Accepted Manuscript

Drag reduction promoted by repetitive bubble injection in turbulent channel

flows

Hyun Jin Park, Yuji Tasaka, Yoshihiko Oishi, Yuichi Murai

PII: S0301-9322(15)00120-2
DOI: http://dx.doi.org/10.1016/j.ijmultiphaseflow.2015.05.003
Reference: IJMF 2220

To appear in: International Journal of Multiphase Flow

Received Date: 28 November 2014

Revised Date: 1 April 2015
Accepted Date: 6 May 2015

Please cite this article as: Park, H.J., Tasaka, Y., Oishi, Y., Murai, Y., Drag reduction promoted by repetitive bubble
injection in turbulent channel flows, International Journal of Multiphase Flow (2015), doi: http://dx.doi.org/
10.1016/j.ijmultiphaseflow.2015.05.003

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Drag reduction promoted by repetitive bubble injection in turbulent channel flows

Hyun Jin Park*, Yuji Tasaka, Yoshihiko Oishi, Yuichi Murai

Laboratory of Flow Control, Division of Energy and Environmental Systems, Graduate School of
Engineering, Hokkaido University, N13 W8, Kita-ku, Sapporo 060-8628, Japan

* Corresponding author. Tel: +81 11 706 6373; fax: +81 11 706 6373.
E-mail address: [email protected] (H.J. Park).

1. Introduction
Turbulent boundary layer control by injection of bubbles into the boundary layer is expected to enhance
the energy efficiency of vessels by reducing the frictional drag that constitutes nearly 80% of the total
drag acting on large class vessels. This is also attractive to engineers because it offers installation
simplicity and is pollution-free. This technique has been studied in a number of institutes to clarify the
drag reduction mechanism and to enable practical use on actual vessels since McCormick and
Bhattacharyya (1973) first introduced the technique. Some studies indicate the importance of
modifications to the vortical flow structures in turbulent boundary layers caused by the fragmentation and
deformation of bubbles; these structures create Reynolds shear stress that dominates the skin friction drag
in turbulent flows (Meng and Uhlman 1989; Kawamura and Kodama 2002; Xu et al. 2002; Kitagawa et al.
2005; Jacob et al. 2010). As a practical demonstration, Kodama et al. (2005 and 2008) evaluated the
performance of this technique on a real vessel, a cement carrier named the Pacific Seagull, and a net
power saving of approximately 5%, calculated based on the fuel consumption and the energy
consumption required to inject bubbles, was reported. Other groups experimenting on the same vessel
also reported that the maximum average drag reduction reached approximately 11% in their experiments
(Hoang et al. 2009). These experimental results using a real vessel posed two problems, which must be
addressed for practical applications: (i) high energy consumption occurs when injecting bubbles at the
bottom of deep draft ships against hydrostatic pressure; (ii) only a small bulk drag reduction effect is
obtained, whereas sufficient local drag reduction can be realized. Adoption of huge vessels with shallow
draft can avoid the first problem but restricts general use of the technique for various other types of vessel.
In fact, to cause relatively low or negative drag reduction effects, researchers have indicated that
two-phase flow structures in the case of low void fractions (i.e., the volume fraction of the bubbles)
should be studied (Kato et al. 1999; Berg et al. 2007). Fatter bubbles that are comparable in size to the
boundary layer thickness also increase the wall shear stress (Murai et al. 2007). Figure 1 summarizes
these known facts schematically. Figure 1(a) represents the existence of a critical void fraction, αc, at
which the wall shear stress becomes smaller than the original value without the bubbles. The value at

1
which αc appears depends on the Reynolds, Froude, and Weber numbers. Typically, αc has a value of more
than 0.1 in the case of low Reynolds number turbulent flows, and tends to infinity for laminar boundary
layers containing spherical bubbles. Figure 1(b) illustrates the bubble size (Db) dependency: microbubbles
and air films achieve drag reduction by different mechanisms (Elbing et al. 2008), but very often
intermediate-sized bubbles conversely increase the wall shear stress. The bubble-to-liquid interactions of
these intermediate bubbles were measured by Oishi and Murai (2014), and their results indicate that
inclusion of such fat bubbles should be avoided to obtain stable drag reduction performance. Also, the
size of the bubbles generated by the most commonly used types of bubble generators, such as blowers,
changes with the void fraction. The bubbles are naturally small in size at low void fractions and become
larger with increasing void fraction. This means that conventional bubble generators cannot control the
bubble size and the void fraction independently, which means that the drag reduction performance could
not be maximized using these generators.
To improve the efficiency of drag reduction produced by bubble injection, we propose a novel bubble
injection control method that enables the historically accumulated knowledge of the parametric
dependency to be applied effectively. The method involves control of repetitive bubble injection (RBI).
This control is realized by simple open-close iteration of a valve for the bubble supply, but provides
complex variability in the bubbly two-phase turbulent boundary layers to lead to new phenomenological
discussions. A preliminary experiment on the effects of RBI was reported by our group (Park et al. 2009),
and we confirmed the feasibility of this method for improvement of the drag reduction performance. This
RBI scheme is expected to produce (i) concentration of the air resource to increase the local void fraction
so that the system avoids a drag-increasing regime at low void fractions, (ii) reduction of the air volume
flow rate required to introduce bubbles at deep locations where the hydrostatic pressure is high, and (iii)
repetitive renewal of the vortical flow structures that develop inside the turbulent boundary layers. The
first and second expectations are obvious, as we mentioned in the previous paragraph. Our recent work on
flow visualization (Park et al. 2014) indicated that the RBI system provides reproducible bubble swarms
in the downstream region with leading air films that insulate the vortical structures present in the
turbulent boundary layer from the wall. Interestingly, we found that the most of the vortical structures
survive underneath the bubble swarms and their capability to create frictional drag on the wall, as in the
single-phase condition, may be restored after the passage of the leading air films, although with a
considerable delay. This series of visualizations has indicated to us that the renewal time of the vortical
flow structures promotes drag reduction as a third effect of RBI, and this effect can be added to the
previously mentioned purposes of RBI.
The RBI method that is adopted in this paper injects bubble swarms with locally high void fractions
into a turbulent boundary layer at controlled intervals. Even if the mean void fraction is set to be low,
these bubble swarms produce strong void fraction fluctuations within the boundary layer to maintain the
two-way interaction between the bubbles and the liquid flows. As we will show later in the paper,
individual bubble swarms are always led by local air films that refresh the turbulent boundary layer that is

2
developing spatially in the streamwise direction. In this paper, we report a new series of experimental data
that were obtained by multiple diagnoses, including the ultrasound Doppler method, which leads to an
in-depth discussion of the improved performance of RBI-based drag reduction. As a platform for these
investigations, we use a turbulent channel flow at relatively low Reynolds numbers, where Re ~ 103. In
this regime, bubbly drag reduction hardly occurs in the case of continuous bubble injection (Oishi and
Murai 2014), and, contrastingly, the effects of RBI can be clearly distinguished. Since viscous
modification by bubbles remains significant in this low Reynolds number turbulent flow regime, we can
discuss the reason why the drag reduction is promoted by RBI, based on comparing the wall shear stress
and the Reynolds shear stress profiles.

2. Experimental setup

2.1 Setup and measurement equipment

A schematic diagram of the experimental apparatus is shown in Fig. 2. The test section is a horizontal
rectangular channel made from transparent acrylic resin, and is 40 mm high (H = 2h), 160 mm wide (W)
and 6000 mm long. Silicone oil (KF-96-10cs, Shin-Etsu Chemical Co., Ltd.) is used rather than water as
the working fluid to ensure the reproducibility of the experimental results by avoiding any uncontrollable
influences of contamination of the bubble interfaces by tracer particles and other impurities, because the
oil is a non-polar liquid. The kinematic viscosity (ν), density (ρ0) and the surface tension of the silicone
oil at 25°C are 10 × 10-6 m2/s, 930 kg/m3 and 20.1 mN/m, respectively. Spherical polyolefin fine powders
called FLO BEADS (CL-2507, Sumitomo Seika Chemical Co., Ltd.) are adopted as tracer particles to
measure the velocity vector fields. The average diameter of these particles (Dp) is 180 µm and the particle
density (ρp) is 920 kg/m3. Concerning the traceability in turbulence, relaxation time of the particles in the
liquid phase (tp) is estimated to be 0.27 ms as defined by tp = ((2ρp +ρ0)Dp2)/(36νρ0). In the case of fully
developed turbulent flow at the maximum bulk velocity at Ubulk = 2.0 m/s, Stokes number (St) of the
particle, St = tp/tη, being estimated from the equivalent Kolmogorov time scale (tη) is smaller than 0.07,
thus sufficiently smaller than unity. A bubble injector plate, which has a single open slit with 10 mm in
the streamwise side and 120 mm in the spanwise length, is mounted on the upper wall of the channel
without step at x/H = 43.75 from the channel inlet, where the x, y, and z coordinates are defined as the
streamwise distance from the channel inlet, the vertical downward coordinate from the upper wall and the
spanwise location from the channel side wall, respectively. The injector is connected to an
electromagnetic valve so that the arbitrary temporal fluctuation in the gas flow rate is produced by a
function generator. Bubble size is naturally determined by the balance between air flow rate and shear of
the liquid flow at the outlet of the gas supplying slit. The balance changes also with operation scheme of
the repetitive bubble injection which is explained in section 2.2. Bubbles that reach the end of the channel
are removed from the oil in a swirling tank set at the channel outlet before re-entering into a pump.

3
Equipment to be used for the various measurements is located more than 20H away from the bubble
injector to investigate the downstream development of bubbly flows.
Two 4 MHz ultrasonic transducers are attached to the bottom wall of the channel 25H away from the
injector to act as ultrasonic velocity profilers (UVPs) and measure the internal flow structures. The
distance between the two transducers is 0.25H in the streamwise direction and the transducers are set at
different angles, ±8° from the vertical axis, while maintaining line symmetry. The active diameter of the
transducer used to generate the ultrasonic beam is 5 mm and the divergence half-angle of the beam is 2.2°.
In this set-up, UVP detects local instantaneous velocity of liquid phase along the measurement line at the
spatial resolution of 5 mm in the diameter and 0.5 mm in the beam direction. The resolution is insufficient
to detect velocity fluctuation smaller than the Kolmogorov length scale, η ~ 0.20 mm, in the channel
flows. However, the turbulence larger than the resolution is correctly measured in a non-invasive way.
Furthermore, local instantaneous variation of the gas-liquid interface is also captured by the same set-up
of UVP so that two-phase flow structure is visualized quantitatively as a function of time. The ultrasonic
beams generated by the two transducers cross near the upper wall of the channel, and thus the interactions
between bubbles that are migrating close to the upper wall and the liquid flow structure are investigated.
From the instantaneous velocity vector information obtained in the crossed area, which is obtained by the
two synchronized UVPs, turbulence characteristics such as the Reynolds shear stress profiles are
evaluated in a similar manner to the system used by Taishi et al. (2002) for turbulent pipe flows.
Ultrasonic absorbing boards are attached to both the upper and lower walls of the channel around the
measurement line of the UVPs to eliminate multiple reflections from ultrasonic waves that would
otherwise remain around the measurement section. We set the absorbing board to be a part of the channel
wall such that it would not affect the local boundary layer structures. To estimate the wall friction, a shear
transducer, which measures the local wall shear force directly, is mounted 21.25H away from the injector
on the upper wall. Similar sensors were used to investigate the wall shear stress of bubbly flows in several
previous studies (Guin et al. 1996; Kodama et al. 2000; Moriguchi and Kato 2002).
All of these measurement instruments are synchronized in the present set-up, so that we can
simultaneously capture data for multiple different physical quantities to evaluate the internal frictional
drag reduction mechanism enabled by RBI. The measurement conditions are summarized in Table 1. In
the table, the spatial resolution of 0.5 mm means a velocity profiling resolution in ultrasound beam
direction, which is determined by a half wavelength of ultrasonic pulse, cC/2fu, where c, C and fu are the
speed of sound in the liquid, the cycle number of waves constituting the ultrasonic pulse, and the
ultrasonic frequency, respectively. Since the velocity profile obtained from a single shot of ultrasonic
pulse remains noisy, UVP signal processor reduces the noise in the profile using repetition of ultrasonic
pulses to take short time average during tens of pulse repetition. In our experiments, the ultrasonic pulse
emission was repeated 28 times for noise reduction, which results in the temporal resolution at 10 ms.
The Reynolds number of the flow is defined as Re = Ubulkh/ν = Qliquid /(2Wν), where Ubulk, ν and
Qliquid are the bulk mean velocity, the kinematic viscosity and the volumetric flow rate of the liquid phase.

4
The hydraulic equivalent diameter of the channel flow is calculated to be 4.5H, and the Reynolds number
of a pipe-equivalent is therefore 4.5 times larger than the Re that was defined for the channel flow. In this
paper, the Re is defined under liquid single-phase conditions and is unmodified when considering the
bubbles’ contributions to the mean density, the effective viscosity, and the acceleration of the liquid flow
velocity. According to the friction coefficient (Cf) data measured from the shear transducer, there is a
transition regime from laminar to turbulent flows in the 800 ≤ Re ≤ 1200 range, as shown in Fig. 3. The
experiments described in this paper target flows where Re > 1200 to assess the effects of RBI in the
turbulent flow regime.

2.2 Repetitive bubble injection

The electromagnetic valve used to control the bubble injection has three different states, which are
switched using control signals created by the function generator: these are the always closed state, the
always open state and the periodically opened state. The definition of the temporal opening control of the
valve is illustrated in Fig. 4, where P, topen and Qgas are the valve operation period, the opening time of the
valve and the instantaneous gas flow rate, respectively. When the valve opens, the injected bubbles have a
local void fraction that is estimated using αlocal = Qgas/(Qliquid + Qgas).
Figure 5(a) to (d) shows photographs of bubble images taken at each phase Φ in a single injection,
from the top of the channel at the downstream measurement location where the bubbles flow from left to
right. The RBI conditions in this case are αlocal ~ 1.50%, P = 2.00 s and topen = 1.00 s. The streamwise
length of the bubble swarm is much longer than the side of the square photograph under these conditions.
Each part of the bubble swarm is composed of bubbles of different sizes, ranging from 1 mm up to 100
mm. The front region of the bubble swarm is led by some split air films. Because the initial state of the
bubbles right after bubble injection is confirmed to be homogenous in terms of bubble size, the air film
occupation at the front part of the swarm at the downstream location indicates that the air films migrate
slightly more quickly than the other bubbles. The evidence is shown in Fig. 6, where the averaged
streamwise velocities of bubbles are measured separately from the optical visualized images, i.e. Fig. 5.
This is consistently explained by the drag reduction effect of the air films, i.e., they accelerate the liquid
flow for reduced friction. The central part of the bubble swarm is composed of bubbles of various sizes.
In this region, the bubbles change their sizes because of fragmentation and coalescence. Finally, at the
rear part of the swarm, only small bubbles remain but some intermediate sized bubbles, 1 mm < Db ≤ 8
mm, form streamwise chains. This phenomenon is explained by bubble accumulation into low-speed
streaks of the turbulent boundary layer, where the buoyancy of the bubbles also supports their
accumulation (Harleman et al. 2011; Park et al. 2014). However, bubbles of Db ≤ 1 mm behave similarly
to passive scalar along the wall, which is quasi-uniformly dispersed in the channel near the wall
(Chouippe et al. 2014). Figure 5(e) shows a line-scanned image of the bubble swarm in two RBI cycles,
where time progresses from left to right, and the spatial structure of the swarm is interpreted as the
structure running from right to left. It has thus been confirmed that RBI produces the bubble swarms

5
periodically and stably with good similarity.
Figure 7 shows a sample of the measurement results obtained by the shear transducer under the same
conditions that were used for the previous visualization. The gray flat line in Fig. 7 indicates the averaged
shear stress obtained under the same conditions but without bubble injection. The wall shear stress
obtained from the shear transducer under the periodic passage of the bubble swarms also largely
fluctuated as expected from the bubble swarm visualized in Fig. 5. The graph shows the fluctuations in
two cycles of RBI, where there is reduced shear stress for much of the time, but several moments show
spiked increases in the stress above the average stress of the single-phase flow. In section 3, we will
present the statistical trends of the wall shear stress in detail.

2.3 Gas–liquid interface detection

The gas–liquid interface can be detected using the echo signal from pulsed ultrasound (Matikainen and
Irons 1986; Murakawa et al. 2008). The authors’ group also developed a combination of three different
algorithms that enabled various types of gas–liquid interface to be detected using a single ultrasonic
transducer (Murai et al. 2010). The echo amplitude of the ultrasonic pulse in the flow configuration
presented here is shown in Fig. 8(a). The data is obtained from synchronized measurements with the wall
shear stress shown in Fig. 7. In the graph, the values of y/h = 0 and 1 correspond to the upper wall surface
and the center of the channel, respectively. The bubbles and the tracer particles reflect ultrasonic pulses
with different reflection ratios, depending on their acoustic impedance relative to the surrounding liquid.
The echo amplitude provided by the tracer particles distributed in the liquid phase has an almost equal
value of approximately 250 ± 50 in an arbitrary digit code, which is expressed using white color. This
level is unmodified over time. In contrast, the ultrasound is unable to penetrate the gas phase and this
leads to total reflection from the first gas–liquid interface along the path of the ultrasound pulse. The echo
amplitude at the interface thus jumps up as it is reflected back to the ultrasonic transducer (see the red
spots in Fig. 8(a)). This point indicates that the gas–liquid interface is almost perpendicular to the
ultrasonic path. An interface with a larger displacement angle to the ultrasonic path provides an oblique
reflection of the ultrasound, which deflects the pulse away from the transducer. The echo amplitude is
thus found to have a smaller value than that of the tracer particles that are dispersed in the surrounding
liquid, which are identified by the blue spots in the figure. Consequently, both the red and blue spots in
the figure form a profile of the location and the shape of the gas–liquid interface. Thus, detection of the
variously-sized bubbles produced by RBI is possible if an adequate threshold is applied to the echo
amplitude distribution that is measured in the two-dimensional space-time domain. The corresponding
interface detection result is shown in Fig. 8(b) where the regions above the first interface, i.e., the bottom
surfaces of individual bubbles is colored black. The interface detection is performed by two-value
thresholding; the echo intensity under 180 or over 290. When compared with the bubble distribution
recorded by the camera (Fig. 5(e)), it is confirmed that the technique presented here using the echo
information works appropriately. However, it was noted that bubbles smaller than 2 mm will not always

6
be detected because of weakened mirror reflection within the ultrasonic beam diameter of approximately
5 mm.

2.4 Velocity vector measurements

The UVP measurements provide a velocity component in the beam direction at all positions along the
beam (Takeda 1991; 2012). Therefore, if two ultrasonic transducers are arranged in line symmetry as
shown in Fig. 9, a local velocity vector field is obtained near the beam crossing point. The velocity vector
at the beam crossing point, u, is calculated as u = (u, v) = ((ξ1 − ξ2)/2sinθ, (ξ1 + ξ2)/2cosθ), where ξi is the
velocity component along the ultrasonic beam direction, (u, v) are the streamwise and vertical velocity
components of the liquid flows, respectively, and θ is the angle of the beam from the vertical axis. In this
experiment, the transducers were arranged to measure the velocity vectors close to the upper wall: the
beam-crossing point is located at y/h ~ 0.22. For measuring the velocity vectors at all the position in
channel height, we apply a frozen hypothesis in the narrow region between two ultrasound beams;
velocity vectors due to turbulence being invariant during the short horizontal advection between the
beams. To perform this measurement with higher spatial resolution, the distance between the two
ultrasonic transducers was set to be short enough at 10 mm, while θ was set to be small enough at ±8°.
Reynolds shear stress from the velocity vector measured in the beam crossing volume is always
underestimated due to lack of spatial resolution. The underestimation expands as separating from the
beam cross point as reported by Taisi et al (2002). In our approach, the bias error is reduced by setting the
two transducers installed narrowly so that beam crossing volume is elongated in the vertical direction as
shown in Fig. 9. The beam crossing angle at θ = ±8° allows us to secure the beam crossing volume
reaching entire range of the half height of the channel; 0<y/h ≤ 1.11. Below this angle hinders accuracy of
streamwise velocity obtainable from the UVPs. When combining the detected gas–liquid interface and the
velocity vector fields in the liquid phase, the characteristics of the RBI-originated flow structure of
bubbly two-phase flows in the channel can be fully and consistently investigated using the UVPs.

2.5 Controllability of repetitive bubble injection

Figure 10 shows a probability distribution for bubble existence in a single period of RBI at the
downstream measurement location. The data is calculated by phase averaging of 12 swarms taken from
the ultrasonic echo information that was explained in the previous section, where the valve settings for
RBI are P = 2.00 s and topen = 1.00 s. Figure 11 shows the average duration of a bubble swarm, i.e., the
swarm persistence length in the time dimension, which is computed from the phase-averaged data of the
bubble probability by thresholding with a half value (0.5). The error bars in the figure represent the
standard deviation that occurs out of phase in RBI, which means cycle-by-cycle fluctuations in the swarm
length. The results show that the duration is longer than the corresponding injection time, topen, under all
valve conditions. This is explained by the streamwise diffusion of the bubbles, which is subject to a
velocity gradient with turbulence. In particular, the tail of the swarm, which consists of small bubbles,

7
tends to migrate more slowly than the head of the swarm, as mentioned earlier. This gradually expands
the swarm in the streamwise direction to elongate it from 25% up to 100% at x/H = 20 from the injection
point. However, the duration is controlled in accordance with the given injection time, and increases
monotonically with that injection time. It is also important to note that the duration is unaffected by
changes in the valve operation period, P. In addition, the standard deviation is restricted to remain
sufficiently small in comparison to the mean value. Therefore, we are able to conclude that the bubble
swarms generated by the RBI instruments presented here have sufficient controllability and
reproducibility to allow us to commence the parametric study.

3. Results and discussion

3.1. Drag reduction by continuous bubble injection

To discuss the effects of RBI, it is necessary to understand the drag reduction enabled by the commonly
used continuous injection style. Figure 12 shows the envelopes of the bubble interfaces in continuously
injected bubbles at αmean = 1.50% in a turbulent channel flow of Re ~ 2200, where αmean is the time-mean
void fraction, which is the same as the local void fraction (αlocal). The upper wall is almost entirely
covered with many bubbles under this condition. Figure 13 shows velocity distributions in the
single-phase flow and the two-phase flow. Relative gradual increase of the averaged velocity at the
single-phase flow indicates that the flow is an undeveloped turbulent flow, which is intentionally targeted
in the present study. There is a dip of velocity near the center of channel, y/h ~ 1, due to measurement
error of UVP from multi-reflection of ultrasonic wave. This technical problem is known among UVP
users; UVP inevitably produces erroneous velocity data in a limited range along the measurement line
when applied to the acoustic environment causing multi-reflection of ultrasonic wave. Channel flow
confined narrowly by two parallel plates is typical cases, and thus we encounter the difficulty to
completely avoid the appearance of local inaccurate data points (Murai et al. 2006). The position of
erroneous data appearance is adjustable by modifying ultrasonic pulse repetition intervals. Our solution
employed is to shift the erroneous range to the channel center so that the velocity profile near the wall is
measured properly.
Because a boundary of the turbulent buffer layer is estimated to lye around 40 wall units from the
upper wall in the single-phase flow case, almost all of these bubbles exist inside the original turbulent
boundary layer, as profiled in Fig. 13. The projection void fraction (αproj.), which is scaled on the top of
the graph, is defined as the time-mean ratio of the bubble segment at each depth from the wall to the total
sampling time obtained from the UVP echo signal. From the measured liquid streamwise velocity profiles,
it is clearly seen that the velocity with bubble injection is faster than that of the single-phase flow alone.
There are two reasons for this fact. One is decrease in hydrodynamic effective area for liquid flow due to
occupation of gas layer in the channel. This causes global velocity increment, being estimated by equation
of continuity. Another is alternation of stress boundary condition at the top of liquid phase from solid wall

8
to free surface. This provides asymmetric velocity profile with increasing of the velocity in the top half
and decreasing in the bottom half of the channel. These two effects appear simultaneously to result in the
measured liquid velocity profile that is largely accelerated in the upper half of the channel.
To obtain the basic drag reduction performance produced by continuous bubble injection, the wall
shear stress of the upper wall is measured as a function of the bulk-mean void fractions in the 0% ≤ αmean
≤ 1.50% range. The maximum value of αmean =1.50% is set thus because the upper wall is almost
completely covered with bubbles at this value. Figure 14 shows the time-averaged percentage of the drag
reduction effect, DR, on the upper wall, where the error bars represent the standard deviation of the
temporal fluctuation component. This effect is defined as DR = 1 − τmean/τ0, where τmean and τ0 are the
time-mean shear stress of the bubbly flows and the single-phase flow, respectively. The frictional
Reynolds number in the single-phase flow, which is defined by Reτ0 = huτ0/ν = h(τ0/ρ0)1/2/ν, is 160, where
the frictional velocity, uτ0, is determined using uτ0 = (τ0/ρ0)1/2, where ρ0 is the density of the liquid and h is
the channel half height. In the continuous injection case, drag reduction is observed at higher void
fractions, while the drag actually increases with bubble injection at lower void fractions, i.e., where αmean
≤ 0.50%. The same trend has been reported by a number of previous researchers (e.g. Murai et al. 2007;
Ceccio 2010). To estimate the drag reduction sensitivity, a gain factor is estimated from the measured wall
shear stresses; this factor is simply defined as G = DR/αmean. The gain indicates the magnitude of the drag
reduction percentage per unit of the percentage void fraction. If G exceeds unity, then it implies that the
drag reduction is effectively amplified by more than the density reduction of the fluid caused by the
mixing of the bubbles. Further large positive values of G indicate higher bubbly drag reduction
efficiencies. The gain factor of the continuous bubble injection as a function of the mean void fraction is
shown in Fig. 15, where the error bars indicate the standard deviations of the data. The results show that
the gain factor has a high value of around 17 at αmean ~ 1.50%, and that it gradually descends with
decreasing αmean. The descent trend becomes steep at around αmean < 1.0% and the gain factor falls to a
negative value for αmean < 0.5%. From the data acquired, we require bubbles that are at least larger than
0.5% in terms of their mean void fraction to obtain acceptable drag reduction performance, as long as the
bubbles are injected continuously over time.

3.2. Drag reduction with repetitive bubble injection

The time chart of RBI is defined by opening the electromagnetic valve for short times (topen) during the
RBI operation period (P) with a fixed local void fraction (αlocal), as shown in Fig. 16. The simplest
expectation is that the local value of the gain factor (G) during the passage of the bubble swarms would
maintain the same value under continuous bubble injection conditions at the same αlocal. In this case, the
total drag reduction effect, DR, is estimated using the occupation rate of the gas–liquid two-phase
condition in time as:
DR = topenP−1 αlocalG. (1)

9
In comparison with this, various complex phenomena would have to be combined to produce extra drag
reduction effects in reality, which can be expressed as deviations (DR+) from the above formula as
follows:
DR = topenP−1αlocalG + DR+. (2)
To evaluate the sign and the magnitude of DR+ clearly, we set a sufficiently high local void fraction,
where αlocal ~ 1.50%, in this series of RBI performance tests. We set this void fraction for two reasons.
The first is scientific interest, i.e., the drag reduction retains large sensitivity around this void fraction, as
we confirmed for continuous bubble injection (see Fig. 15), and it is therefore interesting to see how the
effect is amplified by RBI for the same local void fraction. Another reason is a technical restriction, i.e.,
the extra component DR+ may be buried within the standard deviations of the shear stress when a low
local void fraction is used for the RBI. Table 2 shows the experimental conditions used to estimate the
drag reduction by RBI in the flow at the fixed Reynolds number of Re ~ 2200, which was also used in the
previous chapter.
Figure 17 shows the percentage of the time-averaged drag reduction (DR) during RBI as the mean
void fraction αmean increases. Here, αmean is varied using a combination of two parameters, i.e., the valve
operation period (P) and the valve opening time (topen). The plots and the error bars denote the mean
values and their standard deviations, respectively, which are statistics taken from 20 measurement cycles,
each of which takes a sampling time of 20 s. It is confirmed that the DR always has positive values over
the entire tested mean void fraction range from 0 to 1.50%. Thus, the increasing drag region has been
completely eliminated by RBI. The dashed line in the figure indicates the expected value of DR that was
calculated using Eq. (1). When αmean ≥ 0.30%, the DR curve shows a convex variation because the
measured value is larger than the expected value. This new result means that RBI provides a positive
extra effect in terms of promoting drag reduction, which is associated with more complex responses, and
happened to a two-phase turbulent boundary layer when subjected to periodic passage of bubble swarms.
Figure 18 shows the extra component of the drag reduction percentage, DR+, which was defined by
and computed using Eq. (2). Although the standard deviation cannot be ignored in any case, the extra
effect obtained is approximately 2% for αmean < 0.3%. This value of 2% might be considered to be low in
engineering applications, but we must emphasize the fact that a positive value is obtained, which infers
that clarification of the internal two-phase flow structure could lead to further amplification of DR+.

3.3 Modification of the velocity field by RBI

The RBI provides not only void fraction fluctuations, but also acceleration and pulsation of the liquid
flow both in and out of the boundary layer. These influences on each phase interact to produce a new
effect in the form of the extra drag reduction mentioned earlier. Figures 19 and 20 show the
time-averaged distributions of the projection void fraction under each RBI condition and the
corresponding distributions of the mean streamwise velocity in the liquid phase, which is accelerated by
the bubbles. The data for the single-phase flow and the continuous bubble injection regimes are attached

10
to these graphs for comparison. From Fig. 19, it is confirmed that the local void fraction increases
monotonically with increasing valve-opening time in any valve operating period case. The void fraction
profiles seem to be similar to each other among all cases, including that of continuous bubble injection.
Thus, RBI does not lead to any special phenomenon occurring in the vertical profile of the time-averaged
void fraction. In contrast, the data in Fig. 20 shows that the mean liquid velocity in the RBI case remains
almost the same as that of the single-phase flow case, unlike the continuous bubble injection case. This is
explained by the disappearance of the bubble blockage effect, which is significant only for continuous
bubble injection. By considering the mean liquid velocity profiles near the upper wall, it is also found that
the velocity is locally accelerated with RBI in comparison to the single-phase flow value. This fact is
consistent with the interpretation that the liquid phase flows smoothly near the upper wall when RBI
provides a high level of drag reduction. From these two findings, it can be concluded that RBI
concentrates all the effects of the bubbles into the boundary layer structure without changing the mean
flow field out from the boundary layer.
To clarify the scenario described above, the phase-averaged streamwise velocity of the liquid phase
is calculated. Here, the existence of bubbles at each measurement point is distinguished by the echo
intensity, and only the velocity data corresponding to the liquid phase is used to calculate the streamwise
velocity. One of the sample results is shown in Fig. 21, where points at which all of the velocity data are
distinguished as being gas phase data are represented as a white area. The experimental conditions used
for the sample were P = 2.00 s and topen = 1.00 s, which are the same as those for Fig. 8, Fig. 10 and Fig.
20(d). The figure clearly indicates that the streamwise velocities of the liquid phase under the bubble
swarm are faster than those before the passage of the swarm. This becomes so because hydrodynamic
effective cross-sectional area of the liquid flow is narrowed due to occupation of gas layer near the upper
wall. Also, the replacement of the upper boundary for liquid flow from solid wall to free surface induces
asymmetric base flow, partly analogous to open channel flows. These two effects make the streamwise
velocity larger in the upper half of the channel. On the other hand, the velocity pulsation of the liquid
phase is insignificant in far region from the bubble swarm, y/h > 0.4. It is because the liquid phase near
the bubble swarm, y/h < 0.3, is accelerated enough to conserve volume flow rate of the liquid phase. In
general, artificial pulsations into a wall-bounded flow increase the average wall shear stress in turbulent
flow regimes (Scotti and Piomelli 2001; Blel et al. 2009). However, our RBI result does not belong to this
category and behaves rather more like a shear-thinning fluid, which has lower flow resistance in pulsatile
flows.
To discuss the pulsation effect that is caused in the liquid phase, we must consider the possibility
that the pulsatile flow itself promotes drag reduction, regardless of the material used for repetitive
injection. To evaluate this consideration experimentally, the drag reduction percentage (DR) is measured
in the case where the same amount of liquid is injected instead of the air bubbles at the same local volume
fraction, αlocal ~ 1.50%, from the injector. The results are shown in Fig. 22, where the repetitive injection
conditions are topen = 1.00 s, P = 2.00 s and αmean ~ 0.76%. In both the cases of continuous and repetitive

11
injections, the injection of the liquid causes large increases in the frictional drag, which results in negative
DR values. Moreover, the repetitive injection makes the drag increment more than that with the
continuous injection although amount of injected liquid is sufficiently low. In contrast, air bubble
injection maintains positive DR values. Thus, the pulsatile effect contributes to the drag reduction only in
the air bubbles case. When we consider the regeneration of the vortices by the pulsatile flow (Sykes et al.
1986) and the increase in the mean velocity caused by the liquid injection, the increasing percentage of
drag caused by the liquid injection is understood to be reasonable. If the pulsations on the main flow
promote drag reduction with RBI, then the wall shear stress should be lower than that in the continuous
injection case, even with liquid injection. However, with repetitive liquid injection, the shear stress
becomes higher than that in the continuous injection case. Kim and Sung (2003) simulated the effects of
periodic blowing in a wind cross-flow and concluded that the turbulence intensities and the Reynolds
shear stress are enhanced by injection in the downstream regions. Their conclusions agree with the
presented experimental indications under the liquid injection conditions.
Based on these facts, we reach an understanding that the pulsatile flows of the liquid phase that are
caused by RBI will enhance the wall shear stress, but this is not the case when the high-speed flow region
coincides with a region that has a gas–liquid interface as a near-wall side boundary. In parallel, the local
drag reduction caused by the bubble swarm allows the liquid beneath the swarm to flow more quickly.
These two phenomena, which happen naturally at the boundary layer during RBI, provide stable and extra
drag reduction.
For a deeper understanding of what occurs during RBI in the turbulent channel flows, the Reynolds
shear stress profiles are calculated as averages of u'v', where u' and v' are the time fluctuation components
of the velocity vector field in the liquid phase that were measured using the pair of ultrasound transducers.
The profiles, which were normalized with reference to uτ0, are shown in Fig. 23. It should be reminded
that the Reynolds shear stress profiles obtained using the two synchronized UVPs are generally
underestimated as explained before. The profiles presented here therefore show the Reynolds shear stress
components organized by large eddies resolvable by the UVPs. The profile for the single-phase condition
at least agrees with a previous report measured at a similar Re (Wei and Willmarth 1989), although it
causes 11% underestimation in the case of UVP. Furthermore, let us reconsider that the flow targeted here
is not fully developed turbulent channel flow, but is in development at low Re numbers. In the developing
state, local laminar velocity profiles passes occasionally so that the near wall region has a velocity
gradient calmer than fully turbulent flows as we already mentioned previously. This is why the
time-average Reynolds shear stress is earlier grown in the channel center so that it is reversely measured
lower near the wall. Consistently, the local inclination of the Reynolds shear stress at the channel center
for the single-phase flow is almost the same as that estimated from the wall shear stress indicated by the
grey diagonal line shown in the figure. Although the underestimation in Reynolds shear stress occurs,
they do represent the same tendencies as the real values if we assess only the relative change. When
compared with the single-phase condition profile, all profiles in the bubbly flows vary authentically. In

12
the bubbly flow, the Reynolds shear stress has higher values than that in the single-phase flow at the
far-wall region, where y/h > 0.4. There are two reasons for this behavior: the downward displacement of
the shear layer caused by the intrusion of the bubbly two-phase layer near the upper wall, and the
accelerated liquid flow in the central part of the channel for a certain blockage effect. The increase in the
peak value explains that the momentum-transferring eddies are activated at the position, being consistent
to our previous work of flow visualization (Park et al, 2015) which found that turbulent eddies are
survived and conveyed below split air-films. However the peak does not directly affect the wall shear
stress due to shielding effect of void layer near the wall. The shielding effect means a phenomenon that
turbulence can’t migrate beyond the air-films as the films are longer than the turbulent eddies. This is
consistently explained by the relaxation of Reynolds shear stress gradient roughly identified in the graphs.
The Reynolds shear stress gradients become about 0.3 to 0.4 times smaller than those of the single-phase
flow, and this tendency is qualitatively analogous to the data of Kitagawa et al. (2005). Since the
Reynolds shear stress gradient is proportional to the wall shear stress in fully developed turbulent flows,
the relaxation of the gradient in bubbly flow explains delay of turbulent momentum transfer in the
wall-perpendicular direction. Therefore, repetitive bubble injection provides turbulence-modification type
of drag reduction due to intermittent shielding effect of the air films which intensifies the local peak
below the air films but calms down the gradient of Reynolds shear stress far from the wall.
Here we need to clearly distinguish the terminology on the effect of void layer; “blockage effect”
and “shielding effect” which have different meaning. Blockage effect induces liquid velocity increment
entirely in the channel as liquid volume flow rate is fixed during bubbles occupy the near wall layer.
Shielding effect is cut-off of turbulence momentum diffusion beyond the layer of individual air-films. In
RBI operation, a point of study is the fact that these two effects are mixed and interacted unsteadily as
being paid attention in the next section.

3.4. RBI sensitivity to different Reynolds numbers

In this section, the sensitivity of RBI to the Reynolds number (Re) is investigated to evaluate applicability
of RBI to higher Re conditions. We examined the single-phase flow case, the continuous bubble injection
case and two different cases of RBI. The gas flow rate is fixed as a constant in all cases, while the liquid
flow rate is increased to realize high Re flows so that the time-mean void fraction αmean decreases for
higher Re conditions. The experimental conditions are summarized in Table 3.
The drag reduction obtained for each Re is shown in Fig. 24. Because the mean void fraction is set to
be low (αmean < 0.55%; see Figs. 13 and 14), continuous bubble injection does not significantly modify the
frictional drag under all Re conditions (see the dark gray boxes). The drag actually increases under lower
Re conditions, which means that the drag reduction produced by bubble injection is advantageous for
higher Re conditions. This trend has already been reported in many types of flow facility (Berg et al.
2007; Murai et al. 2014). In contrast, RBI produces a different drag reduction tendency (see the light gray
and white boxes). At Re=1700, the drag is increased by RBI to about 13%, which is 10% higher than that

13
for continuous bubble injection. However, RBI does not change the drag so greatly at Re=2300. At the
highest condition of Re=2700, RBI clearly starts to reduce the drag, particularly under the gently injected
bubble conditions, where topen = 1.00 s. If the bubbly drag reduction occurs due to the modified viscosity,
percentage of drag reduction would be decreased as the Reynolds number increases. However, the data
shows that drag reduction is rather promoted. This implies that there is insignificant role in the modified
viscosity in present operation of RBI.
The modification of the Reynolds shear stress events has already been shown and discussed.
However, the Reynolds shear stress is a time statistic-based value, and it is therefore difficult to discuss its
relationship to the internal unsteady flow structure of the liquid caused by RBI. Here, we visualize
time-dependent profiles of the Reynolds shear stress that is modified by the bubble swarms. Based on the
reproducibility of both the bubble swarms and the corresponding flow field, the time-dependent shear
stress field is computed as the phase-average distribution. Figure 25 shows the result for topen =0.50 s for
three different Re, for which the color bar is normalized relative to the squared friction velocity of the
single-phase condition, uτ0. The data in the figure indicate the differential component of the local value of
u’v’ from the time-average distribution of the single-phase flow. Thus, the red and blue regions in the
figure indicate increasing and decreasing local Reynolds shear stress, respectively. The bubble swarm is
painted black in each figure, where the front of each swarm is set at the initial phase angle, Φ = 0 rad. The
bubble swarms depicted are defined by their phase-averaged structures, which are distinguished as areas
that have gas-existence probabilities that are higher than 0.50.
At low Re, the bubble swarm accompanies an area of high local Reynolds stress underneath the
swarm (Fig. 25(a)). This means that the bubble swarm produces velocity fluctuations to its ambience
because the flow field remains in a nearly laminar state at this Re. With increasing Re (Fig. 25(b)), the
velocity fluctuations can be transported downstream and can survive against viscous dissipation, and this
promotes the flow transition to a turbulent flow as a trigger of the disturbance. At even higher Re, the high
Reynolds shear stress is no longer produced beneath the bubble swarm (Fig. 25(c)), and in contrast, the
region turns towards reduction of the shear stress. This local shear stress reduction is explained by the
relaxation of the turbulence in the vicinity of the gas–liquid interface, i.e. the free-slip stress condition
halts the growth of the turbulence eddies. In addition, the bubble swarm in the drag reduction state can
flow downstream slightly more quickly than the liquid flow velocity. With this relative motion, the bubble
swarm sweeps the fluid containing the vortical structures of the turbulence ahead of the swarm. This
phenomenon was visualized in our previous work (Park et al. 2014), and we succeeded in obtaining
consistent data for the Reynolds shear stress distribution here. In the previous paper, we confirmed that
the streamwise vortices in the buffer layer are swept under the bubble swarm and restore themselves after
passing the swarm. It is generally known that these streamwise vortices and their bursting promote
momentum transfer that results in the Reynolds shear stress. Therefore, a combination of the observed
behavior of the streamwise vortices and the measured Reynolds shear stress reduction under the bubble
swarm indicates that the swept streamwise vortices play a major role in lowering the momentum transfer.

14
Figure 26 shows the quadrant expression for the velocity fluctuations in the same layer, where y/h ~ 0.25,
in the single-phase flow and under the bubble swarm, in the 0 rad < Φ < 2π/3 rad range, generated by RBI
with topen = 0.50 s at Re ~ 3000. Figure 26(a) and (b) show values calculated from the instantaneous
velocity vectors obtained by the UVPs. Under the bubble swarm, the scattering of the velocity fluctuation
shrinks in the vertical direction and lies more in the streamwise direction. This means that the turbulent
vortical structure only survives in the tangential direction to the bottom surface of the bubble swarm, and
loses the momentum transfer action.

4. Conclusions
We proposed the RBI technique to promote frictional drag reduction. RBI is proposed specifically to
improve the drag reduction performance at low mean void fractions. The performance of RBI is
investigated comprehensively by measurement of the temporal variation of the local wall shear stress, the
velocity vector field of the liquid phase, and the gas–liquid interface in turbulent horizontal channel flows.
The flow conditions examined for the present RBI performance test are Reynolds number (Re) of 1700 ≤
Re ≤ 3000, mean void fraction (αmean) ofαmean ≤ 1.5%, and bubble size (Db) of Db ≤ 100 mm, respectively.
RBI successfully generates bubble swarms periodically with suitably high reproducibility; these swarms
have split air films on their front and small bubbles on their tails. These bubbles flow near the upper wall
of the channel, which affect the upper boundary layer structure. This induces liquid flow velocity
increment due to blockage effect for void occupation near the wall, and provides asymmetric mean
velocity profile respect to the centerline of the channel. Another impact is alternation of upper stress
boundary condition for liquid phase, which produces shielding effect of turbulence, i.e. presence of free
surface near the solid wall. In RBI operation, these two effects take place unsteadily and interact to each
other so that a new phenomenon comes up. As we expected prior to our experiments, the bubble swarms
maintain a high gain factor for drag reduction because they provide locally high void fractions
intermittently at low mean void fractions, which means that strong two-way interactions between two
phases occur inside the turbulent boundary layer. When compared with continuous bubble injection, it is
confirmed that RBI eliminates the drag increment region that previously occurred at low mean void
fractions, and also produces an extra drag reduction component in addition to the void concentration
effect. This extra effect is assessed via measurement of RBI’s pulsatory effects on the liquid flow field. It
is found that the mean liquid velocity at the far-wall region is not modified by RBI, while it is accelerated
in the continuous bubble injection case. Instead, the near-wall structure of the turbulence is effectively
altered by RBI. Evaluation of the wall shear stress for different Reynolds numbers indicated that RBI has
advantages for application to the higher Re turbulent flows in the tested range. The Reynolds shear stress
measured beneath the bubble swarm is obviously reduced for higher Re, which indicates relaxation of the
turbulence because of the gas–liquid interface that lies close to the upper wall. When we consider these
facts along with the results of our previous report (Park et al. 2014), it is suggested that the vortical

15
structures that were swept by the local air films that lead the bubble swarm are reduced by these bubble
swarms. In summary, the promotion of bubbly drag reduction by RBI is found to be provided by the
following two major effects: i) isolation of the wall surface from the liquid flow by the reproducible
intrusion of air films that naturally emerge in front of the bubble swarms during their advection; and ii)
relaxation of the Reynolds shear stress events beneath the bubble swarm through sweeping of the vortical
structures. Concerning the application of RBI to small bubbles comparable to turbulent eddy scales, we
could expect different mechanism of drag reduction in which certain relaxation time outstands to sway the
time-average drag. This is regarded as a separate topic, and set as our next study.

Acknowledgments
This work was supported by the Fundamental Research Developing Association for Shipbuilding and
Offshore (REDAS), and by JSPS KAKENHI (Grant Nos. 24246033 and 23760143). The authors would
like to express their thanks for this support.

References
Blel, W., Le Gentil-Lelièvre, C., Bénézech, T., Legrand, J., Legentilhomme, P., 2009. Application of
turbulent pulsating flows to the bacterial removal during a cleaning in place procedure. Part 1:
Experimental analysis of wall shear stress in a cylindrical pipe. J. Food Eng. 90, 422–432.
Berg, T.H. van den, Gils, D.P.M. van, Lathrop, D.P., Lohse, D., 2007. Bubbly turbulent drag reduction is a
boundary layer effect. Phys. Rev. Lett. 8, 98, 084501.
Ceccio, S.L., 2010. Frictional drag reduction of external flow with bubble and gas injection. Annu. Rev.
Fluid Mech. 42, 183–203.
Chouippe, A., Climent, E., Legendre, D., Gabillet, C., 2014. Numerical simulation of bubble dispersion in
turbulent Taylor–Couette flow. Phys. Fluids. 26, 043304.
Dean, R.B., 1978. Reynolds number dependence of skin friction and other bulk flow variables in
two-dimensional rectangular duct flow. Trans. ASME J. Fluids Eng. 100, 215–223.
Elbing, B.R., Winkel, E.S., Lay, K.A., Ceccio, S.L., Dowling, D.R., Perlin, M., 2008. Bubble-induced
skin-friction drag reduction and the abrupt transition to air-layer drag reduction. J. Fluid Mech., 612,
201–236.
Guin, M.M., Kato. H., Yamaguchi, H., Maeda, M., Miyanaga, M., 1996. Reduction of skin friction by
microbubbles and its relation with near-wall bubble concentration in a channel. J. Mar. Sci. Technol.
1, 241–254.
Harleman, M.J.W., Delfos, R., Terwisga, T.J.C.V., Westerweel, J., 2011. Dispersion of bubbles in fully
developed channel flow. J. Phys.: Conf. Ser. 318, 052007.

16
Hoang, C., Toda, Y., Sanada, Y., 2009. Full scale experiment for frictional resistance reduction using air
lubrication method. International Offshore and Polar Engineering Conference. 812–817.
Jacob, B., Olivieri, A., Miozzi, M., Campana, E.F., Piva, R., 2010. Drag reduction by microbubbles in a
turbulent boundary layer. Phys. Fluids. 22, 115104.
Kato, H., Iwashina, T., Miyanaga, M., Yamaguchi, H., 1999. Effect of microbubbles on the structure of
turbulence in a turbulent boundary layer. J. Mar. Sci. Technol. 4, 155–162.
Kawamura, T., Kodama, Y., 2002. Numerical simulation method to resolve interaction between bubbles
and turbulence. Int. J. Heat and Fluid Flow. 26, 627–638.
Kim, K., Sung, H.J., 2003. Effects of periodic blowing from spanwise slot on a turbulent boundary layer.
AIAA J. 10, 41, 1913–1921.
Kitagawa, A., Hishida, K., Kodama, Y., 2005. Flow structure of microbubble-laden turbulent channel flow
measured by PIV combined with the shadow image technique. Exp. Fluids. 38, 466–475.
Kodama, Y., Kakugawa, A., Takahashi, T., Kawashima, H., 2000. Experimental study on microbubbles
and their applicability to ships for skin friction reduction. Int. J. Heat and Fluid Flow. 21, 582–588.
Kodama, Y., Takahashi, T., Makino, M., Hori, T., Ueda, T., Kawamura, N., Shibata, M., Kato, H., Inoue,
T., Suzuki, T., Toda, Y., Yamashita, K., 2005. Practical application of microbubbles to ships – Large
scale model experiments and a new full scale experiment. International Symposium on Smart
Control of Turbulence. 213–218.
Kodama, Y., Hinatsu, M., Hori, T., Kawashima, H., Takeshi, H., Makino, M., Ohnawa, M., Sanada, Y.,
Murai, Y., Ohta, S., 2008. A Full-Scale Air Lubrication Experiment Using a Large Cement Carrier
for Energy Saving (Result and Analysis). The Japan Society Naval Architects and Ocean Engineers.
163–166. (in Japanese)
Matikainen, L., Irons, G. A., 1986. Ultrasonic system for the detection of transient liquid/gas interfaces
using the pulse-echo technique. Rev. Sci. Instrum. 57, 1661–1666.
McCormick, M.E., Bhattacharyya, R., 1973. Drag reduction of a submersible hull by electrolysis. Naval
Engineers J. 85, 11–16.
Meng, J.C.S., Uhlman, J.S., Jr., 1989. Microbubble formulation and splitting in a turbulent boundary layer
for turbulence reduction. Advances in Fluid Dynamics. Springer New York, 168–217.
Moriguchi, Y., Kato, H., 2002. Influence of microbubble diameter and distribution on frictional resistance
reduction. J. Mar. Sci. Technol. 7, 79–85.
Murai Y., Fujii, H., Tasaka, Y., Takeda, Y., 2006. Turbulent bubbly channel flow investigated by
ultrasound velocity profiler. J. Fluid Sci. Tech., 1, 12-23.
Murai, Y., Fukuda, H., Oishi, Y., Kodama, Y., Yamamoto, F., 2007. Skin friction reduction by large air
bubbles in a horizontal channel flow. Int. J. Multiphase Flow. 33, 147–163.
Murai, Y., Tasaka, Y., Nambu, Y., Tasaka, Y., Gonzalez, S.R., 2010. Ultrasonic detection of moving
interfaces in gas-liquid two-phase flow. Flow Measurement and Instrumentation. 21, 356–366.
Murai, Y., 2014. Frictional drag reduction by bubble injection. Exp. Fluids. 55, 1733.

17
Murakawa, H., Kikura, H., Aritomi, M., 2008. Application of ultrasonic multi-wave method for two-phase
bubbly and slug flows. Flow Measurement and Instrumentation. 19, 205–213.
Oishi, Y., Murai, Y., 2014. Horizontal turbulent channel flow interacted by a single large bubble. Exp.
Thermal and Fluid Sci., 55, 128–139.
Park, H.J., Oishi, Y., Tasaka, Y., Murai, Y., Takeda, Y., 2009. Turbulent shear control with oscillatory
bubble injection. J. Phys.: Conf. Ser. 147, 012037
Park, H.J., Tasaka, Y., Murai, Y., Oishi, Y., 2014. Vortical stuctures swept by a bubble swarm in turbulent
boundary layers. Chem. Eng. Sci. 116, 486–496.
Scotti, A., Piomelli, U., 2001. Numerical simulation of pulsating turbulent channel flow. Phys. Fluids. 13,
1367–1384.
Sykes, R., Lewellen, W., Parker, S., 1986. On the vorticity dynamics of a turbulent jet in a crossflow. J.
Fluid Mech. 168, 393–413.
Taishi, T., Kikura, H., Aritomi, M., 2002. Effect of the measurement volume in turbulent pipe flow
measurement by the ultrasonic velocity profile method (mean velocity profile and Reynolds stress
measurement). Exp. Fluids. 32, 188–196.
Takeda, Y., 1991. Development of an ultrasonic velocity profile monitor. Nucl. Eng. Des. 126, 277–284.
Takeda, Y. (Ed.), 2012. Ultrasonic Doppler velocity profiler for fluid flow. Springer Japan.
Wei, T., Willmarth, W.W., 1989. Reynolds–number effects on the structure of a turbulent channel flow. J.
Fluid Mech. 204, 57–95.
Xu, J., Maxey, M.R., Karniadakis, G.E.M., 2002. Numerical simulation of turbulent drag reduction using
micro-bubbles. J. Fluid Mech. 468, 271–281.

18
Fig. 1. Schematic representations of wall shear stress modified by bubble injection parameterized in terms
of (a) mean void fraction and (b) bubble size, where the flat lines represent the time-mean shear stress in
the single-phase flow (τ0), αc is the critical void fraction required to obtain drag reduction, and δ is the
turbulent boundary layer thickness.

19
Fig. 2. Schematic diagram of experimental setup: (a) overview and (b) details of the test section of the
channel, where x and y are the distances from the channel inlet and from the upper wall, respectively, and
the dashed lines indicate the ultrasonic beam paths from the ultrasonic transducer pair.

20
Fig. 3. Relationship between the friction coefficient (Cf) and the Reynolds number (Re) under
single-phase conditions for the presented experimental channel (Park et al. 2014), where the solid lines
indicate the variations in Cf reported by Dean (1978) and the dots are experimental data obtained using
the present facility.

21
Fig. 4. Time chart of the controlled gas flow rate for bubble swarm generation by the electromagnetic
valve, where P, topen and Qgas are the valve operation period, the valve opening time and the instantaneous
gas flow rate, respectively.

22
Fig. 5. Photographs of bubble swarms generated by RBI that were taken by a camera located above the
ultrasonic transducers; (a) – (d) show pictures at Φ = 0.3π rad, 0.8π rad, 1.3π rad and 1.8π rad,
respectively, and (e) shows a line scanned image, where Re ~ 2200 in the horizontal channel flow and the
RBI is controlled with αlocal ~ 1.50%, P = 2.00 s and topen = 1.00 s.

23
Fig. 6. Averaged streamwise velocity of bubble swarms, which is shown in Fig. 5, where the error bars
and the gray line indicate the bulk velocity of test fluid and the standard deviation, respectively.

24
Fig. 7. Sample of the wall shear stress with bubble swarms generated by RBI, which is controlled using
αlocal ~ 1.50%, P = 2.00 s and topen = 1.00 s, where Re ~ 2200 in the horizontal channel flow and the gray
line indicates the average shear stress value without the bubbles.

25
Fig. 8. Sample of results obtained by the UVPs: (a) the original echo amplitude distribution and (b) the
streamwise velocity distribution in the liquid phase with the detected gas–liquid interface, where the echo
and the velocity are measured simultaneously with the wall shear stress shown in Fig. 7. (For
interpretation of the references to the colors shown in this figure, the reader is referred to the web version
of this article.)

26
Fig. 9. Relationships among the measured velocities and the flow velocity vector, where ξi is the velocity
component measured by UVP and u = (u, v) is the flow velocity vector.

27
Fig. 10. Probability distribution of bubble existence calculated based on the phase statistics of the gas
phase position detected from the ultrasonic echo information, where the RBI setting values are P = 2.00 s
and topen = 1.00 s.

28
Fig. 11. Durations of the bubble swarms recorded at the fixed point on the upper wall, where the error
bars indicate the standard deviation.

29
Fig. 12. Envelope of bubble interfaces for continuously injected bubbles with αmean ~ 1.50% in turbulent
flow with Re ~ 2200.

30
Fig. 13. Mean streamwise velocity distribution in the liquid phase modified by continuous bubble
injection (square symbol) and the distribution of the projection void fraction (solid line) obtained from the
UVP echo signal, which was measured at the bulk mean void fraction of αmean ~ 1.50%. The gray dashed
line indicates the boundary between the buffer layer and the logarithmic layer in the single-phase flow
that corresponds to 40 wall units.

31
Fig. 14. Drag reduction at the upper wall in continuous bubble injection regime, where the error bars
indicate the standard deviations of the data.

32
Fig. 15. Gain factor of drag reduction in continuous bubble injection regime, where the error bars indicate
the standard deviations of the data.

33
Fig. 16. Concept of RBI, where DR+ indicates extra drag reduction effects on the wall shear stress
produced by RBI.

34
Fig. 17. Time-averaged drag reduction for wall shear stress with RBI, where the dashed line indicates the
drag reduction predicted from that of the continuous bubble injection regime, and the error bars denote
the standard deviations of the data.

35
Fig. 18. Extra drag reduction produced by RBI (DR+), where the error bars indicate the standard
deviations of the data.

36
Fig. 19. Projection void fraction distributions under each of the RBI conditions: (a) P = 5.00 s, (b) P =
4.00 s, (c) P = 3.00 s and (d) P = 2.00 s.

37
Fig. 20. Mean streamwise velocity distributions in the liquid phase when accelerated by bubble injection:
(a) P = 5.00 s, (b) P = 4.00 s, (c) P = 3.00 s and (d) P = 2.00 s.

38
Fig. 21. Phase-averaged streamwise velocity of the liquid phase flow when modified by controlled bubble
injection under valve conditions of P = 2.00 s and topen = 1.00 s, where the points at which all of the
velocity data are distinguished as being gas phase are represented as white areas. (For interpretation of the
references to the colors shown in this figure, the reader is referred to the web version of this article.)

39
Fig. 22. Modifications to the drag reduction caused by injection of bubbles or the same amount of liquid
with αlocal ~ 1.5%, where the gray line expresses each value for the single-phase flow and the repetitive
injection conditions are P = 2.00 s, topen = 1.00 s and αmean ~ 0.76%, which is half of the value used for
continuous injection.

40
Fig. 23. Reynolds shear stress profiles for different bubble injection periods: (a) P = 5.00 s, (b) P = 4.00 s,
(c) P = 3.00 s and (d) P = 2.00 s, where the grey lines denote the wall shear stresses for single-phase
flows obtained using a shear transducer.

41
Fig. 24. Drag reduction at each Reynolds number, where the mean void fraction is fixed to be the same at
each Reynolds number.

42
Fig. 25. Phase-averaged Reynolds shear stress for RBI with topen = 0.50 s: (a) Re ~ 1700, (b) Re ~ 2300
and (c) Re ~ 3000, where the black areas represent bubbles. (For interpretation of the references to the
colors shown in this figure, the reader is referred to the web version of this article.)

43
Fig. 26. Velocity fluctuation distributions at y/h ~ 0.25 and Re ~ 3000: (a) single-phase and (b) under
bubble swarms, in the 0 rad < Φ < 2π/3 rad range, generated by RBI with topen = 0.50 s, where the black
lines are linearly fitted gradients of the plots and the values given at each corner denote the fraction of the
plots that exist in each area.

44
Table 1. Measurement conditions.
Conditions for tracer particles
Particle relaxation time (tp) 0.27 ms
Maximum Stokes number (Stmax) 0.07 -
Conditions for UVPs
Ultrasonic frequency 4 MHz
Cycles of ultrasonic pulse 4 -
Pulse repetitions 28 -
Time resolution 10 ms
Velocity resolution in bream direction 2.37 mm/s
Spatial resolution in bream direction 0.5 mm
Ultrasonic beam diameter 5 mm
Conditions for shear transducer
Measurement area 25π (mm)2
Shear stress resolution 0.05 Pa
Time resolution 50 ms

45
Table 2. Experimental conditions.
Reynolds number (Re) 2200 -
Frictional Reynolds number in single-phase flow (Reτ0) 160 -
-2
Local void fraction (αlocal) 1.5 × 10 -
-2
Averaged void fraction (αmean ) 0 – 1.5 × 10 -
RBI period (P) 2.00, 3.00, 4.00 and 5.00 s
Injection time (topen) 0.50, 1.00 and 1.50 s

46
Table 3. Experimental conditions.
Reynolds number (Re) 1700 2300 3000
Frictional Reynolds number in single-phase flow (Reτ0) 126 160 200
Mean void fraction (αmean) 0.44 % 0.33 % 0.26 %
Injection time (topen) = 0 s Single-phase flow (αlocal = αmean = 0)
Injection time (topen) = ∞ s Continuous injection (αlocal = αmean)
Injection time (topen) = 1.0 s RBI period (P) = 3.00 s (αlocal = 3αmean)
Injection time (topen) = 0.5 s RBI period (P) = 3.00 s (αlocal = 6αmean)

47
Highlights

 Effects of repetitive bubble injection for drag reduction have been examined.

 The repetitive method greatly improved drag reduction even at low void fractions.

 Reproducible bubble swarms followed by air films were main factors in improvement.

 Turbulence modification realized using the method has been measured and discussed.
Keywords: Drag reduction; Gas-liquid two-phase flow; Boundary layer control; Turbulent flow

Abstract: To promote the efficiency of frictional drag reduction using bubbles, we designed a novel
bubble control method that involves repetitive injection of bubbles rather than the conventional
continuous bubble injection approach. Even if the mean void fraction of bubbles to be injected into the
turbulent boundary layer is set to be low, repetitive bubble injection (RBI) maintains the frictional drag
reduction by generating bubble swarms. The enhanced drag reduction mechanism and the
effectiveness of the RBI approach are investigated by studying wall shear stress and the velocity vector
field in the liquid phase when measured in a turbulent horizontal channel flow. The bubble swarms
generated by RBI consist of bubbles of various sizes with leading large air films, which have high
reproducibility. The leading air films, which are a result of the concentrated void fraction, maintain a
high drag reduction effect by air lubrication and by suppression of Reynolds shear stress events in the
turbulent vortical structures beneath the bubble swarm. The latter effect of RBI in particular plays a
significant role at higher Reynolds numbers. Based on the combination of these effects, we confirmed
that RBI provides an extra drag reduction effect when compared with continuous bubble injection.