You are on page 1of 6

P. A. Durbin G. Medic J.-M. Seo J. K. Eaton S.

Song
Mechanical Engineering Department, Stanford University, Stanford, CA 94305-3030

Rough Wall Modification of Two-Layer k


A formulation is developed to apply the two-layer k model to rough surfaces. The approach involves modifying the l formula and the boundary condition on k. A hydrodynamic roughness length is introduced and related to the geometrical roughness through a calibration procedure. An experiment has been conducted to test the model. It provides data on ow over a ramp with and without surface roughness. DOI: 10.1115/1.1343086

Introduction

Surface roughness can have a profound effect on heat transfer to surfaces beneath turbulent ow. The viscous sublayer adjacent to a smooth wall presents a high impedance to transport to and from the surface. Protrusions that penetrate the viscous layer increase transfer rates between the surface and the uid. They do so by generating irregular, turbulent motion and by extending the surface into the ow. Incorporating rough wall capability into computational methods is a considerable practical need. The present paper is in the vein of providing a practical extension to an existing, popular turbulence model. From an operational perspective, the inuence of surface irregularities must be represented indirectly through a mathematical scheme. The rough surface is replaced by an effective, smooth surface on which modied boundary conditions are imposed. Inuences of roughness are represented by a quantity called the hydrodynamic roughness length. Hydrodynamic roughness modies a turbulence model in order to reproduce averaged effects of the true roughness upon the mean ow and upon the turbulence. The exact prescription of hydrodynamic roughness is, therefore, a function of the turbulence model. The k model is widely used in computational uid dynamics; that is the motive for adopting it here. One could debate whether k is the most effective model available; probably it is not. A notable aw is its inability to describe the near-wall zone. The two-layer approach 1 seems to be the most promising method to x that aw. It generally gives more accurate ow predictions than other schemes 2, such as wall functions or low Reynolds number modications. It is well suited to our present purposes; but it is worth noting that the procedure to treat roughness described herein can be applied to other models. Many aerodynamic and heat transfer applications require a model that is valid for the whole range from smooth, to intermediately rough, to fully rough walls. The high pressure turbine blade is the application that provided an immediate motivation for the present work. In the course of their lifetime blades are subject to erosion by impinging combustor air. The roughened surface experiences an increased rate of heat transfer from the gas stream 3. Models for roughness have a long history. A good survey of the literature is provided by Raupauch et al. 4. Both engineering and meterological concerns are covered. The authors state that in the engineering approach the log-law is modied by an additive roughness function, while in the meteorological approach it is represented by a roughness length. In the present paper we refer to the latter as the hydrodynamic roughness length, to distinguish it
Contributed by the Fluids Engineering Division for publication in the JOURNAL OF FLUIDS ENGINEERING. Manuscript received by the Fluids Engineering Division May 8, 2000; revised manuscript received November 17, 2000. Associate Editor: J. Lasheras.

from the true, geometrical roughness. However, we have to disagree with the distinction made in 4 between the engineering and meteorological approaches. Present concern is with engineering problems, but we make use of both the additive form and of the hydrodynamic roughness length. The additive form is simply 5 U / u 1/ log(y)Br . Data * on the additive coefcient ( B r ) as a function of roughness height constitute the empirical input to the present formulation. A longstanding method used in engineering closure is to modify a mixing length prescription by adding a hydrodynamic roughness length to the wall distance 6: e.g., l ( y y 0 ). The present model is of a different form, but y 0 is made use of. Finally, the present two-layer formulation requires a modied k boundary condition. The rough wall version of the k model 6 shares this feature of incorporating roughness into the boundary condition, although in that case it is in the -boundary condition. This list of methodologies is not meant to be exhaustive; it cites approaches that bear on the present paper. Other methods to represent roughness include adding body forces 7.

The Two-Layer Formulation

The formulation begins by reviewing the two-layer method. It consists in patching together the k l and k models. The standard model equation for turbulent kinetic energy is

t k U k 2 T S 2 T k

(1)

where S i j 1/2( j U i i U j ) is the rate of strain tensor and T is the eddy viscosity. It is used in both the k l and k formulations. In the k model equation 1 is supplemented by

t U

2 C 1 T S 2 C 2 T

(2)

where the turbulent time-scale is T k / and the eddy viscosity is

T C kT .
The standard model constants are C 1 1.44; C 2 1.92;

(3) C 0.09.

1.3;
k 3/2

In the k l model, the dissipation rate is represented by and the eddy viscosity is

(4)

T C k l .

(5)

These supplement the k-equation 1. The VonDriest form for the length scales will be adopted here:

l C l y eff 1 e R y / A ;

l C l y eff 1 e R y / A .

(6)

16 Vol. 123, MARCH 2001

Copyright 2001 by ASME

Transactions of the ASME

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 03/04/2014 Terms of Use: http://asme.org/terms

R y y effk / is a wall-distance Reynolds number in which, for smooth walls, y eff is distance from the wall. It subsequently will be modied by adding the hydrodynamic roughness to the wall 3/4 distance. The log-layer solution requires that C l / C where is the VonKarman constant. The widely accepted value of 0.41 gives C l 2.5. The requirement that k y 2 /2 as y 0 gives A 2 C l 5.0 1. There is only one free constant, A . We have selected the value A0 62.5 for use with the two-layer formulation. That value gave very good agreement between C f versus R predictions and experiments for zero pressure gradient, at plate boundary layers. The two-layer formulation consists simply of using 5 and 4 near a wall, and switching abruptly to 3 and 2 at a patching point. The patching point is dened as the location where the 0 damping function 1 e R y / A equals 0.95; i.e., where y log(20) A 0 / k .

for the surface stress completes the roughness formulation. At least, it would complete the formulation if y 0 were known as a function of the geometrical roughness r. y 0 is not a physical length; it is an artice added to produce a suitable mean velocity.

Hydrodynamic Roughness

Surface protrusions will increase the drag force exerted by the wall on the ow. In a channel ow with given pressure drop, the increased drag would decrease the mass ux and the centerline velocity. The additive constant, B, in the log-law should therefore be decreased by roughness. Assume that the roughness is of the random, sandgrain, variety. The log-law can be written U u 1/ log y B B r r (10) * where r was dened above equation 7. The function B r ( r ) represents the alteration of the additive constant by roughness. The function B r ( r ) has been measured experimentally by Nikuradse; 5 t the curve B r 0 B r 8.5 B 1/ log r B r 8.5 B 1/ log r ; ; ; r 2.25 2.25 r 90 r 90

Modications for Roughness

Roughness has the effect of disrupting the viscous sublayer. We dene the y-origin to be where the mean velocity is extrapolated to zero: that is U ( y 0) 0, by denition. This location of y 0 need not be the effective origin for the turbulent kinetic energy. In fact, for a fully rough wall, the log-layer extends to the origin, at which point k cannot vanish. By denition U is of the form u U * log y y 0 / y 0 under fully rough conditions. Then the eddy viscosity is T u 2 / y U u ( y y 0 ). Hence, the turbulence has an effective * * origin at y 0 . This shifted origin could be considered as a denition of the hydrodynamic roughness, y 0 . In Eq. 6 let y effyy0 . This will accommodate the effective origin of the turbulenceprovided that y 0 is appropriately specied. Just modifying y eff does not accommodate the fully rough condition, that the log-law extends to y 0; the damping function should also be deleted. A is sufciently small that the l damping has little effect under fully rough conditions: so, invoking occams razor, we will leave it as. However A is not small; it must be reduced with roughness. The data cited by 4 give the criterion r 90 for fully rough ow. Here the non-dimensional parameter r is dened as ru / where r is the geometrical * roughness height. The simple, linear interpolation A max 1,A 0 1 r /90 (7)

(11)

through his measurements. This formula is broken into three regions: effectively smooth, transitionally rough, and fully rough. The interpolation function in 11 is

sin

/2 log r /2.25 log 90/2.25

which increases from 0 to 1 through the transitionally rough range 2.25 r 90. Under fully rough conditions 11 and 10 give U / u 1/ log y / r 8.5. * While Eqs. 11 are t to data on sandgrain roughness, other roughness geometries only alter the transitional range. Ligrani and Moffat 5 discuss the case of hemispherical roughness. The rough wall law 10 is commonly represented by u U * log y / z 0 where z 0 re B r log r can be called the hydrodynamic roughness length. In the fully rough regime, r 90, the last of 11 gives z 0 r / e 8.5 .031r the hydrodynamic roughness length is a small fraction of the geometrical roughness size. In the smooth wall regime, r 2.5, z 0 e B / u , which recovers the smooth * wall log-law. We have used B 5.5 for the smooth wall additive constant. The two-layer roughness model is to be solved along with the usual mean ow equation

has the desired effect of deleting the damping. The lower limit of 1 was used, instead of 0, in the max function to avoid division by zero. Now, when r 90, l C l ( y y 0 ). The well known loglayer solution 8 k u 2 / C gives T u ( y y 0 ), as desired. * * If the log-layer is indeed to extend to the origin of y under fully rough conditions, the boundary condition on k must become k (0) u 2 / C . But on a smooth wall k (0) 0. Again, a simple * interpolation is used; however, in this case is it quadratic. The boundary condition on k is k 0 u2 * min 1; r /90 2 . C (8)

The reason for quadratic interpolation is that the dissipation formula 4 gives y 0 u4 k 0 r * 2 y 0 C l A C l A C 90y 0

as y 0 , r 0. The use of quadratic interpolation leads to a nite, non-zero value for 0 in the smooth wall limit. The formula u2 T yU0 * Journal of Fluids Engineering (9)
Fig. 1 Computed log layers compared to data correlation

MARCH 2001, Vol. 123 17

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 03/04/2014 Terms of Use: http://asme.org/terms

B com p r , y 0 B r r y 0

(13)

that was solved to nd as a function of r . Because B com p is only known through a boundary layer solution, the boundary layer code was put into a loop and Eq. 13 was solved by Newton iteration. Figure 1 contains a few comparisons between the computed boundary layer prole and the formula of 5 after the iterations had converged: this simply illustrates the manner in which y 0 was found. The solution for y 0 as a function of r obtained by the above procedure is shown in Fig. 2. The circles are values computed by the Newton iterations. The solid line was t through those values to give a continuous calibration curve.

5
Fig. 2 Calibration curve for hydrodynamic roughness

Tests of the Model

t U U U 1/ p T t U U

(12)

for incompressible ow. To calibrate the model, a zero pressure gradient, boundary layer approximation U x U V y V y T y U is solved, along with the two-layer k equations. The solution is a function of r and y 0 . A boundary layer at the representative Reynolds number of R 5,000 was computed. The range 100 y 160 was used as a sample of the velocity in the logarithmic region. In that region, we dene the computed additive constant by B comp r , y 0 U 1/ log y B , averaged over the sample points. Equating this to the experimental data 11 provides an implicit equation

The roughness formulation was tested by computing the experiment of Coleman and Moffat 9. The experiments were on a at plate boundary layer subjected to acceleration of the free-stream. Both rough and smooth wall data were measured. The acceleration parameter was dened as K r ( rdU / dx )/ U . Data sets are for K r 0, 1.5 10 4 , 2.9 10 4 . The inlet condition in the computations is a fully developed, zero pressure gradient boundary layer at the experimental inlet R . The fully developed state implicitly species proles of U, k and . For the three pressure gradients, the inlet Reynolds numbers were R 2,200; 3,740 and 2,320, respectively. Figure 3 includes the measured values of r . The roughness consisted of close packed hemispheres. The equivalent sandgrain roughness height of the hemispheres was constant and equal to 0.31 in. All the data are near to fully rough. Under fully rough conditions, the shape of the roughness elements becomes irrelevant. The other data in Fig. 3 consist of the skin friction coefcient and Stanton number. Because of the pressure gradient, the Reynolds analogy between heat transfer and skin friction, St 1/2C f , does not hold.

Fig. 3 Rough wall in zero and favorable pressure gradients. Clockwise from upper left: k r 0, K r 1.5 10 4 , K r 2.9 10 4 , r 10, St, C f .

18 Vol. 123, MARCH 2001

Transactions of the ASME

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 03/04/2014 Terms of Use: http://asme.org/terms

Fig. 4 Predicted effect of adverse pressure gradient on smooth and rough wall

The agreement of the model solution to the data in Fig. 3 is generally quite good. The computations used the experimental free stream velocity, U ( x ), which was available at the measurement stations. The pressure gradient was computed from a parabolic t through the velocity data. Therefore, the input pressure gradient was piecewise linear; this is why the computed curves appear to be irregular. The level of agreement between model and data is sufcient to initially validate the present roughness formulation. A curious phenomenon was observed when the model was applied to an adverse pressure gradient boundary layer. The smooth wall case of Schubauer and Spangenberg 10 was computed with a rough wall. Roughness initially increased the skin friction, but subsequently caused it to decline more steeply than on a smooth wall, and the boundary layer separated. This thought experiment is shown in Fig. 4. On further consideration, the behavior is quite plausible: roughness thickens the boundary layer and makes the near-wall region more susceptible to ow reversal. Experimental data on roughness induced separation are not available; so the prediction that roughness can provoke separation inspired us to conduct a new experiment. An existing ramp was installed in the test section of the wind tunnel described in 11. The geometry is shown in Fig. 5 with a computational grid. First, mean ow proles were measured above the smoothwalled model. Then the segment from the inlet to the bottom of the ramp was covered with sandpaper and rough wall data were obtained. The lower wall of the windtunnel, downstream of the ramp, was left smooth. For computations, and in the results to be presented, the curved section of the ramp is taken to be the unit of length and the origin is placed at the top of the ramp. Hence the ramp extends from x 0 to x 1. In these units, the inlet of the computational domain is at x 2, and the exit is at x 4.7. The height of the inlet is 1.87 and it extends from y 0 to y 1.87. After the end of the ramp the lower wall has dropped to y 0.294 in these units.

Under smooth conditions, the inlet momentum thickness Reynolds was R 3,400, while under rough conditions it was 3,800. The roughness height was constant; at the inlet r 360. Of course, at separation r 0, so the full range from fully rough to smooth occurs in this ow. At the computational inlet a fully developed boundary layer, with the appropriate thickness was prescribed for the calculation. It was obtained from a separate computation using a boundary layer code. The upper wall also was provided an inlet boundary layer having R 3,400. The geometry in Fig. 5 is such that separation will occur before the foot of the ramp at x 1 in the smooth wall case. So the geometry is not ideal for demonstrating the predicted ability of roughness to cause separation, but it can be seen from the data in Fig. 6 that separation occurs earlier and is more extensive when the ramp is covered with sandpaper. At the rst three locations of Fig. 6 the rough wall model shows better agreement to data than the smooth wall calculation. However, both are in good agreement and illustrate the larger velocity defect of a rough wall layer. Separation onset is seen to occur upstream of x 0.74 over the rough wall, where the smooth wall ow is just on the verge of separation. At x 1 both cases are separated, but when the wall is rough the separation bubble is higher. The qualitative prediction of roughness induced separation is substantiated by these experiments and the agreement between predictions and data is not bad. It should be noted that the model was developed before the experiment was conductedindeed it was the model prediction that motivated the measurements. The nal test case is the sand dune experiment of 12. This experiment is an open channel ow over an articial dune, consisting of a backward facing ramp. More precisely, over a train of two-dimensional ramps, identical in size and shape. This conguration allows the introduction of a periodic condition in the axial direction, reducing the problem to the single dune presented in Fig. 7. The dune bed has the equivalent sand-grain roughness height k s / d 0.0055. We will concentrate on the case designated in 12 as T6: water depth d 0.292m , with L 1.6m and h 0.08m Fig. 7. With this depth, the water surface can be treated as a plane, free-slip boundary. The Reynolds number based on the bulk velocity and water depth d is R d 1.75 105 . Computations of this conguration have been presented previously by 13, who utilized the k model, with the rough wall boundary condition proposed by Wilcox 8. Patel and Yoon 13 used a computational grid consisting in 82 stream-wise points and 69 cross-stream points, whereas we have used 160 stream-wise points and 50 cross-stream points. In both cases, the distance of the rst grid point from the bed was approximately at 10 6 d , corresponding to y 0.25. In both analyses, the free surface was assumed to be a plane of symmetry. Computed streamlines are included in Fig. 7. They display a region of separated ow along the ramp, with reattachment downstream, on the horizontal lower wall. Figure 8 presents a comparison of friction coefcient C f to the experimental data and to the k computations transcribed from

Fig. 5 Ramp geometry for present experiment

Journal of Fluids Engineering

MARCH 2001, Vol. 123 19

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 03/04/2014 Terms of Use: http://asme.org/terms

Fig. 6 Proles of U velocity component at various downstream positions. Experiment , rough wall; , smooth wall. Computation , rough wall; ----, smooth wall.

13. A smooth wall computation with the two-layer k model is also shown. The introduction of roughness slightly increases the predicted recirculation region, but the largest effect is after reattachment. The roughness model brings C f predictions to the correct level. Both turbulence models are in agreement with the data.

Finally, we compare velocity proles from the roughness simulation to data at measurement stations 4 ( x 0.13L ), 7 ( x 0.27L ) and 14 ( x 0.7L ). The x-locations are measured from the inlet, corresponding to the diagram, Fig. 7. The magnitude of the backow in the separated region is seen to be predicted correctly by the upper left panel of Fig. 9. U-velocity proles at the other locations are also in satisfactory agreement to the data. The V -proles in Fig. 10 also agree well with experiments, except near the wall at station 7. This is on the at bottom wall, shortly after

Fig. 7 Flow domain for sand dune test case. Dimensions are in mm. The lighter lines show streamlines of the separated ow over the roughened ramp.

Fig. 8 Friction coefcient on lower wall. , experiment; present; -X--, k ; "", smooth wall.

Fig. 9 U-velocity proles at stations 4, 7, and 14, proceeding clockwise from the upper left. Lines are computations, squares are experiment.

20 Vol. 123, MARCH 2001

Transactions of the ASME

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 03/04/2014 Terms of Use: http://asme.org/terms

Fig. 10 V-velocity proles at stations 4, 7, and 14. Lines are computations, squares are experiment.

reattachment. It may indicate a slight misprediction of the reattachment location. Note, however, that the values of V are an order of magnitude less than U.

References
1 Chen, C.-J., and Patel, V.-C., 1988, Near-wall turbulence models for complex ows including separation, AIAA J., 26, pp. 641648. 2 Rodi, W., 1991, Experience using two-layer models combining the k model with a one-equation model near the wall AIAA paper 91-0609. 3 Bogard, D. G., Schmidt, D. L., and Tabbtta, M., 1998, Characterization and laboratory simulation of turbine airfoil surface roughness and associated heat transfer, ASME J. Turbomach., 120, pp. 337342. 4 Raupauch, M. R., Antonia, R. A., and Rajagopalan, S., 1991, Rough-wall turbulent boundary layers, Appl. Mech. Rev., 44, pp. 125. 5 Ligrani, P. M., and Mo ffat, R. J., 1986, Structure of transitionally rough and fully rough turbulent boundary layers, J. Fluid Mech., 162, pp. 6998. 6 Kays, W. M., and Crawford, M. E., 1993, Convective Heat and Mass Transfer, 3rd ed., McGraw-Hill, NY. 7 Taylor, R. P., Coleman, H. W., and Hodge, B. K., 1985, Prediction of turbulent rough-wall skin friction using a discrete element approach, ASME J. Fluids Eng., 107, pp. 251257. 8 Wilcox, D. C., 1993, Turbulence Modeling for CFD, DCW inc. 9 Coleman, H. W., Moffat, R. J., and Kays, W. M., 1976, Momentum and energy transport in the accelerated fully rough turbulent boundary layer, Report HMT-24, Dept. of Mech. Eng., Stanford University. 10 Kline, S. J., Morkovin, M. V., Sovran, G. and Cockrell, D. J., eds., 1968, Computation of Turbulent Boundary Layers, v. 1, AFOSR-IFP-Stanford conference. 11 DeGraaff, D. B., and Eaton, J. K., 2000, Reynolds number scaling of the turbulent boundary layer on a at plate and on swept and unswept bumps, J. Fluid Mech., 422, pp. 319386. 12 Mierlo, M. C. L. M., and De Rytter, J. C. C., 1988, Turbulence measurements above articial dunes, Report, Q789, Delft Hydraulics Laboratory, Netherlands. 13 Patel, V. C., and Yoon, J. Y., 1995, Application of turbulence models to separated ow over rough surfaces, ASME J. Fluids Eng., 117, pp. 234241.

Conclusion

The overall objective, to develop a rough wall formulation for the two-layer k model has been met. The basic method consists of inserting a hydrodynamic roughness length into the equations and modifying the k boundary condition to interpolate between the fully rough and smooth limits. The success of the method depends on a calibration procedure that relates hydrodynamic roughness, y 0 to the true, geometric roughness, r. The data selected here are for sandgrain roughness. Other roughness geometries would produce an altered calibration curve in the regime of intermediate roughness. The basic approach of modifying the smooth wall formulation, then calibrating y 0 by iterating on boundary layer solutions, could be applied to other models. Model predictions of roughness induced separation motivated an experiment to verify this behavior. The experiment was conducted in an existing facility and consisted of comparing rough and smooth wall ow over a backward facing ramp. The predicted behavior was conrmed.

Acknowledgment
We gratefully acknowledge support from General Electric Aircraft Engines and from the Ofce of Naval Research, contract N00014-99-1-0300-P0001.

Journal of Fluids Engineering

MARCH 2001, Vol. 123 21

Downloaded From: http://fluidsengineering.asmedigitalcollection.asme.org/ on 03/04/2014 Terms of Use: http://asme.org/terms

You might also like