Contents lists available at ScienceDirect
Experimental Thermal and Fluid Science
journal homepage: www.else vier.com/locate/etfs
A critical review of the measurement techniques for the analysis of gas microﬂows through microchannels
Gian Luca Morini ^{⇑} , Yahui Yang, Habib Chalabi, Marco Lorenzini
Dipartimento di Ingegneria Energetica, Nucleare e del Controllo Ambientale, Alma Mater Studiorum, Università di Bologna, Viale Risorgimento 2, I 40136 Bologna, Italy
article info
Article history:
Received 12 November 2010 Received in revised form 21 February 2011 Accepted 21 February 2011 Available online 25 February 2011
Keywords:
Microﬂuidics Microconvection Friction factors in microchannels Rareﬁed gases Compressible gases
abstract
In Microﬂuidics, a large deviation in the published experimental data on the dynamic and thermal behav ior of microﬂows has been observed with respect to the classical theory but, from a chronological analysis of these experimental results, it can be realized how the deviations in the behavior of ﬂuid ﬂows through microchannels from that through largesized channels are decreasing. Today, it seems to be clear that some of the inconsistencies in the data were originated from the experimental methods used for the investigation of convective microﬂows. This fact highlights the need for the development of speciﬁc mea surement techniques for Microﬂuidics. In this work, we explore and categorize different approaches found in literature for measuring microﬂow characteristics, especially for gas ﬂows, and the geometry of the microchannels pointing out the advantages and disadvantages inherent to each experimental tech nique. Starting from the operative deﬁnition of friction factor, the main parameters that must be checked in an experimental work in order to characterize the ﬂow are reviewed. A discussion based on uncer tainty analysis will be presented in order to individuate the main operative parameters that one must be able to measure accurately to determine pressure drop in the microchannels with a low level of uncer tainty. In the paper each measurement technique is critically analysed to evidence the important issues which may have been overlooked in previous researches. The main goal of this study is to give a summary of experimental procedure and a useful guideline for experimental research in Microﬂuidics. 2011 Elsevier Inc. All rights reserved.
1. Introduction
During the last 10 years a rapid development of new microﬂow devices (MFD) in several scientiﬁc ﬁelds has taken place. Nowa days, the manufacture of MFDs, like micropumps, microvalves, microcold plates, microheat exchangers, and other microcompo nents and sensors used in chemical analysis, in biomedical diag nostics or in ﬂow measurements, is a consolidated reality. The design of new MFDs requires a deep knowledge of the ﬂuid dynamic and heat transfer phenomena within microchannels in which a liquid or gas ﬂows. For this reason, many experimental studies have been con ducted in order to analyze the behavior of convection through microchannels, of which a review is given in [1–3] . In particular, the main goal of these studies was to determine the friction factors and the convective heat transfer coefﬁcients through microchan nels in which a pressuredriven ﬂow was imposed. These experimental results have been used in order to verify if the laws governing transport phenomena within channels of mac
⇑ Corresponding author. Tel.: +39 051 2093381; fax: +39 051 2093296. Email address: gianluca.morini3@unibo.it (G.L. Morini).
08941777/$  see front matter 2011 Elsevier Inc. All rights reserved. doi: 10.1016/j.expthermﬂusci.2011.02.005
roscopic dimensions still hold at the microscale, and, if not, which new effects must be taken into account at the microscale. A large scatter in published experimental data and inconsisten cies have been observed with respect to the classical theory but, from a chronological analysis of these results, it is possible to extrapolate how the deviations between the behavior of ﬂuids through microchannels and through largesized channels are decreasing. The last experimental works in Microﬂuidics seem to highlight that some of the observed discrepancies in the data were originated from the experimental methods used for the investiga tion of convective microﬂows. In fact, in the last years a dramatic improvement of the tech niques of microfabrication has enabled a more accurate control over the geometry of microchannels and innovative and more accurate measurement techniques for microﬂows have been pro posed with a general improvement of the reliability/accuracy of the experimental data reported in the literature: these latest data seem to be in agreement with the classical theory. This highlights the need for the development of speciﬁc mea surement techniques for the Microﬂuidics ﬁeld or a reﬁnement and adaption of the ones used at larger scales. In this work, we explore and categorize different approaches found in literature for measuring microﬂow characteristics – especially for gas ﬂows – and the geometry of the microchannels
850
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
Nomenclature
d 
inner diameter of circular microtube (m) 
Greek symbols 

D _{h} 
hydraulic diameter of channel, 4 X / C (m) 
a 
coefﬁcient deﬁned in Table 3 
f 
friction factor 
b 
coefﬁcient deﬁned in Table 3 
h 
height of rectangular microchannel (m) 
c 
speciﬁc heat ratio 
t 
time (s) 
C 
wetted perimeter (m) 
L 
channel length (m) 
k 
gas mean free path (m) 
K 
pressure loss coefﬁcient 
l 
dynamic viscosity (Pa s) 
Kn 
Knudsen number, k/ D _{h} 
q 
gas density (kg/m ^{3} ) 
_ 
mass ﬂow rate (kg/s) 
coefﬁcient related to channel geometry in Eq. (12) 

m 
r 

Ma 
Mach number 
s 
viscous stress (Pa) 
p 
pressure (Pa) 
X 
area of crosssection (m ^{2} ) 
R 
speciﬁc gas constant (J/(kgK)) 

Re 
Reynolds number 
Subscripts 

T 
temperature (K) 
cr 
critical value 
v 
average axial velocity (m/s) 
in 
inlet value 
V 
volume (m ^{3} ) 
out 
outlet value 
w 
width of rectangular microchannel (m) 
w 
pertains to the channel wall 
W 
mean ﬂow velocity (m/s) 
z 
axial direction 
by examining the advantages and disadvantages inherent to each of them. Starting from the operative deﬁnition of friction factor, the main parameters that must be checked in an experimental campaign in order to characterize the ﬂow are highlighted. A discussion based on uncertainty analysis is presented in order to pinpoint the main operative parameters that must be measured accurately to determine the pressure drop in the microchannels with a low level of uncertainty.
2. Pressure drop and friction factor
The behavior of isothermal pressuredriven single phase ﬂuid ﬂows in microchannels can be studied by determining the velocity distribution in the ﬂuid region with the help of the mass conserva tion principle (continuity equation) and the equations of conserva tion of ﬂuid momentum (Navier–Stokes equation). Consider a ﬂow through a straight microchannel having an axi ally unchanging and uniform crosssection with an area equal to _{X} and a wetted perimeter equal to C ; it is possible to deﬁne the hydraulic diameter of the microchannel as:
D _{h} ¼ ^{4}^{X}
C
ð
1Þ
and the mean ﬂuid velocity as the integrated average axial velocity over the ﬂow area _{X} :
W ¼
1
X Z X
v _{z} dX ¼
_
m
qX ^{¼}
RT m
_
X
p
ð
2Þ
In Eq. (2) the relationship between the average axial velocity
and the measurable quantities, like mass ﬂow rate m , density _{q} and crosssection area X or pressure p and temperature N for gases is highlighted ( R is the speciﬁc gas constant). It is important to highlight that Eq. (2) has been written for an ideal gas: this means that we have restricted our analysis to pressures below the critical one ( p _{c}_{r} ) and/or to temperature larger than two times the critical temperature ( T _{c}_{r} ) for the gas considered. This hypothesis holds in Microﬂuidics because the gas temperature is generally larger than the critical one. During experiments in microchannels it is difﬁcult to directly derive the mean ﬂuid velocity along the channel from the crosssectional velocity proﬁle, especially when the micro channel is very small and without optical access which can enable the use of nonintrusive speciﬁc optical techniques developed with
_
this aim as the microParticle Image Velocimetry ( _{l} PIV) [4] . The limitations of velocimetry techniques for gas ﬂows in microchan nels will be discussed in the next sections. Therefore, the mean velocity of ﬂow is often calculated measuring the microchannel crosssectional area, the ﬂuid density and the mass ﬂow rate, as per the RHS of Eq. (2) . The pressure drop D p along the channel and the viscous stresses at the channel walls can be related to the mean ﬂuid velocity by means of the following expressions:
js _{w} j ¼ f _{F} ^{q}^{W} 2 ! Dp ¼ f _{D}
2
L
qW ^{2}
D
h
2
¼
ð4f _{F} Þ
L
qW ^{2}
D
h
2
ð
3Þ
where L is the microchannel length. These equations are used as deﬁnition for the Fanning friction factor ( f _{F} ) and the Darcy friction factor ( f _{D} = 4f _{F} ). Eq. (3) evidences that the pressure drop along a microchannel can be predicted if one knows the value of the friction factor. For an incompressible, isothermal ﬂow the Darcy friction factor can be operatively deﬁned as:
f _{D} ¼ ^{D} ^{h}
L
2q
_
m
X
_{2}
ð p _{i}_{n} p _{o}_{u}_{t} Þ
ð 4Þ
where p _{i}_{n} and p _{o}_{u}_{t} are the pressure at the channel inlet and outlet respectively. For isothermal compressible gas ﬂows, using the state equation
of an ideal gas, the expression of friction factor can be rewritten as
follows:
f _{D} ¼ ^{D} ^{h}
L
p
2
in ^{} ^{p} out ^{2}
RT
_
m
X
2
2 ln
^{p}
in
^{p}
out
!
ð 5Þ
where T is the gas temperature, and R the gas constant respectively. Since in microchannels the pressure drop along the tube length can be much larger than for conventionalsize tubes, the effects of gas acceleration cannot be neglected in general, even for very low values of the Mach numbers as observed theoretically and experi mentally by several authors [1,3,5]. The gas acceleration leads to the velocity proﬁle change not only in magnitude but also in shape; the magnitude increment produces an additional pressure drop while the continuous variation in shape of the velocity proﬁle means no fully or locally fully developed ﬂow occurs. In heated microchannels, since the ﬂuid radial temperature proﬁle is
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
851
strongly dependent on the radial velocity proﬁle no fully developed temperature proﬁles will occur as long as the ﬂow is developing. In tubes having an inner diameter less than 500 _{l} m and with a large lengthtodiameter ratio (long tube) high pressure ratios be tween inﬂow and outﬂow and high heat transfer ratios produce a signiﬁcant increase of the Mach number along the tube and high Mach numbers yield higher friction factors and heat transfer rates. In adiabatic and in heated microchannels, the compressibility ef fects can be monitored by using the diametertolenght ratio ( d/ L ) of the microtube and the outlet Mach number deﬁned as follows:
Ma out ¼
^{4}
_
m
^{p}^{d} 2 ^{p} out
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
s
RT out
c
ð
6Þ
where c is the speciﬁc heat ratio of the gas considered and T _{o}_{u}_{t} and p _{o}_{u}_{t} are the gas bulk temperature and pressure at the outlet of microchannel. The conventional theory states that if the average value of the Mach number ( Ma _{a}_{v}_{g} ) along a channel is less than 0.3 the ﬂow can be modeled locally as incompressible. In addition, when the pressure drop along the tube length is lower than 5% of the inlet pressure, the effect of the acceleration of the gas ﬂow in the axial direction can be neglected (in other words, the variation of the gas density along the channel can be neglected):
(
Ma _{a}_{v} _{g} > 0: 3
^{p} in ^{} ^{p} out
^{p} in
> 0: 05
ð 7Þ
When the ﬁrst inequality is satisﬁed, the gas ﬂow is not locally incompressible and the momentum and energy equation have to
be solved as coupled. When the second inequality is satisﬁed, even
if the gas ﬂow can be locally modeled as incompressible, the den
sity variation along the tube cannot be ignored. In this case, the ef fects due to the gas acceleration along the tube become important, even if the Mach number is low. The gas acceleration losses are ta ken into account in the evaluation of the friction factor through the second term of the righthand side of Eq. (5) . For microchannels, it has been demonstrated by Yang et al. [6] that the friction factor is inﬂuenced by the value of the Mach number at the outlet (Eq. (6) ); when Ma _{o}_{u}_{t} is larger than 0.3 a deviation of the friction factor from the prediction of the conventional theory is evidenced. For this rea son, the two constraints of Eq. (7) (with Ma _{o}_{u}_{t} instead of Ma _{a}_{v}_{g} )
have always to be veriﬁed when an experimental determination of the friction factor in a microchannel for a gas ﬂow is made in or der to know whether compressibility effects can be considered negligible. In Fig. 1 the trend of the Darcy friction factor as a function of the Reynolds number for a nitrogen ﬂow through a microtube having a circular crosssection, with an inner diameter equal to 172 _{l} m and
a length of 10 cm is shown [6] . The error bars related to the exper
imental friction factors are not shown in order to improve readabil ity; the maximum relative uncertainty on the friction factor is equal to ±10% as indicated in [6] . It is possible to see that the agree
ment between the experimental data and the Poiseuille law ( f _{D} = 64/ Re ) is very good in the laminar regime, at least up to Re equal to 1200. For Re > 1200 it is evident a departure of the friction factors from the Poiseuille law; this behavior can be explained by observing the value taken by the outlet Mach number ( Ma _{o}_{u}_{t} ) in this case, as it becomes larger than 0.3. This result is typical for microtubes having an inner diameter smaller than 500 _{l} m. In the laminar regime, the deviation from the Poiseuille law can be theoretically explained by solving the Navier Stokes equation for compressible ﬂows [5,7] . When the pressure ratio increases, the outlet Mach number increases, especially for short microtubes, and the friction factor increases with the Mach number. Asako
Fig. 1. Darcy friction factors for nitrogen ﬂow through a microtube in adiabatic conditions as a function of the Reynolds number (from Yang et al. [6] ).
et al. [7] have calculated numerically how the friction factor de pends on the Mach number in the laminar regime and proposed the following correlation:
2
f _{D} ¼ ^{6}^{4} þ 2:703 ^{M}^{a} þ 93: 89 ^{M}^{a} _{R}_{e}
Re
Re
ð 8Þ
As can be noticed from Eq. (4), for incompressible ﬂows the knowl edge of the differential pressure between the microchannel inlet and outlet is required only in order to determine the friction factor. On the contrary, Eq. (5) shows that for a compressible ﬂow even the absolute value of the gas pressure at the inlet or at the outlet is needed. This aspect underlines that the experimental strategies for the determination of the friction factor must be different whether compressibility effects are negligible or not. The local measurement of pressure is problematic for micro channels: drilling pressure taps along the channel can be challeng ing due to the small dimensions involved. For this reason, in the most accurate experimental works, special microchannels are de signed with customized pressure taps along their length (Bavière et al. [8] ). On the other hand, when commercial microtubes or standard microdevices are used in the tests, the measurement of the pressure along the microchannel becomes impossible. In these cases, pressure measurements are usually performed by connect ing with appropriate ﬁttings both sides of the microchannel to a larger reservoir, where the pressure taps can be drilled without difﬁculties. However, in this way the pressure measured is not that at the inlet and outlet, which is associated with the change of section. These nonnegligible minor losses ( _{D} p _{i}_{n} , _{D} p _{o}_{u}_{t} ) are due to the abrupt change in ﬂow area between the manifolds and the channel. The total pressure drop, which is the measured pressure difference between the reservoirs, can be expressed as follows:
Dp _{t}_{o}_{t}_{a}_{l} ¼ Dp _{i}_{n} þ ðp _{i}_{n} p _{o}_{u}_{t} Þ þ Dp _{o}_{u}_{t}
ð 9Þ
For perfect gas ﬂows, the minor losses at the inlet and outlet can be determined as follows:
Dp _{i}_{n} ¼
K in
2
_
m
X
2
RT
^{p} in
;
Dp out ¼
K out
2
_
m
2
RT
^{p}
X
out
ð
10Þ
For conventionalsized tubes the values of K _{i}_{n} and K _{o}_{u}_{t} have been established for a number of simple entrance and exit geometries and are quoted in many textbooks (e.g. Idelchik [9]). These coefﬁ cients have been evaluated under the assumption of essentially uni form velocity at the inlet but considering a fully established velocity
852
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
proﬁle in the channels (that means L D _{h} ). These assumptions can be considered veriﬁed for microchannels, where L > 100 D _{h} as a rule. In applying the data related to the values of K _{i}_{n} and K _{o}_{u}_{t} it should be remembered that these coefﬁcients already include the pressure change associated with the change in the velocity proﬁle and depend on the geometry of the ends and of the channels. In addition, for the estimation of the minor losses it is mandatory to know the absolute value taken by the pressure at the inlet or at the outlet, as evidenced by Eq. (10) . In Fig. 2 the minor losses at the inlet ( _{D} p _{i}_{n} ) and at the outlet ( D p _{o}_{u}_{t} ) are compared with the value of the net pressure drop ( D p _{n} = p _{i}_{n} p _{o}_{u}_{t} ) for two microtubes (5 cm and 50 cm long) with an internal diameter equal to 100 _{l} m [10] when a nitrogen ﬂow is imposed through the microtubes. From Fig. 2 it is evident that the minor losses at the outlet of the microtubes are one order of magnitude lower than the net pressure drop while those at the in let are two order of magnitude less. The relative weight of the min or losses increases for large values of the Reynolds number and the contribution at the inlet becomes more important for short micro tubes. On the contrary, the value of the minor losses at the outlet is independent of the length of the microtube because in the experi mental tests of Morini et al. [10] the gas at the exit vented directly into the atmosphere and hence the pressure at the exit was always equal to the atmospheric value. Fig. 2 underlines that the role of the minor losses on the evaluation of the friction factor in laminar regime has to be considered of importance for short microtubes ( d/ L > 0.01) having a low value of the inner diameter. The correct evaluation of these losses is a crucial point for the accurate calculation of the friction factors for gas ﬂows in micro channels. For example, to the best of the Authors’ knowledge, the effect of the rarefaction of gases on the K _{i}_{n} and K _{o}_{u}_{t} have not been systematically investigated up to now and this fact can be an important source of error for the experimental investigation of the pressure drop for rareﬁed ﬂows through microchannels. When the effects related to rarefaction become signiﬁcant the friction factor is inﬂuenced by the value of the Knudsen number, the ratio of the gas mean free path ( k ) and the characteristic dimension (usually D _{h} ) of the channel:
Kn ¼ ^{k}
D
h
¼ ^{l}
D _{h} p
r
ﬃﬃﬃﬃﬃﬃﬃﬃﬃ
pRT
2
ð11Þ
where l is the dynamic viscosity of the gas, p is the local value of the pressure and the gas mean free path is evaluated with the
Fig. 2. Minor losses and net pressure drop ( D p _{n} = p _{i}_{n} p _{o}_{u}_{t} ) for nitrogen ﬂow through microtubes 5 cm long (empty symbols) and 50 cm long (full symbols) with an internal diameter equal to 100 _{l} m (from Morini et al. [10] ).
formula k _{M} ¼
1879 [11] . The analysis of gas rarefaction effects in internal ﬂows is a spe ciﬁc feature of the analysis of convective ﬂows through microchan nels. When the Knudsen number is less than 0.001, rarefaction effects can be considered negligible; for 0.001 < Kn < 0.1 the slip ﬂow regime occurs. From a theoretical perspective, the friction fac tor tends to be reduced when the Knudsen number increases be cause of the slip existing between the gas and the channel walls. It has been demonstrated that in laminar regime the friction factor depends on the Knudsen number and on the crosssection geome try of the microchannel as follows:
RT proposed originally by Maxwell in
ﬃﬃﬃﬃﬃﬃﬃﬃﬃ
p
p
= 2
l = q
p
ﬃﬃﬃﬃﬃﬃ
f _{D} ð Kn Þ ¼ f _{D} ð Kn ¼ 0 Þ
1
1 þ _{r}_{K}_{n}
ð
12Þ
where f _{D} (Kn = 0) indicates the value of the Darcy friction factor in the laminar regime for the channel geometry considered when the rarefaction is negligible ( Kn = 0); the friction factor reduction is related to the channel geometry through the coefﬁcient _{r} and to the Knudsen number. The value of _{r} for channels with different geometries, typical of microﬂuidic applications (i.e. circular, rectan gular, trapezoidal and doubletrapezoidal) and different aspect ra tios was reported by Morini et al. [12] . Observing Eqs. (4)–(12) it is evident that, in order to character ize and analyze a microchannel ﬂow in terms of friction factor, the following quantities should be determined:
(a) microchannel geometry, including the roughness of the channel wall, the hydraulic diameter: this means to know the geometrical characteristics of the crosssection, the geometry of the connections between the microchannels and the reservoirs (if any) and the microchannel length; (b) operating conditions including quantities such as ﬂuid tem perature, differential pressure between inlet and outlet, absolute pressure at one end of the channel and mass ﬂow rate.
The above observations can be summarized by stating that the friction factor for a microchannel can be considered in general a function of the following quantities:
f _{D} ¼ f _{D} ð Dp; p ; m; T ; C; X; L ; K _{i}_{n} ; K _{o}_{u}_{t} Þ
_
ð
13Þ
Among the other variables appearing in Eq. (13) the friction factor can depend in turbulent regime also on the wall relative roughness (e /D _{h} ) of the microchannel; it has been demonstrated that, even in laminar regime, very large values of the relative roughness ( _{e} / D _{h} > 5%) can inﬂuence the value of the friction factor [13] . Using the physical variables recalled in Eq. (13) one can calcu late also the Mach number by means of Eq. (6) and the Knudsen number by means of the Eq. (11) ; through these values one can check the effect of compressibility and of rarefaction on the friction factor and compare the results with the theoretical predictions for compressible and rareﬁed gases. With these parameters in mind, special consideration must be paid both to the design of the experimental setup and to measure ments during a speciﬁc test round at such a small scale to ensure a certain accuracy for reasonable results. Focussing on the experi mental determination of the friction factor in microchannels for gas ﬂows, with the aim to discuss the role played by each param eter on its determination, a detailed analysis of the most interest ing experimental techniques for the measurement of the channel dimensions, gas pressure, temperature, velocity and mass ﬂow rate, together with the analysis of the typical values of their uncer tainty, is presented in the coming sections.
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
853
3. Measurement techniques for microchannels
In this section the most interesting measurement techniques proposed for the determination of the physical quantities useful for the determination of the friction factor in microchannels are summarized.
3.1. Measurement of microchannel dimension and of its roughness
In the experimental tests devoted to the pressure drop analysis in microchannels some methods have been used in order to esti mate the inner diameter of microtubes. Poiseuille [14] in 1840 used mercury to ﬁll glass capillary tubes (down to 15 _{l} m of diam eter) of a ﬁxed length in order to determine the average diameter by weighing the tube before and after ﬁlling. This technique is very simple and enables an accurate determination of the average ﬂow ing area of a microchannel along its whole length. One limitation is that, in order to obtain a good accuracy of the measurement for very small tubes, one has to use very long microtubes. An advan tage of this method is that only a precision balance is required to estimate the ﬂowing area. One interesting development of this technique is based on the accumulation of a heavy liquid (water, mercury, etc.) ﬂowing through a microtube under a constant pressure drop, which gener ates a constant mass ﬂow rate through the channel [7] . If one knows the length of the microchannel, the ﬂuid density and the viscosity, the crosssectional area of the microchannel can be determined by a careful measurement of the liquid weight passing through the microtube over a certain time interval: this implies that the Poiseuille law is assumed to hold at the microscale too. This technique can be very useful to reduce the uncertainty on the determination of the inner diameter of a microtube as demon
strated by Asako et al. [7] , who reported an uncertainty of the order of ±0.2 l m on average inner diameter measurement for a tube with
a nominal diameter of 150 l m (±0.13%). For very small mass ﬂow rates this measurement technique requires very long time for li
quid accumulation. As underlined by Asako et al. [7] , when water
is used as test ﬂuid, care must be used in order to estimate the lost
weight of the accumulated liquid due to evaporation during the experimental tests, which strongly depends on the environmental condition. This aspect can be very critical when lower mass ﬂow rates are generated in the smaller microchannels. Currently, one of the most common methods to measure micro channel dimensions is the scanning electron microscopy (SEM), which images the proﬁle of the channel crosssection with a high resolution. When the tested microchannels are closed (e.g. com mercial microtubes), this technique is applicable to the inlet and outlet sections only; no information can be derived on the axial variation of the channel geometry and on the roughness distribu tion along the channel. This can represent a strong limitation for the correct evaluation of the average channel roughness because the inlet and the outlet sections may be nonrepresentative of the dominant crosssection along the length due to the cutting which may deform the ends. In this case, only with destructive tests after the experimental investigation can one try to recon struct the axial variation of the roughness and of the crosssection by grinding the channel open. Multiple cuts along the channel may prove useless because the action can change the channel’s cross section permanently. On the contrary, for open or optically acces sible microchannels (like silicon rectangular microchannels bonded with a silicon/glass cover) this technique can be applied before the specimens are sealed. With magniﬁcation and image processing, the dimensions of the channel crosssection and the topology of the roughness can be determined with a good accuracy. Since this technique is convenient and exhibits satisfying accuracy,
it has been employed by many investigators (Tang et al. [15] , Celata et al. [16] , Zohar et al. [17] , Morini et al. [10,18] ). It is inter esting to note that the Poiseuille [14] and Asako et al. [7] methods allow to measure directly the average ﬂowing area along the whole length of the microchannel, which is a signiﬁcant parameter for the evaluation of the friction factor as compared to the algebraic aver age of the inlet and outlet area obtained by SEM imaging. Still concerning the roughness, in noncircular microchannels made by chemical or ion etching on silicon wafers and sealed with different materials (glass, Pyrex, silicon) the distribution of the wall roughness along the wetted perimeter of the channel can ex hibit a strong nonuniformity. However, in this case the measure ment of the wall roughness is generally possible by using scanning electron microscope or proﬁlometers along the channel before sealing it. Another characteristic dimension of the microchannels which must be evaluated for the determination of the pressure drop is the channel length L . For long microchannels, the length L is in gen eral larger than the inner diameter; typical values of the D _{h} /L ratio fall within the range 0.1–0.001. In this case, the measurement of L is easier than the measurement of the inner diameter. On the con trary, when the behavior of rareﬁed gases through microchannel is studied in order to verify the predictions of the mathematical mod els based on the Boltzmann equation for transitional and free molecular regimes, very short microchannels must be tested which are characterized by D _{h} /L ratios near unity (Marino [19] ). In this case, the measurement of L can become problematic and, as gen eral rule, characterization of the microchannel must take also the tube length into account, rather than just the hydraulic diameter, as is otherwise customary.
3.2. Measurement of pressure
In the experimental works devoted to pressure drop analysis through microchannels, pressure measurement has been carried out: (i) only at the inlet and at the outlet of the channel (Kandlikar et al. [20] , Asako et al. [7] , Morini et al. [18] , Celata et al. [16] , Tang et al. [15] , Demsis et al. [21] ); (ii) and/or along the channel in order to reconstruct the axial distribution (Zohar et al. [17] , Hsieh et al. [22] , Jang and Wereley [23] , Turner et al. [24] , Kohl et al. [25] ). In the latter case, the knowledge of the axial pressure trend enables to identify special effects, as those related to compressibility and rarefaction, which can play an important role in the determination of the pressure drop; however, the microchannel must be speciﬁ cally provided with pressure taps along its length. If traditional pressure transducers are used, these are generally placed in the reservoirs connected to the microchannel inlet and outlet. Their presence introduces pressure losses at the inlet and the outlet of the channel which can be determined with Eq. (10) . In order to eliminate the effects of these minor losses the tube cutting method has been proposed by Du et al. [26] and other authors (Asako et al. [7] , Celata et al. [16] ). This technique consists of repeating the same mass ﬂow rate conditions of a long tube in a short one with the same inner diameter. Since the effect of the minor losses is approx imately identical for both tubes, one can eliminate it by subtracting the total pressure drop determined for the two. The short and the long tube can be selected from a set of identical tubes or one can obtain the short one by cutting the long tube. In the latter case, care must be exerted in minimizing the effect of the cut on the geometry of the microchannel. When the selection of the short tube and the long one is made among commercial tubes with the same nominal diameter, Asako et al. [7] demonstrated that the real inner diameter may actually vary markedly: in their work in order to obtain a diameter difference in each pair chosen less than 0.1 l m only three pairs of tubes were selected from precise inner diameter measurement among more than 30 specimens, which
854
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
have a nominal diameter of 150 l m but actually vary from 148 l m to 154 l m. This kind of difference in diameter is stochastic and uncontrollable, which means that with the tube cutting method, the smaller the difference in tube pairs one desires, the larger the number of the sample one has to provide. With this method, the value of the friction factor for a microtube can be calculated as follows (as indicated in Morini et al. [18] ):
f ¼ ^{2}^{q}
l
2
Dp
total
ð L Þ Dp
1
total
ð L Þ
2
L
1
L
2
D
^{3}
h
Re
_{2}
ð14Þ
where L _{1} and L _{2} are the lengths of the longer and shorter tube respectively. It has been demonstrated that this method can only be applied correctly when the pressure drop in the tube is linear, as for incom
pressible ﬂows (liquids or gases under low pressure ratio p _{i}_{n} / p _{o}_{u}_{t} ). When the pressure distribution along the channel is not linear, as
in the case of compressible and/or rareﬁed ﬂows, this method can
not be applied. In fact, in this case the effect of the minor losses
cannot be eliminated completely, as demonstrated by Morini
et al. [18] . This means that for gas ﬂows this method holds only
for low Reynolds numbers which corresponds to low Mach num bers at the outlet. When the goal is to determine the axial distribution of the pres sure along a microchannel other techniques can be employed. A
nonintrusive one for the pressure determination along a micro channel is based on PressureSensitive Paints (PSP). PSPs are optical ‘‘molecularsensors’’ which enable the measure of the pressure over a surface. When excited by an outer light source of a certain wavelength, the luminescent molecules with which the surface of the channel cover is coated will emit lumines cence of a longer wavelength. By appropriate ﬁltering, the emitted luminescence can be detected. The luminescent intensity is sensi tive to oxygen molecules near the cover surface and for this reason this technique has been proposed for the analysis of gas ﬂows. Spe ciﬁcally, an increase in the oxygen concentration causes a decrease
in the intensity of the luminescence, which is known as oxygen
quenching (Huang et al. [27] and Liu et al. [28] ). After calibration,
a relation between pressure and luminescent intensity can be
established. This nonintrusive measurement technique can pro vide pressure data with high spatial resolution both along the
channel and at the channel entrance and exit. On the other hand,
as this technique is based on oxygen quenching, it cannot be ap
plied to test of other pure gases, such as N _{2} and He. Moreover, the pressure sensitive paint is coated on a transparent cover of
the microchannel to ensure the direct contact with oxygen and excitation of luminescent molecules from an external light source.
It is evident that the main disadvantage of this technique is tied to
the shape of the microchannel tested: circular or elliptical micro tubes make it impossible to coat the pressure sensitive paint onto the inner wall, and passages which do not have a transparent side cannot beneﬁt from this optical technique.
The pressure sensitive paint (PSP) technique has been applied to the measurement of the pressure distributions in high Knudsen
number regimes by Liu et al. [29] and Bell et al. [30]. Because the PSP works as a socalled ‘‘molecular sensor’’, it is considered suit able for the analyses of high Knudsen number ﬂows, which require diagnostic tools at the molecular level. However, an application of PSPs to microdevices is very difﬁcult, because conventional PSPs are very thick compared to the dimension of microdevices owing
to the use of polymer binders. Moreover, they do not have sufﬁ
cient spatial resolution for the pressure measurement of micro ﬂows due to the aggregation of luminescent molecules in polymer binders as indicated by Mori et al. [31] . More recently, for localmeasurement of pressure in high Knudsen numbers Matsuda et al. [32,33] developed a pressuresensitive
molecular ﬁlm method (PSMF) by using the Langmuir–Blodgett (LB) technique. Before applying PSMF to micro gas ﬂows, the authors have tested the dependence of temperature on pressure sensitivity, which is the main factor of the measurement error for this kind of pressure sensor. Matsuda et al. [33] have demon strated that their paint composed of Pt(II) Mesoporphyrin IX and arachidic acid showed a small dependence of pressure sensitivity on temperature, while that of luminescent intensity is signiﬁcant.
In addition, the authors have demonstrated how, compared with the classical PSP method, the PSMF method can offer more uniform pressure sensitivity and an enhanced intensity of luminescence emission. By applying their technique to the determination of the pressure ﬁeld for gaseous ﬂow through a 2D nozzle [32] , they con cluded that the spatial resolution of PSMF is high enough for microscale ﬂow measurements whose characteristic length is over 50 l m. In order to measure the local pressure along a microchannel Kohl et al. [25] explored and proposed the optical lever method for pressure measurement. The microchannel presents a number of micrometric pressure taps connected to silicon membranes which deform according to the local pressure. The deformation was measured recording the change in deﬂection angle of a ﬁxed incident laser targeting the membrane surface. The change in deﬂection angle was measured by a photodiode sensor which can be precisely moved and positioned. Based on this principle, an inte grated pressure sensor can be produced. The sensitivity of this integrated pressure sensor can be easily adjusted by changing the spatial resolution (the distance between the membrane and the photodiode sensor). The uncertainty on the measured pressure ranged from ±2.4% to ±13.3%. Since a very precise optical system to test quantitatively the slight slope of the pressurized membrane is involved, it requires great labor to adjust this system. Also such optical system, which includes a laser, a lens and several adjustable stands for precise positioning, is expensive. In addition, the range of the measurable pressure is limited by the mechanical strength of the membrane material and the channel must be customized for use with this technique. Even microstrain gauges membranes can be employed to sense the pressure induced at the tap from the microchannel. The defor mation of the membrane results in the variation of the electrical resistance of the material making up the membrane, therefore information on pressure can be transformed into an electrical sig nal, which is usually magniﬁed for better accuracy via a Wheat stone bridge on the membrane, as in the work by Zohar et al. [17] and Bavière and Ayela [8] . After careful calibration, a correla tion between pressure and voltage in the circuit can be given. This pressure sensor can be used to test local pressure along the micro channel and has very low temperature sensitivity (±0.011% C ^{} ^{1} ), as reported by Bavière and Ayela [8] , who found no effects which might actually be the result of inaccurate measurements at the small scale. The constant coefﬁcient which links pressure and elec trical voltage varies from one membrane to another, and each strain gauge needs to be calibrated over a certain pressure range before measurement. Even in this case, the microchannel and the membranes must be speciﬁcally designed in order to use this tech nique. For liquids through very small microchannel, Celata et al. [16] demonstrated that the measurement of the differential pres sure along the channel can be linked to the ﬂuid temperature rise between the inlet and the outlet of the channel due to the viscous dissipation. This effect is more evident for ﬂuids having a low spe ciﬁc heat and a large viscosity (i.e. isopropanol). For gases, the temperature rise due to the viscous dissipation is compensated by the cooling due to the gas acceleration along the channel. For this reason, this effect is not useful for the determination of the dif ferential pressure along a microchannel for gas ﬂows.
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
855
3.3. Measurement of temperature
The measurement of the bulk ﬂuid temperature along a micro channel cannot be made directly without large disturbance on the ﬂuid ﬂow. Suspended micro temperature sensors within micro channels have been proposed in the past, as reported in an inter esting review paper of Nguyen [34] . The ﬁrst prototypes of suspended sensors made by Petersen et al. [35] and Lammerink et al. [36] demonstrated that full integration between this kind of temperature sensors and the microchannels was possible but their impact on ﬂuid ﬂow was not negligible at all. Nowadays microfabrication offers a large variety of devices to achieve local temperature measurements. Among the existing devices it is pos sible to distinguish three groups: (i) thin ﬁlm resistance tempera ture detector ( l RTD), (ii) thin ﬁlm thermocouple ( l TFTC) and (iii) semiconducting sensors (SC). In microchannels, to avoid disturbances of the ﬂuid ﬂow, in many cases only the wall temperature along the channel is deter mined by means of integrated temperature sensors and the direct determination of the ﬂuid bulk temperature can only be obtained at the inlet and outlet sections of the channel by using speciﬁc ple nums. Sometimes, the thermal inertia of the temperature sensors placed at the walls of the microchannel is larger or comparable to that of the channel walls. In this case, the presence of the sen sors can modify the wall temperature distribution. The problem is how to calculate the ﬂuid temperature when the wall tempera ture is known. In adiabatic ﬂows this is simple if one can assume that thermal equilibrium exists between the walls and the ﬂuid. This is not the case when the ﬂow is heated. In the following a summary of the most interesting temperature microsensors developed in the last years is given with the indica tion of the main advantages and drawbacks of each one. In Fig. 3 a schematic representation of these three different kind of temper ature microsensors together with their characteristic calibration curves is given in order to highlight the main differences among l RTD, l TFTC and SC due to their different working principle. The ﬁrst kind of temperature microsensors which can be inte grated in microchannels are the Microresistance Temperature Detectors ( l RTDs). This type of temperature sensors is widely used
in microtechnology devices ﬁrstly because it is easy to make and secondly because it generally is characterized by a linear response. The functioning principle of a l RTD is the same as that of the clas sical RTD in which the temperaturedependent electrical resistivity of a material is used in order to estimate the temperature ( Fig. 3 a). Among the several materials used for the realization of the l RTDs the most common are Pt, polySi, Al, Ni, W, Au or yet Ag [37–43]. The fabrication of the sensitive metal elements for _{l} RTDs is well controlled by using sputtering or evaporation CVD pro cesses. Platinum is commonly used for different applications in microtechnology for its catalytic property in combination with some gases or, at other times, for its inert properties in combina tion with a large variety of ﬂuids. However, platinum ﬁlms can present some drawbacks: the ﬁrst one is the need of an adhesion layer on almost all the substrates. In many cases the deposition of the platinum ﬁlm generates an alloy with different thermal properties than the pure material and the characteristic curve of the electrical resistance vs temperature of the _{l} RTD can change completely. Moreover, platinum ﬁlms can be damaged if the tem perature becomes larger than 550 C. Other drawbacks of the plat inum ﬁlms is their cost and their CMOS incompatibility [39] . For these reasons, some microﬂuidics technical applications of _{l} RTD use polysilicon (polySi) [40] . Thin Film Thermocouples ( l TFTCs) are active elements which use the Seebeck effect to measure the temperature of a twometal junction ( Fig. 3 b). Microfabrication of _{l} TFTCs can be more or less complex depending on the existing material constraints for the speciﬁc microdevice needed. Zhang et al. [44] proposed a chromel thin ﬁlm thermocouple embedded on a Ni substrate and demon strated that their behavior is similar to the standard KType ther mocouples. In fact, the embedded TFTCs of Zhang et al. [44] on Ni substrates have shown that their thermal sensitivity is not af fected by the junction size between 25 _{l} m and 80 _{l} m. This ther mal sensitivity has been precisely measured for a chromel ﬁlm of 100 nm and a junction of (60 l m 60 l m) up to 900 C; the See beck coefﬁcient of this l TFTCs was 40.6 l V C ^{} ^{1} in agreement with the standard value of this coefﬁcient for the classical KType ther mocouples. The results obtained by Zhang et al. [44] were con ﬁrmed by Choi and Li [45] who showed as a K _{l} TFTCs with a
Fig. 3. Schematic representation of the most common temperature microsensors with the indication of their characteristic curve; (a) thin ﬁlm resistance t emperature detector ( l RTD); (b) thin ﬁlm thermocouple ( l TFTC) and (c) semiconducting sensors (SC).
856
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
junction of 25 l m 25 l m and a ﬁlm thickness of 150 nm were able to provide a good sensitivity (40.4 _{l} V C ^{} ^{1} ) up to 800 C and a fast response time (28 ns). However, the authors underline that a microfabrication trouble for this kind of microcomponents is linked to the surface quality (especially roughness) of the thin ﬁlm layers used in these _{l} TFTCs which can strongly inﬂuence the stability of these temperature microsensors. With the aim to take advantage of the CMOS compatibility pro cess (integration, low cost) an alternative for the temperature mic rosensors are the Semiconducting sensors (SCs) which use Polysilicon active elements ( Fig. 3 c). Bianchi et al. [46] present a state of the art of integrated smart temperature SCs sensors like MOS transistors, bipolar transistors and diodes. The use of MOS transistors as temperature sensors needs the use of the transistors in weak inversion in order to obtain a linear relation between tem perature and their electrical parameters even if some limitations due to leakage currents at high temperature can exist [47,48] . Bipolar transistors and diodes present directly a junction voltage proportional to the absolute temperature. Generally, at room tem perature, silicon p – n junctions have a forward voltage drop of 0.7 V, and this voltage decrease by 2 mV for every degree of in crease [49] . Typical temperature range of these sensors could be 55 to 175 C with a typical accuracy between ±0.1 and ±3 C [46] . Transistors with CMOS or bipolar IC technologies enable the realization of all the needed circuits in order to integrate directly the temperature compensation or calibration functions to the sen sors. Filanovsky and Lee [48] realized two temperature sensors with signalconditioning ampliﬁers to extract directly from a MOS transistor the linear temperature dependence of V _{T} . Using BiCMOS technology, the ﬁrst solution chosen was a threshold extractor circuit which gave a linear temperature characteristic be tween 40 and 150 C but showed some disparities linked to fab rication. The second one was a resistive Wheatstone bridge which gave a nonlinear temperature response but with a good repeat ability in the same range of temperature. Finally, planar diffused silicon diodes seem to be a simpler way to measure temperature when employing SCs sensors. Guha et al. [39] showed a high temperature gas sensor consisting of an embedded diode able to give a linear response up to 260 C accord ing to the calibration with a sensitivity of 1.2 ± 0.005 mV/ C. In a pH sensitive ISFET chip, Chin et al. [50] have integrated a pn diode which presented good linearity with a sensitivity of 1.51 mV/ C (in the range 0–50 C) and which also have been used with a spe ciﬁc compensation circuit dedicated to the application. Even if, up to now, CMOS technologies are not so frequently used in Microﬂuidics applications, in the future SC sensors could have a wide range of applications in this ﬁeld due to their ﬂexibil ity and to their high scale of integration that enables to expand the single sensor to an array of sensors. For example, Han and Kim [51]
have developed a 32 32 diodes (1024 diodes) array for measur ing the temperature distribution on a small surface (8 mm 8 mm). This kind of sensor reduce the 2048 interconnections pads needed for 32x32 RTD or TFTC sensor array to only 64 interconnec tions pads that involve a tiny chip and the simplest design. In Han and Kim’s work [49] , calibration experiments evidenced the linear output temperature signal of this array in the range 0–100 C with an accuracy of ± 0.5 C. In compensation of this ﬂexibility at the fabrication level, array sensors need a multiplexer to read the measurements. In Table 1 a selected list of microsensors for temperature mea surement is shown with the indication of the main technical char acteristics (when available) of these sensors (thermal sensitivity, accuracy and range of application). In many of the published scientiﬁcal works temperature mea surements are still obtained using traditional sensors like thermo couples or Platinum RTD. Bavière et al. [52] measured the temperature along a rectangular channel by arranging four ther mocouples uniformly in line with the channel axis. The thermo couples were planted in blind holes 500 _{l} m from the channel wall surface. To predict the temperature of the wall surface a cor relation was presented based on calibration experiments and the temperature measured by thermocouples a constant distance away from the wall surface. This approach provides an indirect way to obtain the local wall surface temperature along the whole channel, which makes it possible to analyze the local heat transfer charac teristics. However, the existence of these holes near the channel walls may alter the temperature distribution inside because they introduce a heterogeneous thermal conductivity. Demsis et al. [21] investigated the convective heat transfer coefﬁcient in a coun ter ﬂow, tubeintube heat exchanger. Two ports perpendicular to the axis of the microchannel were machined and connected to the inlet section of the inner tube 35 mm before the heated section, one for the measurement of temperature and the other for pres sure. The measurement of the outlet pressure and temperature was proposed in the same way through two ports 35 mm after the heated section. The risk of this technique is linked to the erro neous evaluation of the ﬂuid outlet temperature: if the part of tube between the end of the heated section and the outlet side port is not thermally insulated with care, the ﬂuid temperature decreases along this stretch and the temperature recorded at the exit side port differs from the ﬂuid temperature at the end of the heating section. Also, the side ports may disturb the main stream inside the tube. This kind of side ports can be readily applied to non circular microchannels or circular microtubes with relatively large inner diameter; however, if the inner diameter is very small, below 0.3 mm, it becomes difﬁcult to manufacture them. Besides thermocouples or diode arrays which can provide single or multipoint measurement of temperature, some speciﬁc optical techniques have been used for the measurement of a
Table 1 Characteristics of the microsensors for temperature measurement.
Authors 
Material 
Thermal sensitivity 
Accuracy 
Temperature range 

RTDs 
Schöler et al. [38] Wu et al. [42] 
Platinum 
20–80 C 45–105 C 

Silver paint 

Wu 
et 
al. 
[40] 
Silicon 
0.05 C 

Choi et al. [43] 
Gold 

TFTCs 
Zhang et 
al. [44] 
Chromel 
40.6 _{l} V C ^{} ^{1} 40.4 _{l} V C ^{} ^{1} 
20–900 C 20–800 C 

Choi and 
Li [45] 
Chromel 

Kim and Kim [61] 
Chromel 

SCs 
Guha et al. [39] Bianchi et al. [46] Chin et al. [50] Han and Kim [51] 
Silicon 
1.20 mV C ^{} ^{1} 
0.005 mV C ^{} ^{1} ±0.1–3 C 
20–260 C 55–175 C 10–50 C 0–100 C 

Silicon 

Silicon 
1.51 mV C ^{} ^{1} 

Silicon 
±0.5 C 
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
857
twodimensional temperature ﬁeld on a solid surface at micromet ric scales, such as liquid crystal thermography (Muwanga and Hassan [53] ), infrared thermography (Mosyak et al. [54] ) and temperature sensitive paints (Liu and Sullivan [28] ). The most important constrain of these optical techniques is that these methods can be successfully applied only in the cases in which the external surface of the test section is not thermally insulated.
3.4. Measurement of local velocity data
There is a crucial need of local experimental gas velocity data through microdevices but all the velocimetry techniques proposed for liquid ﬂows in microchannels present strong limitations if ap plied to the analysis of gas ﬂows through microchannels. For the determination of the local velocity of a gas in a microchannel three optical velocimetry techniques have been proposed in the open lit erature up to now: (i) the LaserDoppler Anemometry (LDA); (ii) the microParticle Image Velocimetry ( _{l} PIV) and (iii) the Molecu larTagging Velocimetry (MTV). The LaserDoppler Anemometry (LDA) has been preliminary used by Ladewig et al. (cited in [55] ) in order to measure gasphase data in an operational Ushaped minifuel cell with channels of 2 mm 2 mm rectangular crosssection. The ﬂow was seeded with water droplets, but the velocity data obtained were not yet com plete enough to fully characterize the ﬂow inside the minichannels. Furthermore, LDA is a pointmeasurement technique; in order to obtain a more representative visualization of the whole velocity ﬁeld in a microdevice a ﬁeldmeasurement technique as the micro Particle Image Velocimetry can be more useful. For the problems related to the systematic use of the LDA technique for the recon struction of the velocity proﬁles in microchannels the LDA tech nique can be considered useless in measuring the velocity within channels having an inner hydraulic diameter below 1 mm. On the contrary, for the investigation of the velocity ﬁeld in micro channels having inner diameters below 1 mm the microParticle Image Velocimetry is nowadays a consolidated technique but only for liquid ﬂows. An interesting and complete review of the progress of this technique in the recent years is due to Lindken et al. [4] . Up to now, few works have described the application of this technique for the analysis of gas ﬂows through microchannels. One reference paper in this ﬁeld is due to Yoon et al. [55] ; in this paper the micro PIV technique has been used to directly measure gasphase velocities in situ in an operational fuel cell with chan nels having characteristic dimensions of 1–2 mm. For PIV measure ments, in macroscale gas ﬂows, the gas is often seeded with olive oil particles generated by an atomizer. To this aim, high pressures and a high ﬂow rates are needed for the generation of the oil seed particles. On the contrary for microscale gas ﬂow applications, as microfuel cell, the gas ﬂow rate is relatively small (Reynolds num bers of the order of hundreds to a few thousands) and this can cre ate problems in the generation of appropriate tracer particles. Since PIV is based on imaging of particulates introduced into the ﬂow, the accuracy of the gasphase ﬂow velocity measurement is strongly dependent on the ability of the tracer particles to follow the stream. In the work of Yoon et al. [55] , water droplets have been used as seed particles for measuring the ﬂow in a straight minichannel. This choice was motivated for fuel cells with the opportunity to introduce the water particles in the gas ﬂow using the humidiﬁer section without contaminating the system. The re sults of Yoon et al. have demonstrated that certain limitations are encountered when trying to use such tracer in ﬂows with sig niﬁcant decelerations, such as in a 180 switchback turn. In partic ular, the experimental results of Yoon et al. have evidenced that tracer particles smaller than 1 l m are required to accurately follow the ﬂow in such regions. This aspect is one of the most important constraints in the application of the _{l} PIV technique for the analysis
of gas ﬂows in microchannels. Sugii and Okamoto [56] have ap plied the PIV technique in order to investigate the velocity distri bution in a fuel cell with 1 mm 0.5 mm rectangular polymer microchannels. Even in this paper, it is stressed how the main problem for the application of this technique to the investigation of micro gas ﬂow is the generation of proper seeding. In this work ﬂuorescent oil particles having a diameter between 0.5 and 2 _{l} m were used as tracer particles in a nitrogen ﬂow for low Reynolds numbers between 26 and 130. MolecularTagging Velocimetry (MTV) is a wholeﬁeld optical technique that relies on molecules that can be turned into long life time tracers upon excitation by photons of appropriate wavelength [57] . These molecules can be either premixed or naturally present in the ﬂowing medium (unseeded applications). Typically, a pulsed laser is used to ‘‘tag’’ the regions of interest, and those tagged re gions are interrogated at two successive times within the lifetime of the tracer. Lagrangian displacement vector provides the estimate of the velocity vectors. This technique can be thought of as essentially a molecular counterpart of Particle Image Velocimetry (PIV), and it can offer advantages compared to particlebased techniques where the use of seed particles is difﬁcult, or may lead to complications as for gas ﬂows in microdevices. However, this technique has been only used for the analysis of liquid ﬂows in microchannels [58,59]. On the contrary, up to now the application of this technique for the analysis of gas ﬂows in microchannels is not completely consolidated. A speciﬁc European research project (GASMEMS project) devoted to the development of velocimetry techniques for gas ﬂows in microchannels has been launched in order, among others, to demonstrate the possibility of using _{l} PIV and MTV for the analysis of the gas ﬂows in channels having inner dimensions below 1 mm [60]. It is possible to conclude that for the measurement of the local velocity data in microchannels much work is still needed in order to adapt the velocimetry techniques proposed for liquid ﬂows through microdevices to the analysis of gas ﬂows; this ﬁeld is of strategic importance especially for the optimization of microﬂuidic components as, for example, the micro and minifuel cells.
3.5. Measurement of ﬂow rate
The ﬂow rate through a microchannel can be measured directly by means of appropriate instruments (mass or volumetric ﬂow me ters) or in an indirect way by checking the value taken by other measurable quantities, like pressure, forces, weight, volume, tem perature or a combination thereof. As a rule of thumb, it is possible to use a commercial mass or volumetric ﬂow meter for gas ﬂows only if the gas mass ﬂow rate through a microchannel is larger than 10 ^{} ^{8} kg/s (0.1 Nml/min). On the contrary, for very low mass ﬂow rates (<10 ^{} ^{8} kg/s) indirect methods can be considered more reli able in order to determine the ﬂow rate with low values of uncertainty. The commercial ﬂow meters for gas ﬂows can be divided in two groups: the volumetric ﬂow meters and the mass ﬂow meters. The volumetric ﬂow meters give an indication of the ﬂow rate regard less of the ﬂuid tested; the mass ﬂow meters, on the contrary, can be used only with the ﬂuid for which they have been calibrated. The sensors of the commercial gas ﬂow meters can be divided in nonthermal ﬂow sensors and thermal ﬂow sensors. The ﬁrst kind is classiﬁed according to the mechanical working principle by means of which the ﬂow rate can be measured indirectly:
(1) by the drag force (i.e. using silicon cantilevers); (2) by pressure measurements (i.e. using capacitive and/or piezoresistive pressure sensors); (3) by using the Coriolis principle.
858
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
The main disadvantage of the nonthermal ﬂow sensors is linked to the dependence of the force, pressure difference and Cori olis force on the density of the working ﬂuid. Since the gas density depends on the gas temperature, a temperature compensation is a compelling need for this kind of sensors. As indicated by Nguyen [34] , the thermal ﬂow sensors em ployed in the commercial ﬂow meters can be classiﬁed as follows:
(1) thermal mass ﬂow sensors which measure the effect of the ﬂowing ﬂuid on a hot body (micro hotwire and micro hot ﬁlm sensors); (2) thermal mass ﬂow sensors which measure the asymmetry of temperature proﬁle around a micro heater which is modu lated by the ﬂuid ﬂow (calorimetric sensors); (3) thermal mass ﬂow sensors which measure the delay of a heat pulse over a known distance (timeofﬂight sensors).
The most popular sensors among the commercial ﬂow meters are the calorimetric sensors: this kind of ﬂow meters are able to predict the mass ﬂow rate through the sensor by means of the fol lowing balance equation:
_
m ¼
Q th
c p ð T b; out T b;in Þ
^{ð}^{1}^{5}^{Þ}
which puts in evidence how the gas mass ﬂow rate can be deter mined by means of two local measures of temperature (T _{b}_{,}_{o}_{u}_{t} and T _{b}_{,}_{i}_{n} ) and the knowledge of the heat power (Q _{t}_{h} ) transferred to the gas by the heater. By using microfabrication, this kind of thermal mass ﬂow sensor can be directly integrated into a microchannel. Schöler et al. [38] have shown as two classic Pt RTDs and a microheater made on a glass substrate can be directly integrated on a SU8 epoxy resist microchannel (50 l m 15 l m 8000 l m). As demonstrated by Kim and Kim [61] by using a capacitive mass ﬂow sensor using two KType TFTC (150 l m 150 l m) associated to a PDMS micro channel (800 l m 800 l m 4 cm), the accuracy of this kind of mass ﬂow meters can be strongly inﬂuenced by the position and the thermal insulation of the microsensor. When the gas mass ﬂow rates that one wants to measure through microchannels are very low (<10 ^{} ^{8} kg/s) the use of com mercial devices becomes unsuitable because of the high level of uncertainty afﬂicting the measurements. In addition, accurate measurements of mass ﬂow rate in microchannels are challenging for gas ﬂows, which have a density very sensitive to the room pres sure and temperature ﬂuctuations. Low gas ﬂow rates can be determined using the general expres sion of the mass ﬂow rate as the change of mass over time:
_
m
¼ ^{d}^{m}
_{d}_{t}
ð16Þ
If the gas can be considered as an ideal gas, the mass ﬂow rate can always be expressed as follows:
m ¼ ^{p}
_
dV
V
dp _{} pV dT
RT ^{2}
dt
RT dt ^{þ} RT dt
ð17Þ
It can be noted from Eq. (17) that if two of the three variables (vol ume, pressure and temperature of the gas) are kept constant while the third one is monitored during the experiment, the mass ﬂow rate of the gas can be determined. This provides the basic insights into gas mass ﬂow rate measurements. In practice, it is impossible to keep two of the three variables strictly unchanged, so great ef forts are made to minimize the ﬂuctuations of the nominally un changed variables to determine the mass ﬂow rate as accurately as possible. These ﬂuctuations are usually the main source of errors in low mass ﬂow rate measurements for gas microﬂows.
Based on Eq. (17) several techniques have been proposed for the determination of the mass ﬂow rate below 10 ^{} ^{8} kg/s. In the droplet tracking technique the gas volumetric ﬂow rate is measured by introducing the outﬂow gas into a pipette, in which the gas pushes a droplet forward. By recording the position of the droplet versus time under the assumption of constant pressure and temperature, the velocity of the droplet can be determined and knowing the diameter of the pipette the mass ﬂow rate of gas can be calculated. In this case, Eq. (17) can be rewritten as:
m ¼ ^{p}
_
dV
¼ ^{p}
RT
2
pip
pd
Dl
4
_{D}_{t}
RT dt
ð 18Þ
where d _{p}_{i}_{p} is the diameter of the pipette and D l is the distance the droplet moves during a certain time interval (D t ). This technique does not require expensive meters and can be performed by using multiple microchannels. The lowest mass ﬂow rate that can be mea sured with this technique depends on the speciﬁc setups and on the maximum number of microchannels tested in parallel. As rule of thumb, the lowest value of the mass ﬂow rate that can be measured by using the droplet tracking technique is of the order of 10 ^{} ^{1}^{0} kg/s. However, it is possible to obtain lower mass ﬂow rates if the mea surement is made using hundreds of identical microchannels in parallel in a single test section. This is the case of Shih et al. [62] who measured the mass ﬂow rate of helium through microchannels in the order of 10 ^{} ^{1}^{2} kg/s, which is the same order measured by Ewart et al. [63] with nitrogen. Colin et al. [64] obtained the mass ﬂow rate down to the order of 10 ^{} ^{1}^{3} kg/s for helium, while Celata et al. [16] reached ﬂow rates of the order of 10 ^{} ^{8} kg/s for the same type of gas with this technique. However, it is technically difﬁcult to maintain a constant moving speed for the droplet. Another problem is the accurate measurement of the pipette inner diameter, which weighs most in determining the mass ﬂow rate, as shown in Eq. (18), and, in the case of parallel channels, their dimensions may vary signiﬁcantly and thus the value obtained is only an average. If gas ﬂow is accumulated in a tank for a certain time interval, there will be a pressure rise in the tank during the ﬂow. If the pres sure in the tank is under reasonable control and not very high, the expansion of the tank or the increase in its volume can be safely neglected. In this case Eq. (17) becomes:
_
m
¼ ^{V}
dp _{} pV dT
RT dt
RT ^{2}
dt
ð
19Þ
If the second term in the righthand side of Eq. (19) is very small compared to the ﬁrst one (i.e. <1%) it can also be dropped and by careful measurement and recording of pressure rise inside the tank versus time the mass ﬂow rate can be determined. This is usually referred to as constant volume technique. Based on this principle, Ewart et al. [65] built an experimental setup in which the outlet of the microchannel is connected to a large tank. By recording the pressure versus time inside the tank, the mass ﬂow rate can be determined. The lowest mass ﬂow rate measured with this tech nique was of the order of 10 ^{} ^{1}^{3} kg/s. The volume of the tank should be carefully chosen so that the pressure change is detectable by a pressure sensor for mass ﬂow rate calculation and at the same time negligible to remain a nearly constant outlet pressure. Great care should also be paid to the temperature ﬂuctuation of the gas as it is compressed into the tank, so that the temperature change can be neglected, which greatly simpliﬁes the calculation of the mass ﬂow rate. Arkilic et al. [66] used a twotank, modiﬁed constantvolume accumulation technique to measure the mass ﬂow rate of gas, as the singletank procedure would not ensure the thermal stability needed for their experiment. In this case, Eq. (19) is not sufﬁcient to calculate the mass ﬂow rate, as both upstream and downstream pressures vary during the experiment. Instead, the pressure
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
859
difference between the two tanks (both located downstream of the microchannel) is measured. During the test one tank is used to accumulate the gas ﬂowing out of the microchannel and the other one remains in steady state which provides a pressure reference. The pressure difference between the two tanks is measured instead of the absolute pressure for the determination of mass ﬂow rate, as given by:
m ¼ ^{V}
_
RT
dð DpÞ _{} DpV dT
dt
RT ^{2}
_{d}_{t} ð
20Þ
where Dp is the pressure difference between the reference tank and the ﬂow tank. As the two tanks are connected before the ﬂow starts, this quantity is 0 at the beginning. Due to the relatively small mass ﬂow rate of gas through the microchannels, when the ﬂow ﬁnishes the pressure difference can be very small, quite several orders of magnitude smaller than the absolute pressure in Eq. (19). Therefore, the sensitivity of mass ﬂow rate to the tank temperature ﬂuctuation is reduced by several orders, and the tank for gas accumulation can be reasonably regarded as isothermal during the experiment. In addition, this technique requires that the two tanks have identical temperature and undergo the same thermal ﬂuctuations, which can be realized by a good design, arrangement and thermal insula tion of the tanks. For this reason, in the work by Pitakarnnop et al. [67] the whole setup was thermally regulated with Peltier modules to maintain a constant temperature. This method provided data for mass ﬂow rates of an order of 10 ^{} ^{1}^{1} kg/s and a sensitivity as low as 7 10 ^{} ^{1}^{5} kg/s (Arkilic et al. [68] ) was reported; Pitakarnnop et al. [67] developed a new setup for gas microﬂows in which the constant volume method for mass ﬂow rate measurement was implied to both the upstream and downstream ﬂow. This offers a doublecheck providing the possi bility to compare the measured results for the same ﬂow. Their measured values of mass ﬂow rates can be as small as to 7.1 10 ^{} ^{1}^{4} kg/s.
To achieve a constant mass ﬂow rate with this technique, the pressure difference between the two tanks should increase linearly with time during the experiment. As the outlet of the microchannel is directly connected to the tank, the outlet pressure keeps chang ing and is uncontrolled. The inlet pressure should be very carefully adjusted with the passage of time to achieve a linear increase of pressure difference between the two tanks. This is very difﬁcult to achieve in practice. Thus, the process of gas ﬂow in the micro
channel becomes timedependent with this method. In order to complete this review, it is possible to highlight that the most sensitive technique for the gas detection and for the mea surement of very low mass ﬂow rates in leaks is based on mass spectrometry and it has been presented by Tison in 1993 [69] . Tison developed a speciﬁc setup to measure gaseous capillary leak rates ranging from 10 ^{} ^{6} mol/s to 10 ^{} ^{1}^{4} mol/s. A series of complex operation steps are involved in the test so that the ﬁnal determina tion of ﬂow rate is dependent on the ratio of measured quantities instead of their absolute values, which reduces the sensitivity of possible ﬂuctuations (temperature, pressure and so on) and ex pands the measurement to very low range (pressures between 10 ^{} ^{4} and 10 ^{} ^{8} Pa). In this work, however, the focus is on the deter mination of leak rates of a speciﬁc gas, helium, which, at such pres sures, is actually part of a mixture, so that it is the partial pressure which has to be determined, and either ion gauges or mass spec trometers are used to this aim. It can be highlighted that Tison’s technique is timeconsuming (6–8 h for a single test) with respect the other techniques reviewed and its use in microﬂuidics can be considered useful only for speciﬁc tests in which the gas mass ﬂow rates are very low (<10 ^{} ^{1}^{5} kg/s). To summarize the main observations made in section for each technique, Table 2 shows the most important methods proposed for the ﬂuiddynamic investigation in microchannels. For each technique, the typical ranges of values for which they have been applied together with the typical values of uncertainty
Table 2 Summary of the experimental techniques proposed for the analysis of the gas ﬂows in microﬂuidic applications.
Physical 
Technique 
Range 
Typical uncertainty 
References 
quantity 

Mass ﬂow 
Flow meters 
50000–0.05 Nml/min 
0.6%, 1%, 2% 
Morini et al. [18] , Celata et al. [16] , Tang et al. [15] Celata et al. [16] Shih et al. [62] , Ewart et al. [63] , Colin et al. [64] Ewart et al. [65] , Pitakarnnop et al. 
rate 

Droplet tracking 
10 ^{} ^{8} kg/s 10 ^{} ^{1}^{0} –10 ^{} ^{1}^{3} kg/s (hundreds of parallel microchannels) 10 ^{} ^{1}^{3} kg/s 
9.64% (calculated) 

3–4% 

Single tank constant volume 
4.5% 

[67] 

Twotank constant volume 10 ^{} ^{1}^{1} kg/s 
– 
Arkilic et al. [66,68] Tison [69] 

Mass spectrometry method 
4 10 ^{} ^{1}^{7} kg/s–4 10 ^{} ^{8} kg/s 
1–8% 

Diameter 
Accumulation of heavy liquid SEM 
150 l m 0.5–40 l m 133–730 l m 10–300 l m 30–254 l m 
0.13% 
Asako et al. [7] Zohar et al. [17] Morini et al. [18] Tang et al. [15] Celata et al. [16] 
1.25–2% 

2% 

0.3–2.98% 

2.48–3.67% 

Temperature Thermocouples (directly, at inlet and outlet) Thermocouples (indirectly, along microchannel) Thermocouples (directly, after inlet and before outlet) 
0–200 C 
0.17%, 0.25% 
Tang et al. [15], Morini et al. [18] 

– 
Systematic error reduced by around 0.8 K 
Baviere et al. [52] 

– 
Inlet and outlet effects minimized 
Demsis et al. [21] 
Infrared thermography 50–130 C
Sensitivity: 0.1 C Mosyak et al. [54]
Liquid crystal thermography 43–50 C 
1.1–1.5 C 
Muwanga et al. [53] 

Temperature sensitive paint 
10–100 C 
Sensitivity. 1% C ^{} ^{1} Liu et al. [28] 

Pressure 
Pressure transducer 
0–1.5 MPa 
0.25%, 0.5% 
Tang et al. [15], Morini et al. [18] , 

Tube cutting PressureSensitive Paints Pressuresensitive molecular ﬁlm 
From near vacuum to 2 atm. 10 ^{} ^{2} –10 ^{4} Pa 
Minor losses minimized 1 mbar – 
Asako et al. [7] Asako et al. [7] , Celata et al. [16] Huang et al. [27] , Liu et al. [28] Matsuda et al. [32,33] 

Microstrain gauge 
membrane 
0.1–0.4 MPa 
1% 
Zohar et al. [17] , Baviere et al. [8] 

Optical lever 
0–1.4 MPa 
2.4–13.3% 
Kohl et al. [25] 
860
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
are shown. The data in Table 2 may be employed during the design of a new test rig in order to select the best measurement tech niques with the goal to minimize the global uncertainty on the de rived quantities, as shown in the next section.
4. Uncertainty analysis
Eq. (13) highlights that the friction factor is a function of
many parameters. Eqs. (4) and (5) link the friction factor and the measurable quantities. The friction factor is a general function
f
¼ f ð x _{1} ; x _{2} ;
;
x _{n} Þ in which
the single measurable variable x _{i} is
characterized by a known absolute uncertainty d x _{i} which corre sponds to the relative uncertainty ^{d} ^{x} ^{i} . Based on the theory of uncertainty propagation, the absolute uncertainty on the friction factor is given by [70] :
x
i
df ¼
s
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
X
_{@} _{x} j covð x _{i} ; x _{j} Þ
f
@
@
x _{i}
2
n
i¼1
ð dx _{i} Þ ^{2} þ 2 ^{X}
n
i¼1 X j¼iþ1
n
@ f @ f
@ x _{i}
ð21Þ
where cov ð x _{i} ; x _{j} Þ is the covariance associated with x _{i} and x _{j} . If x _{i} and x _{j} are independent, their covariance is 0; if the two variables are correlated, their covariance will not be 0. However, in experiment the singly measured variables are independent of each other, which greatly reduces the complexity of calculation. Even with such sim pliﬁcation, the measurable variables might be correlated by the instrumentation used in the experimental setup (i.e. the use of the same pressure sensor or temperature sensor and so on) and/ or by the room conditions (via data acquisition system). However, this kind of correlation is relatively weak (i.e. Agilent 34420A digital multimeter for data acquisition presents a slight sensitivity of the magnitude of 0.0008% with regard to room temperature) and can be either negative or positive, which tends to balance out the total contribution to the overall uncertainty of the function f . For this rea son, in the following discussion the measurable variables will be considered independent to each to other, and Eq. (21) can be simply expressed as follows for the total relative uncertainty of f :
df
f ^{¼}
s
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2
@ f
x
1
2
d
x _{1}
2
þ
2
f
@
x
2
dx
_{2}
@ x _{2} f
x
2
2
þ
@ f
x
n
2
dx
_{n}
@ x _{1} f
x
1
@ x _{n} f
x
n
ð22Þ
For ﬂow in microtubes, by considering all the directly measurable quantities (inner diameter, channel length, temperature, pressure drop, outlet pressure and mass ﬂow rate) in the determination of the value of friction factor, the relative uncertainty of friction factor is given by:
df
f ^{¼}
s
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2
k
^{2}
1
d
D
_{h}
D
h
2 þ k ^{2}
2
d
L
L
2
2
3
d
T
T
2
þ k
2
4
dD p
p
2 þ k ^{2}
5
^{d}^{p} out
^{p}
out
2 þ k ^{2}
6
d
_
m
_
m
þ k
ð
23 Þ
where the sensitivity coefﬁcients, k _{1} , k _{2} ,
uncertainty of each measurable parameter contributes quantita tively to the uncertainty of friction factor, are listed in Table 3 for both incompressible and compressible ﬂow. The analytical expres sion of these coefﬁcients, reported in Table 3 , can be calculated by Eqs. (22) and (23) for which:
, k _{6} , which show how the
k _{i} ¼ ^{@} ^{f}
x
i
@ x _{i} f
ð
24Þ
and depends on the ﬂow conditions in the experimental test as well on the measurement strategies adopted (i.e. the use of two absolute pressure sensors at the inlet and at the outlet of the microchannel
or the use of one differential pressure sensor with one absolute
pressure sensor and so on). Eq. (23) highlights that the relative uncertainty of the friction factor is more sensitive to the measured variables which are asso
ciated to larger sensitivity coefﬁcients and the largest one domi nates the total relative uncertainty. Speciﬁcally, the accuracy of microchannel diameter measurement has the greatest weight in the determination of the friction factor uncertainty, followed by mass ﬂow rate and pressure (for compressible ﬂows) measure ment, while the accuracy of length and temperature measurement is comparatively less important. This analysis suggests that for accurate calculations of the friction factor the measurement of the channel diameter is of crucial importance: even a small error may induce a large error in the calculated quantity. When compressibility effects become important and for very low mass ﬂow rates (<10 ^{} ^{8} kg/s), uncertainty in ﬂow rate measure ments becomes large; as explained in the previous section, in these cases the ﬂow rate is not directly measured but it is calculated with Eqs. (17)–(20) as a function of temperature, pressure and volume of the system: this increases the inﬂuence of such quantities mea surement on the total uncertainty of the friction factor. It can also be noted from Table 3 that, for incompressible ﬂows and compressible ﬂows with low pressure drop, the uncertainty of friction factor can be directly calculated from the error of each measurement, without knowing the absolute values of the mea sured parameters. On the contrary, for compressible ﬂows with comparatively large pressure drops (which is usually the case in microchannels), the friction factor uncertainty depends on the absolute values of the following experimental parameters:
a ¼ ^{D}^{p} ;
^{p}
out
b ¼ ^{2}^{D} ^{h}
fL
ð
25Þ
where D p is the pressure drop along the whole microchannel, p _{o}_{u}_{t} is the outlet pressure, f is the friction factor and L in the total length of
the microchannel.
For compressible ﬂows different strategies can be used in order
to determine the friction factor depending on the type of pressure
sensors involved in the measurement.
Table 3 Sensitivity coefﬁcients of operative parameters in the calculation of friction factor.
Sensitivity 
Incompressible 
Compressible ﬂow (negligible _{a} ^{a} ) 
Compressible ﬂow (nonnegligible a ^{a} ) Strategy #1 
Compressible ﬂow (inlet and outlet pressure measured) Strategy #2 

coefﬁcients 
ﬂow 

k _{1} (diameter) k _{2} (length) k _{3} (temperature) k _{4} (pressure drop) 
5 
5 
5 + 4 b ln(1 + a ) 
5+4 b ln(1 + a ) 

1 
1 
1 
1 

0 
1 
1 + b ln(1 + a ) 
1+ b 
ln(1 + a ) 

^{1} 
^{1} 
1 þ a 2 þ a ^{} ^{b} _{1} _{þ} _{a} þ b 1 þ a

2 a þ a 
ln ð 1 þ a Þ 
2 þ _{a} _{2} _{þ} _{2} _{a} b þ 2 b 1 þ of D p ) 2

1 a 2 þ 2 a 
ln ð 1 þ a Þ (for p _{i}_{n} instead 

k _{5} (outlet pressure) k _{6} (mass ﬂow rate) 
0 
1 
_{1} þ 
h 
2 þ 
i 
2 _{a} _{2} _{þ} _{2} _{a} ½ 1 þ b ln ð 1 þ a Þ b 

1 
a 
þ b 
2 
ln ð 1 þ a Þ þ 
1 
a þ a 

2 
a 

2 
2 
2+2 b ln (1 + a ) 
2+2 b ln (1 + a ) 
^{a} a = D p / p _{o}_{u}_{t} ; b = 2 D _{h} /( fL ).
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
861
For example, one can measure:
the total pressure drop due to the microtube by using a differ ential pressure sensor together with the determination of the absolute value of the outlet pressure (indicated as Strategy #1 in Table 3 ) or the absolute value of the pressure at the inlet and at the outlet of the microtube (indicated as Strategy #2 in Table 3 ).
Of course, other combinations of pressure sensors are possible. Each different strategy causes some difference in the expression of friction factor and thus changes the sensitivity coefﬁcients of each measurement (Eq. (24)), as shown in Table 3 . In order to give some numerical examples, the inﬂuence of sensitivity coefﬁcients of the six measurable quantities (inner diam eter, microtube length, temperature, total pressure drop, outlet pressure, mass ﬂow rate) on the uncertainty of the friction factor is investigated quantitatively with a set of errorfree data generated from gas ﬂow in a 100 l m microtube with a length of 30 mm. The ﬂow is assumed to be compressible, isothermal and in the laminar regime, and the gas is discharged to the atmosphere. The typical value of uncertainty for each measurement can be chosen by using the data in Table 2 ; we have assumed an uncertainty of ±2% for inner diameter measurement, ±0.3% for length measurement, ±0.25% for temperature measurement, ±0.5% for pressure measurement and ±2% for mass ﬂow rate measurement. Fig. 4 shows the inﬂuence of each measurement on the uncer tainty of friction factor. It can be seen that the accuracy of friction factor is most sensitive to the measurement of inner diameter, which results in an uncertainty of more than 10% on the friction factor. This contribution grows gradually with the increase of Reynolds number. The importance of mass ﬂow rate measurement takes the second place and a 2% deviation results in an error of 4–5.5% in friction factor. Compared with diameter and mass ﬂow rate, the inﬂuence of other quantities (pressure drop, outlet pres sure, length and temperature) on the accuracy of friction factor is small. In practice, the uncertainty evaluation is more complicated than the previous simulation, which based on assumption of constant values for the relative uncertainty of each measured quantity. Dur ing a speciﬁc test round for isothermal ﬂow, four out of the six quantities to be measured (inner diameter, microtube length, tem perature, total pressure drop, outlet pressure, mass ﬂow rate) al most remain constant when the imposed Reynolds number is changed, namely inner diameter, length, temperature and outlet
pressure (here it is assumed that the gas is discharged directly into the atmosphere). On the other hand the value of the pressure drop and the mass ﬂow rate increase greatly in microchannels when the Reynolds number increases. This causes a change in the relative uncertainty of these two quantities, which tends to be larger when the pressure drop and mass ﬂow rates are farther from the full scale (FS) value of the instrument. To further explore this effect, we assume the relative uncer tainty for pressure drop and mass ﬂow rate is ±0.5% of the full scale (FS) reached when the Reynolds number is 2300. Fig. 5 shows the single contribution of the uncertainty of these two parameters to the uncertainty on friction factor. The total fric tion factor uncertainty is calculated based on the changing uncer tainty of pressure drop and mass ﬂow rate, as well as the constant relative uncertainty of inner diameter (±2%), length (±0.3%), outlet pressure (±0.5%) and temperature (±0.25%). It can be seen that for low Reynolds number (<500), or more generally when the mea sured values are far from the full scale, the uncertainty of friction factor becomes very large and sometimes this leads to unreason able results in Microﬂuidics experiments. In order to control the uncertainty during experimental tests over a wide range of Reynolds numbers it becomes mandatory to use a series of devices with different full scale (FS) values. As shown in Fig. 6 , if two pressure meters and two ﬂow rate me ters are used having their FS at Re = 500 and at Re = 2300 when the
Re
Fig. 5. Contribution of the differential pressure sensor and of the mass ﬂow rate sensor having an uncertainty equal to ±0.5% of FS (Re = 2300) on friction factor uncertainty.
Re
Fig. 4. Inﬂuence of each measured parameter with typical uncertainty for a microtube having an inner diameter of 100 _{l} m and a length of 30 mm.
Re
Fig. 6. Effect of the use of two differential pressure sensors and mass ﬂow rate sensors having different FS (Re = 500 and Re = 2300) on friction factor uncertainty.
862
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
Reynolds number becomes lower than 500 the measurement can be switched to the set of devices having their FS at Re = 500 and this can greatly reduce the overall uncertainty of friction factor at lower Reynolds number. By comparing the data of Fig. 6 with those of Fig. 5 it is evident that using more devices having different full scale values it is pos sible to expand the range of the Reynolds number in which the fric tion factor uncertainty stays below ±14%.
Re
Fig. 7. Comparison of the inﬂuence of different strategies for the pressure measurement on the sensitivity coefﬁcients of the inlet, outlet and differential pressure (sensors with a ﬁxed uncertainty (±0.5%)).
As shown in Table 3 , using pressure sensors of identical uncer tainty but different strategies of pressure measurement may result in a different accuracy for friction factor too. In this sense, an ‘‘a priori’’ uncertainty analysis can become an important design tool for microﬂuidics experiences as underlined in [71] . Fig. 7 indicates that if the inlet and outlet pressure are mea sured for the determination of the friction factor for a compressible gas microﬂow (Strategy #2), the sensitivity coefﬁcients for friction factor will be very large at lower Reynolds number compared with the measurement of pressure drop and outlet pressure. It is evident that for Reynolds number larger than 500, it is more advisable to choose an absolute pressure sensor with a higher accuracy for the inlet section in order to reduce the total uncertainty on friction factor. On the contrary, if the pressure drop and outlet pressure are measured (Strategy #1), it is more beneﬁcial to use a more accu rate differential pressure transducer in order to obtain a low total uncertainty. Uncertainty analysis can also be used to check the validity of the results published in the literature: sometimes discrepancies from the predictions of the theory can be traced back to the use of inaccurate instrumentation for the measurements. A list of experimental papers has been examined in order to check the level of uncertainty associated to the measured values of the friction fac tors for liquids and gases. In Table 4 the declared uncertainties on the friction factors re ported by Ferguson et al. [72] for experiments with liquid ﬂows are summarized in order to have a reference for a comparison with the values described in this paper for gas ﬂows. It is interesting to
Table 4 Declared uncertainties of quantities measured in experiments with liquid ﬂows in percentage (from Ferguson et al. [72] ).
Authors 
Geometry 
Pressure (%) 
Mass ﬂow rate (%) 
Re (%) 
Friction factor (%) 

d (%) (circular) 
h, w (%) (rectangular) 
L (%) 

Li et al. [73] Liu and Garimella [74] Wu and Cheng [75] Mala and Li [76] Judy et al. [77] Chen et al. [78] Celata et al. [16] Peng et al. [79,80] 
2 
– 
0.1 
1.5 
2 
– 
– 
– 
– 
3.9 
0.25 
1.01 
10.5 
11.8 

1.83 
– 
0.37 
0.68 
1.69 
– 
– 

2 
– 
0.2 
2 
2 
3 
9.2 

2.5 
2.5–5 
– 
0.25 
– 
– 
– 

– 
– 
– 
2 
5 
10.5 
5.4 

– 
– 
–– 
7 
5 
7 

– 
– 
– 
1.5 
2.5 
8 
10 
Table 5 Declared uncertainties of quantities measured in experiment with gas ﬂows in percentage.
Authors 
Geometry 
Pressure 
Temperature 
Mass ﬂow rate (%) 
Re 
Friction factor (%) 

d (%) (circular) 
h, w (%) (rectangular) 
L (%) 
(%) 
(%) 
(%) 

Ewart et al. [63] 
1.39 
– 
– 
1.5 
0.02 
4.5 (const. volume); 4.2 (drop tracking) 
– 
– 
Ewart et al. [65] Celata et al. [16] 
– 
h : 2.13; w : 0.20 – 
1.06 
0.5 
0.02 
4.5 
– 
– 19 (ﬂow meter);27 (pipette) 
2.48–3.67 
– 
– 
– 
– 
– 

Asako et al. [7] Zohar et al. [17] Hsieh et al. [22] Maurer et al. 
0.13 
– h : 2; w : 1.25 h : 1; w : 1 
0.003 
0.42 
– 
0.98 
– 
12.5 
– 
– 
<1 
<1 
<8.5 
– 
– 

– 
– 
0.7 
0.04 
4.0 
1.46 
1.82 

– 
h : 2; w :1 
1 
2 
– 
2 
– 
– 

[81] 

Morini et al. [18] Araki et al. [82] 
2 
– 
0.3 
0.5 
0.25 
0.5–0.6 
3 
10 
triangle and trapezoid D _{h} : 3.23 – 
0.2 
5.13 
– 
2.2 
– 
10.9 

Turner et al. [83] Tang et al. [15] Colin et al. [64] 
– 
– 
– 
– 
4.9 
4.8 

0.3–2.98 
0.02–0.07 
0.25 
0.17 
2 
4.5 
5.9 

– 
h : 2.23–18.5 w : 
0.2 
– 
– 
– 
– 
– 

0.58–1.4 

Pitakarnnop et al. [67] 
– 
h : 5.3 w : 1.4 
0.2 
0.5 
0.2 
4 (const. volume); 3.1 (drop tracking) 
– 
– 
G.L. Morini et al. / Experimental Thermal and Fluid Science 35 (2011) 849–865
Table 6 Typical values of parameters in experiments with gas ﬂows.
863
Authors 
Geometry and size ( l m) 
Length (mm) Pressure (kPa) 
T (K) 
Mass ﬂow rate (kg/s) 
Re 

Inlet 
Outlet 

Ewart 
et 
al. 
[63] 
Circ. d : 25.2 l m Rect. h: 9.38; w: 492 Circ. d : 30–254 Circ. d : 150 
53 
1.22–12.11 
0.25–2.47 
296.5 
0.02–2 10 ^{} ^{1}^{0} 0.0049–22.5 10 ^{} ^{1}^{0} – 4 10 ^{} ^{6} 1.5–6 10 ^{} ^{9} 
0.0018–2.5 

Ewart et al. [65] Celata et al. [16] Asako et al. [7] 
9.4 
0.060–115.47 0.012–32.65 – 
– 

50–91 
Up to 1000 
100 
– 
0.8–500 

30–50 
300 
190 
300 
1508–2188 

Zohar et al. [17] 
Rect. h: 0.5 and 1; w : 40 
4 
Up to 400 
100 
– 
– 

Hsieh 
et al. [22] 
Rect. 
h: 
50; w : 200 
24 
2.7–64.63 
0.68–13.19 
300 
0.88 10 ^{} ^{8} –40.9 10 ^{} ^{8} 2.6–89.4 

Maurer et al. [81] Morini et al. [18] 
Rect. h: 1.14; w : 200 Circ. d : 133–730 
10 
140–500 
48–100 
296 
6 10 ^{} ^{1}^{2} 5 10 ^{} ^{1}^{0} 
0.001–0.07 

200–1000 
– – – – Up to 300 
– 
– 
– 0.2 10 ^{} ^{1}^{0} 1.0 10 ^{} ^{1}^{0} – 4 10 ^{} ^{1}^{3} 2 10 ^{} ^{9} 8 10 ^{} ^{1}^{4} 2 10 ^{} ^{1}^{1} 
100–5000 

Araki et al. [82] Tang et al. [15] Colin et al. [64] 
Triangular and trapezoidal D _{h} : 3.92–10.3 15 and 25 
100 
295 
0.042–4.19 

Circular and rectangular D _{h} : 10–300 Rect. h: 0.5–4.5; w : 20, 50 
27.5–100 
100 
295 
3–6200 

5 
65–200 
294 
– 

Pitakarnnop et al. [67] Rect. h: 1.88; w : 21.2 
5 
2–50 
298.5 
– 
note that the typical values of the friction factor uncertainties vary from ±5.4% to ±11.8% and, in many papers, the most important uncertainty, related to the inner diameter, was not declared. In these works the ﬂow was adiabatic and in many cases the temper ature of the test ﬂow was not checked. Table 5 reports a collection of declared uncertainties for exper iments with gases. Comparing the data of Table 4 with those of Ta ble 5 it is seen that for gas ﬂows the uncertainty on the friction factor can be larger than ±12% due to the smaller dimensions of the tested microchannels (less than 150 _{l} m generally) and to the larger uncertainties on mass ﬂow rate measurement. In fact, very low ﬂow rates and low values of pressure are used in these exper iments in order to obtain large values of the Knudsen number (large rarefaction effects). By observing the total uncertainties of the friction factor of Ta ble 5 it is interesting to note that in many experimental works the data about the uncertainty analysis are incomplete and it is impos sible to know the real value of the total uncertainty on the exper imental friction factors. The typical values of temperature, pressure and mass ﬂow rate for these works are given in Table 6 , together with the range of Reynolds numbers investigated. It is interesting to highlight that in the experimental works in which the tested Reynolds numbers were larger than 10 only com pressibility effects were evidenced and analysed; on the contrary, in order to study the rarefaction effects on the gas microﬂows the Reynolds number must be less than 10 (very low mass ﬂow rates). In fact, in order to study the gas rarefaction effects small channels (usually less than 20 l m) and low values of pressure (less than 0.1 bar) must be used. Under these conditions the values of the mass ﬂow rate are very low (10 ^{} ^{6} 10 ^{} ^{1}^{4} kg/s) and
Viel mehr als nur Dokumente.
Entdecken, was Scribd alles zu bieten hat, inklusive Bücher und Hörbücher von großen Verlagen.
Jederzeit kündbar.