Professional Documents
Culture Documents
McClain
Assistant Professor
Department of Mechanical Engineering,
The University of Alabama at Birmingham,
1530 3rd Ave. S., BEC 358B, The Importance of the Mean
Birmingham, AL 35294-4461
S. Patrick Collins
Elevation in Predicting Skin
Graduate Research Assistant
Department of Mechanical Engineering,
The University of Alabama at Birmingham,
Friction for Flow Over Closely
1530 3rd Ave. S., BEC 257,
Birmingham, AL 35294-4461
Packed Surface Roughness
B. Keith Hodge The discrete-element surface roughness model is used to provide insight into the impor-
Professor tance of the mean elevation of surface roughness in predicting skin friction over rough
Department of Mechanical Engineering, surfaces. Comparison of experimental data and extensive computational results using the
Mississippi State University, discrete-element model confirm that the appropriate surface for the imposition of the
P.O. Box ME, no-slip condition is the mean elevation of the surface roughness. Additionally, the use of
Mississippi State, MS 39762 the mean elevation in the Sigal-Danberg approach relating their parameter to the equiva-
lent sand-grain roughness height results in replacing three different piecewise expres-
Jeffrey P. Bons sions with a single relation. The appropriate mean elevation for closely-packed spherical
Associate Professor roughness is also examined. 关DOI: 10.1115/1.2175164兴
Department of Mechanical Engineering,
Brigham Young University,
435 CTB, P.O. Box 24201,
Provo, UT 84602-4201
冉 冊
including Schlichting’s 1937 data set.
−1.6
S pa A f The basis for the discrete-element model was also described by
⌳s = 共1兲
S f As Schlichting in his 1937 paper. In attempting to explain the effect
of roughness element density on the effective sand-grain height,
where S pa is the plan area 共as seen from above兲 of the surface and Schlichting suggested that the total drag on a rough surface is the
S f is the total frontal area of the roughness elements, A f is the sum of the skin friction on the flat part of the surface and the form
frontal area of the elements, and As is the windward wetted sur- drag on the individual roughness elements. Methods to solve the
face area of the roughness elements. While the Sigal-Danberg boundary-layer equations for flow over rough surfaces, such as
parameter has demonstrated the most promise for correlating the Finson 关9兴, Adams and Hodge 关10兴, and Lin and Bywater 关11兴,
available equivalent roughness height data, the best correlation incorporated momentum sink terms into the boundary-layer equa-
based on the parameter is still limited. Bons 关7兴 has also shown tions based on Schlichting’s suggestion. The form of the discrete-
element roughness model presented in this paper originated in the
Contributed by the Fluids Engineering Division of ASME for publication in the
work of Finson 关9兴 and was rigorously derived by Taylor 关8兴.
JOURNAL OF FLUIDS ENGINEERING. Manuscript received February 4, 2005; final manu- The discrete-element model has been validated using many data
script received October 25, 2005. Review conducted by Joseph Katz. sets for surfaces with ordered roughness elements with the same
Journal of Fluids Engineering Copyright © 2006 by ASME MAY 2006, Vol. 128 / 579
冋冉 冊册
the elements were very closely packed. The roughness elements
on the other surface were moderately packed as compared to sur- u u u
 x u + y = − 共x P兲 +  y − u ⬘ ⬘
faces tested by Schlichting 关1兴, Scaggs et al. 关12兴, Hosni 关13兴, x y x y y
Chakroun 关14兴, and Taylor et al. 关15兴. The two cone-roughness Nr
1 u2
surfaces were created with characteristics congruent with the real
random roughness described in the works by Bons 关7兴 and Mc-
−
2 L pLt i=1 兺
CD,idi 共4兲
Clain et al. 关16兴. Measurements of skin friction for turbulent flow
over the conical roughness surfaces were made for comparison to The turbulence model is not modified to include roughness ef-
predictions using the discrete-element model. fects since the physical effects of the roughness on the flow are
Since the DEM has been extensively validated for the afore- explicitly included in the differential equations. In this study, the
mentioned data sets and since the DEM captures the relevant Prandtl mixing length with van Driest damping is used for turbu-
physics 共in the spatially averaged sense兲 of a wide range of rough- lence closure. Hence
ness geometries, the DEM provides a proven tool for analysis and
design procedures for turbulent flow over rough surfaces. Hence,
given a detailed description of the surface roughness, the DEM
− u ⬘ ⬘ = l m
2
冉 冊冏 冏
U
y
U
y
共5兲
has been shown to posses the capability to properly account for where
再
the surface geometry effects on the surface measurables. When
incorporated into a design code the DEM will capture the salient, 0.41y关1 − exp共− y +/26兲兴 for lm 艋 0.09␦
lm = 共6兲
spatially averaged effects of the surface roughness. 0.09␦ otherwise
In addition to the usual turbulence modeling closure require-
ments for −u⬘⬘ and u⬘2, the roughness model has closure re-
The Discrete-Element Model quirements for CD. CD is formulated as a function of the local
The discrete-element model is formulated for roughness ele- roughness element Reynolds number
ments with three-dimensional shapes for which the element cross
section can be defined at every height, y. The differential equa- u共y兲d共y兲
Red = 共7兲
tions including roughness effects are derived by applying the basic
conservation statements for mass and momentum to a control vol- thus directly including information on the roughness element size
ume such as that shown in Fig. 1. Basic to this approach is the and shape. The functional forms for CD used in this study for
idea that the two-dimensional, time-averaged turbulent boundary- circular roughness elements is
冉 冊
layer equations can be applied in the flow region below the crests
冦
−0.125
of the roughness elements. The flow variables are spatially aver- Red
aged over the transverse 共z兲 direction and the streamwise 共x兲 di- Red ⬍ 60,000
CD = 1000 共8兲
rection. The physical effects of the roughness elements on the
0.6 Red ⬎ 60,000
fluid in the control volume are modeled by considering the flow
blockage, the local element heat transfer, and the local element The functional form for CD was determined by extensive calibra-
form drag. The void factors, , are defined as the fraction of the tions using a number of deterministic surface roughness data sets
area open to flow. The form drag force on the control volume is 关8,17兴.
Cf =
共1 − ␣兲
du
dy
冏 y=y me +
1 1
2 L tL p
冕冉
⬁
y me
u2
Nr
兺C
i=1
D,idi 冊 dy
1 2
U
2 e
共12兲
冏 冕冉 冊
erated to have characteristics congruent with two randomly rough
⬁ Nr
surfaces created from scaled profilometer traces taken from real
兺C
du 1 1
共1 − ␣兲 y=0 + u2 D,idi dy gas turbine surface roughness 关7,16兴. The conical roughness ele-
dy 2 L tL p i=1
0 ments on Surface 1 were created with a height equal to three times
Cf = 共10兲
1 2 the rms roughness height, Rq, of the deposit surface described by
U McClain et al. 关16兴. The rms roughness height, Rq is determined
2 e
using
y2
N i=1 i
共13兲
flows over randomly rough surfaces. The adaptation made for where y i is measured from the mean elevation of the surface. The
flows over random roughness that is relevant to this study is the height multiplier of three is a nominal value based on the discus-
adoption of the mean elevation as the datum or location of the sion by Tolpadi and Crawford 关18兴 regarding the relationship be-
no-slip surface. Figure 2 presents a three-dimensional representa- tween the statistical parameters of a randomly rough surface and
tion of a randomly rough surface. The most obvious difference its equivalent sand-grain height.
between the surface depicted in Fig. 2 and the cone surface rep- For the deposit surface, the correlation length, c, was found
resented in Fig. 1, is that there is no “flat” part of the randomly using Fourier decomposition. The dominant frequencies were
rough surface from which to reference the discrete-element found and the correlation length was calculated as the distance at
boundary-layer equations. which the autocorrelation function falls to a value of 0.1. The
A reference or datum is needed for the application of the diameter of the cones on Surface 1 was set to one-half the corre-
boundary conditions in solving the boundary-layer equations. A lation length, and the spacing was set equal to the correlation
reference surface at the minimum elevation makes sense for the length.
blockage-fraction evaluation, but the roughness model needs both The roughness elements on Surface 2 were created with a
blockage-fraction and blockage-diameter distributions. For ran- height equal to three times the Rq of the erosion-2 surface as
dom roughness with closely-spaced roughness features below the described by McClain et al. 关16兴. The diameter of the elements on
mean elevation, the concept of a blockage diameter is not appro- Surface 2 was set as twice the correlation length for the erosion-2
priate. Because all of the blockage elements are connected, the surface. The spacing of the cones was set so that distance between
effective diameter of the blockages becomes the length of the the rows of cones in the direction of the flow equaled the base
profilometer trace as the minimum elevation is approached. diameter of the cones.
McClain et al. 关16兴 determined that the mean elevation serves The two cone surfaces were generated using the three-
as the computational location of the no-slip surface for randomly dimensional 共3D兲 printer and studied in the wind tunnel. A Strata-
rough surfaces. Since the mean elevation is the mean location of sys Genisys Xs three-dimensional printer was used to create cou-
the no-slip surface, applying the no-slip surface at the mean el- pons of the cone roughness surfaces. The Genisys Xs printer
evation is consistent when using a spatially averaged model such generates 3D models with a resolution of 0.33 mm using a durable
as the discrete-element model. Using the mean elevation as the polyester. The printed surface coupons were fitted and connected
location of the no-slip surface produced very good agreement with together to make roughness panels to be placed in a wind tunnel
experimental measurements of skin friction for turbulent flows for testing. Table 1 presents the measured characteristics of the
over randomly rough surfaces. cone surfaces from the three-dimensional printer.
Surface 1 4.420 5.969 12.065 10.449 12.065 2.021 Surface 1 944,000 2600 300.8 0.20
Surface 2 1.905 4.445 5.207 4.509 5.207 1.171 Surface 1 525,000 1600 171.2 0.18
Surface 2 923,000 2300 107.4 0.09
Surface 2 517,500 1400 59.0 0.08
The height of the cones above the mean elevation and the di-
ameter of the cone at the mean elevation, as depicted in Fig. 3, are
calculated using Eqs. 共14兲 and 共15兲, respectively. and a low Rex around 500,000. Table 2 summarizes the boundary
layer and freestream conditions for the rough-surface skin friction
d20k measurements. The momentum thickness and boundary-layer
keff = k − 共14兲
12L pLt thickness were not measured directly in the experiments. The val-
ues of Re and k / ␦ presented in Table 2 are based on the
keff boundary-layer code predictions considering the roughness ef-
d0,eff = d0 共15兲 fects.
k
The Reynolds numbers’ designations as “high” or “low” are
The heights and diameters of the Surface 1 cones above the mean compared only to the Reynolds numbers used in this study. A
elevation are 4.093 and 5.527 mm, respectively. The heights and typical turbine blade chord Re is 500,000 to 2,000,000. The finite
diameters of the cones on Surface 2 above the mean elevation are range used in this study is in the low to mid-range of typical
1.485 and 3.466 mm, respectively. gas-turbine chord Reynolds numbers.
Skin Friction Measurements
Results and Discussion
The wind tunnel used in this study is housed at the Air Force
Research Laboratory 共AFRL兲 at Wright-Patterson Air Force Base The discrete-element model was run for the cone surfaces using
in Dayton, OH. The wind tunnel is an open-loop tunnel with a each of the following options:
nominal cross section of 0.24 m by 0.38 m in the test section. For 共1兲 the full cones placed on the surface floor with shear
the tests presented here, the wall was adjusted to produce zero acting on the surface floor
freestream acceleration 关7兴. Flow enters the test section through a 共2兲 the partial cones referenced to the mean elevation with
main conditioning plenum and an entry section. As the flow exits shear acting on the area not blocked at the mean
the entry section and just before the flow enters the test section of elevation
the tunnel, the boundary layer is bled from the flow using a knife-
edge boundary-layer bleed and a boundary-layer suction blower. The discrete-element predictions using the floor of the cone sur-
This creates the situation where the boundary-layer thickness is face as a reference, C f,fl, the discrete-element predictions using
zero at the entry of the test section. For all skin friction measure- the mean elevation as a reference, C f,me, and the experimentally
ments, the flow was tripped turbulent using a 1.6 mm diameter measured skin friction coefficients, C f,meas, are presented in Table
cylinder placed 2.54 cm from the knife-edge. Roughness panels, 3 for the high Reynolds number data and in Table 4 for the low
0.32 m in length, were placed 1.04 m from the knife-edge of the Reynolds number data.
test section. The tunnel then continues 0.62 m beyond the trailing The C f predictions for the smooth surface are within 3.5% of
edge of the roughness panels. The freestream turbulence intensity, the experimentally measured values. Tables 3 and 4 also show that
⬘ / Ue, was measured using a hot-wire anemometry system
Tu= urms
and was found to be consistently 0.9% for all of the skin friction
measurements. Table 3 Comparison of the cone and smooth surface skin fric-
The skin friction coefficients were determined by suspending tion results and discrete-element predictions at high Reynolds
the roughness panels in the wind tunnel using wires attached to numbers
the wind-tunnel framing and measuring the downstream move-
ment of the panels during a test. Detailed information on the wind C f,me C f fl
Surface Rex̄ C f,meas 共% diff兲 共% diff兲
tunnel and the average skin friction coefficient determination tech-
nique can be found in Bons 关7兴. Bons 关7兴 reports a systematic Surface 1 944,000 0.01450 0.01396 0.01514
uncertainty 共bias兲 of 0.00022 in the measured skin friction coeffi- 共−3.72兲 共4.41兲
cients with a 2.8% random uncertainty 共repeatability兲. Detailed Surface 2 923,000 0.01040 0.01003 0.01278
information on the simulations and their comparisons to the ex- 共−3.56兲 共22.88兲
perimental data can be found in McClain 关19兴. Smooth 917,500 0.00349 n/a 0.00356
Skin friction measurements were made for each surface at two 共2.0兲
Reynolds numbers based on the distance from the leading edge to
the center of the roughness panel: A high Rex around 1,000,000 Table 4 Comparison of the cone and smooth surface skin fric-
tion results and discrete-element predictions at low Reynolds
numbers
C f,me C f fl
Surface Rex̄ C f,meas 共% diff兲 共% diff兲
the discrete-element predictions using the mean elevation as the decrease in roughness height on the transition of flow from lami-
reference elevation are all within 10% of the measured skin fric- nar to turbulent flow. Even if the flow does not separate, the ac-
tion coefficients. While 10% agreement may not be considered celeration of the fluid at the beginning of the roughness and the
acceptable for the prediction of C f over smooth surfaces, a 10% deceleration of the fluid at the end of the roughness creates an
agreement with experimental data provides a good metric for the increase in form drag on the cone roughness elements.
accuracy of the DEM in comparison to the alternative prediction Neither the effects of the extra form drag on the upstream ele-
techniques for flows over rough surfaces. ments below the mean elevation nor the effects of flow separation
For Surface 2, the discrete-element predictions using the floor are considered in the discrete-element predictions. Taylor 关8兴 cali-
of the cone surface as a reference overestimate the skin friction brated the discrete-element model using Schlichting’s corrected
coefficients by a minimum of 20%. Using the mean elevation as a data and compared the discrete-element predictions to the experi-
reference under predicts the skin friction coefficient by as much as mental data of Healzer 关21兴, Pimenta 关22兴, and Coleman 关23兴. In
7.8% for Surface 2. Since the Surface 2 cones are closely packed, each of the cases studied by Taylor, the roughness started at the
the amount of over prediction by the discrete-element model using “knife-edge” or start of the wind or water tunnel. Skin friction
the flat surface as a reference suggests that using the mean eleva- measurements were then determined using the momentum-defect
tion as a reference is appropriate for closely packed surfaces. method or using the measured Reynolds stresses and mean veloc-
For Surface 1, using the mean elevation underestimates the skin ity profiles. In those test cases, the flow did not transition from
friction coefficients. At the low Reynolds number, using the mean smooth surface to rough surface and back to smooth surface. The
elevation underestimates C f,meas by over 9%. At the high Reynolds changes in mean elevation experienced by the cone surfaces of
number, referencing the flow to the flat part of the cone surface this study cause some of the differences between the discrete-
produces less than 5% difference from the measured skin friction element predictions and the experimentally measured skin friction
coefficient. At the low Reynolds number, referencing the model to coefficients.
the flat part of the cone surface under predicts the skin friction
coefficient by 2%. Implications for the Equivalent-Sand Grain Roughness
Because using the mean elevation as a reference underestimates Model
the skin friction coefficients for Surface 2 and because referencing
the model to the flat part of the Surface 1 underestimates C f,meas Since the effects of the mean elevation are clearly evident in the
by more than the experimental uncertainty, other factors must be discrete-element predictions for flows over closely packed cone
influencing the flow over the cone surface as tested in the experi- surfaces, the same effect should be evident in the data used in the
mental apparatus. One possible cause for the higher measured skin equivalent sand-grain roughness approach. The most widely used
friction coefficients is the change in mean elevation as the flow correlation for using the physical characteristics of a surface to
transitions from smooth surface to rough surface and back to predict the equivalent sandgrain roughness height is based on the
smooth surface along the wind tunnel model. Figure 4 shows the Sigal-Danberg parameter, ⌳s.
configuration of the random roughness panels and the cone rough- Figure 5 presents the ratio of the equivalent sand-grain height
ness panels in the experimental facility. over the ratio of the roughness element height 共ks / k兲 versus ⌳s as
The randomly rough surfaces studied by Bons 关7兴 and McClain traditionally defined by Sigal and Danberg 关6兴 for the classical set
et al. 关16兴 were placed in the wind tunnel such that the mean of surfaces with ordered, three-dimensional roughness elements
elevation of the roughness panels was at the same elevation as the from Schlichting 关1兴关1, as corrected in Ref. 24兴 and for the cones
flat surface of the upstream Plexiglas floor of the wind tunnel. studied by Bogard et al. 关25兴. To investigate the effects of the
Discrete-element predictions were within ±7% of the experimen- mean elevation on surfaces with ordered roughness elements of
tally measured skin-friction coefficients for the randomly rough the same three-dimensional shape, the Sigal-Danberg parameters
surfaces 关16兴. However, the cone surfaces were installed in the were recalculated for the classical surface set using the character-
wind tunnel with the floor of the cone surfaces at the same eleva- istics of the roughness elements above the mean elevation. Figure
tion as the upstream Plexiglas. For the cone surfaces, the place- 6 presents the ratio of the equivalent sand-grain roughness height
ment of the roughness elements creates a situation where the mean to the effective height of the roughness elements above the mean
freestream velocity accelerates as the flow transitions from elevation 共ks / keff兲 versus the Sigal-Danberg parameter evaluated
smooth surface to rough surface. The freestream velocity then using the characteristics of the roughness elements above the
decelerates at the downstream portion of the roughness panel as mean elevation 共⌳s,eff兲.
the flow transitions from rough surface back to smooth surface. If Figure 6 demonstrates that when the mean elevation is consid-
the deceleration is strong enough, the flow will separate as found ered in the determination of the Sigal-Danberg parameter, the val-
by Pinson and Wang 关20兴 when they studied the effects of a step ues of ks / keff correlate well with an exponential function based on
冉 冊冉 冊 冉 冊
at a considerably larger spacing. Once that critical spacing was
2
reached, the equivalent sand-grain roughness height decreased as Sf uk Asm
the spacing was increased. C f = CR + C f,s 共19兲
S pa U⬁ S pa
To explain this decreasing ks as the spacing increased, Schlich-
ting 关1兴 postulated that the total drag force 共F兲 on a rough surface Inspecting Eq. 共19兲, at a set location on a surface with roughness
elements of the same size, as the roughness elements are spaced
was the sum of the skin friction of the flat part of the surface 共Fs兲
further apart, the dominant term affecting the skin friction is the
and drag on the roughness elements 共FR兲. ratio of the frontal area of roughness elements over the plan-form
F = FR + Fs 共16兲 area of the surface. For a set element size, the value of S f / S pa
decreases as a function of the square of the average element spac-
Schlichting 关1兴 then identified a resistance coefficient, CR, of the ing. While uk / U⬁ will change as the elements are spaced further
roughness elements as
apart, the value of uk / U⬁ ranges from 0 to 1.0, and the changes
FR will not be as big as the dramatic decrease in the value of S f / S pa.
CR = 共17兲 Figure 7 demonstrates the expected variation of the skin friction
1 2
u S f coefficient at a set location on a rough surface with a fixed rough-
2 k ness element shape as the spacing between the roughness ele-
Where uk is the velocity of the fluid at the highest elevation of the ments increases. If the effective height of the elements above the
roughness elements and S f is the projected frontal area of all of mean elevation were not relevant, Eq. 共19兲 implies that the skin
friction coefficient of a rough surface would start at a high value
for the most closely packed configuration, then decrease as the
square of the average spacing decreases. If the effective height is
important, the roughness elements would appear to be much
shorter to the flow than the actual height of the elements for the
most closely packed configuration. Thus, the effective projected
frontal area would be much less than the actual projected frontal
area of the roughness elements, and uk / U⬁, would also be signifi-
cantly lower. As the roughness element spacing is increased, the
combined effects of the increasing frontal area and the increasing
uk / U⬁ outpace the increases in S pa until the critical spacing is
reached. When the spacing increases beyond the critical spacing,
the skin friction coefficient then begins to decrease. Recalling that
for a constant element shape that ⌳s varies proportional to the
square of the element spacing and realizing that the equivalent
sand-grain roughness height is directly related to the skin friction
coefficient, the curve reflecting the effect of keff on C f in Fig. 7
explains the trends exhibited by the data in Fig. 5.
Figure 6 also demonstrates that for the surfaces with closely
packed spherical roughness studied by Schlichting 关1兴, the fluid is
Fig. 6 ks / keff vs ⌳s evaluated using the characteristics of the skimming over the roughness elements. In Fig. 6, there are two
roughness elements above the mean elevation sets of points connected by arrows. The set of connected points
labeled “B” represent the surface with the closest possible spacing on Fig. 6. From Eq. 共20兲, the expected skin friction coefficient is
of spheres. The spheres for the set labeled “A” have an average 0.008, which is 20% lower than the measured skin friction coef-
spacing over diameter ratio of 1.5. The closed-circle points at the ficients at either Reynolds number tested.
beginnings of the arrows represent the values of ks / keff and ⌳s
evaluated for the surfaces with closely packed spheres when ref- Conclusions
erenced to the mean elevation above the wall. The open-circle Skin friction coefficients were measured for two cone surfaces
points at the end of the arrows represent the values of ks / keff and at two turbulent Reynolds numbers. Discrete-element model pre-
⌳s evaluated as referenced to the mean elevation above the cen- dictions were made using the modifications to the DEM made for
terline of the spheres. Figure 8 demonstrates the differences in the randomly rough surfaces, the most important of which being the
mean elevation used for the sets of points for the surface most use of the mean elevation as the location of the no-slip plane. The
closely spaced spheres. effects of considering the mean elevation in the evaluation of the
The open-circle points referenced to the mean elevation above equivalent sand-grain roughness heights and in the calculation of
the sphere centerline were calculated after the values of ks / keff and the Sigal-Danberg roughness density parameters was also ex-
⌳s for the most closely packed spheres based on the mean eleva- plored. From the investigation, four major conclusions were
tion above the wall were found to occur in almost the same loca- made:
tion in Fig. 6 as do the points for the closely packed hemispheres
in Fig. 5. The fact that the open-circle points line up very well 共1兲 The relevant surface datum for the evaluation of turbu-
with the trend of the other data representing surfaces more lent boundary layer skin friction over rough surfaces
sparsely spaced roughness elements confirm that the flow below with either the equivalent sand-grain or the discrete-
the centerline elevation is not contributing to skin friction. For the element method is the mean elevation. This is true for
closely packed spheres, the flow is skimming over the spheres. randomly rough and ordered roughness surfaces, though
Morris 关26兴 elaborates on skimming flows similar to the flow over the result only becomes significant as the ordered sur-
the closely packed spheres. face elements become more closely packed.
Comparing the predictions from the discrete-element model and 共2兲 Accounting for the mean surface elevation allows for a
the equivalent sand-grain model would be useful and enlighten- unification of all previous ordered three-dimensional
ing. For the experimental situation studied, however, equivalent roughness element data in the literature. This implemen-
sand-grain model predictions could not be made using the avail- tation replaces the three different equations relating ks / k
able correlations for skin friction coefficient based on equivalent to ⌳s, as reported by Sigal and Danberg, with one uni-
sand-grain height. In the experiments, the flow transitioned from versal relation.
smooth surface, to rough surface, and then back to smooth sur- 共3兲 The mean elevation must be accounted for in experi-
face. The correlations available, such as the one provided by mental research with densely packed roughness surfaces
Schlichting 关27兴, when there are abrupt transitions from the smooth to
冋 冉 冊册 −2.5
rough wall condition. Unless some accommodation is
x made, fluid acceleration over the rough wall region will
C f = 2.87 + 1.58 log 共20兲
ks create elevated levels of skin friction relative to simula-
were developed for flows over surfaces with constant surface tions using a zero freestream velocity gradient assump-
roughness. As discussed earlier, at the beginning of the roughness tion.
section, the skin friction is elevated as the flow adjusts to the 共4兲 For roughness elements that do not have continuously
roughness elements. At the end of the roughness section, the flow decreasing cross section away from the wall 共e.g.,
may separate depending on the magnitude of the change in mean spheres兲, close packing renders the portion of the ele-
elevation from the rough region to the smooth region. Because the ment below the maximum cross-section insignificant for
correlations were not constructed to capture the transition of sur- Cf determination.
face roughness, skin friction coefficient predictions constructed
with the correlations for the current experimental data are much Acknowledgment
lower than either the discrete-element prediction or the experi- This work was performed under partial sponsorship from the U.
mentally measured skin friction coefficient. For example, for Sur- S. Department of Energy, the National Energy Technology Labo-
face 2, the effective Sigal-Danberg parameter is 37.38 and the ratory, and the South Carolina Institute for Energy Studies through
corresponding value of ks / keff 2.2 from the correlation provided the Advanced Land-Based Gas Turbine Systems Research Consor-