. The STD model was used with grids 4 and 5. 1989.Grid 1(70x32) Eqn. Narasimha et al. TRANSACTIONS OF THE ASAE . Launder and Rodi.5. 0. 1973).0- (b) Figure 4-Resiilts of the plane-free jet simulation .-" 12 y" 50 100 x/b 150 200 250 (a) .(a) jet spread and (b) entrainment ratio. 1983.(a) velocity decay and (b) velocity profile. The velocity decay was virtually the same as equation 5. (5) o Grid 4 (120x74) A Grid 3 (120x60) ^ Grid 2 (100x52) .Grid 4 (120x74) D Grid 5 (120x92) y/yO. The effects of different grids using the LBW model are shown in figures 5 and 6. and 4) showed good agreement with published data. Grid 1.Grid 1(70x32) •Jfi / ^ w 50 100 150 x/b 200 250 The plane-wall jet was simulated using the STD. 694 Figure 5-Results of the plane-wall jet simulation using the LBW model .Grid 1(70x32) 18 The plane-free jet was simulated using the STD model and four of the five grids as shown in table 3.(a) velocity decay and (b) velocity profile. The jet spread prediction is slightly higher (< 10% higher) than the experimental result of equation 3. V W] Eqn. (2) o Grid 1(70x32) A Grid 2 (100x52) ^ Grid 3 (120x60) .5 y/yO.2 0 1 50 i 100 x/b 1 150 " " i• 200 • 250 250 (a) (a) Eqn.6 0.Grid 4 (120x74) n Grid 5 (120x92) 50. LB.(l) o Grid 4 (120x74) A Grid 3 (120x60) ^ Grid 2 (100x52) . The results of the plane-free jet with adequate grid resolution (grids 2..4 0. 6) near the exit. overpredicted the jet spread and entrainment ratio and underestimated the velocity decay. grids 2.Grid 1(70x32) Eqn. and LBW models with some or all of the five different grids (table 3). which was the coarsest grid with one grid point inside the inlet width. Both grids had grid points in the wall region with y+ < 11. As shown in the figures.8 Eqn. Effects of Different Grids.(3) o Grid 4 (120x74) A Grid 3 (120x60) V Grid 2 (100x52) . but convergence was not achieved. PLANE-WALL JET . 3. indicating adequate resolution with grid 2. (7) o Grid 4 (120x74) A Grid 3 (120x60) V Grid 2 (100x52) .

As shown later in the results. with grids 1.6 0. all predictions were within 10% of the experimental results. The predictions were within 10% of the experimentally measured profile for most of the jet and within 30% at the outer region of the jet where the velocity was low. However. 6b).(8) o Grid 1(70x32) A Grid 2 (100x52) V Grid 3 (120x60) . Grid 4 did not show a clear advantage in entrainment ratio prediction as it did for the jet spread prediction. The LB model overpredicted the jet spread by about 40%.250 100 x/b 150 200 < ^ 250 (a) (b) Figure 7-Results of the plane-wall jet simulation using grid 3 (a) velocity decay and (b) velocity profile. Grid 5 was the densest grid used but it did not have the best performance. Based on the above analysis. The STD model and tiie LBW model gave predictions that were very close to each other and were about 30% larger than the experimental results. The STD model and LBW model gave identical results.k.^ ' 1 o Grid 1(70x32) A Grid 2 (100x52) V Grid 3 (120x60) • Grid 4 (120x74) n Grid 5 (120x92) 6 0 w ^ ^^ 100 x/b 150 200 250 50 (a) Eqn. The 30% difference between the experimental results and the STD model prediction was in agreement with the conclusion of Launder and Rodi (1983). the LB model overpredicts the jet spread. The results of using other grids were similar to grid 3. The effects of different turbulence models using grid 3 are shown in figures 7 and 8. 2. Predictions using grids 1 and 4 showed slighdy larger errors than the other grids. All models significantly overpredicted the jet spread. Further investigation is needed to study if the same grid impact is true for jets with separation. all of which had y"^ > 11. ^ .Grid 4 (120x74) a Grid 5 (120x92) 250 (b) Figure 6--Results of the plane-wall jet simulation using the LBW model .0 to the LBW model.5 (table 3) where LBW is effectively the same as the LB model. Grid 4 gave the best performance in terms of jet spread prediction. They concluded 1. The velocity profiles at x/b « 150 are shown in figure 7b. Comparing the results of grids 1.30 24 = 18 Eqn. and 5 slightly better than the other two grids (fig.(4) J^?-. 695 . the predicted profiles showed a peak velocity closer to the wall than what was observed in experiments. The jet spread predictions were 20 to 30% higher than the experimental results (fig. The velocity profiles were at the location of x/b closest to 150 (depending on the grid. Figure 5a shows the velocity decay using different grids. Figure 7a showed the velocity decay predictions using different turbulence models. making it VOL. no grid showed a clear advantage in terms of better agreement with the experimental results for the plane-wall jet. The velocity profiles showed little difference at different locations except the initial part of the jet (up to x/b = 40) and at the exit. The overpredictions of that grid in velocity profile and velocity decay eroded the advantage in jet spread. The jet spread predictions with different turbulence models are shown in figure 8a.(5) 0. The LB model prediction showed a slightly larger error than the other two models. grid 2 showed larger overprediction of jet spread and entrainment than grids 1 and 3. The reason may be that all the grid points adjacent to the wall boundary for grid 5 have y+ values less than 11.(a) jet spread and (b) entrainment ratio.5 and used wall functions. Changing from grid 4 to grid 5 also changed the predicted results.8o STD A LBW . x/b ranged from 145 to 155). The predicted results were all in good agreement with the experimental results. Velocity profiles predicted are shown in figure 5b. difficult to judge what should be the adequate grid resolution. The entrainment ratios were overpredicted by about 30%. 4. 39(2):689-698 0. Figures 6a and 6b show the predicted jet spread and entrainment ratios. Effects of Different Turbulence Models.4- 10.LB Eqii. and 3. 6a). The three models showed almost identical results and the results were in good agreement with the published data. As shown.

C2. it is concluded that the LB model is not as good as the other two models for wall jet prediction. All three turbulence models gave prediction of velocity decay and velocity profile close to the published data. no one grid showed a clear advantage over the others among the five grids used. The LBW model had an advantage in that it could be used for a region where the flow was not fully turbulent. The overprediction in jet spread and entrainment ratio may also exist in room air movement involving wall jets. 8b). and jet entrainment ratio were well established by experimental and analytical studies. The STD model and the LBW model showed almost identical results which were better than the LB model. Error due to inadequate grid resolution may have offset other errors for the coarse grids.(a) jet spread and (b) entrainment ratio. CH turbulence model constants error function: erf erf(x) = . • that "the main cause of this discrepancy is the wall's damping of the normal velocity fluctuations". Contrary to the common practice that an unconfined plane-wall jet can be treated as one half of the free jet. A room of the same length will have a smaller overprediction with a lower room height because the jet's expansion is more likely to be restricted. The LBW and LB model did not show any improvement over the STD model in terms of better agreement with the published data. A grid that was too coarse (one grid point inside the inlet) overpredicted jet spread and underpredicted velocity decay for the plane-free jet. half of the inlet height for plane-free jet Ci. the LB and LBW models took about three times as long to complete one iteration compared with the standard k-e model. indicating a need for better turbulence models to predict plane-wall jets. The LB model gave the worst performance for the grids tested and the STD model may not converge if all the grid points are not in the fully turbulent region. The overprediction is likely to be smaller than that of an unconfined jet because the surroundings also affect jet expansion. However. the STD model did not converge if grid points were not in the fully turbulent region (grids 4 and 5). Based on this analysis. velocity profile. the LBW model was better than the other two models.| -t^dt G k Jo P Q Rk Rt turbulence model functions turbulence generation specific turbulent kinetic energy momentum flux at the inlet mean pressure volumetric flow rate turbulence Reynolds number (k^/^y/x)) turbulence Reynolds number [k2y/(ve)] 696 TRANSACTIONS OF THE ASAE . For the plane-wall jet. The solution was in good agreement with the experimental results."QS s * ^l "C^ > . experimental results showed that the plane-wall jet has 30% less jet spread and entrainment ratio than the corresponding plane-free jet.(8) o"* Y J L / _ 50 • 100 150 (b) 200 250 Figure 8-Resiilts of the plane-wall jet simulation using grid 3 . For the plane-free jet. Using a polynomial approximation to replace exponential terms significantly improved the speed of the low Reynolds number model used.30 CONCLUSIONS • o STD A LBW . numerical simulation using the STD model showed that adequate grid resolution was reached with five grid points in the inlet for the jet studied. This is especially true for jet spread prediction. This damping is "not just in the immediate vicinity of the surface. but in the outer 'free-shear-layer' region beyond the velocity maximum". NOMENCLATURE VARIABLES b jet inlet height for wall jet.^ 6 0 M DO " ^ ^ 100 x/b 150 200 250 50 • (a) r^*"*" j^' ^ ^ . But they all overpredicted the jet spread and entrainment ratio significantly (20 to 40%). The STD model and the LBW model gave comparable results. jet spread. For the plane-wall jet. Similar conclusions were also drawn by Cazalbou and Bradshaw (1993). It was found that the jet velocity decay. • Published data on unconfined plane-free and planewall jets were summarized. Because of the exponential terms used in the low Reynolds number model."A 6^ o STD A LBW LB Eqn.(4) 24 = 18 i . Further studies on turbulence models are needed to improve the performance of jet spread prediction. The overprediction in jet spread resulted in the overprediction of the entrainment ratio (fig.LB Eqn.

Kelso. J. Environment Control for Animals and Plants. ASHRAE Handbook (Fundamentals). 1988. Patel. 1994. London. and P. S. 19:53-74. N. and R. The numerical computation of turbulent flows. E.. Ventilation of Buildings. M. 1990.: ASAE. 100(2):669-676. 1990. Cazalbou. M.. Building and Environment 27:321-338. turbulent flow by a low-Reynolds-number k-8 model. Fluids Eng. S. 1989. St. Huber. Transactions ASHVE56:A59-An%. B. P. 19:81-128. Jiang. B. Numerical Heat Transfer and Fluid Flow. Evaluation of model coefficients using direct numerical simulations.D. The turbulent wall jet-measurements and modeling. 1992. V Um s y^ yo. Narasimha. Choi. Murakami. Prog. and H. Hoff. Chon and J. J. Baker. A numerical study of indoor air quality and thermal comfort under six kinds of air diffusion. I. 95-4476. Bundy. 1987. R. B. 1990. J. J. B. England: E&FNSPON. T. 77:354-359. P. Spalding. D. 1972. Ph. andEng. 39(2):689-698 697 . Plane turbulent wall jet flow development and friction factor. K. C. 1980. T. Mass. Computational determination of the behavior of a cold air ceiling jet in a room with a plume. Physics of Fluids 5(12):3233-3239. Mich. Awbi. Turbulence model for boundary layers near walls. Kato and R. Hermann and G.y < > 1 ^ V Mt ^t'^e P 11 SUBSCRIPTS r generalized source term velocity components jet velocity decay or peak velocity dimensionless distance from the solid surface location where u = u^^/l minimum distance to any solid surface dissipation rate of k coordinates generalized variable laminar dynamic viscosity laminar kinematic viscosity turbulent viscosity diffusion Prandtl numbers density diffusion coefficient ^ = y/yo. E. Press. Q. P. Myers.. 1950. S. Murakami. and R. Kato and H. Numerical simulation of the performance of building ventilation systems. 1990.. A. and D. C. Joseph. G. Nakagawa. and D. White. A. Mich. Numerical prediction of a turbulent fountain in a room. Transactions of the A&4£: 36(5): 1449-1459. ASHRAE Transactions 96(2): 361-366. The Theory of Turbulent Jets. 1990. 1995a. R.. Turbulent transport in wall-bounded flows. 1994.: ASHRAE. Liu.I. Turbulent secondary flows. Whittle. G. ASHRAE Transactions 98:929-939. E. N. 1993. Development of a robust finite element CFD procedure for predicting indoor room air motion. J.: ASAE. C . 1981. H. Moser and P. Transactions of the A^A^" 31(6): 1804-1814. Albright. A comparison of three proposed ventilation criteria. L. 1969. The Physics of Fluids 15(1): 20-30. Chen. and Z. Numerical Heat Transfer 3:189-207. 1991. A.: ASAE. New York: McGraw-Hill. 1991. J. Modeling particle transport in slot-inlet ventilated airspaces. Koestel. 15:429-459. S. B. Significant questions in predicting room air motion. Launder. A. Chieng. J. Patel. 1980. Ooka. L. 113:579-586. AAIW. 1993. ASAE Paper No. The turbulent wall jet. H. Building and Environment 24:73-84. 1983. Hoff and D. Spalding. ASAE Paper No. Gan. 1987. Patankar. Joseph. E. ASAE Paper No.U. Q. M. Cambridge. H. Narayan and S. J.. B. E. Basic Eng.. and G. 1988. St. Ann. 1974. C. 1973. ASM. ASHRAE. V. Ng. and K.. Comparison of numerical predictions of horizontal nonisothermal jet in a room with three turbulence models . T. Plane non-isothermal air jets discharging along a smooth ceiling. 1989. Q. and B. Albright. Fang. Gartshore and M. Bremhorst. 1992. . Numerical simulation of the indoor environment.Current status and capabilities. and H. Unpub. 100(2):677-684. H. Yoon. diss. Transactions ofASME 85:47-54. ASHRAE Transactions 100(2): 697-704. Bergstrom. Computer Methods in Applied Mech. Schauer and R. Building and Environment 29(4):449-459. Numerical Heat Transfer. and DSM. J.. Fluid Mech. Y. Turbulent flow in a channel with a wavy wall. N. Salcudean. Launder. Numerical prediction of horizontal nonisothermal 3-D jet in room based on the k-e model. Turbulence models and their applications to the prediction of intemal flows: A review. ASHRAE Trans. . Williams and R. Mich. A.5 ymin £ x. 1995b. AIAA J. S. 26(6):641-648. Kirkpatrick. P. D. Kelso. 1981. Manbeck. VOL. A. M. Ga. Univ. Djilali. Q. Transactions ofASME 103:456-460. Baker. Aerospace Sci. 1992. 95-4642. Computational fluid dynamics for building air flow prediction . C . Three-dimensional air speed and temperature distributions in a scaled-model livestock confinement facility. Mich. 1991. Warhaft.k-£ EVM. Moser and A. An application of the k-epsilon turbulence model to predict air distribution in a slot-ventilated enclosure. D. 1989. P. Black. Joseph. 1994. Calculation of convective heat transfer in recirculating turbulent flow using various near-wall turbulence models. Nallasamy. Awbi. A modified form of the k-e model for predicting wall turbulence. J. Joseph.. Knappmiller. Suter. D.5 turbulent effective grid number at inlet X location t off ij 0 X REFERENCES Abramovich. H. G. Chen. Prediction of buoyant. J. B. 1963. J. Timmons and Z. 69-450. B. 3:269-289. On validation of computational fluid dynamics procedures for room air motion prediction. On the calculation of turbulent heat transport downstream from an abrupt pipe expansion. Eustis. Rodi. Comparative study of ventilating jets from various types of outlets. Minneapolis. V.. Jones. T. Walker and G. Bradshaw. Aeronautical J. Atlanta. Parthasarathy. J. Rev. Launder. K. C. Maghirang.: The M.. K. Advances in computational fluid dynamics: A review from a ventilation simulation perspective. St. Building and Environment 29(4): 261-273.: ASAE. ASHRAE Transactions 97:38-48. L. ASHRAE Trans. 1994. Y. Parametric analysis of turbulent wall jets in still air. ASHRAE Transactions 96(1): 760-774. S. Fluid Mech. Chen. S. St. Computers & Fluids 15(2): 151-194. 1963. ASHRAE Transactions 98(1):203-217. Near-wall turbulence models for complex flows including separation..T. and V. M. J. Tuve. ASHRAE Transactions 96(l):564-573. 1994. H. K. Part A 16:189-212. Bradshaw. S. . Chen. Grot. C. and A. Application of computational fluid dynamics in room ventilation. and W. Lam. of Minnesota. B. Rev.

Tuve. Turbulence models for near-wall and low Reynolds number flows: A review. Schwarz.und Stojfubertragung 3:85-95. Transactions oftheASAE 38(l):231-239. 1992. The two-dimensional turbulent wall-jet. 1993. Wilson. L. thesis. Wall-jet velocity and temperature profiles resulting from a ventilation inlet. 1970. Y. 1970. I. Scheuerer. L. W. J. Warme. A two-parameter model of turbulence. Rodi and G. N.K. Rodi. 698 TRANSACTIONS OF THE ASAE . Walker. Napier Polytechnic. Setrak. 31:1414-1421. Turbulent Jets. Esmay and S. Thangam. and W. 1984. Manbeck. J.Patel. Speziale. Ph. 1976. Cosart. of Fluid Mech. 1991. 1992. Fluid Mech. S. B. AIAA J. G. W. Wilcox. A. AMA 7. 30:1314-1320. N. and H. Wygnanski. Worley. and its application to free jets. D.. Transactions oftheASAE 20:517-522. Persson. C. 1977. A. The Netherlands: Elsevier. G. U. Transactions oftheASAE 13(2):77-81. D. 7. 1995. The effect of rectangular obstacles on the diffusion of wall jet. A. J. AMAy. White. New York: McGraw-Hill. B. H. Turbulent flow past a backward-facing step: A critical evaluation of two-equation models. 62:873-877. Horev. 1988. 10:481-495. S. Transactions ASHVE 59:261-279. C . Measurements of skin friction in a plane turbulent wall jet. Katz and E. Comparison of two-equation turbulence models for boundary layers with pressure gradient. and C. 1953. P. On the applicability of various scaling laws to the turbulent wall jet. Spalding.. of the Royal Aeronautical Soc. J. 1961. F. 1958.D. 23:1038-1319. Modeling particle transport and air flow in ceiling inlet ventilation systems. Review of the theoretical relationships of isothermal ventilating air jets. Sigalla. Amsterdam. Rajaratnam. M. W and D. V. Air velocities in ventilating jets. 234:669-690. M. Viscous Fluid Flow.