1-s2.0-s0009250914006198-main
TRANSCRIPT
-
8/10/2019 1-s2.0-S0009250914006198-main
1/10
Numerical evaluation of stresses acting on particles in high-pressure
microsystems using a Reynolds stress model
S. Beinert n, T. Gothsch, A. Kwade
Institute for Particle Technology, TU Braunschweig, Volkmaroder Strasse 5, 38104 Braunschweig, Germany
H I G H L I G H T S
We investigate the multiphase ow elds of different micro-channels by CFD.
A new model to calculate acting stresses on particles was used. Based on the Reynolds stress tensor and a stationary particle tracking a quantication of acting stresses was possible. Effect of steam and cavitation was reproduced. Peak stress seems to be the leading breakage mechanism in turbulent microchannels.
a r t i c l e i n f o
Article history:
Received 28 August 2014
Received in revised form
17 October 2014
Accepted 22 October 2014
Available online 30 October 2014
Keywords:
Simulations
Multiphase
Reynolds-stress model
Stress analysis
Dispersing
a b s t r a c t
In the present study a model was developed which enables to calculate the stress acting on particles
induced by the uid ow in microchannels. Computational uid dynamics (CFD) was used to simulate
the ow eld in ve different microchannel geometries. With the help of aReynolds-stress model and a
stationary particle tracking a quantication of stresses in different geometries at varying pressure
differences is possible. Furthermore, the effect of cavitation which occurs if the uid expands to ambient
pressure was investigated. The mass ow rates determined by the simulations are in good agreement
with the ones determined experimentally. Except for the z-channel the computed stresses are in good
agreement with previously conducted dispersion experiments. Additionally, the computed uid stresswas compared with the calculated Kolmogorov length scale to validate the results.
& 2014 Elsevier Ltd. All rights reserved.
1. Introduction
In the last couple of years the impact and inuence of
numerical methods for process design and optimisation largely
increased. Simulations are used to get a better understanding
of different processes, for the prediction of the results or as a
additional tool to investigate acting mechanisms. Due to the
increased computational power new, more detailed simulationscan be conducted which improved the understanding of unit
processes. Especially, if experimental methods are not applicable
or if they are expensive (in time and costs) simulations become
more and more the method of choice. In this work simulations
were used as they are (often) faster, reliable and enable informa-
tion which cannot be fetched out of experimental investigations.
They were used to compute different types of microchannel
geometries and to identify the dominant stress mechanism for
the breakage of agglomerates or aggregates (in the following
named particles).
The usage of microchannels in chemical and pharmaceutical
engineering offers a couple of advantages, like high energy input
per volume, high efciency, narrow residence time distributions
and reproducible results. As nanoparticles have become more and
more important for industrial applications the microchannels areused to produce particles with a dened size and a dened,
narrow size distribution. As the geometry of these devices is very
exible investigations to optimise the geometry are required.
The number of possible designs is extremely high and, hence,
simulations are required to get a understanding of the leading
breakup mechanism. If the mechanism for the breakage of the
particles in a microchannel is known the geometry can be adapted
in a way which supports this mechanism.
Wengeler and Nirschl (2007)andWengeler (2007)investigated
the dispersion behaviour in nozzles with a circular cross-sectional
area. They used a stress model with four different kinds of stresses
Contents lists available atScienceDirect
journal homepage: www.elsevier.com/locate/ces
Chemical Engineering Science
http://dx.doi.org/10.1016/j.ces.2014.10.042
0009-2509/&2014 Elsevier Ltd. All rights reserved.
n Corresponding author. Tel.: 49 531 391 9621; fax: 49 531 391 9633.
E-mail address:[email protected](S. Beinert).
Chemical Engineering Science 123 (2015) 197206
http://www.sciencedirect.com/science/journal/00092509http://www.elsevier.com/locate/ceshttp://dx.doi.org/10.1016/j.ces.2014.10.042mailto:[email protected]://dx.doi.org/10.1016/j.ces.2014.10.042http://dx.doi.org/10.1016/j.ces.2014.10.042http://dx.doi.org/10.1016/j.ces.2014.10.042http://dx.doi.org/10.1016/j.ces.2014.10.042mailto:[email protected]://crossmark.crossref.org/dialog/?doi=10.1016/j.ces.2014.10.042&domain=pdfhttp://crossmark.crossref.org/dialog/?doi=10.1016/j.ces.2014.10.042&domain=pdfhttp://crossmark.crossref.org/dialog/?doi=10.1016/j.ces.2014.10.042&domain=pdfhttp://dx.doi.org/10.1016/j.ces.2014.10.042http://dx.doi.org/10.1016/j.ces.2014.10.042http://dx.doi.org/10.1016/j.ces.2014.10.042http://www.elsevier.com/locate/ceshttp://www.sciencedirect.com/science/journal/00092509 -
8/10/2019 1-s2.0-S0009250914006198-main
2/10
(compressive, shear, tensile and turbulent stress) for a two-
dimensional simulation to evaluate the acting stress. This model
was enhanced to three dimensions and extended with a stationary
particle tracking to enable a quantication of different microchan-
nel geometries (see Beinert et al., 2012). Furthermore, in these
previous published results a k model was used to calculate the
ow elds and compute the stresses acting on the particles.
As the dispersion of solid particles in microchannels is a
relatively new process and not yet common in industrial applica-tion there is not much research done on it. Many research projects
were done in the eld of (laminar) mixing and chemical reactions
in micro-devices and particle laden gas ows. The laminar mixing
of different species and a possible reaction is based on the
diffusion process and the interface which is very large in such
micro-devices. Particle laden uid ows, especially turbulent
ows, are very common for diesel injection, soliduid transpor-
tation or uidized beds.
1.1. Mixing and chemical reactions in micro- and macro-devices
Among others, micro-devices are used for continuous mixing
and/or for continuous chemical reactions. Due to the fact that
laminar ows are volitional only low uid velocities, low forcesand low pressures occur. Therefore, established experimental and
numerical methods for laminar ow can be used to characterise
and improve the geometry of the micro-devices. Pennella et al.
(2012) investigated a new designed micromixer by the help of
particle image velocimetry (PIV) and computational uid
dynamics (CFD). They showed that the current methods can be
used to largely improve the geometry in terms of mixing quality,
costs and energy input. On the other hand the micro-scale of
reactive ows was investigated. For example Porta et al. (2012)
looked at the global reaction rate and the spacetime distributions
in dependency of dimensionless quantities like Peclet or Dam-
koehler number.
In the work ofMakowski et al. (2012) large eddy simulations
(LES) were used to calculate the mixing process of particles in arectangular channel reactor. The results were compared with
different simulations and experimental data. They showed that
for high Reynolds numbers and low concentration the k model
offers similar results. With decreasing Reynolds numberand
higher concentration large scale inhomogeneities and non-linear
effects have to be taken into account and, therefore, the usage of a
LES is required.
Similar investigations were performed in Gradl et al. (2006)
and Gradl and Peukert (2009). They investigated a t-mixer with
the help of experimental and numerical methods. They used
coupled direct numerical simulations (DNS) and population bal-
ance equations (PBE) to model the inuence on the resulting
particle size distribution. With the help of a derived micro-mixing
model they were able to predict the particle size distribution inthe precipitation process. For unstable conditions they extended
their coupling with the DLVO theory which improves the possibi-
lities to stabilise the production of particles.
1.2. Particle ladenows
In contrast to the work shown above a lot of investigations have
taken place in the eld of particle laden turbulent ows in non-
micro-devices. In Messa et al. (2013, 2014) different process para-
meters for turbulent ows were computed and compared with
experimental data. They compared their computed results based on
a turbulence model including physical mechanisms with experimen-
tally measured pressure drops, solid volume fractions or spatial
particle distributions and particle velocities. Furthermore, DNS can
be used to calculate micro- and macro-time scales (compare the
work ofWang and Manhart, 2012).
(Gui et al., 2010) simulated the particlevortex interaction in
turbulent, swirling jets with the help of DNS and studied the
inuence of the particles on a plane jet in Gui et al. (2013). They
showed that the DNS is a good method to simulate those interac-
tions if the Reynolds number is not to high due to the rapidly
increased computational costs. Furthermore, they pointed out that
for high volume fractions of the dispersed phase the two-wayinteractions between the uid and the particles have to be taken
into account.
1.3. Stress mechanism in dispersion processes
A general overview of the acting stresses on particles in a uid
is given in Raasch (1961) and a general overview of dispersion
experiments in different dispersing devices is given by Schilde
et al. (2010, 2011).
Investigations of dispersion and aggregation behaviour of parti-
cles in non-uniform shear ows were performed and described by
Masuda et al. (2013b) (backward facing step) and Masuda et al.
(2013a) (parallel plates). These investigations were performed for
lowReynolds numbers and two-dimensional
ow simulations. Theyintroduced a model which enables the prediction of size distribution
of particle clusters (based on coagulation and breakup) and com-
pared the gained results with experimental data.
Rotorstator devices are often used for the production of disper-
sions or emulsions as they produce high shear stresses. Jasiska et al.
(2013, 2014)simulated the three-dimensional ow in a rotor stator
with the k model to predict the power input and turbulence
within the ow and/or linked it with the population balance
equations to predict the droplet size distribution over time.
CFD coupled with the discrete element method (DEM) can be
used to evaluate the stress acting on agglomerates in an pharma-
ceutical dry powder inhaler. Tong et al. (2013)used this approach
to identify agglomerate-device impacts as the dominant stress
mechanism beside multiple impacts between agglomerates. A
similar approach to describe the dispersion progress in a stirred
media mill is shown in Beinert et al. (2014). Coupled CFD-DEM
simulations were performed to evaluate the contacts between
grinding media to estimate with the help of PBE the particle
size distribution over time.
2. Methods and models
In the rst section the microchannels, experimental set-up
and material properties are described. Further information on
the experimental part of this project are given in Gothsch et al.
(2011, 2014). The second part describes the software and used
boundary conditions while the third part describes the turbulent
model. The fourth part denotes how the acting stress is calculated,while the last section takes the effect of turbulence on particles
into account.
2.1. Experimental set-up
A uid was pumped through the microchannels (containing
particles) with a pressure difference up to p 500 bar andexpanded to ambient pressure or into a pressure vessel at various
pressures. These microchannels were manufactured by the Insti-
tute for Microtechnology of TU Braunschweig (compare Lesche
et al., 2011). The pressure in the vessel is called backpressure
while the difference between the inlet and the outlet is called
pressureor pressure difference. The backpressure experiments
were performed to investigate the inuence of cavitation. Two
S. Beinert et al. / Chemical Engineering Science 123 (2015) 197206198
-
8/10/2019 1-s2.0-S0009250914006198-main
3/10
different types of experiments were used to compare the results
with the numerical gained data: mass ow measurements (with or
without backpressure) and dispersing experiments. These results
were performed in ve different microchannels which are shown
in Fig. 1. The upper four geometries are microchannels of older
designs which were investigated in previous works. The acting
stresses based on a k model were computed (compareBeinert
et al., 2012) and are used in this work to evaluate the new model.
As the previous experimental results showed that the orice
channel was the most effective one, a new geometry (the lowest
one inFig. 1) was fabricated in silicon and stainless steel to gain
higher pressures. All microchannels of the rst generation offer a
height ofh 68m while the second generation offers a height ofh 50m.
In order to simulate the ow in a correct manner material
properties have to be known. In Table 1 all required material
properties are listed. For the dispersion experiment aluminium oxide
Al2O3 (Aeroxides
AluC, Evonik) produced by pyrolysis was used.
The solid density of aluminium oxide (Al2O3) cannot be used as
the particle's density. They consist of primary particles which are
formed to aggregates (solid strong bonds) and/or agglomerates
(solid or liquid weak bonds) with a fractal shape due to the
pyrolysis production process. This shape can be quantied by the
fractal dimension d. According toStiess (2009)the fractal dimen-
sion is dened as
n ra
rp d
; 1
where n is the number of primary particles, ra the radius of
gyration of the agglomerate or aggregate and rp the radius of the
primary particles. A fractal dimension ofd 1 represents a row of
primary particles while a fractal dimensiond 3 represents a solid
sphere. This relation can be used to calculate the volume ratio of
solid v within a sphere with the radius ra:
v ra
rp
d 3: 2
The fractal dimension of aluminium oxide (Aeroxides AluC,
Evonik, was used, too) was determined as d 2.7 (compare
Capitaine and Hill, 2012). Therefore, the density of spherical
particles which represents the investigated aluminium oxide
particles with the radius of the agglomerates ra 115 nm (the
mass-weighted measured median particle size is x50 230 nm,
compare Gothsch et al., 2011) and the radius of the primary
particlerp 13 nm (taken from producers data sheet) is
a H2O ra
rp
d 3 Al2O3 H2O
2179 kg=m3: 3
This density was used to depict the shape of the agglomeratesas only a spherical drag law was used for the stationary particle
tracking. The reduced density is comparable to a higher drag
coefcient which is caused by the fractal dimension of the
agglomerates.
2.2. Numerical set-up
The simulations were performed with the commercial software
package ANSYSs Fluents
14.0.0 based on block structured meshes
created with ANSYSs ICEM CFDs 14.0. The mesh of the micro-
channels of the 2nd generation is resolved ner to improve the
quality of the simulations (four million cells increased to seven
million cells). To reduce computational cost the symmetry of
microchannels was taken into account. As turbulence model aReynolds-stress model was used. The pressure difference between
inlet and outlet was specied and the uid ow was calculated.
Due to the fact that the uid expands to different pressures
including ambient pressure cavitation occurred. To include the
effect of cavitation the mixture model was used and as interaction
between steam and water the ZwartGerberBelamri modelwas
applied. The simulations were transient to enable non-steady
behaviour of the different microchannels. If a stationary value
occurred a stationary particle tracking was conducted to measure
acting stresses, uid and particle velocity (hence, the relative
velocity) and volume fraction of surrounding steam. To quantify
the dispersing behaviour of the different geometries the seeding
points were spatial distributed all over the inlet (distance between
two seeding points: d 8:7 m, leading to 10,201 seeding pointsover the entire inlet). The normal uid velocity component at
every seeding point was used to weight the tracked values. The
stresses acting on the particles were calculated by variables of the
uid as described in the following.
The time step size varied between t 2 10 92 10 8 sdepending on the pressure difference p as it is nearly propor-tional to the maximum velocity. The Courantnumber was partly
above 1 without any stability issues. For all channels, expect the
straight channel, the mass ow rate over time behaves like an
under- or an overdamped system (a maximum of 150,000 time
steps were required). As convergence criteria the absolute value of
the residuals for every solved equation had to be lower than 104,
which was only critical for the rst couple of time steps.
2.3. Numerical models
In this model the uid stress is taken as an indicator for
breakage. For very small particles the validation of this assumption
has to be proven (compare Section 2.5), for very large particles and
high volume fraction of the particles the uid ow would be
inuenced which is not taken into account. The investigated
particles are small and the volume fraction is very low
(x50 230 nm and m 0:010 which is v 0:003, compare
Gothsch et al., 2011).
If the uid is incompressible the NavierStokes equations are
uit
xjuiuj
pixi
xj
uixj
ujxi
ki: 4
Fig.1. Geometries of the four microchannels of the 1st generation (upper four) and
the orice channel of the 2nd generation (lowest one).
Table 1
Properties of used materials.
Material Density(kg/m3) Viscosity(Pa s) Vapour pressure(Pa)
Watera 997.05 8.901 104 3230
Steama 0.023 9.867 106
Al2O3b 3270
a Values taken fromVDI (2013).b Values taken from producers data sheet.
S. Beinert et al. / Chemical Engineering Science 123 (2015) 197206 199
-
8/10/2019 1-s2.0-S0009250914006198-main
4/10
To reduce computational costs it is common to use turbulence
models which need additional transport equations. With higher
amount of additional transport equations the computational costs
and the accuracy increase. Reynolds proposed an averaging to
reduce the effects of turbulence in the Reynolds-stress tensor
(RST). Therefore, the velocity u and the pressure p are split into
an average and a uctuation value:
ui ui u0i and pi p i p
0i: 5
With these new quantities theNavierStokes equations could be
transformed into the Reynolds-averaged-NavierStokes equations
(RANS equations):
uit
xjuiuj u
0iu
0j|fflffl{zfflffl}
RST
0B@
1CA p i
xi
xj
uixj
ujxi
ki 6
The Reynolds-stress tensorin index notation is
u0iu0
j
u01u01 u
01u
02 u
01u
03
u02u01 u
02u
02 u
02u
03
u03u01 u
03u
02 u
03u
03
0BB@
1CCA: 7
It is easy to observe that the Reynolds-stress tensoris symmetric
and, hence, six additional transport equations are required to solvethe Reynolds stresses. Furthermore, the diagonal elements are
tensile stress while the others represent shear stress. To receive
the acting stresses at every place in the solution area the stresses
out of the averaged velocity and due to the Reynolds-stress
tensor is required. The resulting stress tensor is
sij
2uixi
u0iu0i fori j
uixj
ujxi
u0iu
0j foriaj
8>>>>>:
8
2.4. Stresses on particles
In this work only spherical particles are investigated and,hence, the differentiation between tensile and shear stress is not
useful as a different coordinate system will lead to different tensile
and shear stresses. The principal axis transformation can be used
to transform sij into a tensor tij which consist only of diagonal
elements. This is done by the Jacobi method and leads to
tij
11 0 0
0 22 0
0 0 33
0B@
1CA: 9
11, 22 and 33 are the principal stresses and can be used to
detect material failure. It is a common method (compare max-
imum shear stress criterion) to calculate if a material breaks. It can
be useful to reduce the stresses to one value. Hence, the principal
stresses are used to dene a vector s! which norm could be usedas acting stress:
s!
11
22
33
0B@
1CA and
s ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
211 222
233
q : 10
2.5. Turbulent stress on particles
The next section discusses the stress acting on the particles.
First, it is interesting to compare the stress induced by turbulence
and the stress induced by the averaged velocities as proposed in
this model. Second, it is useful to compare the stress acting on a
particle taking the ratio of the size of the particles and the size of
the smallest eddies into account. Third and last, it is necessary to
consider the character of the stress (like inertia or viscous stress).
2.5.1. Order of turbulent stress
In this section the order of the turbulent stress is compared
with the order of the stress induced by averaged velocities. For
every Reynolds-stress model the averaged velocity gradients and
the uctuation of every velocity component are calculated. The
turbulent stress su0 and the stress induced by the averaged
velocitiess u are
su0 =k
1=2and su
uixi
: 11
where kis the Kolmogorov time scale, the turbulent dissipation
rate and the kinematic viscosity. The ratio is
su0
su
ffiffiffiffiffiffiffiRet
p
u0il
1
ui=xi; 12
if the turbulent dissipation rate , which is dened using the
velocity uctuationu 0i and the scale of the larger eddies l,
k3=2
l pu0i3
l ; 13
and the turbulent Reynolds number
Ret u0i l
14
is used. One can take u0i=l ui=xi as the frequency of large eddies
and the frequency of the mean ow should be equal or at least
similar. Therefore, the ratio of the stresses is
su0
su
ffiffiffiffiffiffiffiRet
p 15
and it is easy to observe that for fully turbulent ow the stresses
induced by the mean ow are negligible. In this work different
pressure differences and different geometries are investigated.Some ows are not fully turbulent and, hence, it is necessary to
take both kinds of stresses into account. Nevertheless, the turbu-
lent stress should be the dominant one.
2.5.2. Order of turbulent stress on particles
The turbulent stress acting on particles su0 ;p in dependency on
the particle's diameter dp can be estimated using the uid's
density and the turbulent dissipation rate :
su0;p dp 2=3
16
The ratio between the turbulent stress acting on particles su0;pand the stress induced by averaged velocitiessu can be estimated
in the following manner:
su0;psu
ffiffiffiffiffiffiffi
Retp
su0;psu0
ffiffiffiffiffiffiffi
Retp
dp 2=3
= 1=2
ffiffiffiffiffiffiffi
Retp
dp
k
2=3with k
3
1=417
This leads to the conclusion that with increasing turbulence
and with increasing ratio of particle size and size of the smallest
eddies the turbulent stress will become dominant. As the range of
investigated ows includes laminar ows and ows in transition
as well as particles with different size the uid stress due to
gradients of the averaged velocity is still necessary to take into
account. Again, the expected turbulent ow leads to the assump-
tion that the turbulent stress is the important one.
S. Beinert et al. / Chemical Engineering Science 123 (2015) 197206200
-
8/10/2019 1-s2.0-S0009250914006198-main
5/10
2.5.3. Order of inertial turbulent stress
The inertial stress acting on a particle is the energy input of the
active uidEfper volume of the uid itselfVf. The energy input is
caused by the smallest eddies which dissipate the uids energy
due to viscous effects. It can be written as
su0 ;i EfVf
mf u
2
mf=
u2p dpk
2
d2p: 18
Therein, mf is the mass of the active uid and k is the
Kolmogorov time scale. As the inertial effects on a particle can
only be based on eddies of at least the same magnitude the acting
velocity should be proportional to the diameter of the particle and
theKolmogorov time scale. Comparing these inertial uid stresses
with the turbulent stress due to viscous effects leads to
su0 ;isu0
d
2p
1=2 dpk 2
19
Hence, with decreasingdp=k the inertial uid stress decreases
and becomes negligible for dp=k{1. In the following these
different conclusions have to be proven and will be an indicator
to evaluate if the model is sufcient to describe the acting stresses.
3. Results and discussion
3.1. Massow and spatial distributions
Due to the fact that the pressure difference is specied the
resulting mass ow can be used to validate the macroscopic
quality of the simulations and used models. In Fig. 2 the mass
ow of water for the four different microchannel geometries of the
1st generation is shown (simulations: pressure difference of
p 50, 100, 200, 300 and 500 bar; experiments: varying pres-
sure differencep). Furthermore, a comparison of thekand theReynolds-stress model with experimental data is shown.
It can be seen that the more sophisticated Reynolds-stress
modelimproves the quality of the simulation. The y-channel is
again in a very good agreement. Additionally, the straight (up to
100 bar, for higher pressures the ow in the straight channel did
not become stationary) and z-channel are in very good agreement
with theReynolds-stress model and justify the increased computa-
tional costs. The difference between calculated and experimentally
measured mass ow in the orice channel is reduced but still
distinct. This could be induced due to small inaccuracies in the
production of the microchannel geometry or the dimensions
of these microchannels. Nevertheless, overall the comparison of
the experimental data with the simulation results based on the
Reynolds-stress modelis in good agreement. Fig. 3 shows the
velocity distribution of the uid at a plane with the height of
h 34m (half of the microchannel height) for the four differentgeometries at a pressure difference between in- and outlet of
p 500 bar.As the pressure difference is responsible for the uid ow
the geometry denes the occurring uid velocity. For the same
pressure in the z- and straight channel only moderate velocities
are reached. The long small channel creates a continuous decrease
in pressure and, hence, a relatively constant average velocity. The
orice and y-channel offer partially very narrow channels and due
to the rapid decrease in pressure high velocities occur. Behind the
narrow parts of the microchannel and in the widening at the end
of the channels regions with very low velocities occur. At these
regions cavitation can develop due to the fall of the static pressurebelow the evaporation pressure. The steam distributions for these
four channels are shown inFig. 4.
If the colour differs from dark blue cavitation leads to steam
until only steam is present for red coloured regions. The steam
distribution of the z- and straight channel and of the orice and
y-channel are similar. For the z- and straight channel steam occurs
at the widening at the end of the channel. For the orice and
y-channel the steam occurs after the narrow section and increases
at the widening at the end. The steam leads to a decreased energy
dissipation due to the reduced density and viscosity. Hence, it
enables the forming of free jets by the meaning of an expansion
into the gas phase (not into vacuum as in other cases).
These gained ow elds can then be used for a stationary
particle tracking. This stationary particle tracking is based on the
Fig. 2. Comparison of the mass ow for the different microchannels and for
different pressure differences.
Fig. 3. Velocity distribution of the uid for the four different geometries at a
pressure difference ofp 500 bar.
Fig. 4. Steam and water distribution for the four different geometries at a pressure
difference ofp 500 bar.
S. Beinert et al. / Chemical Engineering Science 123 (2015) 197206 201
-
8/10/2019 1-s2.0-S0009250914006198-main
6/10
previously described equations and requires the averaged velocity
gradients and the Reynolds-stress tensor.
3.2. Evaluation of the new model
To characterise different microchannel geometries the values of
the different parameters along every possible path are required.
As a Reynolds-stress model is used which is based on the RANS
equations the averaged velocity u and the uctuating velocity u0
are constant at every place. Therefore, it is possible to reduce the
number of tracked particle paths. In Fig. 5the tracked parameters
along a path of a particle with a diameter of dp x50 230 nm
released at the centre of the orice channel for a pressure
difference ofp 500 bar are shown. Therein velocity is theuid's velocity, relative velocity is the difference between the
particle velocity and the uid velocity (negative values represent
the case of a particle which is faster than the uid), steam
fraction is the fraction of steam in the region surrounding the
particle and stress represents the acting stress according to
Eq.(10).
It can be seen that the uid's velocity is very low before the
particle reaches the orice. Short before the orice a rapid increase
in the uid's velocity occurs leading to a continuously deceleration
until the widening is achieved. The relative velocity is an indicator
for the acting stress as the difference in velocity causes forces
acting on the particles. The largest difference occurs when the
uid is accelerated rapidly. The peak difference is vrel 8 m=s.
After the orice the particle is continuously decelerated by the
uid. The steam fraction in the volume surrounding the particle
increases after the orice and remains high until the particle
leaves the microchannel. When the local minimum is reached the
widening of the channel begins.The main stress along the entire path of the particle is tracked,
too. It can be seen that the peak value occurs in the orice itself,
but after the peak value of the relative velocity. Due to the
turbulence after the orice the main stress is higher compared
to the laminar region before the orice. Furthermore, it can be
seen that the relative velocity is an indicator if locally high uid
stress occurs. Due to the fact that a Reynolds-stress model is used,
the stress induced by turbulence cannot be tracked or charac-
terised by the relative velocity. As these stresses could be larger
the main stress has to be used.
To quantify different geometries at different pressures the
peak values (like relative velocity, stress, etc.) of every tracked
path are calculated and weighted by the normal velocity compo-
nent at the inlet of the seeding point. In Fig. 6the probability of a
particle to face a maximum relative velocity for the four micro-
channel geometries for a pressure difference of p 500 baris shown.
The z- and the straight channel offer a similar relative velocity
distribution. Due to the redirections within the z-channel slightly
higher values occur. Expectedly, the y- and, especially, the orice
channel prevail higher relative velocities. Due to the narrow parts
the uid's acceleration at a very limited space leads to higher
differences. If the acceleration or the deceleration of the particles
by the surrounding uid is the leading breakage mechanism
(excluding shear stress due to velocity gradients and turbulence)
the orice channel should offer the best results while the z- and
straight channel should be more inefcient; the y-channel should
be between them.
To predict the dispersion efciency of different microchannel
geometries the acting stresses on a particle were tracked and the
resulting distributions for the orice channel at ve different
pressures (see Fig. 7) and for the four different investigated
geometries at two different pressures (seeFig. 8) are presented.
The left part of Fig. 7 shows the probability for a particle to
suffer a specic peak value of the stress at ve different pressure
differences. It can be seen that an increase in pressure difference
leads to an increased probability to sustain the same stress or it
leads to an increased possibility to sustain a higher stress. To
investigate the dependency of the peak stresses s on the pressure
difference the stress is normalized by the pressure difference:
sn s
p: 20
The probability to face a specic normalized stress is shown
in the right part ofFig. 7. As the shown curves differ for varying
pressure differences it can be stated that there is now lineardependency of the stress on the pressure. The dependency is less
than linear meaning that a doubled pressure difference does not
lead to an doubled stress value or probability.
The comparison between the different microchannel geome-
tries is shown in Fig. 8. For the lower pressure difference
p 100 bar all channels exhibit peak values for every trackedpath within a close range. While the z-channel offers highest peak
values, the three remaining channels behave very similar. For a
pressure difference ofp 500 bar the results differ. The straightchannel offers the lowest peak values and is about one decade
lower than the other channels. While the occurring peak values in
the z-channel are all-over despite the comparatively low relative
velocities the highest, the maximum stress occurs in the orice
and y-channel. This leads to the conclusion that the relative
Fig. 5. Velocities, stress and steam fraction for a single particle passing the orice
channel for a pressure difference of p 500 bar (the axes are assigned to the
colours of the four variables). (For interpretation of the references to colour in this
gure caption, the reader is referred to the web version of this paper.)
Fig. 6. Peak value distribution of the relative velocity between uid and particle for
the four microchannels for a pressure difference ofp 500 bar.
S. Beinert et al. / Chemical Engineering Science 123 (2015) 197206202
-
8/10/2019 1-s2.0-S0009250914006198-main
7/10
velocity, which is based on the averaged uid velocities, might not
be the dominating dispersion quantity.
Beside the peak values of the acting stress the energy input on
the particle or at the surface of it might be the leading dispersing
quantity. Therefore, the energy input per surface Esalong the path
which is dened as
Es E
S
Z lull
sdl with l l andlu f sc 21
was tracked. The critical stress sc is the stress which is required to
act on the particles or at least their structure. As it is unknown in
this work two randomly chosen values are used to characterise the
different geometries. If the stress s exceeds the critical stress sc(the chosen values are sc 10
5 Pa and sc 106 Pa) the lower
boundl l was set; if the stress falls below those critical values the
upper bound lu was set. For every path line the highest integral
was used to quantify the surface energy. The probability to face
such an surface energy for both critical values is shown in Fig. 9
(surface energies with a critical stress ofsc 105 Pa are shown in
bright colours while surface energies with a critical stress of
sc 106 Pa are shown in powerful colours).
It can be seen that there are huge differences between both
tracked surface energies. While the straight channel offers the
highest surface energies for the lower critical stress it does not
even reach the higher critical stress, i.e. no curve is shown in Fig. 9.
Due to the constant velocity and turbulence in the entire micro-
channel the product of stress and distance is high as long the
critical stress is smaller than the stress acting in the channel. By
increasing the critical stress sc the other three channels which
offer higher peak values due to higher pressure gradients enable
higher surface energies.
Furthermore, it can be seen that for a critical stress of
sc 106 Pa between 10 and 40% of the particles does not face a
surface stress. Also, the behaviour of the three remaining micro-channel geometries changes. While the z-channel leads to the
highest probability to face a specic stress the order of the orice
and y-channel changes if the critical stress varies. This behaviour
can be used as an indicator to estimate the real critical stress as
both channels behave similar in the dispersing results. Hence, it
can be stated that if the surface energy is the dominant dispersing
parameter its value should be at least above sc 105 (and possibly
above sc 106 Pa).
3.3. Stresses in the microchannels of the 1st generation
Comparing the experimental performed dispersing results leads
to the assumption that the computed stresses of the z-channel are
overestimated. The straight, y- and orice channel behave like the
Fig. 7. Peak value distribution of the maximum stress (left) and normalized maximum stress (right) acting on the particles for the ori ce channel for pressure differences
betweenp 50 and 500 bar.
Fig. 8. Peak value distribution of the main stress acting on particles for the fourmicrochannels for a pressure difference ofp 100 (bright colours) and 500 bar
(powerful colours). (For interpretation of the references to colour in this gure
caption, the reader is referred to the web version of this paper.)
Fig. 9. Probability for a particle to face a specic surface energy above two different
critical values while passing the four different microchannels at a pressure ofp 500 bar (the straight channel offers no stress values above s 106 Pa). (For
interpretation of the references to colour in this gure caption, the reader is
referred to the web version of this paper.)
S. Beinert et al. / Chemical Engineering Science 123 (2015) 197206 203
-
8/10/2019 1-s2.0-S0009250914006198-main
8/10
experimental results would expect (for peak stresses and surface
energy), while the z-channel defers. Due to the underestimated mass
ow (seeFig. 2) the stress of the orice channel should be under-
estimated, while the others have not to be corrected (correct mass
ow). Compared to the other investigated three channel geometries
the z-channel offers two wide redirections which lead to swirls and
heavily curved streamlines. This behaviour leads to very high
Reynolds stresses which result in higher stresses acting on the
particles and might be the reason for the overestimation.
3.4. Microchannels of the 2nd generation inuence of cavitation
and steam
Another interesting inuence on the dispersing efciency of
different microchannel geometries is the effect of cavitation and,
hence, steam. Therefore, investigations with the orice channel of
the 2nd generation were performed. The effect of cavitation on the
mass ow is presented inFig. 10. For the experimental investiga-
tions an backpressure was applied which increases the static
pressure occurring in the entire channel to a sufcient level which
prevails cavitation and eliminates all steam regions. According to
these results single- and multiphase simulations were performed.
It can be stated that the mass ow is in excellent agreement for
the two investigated pressure differences with or without applied
backpressure. Compared to the results of the 1st generation this
could be based on the ner mesh of the newer microchannel
geometries or the increased dimensions of the channel and the
orice itself (30 68m2 increased up to 80 50m2).In addition, the effect on the mass ow due to the increased
resistance caused by cavitation can be seen, too. The viscosity of
steam is much lower than liquid water and, hence, the inner
friction of the uid is reduced. This would lead to higher mass ow
rates for multiphase systems. But on the other hand, the region
after the orice itself should not inuence the ow before and in
the orice itself. Short before the orice and in the orice the
shear stress of the uid is the highest. The difference in both cases
is that already in the orice cavitation occurs and leads to steam.Even for higher pressures the density of steam is much lower than
in liquid state. Therefore, the creation of steam head to a reduced
cross-sectional area and, hence, to an increased resistance. As long
as steam (created due to cavitation) is present in the orice (but
not in the vena contracta) the mass ow should be lower
compared to the single phase case.
According to the work ofGothsch et al. (2014)there is no linear
increasing mass ow rate between a backpressure of pb 0 bar
andpb 175 bar at constant pressure difference between inlet and
outlet p 200 bar. The mass ow increases up to an back-pressure of pb 90 bar and remains constant for higher back-
pressures. It is expected that the mass ow rate will become
constant if no cavitation occurs as the pressure difference between
inlet and outlet is constant. It could be shown that for a back-
pressure ofpb 90 bar cavitation and, hence, steam is only created
in the vena contracta. If the steam remains in the vena contracta
there is no inuence of it on the uid ow due to the decrease in
density. This explanation is supported by the pressure distribution
shown in Fig. 11. It can be seen that the very low (negative)
pressure (which would lead to steam if no backpressure is applied)
is only present in the vena contracta. Therefore, already low
backpressures can reduce regions especially behind the orice
where steam is present.
In Table 2 the required, calculated backpressures to prevent
cavitation and the ratio between backpressure and pressure
difference between inlet and outlet are listed. Additionally, the
required measured backpressures to receive a constant mass ow
rate and the appropriate ratio are listed, too. It can be seen that
higher pressure differences require higher backpressures to get the
same behaviour. The difference between experimental and simu-
lative gained results is based on the described effect (backpressure
inuence on the mass ow). The equation to calculate the lowest
occurring pressurepminin the simulations to suppress cavitation is
pmin 0:505 p1:10 22
and to calculate the experimentally gained backpressure pb to
reach a constant mass ow is
pb 0:076 p1:33: 23
The italic values in Table 2 represent values gained from
Eqs. (22) and (23). It can be seen that the pressure to reach a
constant mass ow and, hence, to allow cavitation only in the vena
contracta (leading to a constant mass ow) increases rapidly
compared to the lowest pressure which would suppress cavitation
in the whole microchannel. The extrapolated value to gain a
constant mass ow for a pressure difference at p 1000 bar isin good agreement with the experimental results (as a constant
mass ow was not reached).
Fig. 10. Experimentally measured and computed mass ow for different inlet
pressures with/without backpressure for the new orice channel.
Fig. 11. Pressure distribution in the 2nd generation orice channel at a pressure of
p 200 bar without cavitation.
Table 2
Applied pressure differences and lowest occurring pressures which suppress
cavitation (received from simulations) and required backpressures to gain a
constant mass ow rate (received from experiments).
Pressure difference (bar) Lowest pressure
(simulations)(bar)
Back pressure
(experiments)(bar)
100 81.7 35
200 175.7 90
300 274.9 150
500 483.4 3001000 1039.3 Z700 755
S. Beinert et al. / Chemical Engineering Science 123 (2015) 197206204
-
8/10/2019 1-s2.0-S0009250914006198-main
9/10
In Fig. 12 the particle tracking for the new orice channel is
presented. The stress is normalized by the pressure to enable an
easy comparison between different pressures and between the
single- and multiphase simulation. The powerful coloured lines
represent the previously described computed stress. Due to the
fact that the curves with and the curves without cavitation are
almost identical for varying pressure differences it can be con-
cluded that there is an almost linear dependency on the peak
values of the stress and the pressure. Furthermore, it can be seen
that the reduced mass ow of the multiphase simulations leads to
an reduced level of acting stresses. This is in good agreement with
the experimental data.
3.5. Evaluation of the magnitude of the turbulent stress
In the following two different stresses are investigated: rst the
stress 50% of all particles would face, second the stress only 10%
of all particles would face. Those stress values for a pressure
difference ofp 200 bar without cavitation are used to calculatethe shear strain of the uid:
50 smax;50= 1:29 109 1=s and
10 smax;10= 3:24 109 1=s: 24
Dividing the maximum velocity (which is about vmax 250 m=s)
by the shear strain the radius of the required eddies can be
calculated:
d50 2 vmax=50 388 nm andd10 2 vmax=10 154 nm: 25
If the length scale for the smallest eddies dened byKolmogor-
ov is used, the lower limit of occurring eddies could be calculated
using the turbulent dissipation rate found in the simulation:
k;sim 38 nm 26
The eddy size computed with the dissipation rate out of the
computed Reynolds-stress tensoris in congruence. Again, with the
length scale dened by Kolmogorov it is possible to calculate an
lower limit at which the dissipation of energy due to uid friction
occurs. For the second approach, Eq. (13) with the maximum
occurring velocity vmaxand the lowest length (half of the height of
the channel l 25m) is used to receive an lower limit for this
length scale:
k 3
1=4
3 l
v3max
1=4 33 nm od10od50 27
It can be seen that the computed stresses used by theReynolds-
stress tensor and the maximum velocity seem to be in the range
which could be delivered by the occurred turbulence. To identify
the order of the turbulence stress inFig. 12the stress calculated by
the averaged velocity gradients (brighter colours) is shown, too.These computed velocity gradients are lower than the real stresses
and two to three orders of magnitude lower than the turbulent
stress computed by the Reynolds-stress model . Hence, it can be
stated that the stress is caused by fully turbulent ows. The ratio of
particles and Kolmogorov length scale is about
dpk
230 nm
33153 nm 1: 28
Hence, the turbulence stress could act on the particles and
could be the leading breakage mechanism.
4. Conclusion
In the present study a model to calculate the stress based on
uid velocity gradients and theReynolds-stress tensoris presented.
Based on the Jacobi method a single value which represents the
uid stress can be calculated for every position in the microchan-
nel. With a stationary particle tracking these stresses could be
tracked. Beside the stress the relative velocities and the energy per
surface were tracked. With the normal uid velocity component at
the inlet a weighting of all maximum values for every seeding
point was possible to enable a quantication of different geome-
tries for varying pressure differences. Five different microchannel
geometries were investigated: a straight, a z-, a y- and an ori ce
channel of the 1st generation and a orice channel of the 21nd
generation. The rst ones were used to compare the results with
previously gained results based on simulations with a k model.
The mass ow rate was used as an indicator for the quality
of the simulations. Compared to the k model the results from
simulations based on the Reynolds-stress model are an improve-
ment and offer (partly excellent) agreement with the experimental
results. Only the orice channel of the 1st generation differs a bit.
Due to the increase of the cell number of the mesh, the improved
accuracy of the turbulence model compared to the previous work
and the excellent agreement in mass ow rate for the orice
channel of the 2nd generation it can be stated that further
improvements of model and ner meshes will not lead to the
identical mass ow rate. As the dimension of the orice channel of
the 1st generation is the smallest and the fabrication method
changed from the 1st to the 2nd generation the difference could be
induced due to the different dimensions and/or fabrication inac-curacies. The calculated stresses are in good agreement except for
the z-channel which offers two redirections which lead to swirls
and highly curved streamlines. It could be shown that the max-
imum stress as well as the energy per surface can be the dominant
breakage mechanism (while the maximum stress as leading
breakage mechanism is more reasonable).
For the orice channel of the 2nd generation the focus was on
the inuence of the cavitation and steam regions. The mass ow
was in excellent agreement with the experimental data. It could be
shown that the simulations can reproduce the cavitation and
steam effects. Additionally, the experiments and simulations sup-
port the explanation for the reduced mass ow: as long as the
steam is limited to the vena contracta the mass ow is constant. If
the steam regions expand into the orice itself a reduction occurs.
Fig. 12. Peak value distribution of the normalized main stress and stress due to
velocity gradients acting on particles for the new orice channels for a pressure
difference ofp 100 and 200 bar with and without cavitation. (For interpretation
of the references to colour in this gure caption, the reader is referred to the web
version of this paper.)
S. Beinert et al. / Chemical Engineering Science 123 (2015) 197206 205
-
8/10/2019 1-s2.0-S0009250914006198-main
10/10
Furthermore, it was shown that the turbulence ow and the
eddies are in the range which might be responsible for the
breakage of the particles. The stress computed with the developed
model is in the range of reasonable occurring turbulent ows.
Hence, the increased computational costs for solving the six
additional equations for the Reynolds-stress tensorresult in an
improved quality of the owelds and quantication possibilities.
List of symbols
Latin symbols
E energy (J)
d fractal dimension ()
dp m particle diameter
h height, depth (m)
k turbulent kinetic energy (m2=s2)
ki body force, index notation (N/kg)
l length (m)
p pressure (Pa)
b backpressure (Pa)
r radius (m)
s stress (Pa)
sc critical stress (Pa)
sn normalized stress ()
s! stress vector (Pa)
su0 , su stress induced by velocity (Pa)
sij stress tensor (Pa)
S surface (m2)
t time (s)
tij transformed stress tensor (Pa)
ui velocity, index notation (m/s)
vmax maximum velocity (m/s)
V volume (m3)
x50 median particle size (m)
xi coordinates, index notation (m)
Greek symbols
turbulent dissipation rate (m2=s3)
viscosity (Pa s)
k Kolmogorov length scale (m)
kinematic viscosity (m2=s)
density (kg=m3)
ij stress component (Pa)
k Kolmogorov time scale (s)
m mass ratio ()
v volume ratio ()
Acknowledgments
The authors gratefully acknowledge the DFG for nancial
support within the DFG research group 856 Mikrosysteme fuer
partikulaere Life-Science-Produkte (mikroPART) (KW 9/6-2).
References
Beinert, S., Gothsch, T., Kwade, A., 2012. Numerical evaluation of ow elds andstresses acting on agglomerates dispersed in high-pressure microsystems.Chem. Eng. Technol. 35 (11).
Beinert, S., Schilde, C., Gronau, G., Kwade, A., 2014. Cfd-discrete element method
simulations combined with compression experiments to characterize stirred-
media mills. Chem. Eng. Technol. 37 (5) .Capitaine, E.M.R., Hill, J.M., 2012. Determination of the fractal dimension of alumina
using adsorption: challenges of choosing appropriate adsorbates and analysis
methods. Can. J. Chem. Eng. 90 (3).Gothsch, T., Finke, J.H., Beinert, S., Lesche, C., Schur, J., Bttgenbach, S., Mller-
Goymann, C., Kwade, A., 2011. Effect of microchannel geometry on high-
pressure dispersion and emulsication. Chem. Eng. Technol. 34 (3).Gothsch, T., Schilcher, C., Richter, C., Beinert, S., Dietzel, A., Bttgenbach, S., Kwade,
A., 2014. High-pressure micro
uidic systems (hpms):
ow and cavitationmeasurements in supported silicon microsystems. Microuid. Nanouid.,
http://dx.doi.org/http://dx.doi.org/10.1007/s10404-014-1419-6 , online.Gradl, J., Peukert, W., 2009. Simultaneous 3d observation of different kinetic
subprocesses for precipitation in a t-mixer. Chem. Eng. Sci. 64 (4) .Gradl, J., Schwarzer, H.-C., Schwertrm, F., Manhart, M., Peukert, W., 2006.
Precipitation of nanoparticles in a t-mixer: coupling the particle population
dynamics with hydrodynamics through direct numerical simulation. Chem.
Eng. Process.: Process Intensif. 45 (10).Gui, N., Fan, J., Chen, S., 2010. Numerical study of particle-vortex interaction and
turbulence modulation in swirling jets. Phys. Rev. EStat. Nonlinear Soft Matter
Phys. 82 (5).Gui, N., Yan, J., Fan, J., Cen, K., 2013. A dns study of the effect of particle feedback in a
gas-solid three dimensional plane jet. Fuel 106.Jasiska, M., Badyga, J., Cooke, M., Kowalski, A., 2013. Application of test reactions
to study micromixing in the rotor-stator mixer (test reactions for rotor-stator
mixer). Appl. Therm. Eng. 57 (12), 172179. http://dx.doi.org/http://dx.doi.
org/10.1016/j.applthermaleng.2012.06.036 .
Jasiska, M., Badyga, J., Hall, S., Pacek, A.W., 2014. Dispersion of oil droplets inrotor-stator mixers: Experimental investigations and modeling: E. Chem. Eng.
Process. 84, 4553. http://dx.doi.org/10.1016/j.cep.2014.02.008.Lesche, C., Krah, T., Bttgenbach, S., 2011. Microelectro discharge machiningan
innovative method for the fabrication of 3d microdevices. In: Proceedings of
SPIEThe International Society for Optical Engineering, vol. 8066.Makowski, L., Orciuch, W., Badyga, J., 2012. Large eddy simulations of mixing
effects on the course of precipitation process. Chem. Eng. Sci. 77 .Masuda, K., Suzuki, H., Komoda, Y., Hidema, R., 2013a. Aggregation/dispersion
behaviors ofne particles in aow between parallel plates. J. Chem. Eng. Jpn 46
(8).Masuda, K., Suzuki, H., Komoda, Y., Hidema, R., 2013b. Particle dispersion/aggrega-
tion model in a non-uniform shear ow. Nihon Reoroji Gakkaishi 41 (2), 34.Messa, G.V., Malin, M., Malavasi, S., 2013. Numerical prediction of pressure gradient
of slurry ows in horizontal pipes. In: American Society of Mechanical
Engineers, Pressure Vessels and Piping Division (Publication) PVP, vol. 4.Messa, G.V., Malin, M., Malavasi, S., 2014. Numerical prediction of fully-suspended
slurry ow in horizontal pipes. Powder Technol. 256 .
Pennella, F., Rossi, M., Ripandelli, S., Rasponi, M., Mastrangelo, F., Deriu, M.A.,Ridol, L., Khler, C.J., Morbiducci, U., 2012. Numerical and experimental
characterization of a novel modular passive micromixer. Biomed. Microdevices
14 (5).Porta, G.M., Thovert, J.F., Riva, M., Guadagnini, A., Adler, P.M., 2012. Microscale
simulation and numerical upscaling of a reactive ow in a plane channel. Phys.
Rev. EStat. Nonlinear Soft Matter Phys. 86 (3).Raasch, J., 1961. Beanspruchung und verhalten suspendierter feststoffteilchen in
scherstroemungen hoher zaehigkeit (Ph.D. thesis). TH Karlsruhe.Schilde, C., Kampen, I., Kwade, A., 2010. Dispersion kinetics of nano-sized particles
for different dispersing machines. Chem. Eng. Sci. 65 (11).Schilde, C., Mages-Sauter, C., Kwade, A., Schuchmann, H.P., 2011. Efciency of
different dispersing devices for dispersing nanosized silica and alumina.
Powder Technol. 207 (13).Stiess, M., 2009. Mechanische VerfahrenstechnikPartikeltechnologie 1, 3rd Edi-
tion Springer, Berlin.Tong, Z.B., Zheng, B., Yang, R.Y., Yu, A.B., Chan, H.K., 2013. Cfd-dem investigation of
the dispersion mechanisms in commercial dry powder inhalers. Powder
Technol. 240.VDI-Gesellschaft Verfahrenstechnik und Chemieingenieurwesen, 2013. VDI-
Waermeatlas. Berechnungsunterlagen fuer Druckverlust, Waerme- und Stof-
fuebergang, 11th Ed. Springer, Berlin.Wang, B., Manhart, M., 2012. Two-phase micro- and macro-time scales in particle-
laden turbulent channel ows. Acta Mech. Sin./Lixue Xuebao 28 (3).Wengeler, R., 2007. Hydrodynamic Stress Induced Dispersion of Nanoscale Agglom-
erates by a High Pressure Process (Ph.D. thesis). TH Karlsruhe.Wengeler, R., Nirschl, H., 2007. Turbulent hydrodynamic stress induced dispersion
and fragmentation of nanoscale agglomerates. J. Colloid Interface Sci. 306 (2).
S. Beinert et al. / Chemical Engineering Science 123 (2015) 197206206
http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref1http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref1http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref1http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref1http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref1http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref1http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref1http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref1http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref2http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref2http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref2http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref2http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref3http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref3http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref3http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref3http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref4http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref4http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref4http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref4http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref4http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref4http://dx.doi.org/%3Cmac_doitxt%3E10.1007/s10404-014-1419-6http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref6http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref6http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref6http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref7http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref7http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref7http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref7http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref7http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref7http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref7http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref8http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref8http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref8http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref8http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref8http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref8http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref9http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref9http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref9http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.applthermaleng.2012.06.036http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.applthermaleng.2012.06.036http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.cep.2014.02.008http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref13http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref13http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref13http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref13http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref13http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref14http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref14http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref14http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref14http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref14http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref14http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref14http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref14http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref15http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref15http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref15http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref15http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref15http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref17http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref17http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref17http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref17http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref17http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref18http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref18http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref18http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref18http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref18http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref18http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref18http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref19http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref19http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref19http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref19http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref19http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref19http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref19http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref19http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref21http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref21http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref21http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref22http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref22http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref22http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref22http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref22http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref22http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref22http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref22http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref23http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref23http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref23http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref23http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref23http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref24http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref24http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref24http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref24http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref26http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref26http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref26http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref26http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref26http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref28http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref28http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref28http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref28http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref28http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref26http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref26http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref24http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref24http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref24http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref23http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref23http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref22http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref22http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref22http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref21http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref21http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref19http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref19http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref19http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref18http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref18http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref18http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref18http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref17http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref17http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref15http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref15http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref14http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref14http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref14http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref13http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref13http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.cep.2014.02.008http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.cep.2014.02.008http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.cep.2014.02.008http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.applthermaleng.2012.06.036http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.applthermaleng.2012.06.036http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.applthermaleng.2012.06.036http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.applthermaleng.2012.06.036http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.applthermaleng.2012.06.036http://dx.doi.org/%3Cmac_doitxt%3E10.1016/j.applthermaleng.2012.06.036http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref9http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref9http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref8http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref8http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref8http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref7http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref7http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref7http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref7http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref6http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref6http://dx.doi.org/%3Cmac_doitxt%3E10.1007/s10404-014-1419-6http://dx.doi.org/%3Cmac_doitxt%3E10.1007/s10404-014-1419-6http://dx.doi.org/%3Cmac_doitxt%3E10.1007/s10404-014-1419-6http://dx.doi.org/%3Cmac_doitxt%3E10.1007/s10404-014-1419-6http://dx.doi.org/%3Cmac_doitxt%3E10.1007/s10404-014-1419-6http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref4http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref4http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref4http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref3http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref3http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref3http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref2http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref2http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref2http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref1http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref1http://refhub.elsevier.com/S0009-2509(14)00619-8/sbref1