chemical engineering research and design 8 7 ( 2 0 0 9 ) 298306

Contents lists available at ScienceDirect

Chemical Engineering Research and Design

journal homepage: www.elsevier.com/locate/cherd

Determination of microchannels geometric parameters

using micro-PIV
Goncalo Silva, Nuno Leal, Viriato Semiao
Mechanical Engineering Department, Instituto Superior Tecnico, Universidade Tecnica de Lisboa,
Av. Rovisco Pais, 1049-001 Lisbon, Portugal

a b s t r a c t
The precise characterization of microgeometries is a crucial task in the study of ow phenomena at these scales.
Since the use of conventional optical microscopes is somehow limited in terms of accuracy, the present work studies
the use of micro-PIV measurements to characterize a rather irregular microchannel cross-section. The micro-PIV is
employed with a spatial resolution of 23.68 m 23.68 m allowing for the location of the microchannel walls with
an accuracy, in average, 10 times higher than that provided by the use of a conventional optical microscope. The
accuracy of the micro-PIV results was validated by comparing the volumetric ow rate yielded by the integration
of the micro-PIV velocity proles against that supplied by the syringe pump employed in this experimental work.
The 3% difference revealed the good quality of the measurements, demonstrating the potential of the micro-PIV
technique to characterize both the ow kinematic parameters and the enclosing geometry.
2008 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
Keywords: Micro-PIV; Microchannels; Geometry denition; Microuidics

1.

Introduction

The study of uid dynamics at micrometric scale geometries

still presents conicting and controversial results. Despite
being nowadays established that decreasing the length scale
makes relevant the phenomena traditionally neglected in
macroscale ow theory, it is proved that the incorrect estimation of the experimental uncertainties can lead to misleading
reports of those relevant phenomena, Gad-el-Hak (1999),
Hetsroni et al. (2005), Bayraktar and Pidugu (2006) and
Kandlikar et al. (2006). A simple error propagation analysis shows that, no matter what parameter one intends to
measure, the overall experimental error increases due to the
greater inuence of the uncertainty in the geometrical characterization of the uid ow domain as the length scale
decreases (see, e.g. Hetsroni et al., 2005). Moreover, due to
uncertainties in the manufacturing process and, sometimes,
due to the lack of information from the manufacturer, the
geometry dimensions and shape are generally not known to
the desired accuracy.

The inuence of an inaccurate geometrical characterization in microuidic ow studies is not a new issue. In fact,
its inuence has repeatedly been reported as one of the
main causes for obtaining incorrect results in experimental
microuidic studies, Hetsroni et al. (2005) and Magueijo et al.
(2006).
Pressure drop measurements in fully developed ows and
its comparison against the results predicted by the conventional Fluid Mechanics theory applied to macroscale ows is
perhaps the most notorious example of a situation where
the ow geometry has a major inuence in the accuracy
of the nal results. Distinctive examples of this situation
can be found in the open literature. For instance, Guo and
Li (2002) used a 40 optical microscope to characterize the
inner diameter of a circular glass microtube obtaining a
value of 84.7 m. The inclusion of this value in the pressure
drop computation yielded values considerably higher than
those predicted by the conventional HagenPoiseuille theory.
The same measurements conducted with both a 400 optical microscope and a scanning electron microscope (SEM)

Abbreviations: CCD, charge coupled device; NA, numerical aperture; PIV, particle image velocimetry; SEM, scanning electron microscope;
SPM, scanning probe microscope.

Corresponding author. Tel.: +351 218417726; fax: +351 218475545.

E-mail address: [email protected] (V. Semiao).
Received 22 February 2008; Received in revised form 27 August 2008; Accepted 27 August 2008
0263-8762/$ see front matter 2008 The Institution of Chemical Engineers. Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.cherd.2008.08.009

chemical engineering research and design 8 7 ( 2 0 0 9 ) 298306

Nomenclature
ddiff
dp
Dh
e
M
n
U
V
X
Y
zcorr
Z

diameter of point spread function (m)

particle diameter (m)
microchannel hydraulic diameter (m)
CCD pixel size (m)
magnication
refraction index
tted velocity prole (m/s)
velocity vector (m/s)
streamwise ow direction (m)
cross-streamwise direction (width) (m)
depth of correlation, half the thickness of the
micro-PIV measurement volume (m)
cross-streamwise direction (height) (m)

Greek symbols
z
depth-of-eld of a microscope objective (m)

relative contribution of a particle displaced a

distance z from the object plane

wavelength of the emitted light (m)

revealed, for the same microtube, an inner diameter of 80 m,

value that resulted in pressure drop values in good agreement
with the theory. Another problem related to uncertainties in
the geometrical characterization of the ow boundaries is
described by Celata et al. (2006). These authors observed for
a Teon circular microtube with 304 m diameter the presence of slight discrepancies between the measured pressure
drop and the value theoretically expected. The observation
of SEM images of the microtube revealed deformations in
the assumed cross-sectional circularity shape, i.e. instead of
circular the microtube cross-section was best described by
an ellipse. The comparison of the measured pressure drop
data with the predicted HagenPoiseuille results, now for an
elliptical cross-section, showed good matching. Based on the
previous reasoning it is clear that optical microscopes and
their conventional magnications, usually below 100, may
not be particularly recommended if one needs to attain high
accuracies.
A related problem occurs when performing micro-PIV measurements. Since the micro-PIV technique is traditionally
based on the use of optical microscopes, the visual determination of ow features is somehow limited by the accuracy of
the optical system itself. To overcome this, avoiding the use
of expensive equipment like the SEM or the scanning probe
microscope (SPM), and achieving accuracies higher than those
provided by a simple optical microscope, the micro-PIV velocity data can be used. This research work intends to perform
a thorough study on the use of the micro-PIV technique to
accurately characterize the microuidic ow geometry, particularly its inner boundaries. The comparison of the results
provided by micro-PIV measurements against those obtained
through raw visual inspections with a conventional optical
microscope is also performed in order to establish the validity
of the former.

2.

Micro-PIV fundamentals

PIV is a non-intrusive optical technique used to measure the

uid motion. The motion of the uid is determined by measuring the displacement of seeding particles in two recorded

299

images separated by a known period of time. Each image

is divided into a certain number of cells, named interrogation areas, each of these containing a minimum amount
of particles. For traditional cross-correlation algorithms, a
minimum of ve particles per interrogation area is usually
required, whereas for more advanced correlation algorithms
(e.g. ensemble time average algorithms) this requirement
may not be necessary (Meinhart et al., 2000). The mean
displacement of the amount of particles contained in each
interrogation area is determined through a statistical correlation between the two images recorded.
Micro-PIV is a technique that is based on the PIV concept. Yet, micro-PIV is applied to perform measurements
in geometries with much smaller dimensions where much
higher resolutions are required. Wereley et al. (2002) outlined
three main aspects that differentiate micro-PIV from the traditional PIV: the ow eld is volume illuminated instead of
being illuminated by a laser light sheet, producing higher levels of background noise from out-of-focus particles if their
density is too high; if particles are smaller than the laser light
wavelength, which is needed in some studies where spatial
resolutions of a few microns are required, light scattering techniques cannot be employed; Brownian motion of particles with
diameters smaller than 1 m may compromise the evaluation of the ow velocity since that introduces an additional
percentage of uncertainty in the nal result.
The rst work where the PIV technique was successfully adapted for studying microscopic ows was carried out
by Santiago et al. (1998). A pressure-driven Hele-Shaw with
a cylindrical obstruction in the centre of a 120 m2 ow
was studied and results were obtained with 3.45 m vectorto-vector distances with 6.9 m 6.9 m spatial resolution.
Later, Meinhart et al. (1999) measured the velocity eld of
a 30 m 300 m glass channel with a spatial resolution of
0.9 m 13.6 m in the wall normal and streamwise ow
directions, respectively.
Stone et al. (2002) used the micro-PIV technique applied to
the study of ows in microchannels to determine the position
of the microchannel walls within a spatial resolution of tens of
nanometers for a single horizontal plane. The high accuracy
of their measurements was obtained by using interrogation
volumes of 0.9 m 13.6 m 1.8 m that already account for
the existence of 50% overlap in the spanwise and streamwise
directions. These authors were the rst to propose the experimental technique discussed and developed in this work as an
alternative to the application of specic microscopes that, as
mentioned before, possess limitations for measurements in
liquid lled microchannels. Such technique is applied herein
at 61 horizontal measurement planes to reconstruct the entire
microchannel cross-section.

3.

Experimental apparatus

In
this
work
a
microchannel
manufactured
by
DantecDynamics was used, which is shown in Fig. 1.
This microchannel is manufactured by laser ablation in a
polymeric chip (PMMA) and sealed with the same material.
This ensemble allows for micro-PIV measurements with
reduced optical aberrations. The study was performed in the
rectilinear region of the microchannel, denoted in Fig. 1 by
the zone enclosed by the elliptical line.
The pressure required for the uid to ow was provided
by a syringe pump (NE-1000, New Era Pump Systems, Inc. ).
Error tests performed to this pump indicated the uncertainty

300

chemical engineering research and design 8 7 ( 2 0 0 9 ) 298306

Table 1 Maximum spatial resolution of commonly used

air-immersion lenses

Fig. 1 The studied microchannel integrated in a chip

platform supplied by DantecDynamics .
of 1.5 l/min. The uid used in all experiments was deionised
water.
A micro-PIV system was used to study the uid ow
inside the above microchannel. A pulsed Nd:YAG laser New
Wave Solo II-15 ( = 532 nm) was used to illuminate the measurement section of the microchannel. Fluorescent nile red
ow-tracing particles with 1 m of average diameter were
added to deionised water in a proportion of 0.5 l (of a solution
of 3.64 1010 particles/ml) to 1 ml of water. An epi-uorescent
lter was used to prevent the background noise having wavelength below the red light spectrum ( = 575 nm) to reach the
Flowsense 2M CCD (charge coupled device) camera. This
camera has a 1600 1186 pixel2 resolution (pixel size = 7.4 m)
with 8/10-bit intensity resolution. Time delays ranging from
100 s to 500 s between frames and a repetition rate of 15 Hz
were used. In order to obtain magnied particle images the
CCD camera was coupled to a Leica DM ILM microscope with
an air immersion lens of magnication M = 10 and numerical aperture of NA = 0.25. A schematic representation of the
micro-PIV system used in this work is presented in Fig. 2.

4.

Measurement resolutions

4.1.

Optical microscope resolution

The images of the microchannel geometry are captured by a

CCD camera connected to an optical microscope. Therefore,
in a micro-PIV setup, the image properties are inuenced by
both the microscope and the CCD camera optical parameters.
The maximum in-plane spatial resolution of any optical
system is limited by light diffraction, a phenomenon consist-

NA

ddiff (m)

Maximum spatial
resolution (m)

5
10
20
40
63
100

0.12
0.25
0.40
0.55
0.70
0.90

54.1
51.9
64.9
94.4
116.8
144.2

27.05
25.95
32.45
47.20
58.40
72.10

ing of an interference pattern that images the object as a

central disk (the Airy disk) surrounded by concentric rings (the
Airy rings), as for reported for instance by Inou and Spring
(1997) and Raffel et al. (1998). It can be demonstrated, using the
so-called Fraunhofer approximation (Adrian and Yao, 1985),
that the diameter of the Airy disk in the image plane, ddiff ,
is expressed by Eq. (1), where M is the lens magnication,  is
the wavelength of the emitted light and NA the lens numerical
aperture:
ddiff = 2.44 M


2 NA

(1)

The resolution of an optical system can thus be dened as

0.5ddiff , i.e. half the distance between two adjacent Airy disk
centres. The denition of the spatial resolution by this criterion is known as the Rayleigh criterion (Inou and Spring,
1997).
Taking into account that the image in a CCD camera is
formed by an array of pixels, the pixel size emerges as another
limiting parameter to the maximum optical resolution since
no geometry can be resolved below the pixel dimension.
However, considering the pixel size of a CCD camera, which
is usually of the order of 10 m or less, and the optical characteristics of commonly used air-immersion objectives, the
maximum in-plane spatial resolution in a micro-PIV is limited by the light diffraction phenomenon. Table 1 illustrates
several diffraction and resolution values for commonly used
air-immersion lenses.
Besides the maximum in-plane spatial resolution one also
needs to take into consideration the out-of-plane resolution
limit. The out-of-plane resolution is given by the depth-of-eld
of the optical system, Z , which can be viewed as an estimate
of the distance that a microscope slide may be moved in the

Fig. 2 Scheme of the DantecDynamics assembled micro-PIV equipment.

chemical engineering research and design 8 7 ( 2 0 0 9 ) 298306

4.2.

Table 2 Depth-of-eld of commonly used

air-immersion lenses.
M

NA

z (m)

5
10
20
40
63
100

0.12
0.25
0.40
0.55
0.70
0.90

49.3
11.5
4.3
2.1
1.3
0.7

vertical direction while still maintaining focus on an innitely

thin specimen. According to Inou and Spring (1997), the
depth-of-eld of a microscope objective is given by the sum
of the focal depth due to diffraction, the rst term on right
hand side of Eq. (2), and the focal depth due to geometrical
effects, the second term on right hand side of Eq. (2):
z =

n0
NA2

ne
NA M

(2)

In the previous equation, n is the refractive index of the

imaging medium, 0 is the wavelength of the imaged light in
vacuum and e is the smallest distance that can be resolved
by the image detector placed in the microscopes image
plane, which corresponds to the CCD pixel size in the present
case. Table 2 illustrates the out-of-plane resolution achieved
by commonly used air-immersion lenses with n = 1.00,
0 = 532 nm and e = 7.4 m.

301

Micro-PIV resolution

As previously stated, in the micro-PIV technique the image

is divided into a certain number of cells named interrogation areas. To make full use of the information provided by
the images, the interrogation areas are generally overlapped.
As a result, and although each velocity vector represents the
average velocity of the ow inside the interrogation area,
the distance between neighbouring velocity vectors is greatly
reduced due to the overlapping process. The distance between
neighbouring velocity vectors is considered here to dene the
in-plane spatial resolution of the micro-PIV measurements.
The velocity vector yielded by the micro-PIV technique
characterizes the average motion of the ow or, more precisely, the average motion of the total amount of particles
in that region within the interrogation area. Excluding the
presence of severe velocity gradients, which may signicantly
deteriorate the correlation peak biasing the velocity measurement (Westerweel, 2008), as long as all particles stay within
the interrogation area, the computed velocity vector is always
an accurate measure of the mean ow velocity within the
interrogation area. However, by construction, the micro-PIV
technique always places the measured velocity vector in the
centre of the interrogation area. This can lead to signicant
bias errors in the velocity representation, even in the case of
minor velocity gradients. This error exists since one is assuming that the mean velocity is also located at the interrogation
area centre. However, this assumption is only true when there
is no velocity gradient or, when the velocity gradient exists but

Fig. 3 Velocity prole within an interrogation area and corresponding histogram of velocities: (a) linear velocity prole, (b)
histogram of velocities for a linear velocity prole, (c) parabolic velocity prole and (d) histogram of velocities for a parabolic
velocity prole.

302

chemical engineering research and design 8 7 ( 2 0 0 9 ) 298306

without compromising the overall results quality in terms of

valid/rejected vectors, this spatial resolution was intentionally
chosen to demonstrate that, provided the conditions previously discussed are veried, the method studied herein is
applicable.
Regarding the spatial resolution in the out-of-plane direction, since the ow eld in micro-PIV is volume illuminated,
this parameter is not just simply given by the lens depthof-eld. This is so because there are particles that still emit
enough light to contribute to the correlation signal despite
being out-of-focus. Therefore, the resolution in the vertical
direction, named by Olsen and Adrian (2000) as depth-ofcorrelation, is dened as twice the distance that a particle of
diameter dp can be positioned from the object plane so that
the intensity along the optical axis is an arbitrarily specied
fraction of its focused intensity, . Beyond this distance, the
particles intensity is sufciently low and does not inuence
the velocity measurement. For this reason, the depth-ofcorrelation, 2Zcorr , can be viewed as the limiting out-of-plane
spatial resolution of a micro-PIV measurement. According to
the study of Olsen and Adrian (2000) the depth-of-correlation
is given by the following equation:

it is constant, i.e. when the velocity changes linearly within

the interrogation area, see Fig. 3a and the corresponding histogram, Fig. 3b. For non-linear velocity variations, like those
sketched in Fig. 3c, the bias velocity error is present and is
represented in Fig. 3d. Consequently, if one assumes that the
presence of a velocity gradient does not lead to out-of-pair
particles, the error produced by such gradient is only relevant
when the interrogation area is so large that the inaccuracy
resulting from the linear velocity assumption within the area
is signicant. Consequently, if the real velocity prole can be
represented, without signicant loss of accuracy, by piecewise
linear functions (where the length of each section corresponds
to the interrogation area length) the assumption of negligible
velocity gradients is valid.
This analysis leads to the conclusion that small interrogation areas provide the most accurate results since they
minimize this error. However, considering the fact that the
signal strength in the velocity computation process is proportional to the number of particle pairs within the interrogation
area (Keane and Adrian, 1990; Westerweel, 1998), if one
keeps the particles density constant it immediately follows
that the interrogation areas size cannot be decreased below


2zcorr = 2


1

a certain dimension. For that reason, the interrogation area

size must be optimally chosen so that the maximal signal
strength is achieved with the less velocity gradient error contribution possible. A rule-of-thumb generally followed by the
PIV community consists in choosing interrogation areas sizes
that simultaneously ensures that, at least, ve particle pairs
are present (Keane and Adrian, 1990) and that the particles displacement does not exceed 1/4 of the interrogation area length
(Keane and Adrian, 1990; Westerweel, 1998; Raffel et al., 1998).
If the velocity gradient contribution can be neglected or, at
least, if the velocity vector location within the interrogation
area is not too inaccurate, the use of sub-pixel interpolation
schemes (Willert and Gharib, 1991; Prasad et al., 1992) allows
one to estimate the average velocity value inside the interrogation area to sub-pixel accuracy. In fact, as demonstrated
by Prasad et al. (1992), as long as particles are resolved by
23 pixels, the accuracy in locating the maximum peak in the
correlation plane, i.e. the precision in determining the velocity vector within the interrogation area is approximately 0.1
pixels. Therefore, even though the micro-PIV in-plane spatial
resolution is usually of the order of several pixels, the kinematic information supplied by this technique, which is given
by the mean velocity vector inside the interrogation area, is
of the order of 0.1 pixels. If the information provided by the
kinematic eld can be somehow related to the determination
of some spatial feature in the ow eld, one will be able to
spatially characterize that feature with accuracy higher than
that provided by a conventional optical microscopy technique.
Therefore, in order to achieve an in-plane spatial resolution of
less than one pixel size, while ensuring that this resolution
provides an accurate measure of the ow eld, the presence
of velocity gradients within the interrogation areas cannot
be signicant. In this study, and because the ow is fully
developed, as described below, the in-plane spatial resolution
was chosen to have a value of 23.68 m 23.68 m, with 50%
overlap, in a eld of view of 1172 m 887 m. Although the
in-plane spatial resolution could be signicantly decreased

dp 2 (n/NA) 1
4

2  1/2

5.95(M + 1) 2 (n/NA) 1
4M2

(3)

Since the depth-of-correlation value is always larger than

the depth-of-eld value of the used lens, the limiting outof-plane resolution of a micro-PIV measurement is always
larger than that of a conventional microscope measurement.
Table 3 illustrates common values of depth-of-correlations
with air-immersion lenses assuming dp = 1 m and = 0.01 and
absolute differences to corresponding depth-of-eld values,
z = zcorr z . As it can be seen in Table 3 the micro-PIV spatial resolutions in the out-of-plane direction are considerably
larger than those provided by conventional microscopy measurements. This difference decreases signicantly as the lens
magnication and numerical aperture increase. To overcome
this problem the scientic community has associated the
micro-PIV technique with more advanced microscopy techniques. The use of laser scanning confocal microscopy (LSCM)
is one of the most recognized examples.
The use of such microscopes allows one to achieve high
out-of-plane resolutions without compromising the in-plane
spatial resolution, i.e. without having to use high magnications. For instance, Kinoshita et al. (2005), using a micro-PIV
system with a confocal microscope, performed velocity measurements in moving droplets inside microchannels with a
eld of view of 228 m 171 m and a confocal depth of
1.8 m. The major drawbacks of using the confocal technique

Table 3 Depth-of-correlation of commonly used

air-immersion lenses assuming dp = 1 m and = 0.01
and absolute differences to corresponding depth-of-eld
values
M

NA

2Zcorr (m)

5
10
20
40
63
100

0.12
0.25
0.40
0.55
0.70
0.90

161.9
34.2
12.8
6.5
3.7
1.5

z (m)
31.65
5.60
2.10
1.15
0.55
0.05

chemical engineering research and design 8 7 ( 2 0 0 9 ) 298306

are the considerably higher cost of the equipment required

and the relatively lower state-of-development, when compared to the conventional micro-PIV technique. This fact is
denoted by the limited range of applications that can use confocal micro-PIV with success at present (Lima et al., 2006).
Despite of these drawbacks, the virtues of confocal micro-PIV
are undeniable, particularly in situations where measurements with high out-of-plane resolutions are crucial.
Similarly to the analysis performed for the in-plane spatial
resolution, the out-of-plane resolution size, i.e. the thickness
of the measurements control volume, can be considered adequate as long as the displacement of the particles in the
out-of-plane direction does not exceed 1/4 of the control volume thickness and also that, within the interrogation area,
severe velocity gradients in this out-of-plane direction do not
occur. The lens used in this work has M = 10 and NA = 0.25,
which corresponds to a depth-of-correlation value of 34.2 m
(see Table 3). Therefore, in order to perform the reconstruction
of the three dimensional velocity eld with a vertical overlap approximately equal to the horizontal one, i.e. 50%, the
horizontal planes were measured each 15 m in the vertical
direction. This vertical motion is controlled by a stepper motor
with reproducibility better than 1 m and a total precision of
3 m that is controlled by a real time RISC processor. The
stepper motor can be addressed with a resolution of 1.5 nm.
The inuence of the chosen in-plane and out-of-plane resolution values in the accurate description of the velocity eld
is performed below.

303

Fig. 5 Measured velocity prole at X = 200 m of a

horizontal measurement plane at Z = 500 m.

5.
Estimation of the micro-PIV velocity
uncertainty and validity of the ow fully
developed condition
The a priori estimation of the error associated with the velocity measurement with micro-PIV is a complex task since
each micro-PIV realization comprises the participation of several distinctive parameters whose interaction, in practice, is
almost impossible to quantify in detail (a deeper analysis of
some of the most common PIV error sources can be found, for
instance, in Raffel et al., 1998 or Wereley and Meinhart, 2004).
Consequently, and in order to have an order of magnitude of
the micro-PIV error present in the overall velocity measurements, an approach similar to one described in Silva et al.
(2008) is adopted here. With this approach several measurement planes of the velocity eld, as the one depicted in Fig. 4,

Fig. 4 Example of an image of the studied microchannel

at a horizontal plane Z = 200 m with measured velocity
vectors.

Fig. 6 Three-dimensional velocity prole built up from 61

two-dimensional horizontal velocity proles, similar to that
depicted in Fig. 5.
are extracted along the microchannel height. In each plane
Z = const, and for a constant streamwise location, X = 200 m
in the present case, the velocity prole is computed. Fig. 5
shows an example of such velocity prole at Z = 500 m and
X = 200 m. This procedure, applied to all the 61 horizontal
measurement planes, allows for the reconstruction of the
overall three-dimensional velocity prole as depicted in Fig. 6.
The volumetric ow rate given by the volume enclosed by
this three-dimensional velocity prole, which is computed
through numerical integration, is compared against the corresponding ow rate supplied by the syringe pump, which
is considered to provide a theoretical value. The difference
between ow rates provides an indication of the bias error
of the micro-PIV technique in the case under study. For the
volumetric ow rate used herein, which is ca. 1 ml/min, the
difference between the micro-PIV predicted value and the one
supplied by the syringe pump, is ca. 3%. This indicates that,
in average, the velocity information provided by the microPIV technique is expected to have an accuracy of ca. 3%. Since
the micro-PIV data used in this analysis was obtained with
the in-plane and out-of-plane resolutions described in Section 4.2, the accuracy of these resolutions is also denoted by

304

chemical engineering research and design 8 7 ( 2 0 0 9 ) 298306

ca. 3% error. It should be stressed that the computation of the

velocity eld was performed with an ensemble time average
correlation algorithm, as described by Meinhart et al. (2000)
and Delnoij et al. (1999), in order to obtain a velocity eld
with the less unphysical velocity vectors. These unphysical
vectors can be caused by the presence of both a low signalto-noise ratio in the correlation plane, caused by the required
low seeding concentration, and the inuence of the Brownian motion on the computation of the instantaneous velocity
vectors, Devasenathipathy et al. (2003). It should be remarked
that the use of this method presumes the presence of a steady
ow, condition that is here ensured by the syringe pump.
As far as the fully developed ow condition in the streamwise ow direction is concerned, this condition was veried
by comparing the momentum ow rate imbalance at three
distinctive streamwise locations (X = 200 m, X = 400 m and
X = 700 m) against the value here considered to represent
the accuracy of the micro-PIV velocity measurements. The
momentum ow rate imbalance value was of the order of 3.4%,
whereas the corresponding volumetric ow rate imbalance
was 1.5%. Therefore, since the momentum ow rate imbalance
value is affected by the volumetric ow rate one, its intrinsic
value is considered to be less than 3%, i.e. within the experimental uncertainty. As a result, the microuidic ow in the
channel section under study is fully developed within a ca. 3%
uncertainty.

6.
Microchannel cross-section
characterization using micro-PIV
As pointed out in Section 4.2, one can ensure that the velocity eld is measured with an accuracy of the order of the
sub-pixel size under certain conditions. If the advantage of
this precise velocity information could be used to characterize some particular spatial features, as for instance the solid
boundaries, accuracies much higher than those provided by
the direct visual measurements would be achieved. Taking
into account that through an indirect approach, the kinematic boundary condition at a solid wall requires that the
velocity eld at the uid/solid interface should not exhibit
discontinuities, provided that the uid can be modelled as
a continuum (Gad-el-Hak, 1999) and the wall material is
hydrophilic (Tretheway and Meinhart, 2002), the knowledge of
the velocity eld allows one to identify the presence of a wall
as the region where the velocity is nil. However, because in
the walls vicinity, micro-PIV measurements have limited accuracy, it is extremely difcult to directly measure the regions
of nil velocity. In fact, very close to a solid interface, particles tend not to follow exactly the uid motion since physical
and chemical interactions between them and the walls start to
take place. Moreover, due to the proximity of a medium with
a different refractive index, i.e. the wall region corresponds to
a liquid/solid interface, the process of imaging the particles
motion close to the wall is also affected by optical phenomena, which are perceived mainly as severe optical distortions.
To overcome these problems, it is preferable to use data from
regions where the accuracy of the velocity measurements, and
so the sub-pixel accuracy, can be guaranteed. Consequently,
the velocity data measured sufciently apart from the wall is
extrapolated to regions near the wall holding the advantage
of the sub-pixel accuracy.
To achieve this objective, the velocity eld has to be
assumed to be dened by some analytical function V = V(x,y,z)
which can be known from the yielded experimental data.

Since, as it was previously reported, the velocity eld is

fully developed, the eld V = V(x,y,z) is assumed to exhibit a
near parabolic variation along the cross-streamwise directions
y and z and no variation in the streamwise ow direction x,
which is typical of fully developed laminar channel ows.
The HagenPoiseuille solution is only exact for ows inside
circular tubes and between innite parallel plates. For other
simple non-circular cross-sectional shapes analytical solutions in terms of series expansions are available for the
velocity proles (Shah and London, 1978). For more complex
cross-sectional shapes, as the one investigated in this work,
there is no analytical solution to dene the velocity prole.
In spite of that, and although the mentioned complex series
functions could be used, the use of simpler parabolic functions
generally ts the experimental data in a much more accurate
way, as observed by Stone et al. (2002). Summing up, for irregular geometries, as in the present case, where no analytical
solutions are available, the most feasible choice, as experimental evidence of Fig. 5 suggests, is to use parabolic proles
to characterize V = V(x,y,z). Hence, the experimental velocity
vectors at constant cross-streamwise ow direction are tted
to continuous parabolic functions of the type of U = Ay2 + By + C.
The success of this approximation is revealed by the correlation coefcient of the parabolic tting operation, which is
always above 90%. The roots of this analytical function are
computed and their values are associated with the location of
the inner wall region. It should be reminded that this procedure assumes a no-slip velocity condition at the wall. However,
in the case of a slip velocity this procedure remains valid provided that one knows the slip length. Since, in the present
work, the static equilibrium contact angle is below 90 (ca. 74 ),
i.e. the wall inner surface is hydrophilic, the no-slip velocity
condition at the wall is valid.
The characterization of the microchannel cross-section is
carried out by repeating the procedure discussed above for the
61 measuring planes along the microchannel height.
The results of the microchannel dimensions and crosssection are depicted in Fig. 7a and b. As it is observed,
qualitatively, both methods yield a similar microchannel
cross-section with dimensions of the same order of magnitude, the minimum/maximum width dimensions of ca.
142 m and ca. 843 m, respectively, and the height of approximately 900 m, yielding a microchannel hydraulic diameter of
Dh = 637 m.
In these measurement planes statistical variations in the
measured velocities lead to a variation in the roots of the tted parabolas. Therefore, in order to quantify the statistical
variations in the velocity results supplied by micro-PIV, 35
realizations of the same measurement at the same horizontal
plane were carried out and analysed. Since the ow is steady
and laminar, temporal variations should not exist. Hence, the
presence of any temporal statistical variations is interpreted
as irregularities in the measurements. The horizontal error
bars depicted in both Fig. 7a and b quantify this effect on the
wall location value, which is predicted by the roots of the tted
parabolas.
The horizontal error bars magnitude reveal the validity
of the assumptions previously stated regarding the higher
accuracy provided by micro-PIV. In average this accuracy is
approximately 10 times higher. This result is not surprising
if one observes a traditional microchannel image, see Fig. 4.
As it is observed the inner wall region is very difcult to
identify because of the image blurriness in this region. This
feature leads to uncertainties much higher than the pixel size.

chemical engineering research and design 8 7 ( 2 0 0 9 ) 298306

305

Fig. 7 Microchannel cross-section and corresponding errors displayed as horizontal bars yielded by: (a) optical microscope;
(b) micro-PIV measurements.
Moreover, the intrinsic subjectivity of a visual inspection also
contributes for the error depicted in Fig. 7a). As it would be
expected, considering optical arguments, the occurrence of
those facts increases with the top/bottom wall proximity.
The micro-PIV results are also affected by this top/bottom
wall proximity deterioration. In these measurement planes
statistical variations in the measured velocities lead to a variation in the roots of the tted parabolas. As a result, the
accuracy in determining the wall location is decreased. This
fact is revealed by the higher magnitude of the horizontal error
bars in these regions. Despite of that, the effect of this error is
not as severe as in the optical microscope case.
The vertical error bars are not shown since their values
are constant along the microchannel height and equal to
11.5 m for the case of the optical microscope measurements, Fig. 7a), and 17.1 m for the micro-PIV case, Fig. 7b).
This gure reveals that, even though the micro-PIV yields better accuracy in the in-plane direction, the characterization of
the microchannel cross-section along the out-of-plane direction is less precise with micro-PIV. However, the advantage of
the micro-PIV procedure discussed herein in terms of the inplane plane accuracy pays off the slight disadvantage present
in the out-of-plane direction.

7.

Conclusions

The relevance that geometrical parameters have in microscale

ows associated with the relative inaccuracy/high cost of the
experimental techniques currently available were the main
motivations to develop this study. The method discussed is
based on the use of the velocity data provided by the microPIV technique to establish the microchannel dimensions and
its cross-sectional shape. To do so, 61 velocity proles were
measured along the microchannel height, which, in combination with the no-slip velocity condition at the wall, allowed for
the reconstruction of the microchannel cross-section.
The validity of the micro-PIV kinematic data was conrmed
by comparing the volumetric ow rate supplied by the used
syringe pump against that yielded by the integration of microPIV velocity proles. The difference of 3.0% revealed the good
quality of the micro-PIV results.
The superiority of the micro-PIV technique in the accurate
reconstruction of the microchannel boundaries was con-

rmed by comparing its results against those provided by raw

visualizations using the standard optical microscope that is
part of the experimental apparatus. With this procedure, it
was veried that the use of the micro-PIV in combination with
the methodology here applied yielded results, in average, 10
times more accurate. This result demonstrates that even when
considerably large in-plane spatial resolutions are employed
with micro-PIV, 23.68 m 23.68 m as in the present study,
the use of the velocity data can still be applied with success in
the reconstruction of a microchannel cross-section allowing
for the simultaneous study of the ow eld and its boundaries.

References
Adrian, R. and Yao, C., 1985, Pulsed laser technique application to
liquid and gaseous ows and the scattering power of seed
materials. Appl Opt, 24: 4452.
Bayraktar, T. and Pidugu, S., 2006, Characterization of liquid ows
in microuidic systems. Int J Heat Mass Transf, 49:
815824.
Celata, G., Cumo, M., McPhail, S. and Zummo, G., 2006,
Characterization of uid dynamic behaviour and channel wall
effects in microtube. Int J Heat Fluid Flow, 27:
135143.
Delnoij, E., Westerweel, J., Deen, N., Kuipers, J. and van Swaaij, W.,
1999, Ensemble correlation PIV applied to bubble plumes
rising in a bubble column, In Proceedings of the 4th International
Conference GasLiquid and GasLiquidSolid Reactor Engineering
(Delft, NL)
Devasenathipathy, S., Santiago, J., Wereley, S., Meinhart, C. and
Takehara, K., 2003, Particle imaging techniques for
microfabricated uidic systems. Exp Fluids, 34: 504515.
Gad-el-Hak, M., 1999, The uid mechanics of microdevicesthe
Freeman Scholar lecture. J Fluids Eng, 121: 533.
Guo, Z. and Li, Z., 2002, Size effect on micro-scale single phase
phase ow and heat transfer, In Proceedings of the 12th
International Heat Transfer Conference
Hetsroni, G., Mosyak, A., Pogrebnyak, E. and Yarin, L.P., 2005, Fluid
ow in micro-channels. Int J Heat Mass Transf, 48:
19821998.
Inou, S. and Spring, K., (1997). Video Microscopy (Second edition).
(Plenum Press).
Kandlikar, S., Garimella, S., Li, D., Colin, S. and King, M., (2006).
Heat Transfer and Fluid Flow in Minichannels and MicroChannels
(First edition). (Elsevier).

306

chemical engineering research and design 8 7 ( 2 0 0 9 ) 298306

Keane, R. and Adrian, R., 1990, Optimization of particle image

velocimeters. Part I. Double pulsed systems. Meas Sci Technol,
1: 12021215.
Kinoshita, H., Oshima, M., Kaneda, S. and Fujii, T., 2005, Confocal
micro-PIV measurements of internal ow in a moving droplet,
In Proceeding of the 9th ICMSCLS , pp. 629631.
Lima, R., Wada, S., Tsubota, K. and Yamaguchi, T., 2006, Confocal
micro-PIV measurements of three-dimensional proles of cell
suspension ow in a square microchannel. Meas Sci Technol,
17: 797808.
Magueijo, V., Semiao, V. and de Pinho, M.N., 2006, Effects of
ultraltration permeation rates on the hydrodynamics of a
minichannel/slit laminar ow. Chem Eng Sci, 61: 71397150.
Meinhart, C., Wereley, S. and Santiago, J., 1999, PIV measurements
of a microchannel ow. Exp Fluids, 27: 414419.
Meinhart, C., Wereley, S. and Santiago, J., 2000, A PIV algorithm
for estimating time-average velocity elds. J Fluids Eng, 122:
285297.
Olsen, M. and Adrian, R., 2000, Out-of-focus effects on particle
image visibility and correlation in microscopic particle image
velocimetry. Exp Fluids, 29: 166174.
Prasad, A., Adrian, R., Landreth, C. and Offutt, P., 1992, Effect of
resolution on the speed and accuracy of particle image
velocimetry interrogation. Exp Fluids, 13: 105116.
Raffel, M., Willert, C. and Kompenhans, J., (1998). Particle Image
VelocimetryA Practical Guide. (Spinger-Verlag).

Santiago, J., Wereley, S., Meinhart, C., Beebe, D. and Adrian, R.,
1998, A particle image velocimetry system for microuidics.
Exp Fluids, 25: 316319.
Shah, R.K. and London, A.L., (1978). Laminar Flow Forced Convection
in Ducts, Volume Supp. 1 of Advances in Heat Transfer. (Academic
Press).
Silva, G., Leal, N. and Semiao, V., 2008, Micro-PIV and CFD
characterization of ows in a microchannel: velocity proles,
surface roughness and Poiseuille numbers. Int J Heat Fluid
Flow, 29: 12111220.
Stone, S., Meinhart, C. and Wereley, S., 2002, A microuidic-based
nanoscope. Exp Fluids, 33: 613619.
Tretheway, D. and Meinhart, C., 2002, Fluid slip near hydrophobic
microchannel walls. Phys Fluids, 14(3): L9L12.
Wereley, S.T. and Meinhart, C.D., (2004). Micron Resolution Particle
Image Velocimetry, Diagnostic Techniques in microuidics.
(Springer-Verlag).
Wereley, S., Gui, L. and Meinhart, C., 2002, Advanced algorithms
for microscale particle image velocimetry. AIAA J, 40:
10471055.
Westerweel, J., 1998, Fundamentals of digital particle image
velocimetry. Meas Sci Technol, 8: 13791392.
Westerweel, J., 2008, On velocity gradients in PIV interrogation.
Exp Fluids, 44: 831842.
Willert, C. and Gharib, M., 1991, Digital particle image
velocimetry. Exp Fluids, 10: 181193.