Sie sind auf Seite 1von 12

Microfluid Nanofluid (2009) 6:763774

DOI 10.1007/s10404-008-0351-z

RESEARCH PAPER

Performance analysis of a folding flow micromixer


Z. Chen M. R. Bown B. OSullivan
J. M. MacInnes R. W. K. Allen M. Mulder
M. Blom R. vant Oever

Received: 7 July 2008 / Accepted: 28 August 2008 / Published online: 2 October 2008
Springer-Verlag 2008

Abstract The performance of a folding flow micromixer the mixer to allow for this uncertainty. Application of the
in the Stokes flow regime is investigated computationally scaling relation to mixer design is highlighted by a dis-
and experimentally. Consistency with a previously derived cussion of the options available for improving the
general scaling relation is demonstrated and the geometric performance of the experimental mixer.
parameters in the scaling relation are determined for this
mixer. Measured data from a second similar mixer are
correctly predicted using the scaling relation, thus showing
that the approach allows quantitative prediction of mixing. 1 Introduction
This paper focuses on the errors associated with such
predictions. Basic errors, expressed as variations in the Over the last several years, many different microchannel-
standard deviation of the concentration profile, were esti- based approaches for mixing two or more liquid streams
mated to be -10% for the computation and -30% for the have been reported. Among these are direct layering of
experiment at the highest values of Peclet number con- the streams (Koch et al. 1998; Veenstra et al. 1999;
sidered. It is shown that this experimental error was mostly Bessoth et al. 1999), using alternating flow of the streams
due to depth averaging of the spectroscopic technique used (Glasgow and Aubry 2003; MacInnes et al. 2005),
for concentration measurement at the high Peclet numbers. placement of ribs to produce a folding flow pattern
However, extra uncertainty is associated with chip fabri- (Stroock et al. 2002; Howell et al. 2005; Bringer et al.
cation tolerances and this was investigated further. 2004) and use of a microchannel network to produce a
Measurements at the outlet of nine different mixer chips of folding flow by splitting and recombining streams (Gray
notionally identical design revealed variations in mixing et al. 1999; Munson and Yager 2004; Park et al. 2004;
of 26%. This variation was attributed to misalignment of Schonfeld et al. 2004; Chen and Meiners 2004; Simonnet
the glass layers determining the geometry of the mixer in and Groisman 2005; Lee and Lee 2005; MacInnes and
the chip. Thus, the combination of measurement error and Allen 2005; MacInnes et al. 2007). These along with
misalignment means predictions of the concentration several other specialised approaches for mixing are
standard deviation for the mixer may get non-uniformity reviewed in recent papers that consider micromixing
wrong by up to 50%. Ensuring a required uniformity, (Ottino and Wiggins 2004; Hessel et al. 2005; Nguyen
however, simply requires adding a few further elements to and Wu 2005). Comparison of the performance of dif-
ferent types of mixers has led MacInnes and Allen (2005)
to suggest that mixers of the folding flow type are espe-
Z. Chen  M. R. Bown  B. OSullivan  J. M. MacInnes (&)  cially promising for critical mixing applications in the
R. W. K. Allen Stokes regime, where inertial effects are negligible but
University of Sheffield, Sheffield, UK rapid and highly uniform mixing is required.
e-mail: j.m.macinnes@sheffield.ac.uk
A particular folding flow mixer geometry is investigated
M. Mulder  M. Blom  R. vant Oever here and its performance is determined using both com-
Micronit Microfluidics, Enschede, The Netherlands putation and experiment. The mixer uses tear-drop

123
764 Microfluid Nanofluid (2009) 6:763774

shaped elements and, as is usual, these are placed in series


to produce repeated folding and layering of the solutions to
reduce diffusion distance and thereby mixing time. These
elements have been incorporated in an experimental chip Odd number
designed for stop-flow reaction kinetics testing in ionic element
liquids (to be reported elsewhere). The plan view of the
chip can be seen in Fig. 1 with the reaction network on the
left-hand side including a primary 13-element mixer
(Mixer B) to produce the uniform reaction mixture and a
7-element mixer used to stop reaction by dilution, with y Even number
x
subsequent sampling off-chip. On the right hand side of the element
chip are two further mixing chains, each of 16 elements. z
These have a lengthened outlet tail designed to allow
Fig. 2 Computational domains used for the odd number and even
clear optical access for measurements. The elements of the number elements. The sections taken as the beginning and end of each
upper mixer (Mixer A) are identical to those of Mixer B element are shaded dark grey. This geometry corresponds to that of
except for tail length. The network and mixer geometry is experimental Mixer B
formed by two outer glass layers with wet-etched channels
interconnected by powder-blasted holes in a middle glass
layer. These three layers when assembled result in the results in high pressures to produce the necessary flow rates
mixing element geometry of Fig. 2 (which shows the to mix reagents rapidly. Thus, it is crucial to predict the
domain geometries used in computations of the short-tail pressure drop required for a given mixer geometry to
mixer, Mixer B). produce a required mixture uniformity in a specified time.
The ionic liquids for which the chip has been designed MacInnes et al. (2007) derived a general scaling rela-
are Newtonian, have viscosities more than 10 times greater tion which, when fitted to a particular mixer geometry,
than that of water and give species diffusivities that are 100 allows mixing time and uniformity to be predicted. The
times or more smaller than in water. Diffusion time scales applicability of the scaling relation was demonstrated by
are about 10 min in the ionic liquids as compared to 10 s in considering a particular F mixer geometry (shown in
water for small molecules diffusing across a 100 lm Fig. 4) for which accurate three-dimensional finite volume
channel. With reaction time scales of interest being of and Monte Carlo computations were made to determine
order 1 min, the long diffusion times in the ionic liquid are the geometric parameters in the relation for this geometry.
clearly unacceptable for accurate reaction kinetics testing. The work presented here applies the scaling relation to an
The high viscosity further rules out approaches based on experimental mixer (Mixer A) having an entirely different
the complex flow motions associated with inertial flow tear-drop geometry. Parameter values are determined
regimes since, with these liquids, the Reynolds number will using a combination of results from finite volume
be below unity in practical microchannel networks and so computations at low Peclet number and experimental
will be in the Stokes flow regime. High viscosity also measurements at high Peclet number. The scaling relation
is used to predict the performance, i.e. pressure drop and
mixture uniformity, as a function of mixing time and
Mixer B Mixer A number of elements in the mixer. The effect of geometry
is also considered by comparison with the previously
characterised F mixer. A second mixer, similar to the first
but with a shorter tail, is also correctly predicted using
y the scaling relation with parameters adjusted to take
account of the change in non-dimensional residence time
with tail length.
As part of the experimental work, nine different chips
x manufactured from the same mask design have been tested
Fig. 1 Plan view of the experimental chip network. On the left side is to assess the effect of geometric variability associated with
a 13-element mixer (Mixer B) feeding a reactor as part of a stop- fabrication tolerances. Measurements are made at the outlet
flow reaction system. On the right hand side of the chip are two 16- of Mixer B for each one of the different chips and these
element mixers. The network is formed from two glass layers of wet-
indicate that considerable variation in mixture uniformity is
etched channels connected by powder-blasted holes in a third glass
layer between these two channel layers. Top channels are shaded associated with misalignment of the glass layers forming
black and bottom channels are shaded grey the mixer geometry.

123
Microfluid Nanofluid (2009) 6:763774 765

2 Experimental details number and the Peclet number must vary by much more
than 10% to give experimentally detectable changes in
The mixer network is constructed from three layers of glass, mixing.
the top and bottom layers having channels formed by iso- Mixing was assessed experimentally by introducing
tropic wet etching to a depth of 75 lm and the middle glass equal flow rates of two ionic liquid streams, one containing
layer having powder blasted connecting holes. Figure 1 0.6 mM of Rhodamine B fluorescent dye. An excitation
shows a plan view of the network with four feed channels filter (546 nm band pass with 10 nm full width at half-
etched into the bottom layer (shaded grey) connected to a maximum transmission) was used in the illumination path
13-element mixer with elements arranged diagonally through the objective lens of an inverted microscope (Zeiss
(Mixer B). This mixer in turn feeds a 1 mm wide reaction Axiovert 100). Light passing to a charge-coupled device
vessel etched into the top layer (shaded black). The feed (CCD) camera was filtered for the fluorescence emission
channels and the tail of each element are 275 lm wide and (600 nm long pass with a 25 nm transition width). Each
the narrower loop portions of the elements are 180 lm CCD pixel value can be related to the average concentra-
wide. The connecting holes taper from about 200 to 100 lm tion over the channel depth at that pixel location in the
diameter going from the bottom surface to the top surface of image. Thus, the profile of depth-average concentration
the 175 lm thick middle glass layer. Micronit Microfluidics across a channel section can be found and from this, the
fabricated the chips with 12 identical chip patterns on each volume average concentration and an approximation to the
glass layer and the layers were bonded together into a standard deviation of concentration over the section can be
wafer before being diced into individual chips. The four determined. A uniform mixture of equal parts of the dyed
feed streams to Mixer B allow a variety of initial chemical and un-dyed solutions was also used to allow variation in
mixtures to be studied, but in the experiments presented channel depth to be taken into account.
here, just two solutions are mixed in equal parts, one Depth-average concentrations were determined from
solution supplied to the upper two feed streams and the recorded pixel values using three different images for each
other to the lower two feed streams. A second mixer with location: an image (I) of the mixing flow, an image (ID)
elements having a longer tail has been incorporated on the with no illumination to correct for background noise and an
chip so that clear optical access to the tail region is possible. image (Io) with the uniform mixture in the channel. The
This second mixer (Mixer A in Fig. 1) has 16 elements, depth average concentration, Cz at each pixel position
which are arranged in a horizontal array rather than diag- normalised by the concentration of the uniform mixture
onally. Computations of another folding flow geometry concentration could then be determined using
(MacInnes and Allen 2005) showed that the mixing was not I  ID
sensitive to the arrangement (horizontal, diagonal or Cz : 1
Io  ID
otherwise) chosen for the elements. Also, the scaling model
of MacInnes et al. (2007) suggests that the longer tail has a At the edges of the channel, the wet-etched surface
negligible effect on the mixing (when account is taken of curves in an approximately quarter-circular section
the difference in mean residence time) since the loop and resulting in the channel depth approaching zero as the
connecting hole geometry are identical. Because of its edge is approached. The ratio in Eq. 1 becomes inaccurate
longer tail, the pressure drop per element will be greater for near the channel edges and the results there were not
Mixer A than for Mixer B, but for the Stokes flow of included in the determination of the mean and standard
interest, this can be taken into account using the developed deviation of concentration. In the case of 275 lm wide
flow numerical solution of MacInnes et al. (2003). channel sections, the data within 8 lm of either edge were
Flow through either Mixer A or Mixer B is produced by discarded and for the 1 mm wide reactor the data within
application of nitrogen gas pressure (6 bar gauge) to an 33 lm of the edges were discarded. This corresponds to
ionic liquid [bmim][N(Tf)2] (1-butyl-3-methylimidazolium overlooking just 2% of the channel section area and this
bis {(trifluoromethyl)sulfonyl}amide) held in supply res- should introduce negligible error. A composite of three
ervoirs. PEEK capillary tubes convey the liquid to the inlet low-magnification images showing the entire 16 elements
holes of the glass chip and from the outlet to waste at of Mixer A is shown in Fig. 3 along with measured
atmospheric pressure. Flow rates produced in the tests are concentration profiles across the outlet section of every
estimated from the known viscosity of the liquid and the fourth element. One-half of the inlet flow at the lower left
channel dimensions. Average velocities calculated from the in the image contains the Rhodamine B. The increasing
estimated flow rates and nominal dimensions are expected number of striation layers from element to element
to have an uncertainty within about 10%. No direct mea- expected for this type of mixer is evident. It is noted that
surement of average velocity (or flow rate) was made since these small-scale variations in concentration are somewhat
the average velocity is needed only to determine Peclet blurred by internal reflection and other effects causing

123
766 Microfluid Nanofluid (2009) 6:763774

k=4 k=8 k = 12 k = 16
Cz
1.0

0.0
0 100 200 0 100 200 0 100 200 0 100 200
Position (m) Position (m) Position (m) Position (m)

k=4 k=8 k = 12 k = 16

Fig. 3 Composite microscope image of fluorescence intensity (bot- The Peclet number based on channel depth (R) and average velocity
tom) for Mixer A and (top) profiles of measured (depth averaged) (V) is Pe = 3.2 9 105
concentration at selected element outlets (k) derived from the images.

fluorescent light emitted at one position to be received by


Table 1 Experimental conditions
the camera from adjacent positions.
Quantitative comparison with the computations and with q 1,500 kg/m3
the scaling relation is dependent on determining the con- l (at 25C) 0.0497 N s/m2
ditions of the experiment, in particular the Peclet number. D (at 25C) 1.54 9 10-11 m2/s
The Peclet number used here is Pe = VR/D where V is Sc = l/qD 2.1 9 106
average velocity in the 275 lm wide tail section of an R (channel depth) 75 9 10-6 m
element, R is the channel depth (75 lm) and D is the dif- sD 365 s
fusivity of the Rhodamine B in the ionic liquid. The
13-element 16-element
average velocity in the tail section is readily determined (diagonal) mixer (horizontal) mixer
and the channel geometry is known with satisfactory
accuracy, but the diffusivity of the Rhodamine B in the V (in 275 lm section) 0.085 m/s 0.0683 m/s
ionic liquid, although expected to be in the region of Pe = VR/D 4.0 9 105 3.2 9 105
10-11 m2/s based on values for similar molecules diffusing Re = qVR/l 0.19 0.15
in similar ionic liquids, needs to be measured. This was Mean residence time 256 ms 489 ms
accomplished by stopping dyed and un-dyed ionic liquid
solutions side by side in a channel, thereby initiating a two-
dimensional transient diffusion process in the section of the
channel. Images were recorded to give depth-average 3 Computational approach
concentration profiles as a function of time until a uniform
mixture was reached. The diffusion equation was then A continuum computational method is used here to deter-
solved numerically for the two-dimensional channel cross- mine the performance of the experimental mixer. The
section shape, beginning with an initial step profile of computations are, however, limited to low Peclet number
concentration, and the results were fitted to the experi- conditions where sufficiently fine grid size can be used to
mental profiles by adjustment of the diffusivity. The control numerical errors in the solution of the species
diffusivity so determined is judged to have an uncertainty equation. The exact equations for steady Stokes flow
of around 5% and the value determined is listed in Table 1 transporting a dilute species can be expressed in terms of
along with other experimental conditions. non-dimensional variables as:

123
Microfluid Nanofluid (2009) 6:763774 767

op o2 ui ouj ouj C 1 o2 C
2 0 : 2
oxi oxj oxj oxj Pe ox2j

The normalising scales used are R for length, V for velocity


and the viscous scale lV/R for pressure. Concentration is
normalised by that in the inlet stream containing the dye so
concentration ranges from zero in the pure undyed solution
to unity in the pure dyed solution. Fig. 4 Computed contours of concentration at the wall surfaces for
These equations are solved using finite volume discret- the first element of the tear-drop mixer geometry (Mixer B) and for
isation to determine the velocity, pressure and species the F mixer geometry (MacInnes et al. 2007). The Peclet number is
1000 in each case. Red corresponds to one of the pure solutions, blue
fields within the mixer (using the Fluent CFD program,
to the other and intermediate shades to mixtures of the two
Version 5). The computation is made element by element
with the solution at the outlet from one element being
transferred as the inlet boundary condition to the next
element downstream. For each element computation, suf- encouraging to note that in the computation, more of the
ficient length of the downstream element is included for the dyed solution flows into the left hand branch when the flow
results at the element outlet section to be unaffected by the divides at the beginning of the second element, which is
outlet boundary condition imposed further downstream. also found to be the case in the experimental result in
The inlet boundary condition for the first element is a lat- Fig. 3.
eral step function in concentration and a uniform velocity
in the direction parallel to the channel. Three computa-
tional domains are used, one for the first element where 4 Mixer scaling relation
inlet conditions are imposed, a second for even-numbered
elements and a third for odd-numbered elements. Figure 2 The experimental and computational results will be used to
shows the odd and even computational domains with the determine the parameters in the scaling relation developed
coordinate system used. (Note that z corresponds to the by MacInnes et al. (2007) for concentration non-uniformity
direction of depth averaging in the experimental mea- as a function of the number of elements in the mixer and
surements.) The sections corresponding to the beginning Peclet number. The scaling relation allows both the overall
and end of the element are shown as shaded sections in consistency of the results to be judged and the use of
the figure. limited results to predict performance for mixer design
A grid cell size of 0.04 times the channel depth proved purposes. The most direct measure of mixture non-unifor-
to be the practical limit with available computer resources mity is the standard deviation of concentration normalised
(AMD Opteron X50 processors with nominal speed of by the concentration of the fully mixed state. This relative
2.2 GHz). This cell size was used uniformly throughout the standard deviation, r, is defined by
domain to produce a mesh of tetrahedral cells. Second v
u 2
order differencing was used for the convective terms in the u C
r t 2  1 3
species equation and computed values of concentration C
standard deviation are expected to have an error of about
1% at a Peclet number of 1,000, the error increasing with where C is concentration and the over bar represents an
Peclet number. This error was estimated from computa- averaging operation (here volume weighted averaging over
tions at Pe = 1,000 made for a series of different grid sizes a flow section is used). For volumespof two solutions
with small enough grid size being reached to judge the combined in the ratio /, r equals 1 / in the initial
exact result and also from related previous computational unmixed state (MacInnes et al. 2005) and approaches zero
studies where error was assessed by comparison with exact as progress is made towards complete uniformity. So equal
solutions (MacInnes 2002). The error scales directly with flows of two solutions, as is the case here, correspond to an
the Peclet number for fixed grid size, so at the highest initial relative standard deviation of r = 1.
Peclet number computed here (10,000) the error is around The standard deviation is reduced from element to
10%. The computations required up to 60 h CPU time at element by diffusion between layers of the solutions.
each Peclet number, depending on the number of elements MacInnes et al. (2007) derived a general relation for decay
computed. Figure 4 (tear-drop mixer) illustrates the char- of standard deviation with number of elements that applies
acter of the species distribution, where computed over the entire range of Peclet number. The relation blends
concentration at the channel wall surfaces for the first high Peclet number scaling with that at low Peclet number.
element of the mixer at Pe = 1,000 is shown. It is At high Peclet number, mixing is delayed by a certain

123
768 Microfluid Nanofluid (2009) 6:763774

number of elements as the smallest length scales of the non-dimensional variables. Three equations are needed:
concentration field are progressively reduced to the size the scaling relation (Eq. 4) provides one equation, the
where diffusion can play a significant role, the delay definition of Peclet number provides a second and the
increasing with Peclet number. Then, the standard devia- third is provided by the Stokes flow pressure drop relation
tion of concentration decreases exponentially at a rate that for one element (Dpe = j/se where j is the pressure drop
is independent of Peclet number and dependent only on the coefficient and Dpe is the element pressure drop norma-
continual folding produced by flow through the element lised using the viscous scale lV/R). The design constrains
geometry. Three geometric parameters a, k and c are are element size, mixing time and mixture uniformity.
needed to quantify, respectively, the dissipation rate of These three constraints with the three equations determine
standard deviation, the effectiveness of the elements at the number of elements required and the pressure drop
folding the solutions and the delay to the beginning of the and Peclet number of operation. The three design
exponential decay. At low Peclet number the decay rate is equations in terms of the non-dimensional design
proportional to the inverse of the Peclet number and a constraints are:
fourth geometric parameter (b) is needed to quantify the    
s
transition between the two Peclet number regimes. The r exp ase k 2 kr  c max0; lnkse kr =s
b se k r
relation derived for relative standard deviation is
    5
1
r exp ase k 2 k  c maxlnkPe; 0 4 jkr2
b Pe Dp 6
s
where k is the number of elements the flow has passed
k r se
through, Pe is the Peclet number as previously defined and Pe 7
s
se is mean residence time in a single element non-dimen-
sionalised by the convective time scale R/V. The The design criteria are imposed by specification of the
parameters in Eq. 4 are functions only of element geometry required mixture uniformity (r) and mixing time (sr). The
in the Stokes regime considered here. It should be noted mixing time is taken as the total mean residence time of the
that beyond the Stokes regime they become functions of mixer, sr = krseR/V. This mixing time is present in the
Reynolds number as well. above equations as the non-dimensional time s = sr/sD,
The computations consider just the short-tail mixing where sD = R2/D is the diffusion time scale based on
element of Mixer B whereas the most detailed experiments channel size, R. Note that the non-dimensional mean resi-
are made for the long-tail mixing element of Mixer A. As dence time of the mixing element (se) is a purely geometric
has been mentioned, since the part of the geometry parameter and is a known constant for a particular element
responsible for folding is identical in the two geometries, it geometry (values given in Table 2). With r and s specified,
is possible to translate results from one to the other in a the number of mixing elements (kr) is uniquely determined
straightforward manner, taking account of the different by solution of Eq. 5. The non-dimensional pressure drop
values of se and pressure drop coefficient (j) for the two across the mixer (Dp) is then determined by Eq. 6 and
mixers. The product sek determines the scale reduction per the Peclet number by Eq. 7 (note that the physical pres-
element at large Peclet number and the product ase deter- sure drop has been normalised by the stress scale lD/R2 to
mines dissipation of concentration variance over the form Dp).
element. Lengthening the tail does not affect scale at high
Peclet number so the former must remain constant for the
two similar mixers. The later should remain approximately
constant since strongest dissipation occurs in the loop
Table 2 Parameter values in the scaling relation (Eq. 4 or 5) for tear-
where the layers are stretched and diffusion distance is drop Mixers A and B and the F mixer (MacInnes et al. 2007)
reduced, the increase in residence time due to lengthening
Tear-drop mixer F mixer
the tail being countered by a reduction in average dissi-
pation rate and hence in a. Mixer A (long tail) Mixer B (short tail)
Once the scaling relation for a particular mixing element se 26.9 21.6 11
geometry is accurately established, one can reliably design a 21.0 26.2 28
any mixer based on that same element geometry. Questions
k 0.000353 0.000439 0.00162
of optimum number of elements, the effect of size and of
b 3.22 3.22 1.63
performance relative to other element geometries can be
c 0.564 0.564 0.776
decided based on the scaling relation. MacInnes et al.
j 9,610 5,940 1,190
(2007) develop the necessary design relations in terms of

123
Microfluid Nanofluid (2009) 6:763774 769

5 Results element, i.e. three elements are required to develop small


enough layer thickness for significant mixing to begin. The
Computations at four different values of the Peclet number actual data in fact show a gradual transition to the final
up to 10,000 for Mixer B and experimental measurements decay rate, with initially slow decay rather than no decay
at Pe = 320,000 for Mixer A were made. In all cases, the for the first three elements and the final constant decay rate
Reynolds number was much lower than unity, so Stokes being reached by about the ninth element. The fact that
flow prevailed. The experimental condition used corre- standard deviation values in this transition region are rather
sponds to typical operation of the mixer in the reaction erratic and unexpectedly low compared to the computation
device and, indeed, to conditions commonly of interest in at Pe = 10,000, suggests that error due to depth averaging
microchannel networks. Finite volume computations at may be especially important during transition. Clearly,
Peclet numbers greater than 10,000 produce unacceptably though, the scaling relation gives an excellent account of
large errors and experiments at lower Peclet number could all of the data, both computational and experimental, in the
not be controlled accurately using the available apparatus. final decay region, where mixing performance will be
Rather than a direct comparison of computational predic- judged.
tion with experimental measurement, agreement of the Of course, the scaling relation is only as good as the
computation and experiment will be judged indirectly quality of the data on which it is based. As mentioned,
based on how well the scaling relation fits all of the results the computations have a numerical error of about 10% at
together. Pe = 10,000, which decreases with Peclet number, and
Figure 5 shows the results with open symbols for the this error will give a lower standard deviation than is
computations and solid symbols for the experiment. The actually present. A second important error is that caused
lines are the scaling relation (Eq. 4) fit to all data in the by the depth averaging of the experimental measurement
final stage of decay, with the parameters listed in Table 2 of concentration (e.g. Schonfeld et al. 2004; Nguyen and
giving the least-squared-error best fit. It is the final expo- Wu 2005). With variation of concentration in the direc-
nential stage of decay that determines mixer performance tion perpendicular to the image not captured by the
and to which the scaling relation applies. The starting measurement, striation layers that are not perfectly per-
element used to determine the best fit was selected based pendicular to the image plane result in falsely low
on where the error in fitting the final decay settled to an standard deviation values. The magnitude of this error can
approximately steady level, the error increasing as more be estimated using the computational results. Applying
elements from outside the exponential region were inclu- depth averaging to the computational results reduces
ded. This was determined individually for each Peclet values of r by between 0 and 0.6% (depending on which
number data set (computational or experimental). In the element is considered) for Pe = 100, by 2.26.9% for
figure, a significant initial delay at high Peclet number can Pe = 1,000 and by 5.312.1% for Pe = 10,000. This
be observed, as expected, with little mixing occurring increasing effect with Peclet number suggests the reduc-
before the exponential decay begins. The origin of expo- tion at the high Peclet number of the experiment might be
nential decay for Pe = 320,000 is at about the third around 30%. Blurring of image intensity mentioned ear-
lier will introduce somewhat further error also reducing
1.00 the measured deviation.
Thus, while the fitted scaling relation represents the data
well the error present in these data may be allowing a
fortuitously good match with the scaling relation. To test
this, the data at Pe = 10,000 and 320,000 were increased to
Pe = 320,000 account for the errors estimated above (by 10 and 30%,
0.10
respectively) and the scaling relation was refitted to the
adjusted data. The resulting scaling relation gave an
10,000
1,000
equally good fit to the adjusted data as that shown in Fig. 5
100
320 for the original data. Thus, the general scaling relation does
seem to allow an excellent account of the mixing behaviour
0.01
0 5 10 15 20 of the tear-drop mixer considered here, just as was found to
k be the case for the F mixer investigated in MacInnes et al.
(2007). One must bear in mind that since the parameters for
Fig. 5 Comparison of r results from computation (open symbols) and
measurement of Mixer A (solid symbols) with the scaling relation
the tear-drop mixer in Table 2 fit the original unadjusted
fitting all results. Parameter values used in the scaling relation (Eq. 4) data the corresponding scaling relation will under-predict
are listed in Table 2 the mixture standard deviation.

123
770 Microfluid Nanofluid (2009) 6:763774

6 Effect of misalignment estimated for Wafer A to be about 12 lm based on com-


parison of the fabricated channels and the design pattern. In
In addition to the element-by-element measurements for the case of Wafer B, alignment marks on individual glass
Mixer A, further measurements were made at the outlet layers could be used to determine within a couple of
from Mixer B when it was realised that different chips of microns the actual displacement of the top and bottom
identical design gave noticeably different mixing perfor- layers relative to the middle layer. Since chips from the
mance. Chips were available from three different wafers same wafer are bonded before dicing, misalignment of each
(A, B and C) and from different positions on the wafers. chip is determined by the misalignment of the glass layers
Figure 6 shows the fluorescence images of the reactor and the position of the particular chip on the wafer. The
region (i.e. the outlet from Mixer B) and the corresponding misalignment of the glass layers was determined in terms
measured values of relative standard deviation in the of a rotation angle about the wafer centre followed by
reactor produced by nine different chips of identical design translation. The rotation and translation were determined
taken from Wafers A, B and C. The images are arranged in by best fit to the measured offset of the alignment marks.
increasing order of r and one can see the progression quite From this information, the displacement of the top and
clearly from a relatively low non-uniformity of 6.9% to bottom layers relative to the middle layer can be worked
relatively high non-uniformity of 25%. Interestingly, there out for the centre of Mixer B in the case of each chip from
is not necessarily a consistent relationship between uni- Wafer B. The values determined are given in Table 3 and
formity for a chip and the wafer from which it was taken. one can note that these chips have been better aligned than
In the case of the chips from Wafer B, several give the the one from Wafer A, i.e. to within around 5 lm as
lowest levels of non-uniformity while another gives rela- compared with 12 lm. Though not measured, the chips
tively high non-uniformity. from Wafer C are expected to have similar levels of mis-
Misalignment of the glass layers during fabrication is alignment to those of Wafer B.
the most obvious cause of the observed variability. The Returning to Fig. 6, one does find a similarity in the
misalignment of the top and bottom layers could be pattern of dye distribution across the reactor for chips from

Fig. 6 Fluorescence intensity


images showing the mixture
non-uniformity in the reactor
region (outlet from Mixer B) at
Pe = 4 9 105. The wafer from
which each chip has come is
indicated along with the
measured value of r. The
images are arranged in order of
increasing relative deviation
going from left to right and from
Wafer B #10 6.9% Wafer B #8 8.0% Wafer B #11 8.4%
top to bottom

Wafer C #10 9.5% Wafer C #11 10.1% Wafer C #3 10.8%

Wafer C #2 10.9% Wafer B #4 11.8% Wafer A #2 25.0 %

123
Microfluid Nanofluid (2009) 6:763774 771

the same wafer, e.g. those from Wafer B tend to have some offset, while for Case S15, the mixing is improved by the
streaks of high dye concentration to the left of the reactor offset. Relative standard deviation has been altered by
centre, those from Wafer C a sharp high concentration around 50% of the value reached when the channels are
streak just near the reactor centre. This might be expected perfectly aligned. Further computations for these two cases
since all are affected by the same offset rotation and at lower Peclet numbers suggest also that the deviation in
translation. However, quite large variations in mixing mixing due to offset increases with Peclet number. The
evidently result from apparently subtle differences in offset eight different experimental chips from Wafers B and C are
even for chips from the same wafer. For example, chip #4 found to have a 26% variation in mixing and this appears
from Wafer B gives a r value almost twice that of chip #10 to be consistent with the computational result, taking
from the same wafer and yet the offsets are similar in account of the increased variation due to larger Peclet
magnitude (Table 3). Chip #8 has the largest offsets of the number and the decreased variation due to the smaller
four tested, yet the value of r produced is among the offset magnitude for those wafers (of around 5 lm) com-
lowest. pared to that used in the computations. Therefore,
To quantify the effect of misalignment, two further misalignment appears to explain the observed variation in
computations with intentional misalignment were made for mixing for the different chips. For the fabrication toler-
Pe = 1,000. In the first, the top layer of channels were ances maintained for Wafers B and C, two or three extra
offset in the x direction (coordinate system of Fig. 2) rel- elements would need to be added to ensure the required
ative to the middle layer and the bottom layer of channels uniformity is achieved, and perhaps four or five extra for
were offset in the y direction relative to the middle layer, the larger fabrication tolerances of Wafer A.
each by 15 lm (similar to the offset magnitude for Wafer
A). This misalignment moves the outlet of each connecting
hole to one side of the receiving channel (Case R15). In the 7 Application of the scaling relation
second computation, the offset is such that the inlet of each
connecting hole is moved to one side of the supplying At the highest Peclet number, which is most typical of
channel (Case S15). The resulting change in decay of rel- conditions of interest, the standard deviation of concen-
ative standard deviation for these two cases is shown in tration remains around 8% at the outlet of the mixers
Fig. 7. Case R15 worsens the mixing compared with no (Fig. 5 for Mixer A, Fig. 6 for Mixer B). Arguably, for
many applications good uniformity would correspond to
Table 3 Offset coordinates of the centre of Mixer B for chips from r = 1% or lower and thus the performance of the mixers
Wafer B studied here would be unacceptable. The design equations
Chip Top relative to middle Bottom relative to middle
(Eqs. 57) will now be used to assess the ways in which the
mixer could be improved.
dx (lm) dy (lm) dx (lm) dy (lm) First, exactly the same element geometry could be used
#4 -2.9 0.5 -2.5 5.9 but with more elements or operating at a lower flow rate.
#8 -5.7 -3.7 -6.4 0.1 To determine the required adjustments, Eq. 5 can be used
#10 -2.9 -3.7 -2.5 0.1 to develop a general performance plot, i.e. the mixture
#11 -1.5 -3.7 -0.6 0.1 uniformity as a function of non-dimensional mixing time
for mixers having different numbers of elements. For the
Mixer B element (short tail), the plot of Fig. 8 results for
1 mixers differing only in the number of elements, k. For
mixing of Rhodamine B in ionic liquid as in the experi-
Perfect alignment
ments, the channel diffusion time is sD = 365 s. If a
0.1 mixing time of 0.365 s is required, for example, s must be
Case R15 10-3. From the figure, the current 13 element mixer gives r
of about 8% for s = 10-3, and one can see that this could
Case S15
0.01
be reduced to below 1% for the same mixing time by
increasing the number of elements to about 22. However,
Eq. 6 shows that this approximate doubling in the number
Pe = 1000
of elements would be accompanied by a quadrupling of the
0.001
0 3 6 9 12 15 18 21 pressure drop across the mixer. Thus, the 2 bar pressure
k drop required with 13 elements would increase to about
Fig. 7 Computed effect of misalignment on r at Pe = 1,000 for two 8 bar to reduce standard deviation to 1%. Alternatively, the
example offsets, Cases S15 and R15 existing 13-element mixer could be operated at a larger

123
772 Microfluid Nanofluid (2009) 6:763774

1.E+00
Tear Drop Mixer
In practice, one would like to design a mixer in a direct
k =5 (short tail) manner. The problem reduces to seeking the best mixer
9
for a particular application. Two mixing criteria are asso-
1.E-01 ciated with a given application: the mixture uniformity to
13
be produced by the mixer and the time within which that
17 uniformity must be produced. Thus, to determine the mixer
1.E-02
21 design requires r and sr to be decided. A third parameter
must be decided on practical grounds and that is the size of
25
the mixer, i.e. R. Once R, r and sr are decided, there is one
1.E-03
1.E-04 1.E-03 1.E-02 1.E-01 1.E+00 further choice and that is the element geometry to be used.
For, in principle, any particular folding flow mixing
geometry can be operated such that the three criteria are
Fig. 8 Predicted performance of the tear-drop mixer. Curves are the
scaling relation (Eq. 4) with the values listed in Table 2 for Mixer B. satisfied. One simply solves the design equations (Eqs. 5
Symbols show measured values at the outlet of Mixer B for the nine 7) for kr, Dp and Pe, thereby determining the required
different experimental chips from Wafers A, B and C number of elements for that mixer and the conditions of
operation. However, each element geometry will lead to a
value of mixing time. Increasing s to 0.6 reduces r to different operating pressure drop and a different chip
below 1% according to Fig. 8. However, the mixing time is footprint area for the same size R. The pressure drop and
then very long (3.6 min) and similar to that for a simple chip area occupied by the mixer are closely related since,
channel, i.e. the channel diffusion time. Lengthening the for fixed R, each is approximately proportional to the
mixing time reduces the Peclet number (Eq. 7) which overall flow path length. In nearly all applications, mini-
corresponds to increasing the decay rate of r (Fig. 5). mising either the mixer pressure drop or mixer footprint
The data from the nine different chips at the outlet from area or both will be important. One can thus identify a
Mixer B is also plotted in Fig. 8 and the scatter due to theoretical best mixer to be the one giving minimum
misalignment is evident. The scaling relation determined pressure drop.
by computation and Mixer A measurements passes through Identifying the best mixer geometry would be an
the middle of the data, which represents a successful pre- immense task requiring investigation of all possible geo-
diction (albeit a somewhat modest one since the geometry metric variations of all possible mixing element types (F,
is so close to the one used to determine parameter values of tear-drop, ribbed, etc.). It is not clear that such extreme
the scaling relation). The effect of misalignment may be optimisation would ever be advantageous. Gradual
thought of as effectively increasing or decreasing the improvement of mixer geometry accepting the modest
number of elements in the mixer. The highest value of r is penalties in pressure drop and chip footprint, because the
produced by the chip from Wafer A and has in effect design is not absolutely optimum, is likely to be the way
reduced the number of elements by 5 (from 13 to 8). For ahead. One obvious shortcoming of the tear-drop mixer
the data from Wafers B and C chips, the effective number considered is its large aspect ratio. This is a result of the
of elements is only changed by one or two elements. wet-etching employed to create the channels. The large
A further possibility for reducing r is to reduce the size channel width to depth ratio associated with isotropic
of the mixer. To see the effect of size, which is represented wet-etching of glass means that the diffusion distance is
by the channel depth, R, all non-dimensional quantities relatively large compared to the channel depth on which
depending on R must be expanded. One finds that, for a pressure drop directly depends. The large aspect ratio,
given number of elements and for a constant mixing time, perhaps combined with the fact that the connecting holes
pressure drop remains constant as size is decreased but the are small in relation to the channel width, seems also to
Peclet number decreases in proportion to R2. The reduction reduce the effectiveness of the folding produced by the
in Peclet number with size again serves to increase the geometry. In Fig. 4, the uneven bulk distribution of dyed
decay rate of r, but the effect is modest until quite low liquid between the two branches of the second element may
levels of Peclet number are reached. Until one reduces the be caused by these factors. It would be possible to produce
Peclet number to around 1000, corresponding to a reduc- a more compact version of a tear-drop mixer that would
tion in channel depth from 75 lm down to about 4 lm, the give much improved folding efficiency. To illustrate this
improvement remains small. Such small channel size may prospect, the present tear-drop mixer can be compared with
be impractical in relation to achieving required fabrication the more compact F mixer design (Fig. 4; Table 2) char-
tolerances and avoiding blockages. In addition, as channel acterised in MacInnes et al. (2007). Solution of the design
size is decreased, the flow rate decreases in proportion to equations for an assumed mixture uniformity requirement
R3 which may or may not be desirable. of r = 1% produces the plots of Figs. 9 and 10, for

123
Microfluid Nanofluid (2009) 6:763774 773

1.E+12 distribution at the channel wall in the first couple of ele-


ments for the two mixers is shown in Fig. 4. From the
1.E+10 figure, it is clear that the tear-drop element is far less
p compact than the F element, resulting in the far larger value
1.E+08 of non-dimensional residence time listed for the former in
Tear-drop Mixer
Table 2. The parameter a is slightly smaller and k is much
1.E+06
smaller for the tear-drop mixer. As mentioned, k is related
= 0.01
F Mixer to the effectiveness of the folding, i.e. the increase in
mixing interface, produced by passage through each ele-
1.E+04
1.E-04 1.E-03 1.E-02 1.E-01 1.E+00 ment; a represents the effectiveness of dissipation of

concentration non-uniformity associated with the folded
Fig. 9 Non-dimensional pressure drop as a function of the reduction concentration field. Thus, both effects are greater in the F
in mixing time required to give r = 1% mixer element considered, with folding being far more
effective (k = 0.00162 as compared to 0.000439) and this
lesser folding in the tear-drop is perhaps evident in the wall
25
concentration plots of Fig. 4. Other parameter differences
20
to note are the higher values of b and friction coefficient, j,
Tear-drop Mixer for the tear-drop. The latter is an obvious result of the
15 longer flow path of this tear-drop design. The parameter b
k is a factor giving effective diffusion distance relative to
10 channel scale R for the low Peclet number limit. Since
F Mixer channel depth has been used for both mixers, the larger
5
= 0.01
value of b for the tear-drop is simply taking account of the
0
greater channel width to depth ratio in that case. Thus,
1.E-04 1.E-03 1.E-02 1.E-01 1.E+00 differences in the parameter values can be related to dif-

ferences in performance and these could help to identify
Fig. 10 The number of mixing elements required as a function of the shortcomings of a particular element geometry and, hence,
reduction in mixing time, again for r = 1% guide development towards optimum geometries.

non-dimensional pressure drop and number of elements,


respectively, as a function of non-dimensional mixing time. 8 Conclusion
For a given liquid system and channel size, the normalising
scales for pressure and mixing time are fixed and the same A folding flow mixer formed from a chain of 16 tear-drop
for both element geometries. Thus, since the pressure elements (Mixer A) has been investigated in the Stokes
required at any given mixing time can be seen from the flow regime, both computationally and experimentally,
figure to be the greatest for the tear-drop mixer considered, with the aims of establishing the characteristics of this
one can conclude that this F mixer element is more geometry and testing the scaling relation developed by
effective than this tear-drop mixer. In fact, over the entire MacInnes et al. (2007). The results are found to be fully
range of mixing time spanned by the plot, the F mixer consistent with the scaling relation and the approximate
element requires less than 10% of the pressure drop needed performance of the mixer has thus been determined to
for the tear-drop mixer. In addition, considerably fewer within the error of the experiment and computations.
elements are required to achieve the same mixing (Fig. 10), The study also reveals a significant effect on mixing
which demonstrates the relation between pressure drop and from small misalignment of the glass layers that form the
chip footprint. It must be emphasised that this result does mixer geometry. A layer offset of 5 lm (1.5% of channel
not show that the F mixer is inherently better than the tear- width) is found to produce up to about a 26% variation in
drop mixer, only that this particular version of an F mixer the standard deviation of measurements of mixture uni-
is better than this particular version of a tear-drop mixer. formity. The offset can lead to improved or worsened
Indeed, a tear-drop mixer of a more compact design was mixing depending on the precise offsets of the layers. This
considered computationally in MacInnes and Allen (2005) variation can, however, easily be taken into account by the
that would outperform the F mixer of Figs. 4, 9 and 10. addition of a few extra elements, the number added
It is interesting to compare the parameter values in the depending on the fabrication tolerances maintained.
scaling relation determined for the two different mixers. The scaling relation determined for Mixer A is used
These values are listed in Table 2 and the concentration to predict the performance of a second similar mixer

123
774 Microfluid Nanofluid (2009) 6:763774

(Mixer B) incorporated in a reaction test device. Mea- Koch M, Chatelain D, Evans AGR, Brunnschweiler A (1998) Two
surements for nine different mixer chips of identical simple micromixers based on silicon. J Micromech Microeng
8:123126
design show a range of relative standard deviation from Lee SW, Lee SS (2005) Microfabrication of the split and recombi-
7 up to 12% for this mixer, which is consistent with the nation micromixer and the effect of its cross-sectional rotation.
predicted value of 8%. Reducing this relatively high non- Proceedings 3rd international conference on microchannels and
uniformity is considered using the design equations for minichannels, June 1315, Toronto, Canada, ICMM200575187
Munson MS, Yager P (2004) Simple quantitative optical method for
the mixer to quantify the potential improvements. Mix- monitoring the extent of mixing applied to a novel microfluidic
ture uniformity is improved either by increasing the mixer. Anal Chim Acta 507:6371
number of elements, which greatly increases pressure MacInnes JM (2002) Computation of reacting electrokinetic flow in
drop, or by accepting a much longer mixing time. The microchannel geometries. Chem Eng Sci 57:45394558
MacInnes JM, Allen RWK (2005) Mixing strategies for flow in
prospect of improving the performance of a mixer by use microchannel devices. Proceedings 7th world congress of
of more efficient element geometry is also highlighted by chemical engineering, July 1115, Glasgow, UK
comparison with results for an F mixer element charac- MacInnes JM, Du X, Allen RWK (2003) Prediction of electrokinetic
terised previously. and pressure flow in a microchannel T-junction. Phys Fluids
15(7):19922005
MacInnes JM, Chen Z, Allen RWK (2005) Investigation of alternat-
ing-flow mixing in microchannels. Chem Eng Sci 60:34533467
MacInnes JM, Vikhansky A, Allen RWK (2007) Numerical charac-
References terisation of folding flow microchannel mixers. Chem Eng Sci
62:27182727
Bringer MR, Gerdts CJ, Song H, Tice JD, Ismagilov RF (2004) Nguyen N-T, Wu Z (2005) Micromixersa review. J Micromech
Microfluidic systems for chemical kinetics that rely on chaotic Microeng 15:R1R16
mixing in droplets. Phil Trans R Soc Lond A 362:10871104 Ottino JM, Wiggins S (2004) Foundations of chaotic mixing 1. Phil
Bessoth FG, deMello AJ, Manz A (1999) Microstructure for efficient Trans R Soc Lond A 362:937970
continuous flow mixing. Anal Commun 36:213215 Park S-J, Kim JK, Park J, Chung S, Chung C, Chang JK (2004) Rapid
Chen H, Meiners J-C (2004) Topologic mixing on a microfluidic chip. three-dimensional passive rotation micromixer using the breakup
Appl Phys Lett 84(12):21932195 process. J Micromech Microeng 14:614
Glasgow I, Aubry N (2003) Enhancement of microfluidic mixing Schonfeld F, Hessel V, Hofmann C (2004) An optimised split-and-
using time pulsing. Lab Chip 3:114120 recombine micro-mixer with uniform chaotic mixing. Lab Chip
Gray BL, Jaeggi D, Mourlas NJ, van Drieenhuizen BP, Williams KR, 4:6569
Maluf NI, Kovacs GTA (1999) Novel interconnection technol- Simonnet C, Groisman A (2005) Chaotic mixing in a steady flow in a
ogies for integrated microfluidic systems. Sensors and Actuators microchannel. Phys Rev Lett 94:134501
77:5765 Stroock AD, Dertinger SKW, Ajdari A, Mezic I, Stone HA,
Hessel V, Lowe H, Schonfeld F (2005) Micromixersa review on Whitesides GM (2002) Chaotic mixer for microchannels.
passive and active mixing principles. Chem Eng Sci 60:2479 Science 295:647651
2501 Veenstra TT, Lammerink TSJ, Elwenspoek MC, van den Berg A
Howell PB, Mott DR, Fertig S, Kaplan CR, Golden JP, Oran ES, (1999) Characterization method for a new diffusion mixer
Ligler FS (2005) A microfluidic mixer with grooves placed on applicable in micro flow injection analysis systems. J Micromech
the top and bottom of the channel. Lab Chip 5:524530 Microeng 9:199202

123

Das könnte Ihnen auch gefallen