This article was downloaded by: [Indian Institute of Technology - Delhi] On: 11 February 2013, At: 05:04 Publisher:

Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Numerical Heat Transfer, Part B: Fundamentals: An International Journal of Computation and Methodology
Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/unhb20

COMPUTATION OF BOUNDARY LAYER TRANSITION USING LOW-REYNOLDS-NUMBER TURBULENCE MODELS

Amir Radmehr, Suhas V. Patankar Version of record first published: 29 Oct 2010.

To cite this article: Amir Radmehr, Suhas V. Patankar (2001): COMPUTATION OF BOUNDARY LAYER TRANSITION USING LOW-REYNOLDS-NUMBER TURBULENCE MODELS, Numerical Heat Transfer, Part B: Fundamentals: An International Journal of Computation and Methodology, 39:6, 525-543 To link to this article: http://dx.doi.org/10.1080/10407790152034809

PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: http://www.tandfonline.com/page/termsand-conditions This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not

be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

Numerical Heat Transfer, Part B, 39:525^543 , 2001 Copyright # 2001 Taylor & Francis 1040-7790 /01 $12.00 + .00

COM P UTATION OF BOUND ARY LAYER TRANSITION USING LOW -REYNOLD S -NUM B ER TURBULENCE M ODELS Amir Radmehr and Suhas V. Patankar
Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013
Department of Mechanical Engineering, University of Minnesota, Minneapolis, Minnesota, USA
A literature review shows that, for the commonly used turbulence models, the predicted location of boundary layer transition is very sensitive to the initial pro les of turbulence quantities and starting location of calculation. To eliminate these effects, two independent solution approaches are proposed: ( 1) to solve the boundary layer equations over a at plate with the starting location of calculation very close to the leading edge of the plate, and ( 2) to solve the elliptic NavierStokes equations over the whole plate, including the leading edge and some region upstream of it. Computat ions show both approaches lead to identical results. Three well-known low-Reynolds-number ( LRN) turbulence models are evaluated with respect to the transition on a at plate. None of the models are able to predict the quantitative aspects of transition correctly without an ad hoc adjustment. A satisfactory new turbulence model is presented in a companion paper.

INTROD UCTION Over the years, the efciency levels and operating temperatures of gas-turbine engines have continued to increase. In modern gas-turbine engines the thermodynamic efciency ranges from 40% to 50% and the turbine inlet temperature is as high as 1700 K. As a result, accurate prediction of the ow eld and heat transfer over the turbine blades is critical in the design process. One of the important factors in analyzing the ow eld and external heat transfer on gas turbine blades is the boundary layer transition. Transition of ow from the laminar to turbulent regime causes a signicant increase in skin friction and heat transfer coefcients. Transition also affects the boundary layer separation. The turbulent boundary layer has a greater tendency to remain attached to the surface compared with the laminar boundary layer. Even if a good prediction of the laminar boundary layer and of the fully turbulent boundary layer can be made, the prediction capability is not very useful without a reliable method for predicting transition. Thus, an accurate prediction of transition is invaluable in the calculation of uid ow and heat transfer and in the resulting design of gas turbines. Transition is inuenced by many factors such as free-stream turbulence intensity (FSTI), pressure gradient, surface roughness, streamline curvature, compressibility, and heat transfer, all of which
Received 11 December 2000; accepted 6 February 2001. Address correspondence to S. V. Patankar, Innovative Research, Inc., 3025 Harbor Lane N., Suite 300, Plymouth, MN 55447, USA. 525

526

A. RADMEHR AND S. V. PATANKAR

NOM ENCLATUR E
Cf Cm ; C1 ; C2 D; E f1 ; f2 ; fm h k Pr t R ex R et R ey St u ui x ut skin friction coefcient constants in the ke turbulence models empirical functions used in some LRN models empirical functions introduced in LRN models mean static enthalpy turbulent kinetic energy turbulent Prandtl number Reynolds number based on x turbulent Reynolds number dened as k2 =n~ e turbulent Reynolds number p dened as ky=n Stanton number mean velocity in the streamwise direction Cartesian velocity components streamwise distance from the leading edge friction velocity dened as tw =rw 0:5 y y r e ~ e tw m mt sk ; se Subscripts e i cross-stream distance from the wall nondimensional distance from the wall dened as rut y=m uid density dissipation rate isotropic dissipation rate shear stress at the wall molecular viscosity turbulent viscosity empirical constants in the turbulence models

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

denoting free-stream value denoting value at the initial starting location of the calculation denoting value at the wall, that is, y 0

are present in the complex ows occurring in gas turbines. Mayle [1] has shown that, among these factors, the FSTI and the pressure gradient are the most important ones. Accurate prediction of the following three aspects of transition is important for use in practical applications: (1) the location of the onset of transition, (2) the location of the end of transition, and (3) the manner in which the properties of interest in engineering applications (for example, skin friction coefcient and heat transfer coefcient) vary during transition. In spite of a vast amount of computational and experimental research in this area, accurate prediction of the onset and length of transition is still largely an unsolved problem for gas-turbine designers. Among the methods for predicting transition, solving the boundary layer equations with the LRN ke turbulent model is the most common approach. The rst version of the LRN ke model was introduced by Jones and Launder [2]. They added the so-called LRN functions to the standard ke model such that the near wall damping effects were simulated. Jones and Launder [3] used their LRN model to successfully predict the Reynolds number where pipe ows became turbulent. Early attempts to simulate the boundary layer transition include the work of Pridden [4], Wilcox [5], Dutoya and Michard [6], Arad and colleagues [7], Hylton and colleagues [8], Wang and colleagues [9], and Rodi and Scheuerer [10]. However, in all these cases, a satisfactory prediction of transition is achieved by using very specic initial proles of turbulence quantities at a specic starting location of calculation. Fujisawa [11] performed calculations with several

COMPUTATION OF BOUNDARY LAYER TRANSITION

527

LRN models. He showed that the Launder and Sharma [12] model performs better than the others for ow over a at plate without a pressure gradient but fails in situations with a pressure gradient. However, he did not report the sensitivity of results to the initial proles and starting location. Schmidt and Patankar [13] were the rst to extensively study the sensitivity of the predictions to both the initial proles of k and e and the starting location of calculation. They showed that the predicted location of transition is quite sensitive to these factors. This sensitivity appears to decrease when the starting location of calculation shifts toward the leading edge of the plate. However, with these early starting positions, the LRN models predicted transition at unrealistically early locations and signicantly shorter than those found experimentally. Schmidt and Patankar [14] introduced empirical constants and functions into the LRN models to adjust the beginning and end of transition. These modications were very specic to the problem of a two-dimensional boundary layer and were adjusted to force the results to conform to the empirical correlations for at plates developed by Abu-Ghannam and Shaw [15]. Using these modications, Schmidt and Patankar predicted some of the at plate data very well but experienced difculties with more complex situations, such as accelerated ow. However, the main weakness of the model is its use of problem-specic modications that are not meaningful in other ows. Abid [16] also showed that the results of his new LRN ke turbulence model depended on the choice of initial proles and starting locations. Only when he made these choices in a specic manner was he able to get good agreement with experimental data. Savill [17] reports results achieved by the European Research Community on Flow Turbulence and Combustion (ERCOFTAC) to evaluate the performance of turbulence models in predicting transition over a at plate under the inuence of free-stream turbulence. Although the sensitivity of results to the starting location and initial proles were shown in this report, no effort was made to obtain a result independent of these factors. Savill concluded that models that employ damping factors that are functions of a general property of LRN ows (for example, the local turbulent Reynolds number) are more suitable for predicting transition than models that use the distance from the wall (such as y ) in their damping functions. He recommends the Launder and Sharma [12] model for accurate prediction of the mean ow quantities. Wilcox [18] used the ko model to predict the transitional ow. He set the initial prole of k to zero throughout the boundary layer and introduced a new concept known as the numerical roughness strip. By changing the width of the numerical roughness strip he could trigger the transition at the desired location. Although formulas for the upper bound of roughness heights and roughness strip width were provided, it is difcult to estimate the correct values of these quantities for a new application. Instead, Wilcox recommends trying different roughness heights and roughness strip widths to nd their optimum values for a new application. Biswas and Fukuyama [19] developed a new LRN model and evaluated it along with the other models against experimental data for at plates and turbine blades. Except for one of the turbine blade cases, they obtained quite satisfactory results. However, they did not address some of the details in their calculations. They adopted

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

528

A. RADMEHR AND S. V. PATANKAR

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

Rodi and Scheuerers [10] suggestion for initial prole of e. For initial prole for k, they stated that they used the experimental prole. However, they did not dene the shape of this prole. Also, there is no comment regarding the sensitivity of results to the initial proles and starting location. This review of the literature shows the LRN ke models can predict the qualitative aspects of transition. However, the ability of these models to accurately predict the onset and length of transition must be explored. Furthermore, a single model has been evaluated differently by different authors. To demonstrate this, the results reported by Abid [16], Savill [17], and Biswas and Fukuyama [19] using the Launder^Sharma model for the same physical problem are replotted in Figure 1. In all three references the test case T3A of ERCOFTAC workshop [20] has been solved, which is a at-plate transitional boundary layer with 3% FSTI. The variations of friction coefcient versus Reynolds number are shown in Figure 1. As can be seen, different results are obtained, although the physical problem, turbulence model, and numerical procedure are the same. The only explanation that we can provide for this discrepancy is that they used different initial proles and/or starting locations. In summary, each of the previous attempts to predict transition has at least one of the following deciencies:

Figure 1. Friction coefcient predicted using the Launder^Sharm a model for test case T3A of ERCOFTAC experiments from three different references.

COMPUTATION OF BOUNDARY LAYER TRANSITION

529

. .

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

The predicted location and/or length of transition do not match the experimental results. The predicted location of transition depends on both the starting location of calculation and the initial ad hoc proles assumed for k and e. Therefore, if the starting location of calculation or the initial proles of k and e are changed, a totally different result will be achieved. Empirical functions specic to at-plate boundary layer (Schmidt and Patankar [14]) or other specic methods (Wilcox [18]) are employed in the turbulence model or solution scheme to force the transition at the desired location.

Thus, all the models have focused only on the two-dimensional boundary layer on a single wall and employed problem-specic quantities such as the momentum-thickness Reynolds number or the distance from the leading edge. What is needed is a general framework for a transition model that is not specic to a simple ow but is generally applicable to three-dimensional ows, free jets, conned ows, and so on. It is worth noting that the original high-Reynolds-number ke model, although initially developed in the context of two-dimensional boundary layers, had a completely general framework and over the years has been successfully applied to a wide variety of complex three-dimensional ows. The same type of generality needs to be maintained in a satisfactory model for transition. Finding a general way to correctly predict transition under the inuence of FSTI and pressure gradient is the main motivation for this research. To accomplish this task, we examined the performance of some of the existing LRN models in a framework that does not depend on the starting location of calculation or the initial proles of k and e. Previous studies show that starting the calculation close enough to the leading edge of the plate eliminates these effects. In addition, the performance of the existing models was examined using an elliptic approach. In this approach, we consider the whole plate, including the leading edge and a region upstream of the leading edge, as the calculation domain. Therefore, no ambiguity remains in specifying the boundary conditions for k and e at the upstream boundary. The examination of existing LRN models reveals their shortcomings and indicates the need for a new model. Such a model is presented in a companion paper (Radmehr and Patankar [21]).

H ow D oes an LR N ke M odel Simulate Transition? Turbulence is a complex phenomenon. The transition to turbulence is even more complex. It is characterized by the instability of the laminar boundary layer, the formation of turbulent spots or bursts, and their growth and interaction. Some researchers hold the view that a simple steady-state form of the differential equations for k and e does not incorporate the complex physics and therefore would not be expected to predict transition. The same argument then can be used to question the ability of the ke model to predict even the fully turbulent ow. However, this use of the model is well-established. Such models are admittedly simple approximations to a very complex phenomenon. Yet they capture the main features

530

A. RADMEHR AND S. V. PATANKAR

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

of the process and, because of the empirical information built into them, correctly predict the observed behavior. When the use of an LRN ke model is considered for the prediction of transition, we discover that a kind of transition process already has been modeled in describing the behavior of a turbulent boundary layer from the wall to the outer fully turbulent ow. The presence of the wall causes the turbulent uctuations to become damped. This damping of turbulent viscosity is incorporated into an LRN ke model such that the model can predict the viscous sublayer, the buffer layer, and the fully turbulent layer. If such a model is applied to the entire boundary layer, it will be possible to predict not only the ``transition in the cross-stream direction from the wall to the outer region, but also the transition in the streamwise direction. For the very thin laminar boundary layer near the leading edge, turbulence will be damped across the whole layer. As the boundary layer grows and the free-stream turbulence diffuses into the layer, the damping will become less severe, thus causing a transition to turbulence. Figure 2 shows the transition in the streamwise and cross-stream directions. In this manner, it is possible that an LRN ke model can predict transition without an explicit reference to the instability and turbulent bursts. These effects have been built into the model indirectly via the empirical damping functions for the turbulent viscosity.

TH E GOVER NING EQUATIONS AND C OM PUTATIONAL D ETAILS Elliptic Approach The ow is assumed to be steady and two dimensional. Therefore, the standard elliptic time-average d form of continuity, momentum, and energy equations in Cartesian coordinates is used. Using Boussinesqs eddy-viscosity hypothesis, the

Figure 2. Transitional ow on a at plate.

COMPUTATION OF BOUNDARY LAYER TRANSITION

531

turbulent stress tensor u0i u0j and enthalpy ux u0j h0 can be expressed as ru0i u0j mt @ui @uj 2 rkdij @xj @xi 3 mt @h Pr t @xj
1 2

ru0j h0

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

where mt is the turbulent viscosity, Pr t is the turbulent Prandtl number, and k is the turbulent kinetic energy. The turbulent viscosity is dened as k2 3 ~ e The tilde has been placed over e to differentiate between the meanings of e in the ~ various LRN models. e is dened as mt rCm fm ~ eeD
4

where D is the dissipation of turbulence at the wall and is expressed in some models ~ as a function of k. In the fully turbulent regime D vanishes. Therefore, e and e become ~ instead of e is recommended because it allows the same away from the wall. Use of e ~ the convenience of using e 0 as the wall boundary condition in the equation. The ~ transport equations for k and e can be written as @ruj k @ @xj @xj ~ @ruj e @ @xj @xj m mt se m mt sk @k @uj @ui mt @xj @xi @xj @uj ~ re D @xi @uj r~ 2 e C 2 f2 E @xi k
5 6

~ @~ e e @uj @ui C 1 f1 mt @xj k @xi @xj

Contained within this set of equations are the ve empirical constants Cm , C1 , C2 , sk , and se , and the ve empirical functions fm, f1 , f2 , D, and E. These functions are expressedin different modelsin terms of one or more ``turbulent Reynolds numbers or the inner wall coordinate y . The solution procedure used was the control volume methodology described in Patankar [22]. A typical computational domain for the elliptic approach is shown in Figure 3. The lower surface of the solution domain consists of two sections: (1) the solid wall representing the surface of the plate, and (2) the symmetry line before the leading edge of the plate. The use of the symmetry boundary condition implies a thin plate or a sharp leading edge. The upper surface of the solution domain was placed far enough away from the solid wall to ensure that the boundary layer at the end of the plate did not reach the upper surface. The numerical grid was rened in the y direction close to the wall and in the x direction close to the leading edge of the plate. In the y direction at least 40 grid points were placed inside the laminar sublayer. The value of y for the rst grid point was always set to less than 0.005. A sufciently ne grid was used to ensure grid-independent results. One of the major advantages of the elliptic approach for this problem is the simplication of inow boundary conditions. Setting the inow boundary upstream of the plate leads to uniform free-stream boundary conditions for all physical

532

A. RADMEHR AND S. V. PATANKAR

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

Figure 3. Computational domain for the elliptic approach.

quantities. Velocity components and k are zero at the wall. The wall boundary con~ dition for e is different for various models and is given in Table 2. Equations (5) and (6) can be reduced to the following set of coupled ordinary differential equations in the free stream, which can be solved independently to provide the boundary conditions for k and e: ue me Parabolic Approach The governing equations for a parabolic solution are the standard boundary layer equations. The solution procedure is based on the control volume methodology. The parabolic nature of the equations allows us to use the forward-marching technique to obtain the solution over the calculation domain. The calculation starts from an initial station after the leading edge of the plate where distributions of all variables are specied. However, the specication of initial proles for k and e in the laminar boundary layer is not very straightforward. The experimental data available in the literature is very limited and cannot help us in this matter. Rodi and Scheuerer [10] proposed an ad hoc method to specify the initial proles for k and e. Their suggestions have been adopted by almost all researchers who have attempted to predict transition by using LRN models. Based on Rodi and Scheuerers proposal the initial proles of k and e can be specied using the following relationships: k ke e ak @u @y u ue
n

dke ee dx

7 8

dee e2 C2 e dx ke

e > ee

10

COMPUTATION OF BOUNDARY LAYER TRANSITION

533

The exponent n is taken as 2, and a is an empirical function related to FSTI. The grid renement in the y direction is similar to that for the elliptic approach. In the x direction numerical experiments have shown the maximum Reynolds number (based on the step size) to 10 is sufcient to obtain grid-independent results. PR ED IC TION OF TRANSITION USING EXISTING LRN M OD ELS Selected LR N Turbulenc e M odels

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

The performance of three available LRN models in predicting transition will be evaluated against experimental data for at-plate ow with varying FSTI. The models are those of Launder and Sarma [12] (L^S), Lam and Bremhorst [23] (L^B), and Biswas and Fukuyama [19] (B^F). These models were selected because they have shown better performance in predicting transition compared with others. The values of constants and functions for these models are summarizd in Tables 1 and 2, respectively. Table 2 also contains the wall boundary condition for e. Test C ases To evaluate the performance of the selected models, four test cases of ow over a at plate with different FSTI were considered. Two test cases were selected from the experiments of ERCOFTAC reported by Savill [20] and two test cases from the experiments of Rued [24]. The combination of these four test cases provides the wide range of FSTI (2.3% to 8.6%) necessary to truly evaluate the effectiveness of the models. The basic experimental conditions for these test cases are summarized in Table 3. Comparison between the Elliptic and Parabolic Approach A series of computations were performed with the selected LRN models to compare the results obtained by elliptic and parabolic approaches. The basic experimental conditions were the same as test case 1. In the parabolic approach, the initial proles for k and e were specied using Eqs. (9) and (10) with n 2 and a 0:375 (the value recommended by Rodi and Scheuerer [10] at 3% FSTI). Five different starting locations were considered corresponding to R ex;i 101 , 102 , 103 , 104 , 5 104 . To represent the predicted transition process, the coefcient of friction was plotted versus the Reynolds number based on the distance from the leading edge of the plate. In Figures 4 through 6, results of the parabolic
Table 1. Constants for various LRN turbulence models Model L^S L^B B^F Cm 0.09 0.09 0.09 C1 1.44 1.44 1.46 C2 1.92 1.92 1.00 sk 1.0 1.0 1.4 se 1.3 1.3 1.3

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

Table 2. Functions for various LRN turbulence models f2

2

Model D 2m 2mmt E

fm

f1

ew B:C: !2 0

L^S

exp

" 1 0:3 exp R et

3 2

1 R et =50

3:4

p !2 @ k @xi

@2 ui @xj @xk

L^B

1 exp 0:016 Rey

1 0:05=fm

1 exp R et

@~ e 0 @y

534 " 1 0:3 exp 1 exp R ey 10 R et 6:5

2 !#

19:5 R et

B^F

1 exp

Ret 150

18:5 R et

1 0:3 " 2# R et exp 50

@~ e 0 @y

COMPUTATION OF BOUNDARY LAYER TRANSITION Table 3. Experimental conditions for the selected test cases Case 1 2 3 4 Reference Savill Savill Rued Rued [20] [20] [24] [24] Identier T3A T3A Grid 1 Grid 4 FSTI % 3 6 2.3 8.6 Ue 5.2 9.6 47 47 m/s m/s m/s m/s

535

Thermal B.C. N/A N/A Tw 320 K Tw 320 K

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

approach using different starting locatoins are compared with the results of the elliptic approach. An examination of these gures shows that the prediction of transition with the parabollic approach is strongly dependent on the starting location, when the Reynolds number at the starting locaton is higher than a critical value. This critical Reynolds number is 102 for the L^S model, 103 for the L^B model, and 104 for the B^F model. Below the critical Reynolds number, the prediction of transition is independent of the starting location of calculation, and the results are almost identical to the results of the elliptic approach. Also, the predicton of transition is not sensitive to the starting proles of k and e when the calculation starts below the critical Reynolds number. More details about this behavior can be found in Radmehr [25]. Because the parabolic approach is several orders of magnitude faster than the elliptic approach, there is no need to pursue the elliptic approach further.

Figure 4. Comparison between the elliptic approach and the parabolic approach with different starting locations using the Launder^Sharm a model.

536

A. RADMEHR AND S. V. PATANKAR

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

Figure 5. Comparison between the elliptic approach and the parabolic approach with different starting locations using the Lam^Bremhorst model.

Figure 6. Comparison between the elliptic approach and the parabolic approach with different starting locations using the Biswas^Fukuyam a model.

COMPUTATION OF BOUNDARY LAYER TRANSITION

537

The results presented hereafter were obtained using the parabolic approach with Reynolds numbers at starting locations smaller than the critical value.

Results Figures 7 through 10 represent the results of calculations for the four test cases using the selected LRN models. For the rst two test cases, variations of the friction coefcient versus Reynolds number are plotted. For the last two cases, variations of the Stanton number versus the Reynolds number are plotted. The empirical correlations for Cf in the laminar and turbulent boundary layers are also plotted in Figures 7 and 8 for comparison. Using these gures, we can summarize the performance of the selected LRN models as follows: . The qualitative characteristics of the variation of Cf and St during transition seem reasonable for all three models. A continuous and smooth increase in Cf or St from the laminar to turbulent region can be observed. As expected, the onset of ransition moves upstream with increasing FSTI. The L^S and L^B prediction of onset of transition is too far upstream compared with the experimental data for all test cases. The B^F model prediction of onset of transition is further downstream compared with the experimental data. For test case 3 (2.3% FSTI), the B^F model did

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

Figure 7. Comparison of predicted friction coefcient with test case T3A of ERCOFTAC reported by Savill [20].

538

A. RADMEHR AND S. V. PATANKAR

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

Figure 8. Comparison of predicted friction coefcient with test case of T3B of ERCOFTAC reported by Savill [20].

Figure 9. Comparison of predicted Stanton number with test case grid 1 of Rued [24].

COMPUTATION OF BOUNDARY LAYER TRANSITION

539

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

Figure 10. Comparison of predicted Stanton number with test case grid 4 of Rued [24].

not predict transition at all in the computatonal domain. For test case 4 (8.6% FSTI), shown in Figure 10, none of the models could predict transition in the heated region of the plate. The predicted variation of Cf and St in the transition zone is very steep for all three models (that is, the predicted length of transition is too short).

Overall, the L^S model is better at predicting transition than the L^B and B^F models. At this stage, a simple test was performed to provide a better understanding of the role of initial proles and starting location in predicting transition. Two calculations were performed for test case 2 using the L^S model. The rst calculation was performed with the elliptic procedure starting upstream of the leading edge of the plate. The second one was also performed with the elliptic procedure but starting at a location that corresponds to R e 1:8 104 . This is two orders of magnitude greater than the critical Re for the L^S model. For the second calculations, the proles for k and e at the inlet boundary were specied using Eqs. (9) and (10) with n 2 and a 0:3. The condition for the second calculation is similar to one of the tests performed in the ERCOFTAC project in transition modeling reported by Savill [17]. Figure 11 presents the results of this test. The results reported by Savill [17] are also plotted. The graph shows perfect agreement between the results of the second calculation and the results reported by Savill. As expected, the rst calculation predicts the onset of transition upstream of that predicted by the second calculation. Because in both calculations the elliptic

540

A. RADMEHR AND S. V. PATANKAR

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

Figure 11. Comparison of predicted friction coefcient using different computational domains with test case T3B of ERCOFTAC reported by Savill [20].

approach was employed, the only difference is in the proles of k and e at the location that corresponds to R e 1:8 104 . Next, the calculated proles of k and e at this location from the rst calculation were compared with the assumed initial proles from the second calculation. Such a comparison is shown in Figures 12 and 13. These gures show a fairly good agreement between the assumed and calculated k proles. However, this is not the case for the e proles. The values of e are dramatically higher in the assumed prole. This comparison shows how, for a given starting locaton, a specically adjusted initial prole for e can lead to a better prediction of transition. The huge value of e assumed inside the laminar boundary layer leads to larger dissipation of the turbulent kinetic energy transported and produced inside the boundary layer. Consequently, the transition process is delayed. However, using such a strategy to achieve the proper results has two deciencies. First, it makes the solution procedure specic and dependent on the ad hoc assumptions of the initial proles. Second, the initial proles of k and e are obviously inconsistent with the employed turbulence model.

SUM M ARY AND CONC LUSION 1. The results of the elliptic and parabolic approach in predicting boundary layer transition were compared. It was shown that both approaches lead to identical results if the Reynolds number at the starting location of

COMPUTATION OF BOUNDARY LAYER TRANSITION

541

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

Figure 12. Comparison between the calculated and assumed proles of k at the starting location.

Figure 13. Comparison between the calculated and assumed proles of e at the starting location.

542

A. RADMEHR AND S. V. PATANKAR

2.

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

3.

calculation for the parabolic approach is less than a critical value. Also, it was shown that the solution achieved in this way is independent of starting location and initial proles. The accuracy of the most popular LRN turbulence models in the prediction of transition was examined. It was shown that none of them is able to predict the quantitative apsects of transition correctly. Also, it was pointed out that some of the previously reported successful results were achieved by introducing a larger amount of turbulence dissipation inside the laminar boundary layer. This examination of LRN models has highlighted the need for a more general and satisfactory model. Such a model is present in a companion paper (Radmehr and Patankar [21]).

R EFER ENCES
1. R. E. Mayle, The Role of Laminar-Turbulent Transition in Gas Turbine Engines, Journal of Turbomachinery, pp. 509^537, 1991. 2. W. P. Jones and B. E. Launder, The Prediction of Laminarization with a Two-Equation Model of Turbulence, Int. J. Heat Mass Transfer, vol. 15, pp. 301^314, 1972. 3. W. P. Jones and B. E. Launder, The Calculation of Low Reynolds Number Phenomena with a Two-Equation Model of Turbulence, Int. J. Heat Mass Transfer, vol. 16, pp. 1119^1130 , 1973. 4. C. H. Pridden, The Behavior of Turbulent Boundary Layer on Curved Porous Walls, Ph.D. thesis, Imperial College, London, 1975. 5. D. C. Wilcox, Turbulence Model Transition Predictions, AIAA Journal, vol. 13, pp. 241^243, 1975. 6. D. Dutoya and P. Michard, A Program for Calculating Boundary Layers along Compressor and Turbine Blades, in R. W. Lewis, K. Morgan, and O. C. Zienkiewicz (eds.), Numerical Methods in Heat Transfer, J. Wiley & Sons Ltd., London, pp. 413^428, 1981. 7. E. Arad, M. Berger, M. Israeli, and M. Wolfshtein, Numerical Calculation of Transition Boundary Layers, Int. J. for Numerical Methods in Fluids, vol. 2, pp. 1^23, 1982. 8. L. D. Hylton, M. S. Mihelc, E. R. Turner, D. A. Nealy, and R. E. York, Analytical and Experimental Evolution of the Heat Transfer Distribution over the Surfaces of Turbine Vanes, Technical Report NASA CR-168015, 1983. 9. J. H. Wang, H. F. Jen, and E. O. Hartel, Airfoil Heat Transfer Using Low Reynolds Number Version of a Two Equation Turbulence Model, ASME Journal of Engineering for Gas Turbines and Power, vol. 107, pp. 60^67, 1985. 10. W. Rodi and G. Scheuerer, Calculation of Heat Transfer to Convection-Cooled Gas Turbine Blades, ASME Journal of Engineering for Gas Turbines and Power, vol. 107, pp. 620^627, 1985. 11. N. Fujisawa, Calculation of Transitional Boundary-Layers with a Rened Low-Reynolds-Number Version of a ke Model of Turbulence, Engineering Turbulence Modeling and Experiments 1990: Proc. Int. Symp. Turbulence Modeling and Measurements, Dubrovink, pp. 23^32, 1990. 12. B. E. Launder and B. I. Sharma, Application of the Energy-Dissipation Model of Turbulence to the Calculation of Flow Near a Spinning Disc, Letters in Heat and Mass Transfer, vol. 1, pp. 131^138, 1974.

COMPUTATION OF BOUNDARY LAYER TRANSITION

543

Downloaded by [Indian Institute of Technology - Delhi] at 05:04 11 February 2013

13. R. C. Schmidt and S. V. Patankar, Simulating Boundary Layer Transition with Low-Reynolds-Number ke Turbulence Models: Part 1An Evaluation of Prediction Characteristics, Journal of Turbomachinery, vol. 113, pp. 10^17, 1991. 14. R. C. Schmidt and S. V. Patankar, Simulating Boundary Layer Transition with Low-Reynolds-Number ke Turbulence Models: Part 2An Approach to Improving the Predictions, Journal of Turbomachinery, vol. 113, pp. 18^26, 1991. 15. B. J. Abu-Ghannam and R. Shaw, Natural Transition of Boundary LayerThe Effects of Turbulence, Pressure Gradient, and Flow History, J. Mech. Eng. Science, vol. 22, pp. 213^228, 1980. 16. R. Abid, Evaluation of Two-Equation Turbulence Models for Predicting Transitional Flows, Inst. J. Engng. Sci., vol. 31, pp. 831^840, 1993. 17. A. M. Savill, Some Recent Progress in the Turbulence Modeling of By-pass Transition, Near-Wall Turbulent Flows 1993: Proc. of the International Conference on Near-Wall Turbulent Flows (pp. 829^848) , Tempe, AZ, Elsevier, Amsterdam, New York, 1993. 18. W. C. Wilcox, Simulation of Transition with a Two-Equation Turbulence Model, AIAA Journal, vol. 32, pp. 247^255, 1994. 19. D. Biswas and Y. Fukuyama, Calculation of Transitional Boundary Layers with an Improved Low-Reynolds Number Version of the ke Turbulence Model, J. of Turbomachinery, vol. 116, pp. 765^773, 1994. 20. A. M. Savill, A Synthesis of T3 Test Case of Prediction, Research Report of ERCOFTAC Workshop, 1991. 21. A. Radmehr and S. V. Patankar, A New Low-Reynolds-Number Turbulence Model for Prediction of Transition on Gas Turbine Blades, Numer. Heat Transfer B, vol. 39, pp. 545^562, 2001. 22. S. V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere, Washington, DC, 1980. 23. C. K. G. Lam and K. Bremhorst, A Modied Form of the ke Model for Predicting Wall Turbulence, ASME Journal of Fluids Engineering, vol. 103, pp. 456^460 , 1981. 24. K. Rued, Transitionale Grezschichichten unter dem Einuss hoher Freistromturbulenze, intensiver Wandkuehlung und starken Druckgradienten in Heissgasstroemungen, Ph.D. thesis, The University of Karsruhe, Germany, 1985 [in German]. NASA Report No. TM-88524, 1987 [in English]. 25. A. Radmehr, Computation of Boundary Layer Transition for Gas Turbine Blades, Ph.D. thesis, University of Minnesota, Minneapolis, 1997.