You are on page 1of 24

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/262219755

Role of data uncertainties in identifying the logarithmic region of turbulent


boundary layers

Article in Experiments in Fluids · June 2014


DOI: 10.1007/s00348-014-1751-3

CITATIONS READS

33 406

3 authors:

Ricardo Vinuesa Philipp Schlatter


KTH Royal Institute of Technology KTH Royal Institute of Technology
323 PUBLICATIONS 5,783 CITATIONS 423 PUBLICATIONS 13,411 CITATIONS

SEE PROFILE SEE PROFILE

Hassan M Nagib
Illinois Institute of Technology
217 PUBLICATIONS 5,750 CITATIONS

SEE PROFILE

All content following this page was uploaded by Ricardo Vinuesa on 22 May 2014.

The user has requested enhancement of the downloaded file.


Exp Fluids manuscript No.
(will be inserted by the editor)

Role of data uncertainties in identifying the logarithmic region

of turbulent boundary layers

Ricardo Vinuesa · Philipp Schlatter · Hassan M. Nagib

Received: date / Accepted: date

Abstract Composite expansions based on the log law 860, 000, based on displacement thickness and freestream

and the power law were used to generate synthetic ve- velocity. Several artificial errors were added to the ve-

locity profiles of zero pressure gradient turbulent bound- locity profiles to simulate typical measurement uncer-

ary layers in the range of Reynolds number 800 ≤ Reθ ≤ tainties. The effects of the simulated errors were studied

by extracting log-law and power-law parameters from


R. Vinuesa

E-mail: rvinuesa@hawk.iit.edu all these pseudo-experimental profiles. Various techniques

were used to establish a measure of the deviations in the


Department of Mechanical, Materials and Aerospace
overlap region. When parameters extracted for the log
Engineering (MMAE), Illinois Institute of Technology (IIT),
law and the power law are associated with similar lev-
Chicago, IL, 60616, USA.

P. Schlatter els of deviations with respect to their expected values,

E-mail: pschlatt@mech.kth.se we consider that the profile leads to ambiguous conclu-

sions. This ambiguity was observed up to Reθ = 16, 000


Linné FLOW Centre, KTH Mechanics, Royal Institute
for a 4% dispersion in the velocity measurements, up to
of Technology, SE-100 44 Stockholm, Sweden.

H. M. Nagib Reθ = 8.6 × 105 for a 400 µm uncertainty in probe po-

E-mail: nagib@iit.edu sition (in air flow at atmospheric pressure), and up to

Reθ = 32, 000 for 3% uncertainty in the determination


Department of Mechanical, Materials and Aerospace
of uτ . In addition, a new method for the determination
Engineering (MMAE), Illinois Institute of Technology (IIT),

Chicago, IL, 60616, USA. of the log law limits is proposed. The results clearly
2 Ricardo Vinuesa et al.

serve as a further note for caution when identifying ei- The functional form of the overlap region is one of

ther a log or a power law in turbulent boundary layers. the most controversial debates in the turbulence com-

Together with a number of available studies in the lit- munity. Although there is consensus about the proper

erature, the present results can be seen as a additional scaling of the mean velocity in the inner region (U + =

reconfirmation of the log law. U/uτ ), two choices for the scaling of the outer region

lead to two different descriptions of the overlap. If the


1 Introduction
scale is uτ , then the overlap region is described by the

following logarithmic law:


Turbulent boundary layers (TBLs) are divided into two
1
regions according to their classical description (see, for U+ = ln(y + ) + B. (1)
κ
instance, Tennekes and Lumley (1972)). The inner re-
This can be observed in experiments where the skin fric-
gion, which is located close to the wall, is scaled by
tion was independently measured, as in Österlund et al.

the viscous length ` = ν/uτ (where ν is the fluid
(2000). The parameters κ and B in equation (1) have
p
kinematic viscosity, and uτ = τw /ρ is the friction
to be determined experimentally. They are believed to
velocity). The outer region, determined by the convec-
be independent of the Reynolds number, but according
tive effects of the freestream, is usually scaled by the
to Nagib and Chauhan (2008) they vary with the flow
boundary layer thickness δ. In this two-region descrip-
geometry. They also found that for zero pressure gra-
tion, the wall-normal coordinate y is nondimensional-
dient (ZPG) TBLs, the overlap region parameters take
ized using the inner (y + = y/`∗ ) and outer (η = y/δ)
the values κ = 0.384 and B = 4.17.
length scales. In the asymptotic limits of both regions
However, if the outer region is scaled by the freestream
(that is, when y + → ∞ and η → 0 simultaneously),
velocity U∞ , the resulting description of the overlap re-
both descriptions of the velocity profile are valid. This
gion is a power law, as stated by George and Castillo
is the so-called overlap region. The first arguments for
(2006). Their so-called GC power law contains func-
matching the inner and outer parts in the overlap re-
tional forms of the overlap in inner and outer variables,
gion were presented by Millikan (1938). As the theory
given by equations (2) and (3) respectively:
behind singular perturbation problems evolved, deeper γ
U + = Ci y + + a+ , (2)
understanding of multiple scale problems was achieved
 γ
U y+a
(see for example Cole (1968)). = Co . (3)
U∞ δ99
Role of data uncertainties in identifying the logarithmic region of turbulent boundary layers 3

The parameters Ci , Co , γ and a+ are Reynolds number a comprehensive data set of more than 300 mean ve-

dependent. Empirical fits have been developed to ex- locity profiles from more than twenty different sources.

+
press the first three as a function of δ99 in terms of their They concluded that the log law provides a fully self-

respective asymptotic values (Ci∞ = 56.7, Co∞ = 0.897 consistent and accurate description of all the mean quan-

and γ∞ = 0.0362), and a+ is a constant in wall units tities, and demonstrated that the same cannot be achieved

equal to −16. Note that although here we consider the by the competing power law theories. However, other

empirical fits proposed in the original GC formulation more recent publications continue to reach inconclusive

(George and Castillo, 2006), other alternative functions comparisons between the log and power laws using dif-

are available in the literature, such as the ones devel- ferent data sets with varying experimental uncertain-

oped by Barenblatt and Prostokishin (1993). However, ties. For instance, Buschmann and Gad-el Hak (2009)

for the scope of this study we will focus on the GC argue that they found “evidence of nonlogarithmic be-

theory formulation, which is based on the concept of havior” after analyzing a comprehensive database of

“near-asymptotics” to obtain the overlap region solu- low-Re (up to a Reynolds number based on momentum

tion at finite Reynolds numbers. thickness θ of Reθ = 4000) channel flow direct numer-

ical simulations (DNSs), combined with experimental


Although both approaches are based on asymptotic
channel flow data from Zanoun (2003) (up to Reθ =
matching, differences arise from the interpretation of
9000) and the pipe flow data measured by McKeon
available experimental data, as can be observed in the
et al. (2004) at Princeton’s Superpipe (which reaches
work by George (1997), or more recently in the study
Reθ = 1.2 × 106 ).
by Buschmann and Gad-el Hak (2003). Even the exper-

imental data used by Österlund et al. (2000) to support In contrast, Alfredsson et al. (2013) have recently

the log law led to a two-power law description of the presented more arguments in favor of the logarithmic

overlap region when processed and interpreted differ- law as the adequate description of the overlap region,

ently by Barenblatt et al. (2000). As a consequence, with the value of κ = 0.38 which was initially proposed

reliable experimental data are essential to gain some by von Kármán (1930). Our primary objective here is

insight about the appropriate description of the mean to explain the difficulties one encounters in such com-

velocity profile of a TBL. Recently, Monkewitz et al. parisons and to quantify the measures required for an

(2008) evaluated the principal competing theories against accurate assessment. Along these lines, Segalini et al.
4 Ricardo Vinuesa et al.

(2013) have recently published a review showing the particular Reynolds number when they exhibit the

uncertainty in determining κ from experimental data, minimum deviation between the δ, uτ and U∞ val-

when using various methods. ues obtained from both formulations. In order to do

this, the friction velocity uτ and the boundary layer

thickness δ from the log law expansion were used


2 Generation of synthetic data
to calculate the freestream velocity U∞ by means of

The goal of the present study is to evaluate some of the skin friction relation (4). This equation is ob-

the techniques used to analyze experimental data in tained from matching the inner and outer descrip-

wall-bounded turbulent flows. Emphasis was placed on tions of the overlap region in the GC power law

studying how accurately these techniques could deter- theory:

mine the appropriate description of the overlap region,  


+ Ci∞ + γ∞ −A
U∞ = δ99 exp + α , (4)
as well as its parameters. The range of Reynolds num- Co∞ (ln δ99 )

bers Reθ based on momentum thickness θ under con- where the constants A and α take the values 2.9 and

sideration was: 800 ≤ Reθ ≤ 860, 000. Synthetic ZPG 0.46 respectively.

TBL velocity profiles were generated as follows: These parameters were then used to generate the

outer expansion of the GC power law (see Appendix


1. The composite expansion by Chauhan et al. (2009)
for details on its formulation), which is valid in the
(see Appendix for the complete analytical form) was
+
range 30 ≤ y + ≤ δ99 . This outer expansion was
used first to generate the velocity profiles based on
enough to fulfill the purposes of the present study.
the log law. Then the curve was discretized, with a
Also 100 points were used to discretize the power
total of 100 collocation points logarithmically spaced
law profile.
along the profile.

2. An equivalent ZPG TBL profile based on the GC It is important to note that the use of uτ and δ

power law (initially formulated by George and Castillo from the log law in equation (4) to obtain the U∞ for

(2006), and further refined by George (1997, 2007)), the power law is the only possible choice. Doing so,

was used to generate an equivalent set of points. we are using the appropriate length scales in both de-

Note that in this context we consider a log law scriptions. In fact, the freestream velocities obtained for

profile to be “equivalent” to a power law one at a the power law were on average only 3% lower than the
Role of data uncertainties in identifying the logarithmic region of turbulent boundary layers 5

corresponding U∞ values from the log law. Two other two data sets spanning the following Reynolds number

possibilities were investigated, confirming the previous intervals:

statement:

1. Use uτ and U∞ from the log law profile to obtain


1. Purtell et al. (1981): 465 ≤ Reθ ≤ 5100
δ for the power law profile using equation (4). This
2. Smith and Walker (1959): 3000 ≤ Reθ ≤ 50, 000
choice is not correct, since these velocity scales are

valid for the power law, but not for the log law,

so the resulting TBL thicknesses for the power law


Note how the highest Reynolds number of the more
were on average 28% lower than the log law values.
recent dataset by Purtell et al., Reθ = 5100, is the
2. Use U∞ and δ from the log law profile to obtain the
one where discrepancies between log law and power law
uτ for the power law profile by means of equation
profiles start to become noticeable in the beginning of
(4). This procedure is also incorrect, since these are
the overlap region. The power law parameters are Re-
the appropriate outer scales for velocity and length
dependent, but this dependence is introduced through
in a power law, but not in a log law. Consequently,
empirical fits based on a limited Re range. This is the
the calculated friction velocities for the power law
reason why differences between both theories increase
were on average 14 times higher than the log law
with Reynolds number.
equivalent values.

Log law and power law velocity profiles are shown It is also important to state that increasing differ-

in Figure 1 for four Reynolds number cases. It can be ences between both formulations with Reynolds num-

observed how the overlap regions from both theories are ber are found in the overlap region due to the particular

very similar for Reθ ≤ 2, 500. However, for Reθ > 5100, choice of scales described above. In other words, fixing

+
differences between the profiles defined by both theories δ and uτ is equivalent to fixing U∞ , with a different

start to appear around y + = 30. This phenomenon can freestream velocity U∞ . Application of the GC power

be explained by the fact that, as mentioned in the In- law expansion subjected to these constraints leads to

troduction, the power law parameters Ci , Co and γ are the behavior observed in Figure 1, which in fact shows

obtained in terms of their asymptotic values through “equivalent” profiles at the various Reynolds numbers

empirical fitting. These empirical relations are based on in the context discussed here.
6 Ricardo Vinuesa et al.

Reθ = 800 Reθ = 2, 500


35 35

30
a) Log law
30
b)
Power law

25 25

20 20
U+

U+
15 15

10 10

5 5

0 0 1 2 3 4
0 0 1 2 3 4
10 10 10 10 10 10 10 10 10 10
+ +
y y
Reθ = 5, 100 Reθ = 16, 000
35 35

30
c) 30
d)

25 25

20 20
U+

U+

15 15

10 10

5 5

0 0 1 2 3 4
0 0 1 2 3 4
10 10 10 10 10 10 10 10 10 10
+ +
y y

Fig. 1 Mean velocity profiles scaled in inner variables generated from log law and power law theories.

3 Analysis of synthetic data profile, Ξ will take the constant value 1/κ within it. The

so-called von Kármán coefficient κ obtained using this


A total of 22 synthetic profiles were generated, 11 based
technique was denoted by κΞ . All the derivatives were
on each of the two overlap region theories. Then, they
evaluated numerically using a 5 point finite-difference
were analyzed using the following techniques:
scheme with non-uniform grid spacing.

3.1 Log law techniques

An alternative approach based on a logarithmic fit


The log law diagnostic function Ξ is defined as:
to equation (1) in the overlap region was also consid-
dU +
Ξ= y+ + . (5)
dy ered. This leads to the simultaneous determination of κ

Comparing equations (5) and (1) it can be observed and B, but requires the use of a criterion to determine

that if a logarithmic overlap region exists in the velocity where the overlap region is located. The parameters
Role of data uncertainties in identifying the logarithmic region of turbulent boundary layers 7

extracted using this technique were denoted by κF and study, we developed a method to identify the logarith-

BF . mic overlap region, based on the inner deviation from

the log law Ψi and the diagnostic function Ξ. Ψi is de-

3.2 Power law techniques fined as follows:

1
Ψi = U + − ln(y + ), (7)
The power law diagnostic function Γ is: κ

y + + a+ dU + which by comparison with (1) is constant and equal to


Γ = . (6)
U + dy +
B within the overlap region if this is described by a
It can be inferred, from comparison of equations (6)
log law. If one fixes the lower and upper limits of the
and (2), that Γ is a constant equal to γ in the overlap +
logarithmic overlap region ylog,min and ηlog,max (note
region if it is described by a power law. The exponent
how the inner limit scales in wall units and the outer
γ obtained using this approach was denoted by γΓ .
limit scales with δ), it is possible to use a secant algo-
Alternatively, power fits to equations (2) and (3),
rithm with (7) to obtain the von Kármán coefficient κ.
in inner and outer variables respectively, can be used +
The approach we considered here was to set ylog,min ,
to obtain the fitted constants CiF , CoF and γF . As ex-
and then find the corresponding κ for an array of pos-
pected from the construction of the profiles, the value of
sible ηlog,max values. This produces, for each possible
the fitted exponent γF using inner and outer variables +
ylog,min , a curve of κ vs ηlog,max . Then, we use spline
was virtually identical in all the cases. This technique
interpolation to obtain the ηlog,max corresponding to
also requires the use of a criterion to determine the lo-
the κ obtained from the composite fit of Chauhan et al.
cation of the overlap region in the first place.
(2009). We decided to carry out this procedure with

+
ylog,min values of 100 and 200, which are the most widely
3.3 Criteria for determining the limits of the overlap
reported in the literature. Application of this method
region
to the velocity profiles by Österlund et al. (2000) with

A wide range of criteria for a log-law-based overlap re- Reθ > 6000 (which, according to the authors, is the

gion have been reported in the literature, and they have minimum Reynolds number for a significant overlap re-

evolved with time as higher Reynolds number data have gion to exist) led to the two following criteria: y + > 100,

become available. An extensive review of the different η < 0.18 ± 0.02 and y + > 200, η < 0.15 ± 0.02. Figure 2

criteria can be found in Örlü et al. (2010). In the present a) and c) shows Ψi calculated from some sample profiles
8 Ricardo Vinuesa et al.

Log law limits: y+


log,min = 100 & ηlog,max = 0.18 Log law limits: y+
log,min = 100 & ηlog,max = 0.18
5.5 4
a) b)
5 3.5

4.5 3
Ψi

Ξ
4 2.5

3.5 2

3 1.5
2 3 2 3
10 10 10 10
+ +
y y
Log law limits: y+
log,min = 200 & ηlog,max = 0.15 Log law limits: y+
log,min = 200 & ηlog,max = 0.15
5.5 4
c) d)
5 3.5

4.5 3
Ψi

4 2.5

3.5 2

3 1.5
2 3 2 3
10 10 10 10
+ +
y y

Fig. 2 a) and c) Inner deviation from the log law Ψi , b) and d) Log law diagnostic function Ξ, for inner-scaled mean velocity
+
profiles by Österlund et al. (2000) with Reθ > 6000. Dots represent corresponding log law limits given by ylog,min and ηlog,max

on each profile.

by Österlund et al. (2000), together with the overlap re- where perhaps the narrower region from c) can be con-

gion predicted by the two criteria. Note that each curve sidered to be slightly flatter. The diagnostic function Ξ

was determined with the particular κ value of the corre- calculated from the same profiles is plotted on Figure 2

sponding velocity profile (ranging from around 0.382 to b) and d), and in this case the data more clearly shows

0.386), which was calculated from the composite profile that the best criterion to represent the extent of the

of Chauhan et al. (2009). Therefore, in this approach logarithmic region is y + > 200, η < 0.15, since below

Ψi is not used to determine κ: it is used to obtain the y + ' 200 the Ξ function presents a valley which was

extent of the overlap region given the value of κ. Both also observed by Monkewitz et al. (2007). Therefore, a

criteria seem to highlight regions of roughly constant Ψi , logarithmic overlap region can be found for y + > 200,
Role of data uncertainties in identifying the logarithmic region of turbulent boundary layers 9

η < 0.15, a conclusion in agreement with Nagib et al. which are based on the asymptotic expansions by Chauhan

(2007), who used a different analysis method. et al. (2009) and George and Castillo (2006) for the log

The same analysis applied to the profiles from Österlund and power laws respectively. These limits are used to ex-

et al. (2000) with Reθ < 6000 yielded the criterion tract the values of the log law and power law constants

y + > 50, η < 0.2±0.04, which interestingly had already from the velocity profiles. A summary of the cases con-

been proposed by Coles (1954) much earlier. It is inter- sidered in the present study is presented next.

esting to remark that the dispersion associated with


3.4 Summary of Reynolds number cases
ηlog,max is larger for the lower Reynolds number cases

than for the high-Re profiles (±0.04 versus ±0.02). This


Table 1 shows the different cases under consideration in
is a consequence of the various low-Re effects and is in
the present study. Cases 1–4 are considered “low-Re”
agreement with the significant dispersion found in the
since, as already discussed, a significant logarithmic
literature (Örlü et al., 2010) (especially when only low
overlap region exists for Reθ > 6000 (Österlund et al.,
Reynolds number data was available).
2000). Buschmann (2000) observed that for Reθ < 2000
To summarize, the following criteria will be used in
(Cases 1–3) the extension of the logarithmic region tended
our log-law-based analysis for the rest of the paper:
to zero. According to him, these cases exhibit incom-

1. For Reθ < 6000, the criterion will be y + > 50, η < plete similarity (as postulated by Barenblatt (1993, 1996))

0.2. and a power law might be the best fit. However, Buschmann

2. For higher Reynolds numbers, we will consider the (2000) also states that the logarithmic law is the appro-

following criterion: y + > 200, η < 0.15. priate description of the overlap region for high enough

With respect to the power-law-based analysis, we Reynolds numbers.

considered it adequate to use the criterion proposed by Cases 5–9 are considered “high-Re”, since all of them

George (2007), where the overlap region is found in the present a separation of scales which was sufficient to ob-

interval y + > 30, η < 0.1. All the profiles were analyzed serve a significant logarithmic region. Although there is

using this criterion, except the lowest Reynolds Reθ = still some debate about the minimum Reynolds number

800, where the outer limit had to be extended to 0.12. to ensure high-Re conditions (Nishioka (2010) recently

Let us recall that the overlap region limits are not published a theoretical study taking into account the

required to generate the synthetic profiles themselves, non-parallel flow effect suggesting a minimum Reθ of
10 Ricardo Vinuesa et al.

Table 1 Summary of Reynolds number cases and limits considered for the log and power law regions.

+ +
Case Reθ Reδ∗ δ+ 3δ +1/2 ylog,min ηlog,max ypow,min ηpow,max Re type

1 800 1200 405 60 50 0.20 30 0.12 Low-Re

2 1400 2000 675 78 50 0.20 30 0.10 Low-Re

3 2000 2800 920 91 50 0.20 30 0.10 Low-Re

4 2500 3600 1125 101 50 0.20 30 0.10 Low-Re

5 7500 10,000 3050 166 200 0.15 30 0.10 High-Re

6 16,000 22,000 6400 240 200 0.15 30 0.10 High-Re

7 27,000 35,000 10,400 306 200 0.15 30 0.10 High-Re

8 32,000 41,000 12,200 331 200 0.15 30 0.10 High-Re

9 82,000 100, 000 30,600 525 200 0.15 30 0.10 High-Re

10 4.0 × 105 5.0 × 105 1.5 × 105 1162 200 0.15 30 0.10 Extrapolation

11 8.6 × 105 1.0 × 106 3.1 × 105 1670 200 0.15 30 0.10 Extrapolation

20,000), for the purpose of the present study we will for Reθ > 6000, ηlog,max = 0.15 (which is representa-

consider Reθ = 6000 to be sufficient. Case 9 (Reθ = tive of other values reported in the literature between

82, 000) corresponds approximately to the highest Re 0.1 and 0.2), and they consider an increasingly higher

+
data achieved experimentally with good measurement ylog,min value based on the arguments by Klewicki et al.

resolution. Thus, cases 10 and 11 are extrapolations (2009). Note that the goal of Marusic et al. (2013) was

from the available experimental data. not to determine these limits, and they were adopted

for the purpose of calculating overlap region parame-


Table 1 also shows, for each case, the boundary layer
ters. By comparing the values in Table 1 with the cri-
thickness scaled in wall units δ + (which is equivalent to
teria reported here, one can observe how both crite-
the friction Reynolds number Reτ ) and the quantity
ria yield similar values in the first low-Re profile (50
3δ +1/2 . These values are useful to compare the loga-
versus 60) and the first high-Re case (200 and 166).
rithmic region limits reported in this study with the
Higher Reynolds numbers provide increasingly higher
ones discussed by Marusic et al. (2013) in a recent re-
lower limits of the overlap region, reaching values above
view on the logarithmic region in wall-bounded turbu-
1000+ in the extrapolated Reynolds number cases. Note
lence: y + > 3δ +1/2 and η < 0.15. In their study they
that Figure 2 shows a roughly constant lower limit of
adopt the same outer limit as the one we propose here
Role of data uncertainties in identifying the logarithmic region of turbulent boundary layers 11

+
the log region of around ylog,min ' 200 for velocity pro- associated with the power law parameters would be

files at Reynolds numbers up to Reθ ' 25, 000, which much larger than the ones associated with the log

according to the criterion discussed by Marusic et al. law parameters if the initial curve was described by

(2013) would exhibit a logarithmic region starting at a log law.

around 300 wall units. 4. As the introduced error level is increased, the rel-

ative errors from both sets of parameters will be

more similar. Thus, it will be harder to tell if the


4 Simulated measurement errors
original curve was a log law or a power law if the

After generating the synthetic data sets, various mea- measurement errors are high.

surement errors were introduced. Then, those data points


We considered the following simulated errors:
were treated as pseudo-experimental profiles, and were

analyzed according to the following procedure:

4.1 Noise addition


1. All the log law and power law parameters were ex-

tracted from all the pseudo-experimental profiles, Various noise levels were introduced to simulate disper-

regardless of their initial overlap region description. sion in velocity measurements when using hot-wires or

For example, from a log law profile, all the log law Pitot tubes. For a given profile, noise was added only to

(κF , BF , κΞ ) and power law (γF , CiF , CoF , γΓ ) coef- the U + values, and thus we assumed no error in probe

ficients would be calculated using the corresponding positioning. A Gaussian distribution centered on the

analysis techniques. original U + value from the profile was considered to

2. Then the relative errors between the extracted pa- model this effect, and the introduced level of noise was

rameters and the values used to generate the com- controlled by means of its standard deviation. Given a

posite profiles were calculated. Gaussian distribution with mean µ and standard devi-

3. If the introduced measurement error level is low, ation σ, the area under the curve between µ − nσ and

it will be possible to unambiguously identify if a µ + nσ is given by erf(n/ 2), where n is an integer

given pseudo-experimental profile had been gener- and erf is the error function. For n = 3, the area un-

ated from an initial log law or a power law profile. der the curve is 99.7%, which means that this percent-

Following the previous example, the relative errors age of random points will be introduced in the interval
12 Ricardo Vinuesa et al.

[µ−3σ, µ+3σ]. This is the so-called 3σ criterion, which als beyond 1000 profiles. This means that, for each

will be considered in this study to adjust the width of Reynolds number, we added random noise to 1000 an-

the normal distribution governing the added points in alytical profiles as in Figure 3 and extracted eκF from

terms of the level of noise introduced to the analytical each of them; then the average error is the value shown

profiles. The value of σ is then calculated, for each U + in Figure 4. The results are almost identical for κΞ and

value in the profile, as: appear qualitatively the same for BF , although the er-

Noise [%] U + ror levels are slightly larger. When no noise was added,
σ= · , (8)
100 3
the values of κ obtained from the pseudo-experimental
where Noise [%] is the percentage of noise to be added.
profiles generated from a log law showed smaller devi-
As discussed in §2, 100 collocation points were used to
ations than the ones from a power law, as expected.
generate the synthetic profiles. The extracted overlap
However, the case Reθ = 1400 was an exception: even
constants were observed to be independent of the num-
if no measurement error was introduced, for this partic-
ber of points in the profiles when we considered more
ular Reynolds number it was not possible to tell which
than 100 points. Figure 3 shows the effect of 2% Gaus-
one was the original curve (a log-law or a power-law-
sian noise addition to log-law and power-law composite
based profile). When noise was increased to 0.5% (Fig-
expansions at Reθ = 32, 000. Note that in this figure
ure 4b)), the Reθ = 2000 profile also presented ambi-
we show the result of adjusting the standard deviation
guity about which one was the description the original
σ of a single normal distribution following (8).
profile. For higher noise levels like 2% and 4%, ambigu-
Let us analyze the impact of introducing Gaussian
ity was observed up to Reθ values of 7500 and 16,000
noise in the calculated overlap region parameters. Fig-
respectively. Note that a velocity dispersion of 4% is
ure 4 shows the relative error in κF (i.e., determined
above acceptable experimental uncertainty, but this re-
from a fit to equation (1) within the corresponding log
sult is simulated to illustrate the propagation of this
law intervals), as a function of Reθ for various noise
kind of ambiguity to higher Reynolds numbers.
error levels. Since the added noise is a random vari-

able, a stochastic analysis needs to be performed on A similar analysis was done using the χ2 distribu-

a sufficiently large number of samples in order to ob- tion, and no significant differences were observed in any

tain consistent results. The errors shown in Figure 4 of the cases with respect to the Gaussian noise distribu-

were found to be independent of the number of tri- tion. Thus, the previous conclusions based on a centered
Role of data uncertainties in identifying the logarithmic region of turbulent boundary layers 13

40

35

30

25

U + 20
15

10
Log profile
Log pseudo−exp.
5
Power profile (+6)
Power pseudo−exp. (+6)
0
0 1 2 3 4
10 10 10 10 10
+
y

Fig. 3 Pseudo-experimental velocity profiles with simulated velocity dispersion of 2% at Reθ = 32, 000. Comparison of log-law

and power-law composite expansions, where the power law profile is shifted vertically 6 units for clarity. Black squares denote

the extent of the respective overlap regions.

statistical distribution can be extended to non-centered bers lead to an increasingly large contaminated region.

distributions, which will not be further considered on This effect was simulated by generating a shifted y + ar-

+
this study. ray (yshift = y + + d+ ) for each Re case. The new set of

U + measurements was obtained by using spline inter-

polation, but the initial y + array was kept for the subse-
4.2 Uniform vertical shift
quent analysis. This represents the fact that the shifted

This error was introduced to simulate an uncertainty in probe measures a different velocity, but it is believed to

the probe position with respect to the wall. Although be in the unshifted position. Figure 5 shows the effect

the shift d is fixed in physical units, d+ = duτ /ν is Re- of introducing a 400 µm shift to log law and power law

dependent since the friction velocity uτ increases with composite expansions for Reθ = 32, 000. Note that the

Reynolds number. While at low Re only the near-wall effect of the shift, constant in physical units, increases

region is affected by this error, higher Reynolds num-


14 Ricardo Vinuesa et al.

60
Gaussian noise addition: 0% 60
Gaussian noise addition: 0.5%
a) Log profile
Power profile
b)
50 50

40 40
eκF [%]

eκF [%]
30 30

20 20

10 10

0 3 4 5 6
0 3 4 5 6
10 10 10 10 10 10 10 10
Reθ Reθ

60
Gaussian noise addition: 2% 60
Gaussian noise addition: 4%
c) d)
50 50

40 40
eκF [%]

eκF [%]
30 30

20 20

10 10

0 3 4 5 6
0 3 4 5 6
10 10 10 10 10 10 10 10
Reθ Reθ

Fig. 4 Relative error in κF when estimating overlap region parameters by using log law and power law pseudo-experimental

profiles. Gaussian noise was introduced in all the profiles. Vertical dashed lines separate “low-Re”, “high-Re” and extrapolation

cases, whereas the horizontal line represents 5% error level. References from the text are highlighted with black ellipses.

with Reynolds number due to the fact that the viscous profiles. We choose κF and γF to show the trends, which

length scale `∗ gets smaller. are comparable for the rest of characteristic parameters

from log and power laws.


Different accuracies in probe positioning are reported

in the literature. Most of the reported error levels for The first conclusion inferred from Figure 6 is the

hot-wires and Pitot tubes are in the range of 5-200 µm fact that the log law parameters are less sensitive to

(Örlü et al., 2010). In Figure 6a) and b) we show, for ini- uncertainties in wall position than the power law ones.

tial log and power law profiles, the errors in determining If a shift in probe position of 200 µm is artificially sim-

the log law parameter κF for different introduced shifts. ulated, κ becomes ambiguous only up to Reθ = 1400,

Note that different symbols are used in Figure 6c) and whereas the same shift produces ambiguity in the power

d) because there we do the inverse exercise: we extract law parameter γ up to Reθ = 7500. The reason for this

the power law parameter γF from initial log and power is the fact that the beginning of the overlap region for a
Role of data uncertainties in identifying the logarithmic region of turbulent boundary layers 15

40

35

30

25

U + 20
15
Log profile

10 Log pseudo-exp.

Power profile (+6)


5
Power pseudo-exp. (+6)
0
0 1 2 3 4
10 10 10 10 10
+
y

Fig. 5 Pseudo-experimental velocity profiles with a simulated shift in probe position of 400 µm at Reθ = 32, 000. Comparison

of log-law and power-law composite expansions, where the power law profile is shifted vertically 6 units for clarity. Black

squares denote the extent of the respective overlap regions.

power law analysis is y + = 30, while it is y + = 200 for points available for evaluating γF and thus the estima-

a log law analysis with Reθ ≥ 7500. Thus, the begin- tions get progressively better. However, after significant

ning of the overlap region will be contaminated by an numbers of points are found in the overlap region, the

uncertainty in probe position at lower Reynolds num- error becomes magnified with Re because the error in

bers if a power law is considered than if we assume a probe position gets larger when expressed in wall units,

log law. Also note the evolution of eγF in Figures 6c) thus contaminating a larger portion of the overlap re-

and d): it initially decreases for increasing Reynolds gion. Another important conclusion can be observed

numbers, then after Reθ = 7500 it gets progressively in Figure 6d), where the effect of introducing a shift

larger. The initial decreasing region is explained by of 400 µm is shown. If a power law analysis is consid-

the fact that the extent of the overlap region increases ered, even the highest Reynolds number Reθ = 860, 000

at higher Reynolds numbers, therefore there are more becomes ambiguous. As mentioned above, the highest
16 Ricardo Vinuesa et al.

Probe shift: 25 µm Probe shift: 200 µm


60 60
a) Log profile
Power profile
b)
50 50

40 40
eκF [%]

eκF [%]
30 30

20 20

10 10

0 3 4 5 6
0 3 4 5 6
10 10 10 10 10 10 10 10
Reθ Reθ
Probe shift: 200 µm Probe shift: 400µm
60 60
c) Log profile
Power profile
d)
50 50

40 40
eγF [%]

eγF [%]
30 30

20 20

10 10

0 3 4 5 6
0 3 4 5 6
10 10 10 10 10 10 10 10
Reθ Reθ

Fig. 6 Relative error in a), b) κF and c), d) γF when estimating overlap region parameters by using log law and power

law pseudo-experimental profiles. Shifts in probe position were introduced in all the profiles. Vertical dashed lines separate

“low-Re”, “high-Re” and extrapolation cases, whereas the horizontal line represents 5% error level. References from the text

are highlighted with black ellipses.

uncertainty in probe position observed in experimen- this parameter is used to scale both U and y. Tradi-

tal studies of wall-bounded turbulence is on the order tionally, the wall shear stress τw was obtained from

of 200 µm, but we believe it is illustrative to include a fit in the overlap region with universally accepted

this higher simulated shift to observe how quickly am- log law constants (this is the so-called Clauser chart

biguity extends to very high Reynolds numbers with method (Clauser, 1954)). However, independent wall

increasingly high errors in probe location. shear stress measurements are essential in order to ob-

tain meaningful information from experimental data, as

4.3 Uncertainty in wall shear stress measurements pointed out by Nagib et al. (2004b). Let us analyze the

effect of using erroneous uτ values to scale the veloc-


If the friction velocity uτ is not measured accurately,
ity profiles. If the friction velocity is overestimated, the
the whole non-dimensional profile will be affected, since
Role of data uncertainties in identifying the logarithmic region of turbulent boundary layers 17

40

35

30

25

U + 20
15
Log profile

10 Log pseudo-exp.

Power profile (+6)


5
Power pseudo-exp. (+6)
0
0 1 2 3 4
10 10 10 10 10
+
y

Fig. 7 Pseudo-experimental velocity profiles with a simulated uncertainty in uτ of ±3% at Reθ = 32, 000. Comparison of

log-law and power-law composite expansions, where the power law profile is shifted vertically 6 units for clarity. Black squares

denote the extent of the respective overlap regions.

scaled profile will be below the original one, while the to happen was lower if the friction velocity was underes-

opposite will be observed if uτ is underestimated. Both timated than if it was overestimated in the experiments.

cases can be appreciated in Figure 7, where the effect The opposite was observed when the power law param-

of using a friction velocity 3% lower or higher than the eters were analyzed, since an overestimation of 2% in uτ

real one to scale the profile is illustrated. The Reynolds was enough to see ambiguity in three profiles when ex-

number was Reθ = 32, 000 in this case. tracting γF (Figure 8c)). This was not observed when

the friction velocity was underestimated. An extreme

Figures 8a) and b) show eκF for measured uτ values case can be observed in Figure 8d), which shows eCoF

5% higher and 3% lower than the original ones respec- (which is the error in determining the constant Co from

tively. In principle, both cases contain two profiles with a fit to the power law overlap region equation expressed

ambiguity at low Reynolds numbers. Nevertheless, it is in outer units (3)) when uτ was overestimated by 3%:

interesting to note that the error level required for this


18 Ricardo Vinuesa et al.

cases 1–6 where clearly ambiguous (Reθ ≤ 16, 000), and to Reθ = 16, 000 for 4% noise (which is above ac-

cases 7–8 (Reθ = 27, 000 and 32, 000) may also present ceptable levels of experimental uncertainty).

ambiguity. 2. The power law analysis techniques were more sensi-

tive to probe shift errors, since the beginning of the

overlap region is closer to the wall (y + = 30), and


5 Summary and conclusions
thus it can be contaminated by wall-position errors

Several widely used techniques for analyzing experi- more easily. A probe shift of 200 µm (for air flow at

mental data in wall-bounded turbulence were tested. atmospheric pressure), which is representative of the

Synthetic velocity profiles were generated in the range highest errors in probe position observed in experi-

800 ≤ Reθ ≤ 860, 000 by using composite expansions mental studies, produced ambiguous conclusions up

based on log-law and power-law descriptions of the over- to Reθ = 7, 500. In order to show how quickly this

lap region. Different measurement errors were then added effect contaminates the velocity profiles even at high

to the synthetic profiles, and the resulting data were Reynolds numbers, a probe shift of 400 µm (which is

treated as pseudo-experimental data series. All the pro- above acceptable values of experimental error) made

files were analyzed following the same procedure, re- even the highest Re profile (Reθ = 8.6 × 105 ) am-

gardless of their original overlap region description. biguous. This is important since this value exceeds

Subsequently, the log-law and power-law parameters by a factor of 10 the largest laboratory Reynolds

were extracted from all the profiles. In general it was numbers currently available.

easy to identify if the original curve was a log law or 3. When the friction velocity was underestimated, it

a power law for low error levels. However, some inter- was easier to misinterpret an initial power law curve

esting conclusions arose when higher error levels were as a log law curve. An error level of 3% extended

considered: this ambiguity for profiles up to Reθ = 2000. How-

ever, the opposite mistake was more likely to oc-


1. When random noise was added, the techniques un-
cur when uτ was overestimated. An error level of
der study had problems identifying if a given curve
2% was enough to observe ambiguous results in the
was a log law or a power law even at very high
same Reθ range. In addition, an overestimation of
Reynolds numbers. This ambiguity was observed up

to Reθ = 7, 500 for 2% noise addition, and went up


Role of data uncertainties in identifying the logarithmic region of turbulent boundary layers 19

Friction velocity uncertainty: +5% Friction velocity uncertainty: -3%


60 60
a) Log profile
Power profile
b)
50 50

40 40
eκF [%]

eκF [%]
30 30

20 20

10 10

0 3 4 5 6
0 3 4 5 6
10 10 10 10 10 10 10 10
Reθ Reθ
Friction velocity uncertainty: +2% Friction velocity uncertainty: +3%
60
60
c) d) Log profile
Power profile
50 50

40
eCoF [%]
40
eγF [%]

30 30

20 20

10 10

0 3 4 5 6
0 3 4 5 6
10 10 10 10 10 10 10 10
Reθ Reθ

Fig. 8 Relative error in a), b) κF , c) γF and d) CoF when estimating overlap region parameters by using log law and power

law pseudo-experimental profiles. Different levels of uncertainty in uτ were introduced in all the profiles. Vertical dashed lines

separate “low-Re”, “high-Re” and extrapolation cases, whereas the horizontal line represents 5% error level. References from

the text are highlighted with black ellipses.

uτ by 3% produces ambiguous conclusions up to is slightly above reported values in the literature, may

Reθ = 32, 000. lead us to re-think the conclusions we draw from the

highest Reynolds number data currently available.


The fact that, for instance, an uncertainty of 3% in

uτ can lead to ambiguous results up to quite a large The present study should serve as a lesson, not only

Reynolds number of Reθ = 32, 000, has an important on careful experimentation, but more importantly on

impact on the interpretation of data; only Oil Film In- the careful selection of appropriate databases to “learn

terferometry (OFI) and few other techniques not widely about turbulence”. Very subtle effects may lead to ex-

used in the turbulence community are able to provide tended debates such as the log law versus power law,

error levels below this. Besides, 3% dispersion in ve- the extent of the overlap region, or the universality

locity or 400 µm uncertainty in probe position, which and values of the log law parameters; i.e., effects which
20 Ricardo Vinuesa et al.

+
may be in the end caused by small measurement er- The inner profile Uinner was developed by Musker (1979),

and is given by:


rors rather than the physics of the flow. The “evidence

y+ − a R2
 
1
of non-logarithmic behavior” observed by Buschmann U
+
inner
= ln + ×
κ −a a(4α − a)
( p !
and Gad-el Hak (2009) might be explained by the fact × (4α + a) ln −
a (y + − α)2 + β 2
+
R y+ − a

that the computational data they used are at Reynolds )


y+ − α
h    i
α α
+ (4α + 5a) arctan + arctan , (10)
β β β
numbers too low to exhibit a significant overlap re-

gion. Also, as recently discussed by Vinuesa (2013) the where α = (−1/κ − a)/2, β = −2aα − α2 and R =
p
higher-Re experimental datasets exhibit various flaws α2 + β 2 . The parameters κ = 0.384 and a = −10.3061

which question the validity of the conclusions extracted give B = 4.17. The exponential term in equation (9)

from them. The large ambiguity exposed by the present was added to account for an overshoot observed in the

paper demonstrates that care has to be taken when experimental data around y + ' 50, as shown by Monke-

identifying either a log or a power law in turbulent witz et al. (2007).

boundary layers. For instance, the results presented here Following Coles (1956), a wake function W(η) was

question the analysis done by Buschmann and Gad-el introduced in equation (9) to represent the outer re-

Hak (2003) on low Reynolds number profiles, and high- gion, with Π being a parameter associated with the

Re data where a plausible small probe shift in physical wake strength. Experimental measurements in the NDF

units becomes greatly magnified when scaled in wall wind tunnel at IIT (Nagib et al., 2004a) were used by

units. Together- with a number of available studies in Chauhan et al. (2007) to develop and exponential wake

the literature, the present results can be seen as a fur- function Wexp of the form:
 
ther reconfirmation of the log law. 1 − exp −(5a2 + 6a3 + 7a4 )η 4 /4 + a2 η 5 + a3 η 6 + a4 η 7
Wexp (η) =
1 − exp [−(a2 + 2a3 + 3a4 )/4]
 1 
× 1− ln(η) , (11)

Appendix
where a2 = 132.841, a3 = −166.2041 and a3 = 71.9114.

The analytical forms of the composite expansions used The wake parameter Π is calculated following the method
in the present study are included in this section. The
described by Chauhan et al. (2009).
velocity profiles based on a logarithmic description of
Velocity profiles based on the GC power law de-
the overlap region were generated from the composite
profile of Chauhan et al. (2009): scription of the overlap region were generated from the
 2 +

+ +
exp − ln y /30 2Π
Ucomposite = Uinner + + W(η). (9) outer profile proposed by George and Castillo (2006),
2.85 κ
Role of data uncertainties in identifying the logarithmic region of turbulent boundary layers 21

and developed by George (1997, 2007): Barenblatt, G. I. (1996). Scaling, Self Similarity, and
  
+ πy Ci + γ Intermediate Asymptotics. Cambridge Texts in Appl.
Uouter = (0.99 − Co )y sin (δ99 ) +
2 Co
Mathematics. University Press, Cambridge.
+Ci (y + + a+ )γ , (12)
Barenblatt, G. I., Chorin, A. J., and Prostokishin, V. M.
where a+ = −16 and y ≡ y/δ99 . This equation, valid

+
(2000). A note on the intermediate region in turbu-
in the interval 30 < y + < δ99 , was enough for the
lent boundary layers. Phys. Fluids, 12:2159–2161.
comparison purposes of the present study. The param-
Barenblatt, G. I. and Prostokishin, V. M. (1993). Scal-
eters Ci , Co and γ are Re-dependent, and they are ex-
ing laws for fully developed turbulent shear flows.
pressed in terms of the asymptotic values Ci∞ = 56.7,
Part 2. Processing of experimental data. J. Fluid
Co∞ = 0.897 and γ∞ = 0.0362 through the following
Mech., 248:521–529.
empirical relations:
Buschmann, M. (2000). Power law or logarithmic law?
αA
γ = γ∞ +  1+α , (13)
+
ln δ99 – a data analysis for zero pressure gradient turbu-
+
 
Co = Co∞ 1 + 0.283 exp −0.00598δ99 , (14)
( ) lent boundary layers with low Reynolds numbers. J.
Co Co∞ (1 + α)A
= exp  + α
, (15) Thermal Science, 9:23–29.
Ci Ci∞

ln(δ99 )

where γ → γ∞ , Ci → Ci∞ and Co → Co∞ as Re → ∞. Buschmann, M. H. and Gad-el Hak, M. (2003). Gen-

The constants A and α take the values 2.9 and 0.46 eralized logarithmic law and its consequences. AIAA

respectively. J., 41:40–48.

Buschmann, M. H. and Gad-el Hak, M. (2009). Evi-


References
dence of nonlogarithmic behavior of turbulent chan-

nel and pipe flow. AIAA J., 47:535–541.


Alfredsson, P. H., Imayama, S, L. R. J., Örlü, and R. Se-
Chauhan, K. A., Monkewitz, P. A., and Nagib, H. M.
galini, A. (2013). Turbulent boundary layers over flat
(2009). Criteria for assessing experiments in zero
plates and rotating disks - the legacy of von kármán:
pressure gradient boundary layers. Fluid Dyn. Res.,
A stockholm perspective. Euro. J. Mech. B, 40:17–
41:021404.
29.
Chauhan, K. A., Nagib, H. M., and Monkewitz, P. A.
Barenblatt, G. I. (1993). Scaling laws for fully devel-
(2007). On the composite logarithmic profile in
oped turbulent-shear flows. Part 1. Basic hypotheses
zero pressure gradient turbulent boundary layers.
and analysis. J. Fluid Mech., 248:513–520.
22 Ricardo Vinuesa et al.

45th AIAA Aerospace Sciences Meeting and Exhibit, Fluid Mech., 501:135–147.

Reno, NV., page 532. Millikan, C. B. (1938). A critical discussion of turbulent

Clauser, F. H. (1954). Turbulent boundary layers in flows in channels and circular tubes. Proceedings of

adverse pressure gradients. J. Aerosp. Sci., 21:91– the 5th International Congress on applied mechanics.

108. Cambridge, MA, USA., pages 386–392.

Cole, J. D. (1968). Perturbation Methods in Applied Monkewitz, P. A., Chauhan, K. A., and Nagib, H. M.

Mathematics. Blaisdell Publishing Co., Waltham. (2007). Self-consistent high-Reynolds number asymp-

Coles, D. E. (1954). The problem of the turbulent totics for ZPG turbulent boundary layers. Phys. Flu-

boundary layer. ZAMP, 5:181–203. ids, 19:115101.

Coles, D. E. (1956). The law of the wake in the turbu- Monkewitz, P. A., Chauhan, K. A., and Nagib, H. M.

lent boundary layer. J. Fluid Mech., 1:191–226. (2008). Comparison of mean flow similarity laws

George, W. K. (1997). The zero pressure-gradient tur- in zero pressure gradient turbulent boundary layers.

bulent boundary layer. Appl. Mech. Rev., 50:689–730. Phys. Fluids, 20:105102.

George, W. K. (2007). Is there a universal log law for Musker, A. J. (1979). Explicit expression for the smooth

turbulent wall-bounded flows? Phil. Trans. R. Soc. wall velocity distribution in a turbulent boundary

A., 365:789–806. layer. AIAA J., 17:655–657.

George, W. K. and Castillo, L. (2006). Recent advance- Nagib, H. M. and Chauhan, K. A. (2008). Variations

ments toward the understanding of turbulent bound- of von Kármán coefficient in canonical flows. Phys.

ary layers. AIAA J., 44:2435–49. Fluids, 20:101518.

Klewicki, J. C., Fife, P., and Wei, T. (2009). On the Nagib, H. M., Chauhan, K. A., and Monkewitz, P. A.

logarithmic mean profile. J. Fluid Mech., 638:73–93. (2007). Approach to an asymptotic state for zero

Marusic, I., Monty, J. P., Hultmark, M., and Smits, pressure gradient turbulent boundary layers. Phil.

A. J. (2013). On the logarithmic region in wall tur- Trans. R. Soc. A, 365:755–770.

bulence. J. Fluid Mech., 716, R3. Nagib, H. M., Christophorou, C., and Monkewitz, P. A.

McKeon, B. J., Li, J., Jiang, W., Morrison, J. F., and (2004a). High Reynolds number turbulent bound-

Smits, A. J. (2004). Further observations on the mean ary layers subjected to various pressure-gradient con-

velocity distribution in fully developed pipe flow. J. ditions. IUTAM Symp. on One Hundered Years
Role of data uncertainties in identifying the logarithmic region of turbulent boundary layers 23

of Boundary Layer Research, DLR Göttingen, Ger- Tennekes, H. and Lumley, J. L. (1972). A First Course

many, pages 383–394. in Turbulence. Cambridge, MA: MIT Press.

Nagib, H. M., Christophorou, C., Rüedi, J.-D., Monke- Vinuesa, R. (2013). Synergetic Computational and Ex-

witz, P. A., and Österlund, J. M. (2004b). Can perimental Studies of Wall-Bounded Turbulent Flows

we ever rely on results from wall-bounded turbu- and Their Two-Dimensionality. PhD thesis, Illinois

lent flows without direct measurements of wall shear Institute of Technology, Chicago (USA).

stress? 24th AIAA Aerodynamic Measurement Tech- von Kármán, T. (1930). Mechanische ähnlichkeit

nology and Ground Testing Conference, 2392. und turbulenz. Proceedings of the 3rd International

Nishioka, M. (2010). A theoretical study on the overlap Congress on Applied Mechanics, Stockholm, Sweden,

region in a flat plate turbulent boundary layer. Fluid pages 85–93.

Dyn. Res., 42:045502. Zanoun, E. S. (2003). Answers to Some Open Ques-

Örlü, R., Fransson, J. H. M., and Alfredsson, P. H. tions in Wall-Bounded Laminar and Turbulent Shear

(2010). On near wall measurements of wall bounded Flows. PhD thesis, Universität Erlangen-Nürnberg,

flows – the necessity of an accurate determination of Germany.

the wall position. Prog. Aerospace Sci., 46:353–387.

Österlund, J. M., Johansson, A. V., Nagib, H. M., and

H, H. M. (2000). A note on the overlap region in

turbulent boundary layers. Phys. Fluids, 12:1–4.

Purtell, L. P., Klebanoff, P. S., and Buckley, F. T.

(1981). Turbulent boundary layer at low Reynolds

number. Phys. Fluids, 24:802–811.

Segalini, A., Örlu, R., and Alfredsson, P. H. (2013). Un-

certainty analysis of the von Kármán constant. Exp.

Fluids, 54:1460.

Smith, D. W. and Walker, J. H. (1959). Skin fric-

tion measurements in an incompressible flow. NACA

Rep., R26.

View publication stats

You might also like