PHYSICAL REVIEW FLUIDS 3, 114604 (2018)

Characteristics of turbulent boundary layer large scale motions using direct

fluctuating wall shear stress measurements

Rommel J. Pabon,* Lawrence Ukeiley, and Mark Sheplak

University of Florida, Gainesville, Florida 32611, USA

Casey Barnard Keane

Sandia National Laboratories, Albuquerque, New Mexico 87123, USA

(Received 14 June 2018; published 8 November 2018)

This experimental work studies the impact large scale motions in a zero pressure gradient
turbulent boundary layer have on the fluctuating streamwise wall shear stress component
using a recently developed 1 × 1 mm2 floating element differential capacitive shear stress
sensor. The sensing system allows for a flat band response with a bandwidth up to 1.8 kHz
(based on a ±3-dB limit). The streamwise velocity is measured using single component
hot-wire anemometry. The experimental setup is first verified to have a canonical zero
pressure gradient turbulent boundary layer using the mean and fluctuating velocity profiles
as well as fits for the mean wall shear stress with respect to the operating Reynolds number.
Characteristics of the large scale structure are examined spatially using Taylor’s frozen
field hypothesis and the lag time of peak levels of correlation between the shear stress
and velocity signals. The large scale motion inclination angle is determined to be 16◦ . The
coherence between the signals demonstrate that low frequency motions dominate most of
the boundary layer except nearest the wall. In addition, conditional sampling of velocity
on shear stress provides conditional velocity statistics profiles which reveal information on
the entire boundary layer during shear stress events, representing the qualitative features of
the bursting-sweeping process.

DOI: 10.1103/PhysRevFluids.3.114604

I. INTRODUCTION
For the multitude of engineering problems involved with fluid flowing relative to a body, the
knowledge of wall shear stress τw over the body can elucidate much more than just the viscous drag.
Locations of separation, cavitation, and transition to turbulence, as well as local flow direction, and
heat transfer rate can be described in some part through accurate wall shear stress measurement. For
√ mean wall shear stress τw , along with the fluid density ρ, also describe
turbulent boundary layers, the
the friction velocity uτ = τw /ρ. This is an important scaling parameter for physics occurring
very close to the wall where viscous forces become as important as inertial forces. The fluctuating
component of wall shear stress τw = τw − τ w , along with the fluctuating pressure, describes a more
comprehensive picture of the effect turbulent motions have on the surface. This knowledge can lead
to more targeted flow control apparatus for practical situations such as drag reduction, delaying stall,
or localized cooling. However, direct experimental and numerical measurements of τw have, until
recently, suffered from a lack of spatial and temporal resolution to resolve these turbulent motions.

*
[email protected]

2469-990X/2018/3(11)/114604(18) 114604-1 ©2018 American Physical Society

 PABON, UKEILEY, SHEPLAK, AND BARNARD KEANE

Relevant reviews of wall shear stress measurement techniques were given by Winter [1], which
focused on mean measurements, Haritonidis [2], which introduced the concept of micromachined
shear sensors, and Naughton and Sheplak [3], which suggested that the current manufacturing
technology has allowed direct sensing methods to be achievable for measuring the fluctuating shear
stress, particularly with the use of miniature floating element–type sensors. Previous studies have
used thermal anemometry, at the wall, for shear stress measurement, available commercially as
sublayer hot-film probes, but they suffer from various issues, particularly frequency-dependent
conduction effects. This has resulted in the wide range of values for the relative fluctuating
+ 
wall shear stress intensity τw,rms = τw,rms /τ w in canonical turbulent boundary layers that the
literature has provided, ranging from 0.06 to 0.45, collected in a review by Wietrzak and Lueptow
+
[4]. In numerical simulations, values for τw,rms and even averaged shear relations like the skin
+ −2
friction coefficient cf = 2τ w /ρU∞ = 2(U∞ ) in the turbulent regime show discrepancies, due to
2

spatial development of the simulation and sensitivity to the simulation’s conditions, e.g., inflow
boundary conditions and resolution. Until these recent simulations by Schlatter and Örlü [5],
+
which show τw,rms proportional to the logarithm of the Reynolds number up to the O(103 ), the
accepted value was believed to asymptote in the turbulent regime to 0.4 by the seminal work of
Alfredsson et al. [6]. The micropillar method of Brücker et al. [7], an optical technique based
on deflection of small pillars, is able to reconstruct a spatiotemporal field of the wall shear stress
on the wall [8]; however, this indirect method still requires strict assumptions on the flow, such
as the linearity of the viscous sublayer (in an instantaneous sense), and no significant energy in the
resonance for accurate measurements. A recent technique involving microelectromechanical system
(MEMS) based manufacturing is the elastic film based sensor of Amili and Soria [9], which is not
intrusive like the micropillar method and uses direct shear stress-strain relationships to transduce
displacement monitored by optical techniques. This technique shows promise, especially in its
ability to capture the spatial shear stress field and scale well with increasing loading to much larger
shear values, but bandwidth up to 240 Hz limits applicable Reynolds numbers. Ultimately for the
current study, repeatable and accurate wall shear stress measurements are necessary to verify the
footprint of coherent structures on the wall shear stress.
The turbulent boundary layer has been observed to have organized structure to a certain degree;
these recurring patterns must exhibit significant levels of correlation of flow properties for a
substantial spatiotemporal extent. While difficult to exactly define for all cases, coherent structures
provide a framework to view the energy and momentum transfer in a flow as a superposition and
interaction of motions at least larger than the dissipation scales. The characterization of coherent
structures in a turbulent boundary layer has revealed the importance of scale separation, with streaks,
quasistreamwise and hairpin vortices populating the inner layer, and large scale motions (LSMs)
as well as very large scale motions (VLSMs) that are theorized to originate from organization of
the smaller scales [10,11]. The increased wall shear stress intensity in high frequencies during
passage of LSMs associated with the bursting process further suggests the interplay between
motions of vastly different scales [12]. This has been characterized more concretely in the recent
confirmation that the small scale motions near the wall are modulated by the outer layer and large
scale component, especially at higher Reynolds numbers [13].
Using the matching layer between this inner region and the outer layer and assuming the
logarithmic profile matches asymptotically in both regions results in the Coles-Fernholz relation for
the skin friction coefficient cf = 2(κ −1 Reθ + C)−2 [14]. High Reynolds number experiments have
reported the constants for the mean velocity profile U + = κ −1 ln y + + B as κ = 0.384, B = 4.17
[15,16], and C = 4.127 [17]. As accurate measurements have been possible at ever increasing
Reynolds number, questions continue to arise on the veracity of long suggested trends such as the
universality of the logarithmic layer in the overlap region [18], the origin and evolution of these
large scale structures as resulting from organized smaller scales [19], and the k −1 energy spectrum,
expected when more scales begin to overlap with increased energy input [20,21]. The increasing
realization that quality high Reynolds number experiments in canonical turbulent boundary layers
are producing unexpected results indicates that experimental methods in tandem with new sensing

114604-2
 CHARACTERISTICS OF TURBULENT BOUNDARY LAYER …

techniques and simulation verification are the way to understand the increasing effect these LSMs
and VLSMs are having on the wall.
The goal of the present study is to further develop the understanding of the dynamics of wall
shear stress, specifically from the large scales that have a significant impact in momentum transfer
as Reynolds number increases. The primary method used to extract this impact is an independent
direct time resolved measure of wall shear stress sampled synchronously with the streamwise
velocity above. With that aim and design in mind, the transduction scheme chosen was a MEMS
based floating element–type capacitive shear stress sensing system that can capture spatially and
temporally resolved skin friction measurements. This is a direct transduction scheme that does not
rely on restrictive assumptions about the flow in contrast to thermal anemometry, the Clauser chart
method, or sublayer structures like micropillars. Current fabrication methods allow for negligible
misalignment errors and limited flow disturbances. A low-pass filter alleviates the sensor response
near resonance without affecting the frequencies in the flat band region. With a point skin friction
sensor, the conditional friction velocity can be extracted based on a desired condition that best
represents a particular structure. This sensor is thus suited for testing future conditional eddy models
of wall shear stress.

II. EXPERIMENTAL METHODS

A. Experimental facility
An Engineering Laboratory Design (ELD) 407B closed-circuit wind tunnel at the University of
Florida was used to conduct the experiments in this investigation. The ELD tunnel has a square test
section of 0.61 × 0.61 m2 cross section that is 2.44 m long, with sidewalls and floors all composed
of 19.1-mm-thick clear acrylic, allowing optical access to the tunnel without significant optical
distortion. A variable frequency drive outputs a constant frequency power signal to control an in-line
three phase 440 V and 200 HP electric motor which itself drives a two stage axial fan, which allows
for a velocity range up to 90 ms−1 .
Downstream of the fan and first set of turning vanes, an in-line heat exchanger is installed with
an external chiller unit supplying water as the coolant fluid at 13 ◦ C. Coolant supply to the heat
exchanger is regulated by a modulating valve actuator using a temperature set point controller
with the thermocouple downstream of the exchanger. The cooling system is able to sustain a set
temperature within ±1 ◦ C. Thus, using a constant air temperature assumption, other thermodynamic
properties such as density and viscosity are assumed constant, using interpolated values and the
average run temperature. Downstream of this heat exchanger and the second set of turning vanes
is the main flow conditioning section, consisting of a honeycomb section and graduated mesh and
porosity screens to reduce turbulence levels entering the test section. Downstream of this section is
the 25:4 area contraction and then the test section. Measured freestream turbulence intensity levels
are below 0.15% as measured by the hot-wire anemometer at the streamwise location of the shear
sensor, centered in the spanwise direction, and far from the test section walls and flat plate model in
the wall-normal direction. This is a more conservative approach and will result in higher turbulence
intensity levels than measurements at the inlet plane of the test section.
A traversing ceiling is installed into the test section capable of three degrees of freedom linear
translation from Velmex Unislide components driven by dc stepper motors, with the screw drives
controlled by a Velmex VXM controller. This traversing system is utilized to position the hot-wire
probe as desired with respect to the capacitive shear stress sensor (CSSS). The minimum step size
of the traverse system is 6.35 μm in the wall-normal and streamwise directions and 12.7 μm in the
spanwise direction. For the typical run speed in the study, 20 ms−1 , this corresponds to a minimum
step size of 0.32 and 0.64 viscous units, respectively. A supporting arm attached to the screw drive is
used to mount the desired probe, which, in this study, is a hot-wire anemometer. A three-dimensional
printed plastic NACA0012 airfoil profile surrounds the supporting arm to prevent shedding-induced
vibrations as well as support the instrumentation wiring.

114604-3
 PABON, UKEILEY, SHEPLAK, AND BARNARD KEANE

FIG. 1. (a) Schematic of the flat plate model and supports (coloring for clarity) and (b) flat plate model
installed with hot-wire mount descending from the traversing ceiling with capacitive shear stress sensor in
view.

B. Flat plate model

Figure 1(a) shows a schematic of the flat plate model used in this study to create a zero
pressure gradient turbulent boundary layer, which is installed in the test section of the ELD
wind tunnel. The green colored pieces (false color) are stereolithography printed with a plastic
approximating acrylonitrile butadiene styrene properties. The leading edge profile was chosen from
the optimization study of Hanson et al. [22], which sought to create a profile that led to the minimum
length of the nonzero pressure gradient region on the top surface. The trailing edge piece also
consists of a hinge to allow rotation of the aluminum trailing edge piece and can be placed at
a desired angle by tightening the hinge, along with additional rods to support the weight of the
aluminum trailing edge piece, installed from the tunnel floor. The aluminum supports under the
model are mounted to the tunnel and the model itself, with rubber panels between the supports,
model, and the tunnel floor to reduce the effect of tunnel vibrations affecting measurements. A
photograph of the installation is shown in Fig. 1(b), along with the hot-wire probe, mounting arm,
and shear stress sensor.
The second large acrylic panel which contains three smaller acrylic instrumentation windows
is equipped with twenty-eight 1.52-mm-diam static pressure taps spaced 2.54 cm apart along the
length of the panel to measure the mean pressure along the top surface using a pneumatic pressure
scanner with eight channels with an accuracy of ±1.25 Pa. Since a zero pressure gradient is desired,
an experimental parametric sweep over both tunnel velocity and the flap angle of the trailing edge
(TE) is performed. For a given tunnel velocity and TE flap angle, a linear regression is found over
the mean static pressure measurements and their respective locations to return a pressure gradient.
Thus, for each tunnel velocity, a zero for the pressure gradient for a sweep of TE flap angles is
interpolated, resulting in the zero pressure gradient TE flap angle.
The primary sensing location of the wall shear stress sensor used in this study is the center of
the second acylic window, 141 cm from the leading edge. Directly downstream of the leading edge
piece, a distributed roughness element is installed, a sandpaper strip of length 6.5 cm that runs the
full span of the wind tunnel. The roughness height was determined using an assumed distributed
roughness critical trip height Reynolds number of 120 [23], the given downstream distance of the
sensor and a nominal tunnel velocity of 20 ms−1 , resulting in approximately a 0.6-mm distributed
roughness height, to a grit size of P30, used in all experiments.

C. Capacitive shear stress sensor

A MEMS based differential CSSS system, shown in Fig. 2, developed at the University of
Florida’s Interdisciplinary Microsystems Group, was used in this study. The transduction of wall

114604-4
 CHARACTERISTICS OF TURBULENT BOUNDARY LAYER …

FIG. 2. Fabricated CSSS device with H-bar supporting tethers and interdigitated fingers on either side of
floating element, along with hole array on the floating element for pressure rejection. (Figure was adapted from
Ref. [24].)

shear stress is based on the displacements of the capacitive fingers, where a differential scheme
between the sensor and the two external electrodes allows for rejection of common mode deflection
[25]. The design factors critical to the study involve trade-offs between spatial resolution in terms
of sensor area and sensitivity to forces, as well as between sensitivity and attenuation when
fabricating either a smaller capacitive gap or including more interdigitated fingers [26]. For the
current experiments, the floating element has dimensions 1 × 1 mm2 , corresponding to a sensor
length of 51 × 51 squared viscous units or 0.03δ × 0.03δ for the 20-ms−1 runs. The dimensions
will result in spatial averaging only for the smallest structures in the flow, while still being much
smaller than the scales of interest of O(δ) [27].
The MEMS manufacturing processes allow for the sensor and packaging to be hydraulically
smooth, requiring height perturbations to be less than five viscous wall units to allow for viscous
dissipation to suppress flow perturbations [28]. However, a wire bond required for the electronic
connection of the sensor to the interface circuit, in addition to room-temperature-vulcanizing
silicone sealant, has an approximate height of 200 μm, or about ten viscous wall units above
the wall. This height perturbation, required for the electronics of the current packaging, was
positioned away from the sensor in the spanwise direction and in the opposite direction to the
spanwise traverses of the hot-wire probe to reduce impact on the measurement to avoid flow
disruption directly downstream of the sensor. Future design considerations will eliminate this height
perturbation using through-silicon vias for backside electronic connection.
Wall shear stress sensitivity was calculated using an acoustic plane wave tube (PWT) with the
experimental setup and procedures outlined by Sheplak et al. [29] and Chandrasekharan et al. [25].
A summary of the technique is as follows: Standing wave patterns within the PWT generated by
an acoustic driver on one end and a sound hard termination on the other end allow for discrete
nodes of minimum pressure and maximum acoustic velocity separated by half wavelength intervals.
A modified solution to Stokes’s second problem allows for estimation of wall shear stress from
pressure measurements at the sound hard termination of the tube. With a standing wave ratio inside
the plane wave tube greater than 40 dB, locating the CSSS at a pressure node allows for measuring
system response to theoretically pure wall shear stress input. In practice, the finite sensor spatial
size and small location uncertainty cause minor pressure input. A sensitivity of 5.43 mV Pa−1 at
1.128 kHz was measured by increasing the driver amplitude; this frequency corresponds to the
resonance of the PWT with fixed end wall sound hard termination. In addition, the shear stress
range in the current study is within the linear response of the sensor, in both mean and dynamic
component, as seen in Ref. [24]. Mounting the CSSS at the PWT termination axially aligned with

114604-5
 PABON, UKEILEY, SHEPLAK, AND BARNARD KEANE

30 180

25 135

90
20

45
15
0
10
-45

5
-90
3

0 -135
-3
-5 -180
500 1000 1500 1775 2000 2500 3000

FIG. 3. Frequency response of transfer function between CSSS and streamwise aligned wall mounted
microphone in PWT. Black circles (lower curve) show the magnitude response in dB showing a ±3-dB
bandwidth limit of 1775 Hz. Red circles (upper curve) show the difference between the measured phase and
the theoretical phase.

the flow and receiving maximum pressure and no shear stress input yields a pressure rejection ratio
of 71 dB. This was achieved in part by the hole array along the element face. A high pressure
rejection ratio is critical in floating element–type sensors due to the much higher pressure magnitude
at the wall compared to shear stress [3].
A frequency response function is generated by using a variable position end wall termination,
varying the PWT length, and thus modifying the resonant frequency while maintaining the CSSS at
a pressure node for a given input frequency. Measurements are made from 300 Hz to 3 kHz, limited
at the low end by the compression driver output capabilities and by sensor resonance at the high end.
The frequency response between the pressure at the end wall and the CSSS is seen in Fig. 3. Sensor
resonance is located at 2.8 kHz, with the usable flatband based on a ±3-dB standard extending
up to 1775 Hz. The magnitude response was offset to be 0 dB at 1128 Hz, the resonance for the
PWT with the fixed back plate, which has lower uncertainty than the variable position end plate.
The magnitude response drops again below 3 dB before increasing above that threshold near 2.4
kHz, but the more conservative bandwidth limit at 1775 Hz was used in this study. The theoretical
phase between the pressure at the end wall and wall shear stress at the first pressure node is 45◦ .
Figure 3 shows the measured phase between the two with 45◦ subtracted to view differences between
the theoretical and measured phases, and within the sensor bandwidth, no consistent significant
difference in phase was found.

D. Hot-wire anemometry
The streamwise velocity was measured using constant temperature hot-wire anemometry (CTA),
operated at an overheat ratio of 1.7. A Dantec 55P15 boundary layer–type miniature single normal
probe, with a 1.25-mm sensing length and a 5-μm-diam tungsten wire is mounted in the flow on
a Dantec 55H20 straight probe support. This probe support is itself attached to a mounting arm,
surrounded by the NACA0012 shell, that is connected to the traversing ceiling. The scoop shape of
this probe along with the mounting end that holds it allows for the sensing element to be as close to
the wall as possible, without contact interference from the mounting arm. The viscous scaled probe

114604-6
 CHARACTERISTICS OF TURBULENT BOUNDARY LAYER …

sensing length of 63 viscous units is larger than the empirically suggested 20 viscous units to avoid
attenuation of the smallest flow scales and location of the peak of streamwise normal stress intensity,
while the sensing length scaled in outer units is about 0.04δ [27]. Thus spatial attenuation due to
sensing element length is expected for the small scale energy but not the large scale. However, these
small scales are not the focus of these experiments, which are primarily concerned with scales of
O(δ) and above.
In situ calibration was performed in the freestream of the ELD tunnel above the flat plate, far from
both the flat plate turbulent boundary layer and the tunnel walls. Ten calibration points were used,
ranging from 0 to 25 ms−1 , designed for a nominal experimental run speed of 20 ms−1 against the
pitot-static probe of the ELD tunnel. A fifth-order polynomial was used to fit the output voltage from
the CTA bridge to the tunnel velocity. This procedure was performed before and after the traverse
steps were performed, and if the calibrations differed by more than 1% at the desired run speed, the
experiment was repeated. The systems frequency response to a 3-kHz square wave generated by the
CTA bridge was about 33 kHz [30]. All of the measurements in this study were made at sampling
frequency of 10 kHz, low-pass filtered both forward and reverse in time to prevent phase distortion,
using a sixth-order Butterworth filter at a cutoff of 1775 Hz for simultaneous measurement of
shear stress and velocity. For each wall-normal position of the hot-wire probe, a total record of
2.5 × 105 samples of wall shear stress and velocity was taken over a record length of T = 25 s, or
16 490 boundary layer turnover times (T U∞ /δ). This allowed the convergence of the power spectral
densities for the large scales of both wall shear stress and velocity. The unfiltered velocity signal
was also recorded to analyze velocity features outside the bandwidth of the CSSS.
For all the hot-wire probe traverses in this study, the sensing element was positioned 2 mm
downstream of the center of the CSSS floating element, corresponding to 101 viscous units or 0.07δ
to avoid both contact and interference between the two sensors. For the traverse in the wall-normal
direction, the hot-wire probe is aligned with the CSSS in the spanwise direction with logarithmically
spaced points clustered near the wall in the wall-normal direction. The point closest to the wall
corresponds to 28 viscous units, just inside the buffer region. For the spanwise traverse, the hot-
wire probe is positioned 1.98 mm or 101 viscous units above the wall, located in the logarithmic
region, with measurement positions 0.2 mm apart, or 10 viscous units. Taylor’s hypothesis is used
to artificially shift the streamwise location of the velocity measurement upstream to coincide with
the CSSS by means of a convection velocity [31]. The frozen field hypothesis requires first low
turbulence intensity with respect to the local mean u2 /U 2 , which maximizes at about 0.03 for this
study, and second that the wavelength of interest is smaller than δ in a boundary layer [32]. While
this criterion is not strictly met, empirical studies have evaluated the validity of the frozen field
hypothesis in terms of scale of interest and separation of measurement location. Recent experimental
and simulation studies show that the appropriate convection velocity depends on both wave number
and wall-normal position. Even so, the mean convection velocity, averaged over all wave numbers,
deviates only 3.5% from the local mean streamwise velocity [33,34]. The instantaneous velocity
field is well represented by the reconstructed velocity field using the mean convection velocity
only up to streamwise separations of O(δ). Since the current study lacks the spatial information
necessary to calculate appropriate convection velocities with regard to varying scales, the local
mean streamwise velocity will be used as the convection velocity henceforth. It is expected that this
pseudospectrum will alias the large scale spectral energy towards higher wave numbers [35].

III. RESULTS
Table I summarizes important boundary layer parameters calculated from mean measurements
and integral analysis of both the velocity field and shear stress.

A. Wall shear stress measurements

The mean wall shear stress is captured with the CSSS, seen in Table I. Comparison of the
measured cf to the Coles-Fernholz relation cf = 2(κ −1 Reθ + C)−2 and using the constants of

114604-7
 PABON, UKEILEY, SHEPLAK, AND BARNARD KEANE

TABLE I. Boundary layer parameters.

+
Reτ Reθ U∞ τw cf U∞ δ H
1520 4420 19.9 m s−1 0.692 Pa 2.87 × 10−3 25.85 30.4 mm 1.371

Nagib et al. [17] results in a difference of 2.33%. The shear stress spectral density shown in Fig. 4(a)
illustrates both filtered and unfiltered shear responses. The low-pass cutoff frequency corresponds
+
to fcutoff ≈ 0.046 and in outer scaling, fcutoff δ/U∞ ≈ 2.71. Thus, only scales of O(δ) and larger are
completely resolved. Clearly, bandwidth concerns for the wall shear stress sensor must be addressed
for future sensors as the current flow Reynolds number cannot be substantially reduced without
adversely affecting the quality of the turbulent boundary layer. The current MEMS technology does
allow the LSMs and VLSMs to be resolved. √ Energy containing scales of smaller size are filtered,
resulting in biased statistics, such as τx2 /τx , which was observed as 0.11, highly underpredicted
for the experimental Reynolds number [6]. However, this study was focused on capturing larger
scales. The normalized skewness and kurtosis of the wall shear stress are 0.21 and 3.8, respectively;
the corresponding probability density function (PDF) of Fig. 4(b) shows that distribution of the
normalized wall shear stress is well represented by a log-normal distribution, except in the extreme
tails, as are the streamwise velocity fluctuations in the viscous sublayer, shown by Alfredsson et al.
[36]. Gomit et al. [37] show a much more Gaussian profile than the current work, attributed to the
attenuation of the small scales. The current results indicate the existence of flow reversal events at
the wall for the current Reynolds number, albeit with a very low probability. Conventional thermal
sensors are unable to distinguish flow reversal events from low intensity total wall shear stress,
supporting the further use of floating element sensors for higher Reynolds number flows. In other
studies, these events have been shown to increase in probability with increasing Reynolds number,
indicating a growth in the higher moments with Reynolds number [38], but these are extremely
rare events. To ensure that the close proximity of the CSSS and hot-wire probe did not affect either
measurement, the central moments up to fourth order were tracked as the probe approached the
sensor, as well as the mean wall shear stress. All these statistics showed no significant deviation or
trend as the hot-wire approached the CSSS, thus there was no conductive or blockage interference
evident between the two sensors.

(a) (b)

FIG. 4. (a) Unfiltered and low-pass-filtered power spectral density of τx , with a resonance peak at 2.8 kHz.
(b) Normalized histogram simulating the PDF of measured total normalized wall shear stress, alongside a
log-normal distribution matching mean and standard deviations of the data.

114604-8
 CHARACTERISTICS OF TURBULENT BOUNDARY LAYER …

1.2
102

kxT Φuu /uτ2

0.8

λxT /δ
100
0.4

0
10−1 100
y/δ
FIG. 5. Premultiplied and normalized power spectral density of unfiltered streamwise velocity for a variety
of wall-normal positions.

B. Velocity measurements
Figure 5 presents the unfiltered premultiplied streamwise velocity power spectral density at
various wall-normal positions. Both the wavelength λxT and wave number kxT are reconstructed
spatial variables from the temporal data collected here and Taylor’s hypothesis using the local mean
velocity as the convection velocity [33]. The spectral density decreases as distance from the wall
increases, with a clear streak attributed to the large scales at a wavelength λxT ≈ 6δ.
Figure 6(a) displays the mean streamwise velocity profile normalized using the measured friction
velocity from the CSSS. A comparison was made to the Schlatter-Örlü [5] zero pressure gradient
turbulent boundary layer direct numerical simulation (DNS) data set for a similar but smaller
Reθ = 4060, with which the current data show good agreement. The slightly higher plateau reached
by the current data compared to the DNS can be attributed to the difference in Reynolds number,
with the current experiment having the higher Reynolds number. The difference between the CSSS
friction velocity and the friction velocity extracted when fitting the velocity profile to either a

(a) (b)

FIG. 6. (a) Mean normalized velocity profile with friction velocity extracted from CSSS and (b) normalized
streamwise velocity variance, each compared to their respective profile from the Schlatter-Örlü [5] DNS data
for a zero pressure gradient turbulent boundary layer at comparable Reynolds number.

114604-9
 PABON, UKEILEY, SHEPLAK, AND BARNARD KEANE

FIG. 7. Diagnostic plot formulation for the streamwise turbulence intensity compared to the Schlatter-Örlü
[5] DNS data.

Spalding or Musker fit by means of least squares regression was less than 2%. The regression
was run over the friction velocity uτ and the initial probe position offset y0 [39,40]. Approaches
like these for estimating skin friction are common in experiments without direct shear stress
measurements, although they assume some level of universality in the log-law constants with respect
to Reynolds number, initial conditions, and flow type, and even high Reynolds number experiments
show a level of uncertainty in the log-law constants [41]. A similar method reported by Kendall and
Koochesfahani [42] suggests O(1%) error in the friction velocity is expected using only velocity
and probe position measurements, providing a level of validation for the mean streamwise velocity
for the current data. The logarithmic region also agrees well with the log-law fit using the constants
of Österlund [15] of κ = 0.38 and B = 4.1 of the equation U + = κ −1 ln y + + B, well within the
uncertainty levels of these constants suggested by Marusic et al. [41]. The streamwise normal
turbulent stress profile is shown in Fig. 6(b), where, compared to the DNS data set, evidence of the
increase in fluctuation intensity near the wall was not observed. This was due to spatial averaging
along the length of the hot-wire probe (in the spanwise direction) near the wall, where eddies are
expected to be smaller than the hot wire [27]. Good agreement was observed in the logarithmic
region, with the difference in the wake region again attributed to the Reynolds number difference,
with higher intensity observed in the current experiment coinciding with higher Reynolds number.
Chauhan et al. [43] established criteria used as a quality check for turbulent boundary layer ex-
periments which uses an empirical fit with respect to Reynolds number of various flow parameters.
+
In the current study, the shape factor H and velocity scale extent U∞ , which is the ratio between
the freestream velocity and the friction velocity, directly related to the skin friction coefficient, are
+
used as quality check criteria. Here U∞ shows a less than 1% discrepancy with its empirical fit for
Reδ∗ , and the difference between the shape factor H and its empirical fit is less than 0.005, both
of which classify the current study as representative of a canonical zero pressure gradient turbulent
boundary experiment using the Chauhan et al. criteria. In addition, Fig. 7 shows the diagnostic
plot formulation introduced by Alfredsson and Örlü [44], which is the standard deviation of the
streamwise velocity fluctuations normalized by the local mean velocity plotted against the local
mean velocity normalized by the freestream velocity. Normalizing the current data in this manner
shows the expected linear region, independent of Reynolds number, with a slope depending on the
local roughness [45]. The good fit between the data acquired in this study and the DNS in the linear
region further suggests this to be a canonical flat plate turbulent boundary layer. The discrepancies
in the lower speed region (i.e., closer to the wall) are further effects of spatial averaging.

114604-10
 CHARACTERISTICS OF TURBULENT BOUNDARY LAYER …

0.5
1 0.5

0.4 y+ 0.8 0.4

29
0.3 92 0.6 0.3
Rτ u

Rτ u
y/δ
152
0.2 357 0.4 0.2
627
0.1 0.2 0.1

0 0
0 -2 0 2 4
-300 -200 -100 0 100 200

Δt+ ΔxT /δ
(a) (b)

FIG. 8. For a wall-normal traverse, cross correlation coefficient Rτ u for (a) inner normalized temporal
delays t + and (b) transformed to the spatial domain with Taylor’s hypothesis.

C. Correlations and coherence

Figure 8(a) presents the temporal cross correlation between wall shear stress and streamwise
velocity for select wall-normal positions. While these are captured as time series, after the artificial
lag in time to compensate for the streamwise distance, Taylor’s hypothesis is used to convert the
signals to the spatial domain in Fig. 8(b) by imposing the local mean streamwise velocity as the
convection velocity [31]. The large degree of linear correlation in Fig. 8 suggests that low-order
techniques for modeling events stochastically based on correlation between the two would be
successful [13]. Velocity events leading shear events in time, evident in Fig. 8(a), imply the existence
of a large scale motion inclined with respect to the wall. The most common characteristic of the
LSMs is the inclination angle, shown to be Reynolds number invariant near 14◦ [46]. This angle was
calculated by using the frozen field hypothesis at each wall-normal position, with the temporal lag at
the correlation peak corresponding to a streamwise distance. Significant levels of correlation persist
up to the end of the logarithmic region, with two distinct qualitative features evident in Fig. 8(b).
The first is the inclination of the large scale structures (note the different scales for the x and y axes)
and the second is the broadness of the correlation, as opposed to a sharp peak, which diminishes the
certainty by which an inclination angle can be observed. Because of these two characteristics, two
peak tracking methods are used to determine the appropriate lag for calculating the inclination angle.
First, the lag associated with the maximum correlation is tracked with wall-normal position. Second,
the expected correlation lag, or the average lag weighted by the correlation value, in essence, the first
moment of the correlation lag, is tracked at each wall-normal position. This method was proposed
to track the location of the bulk correlation, which differs from the overall maximum due to the
skewness in the correlations. Figure 9 shows the result, with the first peak tracking method finding
an inclination angle of 15◦ and the second 16◦ . The two methods achieve similar results, with
the coefficient of determination r 2 (in this simple linear regression, the square of the correlation
coefficient) of the first and second peak tracking methods being 0.947 and 0.967, respectively.
The hot-wire traverse was also performed in the spanwise direction at y + = 99 with the spatial
reconstruction in the streamwise direction presented in Fig. 10. The size of the correlated region
in the spanwise direction is approximately 150 viscous units (0.11δ) at the wall-normal location
referenced; at the same correlation threshold, the wall-normal extent is about 700 viscous units
(approximately equal to 0.5δ), while the correlation extent in the streamwise direction through
frozen field reconstruction is much larger than both at approximately 5 boundary layer thicknesses,
seen in Fig. 8(b). These spatial extents indicate thin and long structures that reach into the wake
region, qualitatively supporting the hairpin vortex packet model. In Fig. 10, at the traverse points

114604-11
 PABON, UKEILEY, SHEPLAK, AND BARNARD KEANE

FIG. 9. Spatial reconstruction of the large scale motion for the wall-normal traverse using the chosen
correlation lag value as the lag at the highest correlation peak (red circles) and the expected correlation lag
from the wall-normal traverse (black circles).

farther from the shear sensor in the spanwise direction, there is evidence of a bimodal structure at
this particular wall-normal location. This is supported by Ref. [47], where negatively correlated side
lobes between the wall shear stress and streamwise velocity are seen farther from the sensor than
the current domain of wall-normal positions.
The magnitude-squared coherence spectra is given by
|φτ u (y, f )|2
γτ2u (y, f ) = , (1)
φτ τ (y, f )φuu (y, f )
where φτ u (y, f ) is the one-sided cross-spectral density between the wall shear stress and velocity
signals at the given wall-normal position and φτ τ (y, f ) and φuu (y, f ) are the autospectral densities
of wall shear stress and velocity, respectively, for a given probe position. The high linear coherence
levels support these inferred sizing of the structure, decaying rapidly after approximately 700
viscous units in the wall-normal direction in Fig. 11(a) and 150 viscous units in the spanwise

200 0.5

0.4
150

0.3
Rτ u
z+

100
0.2

50
0.1

0 0
-2000 -1000 0 1000 2000
x+
FIG. 10. For a spanwise traverse at y + = 99, the inner normalized temporal cross correlation coefficient
between shear stress and streamwise velocity transformed to the spatial domain in streamwise direction.

114604-12
 CHARACTERISTICS OF TURBULENT BOUNDARY LAYER …

103 0.75 200 0.75

150
0.5 0.5
y+

γτ2u
γτ2u

z+
100

102 0.25 0.25

50

0 0 0
10−3 10−2 10−3 10−2
f+ f+
(a) (b)

FIG. 11. Coherence between fluctuating wall shear stress and velocity signals for (a) a wall-normal traverse
and (b) a spanwise traverse of the hot-wire probe at y + = 99.

direction in Fig. 11(b). However, the spatial reconstruction can only be inferred here with Taylor’s
hypothesis, and since there clearly exist scales larger than δ, applicability of the frozen field
hypothesis is at the very least unsure. The diagnostic plot in Fig. 7 serves a secondary purpose
here in also being a threshold limit for another assumption in Taylor’s hypothesis, low local relative
turbulence intensity urms /U [31,32]. As urms /U increases near the wall, it no longer becomes
negligible and backflow events are possible (evident in the shear stress histogram, but not the
hot-wire data due to the transduction mechanism), which severely weakens the applicability of the
hypothesis. Compounding the issue, low levels of mean shear are an additional assumption. Even
with its prevalence in the hot-wire experimental community, a substitute for Taylor’s hypothesis
must be found if ever increasing Reynolds number experiments want to use point measurements,
regardless of averaging issues that have also plagued these types of measurements [27]. Therefore, at
best, only qualitative statements can be made about the largest scales in this study; higher Reynolds
number experiments need a true spatial measurement like particle image velocimetry, even if the
noise threatens correlation values, if they hope to avoid skewing and over interpreting data.
The coherence is dominated by the lower frequencies, seen in Fig. 11(a), well into the wake
region, as well as in Fig. 11(b) for the spanwise traverse. The high frequencies show low significant
coherence only near the wall (less than 100), implying motions at this scale have no large impact
away from the wall. Measurements closer to the wall with greater bandwidth would be required to
derive higher levels of confidence on the impact the near wall small scales have relating wall shear
stress and velocity, while avoiding many of the issues dealing with anemometry in the region [18].

D. Conditional measurements
The advantages of a direct time resolved measure of shear stress include the ability to condition-
ally sample the velocity field based on a shear stress condition. In a similar study by Hutchins et al.
[48], only the sign of the shear stress fluctuations is used as the threshold, and in a study by Gomit
et al. [37], the signal is divided into four equally represented quartiles for thresholding, although the
measured PDF in that work is notably not skewed.
In this study, a series of thresholds are used on the local shear stress fluctuation τx (t ) to describe
certain velocity field statistics. The resulting conditional mean velocity fields are

hh
U +
(y) = u(y, t )|τw (t ) > στ , (2)

h+ (y) = u(y, t )|τw (t ) > 0 ,

U (3)

114604-13
 PABON, UKEILEY, SHEPLAK, AND BARNARD KEANE

(a) (b)

FIG. 12. Mean streamwise velocity profiles conditioned on shear events compared to the measured mean
profile and DNS data from Schlatter and Örlü [5], normalized by (a) the measured total mean friction velocity
and (b) the conditional mean friction velocity for each threshold.

l+ (y) = u(y, t )|τw (t ) < 0 ,

U (4)

ll+ (y) = u(y, t )|τw (t ) < −στ ,

U (5)

where στ represents the standard deviation of the wall shear stress. These inner scaled profiles can
each be normalized by the unconditional mean friction velocity uτ , shown in Fig. 12(a), or by
the conditional mean friction velocity of the respective threshold used in Eqs. (2)–(5), shown in
Fig. 12(b). Qualitative agreement with the results of Hutchins et al. [48] is seen, and the observed
trend continues for tighter thresholds; for the profile in Fig. 12(a), high shear at the wall corresponds
to a steeper wall gradient, and therefore a raised velocity profile, with the trend continuing for an
increased shear condition. This trend is also symmetric for low shear stress. The collapse of the
mean profiles with unconditional normalization in the outer region suggests that the effect of the
wall shear stress on the outer region of the mean velocity profile is minimal, at least for the current
thresholds. This implies a limit to the extent conditional measurements or even flow control can
have on the outer boundary layer, but a more extensive range of Reynolds number is required to find
that limit and its scaling.
For a given wall-normal position, 2.5 × 105 discrete velocity samples were taken. The thresholds
above and below the mean each include nominally 50% of the measurements. The thresholds above
and below one standard deviation include 16% of the measurements, or about 4 × 104 samples.
Increasing the threshold to study extremely intense but rare events would require more samples to
be statistically converged.
The conditional velocity profiles can also be normalized with a friction velocity representative of
the conditionally sampled shear stress, presented in Fig. 12(b). A profile normalized this way will
follow classic universal boundary layer scaling near the wall but have modified wake regions. This
shows that low shear events represent a profile with a lower friction velocity and a stronger wake,
resembling an adverse pressure gradient profile. Similarly, high shear events have a higher friction
velocity and reduced wake region, resembling a favorable pressure gradient. The collapse near the
wall is evidence of more appropriate conditional scaling and that there is still universality in the
logarithmic region and lower, with the pressure gradient–like effects only seen in the outer region.
Thus, a high shear event presents the wall with an instantaneously more full boundary layer.
Of future interest is the application of these thresholds with flow reversal effects (τw < 0) to
see how a conditional mean profile is shaped under these extreme events. However, either many

114604-14
 CHARACTERISTICS OF TURBULENT BOUNDARY LAYER …

(a) (b)

FIG. 13. Streamwise velocity variance profiles conditioned on shear events compared to the measured
variance profile and DNS data from Schlatter and Örlü [5], normalized by (a) the measured total mean friction
velocity and (b) the conditional mean friction velocity for each threshold.

more samples [O(108 )] than the current study or a much higher Reynolds number with only weakly
increasing probability of flow reversal events would be required to converge the statistics with that
condition. In addition, a higher bandwidth sensor would be required to prevent temporal attenuation
of these scales, so a large Reynolds number range experimental setup can begin to model this trend
for future experiments.
To elucidate the effects of local shear intensity on turbulence, the thresholding procedure was
performed on the streamwise normal stress profiles, as
+

u2 2 
hh (y) = u (y, t )|τw (t ) > στ , (6)
+

u2 (y) = u2 (y, t )|τ  (t ) > 0 , (7)
h w
+

u2 (y) = u2 (y, t )|τ  (t ) < 0 , (8)
l w
+

u2 (y) = u2 (y, t )|τ  (t ) < −σ .
τ (9)
ll w

Higher moments of the velocity field are unable to be extracted using this procedure because of the
larger sample requirement for convergence.
The conditional variance profiles are shown in Fig. 13, again in unconditional scaling shown
in Fig. 13(a) and conditional mean friction velocity scaling for the respective thresholds shown
in Fig. 13(b). For the unconditional normalization, only small differences between the conditional
profiles and the total mean are observed. Nevertheless, negative shear fluctuation conditions result
in slightly higher turbulence intensities in the logarithmic region, with high shear conditions having
the opposite effect, and all the profiles collapse to zero at the boundary layer edge. The conditioning
therefore seems to have the opposite effect on the variance profiles compared to the mean profiles.
When applying the conditional normalization as in Fig. 13(b), these small differences are amplified.
Wall-normal values near the wall are ignored here because of spatial attenuation evident in Fig. 6(b),
deviating from the DNS below y + = 100, to avoid drawing incorrect conclusions about the near
wall velocity field. The result of the conditional normalization in Fig. 12(b) is a collapsed mean
profile near the wall, but the logarithmic region is much more energetic for low shear than normal in
Fig. 13(b), and the local high shear causes a pacified logarithmic variance. This is representative of
the bursting-sweeping process, but the effect of local instantaneous pressure gradients, not measured
in this study, cannot be separated from this result.

114604-15
 PABON, UKEILEY, SHEPLAK, AND BARNARD KEANE

IV. CONCLUSION
A MEMS based CSSS was used to measure aspects of a zero pressure gradient turbulent
boundary layer at Reτ = 1520. The chosen Reynolds number balances the need to create a fully
turbulent boundary layer while minimizing the effects of spatiotemporal averaging and resolving
at least the LSMs. The current direct shear transduction system was shown to produce results
consistent with previous wall mounted shear sensors such as hot film sensors, without the effects of
frequency-dependent conduction loss, rectification, and other issues. This study demonstrated the
ability of the CSSS to capture LSMs in a canonical turbulent boundary layer and their correlated
relationship with the wall shear stress, as well as capturing conditional statistics of high and low
shear events to differentiate their effect on the velocity field.
First, the boundary layer over the flat plate model was characterized and the data verified to
be representative of a canonical zero pressure gradient turbulent boundary layer. However, the
limitations of the floating element sensor include a bandwidth-limiting resonance, requiring a low-
pass filter, so attenuation of different energy scales was carefully analyzed. As such, the fluctuating
shear stress intensity was underpredicted for this Reynolds number. Regardless, simultaneous
measurement with a single component hot-wire anemometer allowed for estimating characteristics
of the LSMs through correlations, as well as conditional velocity measurements during directly
measured shear events to demonstrate the sensor’s ability to resolve the imprint of the turbulent
structures in the boundary layer.
The measured mean wall shear stress was used to independently normalize boundary layer
profiles, showing good agreement with common composite profiles, as well as a DNS data set
at a similar Reynolds number [5,39]. High levels of cross correlation between the simultaneously
sampled shear stress and velocity signals were reported, especially near the wall. High levels of
coherence between the two signals persisted well into the wake region for the low frequencies, with
almost no high frequency coherence except near the wall. Measurements in both the wall-normal
and spanwise directions suggested that the LSMs are larger in the wall-normal direction than in the
spanwise direction, each being much smaller than the streamwise extent.
The inclination angle of the LSMs was identified by tracking the spatial separation of the
peak in the correlations, after applying the frozen field hypothesis to the temporal correlation. A
peak tracking methodology was introduced that is more robust in sensing the large scale angle
as opposed to small scale correlation peaks near the wall, which change the angle measurement.
While a Reynolds number invariance has been shown near 14◦ [46], this study observed the angle at
16◦ . Taylor’s frozen field hypothesis is the most prevalent tool to infer spatial statistics from point
measurements, but its validity has been shown to break down, even at the relatively low Reynolds
number of this study.
The average velocity as a function of wall-normal distance was calculated given threshold
conditions of shear stress, with different normalization schemes elucidating information about
different regions in the boundary layer. The conditional variance profiles, on the other hand, seem to
complicate the theory that different normalization schemes can be applied to collapse the statistics of
the different parts of the boundary layer. Application of the differing shear conditions, while directly
modifying the local mean shear, also have a large indirect impact in conditioning on dynamic
processes.

ACKNOWLEDGMENTS
This material was based upon work supported by the National Science Foundation Graduate
Research Fellowship under Grant No. DGE-1315138 and the Sandia Campus Executive Fellowship
PO No. 1266026. The authors also wish to acknowledge the support of the Florida Center for
Advanced Aero-Propulsion and the National Science Foundation under Grant No. CBET-1744146.

114604-16
 CHARACTERISTICS OF TURBULENT BOUNDARY LAYER …

[1] K. G. Winter, An outline of the techniques available for the measurement of skin friction in turbulent
boundary layers, Prog. Aerosp. Sci. 18, 1 (1977).
[2] J. H. Haritonidis, in Advances in Fluid Mechanics Measurements, edited by M. Gad-El-Hak (Springer,
Berlin, 1989), pp. 229–261.
[3] J. W. Naughton and M. Sheplak, Modern developments in shear-stress measurement, Prog. Aerosp. Sci.
38, 515 (2002).
[4] A. Wietrzak and R. M. Lueptow, Wall shear stress and velocity in a turbulent axisymmetric boundary
layer, J. Fluid Mech. 259, 191 (1994).
[5] P. Schlatter and R. Örlü, Assessment of direct numerical simulation data of turbulent boundary layers,
J. Fluid Mech. 659, 116 (2010).
[6] P. H. Alfredsson, A. V. Johansson, J. H. Haritonidis, and H. Eckelmann, The fluctuating wall-shear stress
and the velocity field in the viscous sublayer, Phys. Fluids 31, 1026 (1988).
[7] C. Brücker, J. Spatz, and W. Schröder, Feasability study of wall shear stress imaging using microstruc-
tured surfaces with flexible micropillars, Exp. Fluids 39, 464 (2005).
[8] S. Große and W. Schröder, High Reynolds number turbulent wind tunnel boundary layer wall-shear stress
sensor, J. Turbul. 10, N14 (2009).
[9] O. Amili and J. Soria, A film-based wall shear stress sensor for wall-bounded turbulent flows, Exp. Fluids
51, 137 (2011).
[10] S. Robinson, Coherent motions in the turbulent boundary layer, Annu. Rev. Fluid Mech. 23, 601
(1991).
[11] R. J. Adrian, Hairpin vortex organization in wall turbulence, Phys. Fluids 19, 041301 (2007).
[12] G. L. Brown and A. S. W. Thomas, Large structure in a turbulent boundary layer, Phys. Fluids 20, S243
(1977).
[13] R. Mathis, I. Marusic, S. I. Chernyshenko, and N. Hutchins, Estimating wall-shear-stress fluctuations
given an outer region input, J. Fluid Mech. 715, 163 (2013).
[14] H. H. Fernholz and P. J. Finley, The incompressible zero-pressure gradient turbulent boundary layer: An
assessment of the data, Prog. Aerosp. Sci. 32, 245 (1996).
[15] J. M. Österlund, Experimental studies of zero pressure-gradient turbulent boundary-layer flow, Ph.D.
thesis, Royal Institute of Technology, 1999.
[16] H. M. Nagib and K. A. Chauhan, Variations of von Kármán coefficient in canonical flows, Phys. Fluids
20, 101518 (2008).
[17] H. M. Nagib, K. A. Chauhan, and P. A. Monkewitz, Approach to an asymptotic state for zero pressure
gradient turbulent boundary layers, Philos. Trans. R. Soc. A 365, 755 (2007).
[18] I. Marusic, B. J. McKeon, P. A. Monkewitz, H. M. Nagib, A. J. Smits, and K. R. Sreenivasan, Wall-
bounded turbulent flows at high Reynolds numbers: Recent advances and key issues, Phys. Fluids 22,
065103 (2010).
[19] M. Guala, M. Metzger, and B. J. McKeon, Interactions within the turbulent boundary layer at high
Reynolds number, J. Fluid Mech. 666, 573 (2011).
[20] M. Vallikivi, B. Ganapathisubramani, and A. J. Smits, Spectral scaling in boundary layers and pipes at
very high Reynolds numbers, J. Fluid Mech. 771, 303 (2015).
[21] S. Pirozzoli and M. Bernardini, Probing high-Reynolds-number effects in numerical boundary layers,
Phys. Fluids 25, 021704 (2013).
[22] R. E. Hanson, H. P. Buckley, and P. Lavoie, Aerodynamic optimization of the flat-plate leading edge for
experimental studies of laminar and transitional boundary layers, Exp. Fluids 53, 863 (2012).
[23] H. Schlichting, Boundary Layer Theory, 4th ed. (McGraw-Hill, New York, 1960).
[24] C. Barnard, J. Meloy, and M. Sheplak, Proceedings of 2016 IEEE Sensors, Orlando (IEEE, Piscataway,
2016).
[25] V. Chandrasekharan, J. Sells, J. C. Meloy, D. P. Arnold, and M. Sheplak, A microscale differential
capacitive direct wall-shear-stress sensor, J. Microelectromech. Syst. 20, 622 (2011).
[26] J. C. Meloy, J. Griffin, J. Sells, V. Chandrasekharan, L. N. Cattafesta, and M. Sheplak, Proceedings of the
41st AIAA Fluid Dynamics Conference, Honolulu, 2011 (AIAA, Reston, 2011), pp. 1–18.

114604-17
 PABON, UKEILEY, SHEPLAK, AND BARNARD KEANE

[27] N. Hutchins, T. B. Nickels, I. Marusic, and M. S. Chong, Hot-wire spatial resolution issues in wall-
bounded turbulence, J. Fluid Mech. 635, 103 (2009).
[28] S. B. Pope, Turbulent Flows (Cambridge University Press, Cambridge, 2000).
[29] M. Sheplak, A. Padmanabhan, M. A. Schmidt, and K. S. Breuer, Dynamic calibration of a shear-stress
sensor using stokes-layer excitation, AIAA J. 39, 819 (2001).
[30] H. H. Bruun, Hot-Wire Anemometry: Principles and Signal Analysis (Oxford University Press, Oxford,
1995).
[31] G. I. Taylor, The spectrum of turbulence, Proc. R. Soc. London Ser. A 164, 476 (1938).
[32] C. C. Lin, On Taylor’s hypothesis and the acceleration terms in the Navier-Stokes equations, Q. Appl.
Math. 10, 295 (1953).
[33] C. Atkinson, N. A. Buchmann, and J. Soria, An experimental investigation of turbulent convection
velocities in a turbulent boundary layer, Flow Turbul. Combust. 94, 79 (2014).
[34] J. C. Del Álamo and J. Jiménez, Estimation of turbulent convection velocities and corrections to Taylor’s
approximation, J. Fluid Mech. 640, 5 (2009).
[35] B. J. Rosenberg, M. Hultmark, M. Vallikivi, S. C. C. Bailey, and A. J. Smits, Turbulence spectra in
smooth- and rough-wall pipe flow at extreme Reynolds numbers, J. Fluid Mech. 731, 46 (2013).
[36] P. H. Alfredsson, R. Örlü, and P. Schlatter, The viscous sublayer revisited-exploiting self-similarity to
determine the wall position and friction velocity, Exp. Fluids 51, 271 (2011).
[37] G. Gomit, R. de Kat, and B. Ganapathisubramani, Structure of high and low shear-stress events in a
turbulent boundary layer, Phys. Rev. Fluids 3, 014609 (2018).
[38] C. Diaz-Daniel, S. Laizet, and J. C. Vassilicos, Wall shear stress fluctuations: Mixed scaling and their
effects on velocity fluctuations in a turbulent boundary layer, Phys. Fluids 29, 055102 (2017).
[39] D. B. Spalding, A single formula for the “Law of the wall”, J. Appl. Mech. 28, 455 (1961).
[40] A. J. Musker, Explicit expression for the smooth wall velocity distribution in a turbulent boundary layer,
AIAA J. 17, 655 (1979).
[41] I. Marusic, J. P. Monty, M. Hultmark, and A. J. Smits, On the logarithmic region in wall turbulence,
J. Fluid Mech. 716, R3 (2013).
[42] A. Kendall and M. M. Koochesfahani, A method for estimating wall friction in turbulent wall-bounded
flows, Exp. Fluids 44, 773 (2008).
[43] K. A. Chauhan, P. A. Monkewitz, and H. M. Nagib, Criteria for assessing experiments in zero pressure
gradient boundary layers, Fluid Dyn. Res. 41, 021404 (2009).
[44] P. H. Alfredsson and R. Örlü, The diagnostic plot a litmus test for wall bounded turbulence data, Eur. J.
Mech. B 29, 403 (2010).
[45] I. P. Castro, A. Segalini, and P. H. Alfredsson, Outer-layer turbulence intensities in smooth- and rough-wall
boundary layers, J. Fluid Mech. 727, 119 (2013).
[46] I. Marusic and W. D. C. Heuer, Reynolds Number Invariance of the Structure Inclination Angle in Wall
Turbulence, Phys. Rev. Lett. 99, 114504 (2007).
[47] J. A. Sillero, J. Jiménez, and R. D. Moser, Two-point statistics for turbulent boundary layers and channels
at Reynolds numbers up to δ + ≈ 2000, Phys. Fluids 26, 105109 (2014).
[48] N. Hutchins, J. P. Monty, B. Ganapathisubramani, H. C.-H. Ng, and I. Marusic, Three-dimensional
conditional structure of a high-Reynolds-number turbulent boundary layer, J. Fluid Mech. 673, 255
(2011).

114604-18