Sie sind auf Seite 1von 191

Diss. ETH No.

16116

TWO-PHASE FLOW CHARACTERISTICS IN


GAS-LIQUID MICROREACTORS

A dissertation submitted to the


SWISS FEDERAL INSTITUTE OF TECHNOLOGY ZURICH

for the degree of


Doctor of Technical Sciences

presented by
Severin Wiilchli
Dip!. Masch.-Ing. ETH
born on the 13th June 1974
citizen of Brittnau (AG), Switzerland

accepted on the recommendation of


Prof. Dr. Ph. Rudolf von Rohr, examiner
Pro. Dr. Th. Rasgen, co-examiner

2005
I

Danksagung

Die vorliegende Forschungsarbeit entstand wahrend meiner Zeit als wis-


senschaftlicher Mitarbeiter am Institut fiir Verfahrenstechnik der Eid-
genassischen Technischen Hochschule (ETH) Ziirich zwischen April 2002
und Juni 2005. Finanziell unterstiitzt wurde sie von der ETH Ziirich
im Rahmen eines ETH Forschungsprojekts und von der Emil Barell
Stiftung, Basel.
Meinem Doktorvater, Prof. Dr. Philipp Rudolfvon Rohr, danke ieh herz-
lieh fiir das mir entgegengebrachte Vertrauen, sein Interesse an dieser
Arbeit, die fachlichen und organisatorischen Ratschlage und die Un-
terstiitzung zu einer selbstandigen Arbeitsweise.
Prof. Dr. Thomas Rasgen danke ich fiir die Ubernahme des Korreferates
und das dem Projekt entgegengebraehte Interesse.
Grossen Anteil an dieser Arbeit hat die Werkstatt-Crew des Instituts fiir
Verfahrenstechnik. Die Herausfoderungen, die die Mikroverfahrenstech-
nik stellt, waren riicht immer einfaeh zu bewaltigen. Bruno Kramer, Pe-
ter Hoffmann, Rene Pliiss und Chris Rohrbaeh haben jedoch alle mit
Bravour gemeistert. Besonders danken maehte ieh Herrn Dr. Werner
Dorfler. Er hat nicht nur alle meine Konstruktionen kritisch beurteilt,
sondern zusatzlich aueh grosses Interesse an meinem Projekt gezeigt
und mieh bei Messungen am PSI tatkraftig unterstiitzt. Fiir die organ-
isatorische Unterstiitzung danke ich Silvia Christoffel herzlieh.
Fiir die Einfiihrungen in die Reinraume der ETH und die Unterstiitzung
bei der Herstellung der Mikroreaktoren mochte ich Herrn Dr. Stefan
Blunier herzlich danken.
11 Danksagung

Herrn Dr. Marco Stampanoni von der Swiss Light Source (SLS) am Paul
Scherrer Institut danke ich fur die Hilfe bei den Vorbereitungen und den
grossen Einsatz bei den langen XTM-Messungen.
Die Herren Oliver Kdiuchi, Patrick Reichen und Marcel Strotz, welche im
Rahmen einer Semester- oder Diplomarbeit meine Arbeit unterstutzten,
danke ich fur ihr Engagement und ihre kritischen Fragen und Ideen.
Fur die tolle Zeit mit unvergesslichen Momenten danke ich allen ak-
tuellen und ehemaligen Kollegen am Institut fur Verfahrenstechnik. Ich
werde das sehr kollegiale Klima und die motivierenden und kritischen
Stunden vermissen. Meinen beiden Nachfolgern, Franz Trachsel und Do-
nata Banowski, danke ich fur die Mithilfe zum Schluss meiner Arbeit
und wunsche ihnen viel Spass und Vergnugen bei der Weiterfuhrung der
spannenden Arbeit.
Ganz besonders danken mochte ich naturlich meinen beiden langjahrigen
Schlieremer Kollegen Adrian Wegmann und Nils Kruse. Long live the
PowerLab!
Ein ganz besonderer Dank gilt meinen Eltern und meiner Familie, fur
ihre Unterstutzung wahrend meiner langen Ausbildungszeit.
Grossten Dank gilt meiner Freundin Sibylle Frei. Sibylle hat seit Beginn
meines Studiums viel Verstandnis fur meine Arbeit gezeigt. Danke, dass
Du immer da warst.
III

Summary

An experimental study on the characterization of two-phase gas-liquid


flow relevant to chemical applications in channels with hydraulic diame-
ters in the sub-millimeter scale is presented. Multi-phase flow in silicon-
based microfabricated channels of rectangular cross-sections is consid-
ered. Information on the occurring flow regime and the local void fraction
were obtained from Laser-Induced Fluorescence (LIF) measurements. A
novel method for the visualization of multiphase flow using X-Ray Tomo-
graphic Microscopy (XTM) is proven to be suitable for microfluidic appli-
cations. The liquid flow velocities were measured using micron-resolution
Particle Image Velocimetry (J1PIV). A chemical model reaction demon-
strated the applicability of conducting highly exothermic reactions in
microstructured chemical devices.
To analyze the influence of the fluid properties, the channel diameter,
and variations in. the volumetric ga..') and liquid flow rates on the two-
phase flow pattern, experimentally derived flow regime maps for different
channel geometries and three different fluids are presented. In contrast
to microchannel flows, a great number of correlations exist for multi-
phase flow characteristics in (round) pipes of diameters> 1 mm. The
experimental results from optical flow visualizations in microreactors are
compared with common flow correlations and regime maps for macro-
and microchannels. The agreement with predictions for large-scale ap-
plications is shown to be insufficient. Generally, prediction models based
on experimental data using the same channel cross-sectional shape yield
better agreement with our data than those using a similar channel hy-
draulic diameter.
IV Summary

The ability of reconstructing a 3-dimensional image of the flow makes


X-ray Tomographic Microscopy (XTM) to a powerful non-destructive
investigation method to analyze and study the multiphase flow pattern
in microchannels. Due to several restrictions, this measurement method
is applicable for stationary flow patterns only. Comparing the demon-
strated XTM results with conventional optical flow visualization tech-
niques, XTM is shown to be an adequate flow visualization method for
annular flow, which is mostly required for multiphase chemical reactions
in microsystems.
The pressure drop in two-phase microchannel flow is shown to be signi-
ficantly dependent on the prevailing flow regime. Annular flow regimes
cause lower pressure drops compared to intermittent flow patterns in
!:lpite of higher total volumetric flow rates. A local maximum of the pres-
sure drop at the transition from intermittent to annular flow is presented.
The agreement with pressure drop correlations is inadequate.
To quantify the local volumetric gas content, the emitted fluorescent
intensity was investigated. The void fraction is shown to remarkably
depend on the occurring flow pattern. The great number of void fraction
prediction models for large scale pipe flow are not able to predict the gas
holdup satisfactorily. Generally, experimental validation of the numerical
prediction models (for flow regimes, pressure drops, and void fraction)
is indispensable for the use of microfluidic applications.
The recirculation motion in the liquid segments of intermittent gas-liquid
flows is analyzed using micron-resolution Particle Image Velocimetry
(ttPIV). The velocity distribution influences the mixing and the mass
transport towards the reactive phase interface dealing with two-phase
chemical reactions. For straight microchannels hardly any mass trans-
port over the center line is quantified. Improved mixing quality can be
achieved in channel bends of meandering channels. For enhanced two-
phase mixing, geometrical adaptations are suggested.
v

Z usammenfassung

Die vorliegende experimentelle Arbeit beschreibt die fur chemis-


che Anwendungen wichtigen zweiphasigen ga...:;-fliissig Str6mungen in
Kanalen mit hydraulischen Durchmessern kleiner als 1 mm. Die
mittels mikrotechnischen Fabrikationsmethoden in Silizium hergestell-
ten Kanale weisen rechteckige Str6mungsquerschnitte auf. Informa-
tionen uber die vorherrschende Str6mungsform und lokale Gasge-
haltwerte konnten mit Laser-Induced Fluorescence (LIF) bestimmt wer-
den. Eine neuartige Anwendungsmoglichkeit von X-Ray Tomographic
Micoscopy (XTM) zur dreidimensionalen Visualisierung von mehrphasi-
gen Stromungen in Mikorkaniilen wurde erfolgreich angewandt. Die
Stromungsgeschwindigkeiten in der Fliissigphase konnten mit Mikro Par-
ticle Image Velocimetry (j.LPIV) gemessen werden. Die Durchfuhrung
einer chemischen Modelreaktion bewies die Eignung von Mikroreaktoren
fur stark exotherme Reaktionen.
Anhand experimentell erarbeiteter Stromungsformkarten wurde der
Einfluss der Fluideigenschaften, des Kanaldurchmessers und der Gas-
und Flussigvolumenstrome auf die Stromungsform bestimmt. Im
Gegensatz zu Stromungen in Mikrokanalen existieren eine grosse
Anzahl numerischer or halh-analytischer Ansatze zur Vorhersage
von Stromungsphanomenen in (meist runden) Rohren mit grossen
Durchmessern (D > 1 mm). Die experimentellen Resultate von
Mikr?,kanalen werden mit diesen bekannten Korrelationen verglichen.
Die Ubereinstimmung ist generell ungenugend. Es kann gesagt wer-
den, dass eine Ahnlichkeit im zugrunde liegenden Kanalquerschnitt die
besseren Resultate liefert als eine Ubereinstimmung des Durchmessers.
VI Zusammenfassung

Die Moglichkeit, ein 3-dimensionales Abbild der vorhandenen Stromung


zu rekonstruieren, macht X-Ray Tomographic Microscopy (XTM) zu
einer geeigneten, nicht intrusiven Moglichkeit, die Stromungsform in
Mikrokanalen zu anaIysieren. Infolge verschiedener, technisch bed-
ingter, Einschrankungen ist diese Messmethode nur fur stationare
Stromungszustande anwendbar. Werden die XTM-ResuItate mit den op-
tischen Stromungsaufnahmen verglichen, beweist sich XTM als geeignet
fur die U ntersuchung von Ringstromung, weIche fur vieIe chemische
Reaktionen erwunscht ist.
Messungen zeigen, dass der Druckabfall in zweiphasigen Stromungen
in MikrokanaIen merklich von der vorherrschenden Stromungsform
abhangt. Ringstromung reduziert den Druckabfall im VergIeich zu
intermittierender Stromung merklich, trotz des hoheren totalen
Volumenstroms. Es herrscht ein lokaIes Druckabfallmaximum im
Ubergangsgebiet zwischen Ring- und intermittierender Str6mung vor.
Die Ubereinstimmung der MessresuItate mit verschiedenen Korrelatio-
nen ist ungenugend.
Der lokaIe Gasvolumengehalt wurde mittels der emittierten FIuD-
reszenzintensitat untersucht. Es kann gezeigt werden, dass der Gas-
gehaIt entscheidend von der Stromungsform abhangt. Die zahIreichen
Modelle zur Berechnung des Gasvolumengehalts fur Stromungen in
Rohren grosser Durchmesser sind nicht geeignet fur Mikroanwendungen.
Grundsatzlich ist fur alle analytischen Berechnungen (Stromungsform,
Druckabfall und Gasvolumengehalt) eine experimentelle Validierung
unerliisslich.
Die Geschwindigkeitsverteilung in der FIiissigphase einer intermittieren-
den Str6mung konnte mit Hilfe von Mikro Particle Image VeIocimetry
(j.LPIV) gemessen werden. Die interne ZirkuIation beeinflusst die Durch-
mischung und den Stofftransport hin zur reaktiven Phasengrenzfl.ache
in chemischen Reaktionen. Fiir gerade MikrokanaIe konnte kein merk-
licher Stofftransport iiber die KanaImittellinie festgestellt werden. Die
interne Durchmischung kann durch die Anwendung gewundener KanaIe
verbessert werden. Fiir gute MischresuItate sollten geometrische Anpas-
sungen im Design der KanaIe in Betracht gezogen werden.
VII

Table of Contents

Danksagung I

Summary III

Zusammenfassung V

Nomenclature XIII

1 Introduction 1

1.1 Motivation for the presented work 1


1.1.1 Micro chemical applications 1

1.1.2 Multiphase flow in microchannels . 3

1.1.3 Measurement methods in microfluidics 5


1.2 Thesis structure and objectives . . . . . 6
1.2.1 structure of the presented work. 6
1.2.2 Project goals . 6
VIII Table of Contents

2 Basics and state of the art 8


2.1 Multiphase flow . . . . . . .... 8
2.1.1 Basics of multiphase flows 8
2.1.1.1 Main equations for multiphase flows 9
2.1.2 Multiphase microfluidics . . . . . . . . 12
2.1.2.1 Low-Reynolds-number flows 12
2.1.2.2 Transport phenomena . . . . 13
2.1.3 Characterization of multiphase gas-liquid flows 14
2.1.3.1 Flow pattern . 14
2.1.3.2 Pressure drop 17
2.1.3.3 Volumetric gas content 18
2.2 Hydrogenation of nitrobenzene in microchemical devices 19
2.2.1 Basic principle of hydrogenation reactions 19
2.2.2 Hydrogenation of nitrobenzene to aniline 20
2.2.3 Gas-liquid microreactors . . . . . . . . . . 21

3 Experimental setup and measurement methods 22

3.1 Experimental setup .. 22


3.1.1 Microchannels 22
3.1.1.1 Manufacturing methods 22
3.1.1.2 Design and layout of the microchannels
and reactors 25
3.1.2 Channel holder . . . . . . . . . . . . . . . . 30
3.1.3 Visualization and measurement equipment . 31
3.1.4 Fluids ........................... 33
Table of Contents IX

3.2 Experimental setup for the hydrogenation reaction 34


3.3 Measurement methods . . . . . . . . . . . 37
3.3.1 Laser-Induced Fluorescence (LIF) 37
3.3.2 Micron-resolution Particle Image Velocimetry
(t.tPIV ) 38
3.3.2.1 Basic principle of Particle Image Ve-
locimetry (PIV) .. . . . . . . . . . . .. 38
3.3.2.2 Micron-resolution Particle Image Ve-
locimetry (t.tPIV) 40
3.3.2.3 Seeding particles . . . . . . 40
3.3.2.4 Two-phase velocity measurements using
t.tPIV . . . . . . . . . . . . . . 42
3.3.3 X-Ray Tomographic Microscopy (XTM) 44
3.4 Measurement accuracy . 49
3.4.1 Flow feed . . . 49
3.4.2 Pressure drop. 49
3.4.3 Void fraction measurement accuracy 49
3.4.4 Accuracy of the velocity measurements. 50
3.4.5 Uncertainty in the measurements using X-Ray To-
mographic Microscopy (XTM) 52

4 Flow regime 53

4.1 Flow regimes in microchannels 53


4.1.1 Universal flow pattern map 57
4.2 Comparison with correlations . . . 62
4.2.1 Flow regime map by Baker 62
4.2.2 Flow pattern map by Zhao 63
x Table of Contents

4.2.3 Flow regime map by 'friplett . . 64


4.2.4 Flow pattern map by Kawahara 65
4.2.5 Flow regime map by Coleman .. 66
4.3 3-dimensional flow pattern visualization using XTM 68
4.4 Summary and conclusions . . . . . . . . . . . . . . 76
4.4.1 Flow regimes in rectangular microchannels 76
4.4.2 Flow visualization using XTM 77

5 Pressure drop 79
5.1 Pressure drop in microchannels 79
5.2 Comparison with correlations . 82
5.2.1 Homogeneous pressure drop model 87
5.2.1.1 Main equations of the homogeneous pres-
sure drop model . . . . . . . . . . . 87
5.2.1.2 Comparison with experimental data 88
5.2.2 Friction pressure drop by Lockhart-Martinelli 89
5.2.2.1 Theory of the Lockhart-Martinelli method 89
5.2.2.2 Comparison with experimental data 91
5.2.3 Friction pressure drop by Chawla . . . . . . . 95
5.2.3.1 Empirical pressure drop model by Chawla 95
5.2.3.2 Comparison with experimental data 97
5.2.4 Friction pressure drop correlation by Friedel . 99
5.2.4.1 Basics................. 99
5;2.4.2 Comparison with experimental data 101
5.2.5 Friction pressure drop method by Muller-
Steinhagen and Heck 102
Table of Contents XI

5.2.5.1 Basics on the correlation .. . . . . 102


5.2.5.2 Comparison with experimental data 103
5.3 Summary and conclusions 104

6 Void fraction 106

6.1 Void fraction results 106


6.2 Comparison with correlations 111
6.2.1 Homogeneous void fraction model 111
6.2.2 Volumetric gas content based on correlations by
Armand and Treschew . . . . . . . . . . . . . . . . 111
6.2.3 Empirical model for the determination of void frac-
tions by Chawla 114
6.2.4 Void fractions in microchannels by Serizawa . 115
6.2.5 Correlation by Triplett . . . . . . 118
6.2.6 Void fraction model by Coleman 120
6.2.7 Void fraction model by Kawahara 121
6.2.8 Comparison of the different models. 124
6.3 Void fraction measurements using XTM 125
6.4 Summary and conclusions . 127
6.4.1 Void fraction measurements using Laser-Induced
Fluorescence (LIF) 127
6.4.2 Determination of the void fraction by X-Ray To-
mographic Microscopy (XTM) 129

7 Liquid velocity measurements 130


7.1 PIV analysis for the liquid recirculation motion in inter-
mittent flow 130
XII Table of Contents

7.1.1 Velocity measurement 130


7.1.2 Vortex identification . 131
7.1.3 Proper orthogonal decomposition (POD) analysis. 133
7.1.4 Average correlation method. . . . . . . . . .. 135
7.2 Liquid velocities in straight rectangular microchannels 136
7.3 Recirculation in meandering channels. 145
7.4 Summary and conclusions . . . 153

8 Hydrogenation of nitrobenzene 156


8.1 Packed-bed reactor . 156
8.2 Coated microreactor 157

9 Concluding remarks and outlook 158


9.1 Concluding remarks . 158
9.1.1 Ecological considerations 159
9.2 Outlook . 160
9.2.1 Reactions with supercritical solvents in microreac-
tors 160
9.2.2 Design of a micromixer for industrial relevant re-
actions. . . . . . . . . . . . . . . . . . . . . . .. 162

References 164

Curriculum Vitae 172


XIII

Nomenclature

Latin letters
A m2 (cross-sectional) area
a,A constant value
A 21 Einstein coefficient for the transition
b m channel width
b,B constant value
c constant value
C constant
C constant
C constant factor (Blasius equation)
Cc contraction coefficient
D m diameter
D m 2 /s diffusion coefficient
d m channel depth
d constant value
Dh m hydraulic diameter
De m effective particle diameter
Dp m particle diameter
DB m diameter of diffraction-limited
point spread function
D 2 -D 2-dimensional local velocity gradient tensor
e m smallest resolvable distance
e constant value
el,e2 empirical parameters (Chawla)
XIV Nomenclature

f constant value
ffric friction factor
F(T) fraction of total population
Fr Froude number
9 m/s2 gravitational acceleration
G kg/(m 2 s) liquid mass flux
G constant parameter (Muller-Steinhagen)
g(w) line shape factor
I laser intensity
'l,J pixel locations
J m/s superficial velocity
K constant value
ks m surface roughness
K/d surface roughness
i,L m length
M kg mass
M magnification
M kg/s mass flow rate
m kg/(m2 s) mass flux
N,n number of realizations
n index of refraction
NT absorber total number density
NA numerical aperture
Nu Nusselt number
P Pa, bar pressure
P constant value
Pfric Pa, bar friction pressure
Pace Pa, bar acceleration pressure
Pred Pa, bar reduction pressure
Pgrav Pa, bar gravitational pressure
Pl,P2 parameters (Mishima)
q constant value
Q electronic quenching rate
r,R cross-correlation factor
r m radius
r constant value
R m spatial resolution
Nomenclature xv

Re Reynolds number
s slip
s m diffusion distance
s -] constant value
SF fluorescence signal intensity
t,T s time
T K , C temperature
t constant value
u m/s velocity
u constant value
v constant value
V m3 volume
V m 3 /s volume flow rate
w m/s velocity
w constant value
We Weber number
x m Eulerian Cartensian coordinate
x mass fraction
x mass transport fraction
X two-phase multiplier (Lockhart-Martinelli)
z m length
z m thickness of scintillator

Greek letters

a W/(m2 K) heat transfer coefficient


a beam angle of incidence
c5x m measurement uncertainty
E void fraction, volumetric content, holdup
E: volume transport fraction
T} Pas dynamic viscosity
'Y area ratio
K Boltzman's constant
). W/(mK) heat conductivity
). baker density parameter
XVI Nomenclature

A m wavelength
A dimensionless parameter
Ar real eigenvalue
Acr complex conjugate eigenvalue (real)
Ad swirling strength,
complex conjugate eigenvalue (complex)
v m 2 /s kinematic viscosity
w l/s vorticity
I> 0
inclination angle
I> proportionality factor (Lockhart-Martinelli)
IT characteristic number
W surface tension parameter
W dimensionless parameter
W fluid properties ratio (Chawla)
{} kg/m3 fluid density
a N/m surface tension, interfacial tension
T N/m2 shear stress
s particle relaxation time

,
Tp

~ relative error
two-phase flow parameter (Chawla)

Abbreviations

(I)CCD (Intensified) charge coupled device


DRIE Deep reactive ion etching
EDP Ethylene diamine pyrocatechel
FEM Finite Element Method
FFT Fast Fourier Transform
GC Gas Chromatography
IMES Institute for mechanical systems
KOH Potassium hydroxide
LDV Laser Doppler Velocimetry
LIF Laser-Induced Fluorescence
MEMS Micro-Electro-Mechanical Systems
MS Mass Spectroscopy
Nomenclature XVII

Nd:YAG Neodymium: yttrium aluminium garnet (Y3 A15 0 1 2) crystal


PDA Particle Density Anemometry
PEEK Polyetheretherketone
PIV Particle Image Velocimetry
j.LPIV Micron-resolution Particle Image Velocimetry
PVD Plasma enhanced vapor deposition
RlE Reactive ion etching
SDD Sample Detector Distance
SEM Scanning Electron Microscope
SLS Swiss Light Source
j.LTAS Micro Total-Analysis-System
TMAH Tetramethyl ammonium hydroxide
VOF Volume of Fluid
XTM X-ray Tomographic Microscopy

Sub- and superscripts

abs absorbed
abs absolute
avg average
Ar Armand
B Brownian motion
Cha Chawla
Col Coleman
emit emitted
enr energy
eth ethanol
exp experimental
G gas
glyc glycerol
H homogeneous model
horn homogeneous
Ka Kawahara
L liquid
Ma Massena
XVIII Nomenclature

map flow pattern map


max maximum
mod model
mom momentum
opt optical
reI relative
TP two-phase
1

Chapter 1

Introduction

1.1 Motivation for the presented work


Due to the ongoing progress in the manufacturing techniques of mi-
crostructures, microfluidic devices became relevant for several applica-
tions such as micro-electro-mechanical systems (MEMS), electronic cool-
ing, chemical process engineering, medical and genetic engineering, bio-
engineering, etc. Microchannel flows are superior to macrochannel flows
in the use of a minimum amount of fluids, the high heat and mass trans-
fer rates, and the favorable electro kinetic and other interfacial surface
effects. This fact permits for example the implementation of precise and
reproducible chemical or biological analysis 01' the design of compact
heat exchangers.

1.1.1 Micro chemical applications

In the field of chemical process engineering, high-throughput catalysts


and catalytic screening (Senkan et al., 1999), "lab-on-a-chip" or micro
total analysis systems (t"TAS) gain interest. In carrying out chemical
reactions, whether for the study of kinetics and reaction mechanisms
2 1. Introduction

or for synthesis, the detailed knowledge of the mass and heat transfer
processes are essential. Many of the microreactor concepts were originally
based on single-phase gas or liquid systems (Ehrfeld et al., 2000) or on
two miscible or immiscible phases (Jensen, 1999). Only recently, gas-
liquid microreactors have been introduced in a systematic fashion (Losey
et al., 2001; de Mas et al., 2003; Jahnisch et al., 2000). For multiphase
reactions, the rate and selectivity of the chemical process can be greatly
influenced by the transport of a. single reagent within and across phases.
The enhanced heat transfer observed in microreactors may be exploited
for highly exothermic reactions due to the removal of heat generated and
the suppression of hot spot formation. Another aspect is the optimiza-
tion of slow reactions getting close to the thermodynamic equilibrium.
An example for this was demonstrated by TeGrotenhuis et al. (1999)
for the water-gas~shift reaction, which is investigated as a gas purifi-
cation reaction in the scope of a future application of microtechnology
for reforming applications. In microreactors, heat-exchanging channels
may be introduced, which allow for temperature profiling of the reactor
according to the specific needs of the reactions.
Jensen (2001, 1999) pointed out the advantages of designing and manu-
facturing chemical reactors similar to micro-electro-mechanical systems
(MEMS): heaters and sensors may be integrated and short response
times for heat exchanging tasks are found due to the thin walls applied.
He propagates the application of aerosol and technologies similar to the
inkjet method besides wet preparation and thin film technologies.
The specific surface area is generally high for microstructured reactors.
This increased surface-to-volume ratio which allows for the suppression
of homogeneous reactions (Veser, 2001), however, may also promote un-
desired follow-up reactions, such as total oxidations, at active, but unse-
lective construction materials, such as stainless steel at high temperature.
Microreactors allow safe processing in otherwise explosive regimes. Many
partial oxidations can thus be carried out using pure oxygen and at
elevated pressures. By this, space-time yields can be increased, as e.g.
demonstrated for the ethylene oxide synthesis (Kestenbaum et al., 2002).
A very instructive example for safe processing is given by the hydrogen-
oxygen reaction, which, according to all experience gained so far with
1.1 Motivation for the presented work 3

microreactors, can be handled very safely when operated within the ex-
plosive regime (Veser, 2001). In this content, intrinsic saftey is an often
quoted and recently theoretically confirmed term for such microreactor
processing (Veser, 2001).
Finally, microreactors allow for easier scale-up, as the geometry of the
"unit cell" (the microchannel) does not need to be changed. However,
when numbering-up of microstructured reactors is done, it needs to be
taken into consideration that the housing material has to be minimized to
reduce heat-losses, start-up energy demand and time demand. Therefore,
the practical needs might result in more macro-sized reactors carrying
microchannels as far as scale-up is concerned.

1.1.2 Multiphase flow in microchannels


Microfluidic multiphase systems differ from their macroscopic counter-
parts in two principal aspects, the absence of turbulence, and the smaller
disparity in length and time-scales. Both effects are direct consequences
of the channel dimensions in the micron scale. Commonly used micro-
reaction devices for example feature dimensions smaller than the Laplace
scale at which interfacial forces become significant, or the diffusive scale
at which micromixing takes place.
The physical properties of single-phase microfluidic systems are well un-
derstood. In the case of technical applications however, mostly multi-
phase systems can be found (evaporation, condensation, multiphase flow
in pipe reactors, etc.). Some complex procedures in multiphase systems
cannot be completely characterized theoretically but have to be appro-
ximated by reliable models. For the design, the development, and the
operation of a complete chemical circuit (e.g. lab-on-a-chip), a detailed
knowledge of the multiphase flow and its properties such as the flow
pattern, the volumetric gas content, the pressure drop, or the liquid film
thickness are of utmost importance.
Recent studies of two-phase flow investigated mainly the flow properties
for circular (macro-scale) pipes of diameters D > 1 mm (Lockhart and
Martinelli, 1949; Baker, 1954; Friedel, 1978). Other studies deal with the
question, if and how multiphase flows in microchannels differ from the
4 1. Introduction

ones in macrochannels (Triplett et al., 1999b; Kawahara et al., 2002).


In channels with diameters above some millimeters, the two-phase gas-
liquid flow patterns are dominated by the influence of the gravitational
force. In microchannels with hydraulic diameters below some 100 /-Lm
however, the flow regimes are mainly a function of the interfacial tension,
the viscosity, and the wall friction force. Thus also the pressure drop and
the volumetric gas content of microchannel flows differ from flow in larger
pipes. Only recently, the flow characteristics of multiphase microchannel
flow was investigated (Kawahara et al., 2002; Coleman and Garimella,
1999; Triplett et al., 1999b, a). These projects are mostly experimen-
tal studies and their results are generally applicable but valuable only
for the specific applied geometry. For the prediction of flow patterns in-
side macro and microchannels, empirical or semi analytical models exist
(Baker, 1954; Taitel and Dukler, 1976; Chawla, 1968; Lockhart and Mar-
tinelli, 1949; Friedel, 1978; Zhao and Bi, 2001; Coleman and Garimella,
1999). Unfortunately these models are mostly independent of the flow
pattern and thus not generally applicable. A major question dealing with
multiphase gas-liquid flow is whether or not two-phase flow patterns in
microchannels are different from those encountered in ordinarily sized
tubes.
Measurements of the void fraction in microchannels are still scarce. Al-
most all the holdup correlations, which have been proposed so far, how-
ever, are based on data from circular tubes with diameters larger than
about 5 mm and non-circular channels with such relatively large cross-
sections.
The pressure drop is among the most important hydrodynamic aspects
of two-phase flow. Two-phase pressure drop in channels in particular
is an essential element for the design of piping and process systems,
and has been the subject of numerous experimental studies and a large
number of empirical and semi-analytical predictive methods. The most
widely-used correlations, in spite of the basic theoretical arguments lead-
ing to their development (which may assume particular flow patterns,
often stratified), are presented as flow-regime independent correlations
for obvious practical reasons (Lockhart and Martinelli, 1949; Chisholm,
1967; Friedel, 1978). Nevertheless, these correlations may perform poorly
when applied to high void fraction annular flow. Furthermore, the bulk
1.1 Motivation for the presented work 5

of the previously generated data and the previously published predictive


methods represent two-phase flow in channels several mm in diameter or
larger.
Systematic experimental studies are required in multiphase microfluidics
since computational fluid dynamics codes, which are sufficient for pre-
dicting flow distributions in many (even highly complex) single-phase
microfluidic networks, are very much limited with respect to gas-liquid
flows. Popular numerical approaches, e.g. the volume of fluid (VOF)
method, involve model assumptions, so that reliable experimental data
are highly desirable for validation purposes.

1.1.3 Measurement methods in microfluidics

Measuring flow properties in a micro-configuration is a challenging task,


because the sensors need to be much smaller than the size of the device
under study. In addition, the momentum and the energy of the flow is
diminutive. Therefore, only an minimized amount of momentum and en-
ergy exchange between flow and sensor is allowed so as not to alter the
flow. For these reasons, a very limited number of sensors have been de-
veloped for microflow measurements. Obviously, intrusive measurement
methods known from macro-scale applications are mostly inapplicable.
Integrated pressure sensors are available for microfluidic devices. A very
small duct is used to connect the sensors to the channel. A time-averaged
pressure can be obtained by this arrangement. If unsteady pressure data
are needed, questions about the calibration procedures and even the
physical meaning of the data may arise.
Flow visualization has been proven to be a very useful technique in macro
fluid-dynamics research, and an extensive effort has been expended in
developing methods for visualizing flows in microchannels and valves
(Lanzillotto et aI., 1997; Santiago et aI., 1998; Meinhart et aI., 1999). For
detailed analysis, visualization using X-rays generated by a synchrotron
are used. An advantage is that X-rays can be utilized for visualizing
flows behind certain silicon structures, such a..~ polysilicon, which are not
transparent to visible light.
6 1. Introduction

1.2 Thesis structure and objectives

1.2.1 Structure of the presented work

This work starts with an introduction in the theory of multiphase flows.


The main equations, describing the flow behavior, as well as the driving
forces of gas-liquid flows are presented. The flow phenomena for mul-
tiphase flow in small scale channels as well as the consequences on the
mass transport are highlighted. Importance is attached to the possibil-
ities to characterize the two-phase gas-liquid flow in microchannels. In
chapter 3 the experimental setup and the applied measurement methods
are explained. The following chapters present detailed results on the ex-
perimental data obtained for the determination of the multiphase flow
pattern (chapter 4), the results of the pressure drop measurements (chap-
ter 5), and the data on the void fraction (chapter 6). Special attention
is turned to the comparison with existing correlations and other experi-
mentally derived data. In chapter 7, the recirculation motion inside the
liquid phase of an intermittent flow is analyzed with the help of velocity
measurements. The reference reaction, carried out in the microreactor,
is presented in chapter 8. The thesis ends with the conclusions and the
outlook to possible further work.

1.2.2 Project goals

In this study, we will identify methods to characterize gas-liquid two-


phase flows relevant for chemical reactions in rectangular and optically
transparent microchannels with channel hydraulic diameters in the range
D h < 1 mm. The project goals according to the research plan were:

An experimental setup for the characterization of flow properties


in microchannels is established.
Detailed measurements of the flow pattern, the pressure drop, and
the void fraction for a wide range of flow conditions. The experi-
mental data is validated with literature data. For the flow visuali-
zation Laser-Induced Fluorescence (LIF) is used.
1.2 Thesis structure and objectives 7

3-dimensional, time-averaged flow visualization and void fraction


determination using X-Ray Tomographic Microscopy (XTM).

The velocity distribution inside the liquid phase as well as the


internal recirculation motion is measured using micron-resolution
Particle Image Velocimetry (J.LPIV).
Carry out a highly exothermic reference reaction at well-defined
gas-liquid flow conditions.
8

Chapter 2

Basics and state of the


art

2.1 Multiphase flow

2.1.1 Basics of multiphase flows

The physical behavior and properties of single-phase flows of Newtonian


fluids are well understood, at lea.."3t for laminar flows. Not only most of
the fluid properties are known, there are also fluid-dynamic formulas
describing the kinetics of single-phase flows. In technical applications
however, mostly multiphase flows are present. Examples are evaporation
and condensation processes in steam-power, chemical or refrigeration
facilities, multiphase flows in pipes, gas washing, rectification, extraction,
sedimentation or fluidized bed applications.
Multiphase systems are systems where phase boundaries are present.
The boundaries can appear as phase interfaces between two different
aggregate conditions of the same fluid (e.g. water - steam) or between two
different fluids. In the case of two different fluids, the aggregate condition
can be equal (e.g. water - oil) or different (e.g. water - nitrogen). Hence
2.1 Multiphase flow 9

at least two phases define a multiphase system, even though three or


more phases are possible.
For two-phase systems, the following combinations are possible:

gas - liquid
gas - solid
liquid - liquid
liquid - solid

solid - solid

Pure solid - solid systems occur in process engineering for example in


coating processes. Gas mixtures are always single-phase. Single-phase
systems are per se homogeneous, meaning that the physical properties
are the same at any point of the mixture. In the case of multiphase sys-
tems, phase boundaries with different physical properties and conditions,
as for example the viscosity or density, are present. Those systems are
heterogeneous.

2.1.1.1 Main equations for multiphase flows

The characteristic of multiphase flows is that the two phases can show
different flow velocities. Generally, the gas velocity is higher than the li-
quid velocity. This phenomena, called phase slip, requires the differentia-
tion between mass fraction and mass transport fraction, and the volume
fraction and volume transport fraction, respectively. This is equivalent
to a differentiation between the ratio of the amounts of fluid in a part
of a channel (volumetric content or void fraction) and the ratio of the
amounts of fluid flowing through a cross-section of the channel.
The following definitions are commonly used to describe the phase frac-
tions in two-phase gas-liquid flows:
10 2. Basics and state of the art

1. Mass fraction of phase i:

. M~
Xi=---- with (2.1)
Ma+ML

2. Volume fraction of phase i:

with (2.2)

Generally, Ci is considered as void fraction or volumetric content


of phase i.
3. Mass transport fraction or quality of phase i:

with (2.3)

4. Volume transport fraction of phase i:

with (2.4)

The phase slip, s, is defined as the ratio of the actual velocities of the
two considered fluid phases.

Wa
s=- (2.5)
WL

As known from single-phase fluid dynamics, dimensionless numbers de-


scribing the physical process are used for reasons reducing the occurring
variables. Most of the dimensionless numbers known from single-phase
applications are used also for multiphase flows. In the following, some of
the most relevant numbers are presented.

Wi D
Reynolds number Rei = - - (2.6)
Vi
2.1 Multiphase flow 11

w? rh?
F'roude number Fri = _1,_ = 1, (2.7)
gD {l~gD

QD
N usselt number NUi= - - (2.8)
A"1,

Weber number (2.9)

The written indexes, i = (L, G), can be used for both, gas and liquid
phase. Furthermore, rhi = lVIi/A stands for the mass flux [kg/m 2 s], a
is the heat transfer coefficient [W/m 2 K], Ai is the heat conductivity
[W/ mK], and a denotes the interfacial tension between the two phases
[N/m]. D stands for the pipe diameter. In case of rectangular channels,
the hydraulic diameter D h is considered,

4A
Dh = d 2 b = 150 ... 266.67 /-lm, (2.10)
2 + .
where A denotes the cross-sectional area of the channel, d is the channel
depth and b the channel width.
For the formation of the above mentioned dimensionless numbers, either
the phase velocities, Wi, and therewith the mass flux, rh i , are considered
using the effective channel cross-sectional area flown through by the fluid
(which is smaller than the total cross-sectional area), or using the total
cross-sectional area. In the latter case, the superficial velocities ji are
considered.

. Vi
Ji = - (2.11)
A
The pha.."e distribution of a multiphase flow is influenced by several forces
as

gravity
12 2. Basics and state of the art

pressure forces
buoyancy forces
interfacial tension forces

friction forces.

The interfacial tension, a, between the two phases tempts to form the gas
phase into spherical shape, and the liquid droplets into round droplets
counteracted by spatial pressure differences, friction forces, or buoyancy
forces.

2.1.2 Multiphase microfluidics

Continuous flow in larger dimensions are fully described by the Navier-


Stokes equations. What happens, if the geometries tend towards smaller
dimensions? In microfluidic applications, phenomena become important
that can be neglected for large-scale systems (Gravesen et al., 1993; Gad-
el Hak, 1999; Stone and Kim, 2001). In contrast some physical properties
can be disregarded in microfluidic flows.

2.1.2.1 Low-Reynolds-number flows

Microchannels manufactured to date have dimensions Dh < 1 mm, and


flow speeds in the order of 1 cm/s (though they are varying widely),
which yield Reynolds numbers Re = {lW D h /17 < 30, where f2 and 17
denote the density and the viscosity of the fluid, and wand Dh are
the fluid velocity and the channel hydraulic diameter. This means that
viscous forces dominate the response and that the flow remains laminar.
Investigations of low-Reynolds-number flows are of great interest to the
fluid-dynamics realm as well a..'i to the chemical engineering community.
The most important new theme introduced by the small length scales
of microfluidic devices are the significant role of surface forces (sur-
face tension, electrical effects, van der Waals interactions, and surface
2.1 Multiphase flow 13

roughness), complicated three-dimensional geometries, and the possibil-


ity, that suspended particles have dimensions comparable to those of the
microchannel (Stone and Kim, 2001). One of the challenges in micra-
chemical engineering is to creatively use or tailor these effects to produce
functional devices.
The motion of Newtonian fluids in these small-scale systems are driven
by the following:

1. Applied pressure differences. The velocity distribution is parabolic


across the channel, as familiar from the classic Poiseuille flow.

2. Electric fields. The flow is referred to as electra-osmotic when bulk


fluid motion is driven by stresses concentrated in charged layers
near boundaries. The velocity profile is generally uniform.

3. Capillary driving forces owing to wetting of surfaces by the fluid.


This leads to pressure gradients in liquids, so it is similar in many
ways to item 1, though the shape of the interface may itself be
important.
4. Free-surface flows driven by gradients in interfacial tension
(Marangoni flows). These can be manipulated using the depen-
dence of the surface tension on the temperature or the chemical
concentration (Gallardo et al., 1999).

2.1.2.2 Transport phenomena

Especially in chemical engineering, the better understanding of the local


fluid-dynamics can produce novel, better and more efficient designs of
microchemical reactors. An important factor for the design and perfor-
mance of microdevices are transport, mixing, separation and fluid ma-
nipulation processes, and the influence of surface forces as the devices
are scaled down.
The interaction of convection and diffusion is crucially important in
many applications, especially those involving chemical reactions. Mi-
xing in pressure-driven microchannels occurs much more rapidly than
14 2. Basics and state of the art

predicted if only -molecular diffusion is considered. This dispersion oc-


curs because thermal fluctuations allow the suspended fluid to sanlple
streamlines that have different speeds. Of course, to utilize device volu-
me more efficiently, the channels are curved and some designs even use
three-dimensional networks.
Mixing is important, but appears more difficult at small scales because
the familiar use of turbulence is unavailable. Often molecular diffusion
is not sufficient at given flow speeds and channel lengths. It is therefore
reasonable that enhancing the mixing of two fluids will be aided sub-
stantially by chaotic paths of the fluid. Microdevices are currently being
designed to take advantage of these manners of mixing (Gad-EI-Hak,
2004; Loseyet al., 2002; Stone et al., 2004; Wang et al., 2002, 2003).

2.1.3 Characterization of multiphase gas-liquid flows

Chemical reactions in pipes or channels require detailed knowledge of


the conditions inside the reaction system. Especially the conditions in
small scale channels (microchannels) vary strongly depending on the gas
and liquid volumetric flow rates. The most important characteristics of
two-phase gas-liquid flows in microchannels are the flow pattern, the
pressure drop and the volumetric gas content.

2.1.3.1 Flow pattern

Multiphase flows, where interactions between the two involved phases


are present (e.g. relative velocities, changes in momentum), are called
heterogeneous flows. Depending on the volumetric flow rates of the two
phases, different flow patterns can be observed. The regimes vary in
the phase distribution, the interfacial area, the fluid velocities, and the
pressure drop (see chapter 2.1.3.2). In large diameter pipes (Dh > 1 cm),
the gravity force is dominant. Hence, in horizontal pipes, the fluid with
the higher density is mostly found in the lower part of the channel, the
gas on the other hand is flowing on top of the liquid. This phenomenon
is observed mainly for low fluid velocities. For fluid systems with low
liquid content and high gas velocities, the forces resulting from the flow
2.1 Multiphase flow 15

outweigh the gravitational force. In this case, flow patterns are observed
that can also be found in vertical flows.
In the intermittent flow regime (slug flow and plug flow) the gas phase
is concentrated in elongated bubbles taking the bigger part of the cross-
sectional area. Intermittent flow is characterized by alternating liquid
and gas segments flowing in the channel.
At high gas velocities, the flow minimizes the energy dissipation due
to friction pressure drop and forms an annular flow pattern. The liquid
phase is pushed towards the channel walls, the gas phase is flowing at
high velocities in the liquid core.
In the case of small hydraulic diameters, as found in microchannels, the
ratio of channel surface to channel volume increases. This leads to a re-
lative increase of the surface forces (chap. 2.1.2) compared to gravity. In
channels with diameters below 1 mm only bubbly (relatively large bub-
bles), intermittent and annular flow are observed for a water-air system.
To predict the flow pattern of a fluidic system, flow regime maps are used.
In these maps, the expected flow regime is plotted a.." a function of the
gas and liquid mass fluxes, gas and liquid superficial velocities or other
flow constants. Commonly used flow regime maps for large diameter
pipes are constructed using a water-air flow at low pressures and fully
developed flows. Especially for fluids differing strongly from the phy-
sical properties of water and air, for large differences in the hydraulic
diameter, and for high pressure applications, the prediction of the flow
pattern results in substantial errors. The lines in the flow reginle maps
indicate mean values for the transition region between two flow regimes.
A sudden change in regimes for small parameter variation does not take
place.
In the early 1950's, Baker (1954) constructed a flow pattern map for
two-pha.."e gas-liquid flow in large diameter pipes.
In this diagram, the value Ma/().:rrD 2 /4) is plotted against the dimen-
sionless ratio MLAW / Ma, where

A = [( fla) (~)]
1.2 1000
1/2 (2.12)
16 2. Basics and state of the art

100
6
4
BUbbly
2

-
'b
~

t::
10
6
4


-
1$
Cl
:::i!:
2

6
4 Stratified
2

0.1
2 468 2 466 2 468
10 100 1000
(A. M L ' I M G)

Figure 2.1: Flow regime map by Baker (1954) for horizontal pipes.

(2.13)

These values are dimensionless parameters, .x is the Baker density param-


eter, and W the surface tension parameter. Both values are normalized
for the corresponding values of water and air.
Other regime maps for large scale channel flows are those by Mandhane
et al. (1975) and Taitel and Dukler (1976) which are not discussed in
detail here.
Only a few experimental studies describing the two-phase gas-liquid flow
characteristics of microchannel flows have been reported so far. Due to
the major influence of the surface forces, the flow regimes for hydraulic
diameters below :::::: 1 mm are expected to be different from those of large
scale systems. Especially the influence of the channel cross-section, the
fluids' viscosities, and the surface tension have a major impact on the
transition lines. Stratified flow is not detected in any regime map for
microfluidic applications. Micron-scale channel flow pattern maps exist
2.1 Multiphase flow 17

by Coleman and Garimella (1999, 2003); de Mas et al. (2003); Kawahara


et al. (2002); Triplett et al. (1999b); Zhao and Bi (2001).

2.1.3.2 Pressure drop

The detailed knowledge of the pressure drop of a technical application is


of great interest. In the field of microchemical applications, especially for
large channel arrays or longer channels, the pressure drop is influencing
the design of the system. As seen for the flow regime, several accurate
pressure drop predictions exist for large diameter pipes. The pressure
drop of large-scale multiphase systems can be calculated without the
knowledge of the flow pattern. In contrast, the improvement in accurracy
is disproportional to the required effort.
Models for the pressure drop calculation can be classified into homo-
geneous and heterogeneous models. Homogeneous models assume equal
velocities for both phases (slip s = 0). These predictions are accurate
only for fine dispersions. Better results are obtained using the hetero-
geneous slip model, which solves separate equations for the two phases
and allows a slip smaller than 1.
The total pressure drop in a pipe section dz

dp dPfric dpacc dpred dPgrav


-dz = -d-z- + -d-z- + -d-z- + ~d'--z- (2.14)

is composed of

the friction pressure drop

dPfric
--- (2.15)
dz

the pressure drop due to acceleration

dpacc
--- = -
d ( (]L' 2 (1- EO) + (]o wo'
wL'
2EO) (2.16)
dz dz '
18 2. Basics and state of the art

the pressure drop due to a reduction in diameter (formula depend-


ing on geometry),
and the gravitational pressure drop

- dPgrav
dz = 9 ( (
{}L' 1 - EG
)
+ {}G EG ). s'l.n<I>.
. (2.17)

For large-scale vertical flows, the gravitational pressure drop is the major
factor in the total pressure drop. For horizontal flows on the other hand
(inclination angle <I> = 0), this portion can be neglected.
Friction causes a pressure decrease along the direction of flow. If a gas-
phase is present, it will expand due to the pressure dependence of the gas
density. The following increase in velocity, according to the conservation
of mass, leads to an additional pressure drop due to acceleration. In
most cases, the pressure drop due to acceleration is a lot smaller than
the frictional pressure drop and can be neglected.
By far the largest contribution to the total pressure drop is provided by
the frictional pressure drop. For multiphase flow in large diameter pipes,
several methods for the estimation of the friction pressure drop exist.
In most cases, the phase distribution and the flow pattern are neglected
and the pressure drop results are therefore inaccurate. Among the most
established calculation methods are those by Lockhart and Martinelli
(1949); Chawla (1968); Friedel (1978), and Muller-Steinhagen and Heck
(1986).

2.1.3.3 Volumetric gas content

Generally, the gas and liquid velocities of a multiphase gas-liquid flow


in pipes or channels are not equal. According to Eq. 2.5, the slip is not
equal to 1, S 1= 1. This is a byword for non-homogeneous flow. For most
technical applications, the mass flux, the pressure drop, and the tem-
perature distribution inside the reactors are known from measurements.
Hence the mass transport fractions, xG and XL, as well as the volume
transport fractions, EG and EL, are known. The local mass fractions (xG
and XL), and the local volumetric gas content (EG) are measurable only
2.2 Hydrogenation of nitrobenzene in microchemical devices 19

with extravagant expenses. These values are connected to the transport


fractions as in Eq. 2.18

Ea(l - Ea) xa(1- Ea) ilL


s- - - (2.18)
- Ea(l - Ea) - Ea(1- xa) Ila'
The knowledge of the slip or the void fraction, respectively, is important
for the determination of residence times or the calculation of heat and
mass transport processes. For multiphase flows, mostly empirical corre-
lations are reported, where the transport fractions and fluid properties
deliver an estimation of the slip or the void fraction.
Armand and Treschew (1946); Massena (1960); Lockhart and Martinelli
(1949); Wallis (1969) and Chawla (1968) presented models for the cal-
culation of the local volumetric gas content for multiphase flows in large
diameter pipes.

2.2 Hydrogenation of nitrobenzene in mi-


crochemical devices

2.2.1 Basic principle of hydrogenation reactions


Catalytic hydrogenation refers to the addition of hydrogen to an or-
ganic molecule in the presence of a catalyst. These reactions are used to
produce a variety of both bulk and fine organic chemicals. In addition,
hydrogenation is often used in various purification processes. Hydrogena-
tion is an exothermic reaction and the equilibrium usually lies far toward
the hydrogenated product under most operating temperatures:

!:i.H < O. (2.19)


Hydrogenation is carried out in both the liquid and the gas phases,
but liquid-phase reactions are much more common. In liquid-phase sys-
tems, transport of hydrogen to the catalyst surface is frequently the rate-
limiting step. To react with the substrate, hydrogen must move from the
gas phase into the liquid and then to the solid catalyst.
20 2. Basics and state of the art

The mostly heterogeneous catalysts can be divided into two types: those
used in fixed-bed processing, in which the catalyst is stationary and the
reactants pass over it, and those used in fluidized-bed or slurry process-
ing. Many metals and metal oxides have general hydrogenation activity,
but palladium, platinum, rhodium, and ruthenium make exceptionally
active hydrogenation catalysts and are used frequently.

2.2.2 Hydrogenation of nitrobenzene to aniline

The hydrogenation process converts nitrobenzene to aniline. Commonly


used catalyst for this reaction is palladium. The course of hydrogenation
of nitroaromatic compounds is complex, and a variety of products can
be obtained. ReaCtion is best under mild conditions in neutral media.

0+
nitrobenzene

C6HsN02
3 H2
-
Pd/C5%

6
aniline

C6H7N
+ 2Hf)

The hydrogenation of nitrobenzene to aniline and other byproducts was


investigated by several research groups in the last years (Yu et al., 2000;
Gelder et al., 2002; Holler et al., 2000; Figueras and Coq, 2001; Rode
et al., 2001). The reaction is usually carried out in isothermal autoclaves
(slurry reactors). Stirrers are used for intensive mixing. Since the reac-
tion produces water, the preferred support of palladium is carbon due to
its hydrophobicity. Typical reaction temperature (if carried out isother-
mally) is 323 K (50C) and constant pressures between 4 to 15 bar. The
time for the complete conversion from nitrobenzene to aniline depends
on the concentration and is in the order of several minutes.
2.2 Hydrogenation of nitrobenzene in microchemical devices 21

2.2.3 Gas-liquid microreactors

Only recently the use of microstructured chemical reactors for gas-liquid


reactions has been investigated. The advantages of the small channel
size, as the improved surface-to-volume ratio, is exploited for the chem-
ical processes. Jahnisch et al. (2000) presented gas-liquid microreactors
(falling film reactor and micro bubble column) for the direct fluorination
of toluene. Losey et al. (2001) developed a microfabricated multiphase
packed-bed reactor for the hydrogenation of cyclohexene. A second de-
vice involves the incorporation of porous silicon as a catalyst support, in
the form of a thin layer covering microstructured channels (Losey et al.,
2002).
Experimental studies of nitrobenzene hydrogenation in a microstruc-
tured falling film reactor are presented by Yeong et al. (2004). An array
of 64 straight parallel channels (300 Mm wide, 100 Mm deep, 78 mm
long) were used. The microchannels were coated with palladium (II) ni-
trate solution and allowed to dry. Experiments were conducted at 60C,
1 - 6 bar inlet pressure and liquid flow rates of 0.5 - 3 ml/min nitroben-
zene. The conversion was found to be affected by both liquid flow rates
and hydrogen pressure.
22

Chapter 3

Experimental setup and


measurement methods

In this chapter, the experimental setup, the microchannel design and


fabrication, as well as the measurement methods are presented.

3.1 Experimental setup

3.1.1 Microchannels
3.1.1.1 Manufacturing methods

Silicon micro-electro-mechanical systems (MEMS), as well as microflui-


dic and microchemical applications are fabricated using a photolithogra-
phy and an etching step. Photolithography is the process sequence that
transfers an image (channel layout/design) from a master to the sur-
face of the wafer by light and etching. As master or mask, a glass plate
of opaque (chromium) and transparent regions is used. A high pressure
mercury lamp (436 and 365 nm) exposes the wafer covered by photosen-
sitive resist. Photo resist is composed of organic solvent, light sensitive
3.1 Experimental setup 23

inhibitor, and resin (e.g. positive photo resist) and used mainly as a mask
for transferring a pattern into an underlying layer by wet or dry etching.
The masks are written using e-beam (or laser beam) exposure machines,
so-called pattern generators. The layout data is digitized and drives the
pattern generator, that directly transfers the patterns to the photosen-
sitized masks. 5-inch wafer masks with little data, large structures, and
limited accuracy (as required in the present experiments) are very cost-
intensive. A cheaper alternative is the use of exposed transparency film.
For the formation of microchannels (bulk micromachining) different fab-
rication methods are available, wet etching techniques and dry etch-
ing methods, respectively. Wet etching techniques are mostly isotropic.
Special etching solutions (e.g. Potassium Hydroxide (KOH), Ethylene
Diamine Pyrocatechel (EDP) or Tetramethyl Ammonium Hydroxide
(TMAH)) etch single crystalline silicon with different etch rates along
different crystal directions. Microchannels fabricated using a wet etch-
ing method often feature triangular or semi-triangular channel cross-
sections. The width to depth ratio is limited by the crystalline composi-
tion of the silicon. The Reactive-Ion Etching (RIE) technique is the most
established dry etching method. Reactive ions are generated in a plasma
discharge. An additional physical activation energy supply assists the
chemical etching process. The anisotropy of this process can be adjusted
by an electrode bias. RIE provides etch rates between 100-1000 nm/min.
An expansion of the RIE technique is the Deep reactive-ion etching
(DRIE) method. DRIE is advantageous for anisotropic dry etching of
silicon for deep structures up to 1000 J.Lm. This is enabled by a high
density (inductively coupled) plasma source and an alternating process
of etching and protective deposition (multiplexing, Fig. 3.1B). Typical
performance data are etching rates between 1 - 3 J.Lm/min, aspect ra-
tios up to 30 : 1, a maximum etch depth of approximately 1 mm. Main
advantages compared to the wet etching techniques is the anisotropy,
which is independent of the crystal orientation, and the possibility of
manufacturing straight sidewalls with high aspect ratios (Fig. 3.1A). In
contrast, DRIE is a single wafer process, which means the process is time
consuming.
All the microchannels and microreactors presented in this work are fabri-
cated in the clean room facilities of the Institute of Mechanical Systems
24 3. Experimental setup and mea..'3urement methods

i id: B
Figure 3.1: SEM image showing the rectangular channel cross-
section with straight sidewalls (A). Due to the relatively high aspect
ratio, the channel ground in the center is etched slightly deeper than
close to the sidewalls (b:.d ~ 10%). On the channel sidewalls, the
single etching steps of ~ 1 pm lead to streamwise trenches (B).
3.1 Experimental setup 25

(IMES) at ETH Zurich. We used double sided polished silicon wafers


with 4" in diameter and 525 J.Lm thickness and the standard photolitho-
graphy technique (positive photo resist) in combination with dry etching
(Deep Reactive-Ion Etching, DRIE) to form the channels. For economi-
cal reasons, due to the time consuming fabrication method, and because
of the fact, that our microstructures have a minimum spatial dimension
of several tens of micrometers, we abandoned the use of high quality
chromium etching masks and used polymer films instead. The exposure
resolution of these films was 6400 dpi, which results in a virtual surface
roughness of k s = 3.9 J.Lm. After the fabrication of the microchannels on
the wafer front side, the same procedure (photolithography and etching
process) is repeated for the inlet and outlet holes on the wafer's backside.
For providing an optical access to the channel, the silicon wafer's front
side is sealed to a glass wafer (4" wide and 700 J.Lm thick) in by anodic
bonding.
We fabricated 6 microchannels with different widths or layouts per 4"
wafer. After completion of the fabrication procedure (photolithography,
etching, and bonding), the 6 independent channels are separated by a
wafer saw. Obviously each layout of the same wafer has the same depth,
because the whole wafer was etched a certain time. Finally, each mi-
croreactor chip has outer dimensions of 46 x 18 x 1.225 mm (length x
width x height).

3.1.1.2 Design and layout of the microchannels and reactors

Flow characterization
The most simple setup for a two-phase gas-liquid microfluidic applica-
tion is aT-shaped channel. The two fluids enter the microchannel, meet
at the T-junction, and form a multiphase flow pattern in the long com-
mon channel section towards the multiphase outlet. Test experiments
with this first configuration showed pulsating, unsteady flow, especially
for wide channels and slow superficial velocities. This disadvantage was
improved by the implementation of a static mixer at the mixing zone of
the T-shaped channel. A small number of round pins with 50 J.Lm in di-
ameter were included (Fig. 3.2A), slightly increasing the pressure drop.
26 3. Experimental setup and measurement methods

Unfortunately still pulsating, alternating feed of liquid and gas instead of


continuous quasi-stationary multiphase flow was observed. To overcome
this circumstance, additional artificial pressure drop traps (constrictions)
at the single-phase gas and liquid inlet channels were implemented (Fig.
3.2B). This final inlet configuration ensured a sufficient continuous mix-
ing of the two phases and therefore steady multiphase flow pattern.

\ ,
oL
oor
000
000
00

38 mm
I I
-
,\
, \c"
\
B
38 mm
I

---;55:;;;;__;;-;;.-;;; ..... .........._.,,,h'o ......


........................__ ... __ .., ....

oor
00
000
000

I
00

/)
Figure 3.2: Schematic of the used channel geometries and enlarge-
ment of the two-phase mixing zone. T-shaped channel with static
mixer to avoid pulsating flow (A), and V-shaped final design of the
microchannels (B).

The microchannels have different widths and depths. Experiments were


made using channels with widths between 150 - 400 J.lm and different
3.1 Experimental setup 27

depths (100, 150, and 200 J-lm). This results in hydraulic diameters

4A
Dh = 2 d +2 b = 120 ... 266.67 J-lm. (3.1)

The length of the multiphase part of the microchannels is uniformly


38 mm, which results in a length, l, to diameter, D h , ratio of

l
D = 142.5 ... 316.7. (3.2)
h

Hydrogenation reaction
The chemical reaction (catalytic hydrogenation of nitrobenzene) makes
great demands on the design of the microreactor. Even tough the mass
and heat transfer rates are higher than in commercial batch reactors, the
reaction process requires a certain residence time. This can be realized by
slower velocities or longer channels. As slower fluid velocities are assumed
to result in less internal recirculation, we concentrated on the alternative
of longer reactors. This is realized by meandering channels.
The hydrogenation of nitrobenzene requires the presence of Palladium
as catalyst. Two different possibilities for the contacting of the educts
with the catalyst are considered, the packed-bed method and the sur-
face coating option. For the packed-bed method (see Fig. 3.3A), activated
carbon particles, covered with Palladium, are inserted directly in the mi-
croreactor. Thus the nitrobenzene and the hydrogen will flow through
this catalyst fill. To prevent the catalyst particles from leaving the mi-
croreactor or from flowing back into the inlet tubing, obstacles at the
beginning of the multiphase part of the reactor and in front of the outlet
are included in the channel design (Fig. 3.3B). To enable the heating
of the flow (reaction requires 50C) and to improve the fluid mixing,
part of the inlet is equipped with mixing pins (as in the characteriza-
tion version of the microchannels, Fig. 3.2B and C). The presence of the
packed-bed of the particles causes a high pressure drop over the reac-
tion channel. Hence the length is limited by the fluid delivery equipment
and the channel sealing. The decreased residence time due to the limited
channel length is compensated by the extraordinary fluid mixing and
28 3. Experimental setup and measurement methods

the excellent catalyst surface to volume ratio. Reactors with different


lengths and hydraulic diameters were fabricated.

Fl " " "'" u.\


Mixing and heating Particle barriers Product outlet

i\. ~~
Educt \.I rf/ //
,
1'/ [ """".

.' ~ Ul \ A

c:=::J
c:=:J
--
c==J
c:J
c==J
c:::::J
-----,Ij
c:::::J
- B c
UqU~~ inlet Hea~?g zone Product outlet

[(umum;/
: .
.''

~.:;,.: ~::Jt-~~

Gas inlet D

Figure 3.3: Layout of the microreactor using a packed bed of cat-


alyst particles between the two barriers (A), and an enlarged view
of the particle barrier (B) is shown. The design of the palladium
covered microreactor (D) and the special gas injector (C) .

A second option for the catalyst-fluid contacting was realized by coating


the surface with a thin layer of Palladium (layout: Fig. 3.3D, coating:
Fig. 3.4). Therefore the channel base and side walls are covered by a
3.1 Experimental setup 29

60 nm thick Palladium layer using the Plasma enhanced Vapor Depo-


sition method (PVD). For a better adhesion of the Palladium to the
silicon, the silicon channel is first covered with a titanium layer of 25 nm
thickness. Due to the absence of a packed-bed, and therefore a reduced
pressure drop, these microreactors could be made longer (improved res-
idence time) and thinner (smaller hydraulic diameter), which improves
the surface-to-volume ratio and thus also the heat flux to and from the
channel. To ensure the proper preheating of the educts even at higher
velocities, the heated channel length is enlarged. This channel design has
the major disadvantage, that the internal mixing and the catalyst sur-
face is a lot smaller in respect to the packed-bed option. Reactors with
different lengths and hydraulic diameters were fabricated. Extra chan-
nels were considered for the temperature measurement (thermocouple
insert, Fig. 3.5).

200IIm EHT" 7.00kV DD :23 Dec: 2004


*ll .. ,OO.OOKX H wo- 5nwn T_ :9:60:20

Figure 3.4: SEM image of the titanium and palladium coating


(total thickness ~ 65 nm) on the microreactor ground.
30 3. Experimental setup and mea..'3urement methods

Figure 3.5: Photograph of two different options for the coated


microreactor.

3.1.2 Channel holder

To feed the microchannels described in chap. :3.1.1 with the two fluids
of the gas-liquid multiphase flow, a connection using tubing was needed.
The tubing and especially the connection to the microchannel has to
fulfill several n~quirements, such as tightness, chemical resistance, or the
possibility to be cleaned. The most common method (e.g. Losey et al.
(2001); de Mas et al. (2003)) is to glue the tubing air tight to the channel.
Main advantage of this connection is the easy production and the tight-
ness. The cleaning of the tubing, especially from seeding particles (see
chap. 3.:).2), can be difficult. Furthermore, if a large number of different
channel layouts are considered, the glued connection may be unhandy.
For this reason wc developed a channel holder.
The microchanm-)l can be placed inside the lower part of the channel
holder. An optical access is enabled by the opening in the bottom side of
the hokler. As scaling, a GoreTex or a Neopren layer with holes for
the fluidic connections is used. Both materials are largely chemically re-
sistant and feature an excellent tightness up to 6 bar fluidic pressure. The
channel is tightened by screwing the upper part to the bottom (compres-
sion scaling). Due to the hardness of the GoreTex sealing, the chance of
breaking a microchannel when tightening the screws is higher than when
3.1 Experimental setup 31

Fluidic connections

Screw holes -----,~.

Channel holder (upper part)

Sealing

Microchannell -reactor

Channel holder (lower part)

Figure 3.6: Channel holder with sealing and connections for the
tubing and the tightening screws.

using the Neopren sealing. The upper part of the holder features three
fluidic connections, one for the gas inlet, one for the liquid inlet, and one
for the multiphase flow outlet. The connections are fully compatible to
the Upchruch tubing connections. For the chemical reaction, an addi-
tional connection was applied for introducing the catalyst particles (see
chap. 8).
The channel holder's dimensions (length and width) correspond to a
commonly used microscope slide. Hence it is possible to clamp the holder
into the corresponding position of the microscope.

3.1.3 Visualization and measurement equipment

An inverted fluorescence microscope (Zeiss Axiovert 200) equipped with


a Rhodamine B filter set and with a dual-frame low noise CCD cam-
era (PCa SensiCam QE, 9 frames per second, 1376 x 1040 pixels) was
used for imaging. A frequency-doubled Nd:YAG laser (TSI, 25 mJ per
pulse, A 532 nm) is used for the illumination. The two camera
frames and the Nd:YAG laser were synchronized using a commercial
timing unit (LaVision). Two different microscope objectives were used
32 3. Experimental setup and measurement methods

A c

Figure 3.7: Channel holder with microchannel in the lower part


(A). Closed holder from top with fiuidic connections (B) and from
the bottom with the optical access to the microchannel (C).

(5x and 10x). Maximum spatial resolution of the experimental system


was 0.54 /-ull/pixel (area of view 0.74 mm x 0.56 mm).

Figure 3.8: Experimental setup with the Zeiss Axiovert 200 micro-
scope (left) and the Nd:YAG laser with the light connection to the
microscope (right).

For the presented two-phase gas-liquid experiments, the liquid phase is


dosed by a syringe pump (Harvard Apparatus, accuracy within 0.35%
3.1 Experimental setup 33

and reproducibility within 0.05%). The volumetric flow rate of the liquid
phase, VL, is varied between 0.05 and 5 ml/min for the flow characteri-
zation experiments. Gas-phase container was a commonly used 200 bar
pressurized bottle using pressure reducing valves. The volumetric flow
rate, VG, was controlled by a mass flow controller (Bronkhorst, accuracy
0.5% of reading plus 0.1% full scale) and is varied between 0.5 and
20 ml/min. The fluids are connected to the channel (channel holder) by
flexible tubing (Upchurch, PEEK, 1/16 inch outer diameter and 0.5 mm
inner diameter).
Pressures are measured, at the inlet and outlet of the microchannel
(single phase inlets, multiphase outlet), by pressure sensors (Endress +
Hauser, accuracy 0.5% of full scale). The multiphase flow at the outlet
is fed into a sample container at ambient pressure.

'\-O-----L_.....1 Micrjannel I

r ,..1

.....D<II----JL..-----L-
j
-~-~:-:=~::j'.;-I
I I
I I

LJ.. . 1
I
I

El..~ CCD camera I

Figure 3.9: Schematic of the experimental setup of the fluid supply.

3.1.4 Fluids

For the presented two-phase gas-liquid experiments fluids of different


properties were used. The flow inside a microchannel is mainly af-
fected by the fluid surface tension and the viscosity. To obtain signif-
icant information on the influence of the fluid properties, pure deionized
34 3. Experimental setup and measurement methods

water, ethanol and aqueous glycerol solutions (10% and 20% glycerol
mass fraction) as liquid phase and nitrogen as gas-phase were used. All
the characterization experiments were carried out at room temperature
(T = 25C). Water and ethanol have similar values for the viscosity,
but differ by a factor of approximately 3 in surface tension towards ni-
trogen (see Tab. 3.1). The glycerol solutions on the other hand features
a surface tension close to the one of water, but has differences in the
viscosity.

Fluid Density Viscosity Surface Tension


3
fJ [kg/m ] TJ [mPa s] a [mN/m]
Deionized Water 997 0.899 71.99
Ethanol 789 1.074 21.97
Glycerol (10%) 1021 1.1470 70.50
Glycerol (20%) 1044 1.5270 69.50
Nitrogen 1.20 0.0179 -

Table 3.1: Properties of the used fluids

3.2 Experimental setup for the hydrogena-


tion reaction

A few adaptations to the above presented experimental setup had to be


made for conducting the hydrogenation reaction in microchannels. The
design of the microreactors is presented in chapter 3.1.1.2. Due to the
necessary higher pressures a stainless steel syringe (Harvard Apparatus,
8 ml) was used. The hydrogen is delivered by a pressurized bottle. To
prevent major pressure variations inside the reaction system, the pres-
sure was ensured and controlled using a back-pressure regulator (VICr
Jour, PEEK, 0 - 7 bar) or a needle valve. Microfilters are implemented
preventing the reaction zone to be clogged by impurities, and the cata-
lyst powder to reach the sensitive flow feed equipment, respectively. As
the reaction may require a running-in period (e.g. heating-up), where
3.2 Experimental setup for the hydrogenation reaction 35

no aniline is produced, two storage bottles are considered to be selected


by a 3-way valve. Therewith, the collection of reaction product at well-
defined conditions can be ensured. A schematic of the reaction layout
can be found in Fig. 3.10.

Storage bottle

H2

Figure 3.10: Schematic experimental setup for the chemical reac-


tion.

The channel holder (sealing) was found to be insufficient for high pres-
I:mre experiments. Another way to feed the fiuids without leaking to the
microreactor had to be considered. The channel was covered with thin
layers of chromium (Cr), nickel (Ni), copper (Cu), and gold (Ag) byelec-
tron beam evaporation. Ag provides complete wetting with silver solder.
Cu prevents the formation of an oxide layer on the nickel layer, which it-
self acts as diffusion barrier. Cl' is the actual undercoating. Stainless steel
capillaries are soldered airtight to the reactor (Fig. 3.11). Experiments
showed that the current reactor and fiuidic connections can withstand
pressures up to 120 bar and temperatures up to 90C.
Different ways for heating the reaction zone to the necessary reaction
start-up temperature were considered, a heat-exchanging device, opera-
ted in counter flow mode, heating the complete microreactor including
the channel holder, heating by means of electric resistance, and the im-
plementation of microchannel heat exchanger integrated in the silicon
36 3. Experirnental setup and measurement methods

Figure 3.11: Microreactor with soldered stainless steel capillaries


for high pressure experiments.

chip, respectively. The latter was abandoned due to the limited space
for connecting the tubing in the channel holder. Due to the high spe-
cific heating power, the control of the required temperature, and the
small number of experimental adaptations (no additional sealing or fit-
ting devices), we decided the use of an active heating foil (Minco AG,
6.4 x 6.4 mm 2 , heating power 2.4 W). A foil thermometer and a DC
temperature control system (Minco AG) were used for the control of the
heating temperature.

Figure 3.12: Back-pressun~ n~gulator with PEEK fiuidic connec-


tions (left), and miniature DC temperature control system (right).
3.3 Measurement methods 37

3.3 Measurement methods

3.3.1 Laser-Induced Fluorescence (LIF)

Laser-Induced Fluorescence (LIF) is a nonintrusive technique for the


quantitative measurement of whole, instantaneous fields of scalar quan-
tities as species concentration (Fairbank et al., 1975; Kychakoff et al.,
1982) or velocity measurements (Hassa et al., 1987). Measurements based
on laser-induced fluorescence yield submillimeter spatial and submi-
crosecond temporal resolution. For common large scale applications, a
sheet of laser light, tuned to an optical resonance of the species being
probed, is transmitted across the flow field under study. The studied
fluid is mixed with a fluorescent dye. For multi fluid experiments (e.g.
mixing), only one of the fluids is made fluorescent. Part of the light is
absorbed by the enriched fluid and reemitted as fluorescence. The fluo-
rescence signal, SF, recorded by a CCD camera, can be related to the
absorber total number density, N r , via the equation

In this equation, C is a constant which includes geometric and efficiency


factors, I is the laser intensity, (1+Q/A 21 )-1 is the photon yield or Stern-
Vollmer factor, F(T) is the fraction of the total population in the stage
being pumped, Q is the electronic quenching rate, A 21 is the Einstein
coefficient for the transition, and g(w) is an appropriate line shape factor.
The proportionality factor between the fluorescence signal and the total
absorber number density can be regarded as constant provided that the
product of the Stern-Vollmer factor, the population factor, and the line
shape factor does not vary significantly throughout the flow field being
studied. For measurements carried out at locally constant pressure and
temperature, the signal strength can be related as linearly proportional
to the absorber total number density (Kychakoff et al., 1982),

(3.4)
38 3. Experimental setup and measurement methods

In contrast to large scale applications using a thin laser sheet for the
illumination of the flow field, laser-induced fluorelicence in microfluidic
experiments uses whole field illumination. LIF can therefore no more be
regarded as a 2-dimensional measurement method. The recorded reemit-
ted light is the sum of the intensities over the whole channel depth.
Hence, in the presented experiments, the LIF images are 2-dimensionally
averaged over the direction of observation.
In the present experiments, Rhodamine B (Fluka) was added to the li-
quid phase (0.5 ml of 25% Rhodamine B solution into 50 mlliquid, 0.25%
by volume). Rhodamine B emits red light (peak excitation wavelength
550 nm, peak emission wavelength 605 nm) when excited by 532 nm
laser light. On the recorded grayscale images, the liquid therefore is rep-
resented by bright colors, the gas-phase has no fluorelicent signal and is
therefore represented by black colors.
To eliminate the uneven laser and background intensity, the experimental
image with signal intensity SF is calibrated uliing the fluorelicent liignal of
a microchannel completely filled with liquid, SF,L, and the corresponding
signal of an empty channel (gas phase only), SF,G, according to Eq. 3.5.

SF - SFG
SF,cal = S S' (3.5)
F,L - F,G

According to this calibration technique, the calibrated signal assumes


values between 1 for 100% liquid and 0 for 100% gas.

3.3.2 Micron-resolution Particle Image Velocimetry


(/lPIV)
3.3.2.1 Basic principle of Particle Image Velocimetry (PIV)

Particle Image Velocimetry (PIV) is a well-established, optical, non-


intrusive method for the measurement of flow velocities (Adrian, 1991;
Adrian et al., 2000). In contrast to other methods (e.g. Laser Doppler
Velocimetry (LDV) or Particle Density Anemometry (PDA)) PIV is not
3.3 Measurement methods 39

a single point measurement method but determines the velocities in a 2-


dimensional plane. For that purpose, seeding particles (see chap. 3.3.2.3)
are mixed into the flow and their velocity is analyzed. In double-pulsed
PIV, positions of the flow-tracing particles are recorded at two known
times by either illuminating the particles using a pulsed light source, or
by illuminating the particles using a continuous light source and gating
the light near the camera using a mechanical or electronic shutter. The
displacement of the particles is then estimated statistically by correlating
the particle image pairs (Meinhart et al., 1993).
In large-scale applications, the flow field to be analyzed is illuminated
by a pulsed laser sheet. The particles scatter light onto a photographic
lens located 90 to the sheet, so that its in-focus object planes coincides
with the illuminated slice of fluid. The scattered light is recorded by
a camera and tra.nsferred to a computer for the analysis. The analysis
of the recorded image field is the most important step in the entire
process, as it couples with the image-acquisition process to determine the
accuracy, reliability, and spatial resolution of the measurements. Usually,
the image field is rich in information, therefore not the entire field can
be analyzed, but the local velocity in a small region of the image field is
found by analyzing these interrogation spots or interrogation windows.
The vector field of the whole image is obtained by repeating this process
on a grid of such interrogation spots.

"oo~"~
.....




~ .. ~

~
~

......... ....... 1 .... ........ ......

Figure 3.13: Schematic of the cross-correlation process.


40 3. Experimental setup and measurement methods

3.3.2.2 Micron-resolution Particle Image Velocimetry (j.LPIV)

The use of PIV for microfluidic applications requires several adapta-


tions to the experimental setup (Santiago et al., 1998; Meinhart et al.,
1999). The area of view becomes very small due to the reduced geome-
tries (in the range of some micrometers). Especially the demands on
the optical setup (camera resolution) and the seeding particles are chal-
lenging. The particles still have to be small enough to follow the flow,
without interfering with it. On the other hand, the particles have to be
large enough to scatter sufficient light to be detected by the camera. In
micron-resolution Particle Image Velocimetry (j.LPIV) the particles are
therefore mostly covered by a fluorescent dye. If a particle is illuminated
by the laser, it emits light of a certain wavelength that can be filtered
out. The optical setup has to be anlplified by a microscope objective.
The laser sheet used in standard PIV application can not be propor-
tionally thinner for the microfluidic applications. In contrast to measure-
ments in larger geometries, which can be regarded as truly 2-dimensional
due to the thin laser sheet, in j.LPIV the whole flow field is illuminated.
Therefore, also particle motion in the axis of view is detected. In j.LPIV,
the measurement domain is defined by the relatively small depth-of-field
of the NA = 1.25 lens. Given a criteria of less than 25% increase in par-
ticle diameter, the depth-of-field for the present system was measured
to be approximately 1.1 j.Lm. Particles outside this depth-of-field are at
least 25% larger than in-focus particles and have significantly lower im-
age intensities.
For the measurements presented here, we used the same experimental
setup as described in chap. 3.1.3 and in Fig. 3.9. The velocity components
were computed using a standard cross-correlation routine (Flowmaster
3 (DaVis 6.2), LaVision).

3.3.2.3 Seeding particles

In Micron-resolution Particle Image Velocimetry (j.LPIV), seed particle


size must be small enough to follow the flow without disturbing the flow
field or clogging the device. At the same time, particles must be large
3.3 Measurement methods 41

enough to be adequately imaged and to dampen the effects of Brownian


motion.
For the present velocity measurements, we used red fluorescing polymer
microspheres (Duke Scientific Corp.) with 1 J.1m in mean diameter and
3
a density of 1.05 g/cm . The aqueous suspensions are packaged as 1%
solids in a multicomponent dispersing system which prevents clumping
and aids in filter testing (approximatively 1.8 x 1010 particles in 1 ml
solution). The coefficient of diameter variation (standard deviation as a
percent of the mean) is below 5%. The seeding particles are tagged with
a fluorescent dye that absorbs at a wavelength of Aabs = 532 nm (green)
and emits at a peak of Aemit = 612 nm (red) (Stokes shift of 80 nm).
Brownian motion
Brownian motion is an important consideration when using particles
with diameters D p < 1 J.1ID to trace velocity fields with low velocities (in
the order of 10 J.1m/s. Given time intervals, ~t, much larger than the
particle relaxation time 'Tp (which is of order 10- 9 s), the mean square
distance of diffusion is

(3.6)

where the diffusion coefficient D is given as

K,. T
D=----- (3.7)
3 . 7r . J.1 . D p

Here, D p is the particle diameter, K, is the Boltzman's constant, T is the


absolute temperature of the fluid, and J.1 is the dynamic viscosity of the
fluid. In regions where velocity gradients are small, the effect of Brown-
ian Diffusion can be modelled as white-noise process that increases the
uncertainty with which the displacement of a particle can be measured.
A particle following a flow with velocity u over a time interval ~t has a
displacement

~x = u ~t. (3.8)
42 3. Experimental setup and measurement methods

Therefore, the relative error due to Brownian motion can be estimated


by

(3.9)

This error establishes a lower limit on the desired measurement time in-
terval, ~t. In the present experiments, the characteristic liquid velocities
are in the order of u rv 10 cm/s and ~t rv 500 ns. This yields a relative
error due to Brownian motion of approximatively 1%. Since this error is
a result of diffusion, it is unbiased and can be substantially reduced by
averaging over several particle images in a single interrogation window
and by ensemble averaging over several realizations. According to San-
tiago et al. (1998), the diffusive uncertainty is proportional to !;B/v'"N,
where N is the total number of independent realizations. In the present
experiments we used N rv 200 realizations to ensemble-average the vec-
tor field. So the error due to Brownian motion reduces to less than 5%0
and therefore can be neglected.

3.3.2.4 Two-phase velocity measurements using j.LPIV

The present experiments are restricted to investigating two-phase gas-


liquid flows. The recirculation motion in the liquid segments of an in-
termittent gas-liquid flow were analyzed using micron-resolution Particle
Image Velocimetry (JLPIV). The velocity distribution influences the mix-
ing and the mass transport towards the reactive phase interface dealing
with two-phase chemical reactions. Due to the fact that intermittent flow
is a non-stationary flow pattern, averaging the velocity components of
the complete area of view of the CCD camera over a large amount of
images was not possible as it is in single-phase applications. A multi-step
procedure was used to get ensemble-averaged velocity data of a liquid
segment of an intermittent gas-liquid flow.
An area of interest (master, see Fig. 3.14) was manually chosen in one
image.
3.3 Measurement methods 43

Figure 3.14: Manually chosen area of interest (master) in a


recorded PIV-irmtge of a particle seeded flow in a rectangular mi-
crochannel of 400 ILm width.

In all of the recorded images, the region of best conformance with the
master had to be found. For this, the well known cross-correlation tech-
nique is used. Cross-correlation can be understood most straightfor-
wardly as an operation in the spatial domain. If the shape of a target
is made into a kernel whose values are multiplied by the pixel values
at every location in the image and normalized for the absolute values
of the pixels, the largest response will result when the image contains
the same pattern of grey scale values. A direct cross-correlation in the
spatial domain is computationally very expensive. Therefore, the opera-
tion is more efficiently performed in the Fourier domain by multiplying
(remembering that these are complex numbers) the transforms of the
target and the image together and performing an inverse transformation
on the result. With either implementation, the locations in the image
that match the target are found efficiently even in the presence of noise,
shading, or partial obscllration of the target (Fig. 3.15).
The cross-correlation images (response) are normalized with the maxi-
mum response, which is obviously obtained from the image, where the
master is taken from. For this image, the cross-correlation factor, r, is
maximal and the normalized r[i, j, Nmaster]max equals 1. For a further
44 3. Experimental setup and measurement methods

processing of each image, we defined a minimum cross-correlation factor


of r max > 0.80 to be required. With this value a misinterpretation or a
wrongly identified liquid plug could be avoided.
For pach recorded imagp N with a r[i, j, N]rnax > 0.80, the raw vpctor
spction of the same size as the master at the position ri,
j] in thp full
Rcale velocity vpctor file (Fig. :1.15) was extracted. For all the pmspntpd
experimentR 200 vector files were averaged.

3.3.3 X-Ray Tomographic Microscopy (XTM)

X-ray Tomographic Microscopy (XTM) is a well-known tpchniqup, which


providps volumetric data of sampl<~s in a non-dpstructivp way. XTM has
bepn applied in many fipldR of resparch (matprial Rcience, microelectron-
ics, medicine, biology, archeology). Until recently, major limitations wpre
imposed by low detection efficiency and a low spatial resolution. With the
advent of third generation synchrotron facilities, excellent high intensity
X-ray sources became availablp. Synchrotron radiation, as a highly bril-
liant X-ray source, pushps high-resolution examinations to the microm-
eter and sub-micrometer rangp and its coherent nature pxtends the clas-
sical absorption tomography towards edge-enhanced and phase-sensitive
investigations. Commonly uspd ddector systems (scintillator properties,
optical light transfer, and CCD granularity) imposp a practical limit of
tIlE) spatial resolution of about onc micrometer. The demands of the pro-
gressing research for an advance into the submicron region were fulfilled
by a fully novel detf~ctor system (Stampanoni d a1., 2002a, b).
We performpd our experiments at the XTM station installed at the Mate-
rials Sciencp Bpamline 4S of the Swiss Light Source (SLS). The detection
system is based on X-ray to visible light conversion by YAG:Ce single
crystals of different thicknesses, and optical magnifications, done by mi-
croscope opticR with magnifications from 4x up to 40x. Spatial resolu-
tions down to 1.04 microns have been detected at 10% of contrast (Stam-
panoni et al., 2002b). In order to detect the thin interface between air
<:md fluid, we exploitpd tll(-) coll(-)rpnt propertips of the synchrotron beam.
In-line phase contrast or, likp in the case of the present experiment, sim-
ple edge-enhancement can be easily obtained simply by increasing the
3.3 Measurement methods 45

..
" - .' , , ". . ".'.".'.0'.'0."., ,._., _.,.~,,
--
~

~_.,.,._~.,.,', ... ~- ~ ._-_._~--.-,-_.,~.,. -_._,


_,.~

_._~-,-, .. ,'-.-
__ _ . , _ . , _ . , _ , _ _ _

'.,,-,-,-.",', .. ', .. 0' _ _ . - _ . _

"" .. ,.""" .
,~.

. ..

-
_.~,~,.'

.~.~
_~,J.J

. -." " . " ,.-


_.- - ...- .... -
,_,_,~_,,_._._._

~.-

-.
-- _. __ ....

-. - .
,,~.,,~.,.,

_~,,_.,-,._

, ,
,..... .
.- , . -
,~,_

Figure 3.15: Schematic of the process steps used for the velocity
measurements of unsteady intermittent gas-liquid flow. The liquid
plug of each image is searched by cross-correlation in the Fouri(:~r
space (F{A} . F {B} * ), delivering the position of the best confor-
mance. With this information the raw velocity vectors of the liquid
plug is extracted.
46 3. Experimental setup and measurement methods

distance between sample and detector (sample detector distance, SDD).


In fact, by carefully adjusting the SDD, onc can enhance the air-liquid
interface by recording an image~ where Fresnel diffraction becomes more
important than the absorption. Since this process is related to the second
derivative of the refractive index, the interface is supposed to give suffi-
cient contrast. By selecting an energy of 17.7 keY and a SDD of 370 mm,
edge-enhanced radiographic projection could be acquired with less than
4 lun spatial resolution. Exposure time was 3 seconds. Rotating the sam-
ple along its main axis, it was possible to acquire a full set of projections
over an angular range of [0,180]. Filtered-backprojection algorithms
have been used to tomographically reconstruct slices perpendicular to
the flow direction, revealing information about the flow pattern and the
void fraction.

Objective Scintillator
~--+----+---=--"7:\

Sample

,~)
Relay mirror ~\

Figure 3.16: Schematic diagram of the detector system showing


the geometry of irradiation.

For the X-Ray Tomographic Microscopy measurements, the microchan-


nel was mounted vertically in a channel holder similar to the one de-
scribed in chap. 3.1.2. In contrast to the mounting system used for the
LIF and PIV measurements, the XTM channel holder features only fluid
inlet connectors (see Fig. :3.17). The channel holder was fixed on a 2-
3.3 Measurement methods 47

axis micrometer positioner/translator. Mounted on the XTM station,


the microchannel had to be positioned exactly in the synchrotron ra-
diation beam. Due to the restricted X-ray beam energy and the small
difference in the X-ray absorption rate of the two fluids, the material
surrounding the microchannel had to be reduced to a minimum to get
a sufficiently detail(~d image of the multiphase flow. Therefore, the part
of the channel with the multiphase flow had to be cut to rectangular
external dimensions of 1.5 mm to 2.4 mm. The multiphase outlet of the
microchannel leads directly to tflE-) outside (atmospheric pressure) and
was absorbed below the outlet hole.

liqul~
PEEK tubing

/ Connectors =
j' [If
3

Ri nln ill
1_:_] .
1
1
I11
:

I J ~~~JJ
Microchannel Channel holder

A B

Figure 3.17: (A) Microchannel setup for the XTM experiments


including the 2-axis micrometer positioner/translator (1), the fluid
inlet tubing and connectors (2), the channel holder (3), the mi-
crochanncl (4) and the multiphase fluid outlet (5). (B) Schematic
of the channel holder.

For a full set of projections of the multiphase flow, we recorded


1000 images with a change of the channel/synchrotron beam angle of
~o: = 180 /1000 = 0.18 between each exposure. The exposure time
0 0

was 3 seconds for each projection. The backprojection algorithm com-


bines all the projections to sinograms (Fig. ::>.18C). These files contain
48 3. Experimental setup and measurement methods

the information of a certain horizontal layer of the channel (perpendi-


cular to thp flow direction) for all the recorded angular views. The final
tomogram reconstructs the slices revealing the flow pattern information.

B C

Figure 3.18: (A) Projection acquired at 0: = 00 (n = 1). (B)


Projetion at rx = 30 (n = 300). (C) Sinogram of one horizontal
slice.

XTM images deliver a time cweraged ;) dimensional flow visualization of


the gas-liquid boundary with high spatial resolution. Using a single plane
perpendicular to the direction of flow, information of the volumetric gas
content can be extracted using the spatial distribution of the involved
fluids. For stationary flow patterns: XTM is a powerful visualization
method to present the phase boundary. Especially for annular flow (see
chap. 2.1) the reconstructed gas-liquid interface is clearly noticeable.
Intermittent or bubbly flow visualizations reveal blurry phase boundaries
due to the temporal smearing.
3.4 Measurement accuracy 49

3.4 Measurement accuracy

3.4.1 Flow feed

The accuracy of the syringe pump (accuracy within 0.35%) and the gas
mass flow controller (accuracy 0.5% of reading plus 0.1 % full-scale),
respectively, lead to a mean error in the determination of the superficial
velocities.
All the presented experiments were conducted at room temperature and
atmospheric pressure at the discharge of the test section. To ensure a
steady state flow condition, the measurement location was chosen in
the last quarter of the microchannel. Therefore, the actual pressure at
the test section is close to the atmospheric pressure, and the volumetric
gas flow rate is assumed to be equal to the standard volume flow rate
(Nml/min). The influence of temperature fluctuations on the fluid prop-
erties are neglected, as a maximum change in temperature of 5C was
observed, leading to property differences below 2%.
For the presented reasons, the error due to the flow feed system and the
ambiance, exp, can assumed to be in the range of exp :::::: 5%.

3.4.2 Pressure drop

The accuracy of the pressure drop measurements are directly related to


the accuracy of the pressure sensors. As seen in chap. 3.1.3, the Endress
and Hauser pressure sensors feature a measurement accuracy 0.5% of
the full-scale, which in our case is 0.03 bar. Due to the unsteady flow
patterns, the recorded pressure drops are averaged over a certain time,
further minimizing the measurement uncertainty. Hence, the error of the
pressure drop measurements is in the order of pressure :::::: 1 %.

3.4.3 Void fraction measurement accuracy

The accuracy of the void fraction measurements mainly depends on the


fluctuations in the illuminating laser intensity, the mixture quality of the
50 3. Experimental setup and measurement methods

fluorescent additive in the working fluid, and the reduction of the illumi-
nating laser intensity along the illumination path inside the microchan-
nel. The intensity fluctuations in the recorded images using liquids en-
riched with Rhodamine B only (no gas flow present), was determined
to be 5%. This uncertainty originated mainly from the fluctuations of
the laser intensity, as we assumed the fluorescent dye to be completely
dissolved in the liquid.
Due to the small depths of the used microchannels (d = 100 ... 200 jJ,m),
and the high laser intensities, the assumption of uniform illuminating
intensity was made. Theoretically, the recorded light intensity value is
the integral of the emitted intensities over the full depth of the chan-
nel. A reduction of the emitted intensity along the illuminating path
is expected to be present at high liquid content locations only. The
influence of off-focus recording is neglected. The accuracy of the void
fraction measurements is therefore determined to be in the range of
~voidfraction :::::::: 5 - 10%.

3.4.4 Accuracy of the velocity measurements

For micron-resolution Particle Image Velocimetry (jJ,PIV), the spatial


resolution is limited by the effective diameter of particle images when
projected back into the flow field. For magnifications much larger than
unity, the diameter of the diffraction-limited point spread function, in
the image plane, is given by

A
D s = 2.44 . M 2 . NA' (3.10)

where M is the total magnification of the microscope and NA is the


numerical aperture of the lens (Born and Wolf, 1997). Using the Zeiss
A-Plan 10x objective with a numerical aperture of NA = 1.25, and as-
suming the wavelength of the recording light is A = 612 nm, the diameter
of the point spread function in our experiment is D s :::::::: 6 jJ,m.
The actual image recorded on the CCD camera is the convolution of
the diffraction-limited image with the geometric image (Adrian et al.,
3.4 Measurement accuracy 51

2000). Approximating both the geometric and diffraction-limited images


as Gaussian functions, the resulting convolution is a Gaussian function
with an effective particle diameter De' where

(3.11)

For a magnification of M = 10, a particle diameter of D p = 1 Jjm, and a


numerical aperture of NA = 1.25, the effective particle image diameter
projected onto the CCD camera is De = 11.6 Jjm. According to Prasad
et al. (1992), the location of a particle-image correlation peak can be
determined to within 1/10th the particle-image diameter, if a particle
image diameter is resolved by 3 - 4 pixels. This yields a measurement
uncertainty of 8x ~ D e /10M. For the parameters considered here, the
uncertainty reduces to 8x ~ 11.6/(10 . 10) = 116 nm, which is approxi-
mately 0.5 pixel of the camera resolution.
The fact that one can measure particle displacement in the range of
0.5 pixels of the CCD camera is surprising. In most microscopic appli-
cations, one is primarily interested in determining the shape of small
objects. Obviously, the smallest resolvable shape is on the order of the
resolution of the microscope. In JjPIV, one knows a priori the particle
shape and is interested only in determining particle position. By over-
sampling the image (i.e. resolving the image with 3 - 4 pixels across the
image diameter), one can determine particle position to within an order
of magnitude better resolution than the diffraction-limited resolution of
the microscope.
In common PIV applications, the thickness of the laser light sheet de-
termines the out-of-plane measurement domain. In JjPIV however, one
is often interested in obtaining velocity measurements with out-of-plane
resolutions on the order of 1-10 Jjm. In contrast to common PIV, JjPIV
uses broad field illumination. The total depth of field of a microscope
system is estimated by Inoue and Spring (1997) as the sum of the depth
of field due to diffraction and geometric affects

n A n e
8z= NA2 + M.NA (3.12)
52 3. Experimental setup and measurement methods

where n is the index of refraction of the immersion medium between the


microftuidic device and the objective lens, and e is the smallest resolvable
distance on the image detector. For the present experiments, Eq. 3.12
yields 8z = 1.1 /l-m.
Due to the strictly laminar flow in the microchannels, the error due to
the interrogation technique is expected to be negligible. The uncertainty
of the velocity measurements is therefore determined to velocity ~ 1%.
Obviously the j.LPIV measurements are more accurate than the flow feed
system itself (exp = 5%). Therefore the error in the experimental setup
is dominant and defines the velocity measurement error.

3.4.5 Uncertainty in the measurements using X-Ray


Tomographic Microscopy (XTM)

The spatial resolution of an image recorded using X-Ray Tomographic


Microscopy (XTM) is determined by the amount of defect of focus of
the image distribution before and behind the focal plane and, further,
by diffraction and spherical aberration arising from the thickness of the
scintillator and the substrate. According to Stampanoni et al. (2002b),
the spatial resolution considering all influences can be approximated with
the function

R= JC:A)2 +(q ozoNA)2 (3.13)

with p = 0.34 /l-m, and q = 0.036. NA is the numerical aperture of


the objective, and z the thickness of the scintillator (1.8 ... 50 /l-m). The
spatial resolution of the XTM measurements for the present experiments
(energy 17.7 keY, sample to detector distance SSD = 370 mm) and the
used optical reconstruction process was slightly less than 4 /l-m. The mean
error in spatial resolution of the XTM system, XTM, was determined to
XTM < 1%.

For a detailed description of the accuracy of the the angular position,


the energy fluctuations, etc. we refer to Stampanoni et al. (2002b, a).
53

Chapter 4

Flow regime

4.1 Flow regimes in microchannels


Flow regime maps for two-phase flow in channels or pipes are used to
predict the flow pattern. In most of the existing maps (Zhao and Bi, 2001;
'friplett et al., 1999b; Kawahara et al., 2002; Coleman and Garimella,
1999) the flow pattern is plotted as a function of the superficial gas and
liquid velocities, jG and jL (Eq. 2.11). Some flow regime maps use the
gas and liquid flow rate as well as fluid properties as coordinates (Baker,
1954). Therefore, most of the common flow pattern maps are valid for a
specific combination of liquids, channel geometry (cross-sectional shape,
hydraulic diameter), and channel material only.
In the present study, flow pattern maps for two-phase ga..~-liquid flows in
rectangular microchannels for several hydraulic diameters D h and three
different liquids were constructed. At room temperature, deionized water
features a viscosity similar to ethanol, 1Jwater/1Jeth = 0.84. The interfacial
tension of water however is larger than the one of ethanol by a factor of
O'water/O'eth = 3.28. The ratio of the surface tension between the aqueous
20% glycerol mixture and water is close to one, O'water/O'glyc = 1.04, the
viscosity ratio on the other hand is much higher, 1Jglyc/1Jwater = 1.70 (see
Tab. 4.1).
54 4. Flow regime

Fluid combination Surface tension ratio Viscosity ratio


Water - ethanol 3.28 0.84
Water - glycerol (20%) 1.04 1.70

Table 4.1: Comparison of the properties at room temperatures of


the fluids employed

Laser-Induced Fluorescence (LIF) was used for the visualization of the


two-phase flow and the pattern recognition. Rhodamine B was added to
the liquid phase. The concentration of the dye was chosen to not change
the fluid properties more than by 1%. For each configuration (hydraulic
diameter D h , gas and liquid flow rates VL,G, and fluid combination) a
sample of 100 images was taken. Especially in the regime transition re-
gions (eg. bubbly/intermittent, intermittent/annular), a unique flow pat-
tern could not be detected. The configuration was categorized according
to the higher number of apparent flow regime. This circumstance clar-
ifies that the transition boundaries in the presented flow pattern maps
are transition regions more than sharp transition lines.
Due to the dominance of surface forces (surface tension) compared to
body forces (gravity), we did not detect any volumetric flow rate com-
bination leading to a stratified flow pattern. The dominant flow regime
for moderatel::luperficial gas velocitiel::l il::l intermittent flow (Fig. 4.1B).
In this type of flow pattern (also called plug flow or Taylor flow) the
air moves as elongated bubbles with diameter slightly less than that of
the tube and the bubbles are separated by liquid segments (plugs). At
even lower gas velocities, the gas bubbles tend to have circular shape and
their diameter decreases. We defined the flow pattern to be bubbly flow
when the maximum bubble diameter is smaller than 80% of the channel
width (Fig. 4.1A). For increased superficial gas velocities, annular flow
patterns can be observed (Fig. 4.1C).
In all the following flow pattern maps, the present experimental data is
plotted as follows: represents intermittent flow, 0 annular flow, and 6
the bubbly flow regime.
Figures 4.2A - 4.2E show flow pattern maps for the different configura-
4.1 Flow regimes in microchannels 55

C
Figure 4.1: Images of two-phase flow in a microchannel at low
gas and high liquid flow rates (bubbly flow, A), at flow rates yield-
ing intermittent flow (B), and at high gas superficial velocities and
moderate liquid flow rates, leading to an annular flow regime (C).
Images recorded using Laser Induced Fluorescence (LIF, Rhodamine
B added to the liquid phase).

tions mentioned above.


Especially for sm<:},ll channel geometries (Fig. 4.2A -C), intermittent flow
is detected for a wide range of fluid flow rates. The low surface tension of
ethanol against nitrogen shifts the transition regime intermittent - annu-
lar flow towards lower superficial gas velocities, wheI'l,~as a high viscosity
value, as found for glycerol, favors the intermittent flow regime. For a
microchannel with a hydraulic diameter D h = 187.5 Mm, bubbly flow
was detected only at the highest possible liquid flow rates.
For larger microchannels (D h = 218 Mm), the range of volumetric flow
rates yielding annular flow increases. The transition boundary intermit-
tent flow - annular flow is shifted towards lower superficial gas velocities.
No remarkable influencp on tlw supprficialliquid vPlocity can bp found.
Bubbly flow is detected at lower liquid flow rates than in smaller chan-
nels.
56 4. Flow regime


I:J.
!i.
/:;.
!:J.
A
li
li


/1



o
0

0

o 0 0
~
S. o oo 00 00 o o 0 0
4 0.1
o o 0 0

0.1
0 0
o o 0 0

o o 0 0
o
o 0
o o 0 0
o o
0
0 0
o o 0 0 0



o 0
0
0 o o 0 0
o 0
o 0
o
o o 0 0
o 0 0

o 0 0 0
0.01-1--.............---r.~........
, ....-.."T""~-..,..,- ......---r.........-.!'""'!",........+ 0.01+-.,..--,-...,....+--...,..,..---..,.....-,.---....,."". + "!,""!.....
1 10 10
Jolm/s) A D
o 0 o 0 0
o 0 0 /1 /1 /1 o 0 0
o 0 0

o o 0 0 0
0 0
o 0 0 0

o 0 0 o 0 0 0

o 0 0
[] D 0 D
0.1
o o 0 0.1
o o 0 0
o o 0
o 0 ODD 0
o
o
0
0
0
0
o o 0 r:J Cl D D

o 0 D 0 D 0
o 0 0 o o 0 o 0 0 0
o o 0 0 0
0.01+_................................,., . -......_....--....--....... .,.,+ 0.01-1-_....................-.h-,......... .---..,..,-....................-.r-T,-.."".+
4 I!i IIi ,. .0' " 15 7 ,
1 10 1 10
j o lml8]
B JOlm/s] E
lJ 0 0 0
6. I:J. o 0



o
o
0
0
o
lJ
0
lJ lJ
0 0
lJ
o
o
0 lJ
0 0
0
0

o
0 0 0



0.1
0
lJ o lJ 0

o o 0 0





o 0
0
0 0o





o
o o 0 0
0
0
o 0 0

O.01-1--.....................-.!'""'!",....' ...--..,..-..,.--............... -.!"'"'!,'.'10.+


0
lJ 0 0 0
O.Ol-+--.,........................,........T'T'"-'--...,..-.................................-.+
" 5 1 4 $ ., ,. III
1 1 10
jG[m/s)
C jG[mlS)
F
Figure 4.2: Flow pattern maps in a rectangular microchannel with
a hydraulic diameter D h = 187.5 J.Lm (left) using water - nitrogen
(A), ethanol- nitrogen (B) and glycerol- nitrogen (C), respectively.
The same fluid configuration (D: water - nitrogen, E: ethanol - ni-
trogen, F: glycerol) were analyzed in a channel with Dh = 218 J.Lm
(right).
4.1 Flow regimes in microchannels 57

4.1.1 Universal flow pattern map

Apparently, the presented regime maps show different transition bound-


aries for any of the analyzed configurations. When designing a microre-
actor for a chemical gas-liquid reaction, where a specific interfacial area
is required, detailed knowledge of the corresponding flow pattern and the
flow rates is inevitable. By just changing the liquid pha.',e from ethanol
to acetone for example, the flow pattern maps presented above and the
existing regime maps in the literature cannot be used anymore. There-
fore, a more universally applicable flow regime map is desired. Due to
the large number of influencing factors, a completely universal pattern
map is unrealistic. Besides the fluid properties and the hydraulic diame-
ter of the channel varied in the present study, also the channel material,
its surface roughness, the cross-sectional shape, and other variables as
temperature or the applied pressure are influencing the appearing flow
pattern.
Taking into account the fluids' kinematic viscosities, VL,a, their surface
tension, (j, the difference of the fluid densities, Dog, the superficial veloc-
ities, jL,a, the hydraulic diameter, Dh, and the channel roughness, ks,
a universal flow regime map is presented for silicon microchannels with
rectangular cross-section. All the experiments were conducted at room
temperature, so the influence of changes in fluid properties by a change in
temperature was not examined, as well as changes in the cross-sectional
shape of the microchannel. For the reduction of the above mentioned
influencing factors, the Buckingham pi-theorem (Zlokarnik, 1983) was
used.
Table 4.2 shows the dimension matrix constructed for the influencing
parameters. This table is structured in a core matrix and a remnant
matrix.
The core matrix is transferred into a unit matrix by linear transforma-
tions (Tab. 4.3).
This matrix analysis leads to the remaining 5 characteristic numbers, IT l
to IT5 .
58 4. Flow regime

core matrix remnant matrix


Ilg Dh (j JL JG 1/L 1/G ks
Mass M [kg] 1 0 1 0 0 0 0 0
Length L [m] -3 1 0 1 1 2 2 1
Time T [s] 0 0 -2 -1 -1 -1 -1 0

Table 4.2: Dimension matrix of the influencing factors according


to Zlokarnik (1983)

unit matrix remnant matrix


Ilg Dh (j
JL JG VL VG ks
M+.!.T 1 0 0 _1 -.! _1. _1. 0
2
(L+3M) + ~T 0 1 0 _12 _1 1
2
1
2
1
-~T 0 0 1 1
2
2
1
2
~
2
t2 0

Table 4.3: Dimension matrix after transfer into a unit matrix by


linear transformations

1 1
JL jL .llg 2 Dh 2
II 1 = A
1
_1 D -- 1 1
I..J.g 2. h 2. (j2 (j2

II2 = JG
A
I..J.g _12 . D h --1... 1
(72

1
VL VL .llg 2
II3 = ---::-----:--- (4.1)
A
I..J.g _.1
2.
D.1 1
h 2 (72
1
Dh 2'(72
1

VG
II4 = ---::-----:---:-
A_1 D 1 1
I..J.{) 2. h 2 '(72
4.1 Flow regimes in microchannels 59

A transformation of these 5 charapteristic numbers results i~ a set of


necessary dimensionless numbers 1Rea, Re L, W ea, We L, ~ to fully J
characterize the transition lines.

2
2 jL . D.g. Dh
IT 1 = (J
= WeL
2
2 jG . D.g . D h
IT 2 = (J
= WeG

-1 jL' Dh
IT 1 . IT3 = = Re L (4.2)
VL

-1 jG' Dh
IT 2 . II4 = = ReG
VG

ks
ITs = -
Dh
where the Reynolds number

jL,G' Dh
R eLG = (4.3)
, VL,a

captures the impact of the hydraulic diameter, the superficial velocity,


and the fluid viscosity, whereas the Weber number

(4.4)

is used to include the influence of the fluid densities and the surface
tension.
The Buckingham pi-theorem states, that there is a function f valid for
the regime transition boundaries (tb), where

(4.5)
60 4. Flow regime

As transition boundaries (tb) the conditions of flow pattern changes from


annular flow to intermittent flow and from intermittent flow to bubbly
flow (transition lines in Figs. 4.2A-F) are considered.
Eq. 4.5 is the maximum information the dimension analysis is able to
provide. The exact form of the function f can only be extracted from
experimental results. In most cases, f is an exponential function. For
the universal flow pattern transition boundaries in the present study, we
used

(4.6)

A parameter variation for the parameter a, b, c, d, e and f, using a least


square fit of data points of both transition lines (annular - intermittent
and intermittent - bubbly) yielded the exponents found in Tab. 4.4. The
improved quality of two separate sets of exponents for the two transition
boundaries, thinking of the large number of additional (not considered)
influences, is negligible. Hence for practical reasons we concentrated on
a single set of exponents for both transition lines, as they can be plotted
in the same graph.

Table 4.4: Exponents for universal flow pattern transition bound-


aries according to Eq. 4.6

According to the results of the parameter variation for the transition


boundaries,

(4.7)
4.1 Flow regimes in microchannels 61

of all the experimental data points (x-axis) is plotted versus

107 . Re~2. We~4. (;;:) 5 (4.8)

0 0

0 0

"'0--.. "'-
IS
0 0 0
0 0 0 0


.ot...-'"
0 0 0 0
0
d~ 0 0 0

;:: 0.\ 0
0.1 0 0 0
0
:l . 0 0 0
ofi' 0 0 0 ~ 0 0 0 0

~rr.
0 0 0
~~ 0 0 0

~ : 0
0
0
0
0
0
0
0
0

0
0
0

0.01

,..
0 0

.. e l'
0

t
10
0 0 0
0.01
.,..
Reo0.2 Wft
o
IU
A Reo" Weo B
0 0

"'- 0 0 0 ~t? 0

IS 0 0 0
0 0

..I t


0
0
0
0
0
0 ....-'" 0
0
0
0
0
0
"( ( 01
D.I
" " " " " "
"
"ot



""
"0
0
0 o~
. " 0 0
rr.
~~
: " "" "" ~rr.
~
0
0
0
0
0
0
0" " " 0 0 0"
"" "" ""
0,01
" " " 0,01
0
" "
~ 0' IjI

weGO'~ Reo0.2 We D.~


R8GO,2 10
C G D

" "
"'2' "" "" "'0--.. " 0 0 0


d~



"
"
0
" "
" " .-
,,~
,2

""
0

""
0

"" "
0

" " " "


.ot
( 0.1
"" "" "
0 0

0
0

0
0

0
:lot
0.1
" " " "
"
""
0

~~ " " "


~~
0 0 0 0
0 0 0 0

0" : " "


0 0
0
0
0
0
0
0

0
0 "
0 0
" 0
0
0 0 0 0
"
...
0
0.01 " " " " 0.01

.. Re:2 Weo"
..
0

10
0 0 0

ReoO,2 We:
E F
Figure 4.3: Flow pattern maps using the universal pattern tran-
sition boundaries for a rectangular microchannel for the same flow
and geometry conditions as in Fig. 4.2 (A: water, Dh 187.5 Mm;
B: ethanol, 187.5 Mm; C: glycerol, 187.5 Mm; D: water, 218 Mm; E:
ethanol, 218 Mm; F: glycerol, 218 Mm).
62 4. Flow regime

For all the presented experimental data, the universal flow pattern map
shows a satisfying agreement. The two boundary lines are located in the
transition regions for all the experimental data. Therefore the universal
flow pattern map is capable of making a prediction of the expected flow
pattern for a large variety of fluid and geometrical conditions. The fact
that additional important influencing factors as temperature, the cross-
sectional shape or the channel material and surface roughness, were not
varied or neglected in the presented flow regime map, shows the indis-
pensable need for the experimental verification of the flow pattern for
the detailed design of a chemical microreactor.

4.2 Comparison with correlations


Our experimentally derived pattern map for the microchannel with a
hydraulic diameter of D h = 210 j.Lm using water and nitrogen as working
fluids was compared to several existing flow regime maps. Flow maps to
which our results are compared are selected carefully from the large range
of existing maps. Among the selection criteria were similar dimensions,
similar substance properties, recent publications, range of Bond number.
Again, in all the figures, the present data is plotted as follows:

represents intermittent flow,


o annular flow and
6 bubbly flow regime.

4.2.1 Flow regime map by Baker


Due to the fact that the flow pattern map by Baker (1954) (Fig. 4.4) is
based on flow through large diameter pipes (D h 210 j.Lm), an accurate
consistency of the flow regime cannot be expected. The experimental
data is shifted towards lower gas flow rates. For the Baker flow regime
map, no weighting factors are present to adapt the transition boundaries
to the phenomena in microchannels. The Baker diagram therefore is not
capable to predict accurately flow patterns for microfluidic applications.
4.2 Comparison with correlations 63

1008
6 Misty
4

N -:;;:-
0
l'e
4

-
6-
'':Ji.l:}
2

1
8
6
4

0.1
2 468 24682468
10 100 1000
(I. M'L'I' I M'G)

Figure 4.4: Two-phase flow regime maps of Baker (circular pipes)


using water as working fluid compared with the present experimen-
tal data of a rectangular microchannel with a hydraulic diameter
Dh = 210 J.Lm. Experimental data: e: intermittent, 0: annular, 6:
bubbly.

4.2.2 Flow pattern map by Zhao

The flow regime map of Zhao and Bi (2001) (Fig. 4.5) is experimentally
derived using an air-water mixture in triangular microchannels with a
hydraulic diameter of Dh = 866 J.1>m. This hydraulic diameter is still
larger than the diameters in the present study by a factor of approxi-
mately 4. The agreement between the present data and the regime map
for the microchannel by Zhao is accurate for the flow pattern change be-
tween intermittent (plug) and annular flow. Due to the larger hydraulic
diameters and the different channel geometry, bubbly flow is predicted
only for very low superficial gas velocities.
64 4. Flow regime

'" '" '" '" '" '" '"


6
'" '" '" '"
(;
4
'" '" '" '" '"
3

2
'" '" '" '"
Intennittent

~
.. 0

.~
0.1
0 0 0

6 0 0 0 0
5
4 0 0 0 0

3 0 0 0 0 0

2 0 0 0 0 0

0 0 0 0 0 0 0

0.01
2 3 4 5 6 7 69 2 3 4 (; 6 7 89
1 10
iG [m/sI

Figure 4.5: Two-phase flow regime maps of Zhao (triangular mi-


crochanneles, Dh = 0.866 mm) compared with the present experi-
mental data of a rectangular microchannel with a hydraulic diameter
D h = 210 /-Lm. Experimental data: e: intermittent, 0: annular, 6.:
bubbly.

4.2.3 Flow regime map by Triplett

The results of Triplett et al. (1999b) (Fig. 4.6) are based on experi-
ments using air and water as fluids in semi-triangular microchannels of
hydraulic diameters D h = 1.09 mm. Even tough the hydraulic diame-
ter and the cross-sectional shape of the channels are similar to the ones
in the experiments by Zhao, the flow regime map by Triplett predicts
the flow patterns of the two-phase flow in microchannels of rectangular
cross-section and hydraulic diameters smaller by a factor of 5 with high
accuracy. Again, no bubbly flow was predicted.
4.2 Comparison with correlations 65

2
bubbl~ chum
~ t::. t::. t::. t::. t::.

6
5 t::. t::.
t::.
t::.
t::.
t::.
t::.
t::.
t::.
t::.

0 0
0
0
4
3 t::. t::. t::.
slug

t::. t::. 0 0 0

2
~
oS 0 0
slug annular
0

.~
0.1
0 0 0

6 0 0 0
5
4 0 0 0

3 0 0 0 0

2 0 0 0
annular.
0 0 0 0 0 0 0

0.01
2 3 4 5 6 7 69 2 3 4 5 6 7 69
1 10
iGlm/sj

Figure 4.6: Two-phase flow regime maps of Triplett (semi-


triangular channels, D h = 1.09 mm) compared with the present
experimental data of a rectangular microchannel with a hydraulic
diameter Dh --:- 210 J.Lm. Experimental data: e: intermittent, 0:
annular, 6: bubbly.

4.2.4 Flow pattern map by Kawahara

Kawahara et al. (2002) used de-ionized water and nitrogen as working


fluids and circular microchannels with an inner diameter D = 100 J.Lm.
Even tough the hydraulic diameter of the present study is closest to
the one shown by Kawahara, the data do not correlate. The transition
between the slug and the annular flow regime (slug-ring / semi annular)
seems to be shifted towards higher superficial gas velocities (Fig. 4.7).
The same effect was already found in the present experiments when
changing to channels with smaller hydraulic diameters (Fig. 4.2).
66 4. Flow regime

t:.
6
5 /';. t:.
t:.
t:.

0
4
3 t:. /';. t:. /';.

slug - ring
/';.

2
~ 0 0 0
..,=t 0.1
0 o 0 0


6 0 0 0 0
5 0 0 0 0
4
3 0 0 0 0 0

2 0 0 0 0 0
semi annular
0 o 0 0 o 0 0

0.01
2 3 4 5 6 7 69 2 3 4 5 6 7 89
1 10
ja [m/sI

Figure 4.7: Two-phase flow regime maps of Kawahara (circular


microchannels, Dh = 100 /-Lm) compared with the present experi-
mental data of a rectangular microchannel with a hydraulic diameter
D h = 210 /-Lm. Experimental data: e: intermittent, 0: annular, 6:
bubbly.

4.2.5 Flow regime map by Coleman

Coleman and Garimella (1999) carried out some of the experiments in


round tubes with an inner diameter of D = 1.30 mm, others in mi-
crochannels of rectangular cross-section with a hydraulic diameter of
D h = 5.36 mm. The used fluids are air as gas-phase and water as liquid-
phase. Figure 4.8A shows the comparison of the present data with the
flow regime map for the circular channel and Fig. 4.8B for the rectangu-
lar microchannel by Coleman. The consistency with the circular channel
is insufficient even though the hydraulic diameter is closer to the one
in the present experiment. All the data points are in the intermittent
(plug) flow regime even though annular flow was detected in the present
experiments.
The agreement with the regime map obtained in the rectangular mi-
4.2 Comparison with correlations 67

,I ,I
10
B
B
4
Bubbly
- - ~isPersed
2
Intennittent
-
B
!:>. !:>. !:>.!:>. !:>. !:>.!:>. -
B
4
!:>. IJ.IJ.IJ.
IJ. IJ.IJ. IJ.IJ.IJ. Wavy
00
Annular
.~ 2
!:>. !:>.!:>. !:>..!:>.
00 annular

o 00 b
0.1 - -
B 0 000
B
4



0
0
000
000

2


0
0
0
0
000
000

0.01
000
.,
0 000

'I
2 4 6 B 2 4 6 B 2 4 6 B
0.1 1 10 100
jG [mls]
A
2

1 !:>. /';. /';. /';.


B Intermittent !:>.
B /';. /';. /';. /';.

4

2
/';.

/';.


/';.

/';.
/';.

/';.

/';.

/';.


/';. /';.

/';. o
Wavy annular
0 0

i 0 o 0 0

.~ 0.1
B o 0 0

6 o
o
0
0
0
0
4 Stratified

0 Wavy
0
o
o
0
0
0
0
2 0
0 o 0 0 o 0 0

0.01
2 3 4 5 6 7 B9 2 3 4 5 6 789
1 10
jG [m/s]
B
Figure 4.8: Two-phase flow regime maps of Coleman (A: circular
channels, Dh = 1.30 mm; B: rectangular channels, Dh = 5.36 mm)
using water as working fluid compared with the present experimen-
tal data of a rectangular microchannel with a hydraulic diameter
D h = 210 J.Lm. Experimental data: .: intermittent, 0: annular, 6.:
bubbly.
68 4. Flow regime

crochannel is accurate for the transition from intermittent to annular


flow. Again, no bubbly flow i~ predicted. The impact on the flow regime
by the cross-sectional shape of the microchannel seems to be more im-
portant than by the hydraulic diameter. The ratio between the hydraulic
diameter by Coleman and the present study is

Dh,Coleman = 5.36 = 25.5 (4.9)


Dh,present 0.210

4.3 3-dimensional flow pattern visualiza-


tion using XTM
Due to the restrictions in the X-ray beam energy (even after the reduc-
tion of the silicon and glass material around the channel) each single shot
of the XTM measurements (1000 shots per experiment series) had to be
exposed 3 t:;econds to produce t:;ufficient information on the interfacial
boundary of the two fluid phases. Including the exposures for calibra-
tion (reduction of backlight) and the change in the beam-channel angle,
a single experiment (each different volumetric flow configuration) lasted
up to 1.5 hours. This makes it obvious that this 3-dimensional flow vi-
sualization technique is limited to spatiotemporally stationary flow pat-
terns. The most suitable flow pattern with a phase distribution, which is
stationary for the required measurement durat~on, is ~nular flow. For
different gas- and liquid volumetric flow rates, VG and VL, we created a
flow regime map (Fig. 4.9) for the used channel geometry (250 J.Lm wide
and 100 J.Lm deep).
According to the information provided by the flow regime map, we con-
centrated on 3 different volumetric flow configurations (VG and VL )
where annular flow was expected (experiment 1 to 3). Additionally, an
experiment (experiment 4) wa..~ carried out using volumetric flow rates
in the regime where plug and annular flows were detected in the optical
reference measurements.
The volumetric flow rates VG and VL , their corresponding superficial
velocities jG and jL, and the Reynolds numbers ReG and ReL can be
4.3 3-dimensional flow pattern visualization using XTM 69

...,,
0.5
,;1
.
I
.~ 0.4
II
Slug
<DD
... 1:i
,,
,'" ""
~
'8 0
. . ... ,,,
,

.. ..
0.3
0 <D
,, " "
CD II III a ,,
.
<><>
>
"C 0 III a ,, Annular
'S "
g 0.2 ,O'll III a,a e 3 " <><> "
~ D DI 0
m
'0
lug-AI1J1ular,'
If all .De~.'" "'92" "
(I) 0.1 D Slug flow
C- " " 0 Annular flow
..
:::J
.' " 1
C/) "
0.0 +-~"'--T--Y---OI,...--,----r---'-I----r----r----r--r--Ir---'1""--'---'
Present study

0.0 2.0 4.0 6.0 6.0


Superficial gas velocity jG [m/sI

Figure 4.9: Flow regime map for the rectangular microchannel used
in this experiments (250 x 100 j1,m 2 , D h = 143 j1,m).

found in Tab. 4.5. The Rcynolds number of each phase is defined as in


Eq.2.6.

Exp. 1 Exp. 2 Exp. 3 Exp. 4


VL [ml/rnin] 0.05 0.20 0.30 0.20
JL [m/s] 0.03 0.13 0.20 0.13
ReL [-] 4.7 19.0 28.5 19.0
Vc [ml/min] 3.00 5.00 5.00 3.00
Je [m/s] 2.00 3.30 3.30 2.00
Ree [-] 21.4 35.6 35.6 21.4
expected flow pattern annular annular annular slug - annular

Table 4.5: Configurations for the flow visualization using XTM

In the first experiment, the Reynolds numbers are Ree = 21.4 and
ReL = 4.7, respectively. The XTM image in the direction of the flow
(in the direction of the channel axis, Fig. 4.10A) clearly shows the ex-
pected annular flow pattern inside the microchannel. For a better vis-
70 4. Flow regime

ibility the image was edited by a color-lockup-table. The bright colors


represent the interfacial boundary between the gas and the liquid phase,
as well as between the channel walls and one of the fluid phases. Due
to the relatively large width-to-depth ratio of the microchannel (width
w = 250 /-Lm, depth d = 100 p,m, w/d = 2.5) most of the liquid phase
volume is located in the corners of the microchannel. The liquid film
on the larger side of the channel is thin. A differentiation between the
liquid-gas phase boundary and the channel wall-liquid pha..,e boundary
is barely possible.

(A) (B)

Figure 4.10: XTM picture showing the phase distribution inside


the microchannel for VL = 0.05 ml/min and Vc = 3.0 ml/min (Exp.
1) in the direction of the flow (A) and the :1 dimensional cross section
(B).

The darker rays coming from the edges of the microchannel are artifacts,
resulting from the deflection of the X-ray beam at the sharp corners of
the silicon or glass.
As can be seen, after the reconstruction of the 1000 projections into the
3-dimensional image, the microchannel appears to be not exactly rect-
angular but slightly distorted. Due to the adaptation of the fluid feeding
system to the experimental S(~tup of the XTM measurement station,
longer and less flexible PEEK tubing had to be used (Fig. :3.17). This
4.3 3-dimensional flow pattern visualization using XTM 71

indesirable higher inelasticity caused a disturbance of the exact angle


adjustment while turning the probe 180 0 during the measurement time.
The small difference between the effective angle of the probe and the
nominal angle used for the reconstruction by the software produces this
slightly distorted image.
Figure 4.10B highlights the stationary interfacial boundary at the narrow
side of the channel, which is therefore pronounced.

(A) (B)

Figure 4.11: XTM picture showing the phase distribution inside


the microchannel for VL = 0.2 ml/min and VG = 5.0 ml/min (Exp.
2) in the direction of the flow (A) and the ;) dimensional crOHH Hection
(B).

The configuration of experiment 2 differs from the configuration of ex-


periment 1 by a much higher liquid volumetric flow rate and just a
little higher gas flow rate. The corresponding Reynolds numbers aTe
ReG = 35.6 and ReL = 19,0, respectively. As seen in the flow regime
map (Fig. 4.9) this configuration also results in annular flow. In the im-
age axial to the channel (flow), a slight increase of the area covered by
the liquid phase can be seen (Fig. 4.11A). Again the water is concen-
trated on the narrow side of the microchannel. The interfacial boundary
in this region of the flow pattern is less distinctive than in the previous
experiment. This is the result of a less stationary flow and an interfa-
72 4. Flow regime

cial boundary fluctuating slightly over the total measurement time of


1.5 hours. ThiH aHsumption can be confirmed by the fact that the volu-
metric gas contents (void fraction, cc) of the configurations are nearly
the same. In experiment 1, it is cc = 0.67, whereas in experiment 2, the
hold up is cC = 0.66. The volumetric gas content was derived using opti-
cal LIF experiments (see chap. 6.1). The similar value of the volumetric
gas content results in a higher liquid velocity and therefore higher shear
rates at the interfacial boundary for the configuration of experiment 2.
The slightly blurred interfacial boundary is constant over the channel
length (direction of the flow) as seen in the a-dimensional image in Fig.
4.11B.
With a further increase of the liquid volumetric flow rate, the observed
effect is intensified. The interfacial boundary becomes even lesH pro-
nounced. For this configuration the Reynolds numbers are Rec. 35.6
and ReI., = 28.5. Again, at higher liquid volumetric flow rates VL , the
phase interface of the annular flow seems not to be Htationary, but mov-
ing inside the microchannel (Fig. 4.12A).

(A) (B)

Figure 4.12: XTM picture showing the phase distribution inside


the microchannel for VI., = 0.3 ml/min and Vc = 5.0 ml/min (Exp.
:~) in the direction of the flow (A) and the a dimensional cross section
(B).
4.3 3-dimensional flow pattern visualization using XTM 73

In the cross sectional image of this configuration (Fig. 4.12B), the inter-
facial boundary is no longer marked in clearly bright colors and therefore
hard to detect.
Despite the higher liquid volumetric flow rate, the volumetric gas content
inside the channel is increased only to EG = 0.72. Again, this has to
be compensated by an even higher liquid velocity. Therefore the phase
interface is unstable. No stationary flow pattern is observed.

Channel Lenqth ["rn1 C)


:;:"00 400 0(10 SOl) tooo 1200 1400 11300 c:;:}

2'50 ~
200 ~
1'30 :::;
100 :;;;
Q~
50 3-
=T==--O ~'
~ ~ ~ ~ 1= 1~ ~
Channel Length [piXl!I]

Figure 4.13: Instantaneom; CCD-image of the flow (Exp. :)), show-


ing the unsteady gas-liquid distribution of the a,nnular flow.

When looking at the optical reference image~ the assumption of the fluc-
tuating gas-liquid interface is confirmed. Fig. 4.1:) shows an instanta-
neous acquisition of the corresponding flow. If a large number of such
single images are averaged, stable annular flow seems to be present as in
the other experiments presented so far and as the XTM measurements
of this configuration suggest.
Starting from the configuration of experiment 2, we decrease the gas
volumetric flow rate from VG = 5.0 ml/min to Vc = 3.0 ml/min, while
the liquid flow rate remains at VL = 0.2 ml/min (ReG = 21.4 and
ReL = 19.0). 'l'his configuration, according to the previously presented
flow regime map, is in the range between slug flow and annular flow. In
the temporally resolved optical measurements, both of the flow patterns
could be detected. The XTM measurements (Fig. 4.14A), however, still
show annular flow. An interfacial boundary in the narrow corners of the
channel is difficult to detect, tough. The same effect can be seen in the
cross section view (Fig. 4.14B).
In additional experim(~nts we analyzecl the differences in the :)-
dimensional flow pattern for the two fluidic systems water - nitrogen and
74 4. Flow regime

(A) (8)

Figure 4.14: XTM picture showing the phase distribution inside


the microchannel for VL = 0.2 ml/min and Vc = 3.0 ml/min (Exp.
4) in the direction of the flow (A) and the ;_~ dinwnsional cross section
(B).

etluUlol - nitrogen. The flow in a rectangular microchannel with hydraulic


diameter D h = 200 J-Lm and volumetric flow rates Vc = 8.0 ml/min and
VL = 0.1 ml/min (jc = 2.96 m/s, .iL = 0.04 m/s) was recorded. Fig.
4.15 shows the 3-dimensional reconstruction of the phase interfaces. The
phase boundaries were extracted using an intensity threshold value in
the XTM reconstructions. The XTM recording displays a clear gas-liquid
interface, coming from stationary flow using water - nitrogen. The re-
construction of the ethanol - nitrogen flow shows artifacts in the region
of the phase interfaces caused by marginal wavy flow. The void fraction
using water is remarka,bly higher than for ethanol. Besides the differences
in the phase distributions, the surface condition of the microchannel can
be visualized. For the channel sidewalls, a wavy pattern perpendicular
to the flow direction is present. This surface roughness is a result of the
lack of resolution in the used etching mask (transparency film, see chap.
:3.1.1). The roughness of the channel floor is significantly lower.
In intermittent flow~ the uneven velocities of the liquid and the gas phase
(slip) cause Cl liquid mass transfer from one liquid plug to the next. This
4.:3 :3-dimensional flow pattern visualization using XTM 75

Figure 4.15: 3-dimensional phase interface reconstruction for a wa-


ter - nitrogen (A) and an ethanol - nitrogen (B) two-phase flow in a
microchannel with hydraulic diameter D h = 200 /-Lm and volumetric
flow rates Vc = 8.0 mljmin and VL = 0.1 ml/min Uc = 2.96 m/s,
.iL = 0.04 m/s).

transfer occurs through small amounts of liquid concentrated in the cor-


ners of the rectangular channels. Little information on the shape and
size of these so called menisci is reported in the literature. As seen in
the previous figures, X-Ray Tomographic Microscopy (XTM) b a suit-
able method to ;j-dimensionally visualize stationary two-phase flows in
microchannels. Flow patterns with unsteady phase distribution result in
blurry reconstructions. For intermittent fiow with elongated gas bubbleH,
the phase interfaces are spatially and temporally stationary in the region
of the elongated bubble (highlighted region in Fig. 4.16A). Hence XTM
is expected to reconstruct a blurry image of the stationary part of the
intermittent flow and therefore a visualization of the menisci responsible
for the mass transfer from one liquid plug to another.
Fig. 4.16 (B and C) clearly shows the possibility of X-Ray Tomographic
Microscopy the visualize the liquid menisci in intermittent flow. The void
fraction in the region highlighted in Fig. 4.16A increases to cc ~ 0.96,
whereas the temporally averaged void. fraction for this flow configuration
is determined as cc; ~ 0.76.
76 4. Flow regime

100 flm

120 flm

Figure 4.16: Intermittent flow with elongated bubble recorded us-


ing LIF (A). Stationary part highlighted. XTM reconstruction show-
ing the liquid menisci (E, C) responsible for the mass transfer be-
tween neighboring liquid plugs.

4.4 Summary and conclusions

4.4.1 Flow regimes in rectangular microchannels

The presented flow regime experiments highlight the problem of mi-


crofluidic applications. In contrast to macro-scale applications, where
the gravity force iH the dominant force a,cting on the multiphase flow,
the surface forces mainly determine the flow pattern in microfluidic de-
vices. The flow regime is influenced not only by the surface tension of
the apparent fluids, but also by the channel surface properties, and, not
negligible, by the channel cross-section itself. For the microchannels uti-
4.4 Summary and conclusions 77

lized in the present work, a detailed set of flow regime maps was derived,
showing the differences in the flow conditions for varying liquid proper-
ties and channel hydraulic diameters. The individual flow patterns show
large transition regions, where multiple flow patterns coexist. Sharply
detached flow regimes are not detectable. In contrast to macro-scale
gas-liquid pipe flow, intermittent flow is dominant in microchannel flow.
Some flow patterns, such as the stratified, the wavy, or the dispersed
flow found in large diameter channels, were not observed for the applied
flow conditions in our microchannels. A low surface tension between the
two fluids favors the annular flow regime, the transition region to inter-
mittent flow is shifted towards lower superficial gas velocities. The same
effect can be found when increasing the channel hydraulic diameter.
The universal flow pattern map presents the regime boundaries for a
wide range of flow and fluid conditions. The flow regime map using di-
mensionless parameters instead of the commonly used superficial gas and
liquid velocities is able to predict sufficiently the apparent flow pattern
in microchannels of rectangular cross-sections. Nevertheless, one has to
consider that important influencing factors as the temperature (affecting
directly the fluid properties), different channel cross-sections, the chan-
nel material, and the surface roughness were not varied or have been
neglected in the flow regime map.
The comparison of the experimental data with flow pattern maps ob-
tained for large scale pipes as well as for microfluidic geometries shows
mostly insufficient agreement. It was shown that the best agreement is
achieved with flow pattern maps constructed for channels with similar
cross-sectional shape (rectangular cross-section), even for large differ-
ences in the hydraulic diameters (Dh,map/ Dh,exp ~ 5).
As a basic principle, an experimental validation of the occurring flow pat-
tern is indispensable for the design of any microchemical or microfluidic
device.

4.4.2 Flow visualization using XTM


Considering the presented experiments, X-Ray Tomographic Microscopy
(XTM) proves to be a suitable method to analyze and study the flow pat-
78 4. Flow regime

tern in microchannels. In addition to the commonly used 2-dimensional


optical flow vit:malization techniques, XTM provides 3-dimensional in-
formation of the multiphase flow. Due to the channel material used and
thus the required long exposure times, this measurement method is ap-
plicable for stationary flow patterns only. Annular flow is shown to be
adequate.
Concerning multiphase gas-liquid chemical reactions in microsystems,
requiring the existence of a catalyst material, which has to be inserted
into the channel, blocking the optical access, the XTM measurement
method enables the analysis of the degree of wetting of the catalyst
structure. Thanks to its maximum possible spatial resolution of less than
2 Jjm, even small structures can be analyzed.
All of the presented experiments are made using the phase contrast
method (edge-enhancement method) of the XTM station. Due to the
small difference in the X-ray absorption coefficient of the water and the
nitrogen, the two phases are hard to differentiate using the absorption
method. Knowing from several applications in the field of medicine, it
is possible to artificially increase the difference in beam absorption by
the addition of an X-Ray contrast medium. For example, iodine could be
added to the liquid phase (soluble in water). Owing to the high atomic
weight of iodine, the contrast medium will absorb more X-rays than the
liquid by itself. Therefore, the water containing the contrast agent will
stand out more clearly in the images. Using the absorption method, it
is possible to gain information on the (temporally averaged) volumetric
gas content by means of intensity values of the images. This volumetric
content could be computed also for not stationary flows and expand the
XTM-measurement method to all multiphase flow patterns.
79

Chapter 5

Pressure drop

5.1 Pressure drop in microchannels


The design of optimal microdevices, for example heat exchangers, re-
quires accurate pressure drop models. Only limited research on two-
pha..,,;;e flow pressure drop in small diameter tubes has been conducted,
even tough it is often the crucial factor for the piping and process sys-
tem. In the literature, a large number of pressure drop correlations exist
for large diameter round tubes. But there are significant differences be-
tween large round tubes and smaller channels in the relative magnitudes
of gravity, shear, and surface tension forces, which determine the flow
regime established at a given combination of liquid and gas-phase ve-
locities. Thus, extrapolation of large round tube correlations to smaller
diameters and non-circular geometries could introduce substantial errors
into pressure drop predictions.
We present experimental results of pressure drop measurements in two-
phaBe gaB-liquid flows using water and nitrogen as working fluids. Ho-
rizontal, rectangular channels with hydraulic diameters between D h =
120 p,m and Dh = 160 p,ffi are considered. According to Triplett et al.
(1999a) and Kawahara et al. (2002), the friction pressure drop is plotted
against the superficial gas velocity.
80 5. Pressure drop

0.80 MO
0.78 Oh -120 J.ll11 0.45
-{}- i L =0.28 m1s
0.10
-0- iL =0.42 m1s
0.40
.. ...
---..
.
;::'

ac.
0.88

0.80
~ iL - 0.56 m1s
ic.
0.3&

0.30 0-
--0-
--0
0.55 0
e e
"0 0.25
"!!:> 0.50
0.45
!!
:> -n
0.20 -.::r
'"'"!! 0.40
'"~
0.. 0.1& Dh -133.3 "m
0..
0.35
0.10
=
-{}- iL 0.25 m/s
0.30 -0- iL " 0.38 m/s
0.25
0.0& =
-&- iL 0.50 m/s

0.20 0.00 I
2 3 4 & & 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.8 4.0 4.8 8.0 s.& M
Superficilll glls velocity i e [mls) Superficial gas velocity iG [m/s)
0.80 040

0.45 Oh -166.1 J.ll11


0.35
-{}- k =0.12 m/s
-0- iL =0.18 m/s
0.40

"'.."
030
~ iL =0.24 m/s
B-
0.35
"'"
.. 025
c. 0.30 c.
e
"0
e
"0
0.25 0.20
!! !!
iil
~
0.20 c--
0.15

0.. 0.18 Oh" 150 I'm ~ " ~


0..
0.10
0.10
-{}- k =0.14 m/s
-0- k =0.21 mls 0.08
0.05 ~ iL =0.28 m/s
0.00 000
0.0 M 1.0 1.5 2.0 2.8 3.0 M 4.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.8 4.0
Superficial gas velocity iG [m/sI Superficial gas velocity ie [m/sI
0.50

0.46 Oh -180 J.ll11


-{}- iL " 0.10 m/a
040
-0- k - 0.15 m/s
i 0.35 ~.,.. i ,. 0.20 m/s
L
ac. 0.30
e
~-~
." O,.2S
!!
iil 020
l!!
0.. 015
0--- ::>-
--0-_
, ~,

--0 0
010

005

0.00
0.0 0.4 0.8 1.2 1.6 2.0 2.4 2.8
Superficial gas velocity i e [m/a]

Figure 5.1: Pre~~ure drop of two-phase nitrogen-water flow in de-


pendence of the channel hydraulic diameter, and the superficial gas
and liquid velocities. Pressures are measured with an accuracy of
O.03 bar.

For all the presented channel diameters D h , it can be seen in Fig. 5.1 that
higher superficial liquid velocities, iL, lead to larger pressure drops. For
5.1 Pressure drop in microchannels 81

larger hydraulic diameters, D h , the pressure drop decreases for increasing


superficial gas velocities, ja. This effect can be explained by a change
in flow pattern from intermittent flow to annular flow. The gas-phase
pressure drop is substantially reduced in annular flow. However, if the
superficial gas velocity was further increased, the pressure drop would
increase again. The two-phase pressure drop features a local minimum
in the annular flow regime.
For the smallest channel hydraulic diameter (D h = 120 j,Lm), all the ana-
lyzed flow conditions yield intermittent flow even at high superficial gas
velocities. Hence the pressure drop increases with increasing superficial
gas velocity. A further increase in the volumetric flow rate would cause
the transition to annular flow and therefore the reduction of the pressure
drop as demonstrated for the larger hydraulic diameter channels.
A similar effect is found for the microchannel with D h = 133.3 j,Lffi, where
all the measured data is in the transition regime between intermittent
and annular flow. The pressure drop is not decreasing significantly due
to the present change in flow regime.
Figure 5.2 shows the pressure drop for a microchannel with rectangular
cross-section and hydraulic diameters D h = 120 j,Lm. For this channel
diameter, the pressure drop increases continuously with increasing gas
and liquid volume flow rate. As seen in Fig. 5.1, intermittent flow exists
for all the measured configurations.
For a channel width of 300 j,Lm (D h = 150 j,Lffi, Fig. 5.3), nearly constant
pressure drops, independent of the gas flow rate, can be found for low
liquid flow rates. The low superficial liquid velocities produce a flow
pattern in the transition region from intermittent to annular flow, or
strictly annular flow. As demonstrated in Fig. 5.1 (Dh = 133.3 j,Lm),
the pressure drop is nearly independent of the superficial gas velocity.
For higher liquid velocities, the pressure drop features a maximum in
the intermittent flow region and a remarkable decrease at higher gas
velocities after the pattern change to annular flow. At constant ga.-';; flow
rates on the other hand, the pressure drop increases with increasing
liquid flow rates.
A similar behavior can be found for the largest channel diameter (Fig.
5.4). At low volumetric liquid flow rates, annular flow is present even
82 5. Pressure drop

for low gas flow rates. The pressure drop is decreasing with increasing
gas flow rates. For higher supprficial liquid vclocitiPH, the pressure drop
maximum at the change of flow regimes is less distinct than for the
D h = 150 J-Lm channel.

Gas mass flux, m' gas [kg/(m2s)]


Liquid mass flux, m' wole, [kg/(m2s)]

Figure 5.2: Pressure drop for a microchannel with rectangular


cross-section and a hydraulic diarrwter of D h = 120 JlIn.

5.2 Comparison with correlations

As seen in chapter 2.1.3.2, the total pressure drop in horizontal chan-


nels is composed of the friction pressure drop, the pressure drop due to
acceleration, the pressure drop due to reductions in diameter, and the
geodetic or gravitational pressure drop (Eq. 2.14). The dominant con-
tribution to the total pressure drop is provided by the friction pressure
drop. The numerous pressure drop models found in the literature pre-
dict the friction pressure drop in pipes and channels. However, in the
5.2 Comparison with correlations 83

0.4"

0.35,

e.. 0.3 ..
a.
e
~ 0.25.
'"
!!:'"'" 0.2 ..

0.15.

".A

Gas mass flux, m' gas [kgl(m2s)J


Liquid mass flux, m' ",mer [kgl(m2sl]

Figure 5.3: Pressure drop for a microchannel with rectangular


cross-section and a hydraulic diameter of D h = 150 {lrn.

previous chapter, the total measured pressure drop was presented. To


adequately compare the measured data with the different friction pres-
sure drop models, the ratio of the friction pressure drop to the total
pressure drop has to be known. Therefore, the pressure drops due to
acceleration, due to reductions in diameter, and due to the gravity were
calculated and then subtracted from the measured total pressure drop.
For the pressure drop due to acceleration we considered the following two
models. Tlw first model (Eq. 5.1) is suggested by Kawahara et a1. (2002)
and uses the volumetric gas content, EG, and the superficial velocities at
thE:~ inlet and the outlet of the microchannel, .7G,L. The void fraction is
calculated from the gas density at inlet and outlet pressure.
84 5. Pressure drop

0.26,

0.24"
~ 0.22,
eo
g. 0.2.
-0
e 0.18,
::I

e0.16"
ID

a.
0.14 ..
0.12,

Gas mass flux, m' D"' [kg/(m2s)]


Liquid mass flux. m' water [kg/(m2s)

Figure 5.4: Pressure drop for a microchannel with rectangular


cross-section and a hydraulic diameter of Dh = 160 /fm.

~Pacc =

(5.1)

Additionally the following simplistic model is used

~Pacc =(JL
( w L2 (1 - cc;) + (JC; Wc;2 cc; ) . (5.2)

The pressure drop caused by the reduction of the diameter is estimated


5.2 Comparison with correlations 85

by a homogeneous model by Kawahara et al. (2002)

where 'Y is the ratio of the cross-sectional areas of the multiphase part of
the microchannel and the inlet cross-sections, and Cc is the contraction
coefficient, which is a function of 'Y

Achannel
'Y= (5.4)
Ainlet

1
(5.5)
Cc = 0.639(1 - 'Y)O.5 + 1.
Not taken into consideration are the pressure drops over armatures and
valves and the gravitational pressure drop.
Figure 5.5 shows the contributions of the friction pressure drop, the
pressure drop due to acceleration, and the pressure drop caused by a
reduction in diameter to the total measured pressure drop for the two
different hydraulic channel diameters Dh = 120 J.lm and Dh = 160 J.lm.
Even though the pressure drop due to area reduction increases with
increasing superficial gas velocity, the contribution to the total pressure
drop remains negligible. The friction pressure drop to total pressure drop
ratio, tlPfric/ tlPtot, varies between 99.5% for large hydraulic diameters
and low superficial gas and liquid velocities, and 87.4% for small diameter
channel using high superficial velocities.
86 5. Pressure drop

100 16

98 o~
14

96
12
'Q. t>
94 '" '0
~
AI
10 JJ

-
B 92
!
a.~
-
<I

a.~
<I
90

88
;;r _.
.. -
__ f;l-
8

S
t>
'0

i
86 0
t:.
~Pt1c (friction) (left axis)
~p... (acceleration) (right axis) ..
84 'l ~Prod (reduction) (right axis)
v .. ia '" 0.20 mls 2
82 - JG '" 0.40 m1s
80 0
0 2 4 6 8
Superficial liquid velocity k[m/s]
A
100 16

98
14

96

94
12 .g
'1*
AI
10 JJ

-
92 0 ~lric (friction) (left axis)
~ ~P... (acceleration) (right axis)
!

-
<I

a.~
<I
90

88
'l ~P..d (reduction) (right axis)
.... ia '" 0.10 m1s
- ia '" 0.20 mls
8

6
t>
'0

i
86
4

84
2
82

80 0
0.0 0.5 1.0 1.5 2.0 2.5 3.0
Superficial liquid velocity jL[m/s]
B
Figure 5.5: Pressure drop due to friction (0, left axis), acceleration
(6.), and reduction (V') (both right axis) in percent of the total
pressure drop for two different superficial gas velocities against the
superficial liquid velocities for hydraulic channel diameters D h
120 /-Lm (top) and D h = 160 J.Lm (bottom).
5.2 Comparison with correlations 87

5.2.1 Homogeneous pressure drop model

5.2.1.1 Main equations of the homogeneous pressure drop


model

The homogeneous pressure drop model calculates the frictional two-


phase pressure drop, (llpj Ill)TP, using

(~n TP = fr"c ~h ~:. (5.6)

For the homogeneous mixture density, (l H, several definitions can be


found in Coleman and Krause (2004):

homogeneous density
XG 1- XG)-l
(lH,hom = (- + --- (5.7)
(lC (lL

momentum density

(5.8)

energy density

{!H,enr =

(5.9)

Both the momentum density model (Eq. 5.8) and the energy density
model (Eq. 5.9) require the use of a void fraction (EG) model (see chap.
6.1).
The friction factor, frric, is calculated using the Blasius equation,

c (5.10)
frric = R m'
eH
88 5. Pressure drop

where C and m are constant factors according to the present flow con-
dition::;. For laminar flows, m is equal to unity, m = 1, and C is 64. For
two-phase flows the friction factor is calculated using the homogeneous
Reynolds number

rh D h
ReH = - - - (5.11)
TJH

Again, for the homogeneous viscosity, TJH, several approaches can be used
(Kawahara et al., 2002):

TJH,l TJL (5.12)


TJH,2 xG TJG + (1 - xG) TJL (5.13)
TJH,3 EG TJc + (1 - EG) TJL (5.14)
TJH,4 TJL(l- EG)(1 + 2.5EG) + TJG EG (5.15)

TJH,5
(XG + 1 - XG) -1
(5.16)
TJG TJL
TJL TJG
TJH,6 (5.17)
TJG + x~4 (TJL - TJG)

The homogeneous pressure drop model is very simplified, but features a


wide application area due to the large number of possible variations and
combinations.

5.2.1.2 Comparison with experimental data

According to the different possibilities calculating the homogeneous den-


sity and the viscosity, a large number of applications of the homogeneous
pressure drop model is possible. This is especially the case if considering
the use of several different approaches to define the volumetric gas con-
tent, CG (Eqs. 5.8 and 5.9). For the comparison with the experimental
data, we used the void fraction correlation suggested by Serizawa et al.
(2002) (chap. 6.2.4), which is only dependent on EG.
5.2 Comparison with correlations 89

In Tab. 5.1 the root mean square errors in percent between the different
possible pressure drop calculations and the experimental data is shown.
It can be seen that the homogeneous density (Eq. 5.7) is not suitable for
microfluidic applications except for the combination with the viscosity
calculated using Eq. 5.14. Surprisingly, this density approach does not
only give the two predictions with the worst conformance (version 1 and
2) but also the version with the best conformance with the experimental
results (version 3, using Eqs. 5.7 and 5.14).

Density
Visc.
llH,hom llH,mom llH,enr
17H,l 1080% 211% 41%
17H,2 1050% 205% 41%
rlH,3 21.5% 66% 83%
17H,4 181% 30% 59%
17H,5 450% 64% 33%
17H,6 805% 152% 21.6%

Table 5.1: Root mean square errors in percent of the different


possible calculation methods using homogeneous density approach.

5.2.2 Friction pressure drop by Lockhart-Martinelli

5.2.2.1 Theory of the Lockhart-Martinelli method

The pressure drop calculation method by Lockhart and Martinelli (1949)


is based the assumption that the friction pressure drop of a two-phase
flow is proportional to the pressure drop of the corresponding single-
phase gas or liquid flow,

( tl
P
)
tll TP
= <P~ (tltllP ) G = P
<Pt (tl )
tll L
(5.18)
90 5. Pressure drop

using <1>i ,G as factor of proportionality. These factors can be calculated


using the two-phase multiplier X,

X = ((Dt.p/ Dt.l)L) 1/2


(5.19)
(Dt.p/ ~l)G '
and according to Chisholm (1967),

<I>b = 1 + ex + x 2 (5.20)

(5.21)

The two single-phase pressure drops in Eq. 5.19 are determined by a


homogeneous, single-phase model. Hence the two-phase multiplier X is
dependent on the single-phase friction factors !rric,L and !fric,G. Chisholm
(1967) defined the factor C = 5 for strictly laminar flow (ReL,G < 1000).
This factor is the only possibility for this large scale model to adapt to
the microfluidic conditions. In the literature, two adaptations can be
found by Mishima and Hibiki (1996) (Eq. 5.22) and by Lee and Lee
(2001) (Eq. 5.23).

C = 21 (1 - e-O.319.Vh) (5.22)

C = A . Aq . wr . Reto (5.23)

The equation of Mishima and Hibiki (1996) is based on an air-water two-


phase flow in round and rectangular channels with hydraulic diameters
D h = 1 - 4 mm. Lee and Lee (2001) also used an air-water system in
horizontal, rectangular channels with diameters of D h = 0.78 - 6.67 mm.
The dimensionless parameters A, W, and ReLo (Reynolds number using
liquid only) are defined as follows:

2
A- 1]L W = 1]L(jG + jL) , (5.24)
- (}L (J Dh ' (J
5.2 Comparison with correlations 91

Lee and Lee (2001) defined the parameters A, q, T, and s empirically for
their channels and laminar-Iaminar flow conditions as A = 6.833 X 10- 8 ,
q = -1.317, T = 0.719, and s = 0.557. Depending on the choice of the
factor C, the model by Lockhart and Martinelli (1949) can be adapted
to small-scale channel flow.

5.2.2.2 Comparison with experimental data

The pressure drop model by Lockhart and Martinelli (1949) was defined
for large diameter pipes, even though no area of validity is mentioned.
The data calculated using the original approach for strictly laminar flow
(C = 5) overestimates the pressure drop with a mean deviation of t1p =
0.55 bar and a standard deviation (j = 0.40 bar.
2.0 - t - -.........- - - ' - - - ' - - - . . l . - - - ' - - - - : : : - ' - - - - - - L - - t _

Ri 1.8
0
So
a; 1.6 0
0
0

a.~
0
0
1.4
<I 0
0
e
a. 1.2 0
(\\)

o 0
'0
~ 1.0
::J
:g 0.8
~
a.
'0 0.6

~ 0.4
:c
~
a.. 0.2

0.0 --io"'==--......... ---.------,,----..-------,----r-.......,.--+


0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Measured pressure drop 6Pexp [bar]

Figure 5.6: Comparison of the two-phase frictional pressure drop


between experimental microchannel data and the model predictions
by Lockhart and Martinelli (1949) using C = 5 as parameter.

Knowing the fact that the model is intended for two-phase flow in large
diameter pipes, one could assume that the agreement with the experi-
mental data of the microchannels with larger hydraulic diameter is bet-
92 5. Pressure drop

ter than for smaller diameters. Figure 5.7 shows a large dispersion in the
pressure drop for all the channel diameters.
If the data is varied by the ratio of the superficial gas to liquid velocities,
it can be seen in Fig. 5.8 that the agreement with the model is better for
flow conditions using smaller ratios jG/jL. This trend can be noticed for
all pressure drop models originally designed for large diameter channels
or pipes. Those models were mainly developed by the oil industries for
predicting the pressure drop in pipelines. Consequently low gas flow rates
were considered.
2.0 -1-_--L-_---L._----''--_..L..-_-'------,,---'-_--I-_ __+_

ffi 1.8
0
~
11 1.6
g 1.4 0
0
0
Q.
<l 0 e:J

eQ. 1.2 0
o
'0
e 1.0
:::J Oh = 120 /lm
l2 0.6 o = 133.3 /lm
e
Q. ~ ~ A
Oh
Oh = 150 /lm ~
'0 0.6 'J Oh = 155.6 /lm

~ 0.4
Oh =160 /lm
'0
e
Q. 0.2
- - - +/- 20%

0.0 ...joooI::=----.----r---,r---.....---,--~-~-_+_

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8


Measured pressure drop ~Pexp [bar]

Figure 5.7: Comparison of the pressure drop depending on the


hydraulic channel diameter.

With the variation of the parameter C, Mishima and Hibiki (1996) and
Lee and Lee (2001) tried to adapt the Lockhart-Martinelli model to
the special conditions of microsystem two-phase flows. Both methods
use empirical parameters (Fig. 5.9). The two models show better agree-
ment with the present experimental data. However, still a large scatter
is present.
5.2 Comparison with correlations 93

la 1.8 ...
a.
li 1.6
"' ...
a.~
...
1.4
<J
a.

l::! 1.2 ...
"C o
...
(\)
~
1.0

~ O.B jG/jL <3


~ o 3<k/jL <=6
a.
"C 0.6 ;~ '" B<JG/J L <"10 ~
<;; 10 <iclIJL <=15
~ 004 ... k I JL > 15
'C
~
Q. 0.2 - _. +/-20%

0.0 --+-"'=---.-----r'- r---...--~-__r_-~-__+

0.0 0.1 0.2 0.3 004 0.5 O.B 0.7 0.8


Measured pressure drop ~Pexp [bar]

Figure 5.8: Comparison of the pressure drop depending on the


ratio of the superficial gas to liquid velocity.

A variation of the parameters in

and C = A . Aq 'liT . Reio (5.25)

to fit the models to the present data does not lead to a significant further
improvement of the pressure drop predictions. The disadvantage that the
obtained set of parameters is valid only for a specific channel geometry
or flow condition outweighs the advantage of the slightly better results.
According to the constant factor C = 5 proposed by Chisholm (1967)
for strictly laminar flow, Kawahara et al. (2002) used the experimen-
tally determined average of the factor C as basis for their prediction
model. As the data is extracted from the experimental results, using
C = 0.24 yielded good agreement of the predicted and the measured
pressure drops. However, the consistency of our experimental data with
the Kawahara prediction is less accurate than with the data obtained
using the correlation by Mishima and Hibiki (1996).
Finally, a comparison of the two-phase frictional pressure gradient data
with the predictions of the Lockhart-Martinelli correlation using different
94 5. Pressure drop

0.8 +-_....L-_...J-_---L..._----L.._--L_--L_.,......l._----::>I-

~
o
! 0.7
"

a.
I 0.6

<J
a. 0.5
o...
"C
o

e 0.4
:::I
:z
...a.
Cl) 0.3

~ 0.2
U i
'6
e
ll.
0.1

0.0 -F----.----,----.------r------,----.----r-__;__
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Measured pressure drop ~Pexp [bar]

Figure 5.9: Comparison of the pressure drop predictions by


Mishima and Hibiki (1996) (filled circles) and Lee and Lee (2001)
(open circles).

C-values is shown in Fig. 5.10, including C = 5, C = 9 . 10- 4 from


Eq. 5.22 using Dh = 133 j.Lm, C calculated from the model by Lee
and Lee (2001) (Eq. 5.23), and C = 0.24 (Kawahara et al., 2002). The
conventional value of C = 5 again significantly over-predicts the two-
phase frictional multiplier of the experimental data. On the other hand,
good agreement with the experimentally derived ~L,G (within 10%)
was obtained with the use of the C-value given by Lee and Lee (2001)
and Mishima and Hibiki (1996) (the data of those two models nearly
coincide) and by Kawahara et al. (2002).
When using the pressure drop prediction method by Lockhart and Mar-
tinelli (1949), the approach for the factor C supposed by Mishima and
Hibiki (1996) is recommended because of the good agreement with the
experimental data and its simplicity.
5.2 Comparison with correlations 95

"'-l
- C 5 (Chisholm)
- - - C 0.0009 (Mishima) A
..
e
,~ 100
C .0.0113 (Lee)
_. - C .0.24 (Kawahara)
0.
'';::
o experimental data
"3
E
iii
g 10
13
:E
Q)
Ul
RI
.
6
~
0.1 +---r--r'''T''T'"","r--r-'''T""'I''''''''I''nT-T'''''''I''""T'''IrTTI'IT"""'''''''""''T'"T'T"l".,.,f-
01 1 10 100 1000
Lockhart-Martinelli parameter X

- C 5 (Chisholm)
- _. C 0.0009 (Mishima)
B
----- C .. 0.0113 (Lee)
- . - c 0.24 (Kawahara)
o experimental data

.,4--.....-.........
7 ,
----"""T'"---.---.----.r---r-.....-..--,......j.
6 7 8 !I
10
Lockhart-Martinelli parameter X

Figure 5.10: Variation of two-phase friction multiplier data with


Lockhart-Martinelli parameter (A) and an enlarged detail (B).

5.2.3 Friction pressure drop by Chawla

5.2.3.1 Empirical pressure drop model by Chawla

The pressure drop model by Chawla (1968) is based on the assumption


of an annular flow pattern and was developed for large diameter pipes.
96 5. Pressure drop

The pressure drop of the gas core flow is calculated using the Blasius
equation. The main equation of this model is

0.3164. m2 x'{.j4 ( 1 -:Ea ) 19/8


------=----=--:....,.... 1+ -.- -- (5.26)
.D ) 0.25 Xc . , . ~GL
2 . Dh . (!c ( ~Gh "'

The idealized model is adapted to the reality by means of the two-phase


ft.ow parameter " depending on ft.uid properties, ft.ow conditions, and
several empirical parameters.

1"_(-3+
~ - El E
_3)-1/3 (5.27)
2

I = exp [0.9592 + In(iJ)]


3
with E (5.28)

with e,3 = exp [ ( 0.1675 - 0.055ln (~) ) In( I]i) - 0.67], (5.29)

where Kid is the surface roughness of the channel, iJ a ratio of the ft.uid
densities and viscosities

(5.30)

with (5.31)

with (5.32)
5.2 Comparison with correlations 97

As seen in Eqs. 5.28 - 5.32, the model suggested by Chawla (1968) uses
several empirical paranleters for the definition of the two-phase flow pa-
rameter ,. With this large number of influencing variables, a parameter
variation can provide acceptable agreement of the calculated pressure
drop with the experimental data. Unfortunately, a single set of parame-
ters is valid only for a specific condition (channel geometry, flow condi-
tions, and fluid properties).

5.2.3.2 Comparison with experimental data

In Fig. 5.11, the present experimental data are compared with the pres-
sure drop model by Chawla (1968). The main equation for the two-phase
flow parameter (is used as presented in Eq. 5.27. A remarkable spread of
the calculated values can be found. The correlation shows no tendency in
over- or underestimating the actual pressure drop. The mean deviation
of the model data is /:::,.p = 0.52 bar with a large standard deviation of
a = 1.31 bar. Thus, the original pressure drop model by Chawla (1968)
is not capable of predicting the pressure drop in microchannels.
In the Eqs. 5.28 - 5.30 Chawla introduced several empirical parameters.
Using the lea..~t squares method, the best fit for our channel geometries
and flow conditions was determined (Tab. 5.2). With this new parameter
set, the mean error of the pressure drop prediction can be minimized to
approximately 30% (Fig. 5.12).
An important factor of this prediction method is the surface roughness
K / d. The influence of K / d on the pressure drop prediction quality is
presented in Tab. 5.3. The mean error of the best parameter fit increases
with decreasing surface roughness.
The original pressure drop prediction method by Chawla (1968) is not
suitable for microfluidic applications. With the help of a parameter vari-
ation for the empirical values introduced, the mean prediction error can
be minimized, but remaining faults on the order of 30% and no distin-
guishable basic over- or underestimation of the present pressure drop
show that this method is inapplicable for small-scale channels.
98 5. Pressure drop

3.0
0
';:'
2.8
1-- - +/0 20 %1
.[ 2.6
0
1i 2.4
~
Cl.
2.2
0
<l 2.0 0
a.
e
u
1.6
1.8
0

0
~ 1.4 0
:;, 0
VI 0
VI 1.2
~
Cl. 1.0
uQ) 0.8
1)
0.6
'5
~ 0.4
a. 0.2
0.0 -+-~~~~Dap.l~--=---,-_~---,....,._-+
0.0 0.1 0.2 0.3 0.4 0.5 0.8 0.7 0.8
Measured pressure drop ~Pexp [bar]

Figure 5.11: Comparison of the two-phase frictional pressure gra-


dient between experimental microchannel data and the model pre-
dictions by Chawla (1968).

Parameter Chawla present study


a 1/6 1/6
b -0.9 -2.0971
c -0.5 -3.7455
d 0.9592 23.0773
e 0.1675 0.0474
f 0.055 0.01893
g 0.67 0.66773

Table 5.2: Parameter settings of the original pressure drop correla-


tion by Chawla (1968) and of the best fit for the present conditions.
5.2 Comparison with correlations 99

0.8
~

J!1li
0.7

0.
~ 0.6
<.1
0. 0.5
e
"l!? 0.4 ---
0
0

:J
~
...
Cl>
0.
0.3

"~ 0.2
0
'0
l!? 0.1
a..
0.0 ~--J;L=;::-----r----r----.--.------r----r---+

0.0 0.1 0.2 0.3 0.4 0.5 0.8 0.7 0.8


Measured pressure drop 6p"xp [bar)

Figure 5.12: Comparison of the two-phase frictional pressure gra-


dient between experimental microchannel data and the model pre-
dictions by Chawla (1968).

Kid 10 1 10- 2 10 .;j 10 4 10 -b 10 '0


mean error [%] 29.9 30.2 30.2 61.5 63.7 63.1

Table 5.3: Influence of the surface roughness on the quality of the


parameter variation.

5.2.4 Friction pressure drop correlation by Friedel

5.2.4.1 l3asics

The pressure drop correlation by Friedel (1978) is based on a large


database of experimental data for two-phase flow conditions in hori-
zontallarge-scale channels. As in the method of Lockhart and Martinelli
(1949), the two-phase proportionality factor <pi is determined, which is
multiplied with the single-phase liquid pressure drop to yield the two-
100 5. Pressure drop

phase pressure drop.

( ~p)
~l TP -
- <1>2
L
(~p)
~l L
(5.33)

if>i . (it't = A + B xb(l- xc)' (~~ r


.. (~~r (1- ~~r F~pWeTP (5.34)

The parameter A is defined a..'l

A -- (1 - XG
.)2 + XG
2 {}L ffric,G
(5.35)
{}G frric,L

The single-phase friction factors are calculated using Eq. 5.10. The two-
phase Froude and Weber numbers are defined as

X
'2
x'2 D h
FrTP = D 2 WeTP = 2' (5.36)
9 h {}m a {}m

For {Jm, the homogeneous densities {JH (Eqs. 5.7 - 5.9) or the liquid
density {JL can be used.
Even for large diameter pipes, different values for the parameters B, 5,
t, U, v, w, p, and r can be found (Friedel, 1978; Triplett et al., 1999a).
This indicates that this pressure drop correlation is able to predict the
pressure drop for a single channel geometry and flow condition with high
accuracy, but the same parameters can yield wrong results for any other
fluidic system.
5.2 Comparison with correlations 101

region of validity
Gas mass fraction Xc < 1 [-]
Total mass flux rh 5 -10330 [kg/m 2 s]
Pressure p 1-171 [bar]
Density ratio eL/ec 6 - 1194 [-]
Viscosity ratio 'r/L/'r/G 6 - 5991 [-]
Hydraulic diameter Dh 1 - 257.4 [mm]
Surface tension a 15 - 80.10- 3 [N/m]

Table 5.4: Area of validity of the pressure drop model by Friedel


(1978).

5.2.4.2 Comparison with experimental data

The pressure drop model by Friedel (1978) is valid for the flow, geometry
and fluid conditions found in Tab. 5.4.
Obviously, the agreement of the predicted pressure drop with the exper-
imentally determined pressure gradient is not expected to be good, as
the diameters of the present microchannels are smaller by a factor of
10 than the required ones.
f'V

According to Fig. 5.13, the model by Friedel (1978) mainly overestimates


the occurring pressure drop. As in the model by Lockhart and Martinelli
(1949) (see chap. 5.2.2) small jc/jL ratios yield better agreement of the
predicted pressure drop with the experimental data.
By means of varying the empirical parameters B, S, t, U, v, w, p, and
r, the model is capable to predict the pressure drop in the present mi-
crochannels accurately (Fig. 5.14). The mean error drops to 16.5% with
a standard deviation of 0.28 bar.
It can be shown that the prediction method by Friedel (1978) can be
very accurate for a specific channel geometry and flow conditions. For
a proper prediction of the pressure drop independent of experimental
measurements (no parameter variation possible), this correlation method
is inappropriate.
102 5. Pressure drop

2.0 +-_---L.._----L_ _'---_...l.-_--'-_----L._----l_--+


o
~ 1.8 1- -- +/0 20%1
B "ii 1.6 o
o
0
o

a.~
o
1.4 o
<1 o
ea. 1.2
0 0 o
"0
~ 1.0 CO::> 0
:;, 0
I/) 0
I/) 0.8 0 0
0 0
~ 0
13
a. 0
"0 0.6 00

~
'6 0.4
~
0.. 0.2

0.0 ....jooo=::;;;;.,.-r-----.-----,.-----....----.-----.--~--+
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Measured pressure drop 6p"xp [bar]

Figure 5.13: Comparison of the two-phase frictional pressure gra-


dient between experimental microchannel data and the model pre-
dictions by Friedel (1978) using the original parameters.

5.2.5 Friction pressure drop method by Muller-


Steinhagen and Heck

5.2.5.1 Basics on the correlation

Muller-Steinhagen and Heck (1986) made the observation that for a gas-
liquid mixtures with a volumetric gas content E = 0.5, the friction pres-
sure drop is similar to the pressure drop of a single-phase gas flow (E = 1),
which features the same mass flux as the two-phase flow.

2
t:J..p) rh (5.37)
A = ( t:J..l L = !fric,L 2 flL Dh

2
t:J..p) rh (5.38)
B = ( t:J..l G = frric,G 2 {!G Dh
5.2 Comparison with correlations 103

't:"
as
So 0.7

CL
i 0.6
<l
CL 0.5
...o
"0
~ 0.4
:::I
11)

~ 0.3
CL
~ 0.2

:a~ 0.1
a.
0.0 ~---r-----r----.-----.--r---~--r---+
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Measured pressure drop 6Pex:p [bar]

Figure 5.14: Comparison of predicted and measured pressure drop


using the method by Friedel. The empirical parameters are chosen
to minimize the error.

Again, the friction factors are calculated using Eq. 5.10. From Eqs. 5.37,
5.38, and

G = A+ B(B - A)xc, (5.39)

the two-phase pressure drop correlation by Muller-Steinhagen and Heck


(1986) can be calculated as

('i:) TP = G (1 - xa )'/3 + B Xb (5.40)

With the exception of the friction factors no adaptation to microfiuidic


systems can be made.

5.2.5.2 Comparison with experimental data

Besides the different approaches to calculate the friction factor (Eq.


5.10), the prediction method by Muller-Steinhagen and Heck (1986) al-
104 5. Pressure drop

lows no parameter variation to adapt the predictions to the special prop-


erties of small scale microchannels. Using the unchanged formulas, the
pressure drop is overestimated by a factor of 4. Obviously, this method f'V

is not applicable for pressure drop estimations in microchannels.


4.0 + - _ - - - L . _ - - l . _ - - - J L . . . . - _ . . . L . . - _ . . . . 1 - _ - - - L . . _ - - - - L_ __f_

; [ 3.5 o
1-- -+/.20%1
o 0
I
0.
3.0

<I 00 o
c.. 2.5 o 0
e
-0
~ 2.0 o
~
o
UJ
UJ
~ 1.5 o o
0. o
o
~ 1.0 o
1:)
ij
~ 0.5
a.
0.0 -+-~~~:;~-.---___.-__r-__r-___,.-_I_
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Measured pressure drop .1.Pexp [bar]

Figure 5.15: Comparison of the two-phase frictional pressure gra-


dient between experimental microchannel data and the model pre-
dictions by Muller-Steinhagen and Heck (1986).

5.3 Summary and conclusions

Pressure drop measurements in small diameter microchannels (D h =


120 ... 160 /Lm) were carried out. Depending on the superficial gas and
liquid velocities, pressure drops up to ~p = 0.8 bar were measured,
yielding a pressure drop per unit length of ~p/~l :::::: 21 bar/m. It was
shown that the pressure drop is strongly dependent on the volumetric li-
quid flow rate (superficial liquid velocity) and the two-phase flow pattern
present. Higher superficial liquid velocities cause an increase in pressure
drop, whereas an augmentation of the volumetric ga..o:;; flow rate can lead
to lower pressure losses, if pattern changes are present.
5.3 Summary and conclusions 105

The pressure drop for intermittent flow is significantly higher than for
annular flow patterns (at moderate flow velocities). Increasing the su-
perficial gas velocities, the pressure drop features a local maximum just
before changing the flow pattern from intermittent to annular flow. For
larger hydraulic channel diameters, the pressure drop at low superficial
liquid velocities is proven to be nearly independent of the gas flow rate.
As in macro-scale applications, the major contribution to the total pres-
sure drop in microchannels is shown to be the friction pressure drop. For
the considered hydraulic channel diameters, the ratio of friction pressure
drop to total pressure drop is determined to be in the range of 85 - 99%.
A comparison with the various friction pressure drop models presented
in the literature can be made. The prediction methods originally devel-
oped for large scale pipe flow (homogeneous pressure drop model, and
correlations by Lockhart and Martinelli (1949); Chawla (1968); Friedel
(1978); Muller-Steinhagen and Heck (1986)) substantially overestimate
the pressure drop for channel sizes below 1 mm. Extrapolating the pres-
sure drop from the larger channel sizes seems to be a wrong approach.
In most cases, the pressure drop prediction for small ratios of superficial
gas to superficial liquid velocities (je/ iL, high volumetric liquid con-
tent flows) are significantly more accurate than those for high gas void
fractions. This tendency is not surprising, thinking of the origin of the
prediction models (flow of oil in large-scale pipelines, low gas content).
In contrast to the velocity ratio, the hydraulic diameter does not have a
major influence on the prediction quality. By means of adaptations to the
models for microfluidic applications (Mishima and Hibiki, 1996; Lee and
Lee, 2001), the prediction quality can be slightly improved. A parameter
variation for the models using empirical parameters, yields better agree-
ment with the experimental data, loosing however the general validity
of the method.
The pressure drop, and therefore the dimensioning of the pumping sys-
tem of any microchemical device (lab-on-a-chip) is often the crucial com-
ponent. As seen for the prediction of the existing flow pattern, an experi-
mental validation of the occurring pressure drop is inevitable, designing a
novel microreactor design, as the known pressure drop prediction models
are not able to predict the pressure drop accurately.
106

Chapter 6

Void fraction

6.1 Void fraction results


The knowledge of the slip or local void fraction is essential in (micro-)
chemical applications, as it determines the residence time. Furthermore
it is an important basis for the calculation of the mass or heat transport.
Kawahara et al. (2002) estimated the void fraction experimentally by
analyzing the low liquid flow rates, where most of the recorded images
showed either liquid flowing alone (Ea = 0), or a gas core flow with a
smooth-thin liquid film or ring-shaped liquid film. For the latter, the void
fraction was assumed to be unity. By counting the number of images con-
taining each flow type, the time-averaged void fraction was determined
from the following expression

Number of gas core images


Ea = -- ----------- (6.1)
Total number of images counted

The experimental determination of the void fraction by Triplett et al.


(1999a) differentiated several flow patterns. The void fraction w~ esti-
mated by analyzing photographs of the test section. In the bubbly flow
6.1 Void fraction results 107

pattern, individual bubbles were assumed to be spheres or ellipsoids, de-


pending on their shape. In plug flow, the Taylor bubbletl were divided
into cylinders and spherical segments. In the bubbly flow, each photo
typically covered a large number of bubbles, thus providing a reasonable
volume-averaged estimate of the void fraction. In plug flow, the flow
pattern is relatively regular, and average Taylor bubble diameters and
liquid plug lengths were calculated from multiple photos and used for
the void fraction calculation. In the annular flow pattern, the vapor core
was divided into several cylinders, and the channel average void fraction
was calculated accordingly. Slug-annular and churn flow patterns were
the most difficult flow regimes to analyze. Triplett et al. (1999a) did not
include the void fractions associated with slug-annular flow in their anal-
ysis due to the high uncertainty. In churn flow, an average of 0.5 local
void fraction was assumed in segments of the flow field, where the gas
phase was dispersed.
The void fraction in the present experiments was estimated by analyz-
ing the images of gas-liquid flow recorded by the CCD camera. Each
image covered a distance of about 1.5 mm in the flow direction. The vol-
umetric gas content is dependent on the emitted fluorescent intensity of
the liquid phase enriched with Rhodamine B (Eq. 6.2). Hence, the void
fraction was determined using the intensity distribution of the recorded
images I(i, j)TP, calibrated with the corresponding intensities of a pure
gas (I(i,j)c, cc = 1) and liquid flow (I(i,j)L, cC = 0) as

_ 1 _ I(i,j)TP - I(i, j)c _ I(i,j)L - I(i, j)TP


(6.2)
cC - I(i,j)L - I(i,j)c - I(i,j)L - I(i,j)c .

We assume equal contribution to the total emitted intensity I(i,j)TP


of every liquid part along the viewing direction (depth of microchan-
nel). This means that an underlying liquid portion contributes the same
recorded intensity as a portion close to the camera. This assumption
is verified by the small sizes in microchannel flows. The illuminating
laser has to be ensured to not saturate the CCD camera to avoid loss of
information.
The experimental results show that differences in the fluid properties
have little impact on the void fraction for smaller channel hydraulic
108 6. Void fraction

diameters. For the investigated microchannel with D h = 150 J.Lm (Fig.


6.1A), the void fractions at low superficial liquid velocities are in the
same range for water and ethanol experiments (cwater ::::::::: Ceth ::::::::: 0.75).
However, going to higher superficial liquid velocities, the volumetric gas
content of the ethanol-nitrogen mixture nearly stays constant, whereas
the void fraction in the water-nitrogen experiments drops to about 50%
of the value at low liquid velocities.
For a larger channel cross-section (Fig. 6.1B. D h = 187.5 J.Lm), the void
fraction of the water-nitrogen flow drops continuously 3.', well a..<;; the
one of the ethanol-nitrogen flow. For the lowest volumetric liquid flow
rate (the same as used in the channel with smaller diameter), the two
configurations already have completely different void fractions (cwater :::::::::
0.9, Ceth ::::::::: 0.6). This requires lower actual gas velocities and higher
actual liquid velocities in the configuration using water. The latter is
shown for intermittent flow in chap. 7.2.
The least effect on the void fraction by the superficial liquid velocity can
be found in all the presented results using the highest volumetric gas
flow rate. This is a result of the occurring flow pattern, which is annular
flow for all the analyzed conditions. This regime shows fewer changes
in the volumetric gas content than changes in the fluid velocities when
changing the superficial velocities. For a microchannel with hydraulic
diameter D h = 187.5 J.Lm and water as working fluid, the change in
the flow pattern is explicitly seen in a change of volumetric gas content
for superficial gas velocities below 2.22 m/so For the latter superficial
gas velocity, annular flow is already present at lower superficial liquid
velocities.
In Fig. 6.2 the streamwise and temporally averaged void fraction is plot-
ted against the normalized channel width for two different hydraulic
channel diameters (D h = 150 J.Lm and D h = 187.5 J.Lm). For both chan-
nels, the void fraction was analyzed using the same constant volumetric
liquid flow rate (VL = 0.05 ml/min) and four different volumetric gas
flow rates. The flow pattern in the smaller channel is intermittent flow
for all the measured superficial gas flow rates. This explains the fact
of non-zero averaged void fraction values at the channel sidewalls. The
wall-normal void fraction distribution is similar for all gas flow rates. For
6.1 Void fraction results 109

1.0-r-...I-_...I-_-L--.....L._--&._-J;===::;1-

0.9 IOh = 150 ~lml A


.L 0.8

" 0.7
C'"
1Il
C 0.6
8
III 0.5
III
0'1
o 0.4
'1:
Q)
E 0.3
;:)

:g 0.2 -o-/-+- h-0.74m1s


-0-/ .... J,,-l.llm1s
__ 1 _ J,,-1.48m1s
0.1
..q-I.- JG - 3.70 mls
O.O-l--"""'T"-~-=r==:;:::=~---r---,--~
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40
Superficial liquid velocity jL [m/s]

0.9 B
I 0.8

'"
C " 0.7
AI
l5 0.6
o
IQ 0.5
0'1
.g 0.4

~;:)
0.3
(5 -0- I ia - 0.44 mls
> 0.2 -0- I JG - 0.67 mla
--l!r-- I JG 0.89 m/a
0.1
-SJ- I JG - 2.22 mla
0.0 -l-r--r---r--......-r-...........,.-::.:::;::;;:::;:::::;::;;:::;:::::;::;~-r--;r-r--r--;r-r--.-t-
0.00 0.04 0.08 0.12 0.16 0.20 0.24
Superficial liquid velocity i L [m/s]

Figure 6.1: Measured void fractions depending on the superficial


liquid velocity for a microchannel with hydraulic diameter Dh =
150 J.Lm (A) and Dh = 187.5 J.Lm (B). Open symbols stand for
water-nitrogen experiments, black symbols for measurements using
ethanol and nitrogen. The same volumetric flow rates were used for
both diameters, leading to different superficial velocities.

the larger channel diameter, all the flow regimes are either in the tran-
110 6. Void fraction

sition region from intermittent to annular flow or in the strictly annular


flow regime. The volumetric gas content at the channel sidewalls drops
to ~ O. The film thickness in the annular flow is seen to increase with in-
creasing superficial gas velocity. The parabolic void fraction distribution
becomes narrower.

100

90
A
eo
~ 70
","
c:
.2 eo
U
,g 50
"0
'g 40

13
..2
30
-0- iG 0.74 mlo. ",.0.67
20 ooQ- i G 1.11 mlo. ",0.69
...... iG 1.49 mlo, ",.0.72
10 ...... k. 3.70 mlo, ",0,$7
0
0.0 0.1 0.2 0.3 0.4 0.5 0,6 0.7 0.8 0.9 1.0
normalized channel position xlb [-]
100

90
B
90
~ 70
<:>
'" eo
<:

~ 50
Jg
"0
'0 40
;>
-<>-- jlJ 0.44 m/. s~ 0.89
~
:11:
30 &"

ooQ- JG 0.67 mls, '0 0.65


20 - jG 0.89 m/. 0, 0.80
..- i 2.22 mll, '0 0.53
10

0
0,0 0.1 0.2 0,3 0.4 0,5 0.6 0.7 0.8 0.9 1.0
normalized channel position xlb [-]

Figure 6.2: Local void fraction distribution versus normalized


channel width for channels with hydraulic diameter D h = 150 J.Lm
(A) and D h = 187.5 J.Lm (B). The volumetric liquid flow rate was
1IL = 0.05 ml/min, yielding superficial liquid velocities jL,150 =
0.04 m/s, and jL,187.5 = 0.02 m/so

Increasing the volumetric liquid flow rate to 1IL = 0.3 ml/min (yielding
6.2 Comparison with correlations 111

superficial liquid velocities jL,150 = 0.2 m/s, and jL,187.5 = 0.13 m/s,
respectively), the mean void fraction decreases (Fig. 6.3). In contrast to
the low superficial velocity in Fig. 6.2, the variation in the superficial
gas velocity yields differences in the local void fraction distribution due
to different plug lengths. For a hydraulic channel diameter of D h =
187.5 j.Lm, the same phenomena a9 for lower superficial liquid velocities
(jL = 0.02 m/s, Fig. 6.2B) can be observed. The film thickness at the
channel sidewalls increases with increasing jc.

6.2 Comparison with correlations


For the calculation of the void fraction, several empirical correlations
based on macro- and microfiuidic systems exist. The experimental data
of chap. 6.1 are compared to each of the following correlation methods.

6.2.1 Homogeneous void fraction model


In the homogeneous model, the void fraction is defined as

Xc . (2L
cC horn = . (.) = E c (6.3)
, Xc {2L + 1- Xc . {2L

The prediction of the volumetric gas content using the homogeneous


model is very inaccurate (Fig. 6.4). For all the analyzed experiments
the void fraction is overestimated and the standard deviation of the
experimentally determined void fraction is close to 50% (0" = 0.30).
In particular the void fractions using water are poorly predicted (0" =
0.36, 67%). The homogeneous model is absolutely not suitable for the
prediction of void fractions for volumetric gas transport fractions EC >
0.4 (see Fig. 6.13).

6.2.2 Volumetric gas content based on correlations


by Armand and Treschew
Armand and Treschew (1946) published a simple estimation for the void
112 6. Void fraction

100 -j---JL-----l-_...L-----l_-L_..L-----L_--l-_.L--+

80
~ 70
<:l
<.>
<: 80
.~
g 50
"0
"g 40

~
30
-0-- k.. 0.74 mlS. EO" 0.45
20 .-0- lo 1.11 mlS. EO" 0.5&
...... Jo .. 1.48 mls, Eo .. 0.57
10 ..... io " 3.70 mls, Eo" 0.63

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.6 0.9 1.0
normalized channel position x/b H
100

90

80
~ 70
<:l
'"
<: 80

j 60

~ 40

~ 30 -0--10 0.44 m/s. EO .. 0.45


-0- io 0.67 ""s. $r. .. 0.56
20 ...... 1" .. 0.89 m/s. '0 0.57
10
..... lo" 2.22 m/s, EO. 0.63

o....!IjpI~r--..,__-r-~___._____y-.---~~
O~ O~ ~2 O~ ~4 ~5 ~6 ~7 O~ o. ,~
normalized channel position x/b H

Figure 6.3: Local void fraction distribution versus normalized


channel width for channels with hydraulic diameter Dh = 150 J.tm
(A) and D h = 187.5 J.tm (B). The volumetric liquid flow rate was
VL = 0.3 ml/min, yielding superficial liquid velocities jL,150 =
0.22 m/s, and jL,187.5 = 0.13 m/so

fraction based only on the volume transport fraction eG, neglecting any
mass flow data.

CG,Ar = 0.83 eGo (6.4)

For eG < 0.9 this method is suitable for large diameter pipes. For large
6.2 Comparison with correlations 113

1.0

.....
...!..
0.9

00
.c
6 0.8 0
ll)
0.1
c::: 000
0
13 0.6 ,-
'0 0.5
"g 0.4
'0
Cb
0.3
U
'6
I!? 0.2
a..
0.1
1---+'-20%1
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.1 0.8 0.9 1.0
Measured void fraction Eexp [-]

Figure 6.4: Comparison of the volumetric gas content of the ex-


perimental microchannel data and the model predictions using the
homogeneous void fraction assumption (EO,hom = EO).

volume transport fractions Eo > 0.9, Massena (1960) proposed the fol-
lowing extension

EG,Ma = (0.83 + (1 - 0.83) xo) . Eo. (6.5)

Figure 6.5 clarifies that this correlation mainly overestimates the void
fraction for microf:l.uidic applications. All the predicted volumetric gas
contents for the channel with the smaller hydraulic diameter are larger
than the mea..",ured data. Especially for high superficial gas velocities,
this trend is confirmed. For D h = 150 J.lm the predictions for both fluids,
water and ethanol, do not differ significantly.
The experimental data correlate remarkably well with the predictions
using the microchannel with D h = 187.5 J.lm and ethanol as working
liquid.
114 6. Void fraction

...... 0.9
..!.. e "" 0
" c!1 ~ ..t.~~ Do 8 &@> "
J::
0
o C e elt:J d5'" 0 0 d; .........
De J.
.-- .... "'" c:..00 ...... 0'"
c:
:fi 0.6
eo.
~ ~... "0,,'"

0.5
'0 o Oh = 150 Ilm. ethanol
'g 0.4
Oh = 150 Ilm. water
-g 0.3 o Oh" 187.5 Ilm. ethanol
U e D." 187.5Ilm. water
'5 0.2
~ " " ,," ---+1-20%
a. 0.1

0.0 *-.,........--,-----.---.---.---r----r-.....,...-~_+
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Measured void fraction &exp [-]

Figure 6.5: Prediction quality of the method by Armand and


Treschew (1946).

6.2.3 Empirical model for the determination of void


fractions by Chawla

An additional model for the calculation of the void fraction, based on


large diameter two-phase flows, is presented by Chawla (1968). This
model is based on water - air and oil - air experiments. The volumetric
liquid content is suggested to be

(6.6)

where

K = e[ (1.1-0.1965.ln( ~)) .In(()+o.868-0.1335.ln( ~)] (6.7)

Here, , is the two-phase flow parameter introduced by Chawla (1968) in


Eq. 5.27. According to Eqs. 6.6 and 6.7, the volumetric gas content is
6.2 Comparison with correlations 115

defined as

(6.8)

The comparison between the experimental data and the predictions using
Eqs. 6.6 - 6.8 does show a large scatter. All the predicted void fractions
are in the range of 0.4 to 0.7, whereas the experimental data show gas
contents up to ~ 0.9. The root mean square of the deviation values is
0" = 0.14 (rv 24%). In contrast to the above illustrated correlation by
Armand and Treschew (1946), the current prediction method consid-
ers not only the gas volume transport fraction, Sa, and the gas mass
transport fraction, xa, but also the fluids' properties. The empirical pa-
rameters of this correlation model are based on water - air experiments
for large diameter pipes. The comparison with the experimental data of
our microchannels (ratio of hydraulic diameters Dh,expl Dh,model :=::::: 100)
shows satisfying agreement using water as working fluid with a stan-
dard deviation of 0" = 0.07 (13%). The agreement of the experimen-
tal data using ethanol with the prediction, on the other hand, is worse
(0" = 0.22, 33%). Especially the void fractions for the configuration
(D h = 187.5 J-Lm I ethanol), yielding the best results using the model by
Armand and Treschew (1946), are not useful for further considerations
(0" = 0.32, 40%).

For the friction pressure drop correlation by Chawla (1968), a parame-


ter variation for the implemented empirical values in the calculation of
the two-phase flow parameter ( (Eqs. 5.27 - 5.32) resulted in a remark-
able improvement of the agreement with the experimentally measured
pressure drops. Using the same set of parameters (see Tab. 5.2), all the
predicted void fractions are in the range 0.2 < ca,model < 0.35. The
standard deviation increases to 0" = 0.33 (56%). This result confirms
the complexity of adapting large scale correlations to microfluidic appli-
cation by parameter variation.

6.2.4 Void fractions in microchannels by Serizawa

Serizawa et al. (2002) investigated the void fractions in two-phase flow


116 6. Void fraction

o Oh =150 "m, ethanol


..... 0.9 Oh = 150 "m, water
...!,..
i 0.8
o

Oh = 187.5 "m, ethanol
Oh = 187.5 "m. water
CJ:IB 0.7
- _. +1-20%
c:
o 0.8
13
~ 0.5 o
"0
'g 0.4
"0
Q) 0.3
1:)
'6
~ 0.2
a.. 0.1

0.0 -F--,---,---r--r---,---,-----,----,---r--+
0.0 0.1 0.2 0.3 004 0.5 0.6 0.7 0.8 0.9 1.0
Measured void fraction &exp [-]

Figure 6.6: The void fraction correlation by Chawla (1968) using


the originally proposed empirical parameters.

1.0 +--'---'----'----'----1..---'---'""'---'-.....,.....--'---"'71-
.....
...!,.. 0.9 o D. =150 "m, ethanol
Oh =150 "m, water
~I 0.8 o Oh'" 187.5 "m, ethanol

I
(,)
0.7
O. = 187.5 "m, water

- - - + 1- 20%
,,
CJ:I
c: 0.6
o
13 0.5

"0 004
'0
>
"0 0.3

~
'6 0.2
~
a.. 0.1

0.0 -F-...,...--,---,---r----r----r---,-----,----,-__+_
0.0 0.1 0.2 0.3 0.4 0.5 0.8 0.7 0.8 0.9 1.0
Measured void fraction &exp [-]

Figure 6.7: The void fraction correlation by Chawla (1968) using


parameters obtained by the parameter variation to fit the pressure
drop correlation to the experimental data (Tab. 5.2).
6.2 Comparison with correlations 117

microchannels. Depending on the volume transport fraction, different


correlations for the void fraction are proposed:

EG < EG,A EG =EG

EG,B < EG < 0.6 EG = 0.833 EG (6.9)


EG,B < EG, 0.6 < EG < 0.95 EG = 0.69 . EG + 0.0858
EG,B < EG, 0.95 < EG EG = 0.83 log(l - EG) + 0.633

The two values of EG,A and EG,B are diameter dependent.


The model by Serizawa et al. (2002) is based on experimental data using
silica and quartz capillary tubes with inner diameters between 10 and
100 ).Lm. In contrast to the previous models (Armand and Theschew,
1946; Chawla, 1968), the hydraulic diameters of the experiment and
the prediction are in the same order of magnitude. Nevertheless, the
predictions are not significantly more accurate than using models for
large diameter pipes (0' = 0.17, 27%). Serizawa et al. (2002) does
not take into consideration any information of the fluid properties or
channel geometries. The same void fraction is predicted for a specific
flow rate configuration for both fluids, water and ethanol, regardless of
the hydraulic channel diameter, whereas the experimental results show
different volumetric gas contents.
Even though the correlation by Serizawa et al. (2002) is based on air-
water experiments, the prediction does not show a better agreement with
our experimental nitrogen-water results than with the nitrogen-ethanol
void fractions (O'water = 0.17, O'ethanol = 0.15). Similarly, a difference
in the prediction quality is not apparent for the two different hydraulic
diameters (0"150 = 0.16, 0'187.5 = 0.16).
118 6. Void fraction

"7" 0.9
ca
~ 0.8

~ 0.7 o

.0.
w
5 0.6
ts
~ 0.5
'0
'0 0.4 o On = 150 ",rn, ethanol
> =
On 150 ",rn, water
'0 0.3 "" o =
On 187.5 I-lrn, ethanol
~ =
On 187.5 ",rn, water
'6 0.2
~ ---+/-20%
a. 0.1

0.0 --I"'=--~---r----.---.---r------,r----r---r---.----+-

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Measured void fraction &exp [-]

Figure 6.8: Comparison of the void fractions predicted by the cor-


relation by Serizawa et al. (2002) and the experimentally determined
gas contents.

6.2.5 Correlation by Triplett

An empirical void fraction correlation based on experimental results is


proposed by Triplett et al. (1999a). The void fraction is defined as

1
EG,Tripplet = ( ) q ( ) r
(6.10)
Ae~X)p ~ ~ +1
where the empirical parameters A, P, q, and r are supposed to be set to
the values in Tab. 6.1 (version 1).
The correlation by Triplett et al. (1999a) is based on experimental data
using air and water in circular microchannels with 1.1 and 1.45 mm
inner diameter and in microchannels with semi-triangular (triangular
with one corner smoothed) cross-sections with hydraulic diameters 1.09
and 1.49 mm. The superficial gas and liquid velocities were of the same
magnitude as in the present experiments. The four empirical parameters
6.2 Comparison with correlations 119

enable the correlation to be adapted to any geometrical and flow con-


dition by means of a parameter variation. Using the originally proposed
parameters, the agreement of the predicted void fractions is the best of
all the investigated models. The standard deviation is U = 0.13, 23%.
Nonetheless, Fig. 6.9 shows large deviations for some of the experimen-
tal data. Especially the volumetric gas content for ethanol-air flows is
poorly predicted (uwater = 0.12, Uethanol = 0.15), as the model shows
no trend in over- or underestimating the present void fraction. As water-
air systems were used in the experiments by Triplett et al. (1999a), the
better agreement with our water-nitrogen flow can be explained. Taking
into account only the experimental data of the water-nitrogen flows, a
parameter variation for the empirical parameters in Eq. 6.10 (version 2
in Tab. 6.1) can minimize the standard deviation to Uwater,mod = 0.06
(Fig. 6.10), whereas a parameter set defined for the ethanol-nitrogen
experiments (version 3 in Tab. 6.1) does not remarkably improve the
corresponding results (Uethanol,mod = 0.15). The poor agreement of the
ethanol data causes also the insignificant improvement of the confor-
mance using a global parameter variation (Umod = 0.11, version 4 in
Tab. 6.1).

Version 1 (Triplett) 2 3 4
A 0.28 0.32 0.01 0.09
p 0.64 0.39 0.42 0.50
q 0.36 0.37 0.31 0.01
r 0.07 0.41 0.90 0.01

Table 6.1: Parameter settings for the void fraction correlation by


Triplett et al. (1999a) (version 1), and for the best prediction of the
present experimental data. Version 2 takes into account only water-
nitrogen experiments, version 3 is the fit for ethanol-nitrogen flows,
and version 4 for the complete data set.

The differences in the parameters originally proposed by Triplett et al.


(1999a) (version 1) and those of version 2 (water - nitrogen only) are rela-
tively small. Larger deviations can be found only for the two parameters
p and r, determining the influence of the mass flux and the influence
120 6. Void fraction

of the fluids' viscosities. In the present study, the impact of the fluid
superficial velocities is increased in favor of the viscosity difference. A
possible explanation is the fact, that 'friplett et al. (1999a) used a much
larger range of superficial velocities (Triplctt: 0.02 m/s < jG < 80 m/s
and 0.02 m/s < jL < 8 m/si present study: 0.44 m/s < jG < 3.70 m/s
and 0.02 m/s < jL < 0.37 m/s). No similarity can be found for the pa-
rameter settings for ethanol experiments only, and the settings for all
the experimental data.
1.0

0.9

..
~

...!..
t 0.8
.0. ~~
~.
0
~
...
~
0 0
'1:
.... ~~ .~)-'i
.D'0
09~

~ 0.7
~~c. o~~~o 0
l::
0 0.8
0 ~
~
0
n .g~. ~..,
~ ~ ~
~

0
~
0
0
~ 0.5 ~
~
~

<0
0
'0
.~ 0.4
~
~
~

0
'0 0 Oh " 150 ",rn, ethanol
~
0.3 Oh =150 ",rn. waler
'is 0.2
0 Oh =187.5 ",rn, ethanol
~
a.
0.1
Oh =187.5 ",rn, water

0.0 ~r----r-....------,r-----r--""::::::;~::;::=::;::::=-+
0,0 0.1 0.2 0.3 0.4 0.5 0.8 0.7 0.8 0.9 1.0
Measured void fraction Eexp [-]

Figure 6.9: The void fraction correlation by Triplett et al. (1999a)


using the originally proposed parameter setting.

Therefore, the void fraction correlation by Triplett et al. (1999a) is shown


to be suitable for the prediction of the volumetric gas content using water
- nitrogen or water - air as working fluids. If fluids differing from the
standard combination are involved, this method can not be applied and
the void fraction has to be determined experimentally.

6.2.6 Void fraction model by Coleman

A further void fraction model for the estimation of the volumetric gas
content in two-phase flow in microchannels is presented by Coleman and
6.2 Comparison with correlations 121

, ,,
,,
...!.. 0.9
~;1:1 0.8
. ...
. , ~
,,
,

10.7
'C
l-
t.)
C
o
0.6 .
. .'.. ....,.,;

,;
,'
;11'.

.,. .,'
,'.
is 0.5
,.... !!I...... "
't:I 0.4 , III '
'0
>
't:I 0.3
(\) Oh = 150 J.lm
n 0.2 Oh" 187.5 J.lm
'0
!
a. 0.1 ~ ~ - +/-20%

0.0 -fC--..------r---r--r---r-----,r------r---r--~__+_

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Measured void fraction &exp [-I

Figure 6.10: The void fractions of water - nitrogen flow predicted


using parameter setting version 2.

Krause (2004). This correlation, based on Huq (1998), does not make
use of any empirical parameter.

2 (1 - XC)2
cc, Col = 1- ---------'------'--------.",..0.""'"5. (6.11)
1 - 2 Xc + [1 + 4 Xc (1 - xc) (~ - 1) ]

The model shows agreement in the prediction of the void fraction within
30% (0- = 0.18). The void fraction is generally overestimated. Good
agreement with the experimentally determined gas contents are detected
only for the ethanol-nitrogen configuration in the microchannel with
D h = 187.5 J.1m (0- = 0.08, 10%).

6.2.7 Void fraction model by Kawahara

Based on experimental data, Kawahara et al. (2002) present a method


for the calculation of the time-averaged void fraction of gas-liquid flows
122 6. Void fraction

::I: 0.9

~ 0.8 l: ~ 0

~ o l:
(J 0.7
lol

.0.
C
o 0.6 De
13
~ 0.5
"0
.~ 0.4
o Oh = 150 llm, ethanol
"0 0.3 Oh =150 llm, water
~ o Oh =187.5 llm. ethanol
'0 0.2 Oh =187.5 llm, water
~
a. 0.1 +/-20%

0.0 ~--r------'----r"--r--......----,.--------r---r--.,...--+
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Measured void fraction &BJ(p [-]

Figure 6.11: Comparison of the measured void fraction and those


predicted using the correlation by Coleman and Garimella (1999).

in round microchannels.

0.03E~l
6'G,Ka = 05' (6.12)
1- 0.97Ed

Kawahara et al. (2002) used deionized water and nitrogen as working


fluids flowing in 100 p,m circular tubes with superficial velocities jG =
0.1 - 60 m/s and jL = 0.02 - 4 m/so As the experimental data was
analyzed using the records of the gas-liquid interfaces in the observation
window, which covered a distance of approximately 1 mm in the flow
direction, the images showed either liquid flowing alone (6' = 0) or a
gas core flow with smooth-thin or ring-shaped liquid film. For the latter
flow pattern, Kawahara et al. (2002) assumed the void fraction to be
unity. By counting the number of images containing each flow type, the
time-averaged void fraction was determined (Eq. 6.1).
Even though the experimental conditions of Kawahara et al. (2002) are
similar to those used in the present study, the predictions of the void
fraction using Eq. 6.12 are unacceptable. Only the homogeneous void
6.2 Comparison with correlations 123

fraction prediction method delivers. less accurate agreement with the


experimentally determined data (0' = 0.22, 37%). In contrast to all the
correlations presented above, most of the volumetric gas contents are
underestimated. It can be shown, that this discrepancy is not caused by
the use of other liquids (uwater = 0.17, Uethanol = 0.26) or by the
difference in hydraulic diameter (0'150 = 0.18, 0'187.5 = 0.26), where
none of the comparisons show acceptable agreement. Independent of the
used liquid or channel diameter, the void fraction of experiments with
high superficial gas to liquid velocity ratio, which causes high volumetric
gas contents, show the best agreement.

::c 0.9
o Oh =150 llrn, ethanol
Oh =150 llrn, water , " "
l!
o =187.5 llrn, ethanol
.
.l! 0.8 Oh
=187.5 "rn, water
!
Oh
,,'DiI
~

a
13

'0
0.7

0.6

0.5
- - - +/-20%
6
...
......"
"
"
0
"
0"

0
0
00
0
0

'0 0.4
o ". c. 0 0 0
>
'0 0.3 o 0
Q)
1:) .::J." 0 0
'5 0.2 .~: iOO
~ o
a.. 0.1

0.0 --toC--....---......---r-----r----,--.,...--....---r------.-~

QO Q1 Q2 Q3 Q4 Q5 Q6 Q7 Q8 Q9 1.0
Measured void fraction &exp [-I

Figure 6.12: Void fraction prediction using the method by Kawa-


hara et al. (2002).

The flow pattern map by Kawahara et al. (2002) (Fig. 4.7) clarifies that
in the experiments only intermittent flow regimes were detected, where
in our rectangular microchannels annular flow is observed. The annular
flow pattern features higher volumetric gas contents than intermittent
flow. The flow pattern (and therefore also the agreement of the void
fraction with the correlation by Kawahara et al. (2002)) seems to be more
dependent on the channel cross-section than on the hydraulic diameter.
124 6. Void fraction

6.2.8 Comparison of the different models

All the presented models try to fit experimental results by adjusting


the volumetric gas transport fraction EG' which is usually known, using
empirical parameters, properties (or their ratios) of the involved fluids
(densities, viscosities, velocities), and geometrical parameters as the hy-
draulic channel diameter or the surface roughness. Therefore most of
the correlations are directly or indirectly a function of the volumetric
gas transport fraction EG. In Fig. 6.13, cG,model = f(iG) is plotted for
the above discussed models.

1.0

0.9 - Homogeneous
..... Armand
......
.2... 8erizawa
0.8 _ Kawahara
Cl

-.!l
lA)

c
0.7

0.6
_ ... Coleman
- Triplett (original)
Triplett (Water)
o Experiment
8
fI) 0.5
CO
0)

-
0
.;:: 0.4
(J) ,
E 0.3
,
::::I
(5 ,
, ,
0.2 o
> .. ~.
0.1 ----_.--
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 O.~ 0.8 0.9 1.0
Volumetric transport fraction I; G [-]

Figure 6.13: Comparison of the several different void fraction pre-


diction models. Plotted are the calculated void fractions CG in de-
pendence of the volumetric gas transport fraction iG and the ex-
perimental data points for water-nitrogen flow.

The trend of overestimating the void fraction can be seen for the ho-
mogeneous model, and the theories by Armand and Treschew (1946);
Serizawa et al. (2002); Kawahara et al. (2002); Coleman and Garimella
(1999) and Triplett et al. (1999a). Especially for high volumetric gas
6.3 Void fraction measurements using XTM 125

transport fractions the models for large diameter pipes (homogeneous


model and Armand and n.eschew (1946)) show insufficient agreement
with the experimental data. Serizawa et al. (2002) adapted their model
by means of case differentiation to minimize the overestimation of the
void fraction for high transport fractions. The other models use non-
linear equations to calculate the void fraction from the volumetric gas
transport fraction. In contrast to Coleman and Garimella (1999), the
models by Triplett et al. (1999a) and Kawahara et al. (2002) use empiri-
cal parameters that can be adjusted to the existing problem while loosing
universal validity. In Fig. 6.13 two versions of the model by Triplett et al.
(1999a) are depicted, the one using the originally proposed parameter
setting and version 2 (see Tab. 6.1). The latter features higher accu-
racy for high volumetric gas transport fractions, however shows lower
agreement for applications at low volumetric gas transport fractions.


6.3 Void fraction measurements uSIng
XTM
X-ray Tomographic Microscopy (XTM) is not only a suitable technique
to visualize the flow regime in microchannels (see chap. 4.3), but can
be used also to investigate the fluid distribution and therefore the void
fraction. However one has to consider that results obtained by XTM
measurements are temporally averaged.
With the help of standard image processing steps (image threshold, edge
detection, area sizing) the area taken by the liquid and the one taken
by the gas can be determined in the 2-dimensional cross-sectional XTM-
images of the two-phase flow (Fig. 6.14). The volumetric gas content is
then defined as

Area of gas ~pixels gas


cG XTM =
, Total area of channel
= Total ~ pixels
(6.13)

As seen in chap. 4.3, XTM detects phase interfaces. Images of tempo-


rally and spatially stationary flows show precise fluid boundaries. The
126 6. Void fraction

fluid distribution as well as the void fraction determination are therefore


expected to be accurate. However, wavy interfaces are represented by
XTM as blurry boundary lines. Using the same image processing algo-
rithms, the obtained value of the volumetric gas content may differ from
the actual value significantly.

Figure 6.14: Void fraction determination using X-ray Tomographic


Microscopy (XTM). The areas taken by the liquid and the gas are
dptermined by standard image processing algorithms.

Table 6.2 shows the volumetric gas content as identified by X-ray To-
mographic Microscopy (XTM), and as determined using the method de-
scribed in chap. 6.1. The XTM void fractions appear to be significantly
higher than the experimental results of the optically derived void frac-
tion. The best results are obtained using ethanol as working fluid. The
hydraulic channel diameter does not seem to have much impact on the
consistency of the two measurement methods. Major differpnces between
ethanol and water are the surface tension (creth/crwater = 0.31), and less
significantly the density ({2eth/ (2water = 0.79), whereas the viscosity is
of the same magnitude (rJeth/rJwatcr = 1.19). Known from the analy-
6.4 Summary and conclusions 127

HiH of the flow pattern experiments, the lower surface tension of ethanol
causes temporally and spatially stable fluid interfaces in the annular flow
regime. This leads to sharper boundary lines in the XTM reconstruction
and therefore, thc reconstructed area assigned to the gas is more ac-
curate than for wavy surfaces. Fig. 6.15 illustrates this phenomena for
th(~ two configurations for ethanol, D h = 187.5 Mm, jL 0.04 rn/s,
.hi = 5.33 m/s (top), and for water, D h = 200 pm, jL = 0.04 m/s,
jc; = 2.97 m/s (bottom).

Dh Liquid .h)jc; EC;,XTM EC;,opt ~E


[pm] [m/s] [-] [- ] [%]
187.5 Ethanol 0.04/3.56 0.65 0.45 -31
187.5 Ethanol 0.04/5.33 0.67 0.45 -33
200 Ethanol 0.02/1.85 0.67 0.48 -28
200 Water 0.02/1.85 0.69 0.40 -42
200 Ethanol 0.04/2.96 0.67 0.43 -36
200 Water 0.04/2.96 0.82 0.32 -61

Table 6.2: Comparison of the obtained values for the void fraction
using the two different methods XTM and optical determination.

6.4 Summary and conclusions

6.4.1 Void fraction measurements uSIng


. Laser-
Induced Fluorescence (LIF)

In contrast to the flow regime and the pressurP drop, which arc both
widely investigated for large scale applications, comparatively little in-
formation on the prediction of the void fraction can be found in the
literature. Thinking of the importance of knowing the void fraction, de-
termining the reHidence time or its infi1Hmcf~ on mass and heat transfer
processes, accurate measurements of the void fraction are desirable. We
present experimental results of void fraction measurements using Laser-
128 6. Void fraction

Figure 6.15: Explanation for the large differences in the void frac-
tion measurements between XTM and the optical measurements.
The more the phase interface is temporally and spatially stable, the
more distinct is the reproduced phase boundary in the XTM recon-
struction, and therefore a more accuntte void fraction is defined.

Induced Fluorescence (LIF). The local volumetric liquid content is di-


rectly related to the emitted fluorescent intensity of the Rhodamine B
added to the liquid phase. A calibration procedure eliminates undesired
background noise. Even tough assumptions, such as the equal contribu-
tion to the total intensity independent of the channel depth, were made,
the uncertainty in the measurements are determined to be in the range
of ~voidfra.ction i:::.:::: 5% (laser intensity fluctuations, chap. :3.4.:3).
The experimental data demonstrates the influence of the channel hy-
draulic diameter, the fluid properties, and the How conditions on the
m{~asured void fraction. It can be seen, that the void fraction in small
cha,nnels at low volumetric liquid flow rates is nearly independent of the
fluid properties (surface tension). For annula.r flow patterns, changes in
the superficial gas velocity l(:md to insignificant differences in the gas
6.4 Summary and conclusions 129

holdup. Therefore, the holdup is mainly a function of the superficialliq-


uid velocity. For larger hydraulic channel diameters, the surface tension
of the fluids does affect the flow pattern occurring, and thus also the
void fraction.
The experimental data is compared to several prediction models found
in the literature. Common ground of most of the models is an adapta-
tion of the homogeneous void fraction approach, where the volumetric
gas content, CG, is supposed to be equal to the volumetric transport gas
fraction, EG. None of the prediction models is capable to predict the
void fraction with sufficient accuracy for a wide range of flow configu-
rations and fluid combinations. A parameter variation for the empirical
parameters in the model suggested by Chawla (1968), using the same
settings as in the pressure drop model, yielded void fraction values in
the range of cG = 0.2 ... 0.35 with a standard deviation of (J' = 0.33.
This results confirms the complexity of adapting large scale correlations
to microfluidic applications by parameter variation.

6.4.2 Determination of the void fraction by X-Ray


Tomographic Microscopy (XTM)

X-Ray Tomographic Microscopy (XTM) is not only a suitable method


to analyze the flow pattern in microchannels, but can be used for a
rough estimation of the void fraction (for annular flow patterns) with
the help of standard image processing algorithms. In a 2-dimensional
cross-sectional image perpendicular to the flow direction, the phase dis-
tribution can be extracted. By simply comparing the number of pixels
representing the gas phase to the total number of pixels, the volumetric
gas content is determined. As the XTM images are temporally averaged,
unsteady phase interfaces lead to blurry boundary lines, which can lead
to misinterpretation in the extraction of the gas phase area. Therefore,
the holdup obtained by XTM measurements is generally higher than the
effective volumetric gas content. The accuracy of the XTM holdup mea-
surements is determined to be ~voidfraction,XTM ~ 30% (chap. 3.4.3).
130

Chapter 7

Liquid velocity
measurements

7.1 PIV analysis for the liquid recirculation


motion in intermittent flow
7.1.1 Velocity measurement
All of the presented PIV experiments are based on the need to characteri-
ze the recirculation motion and mixing quality inside the liquid segment
(plug) of an intermittent gas-liquid two-phase flow. The velocity distri-
bution is an important factor of the internal mixing (recirculation) in
chemical microreactors. The recirculation determines the mass transfer
towards the phase interface and towards a catalyst surface, and therefore
influences the reaction rate of a chemical gas-liquid reaction. The mixing
quality is mainly dependent on the fluid velocities, their properties, and
the channel geometry, which includes not only the reactor cross-sectional
shape and hydraulic diameter, but also the surface roughness.
The absolute velocities in the liquid plug are mainly dependent on the
volumetric flow rates of the gas and liquid phase, VG and VL , respectively.
7.1 PIV analysis for the liquid recirculation motion in intermittent flow 131

At first sight, the velocities of all the analyzed flow conditions show the
parabolic distribution, typical for strictly laminar flows. More interesting
is the internal recirculation motion, that is the relative velocities

Vrel = Vabs - Vbulk (7.1)

which are responsible for the mass transport from the bulk towards the
phase boundaries and the channel walls.
With the help of calculated streamlines of the relative velocities in the
liquid plug, the determination of a mass transfer over the center line of
the microchannel can be analyzed. The transverse fluid exchange can be
crucial for mixing, if a fluid component is injected into complex microre-
actor designs from one side of the channel only. This fluid exchange is
favored by stronger internal recirculation motion. The latter is, amongst
others, represented by high wall normal velocity components and the
strength of the occurring vortices.

7.1.2 Vortex identification

Vorticity analysis is generally used to identify locations of vortices, which


indicate internal recirculation motion, and to calculate vortex statistics
(size, strength, etc.). As described above, this is an important require-
ment in microchemical applications. Several possibilities to identify and
characterize vortices are known in fluid dynamics, such as the vorticity
or the swirling strength. The vorticity, w, is defined as

w=\lxl1. (7.2)

Unfortunately, vorticity not only identifies vortex cores but also any
shearing motion present in the flow. These shearing motions can render
the task of computing reliable vortex statistics virtually impossible.
New methods of extracting underlying structures from velocity fields
which involve critical-point analysis of the local velocity gradient tensor
132 7. Liquid velocity measurements

and its corresponding eigenvalues are proposed (Chong et al., 1990; Dall-
man et al., 1991; Zhou et al., 1999, 1996). In three dimensions, the local
velocity gradient tensor will have one real eigenvalue (A r ) and a pair of
complex conjugate eigenvalues (A cr iAci), when the discriminant of its
characteristic equation is positive. When this is true, the particle tra-
jectories about the eigenvector corresponding to Ar exhibit a swirling,
spiral motion (Chong et al., 1990). A~/ represents the period required
for a particle to swirl once about the Ar-axis. If the flow is pure shear
flow, the particle orbits are infinitely-long ellipses and the orbit period
is also infinite, corresponding to Aci = O. Thus, Aci > 0 corresponds to
shorter, more circular ellipses, i.e. eddies. Zhou et al. (1999) show, that
the strength of any local swirling motion is quantified by Aci, which they
define as the swirling strength of the vortex. Vortex identification based
on swirling strength does not reveal regions which contain significant
vorticity but do not have of any local swirling motion (i.e. shear layers).
Since PIV fields are usually two-dimensional, the full local velocity gra-
dient tensor cannot be formed. However, an equivalent two-dimensional
form of this tensor can be computed in the plane in which the PIV data
are recorded; namely,

(7.3)

where (Xl, X2) and (UI' U2) are the streamwise and wall-normal direc-
tions and velocities, respectively. In this case, D 2-D will either have two
real eigenvalues or a pair of complex conjugate eigenvalues. Therefore,
vortices are easily identified by plotting iso-regions of Aci > O.
Fig. 7.1 shows the sum of all considered raw digital PIV images (A) of
a two-phase flow in a Dh = 150 jjm channel, and the according rela-
tive velocity distribution (B). The streamwise (C) and wall normal (D)
velocity component contour plot indicates two stagnation points on the
channel center line close to the bubbles' trailing and leading end. Note
that in all the velocity plots, the bulk velocity, Vhulk, was subtracted.
The corresponding swirling strength contour plot is shown in (E), and
7.1 PIV analysis for the liquid recirculation motion in intermittent flow 133

the contours of the vorticity in (F). Obviously the swirling strength in-
dicates the position of the two counter-rotating vortices whereas the
vorticity is dominated by the shearing motions.
For the analysis of the recirculation motion, for the determination of the
mass transfer over the center line, and the vortices (strength and posi-
tions) in the upcoming section, we therefore concentrate on the swirling
strength instead of the vorticity.

7.1.3 Proper orthogonal decomposition (POD) anal-


.
YSIS

Proper orthogonal decomposition (POD) is a statistical pattern analysis


technique which captures the dominant spatial flow structures in a data
ensemble ("coherent structures"). The POD has been introduced to the
turbulence community by Lumley (1967). Before that time it was al-
ready known in statistics as the Karhunen-Loeve expansion. The proper
orthogonal decomposition is an appropriate tool to decompose the quan-
titative flow field into a series of eigenmodes which represent the degree
of kinetic energy associated with each flow structure. For a detailed de-
scription of the theoretical basics on POD we refer to Lumley (1967);
Sirovich (1987).
We perform a POD on an ensemble of M = 256 realizations of water-
nitrogen flow (jL = 0.04 m/s, ja = 0.37 m/s) in a channel with hydraulic
diameter Dh = 200 p,m. Only the first 2 eigenmodes, containing 75% of
the kinetic energy, were considered. As higher order eigenmodes rep-
resent flow motion contributing lesser energy, they can be regarded as
noise, which is filtered out using only the dominant eigenmodes. The
velocity distribution is reconstructed from these 2 eigenmodes and com-
pared to the ensemble average of the total 256 realizations (Fig. 7.2. It
can be shown, that POD is a suitable method to filter out measurement
uncertainties as random noise. Due to the strictly laminar flow in mi-
crochannels, the introduced noise in the measurement data is small and
the advantages of a POD is marginal.
134 7. Liquid velocity measurements

Flow 50llm
----- OAm/S

A_ _

.... _---_ .... _..... ---_ ... _"' ... ""


~~~~~--------~,'~~~,
'\

.... ~ ,,; F .. _ - " " "" ""'- "I-


_
_~.~.~ , , ,.~J~~I
'I" '\ .. , ~

Sum of digital images Velocity vectors


(bulk velocity subtracted)

005 [m/s] -0.05

c o

Velocity component in flow direction Wall normal velocity component


(bulk velocity subtracted) (bulk velocity subtracted)

2.0 (-] 00 0.75 -0.75

E F

Swirling strength Vorticity

Figure 7.1: Average of instantaneous flow images of a two-phase


intermittent flow in a D h = 150 J-Lm channel (A) and field of rel-
ative velocities (B), the contour plots of the streamwise (C) and
wall-normal relative velocities (D), the according swirling strength
contour plot (E), and the contours of the vorticity (F).
7.1 PIV analysis for the liquid recirculation motion in intermittent flow 135

Ensemble average POD


(256 realizations) (2 modes)

[1/s2]

Figure 7.2: Comparison of an ensemble averaged velocity field (rel-


ative velocities) of 256 realizations (left) and the POD reconstruc-
tion using the first 2 eigenmodes containing 75% of the kinetic en-
ergy of the flow.

7.1.4 Average correlation method

Since many low Reynolds number flows in microfluidic systems are lami-
nar, it is not necessary to determine instantaneous velocity information.
It is sufficient to measure only time-averaged or phase-averaged (peri-
odic flows) velocity fields. Average velocity fields can be obtained by
first measuring instantaneous velocities and then averaging them in ei-
ther space or time. A alternative method of estimating average velocity
fields directly from the correlation function is the average correlation
method (Meinhart et al., 2000).
Estimation of velocity-vector fields using PIV involves three primary
136 7. Liquid velocity mea..'iurements

steps:

1. Particle Image Acquisition


2. Particle Image Correlation
3. Correlation Peak Detection

In order to obtain an average velocity measurement, one must apply an


averaging operation. The averaging operator is a linear operator, and
can be applied after step (1), step (2), or step (3).The particle image
correlation and peak detection operations are both non-linear, and the
order in which the averaging operator is applied can change the quality
of the resulting signal.
In contrast to the commonly used average velocity method, the corre-
lation function is averaged in the average correlation method, and then
the location of the signal peak location is detected (Fig. 7.4). The in-
stantaneous correlation functions contain significant amounts of noise
that can lead to inaccurate or unreliable measurements. This noise is
substantially less in the average correlation function. Since the average
operator is applied to the correlation function before peak detection, the
probability of erroneous measurements is reduced.
We present a comparison of the two averaging methods (Fig. 7.3) for an
ensemble of 10 instantaneous recordings of a two-phase water-nitrogen
flow in a Dh = 200 /-Lm microchannel. Fig. 7.5 shows the two averaged
velocity fields. As seen in the previous chapter for the filtering using
proper orthogonal decomposition, the introduced noise in the measure-
ment data is small due to the strictly la,minar flow in microchannels and
the advantages of the average correlation method is disproportional to
the increa.<;;cd effort.

7.2 Liquid velocities in straight rectangular


microchannels
We experimentally determined the liquid velocity distribution in inter-
mittent flow in straight and meandering microchannels of rectangular
7.2 Liquid velocities in straight rectangular microchannels 137

Image Image A Image 8 Correlation Peak


Sequence R AB Detection
A
1
1
A1
11
81
1
--.. I RA,B'
1
--.. /
+
2
I A2
11
82
I --.. I R A2B2
I --.. -----"
+
3
I A3
11
83
1
--.. 1
RA3B3
1
--.. /+

+
N
1
AN
11
8N I --.. I R ANBN I --.. /
/
Image Image A Image 8 Correlation
Sequence R AB
Peak
Detection
B
1
I A1
11
81
I --.. I RA,B,

+
I
2
I A2
11
82
I --.. 1
R A2B2

+
I
3
I A3
11
83
I
--.. 1
R A3B3
I
+
+
N
1
AN
11 8N
1--"1 R ANBN
1

1
RAB
1
--.. /
Figure 7.3: Diagrams depicting the different ways in which the
average velocity can be estimated: (A) average velocity method,
(B) average correlation method.

cross-section with low superficial liquid velocities. Three different liquids


were considered, deionized water, ethanol, and a 10% aqueous glycerol
mixture. Nitrogen was used as the gas phase in all cases. The properties
of the fluids can be found in Tab. :3.1.
The internal recirculation was analyzed for different flow conditions and
hydraulic channel diameters. Using a liquid flow rate of VL = 0.1 ml/min,
138 7. Liquid velocity measurements

.+
+
Figure 7.4: Instantaneous cross correlation functions that are aver-
aged together to produce an average correlation function (Meinhart
et al., 2(00). The average correlation function has a much higher
signal-to-noise ratio than the instantaneous correlation functions.

Average velocity method

".-'" ,F ~ , ' _ ~ __.. _ . .-'" ~ ~ _ . 7 ",,_ +,," "" _,., ~.. _ ,~

Figure 7.5: Relative velocity vector fields and streamline plots


showing the results from the two different methods of calcula-
tion: average velocity method (left) and average correlation method
(right).
7.2 Liquid velocities in straight rectangular rnicrochannels 139

a gas flow rate of Va = 1.0 ml/min, and a microreactor with a hydraulic


diameter of D h = 210 /-Llll, resulted in a superficial liquid velocity of jL =
0.032 m/s and a superficial gas velocity of ja = 0.317 m/so According
to the flow pattern maps presented in chapter 4.1, intermittent flow is
apparent. Using water as liquid: these flow conditions are represented by
Reynolds numbers of

. - jL' D h = 7.45, .. _ ja D h
R,CL - R,Ca - = 4.31 (7.4)
VL va
Figure 7.6A shows the velocity distribution inside the liquid water plug
of the intermittent flow. The average streamwiHe liquid velocity (bulk
velocity) is 0.36 m/so To investigate the recirculation motion and the
mass transport towards the phase interphases, the relative velocities are
plotted in Fig. 7.6B (vrel = Vabs - Vbulk), where Vabs are the measured
absolute velocities and Vbulk is the mean velocity. Two counter-rotating
vortices are observable. In the center of the microchannel, low stream-
wise velocities are present (Fig. 7.6C), and hardly any mass transfer
over the center line is detectable. The velocity distribution is symmetric.
Two distinct vortex cores are illustrated using the swirling strength (Fig.
7.6E).
Ethanol has a similar viscosity, but a much lower surface tension than
water (see Tab. a.l). Using the same flow conditions as in the water case
(jL = 0.032 m/s and ja = 0.317 m/s), the liquid Reynolds number drops
to RCL = 4.94. The average streamwise velocity inside the plug increases
to Vbulk = 0.38 m/so The ethanol segment length is slightly smaller than
the water plug. The absolute velocity distribution (Fig. 7.7A) shows sim-
ilar characteristics as the one using water, whereas the relative velocities
(Fig. 7.7B) behave remarkably different. Comparatively large relative ve-
locities are measured throughout the liquid plug and especially along the
center line of the channel. The typical vortices are more distinctive, the
value of the swirling strength is slightly increased compared to the pre-
vious experiment using water as liquid phase (Aci,water,max = 1.93 l/s 2 ,
2
Aci,cth,max = 2.14 l/s ). The area of swirling in ethanol is much more
spatially concentrated, showing large gradients. No stagnation points in
the liquid phase close to the leading and the trailing end of the gas phase,
as seen in the water case, are found.
140 7. Liquid velocity measurements

-1.0 m/s .. 0.2 m/s

0.55

[m/s]

.............
-~_

J _ ~_ _
..
...

-~"
-~., 1,1"'
I""'-~~"""""""""'''''''''''''
- - ...

-',,,. .. "''''
~ ". .. W'~
-_ -
--_ ...
...
o - , ... ,r ,r '"
.:
~~,
~~ ~ ~ ~:',
..... ~_~~
~ - ~ ~

r. _.".". ':.
..
~~~~
r_...........
... ..

. .....
~ ,

0.25 0 0.07

[m/s] [m/s]

-0.25 -0.07

E 2.0 F

Figure 7.6: Distribution of the absolute (A) and relative (B) ve-
locities, inside a water plug. The contours in (A) denote the veloc-
ity magnitude. The values of the streamwise (C) and wall normal
(D) relative velocities as well as the swirling strength (E) and the
streamlines (F) are shown.

Further experiments determining the influence of the liquid properties on


the velocity distribution inside the liquid plug were performed using an
7.2 Liquid velocities in straight rectangular microchannels 141

-1.0 m/s .... 0.2 m/s


B
A 0.55 ~~~~_-:~-:.-::;-;-.----:::
........... - ...._... "'
_~~

. -, -.....-
-~
...

~....--
~~....."...-

w::.~:::::::=-::::
_
~r~r

-.~-
r.

::-
~,
,-_ ....
.. -
_

-_.-
M ....... . . - . . - , . . . . . . . _
...

[m/si
... _
-~--_
~._~~
----.-
-----.

_-- .....
....... - .....
~

~
" .....
. ............
,----~
~.~--~
. . . . __ r ~
~ ~-
~--
-
, ... ...............
' ~~ __

o ~~rL~~~
c 0.25 0 0.07

[m/si [m/si

~0.25 -0.07

E 2.0 F

Figure 7.7: Distribution of the absolute (A) and relative (B) ve-
locities, inside an ethanol plug. The contours in (A) denote the ve-
locity magnitude. The values of the streamwise (C) and wall normal
(D) relative velocities as well as the swirling strength (E) and the
streamlines (F) are shown.

aqueous Glycerol solution as liquid phase. This fluid is characterized by


a viscosity similar to ethanol (Vgl yc = 1.147 mPas, Veth = 1.074 mPas)
142 7. Liquid velocity measurements

and a value for the interfacial tension against nitrogen similar to water
(O"glyc = 70.50 mN/m, O"water = 71.99 mN/m). These properties yield a
liquid Reynolds number of ReL = 6.64. The average streamwise veloc-
ity of the liquid plug is lower than in the two cases presented above,
VtHllk = 0.22 m/so This can also be seen in the distribution of the abso-
lute velocity distribution in Fig. 7.8A. However, the distribution of the
relative velocities resembles the ones found using water as liquid. Low
relative velocities and two stagnation points at the center line are signif-
icant. The distribution of the swirling strength as well as the streamlines
show similar behavior as water. In contrast to the previous experiment
(ethanol - nitrogen), the streamlines and the relative vector plots of the
two configurations using watf~r and glycerol, indicate vortices of nearly
triangular shape, symmetrical not only with respect to the center line of
the microchannel, but also with respect to the wall normal liquid plug
center.
A similar behavior can be found for a microchannel with a smaller hy-
draulic diameter of D h = 187.5 J-Lm. Using the same volumetric flow
rates CVL = 0.1 ml/min, VG = 1.0 ml/min), the superficial velocities
are increased to .iL = 0.044 m/s and ,iG = 0.44 m/s, respectively.
For this geometry, the liquid Reynolds numbers are ReL,water = 9.15,
RCL,eth = 6.06 and RCL,glyc = 5.00. The gas phase Reynolds number is
RCG = 5.:39 for all the analyzed fluid combinations. The average stream-
wise velocities are Vbulk = 0.46 m/s (water), 0.64 m/s (ethanol), and
0.43 m/s (glycerol), respectively.
The highest relative velocities were again measured in the ethanol case
(Fig. 7.9B). The two conditions "water - nitrogen" (Fig. 7.9A) and "glyc-
erol - nitrogen" (Fig. 7.9C) show similar characteristics, including low
velocities at the center line and two stagnation points in the vicinity of
the gas bubbles leading and trailing end. The maximum swirling strength
is measured for the ethanol - nitrogen flow. As a result of the lower bulk
velocity in the glycerol slug, an increased slug length is detected (con-
servation of mass).
In all the presented configurations, the gas bubble trailing end curvature
radius is found to be larger than the leading end curvature radius. For
the ethanol plug however, the leading edge is significantly more curved
than in the two other configurations. This leach; to an asymmetry of the
7.2 Liquid velocities in straight rectangular microchannels 143

-+ 1.0 rn/s ... 0.2 rn/s

A 0.55

"~"
,

........
~

.
- -

~~~--
-

...
~

--_ ,.......
.... ~ .. '

~.
r ,

,.. ,: :::::-::::: : : :
,"~ ~---~~~ ,
., '1,'" __ _ .

~--- -
~ ' ::
~""~~~_r

-- _ ... - - ... ,. ~

[rn/s]

---
__-, .~.w~
_._-~~~~,

- - - - __ - ;.... . .... -_-


- .. .., . . . ~ ~ ~
~

~
~~~

~_"~r"'''~_'~''''''''''''''

...........
--_................. ...,., ..
... " " ,
.... -
, I'.
~

o -
r ~ -
.. ..
_ ~ ~...

... ~
..
,
~
- - . ..
~ ~

, , ... , ..
~
. ~ .
---_.......
..... ~ ~ ~ ~ r r r , .

~--

c 0.25 0 0.07

[rn/s] [rn/s]

~0.25 -0.07

E 2.0 F

Figure 7.8: Distribution of the absolute (A) and relative (B) ve-
locities, inside a aqueous glycerol plug. The contours in (A) denote
the velocity magnitude. The values of the streamwise (C) and wall
normal (D) relative velocities as well as the swirling strength (E)
and the streamlines (F) are shown.

two counter-rotating vortices with respect of the line perpendicular to


the flow direction (Fig. 7.9, right).
144 7. Liquid velocity measurements

--
A ~ \

,\"~---~~~~~
'\ \
\

'I,
~ ~ ~ w

_ .... _ ~<#'''''''.".
.".

.,,. ..
r

,
_

~ -~--

..... - --. "."


, ,""
'\

"~--
".

",."
.".
'"

,
..
,j

..

-
~

.....
.... , _., --
. . . ,, ,, ,, ,, . -
.
~ ~

, , . -
~ ~ '\ ~

- .,. - _
~.., \ > , ~

~ ~

,.. ."" .. -
~
~ ~~-_

"
_.-.,,~,--_
. . .. ---~
-_ ~~
~'''''''''''
'I. ..

,
'\
,

,\

\
~

~ ~
>.
.
... .. ~ ~ r ~ .

... I I I,,' ",-" -

.. .. ---
".1"

-~
... J

B "._---------.".
.......... ~"
","
~
__ '

-_--
..... .... "".,,. _rv

............~~------~~~~... .",~
~,'~ ..... ~~~~-_, _ _ ~ .. " . " r

_.L ~_.,""
.......... ,"",.. - ~;"'''' ,
\-~
.~
-~.., ~
-, ...
,.

.... - ----.-----
_r
~
~
-" ,, ..-
,_
I

~ ......
- ~~-=:-":::;:::""--'-:=-:::::.--~- ,:"_"::.:

,I"r ~ ~
/f' _
1', .,,, " ....
,~. + , , J _ , ~ .. ~ ~ ~....,

~~ ~ - - , , I , , , __ ~~

:::-----------~-~~~~~~----

C -:::::::::::---::::::::::
" ~_ ~"''''~~-~-~~y

'
...... ... _.,r __
~

- _-_
.~, ~-----~~~~~~+.

_
-- .
& ~
~,
,~~------~~~~.,+
_~--
... .....
- +
".,
_-

- -- -" , ... -
---.... , ., _ N ~ ," ~ , _

-"""- ~ ~

-_. _ -
r~--""--"''''''''''
, I ,~~---
r __
_
~
_ , _ , ~ , , .. r _

,- "
-~--_ '."-

-_. ...
.....
-... r r".~ ... _+- _ _" " " " " ' ' ' ' ' ' ' ' ' ' ' ' ,
_~~-+----, ~ .. . "-, .
. -

...
~
~"
...... ~ .
,.~".,---~_

~~~_~~~
I ,
-.
-
-.....
,
~~~
~ ~

~~ __
,
~J~J ....
, _ ..
~

-.
0.5 rnls o 17.0 100 J..Lrn

Figure 7.9: Comparison of the relative velocities (left), the swirling


strength (center) and the streamlines (right) in a liquid plug of
water (A), ethanol (B) and glycerol (C) for a microchannel with a
hydraulic diameter of D h = 187.5 Mm.

Obviously, the surface tension affectH not only the Hhape of the gaH bubble
leading and trailing edge, and therefore the shape of the liquid segment,
but also the velocity distribution inside the liquid phase. Lower surface
tension values tend to increase the internal recirculation velocities and
therefore the mixing quality.
All the presented cases in a straight microchannel of rectangular cross-
Hection Hhow no Hignificant maHH transfer across the centerline of the
7.3 Recirculation in meandering channels 145

microchannel. Especially for catalytic gas-liquid reactions where the cat-


alyst is applied at the reactor walls, this can be a relevant disadvantage.

7.3 Recirculation in meandering channels


Due to the strictly laminar flow behavior in microfluidic devices (ab-
sence of turbulence), geometrical adaptations are required to enhance
the transport across the center line of a microreactor. The best mixing
can be achieved using static mixers (Losey et al., 2002; Wang et al.,
2002, 200;)). The disadvantage of microfluidic static mixers is a signifi-
cant increase of the pressure drop, and the loss of prediction quality due
to the loss of a predefined flow pattern. Both of these handicaps can be
avoided using meandering microchannels instead. It can be shown, that
the flow pattern is insignificantly affected by the turnings of meander-
ing channels. The total pressure drop increases only weakly due to the
elongated channel length.
To study the enhancement of the recirculation, the velocity distribution
in liquid plugs is measured for meandering channels of different hydraulic
diameters and turning radii. Two different hydraulic channel diameters
were considered, a rectangular channel cross-section with D h = 200 Mm
and a square cross-section with D h = 150 Mm. For the latter channel
diameter, turning radii of r = 600 Mm and r = 1200 Mm were considered.
The velocity distribution changes from the point entering the curve to its
end, which implies the necessity of averaging only liquid plugs at the ex-
act same angular position in the curve. To obtain a significant number of
recordings at a specific angular position, an unreasonably large number
of images is requil'(-~d. Therefon~, the effect of the meandering chaIllwl was
analyzed using measurement locations on the straight channel sections
directly before entering and after leaving the channel's bend. Instanta-
neous, non-averaged velocity distributions were recorded to analyze the
changes in the recirculation during the change of flow direction in the
curve of the microchannel. Fig. 7.10 shows the measurement locations
for the channels considered.
In Fig. 7.11 to 7.13, the characteristics of the two-phase Bow entering the
146 7. Liquid velocity measurements

r - - - - - - - - - - - - - - - - - -,
:3 ,
,
,,
I
I
I

\ j.. .,r,- --~


I

:1 2:
I I
,
,,,
I
I
I
I
,,
._-
I

Figure 7.10: Locations of the recirculation measurements for a


meandering channel. The velocity distribution is analyzed directly
before the liquid enters a curve (1) and just after the corresponding
bend (2). Instantaneous recordings of liquid plugs were made in the
curved section of the meandering channel (3).

channel turning are shown on the left, whereas the liquid plug leaving
the bend is found on the right hand side.
For all the presented results, water and nitrogen were used as working
fluids. For the large diameter channel (D h = 200 p,m) with a turning
radius T = 1200 Mm, and je = 0.37 m/s eVe = 1 ml/min), and jL =
0.04 m/s CVL = 0.1 ml/min) as fluid superficial velocities, the liquid
recirculation in the plug just before entering the channel curve (Fig.
7.11, top) shows the same characteristics as for the straight channels in
chap. 7.2. Two counter-rotating vortices symmetrical to the channel axis
can be found. For the liquid plug after the channel bend, however, the
velocities inside the liquid plug are more uniform, which implies lower
relative velocities. The vorticity (top) and the swirling strength (center)
7.3 Recirculation in meandering channels 147

decreased following the channel turning. The streamlines, as well as the


relative velocity vectors, indicate no remarkable increase in mass transfer
across the channel center line.
The same volumetric gas and liquid flow rates (Vc = 1 ml/min, VL =
0.1 ml/min) in a channel with a hydraulic diameter of D h = 150 /-tm
increase the superficial velocities by a factor of 2 (jc = 0.74 m/s, jL =
0.07 m/s). The decrease of the vorticity, the relative velocities, and the
swirling strength during the channel bend is found also for the smaller
channel geometry (Fig. 7.12). The liquid segment is slightly elongated
during the angular motion.
Decreasing the volumetric flow rates to Vc = 0.75 ml/min and VL =
0.05 ml/min, leads to similar superficial velocities as for the channel
analyzed in Fig. 7.11 (jc = 0.56 m/s, jL = 0.04 m/s). However, in Fig.
7.13 no remarkable decrease, neither in the relative velocities, nor in the
vorticity distribution can be found. As in the previous cases, the liquid
plug is elongated during the channel turning. The streamlines indicate
no advantageous mass transfer across the channel center line just after
leaving the bend.
Similar behavior can be found for intermittent flow in a meandering
microchannel with turnings of radius r = 600 /-tm. For the same experi-
mental conditions (water - nitrogen flow, D h = 150 /-tm, jc = 0.55 m/s,
.iL = 0.07 m/s), a slightly increased vorticity can be measured after the
channel bend (Fig. 7.14). The swirling strength is not more decreased,
but the area of the maximum swirling strength is enlarged. The stream-
lines of the relative velocities show hardly any transfer motion across the
channel center line.
The improved mixing predicted for meandering channels can not be con-
firmed with measurement locations just after a liquid segment leaves a
channel for our conditions.
As the liquid plugs entering the channel bend feature nearly the same
shape as those leaving the curve (slightly elongated), higher velocities
at the channel outside are expected in the channel turning, leading to a
velocity gradient at the center line or a fluid tra,nsfer between the two
channel halves. In Figs. 7.16 and 7.15, the instantaneous, non-averaged
148 7. Liquid velocity measurements

-------. 0.5 [m/s]


8

[1/5]

-8
-------. flow direction -------. flow direction
1.2

0.0

Figure 7.11: Characteristics of two liquid plugs entering (left) and


leaving (right) a microchannel bend (r = 1200 /-Lm) for a channel
with D h = 200 /-Lm and a fluid system with ja = 0.37 m/s eVa =
1 mJ/min) and jL = 0.04 m,/s (VL = 0.1 ml/min). The relative
velocity vectors together with the liquid vorticity are shown on the
top comparison. The swirling strengths are depicted in the center,
whereas the Htreamlines of the two configurations can be seen at the
bottom.
7.3 Recirculation in meandering channels 149

- - 0.5 [m/sI
16

[1/s]

-16

12

100 Ilm

Figure 7.12: Characteristics of two liquid plugs entering (left) and


leaving (right) a microchannel bend (r = 1200 J.lm) for a channel
with Dh = 150 J.lm and a fluid system with jc = 0.74 m/s (Vc =
1 ml/min) and jL = 0.07 m/s (VL = 0.1 ml/min). The relative
velocity vectors together with the liquid vorticity are shown on the
top comparison. The swirling strengths are depicted in the center,
whereas the streamlines of the two configurations can be seen at the
bottom.

relative velocity vectors, as well as the vorticity is outlined. The two-


phase flow is flowing clockwise, entering from the right at the bottom
of the figures. After a short inlet zone, strong relative motion diagonally
through the liquid slug is detected. Strong mass transfer over the center-
line is present. The vortex on the channel outside is hardly identifiable,
whereas the vortex on the channel inside is shifted towards the gas bubble
150 7. Liquid velocity measurements

- . 0.5 [m/s]
16

[1/s]

-16

100 J.l.m

Figure 7.13: Characteristics of two liquid plugs entering (left) and


leaving (right) a microchannel bend (r = 1200 Mm) for a channel
with D h = 150 Mm and a fluid system with jG = 0.56 m/s (VG =
0.75 ml/min) and jL = 0.04 m/s (VL = 0.05 ml/min). The relative
velocity vectors together with the liquid vorticity are shown on the
top comparison. The swirling strengths are depicted in the center,
whereas the streamlines of the two configurations can be seen at the
bottom.

trailing end. High relative velocities and strong vorticity at the channel
outside are determined for high superficial velocities (Fig. 7.16).
The calculation of the streamlines and the swirling strength for the in-
stantaneous velocity distributions did not show reasonable information
7.3 Recirculation in meandering channels 151

-+ 0.5 [m/sI
16

[1/5]

-16

~F 1100~m
Figure 7.14: Characteristics of two liquid plugs entering (left) and
leaving (right) a microchannel bend (r = 600 /l-m) for a channel
with Dh = 150 /l-rn and a fiuid HyHtern with jc = 0.56 m/s (Vc =
0.75 ml/min) and jL = 0.04 m/s (VL = 0.05 ml/min). The relative
velocity vectors together with the liquid vorticity are shown on the
top comparison. The swirling strengths are depicted in the center ,
whpn~as thp strpamlines of the two confignrations can be seen at the
bottom.

due to the strong infiuence of erroneous velocity vectors present.


The existence of an increased transfer motion across the channel center
line is further analyzed by calculating the transverse (wall normal) and
the streamwise component of the average velocity dependent on the an-
gular position of the liquid plug in the channel curve (turning angle).
Fig. 7.17 clearly shows an increase of the average wall normal velocity
component (loss of symmetry) and a decrease of the average streamwise
152 7. Liquid velocity measurements

Vorticity:

-16 [1/5] 16

Velocity:
- 0.5 [m/5]

Figure 7.15: Instantaneous relative velocity distributions and vor-


ticity contours for liquid plugs in a microchannel with Dh = 150 Mm
and radius r = 1200 !-tm. The superficial velocities were je =
0.56 m/s and jL = 0.04 m/so The two-phase flow is flowing clock-
WIse.

velocity of the liquid plug. Note that the exact plug position (turning
angle) in the channel bend is difficult to measure in the raw images
((angle ~ 2). Due to the strong dependency of the average velocity
components on the turning angle, the accuracy is limited.
7.4 Summary and conclusions 153

Vorticity:

-16 16

Velocity:
...... 0.5 [m/s]

Figure 7.16: Instantaneous relative velocity distributions and vor-


ticity contours for liquid plugs in a microehannel with D h = 150 jJm
and radius r = 1200 jJm. The superficial velocities were je =
0.74 m/s and jL = 0.07 m/so The two-phase flow is flowing clock-
WIse.

7.4 Summary and conclusions


The recirculation motion inside the liquid segments of a multiphase gas-
liquid How in a microchannel were analyzed using micron-resolution Par-
ticle Image Velocimetry (jJPIV). Due to the unsteady flow pattern, a
simple averaging process is not suitable to study the velocity profile. An
additional cross-correlation step was necessary to extract the position
of the liquid plug. Therewith an averaging of a specific area of interest
154 7. Liquid velocity measurements

0.05
~

--
0
0
E
......
g'
0.04

0.03
>
III o
0.02 o
g' o
>111 o
0.01
~ o 0 Il 0 0
'0
0
0.00 ______ .~-----------------------------L-------

,
Q)
> 0.01
Q)
Cl
~ -0.02
,
~
co -0.03
Q)
>
:;:;
co -0.04
o
Q)
a::: 0.05
0 20 40 60 80 100 120 140 160 180
Turning angle rl
Figure 7.17: Average spanwise (wall normal, transfer motion) and
streamwise velocity component inside the liquid plug relative to the
velocities in a straight microchannel dependent on the liquid plug
position (turning angle).

within the vector file was possible. As the velocity distribution is ex-
pected to change as the fluid advances in curved meandering channels
(change in angular position), averaging is abandoned and the instanta-
neous recirculation motion is analyzed.
Of special interest concerning the mass transfer of liquid towards the
phase boundary or a catalyst material coated onto the reactor walls,
was the mixing q~ality and relative velocity distribution. Especially the
existence of mass transfer across the channel center line is desirable. The
computation of the streamlines and the analysis of the swirling strength
gave additional information on the internal recirculation.
For straight microchannels, an increase in relative velocities for low sur-
7.4 Summary and conclusions 155

face tension liquids was detected. In contrast to experiments using water


or aqueous glycerol as working fluid, the relative velocities at the channel
center line for a hydraulic channel diameter of Dh = 210 /-lm are high
using ethanol. No smooth swirling strength contours were detectable.
The streamlines indicate little mass transfer across the center line. The
same characteristics were found in channels with smaller diameter.
Enhanced mixing can be achieved by geometrical adaption (inclusion of
static mixer) or by implementing meandering channels. For meandering
channels with two different curve radii and hydraulic diameters the re-
circulation motion inside a liquid plug just before entering the curve was
compared to the velocity distribution after leaving the channel bend. In
all the analyzed cases, the channel curvature caused a liquid segment
elongation, and, consequentially, a decrease in bulk velocity. The inter-
nal relative velocities dropped as well as the swirling motion (swirling
strength). Despite the presence of a channel bend, hardly any mass trans-
fer across the center line was observed.
The analysis of the instantaneous velocities along the channel bend in-
dicated the existence of mass transfer across the center line (improved
mixing). Strong diagonal motion is confirmed. Obviously, this effect is of
short duration and disappears as soon as the liquid slug exits the chan-
nel curve. Hence, a meandering channel without any straight channel
section is expected to yield the best mixing results.
For catalytic reactions, the internal mixing process determines the mass
transport towards the catalyst. In the case of a catalyst coated on the
channel walls, additional mixing using static mixers (hair pins) or other
adaptations enhancing the two-phase mixing are suggested. The advan-
tage of improved' catalyst contact outweighs the disadvantages of the
additional pressure drop.
156

Chapter 8

Hydrogenation of
nitrobenzene

The nitrobenzene hydrogenation reaction in most cases is carried out in


common large scale reactors (Yu et al., 2000; Gelder et al., 2002; Holler
et al., 2000; Figueras and Coq, 2001; Rode et al., 2001). Stirrers are used
for intense mixing and the temperature and the pressure are usually kept
constant at T = 50C and p = 4 ... 15 bar.
In the present study, we used different designs of microreactors, packed-
bed reactors and palladium coated reactors (chap. 3.1.1.2 and 3.2).

8.1 Packed-bed reactor

In the packed-bed reactor, the catalyst (palladium) is inserted in the mi-


crochannel in powder form. Activated carbon microparticles covered with
palladium are considered. Unfortunately the particles clogged the imple-
mented barriers completely (Fig. 8.1), leading to an immense pressure
drop inside the microchannel. Due to this circumstances, we abandoned
the use of the packed-bed reactor type.
8.2 Coated microreactor 157

Figure 8.1: Catalyst particles in the microchannels clogging com-


pletely the implemented barriers.

8.2 Coated microreactor

Yeong et al. (2004) used a falling film microreactor for the hydrogenation
of nitrobenzene, featuring 64 straight, parallel channels (300 Mm wide,
100 Mm deep, and 78 mm long). Experiments were conducted at 60C,
varying the pressure from 1 to 6 bar. The liquid flow rates were 0.5 -
3 ml/min nitrobenzene solution. Depending on the applied pressure and
the liquid flow rate, conversion rate between 0.25 and 0.85 were achieved.
According to the results of Yeong et 1'11. (2004), we conducted the hy-
drogenation at different flow and boundary conditions. The microreactor
and the inlet tubing were heated to 60 - 75C and pressures between 5
and 10 bar were applied. The nitrobenzene was diluted in pure ethanol
(1:25 by volume). The liquid flow rate was kept as small as possible to
increase the residence time (5 to 10 MI/min). The gas flow rate was lim-
ited by the applied back pressure regulator, requiring a minimum flow
rate to ensure a specific pressure. Minimum flow rate ensuring a pres-
sure of 10 bar was 0.5 Nml/min hydrogen. The applied microreactor had
a hydraulic diameter of D h = 150 Mm (square cross-section) and a to-
tal two-phase length of 120 mm. Assuming a volumetric ga.-s content of
~ 0.8, the residence time is estimated to be in the order of 3.25 to 6.5 s.
158

Chapter 9

Concluding remarks and


outlook

9.1 Concluding remarks


The ability of microfabrication to produce complex designs of (parallel)
microreactors should invigorate the design of innovative reactors, avoid-
ing some of the disadvantages associated with stirred tanks or trickle
bed reactors. In the development of microreaction technology, it is es-
sential to stay focussed on systems where the use of microreactors can
provide unique process advantages. Such advantages could be derived
from increased niass and heat transfer leading to improved yield and
safety for an existing process. The real value of the miniaturization ef-
fort, however, would be in exploring new reaction pathways and finding
economical and environmentally benign solutions to chemical manufac-
turing (Jensen, 1999).
To move beyond the laboratory into chemical production, microreactors
must be integrated with sensors and actuators either on the same chip
or through hybrid integration schemes. The integration of chemical sys-
tems with sensors in J.LTAS (micro total analysis systems) is already a
9.1 Concluding remarks 159

rapidly expanding field. However, for the proper design of actuators (e.g.
pumps), it is crucial to know the flow characteristics of the desired pro-
cess. It appears to be necessary to accumulate experimental data on flow
characteristics (which includes the flow regime, the pressure drop, and
the void fraction) to develop new equations and accurate models for the
prediction of these characteristics in microfluidic systems. In view of this,
the flow regime, the pressure drop, the void fraction, and the liquid veloc-
ity distribution (recirculation motion) have been studied in the present
research project for two-phase flow in small-diameter microchannels. All
the mentioned flow properties have been shown to behave significantly
different from the large-scale pipe flow. Therefore, the existing correla-
tions are not able to predict the phenomena of microchannel two-phase
flow, as the influence of the hydraulic diameter does not scale linearly.
Furthermore, the increasing importance of the surface forces (and con-
sequently the loss of importance of the body forces) causes novel phe-
nomena, implicating the necessity of further experimental investigation.

9.1.1 Ecological considerations

Despite the advantages of microreaction systems compared to their


macro scale counterparts, also the ecological point of view has to be
considered. The ecological consequences of products and processes are
::;ystematically analyzed using an ecobalance. The influence of the fab-
rication and the application of microreactors on the environment are
investigated.
Considering only the fabrication of the silicon based microreactor, the
influence on the environment of the used materials (silicon, chemicals,
etc.) of the photolithography and the etch processes, as well as the main-
tenance of the clean room facilities are considered.
It can be shown that the maintenance of the clean room facilities con-
tribute the major fraction (:=::::: 66%) to the ecological consequences, mean-
ing that the clean room does the major damage to the environment.
Clean room facilities have to be operated 24 hours a day to ensure the
required degree of purity for the silicon reactor fabrication. The DRIE
etching facility is shown to make the second largest contribution in the
160 9. Concluding remarks and outlook

ecobalance. This is mainly due to the high energy consumption. All other
equipment contribute less than 0.5% to the total energy use. The relative
influence of the clean room maintenance can only be reduced by a higher
efficiency or load of the facility.
Besides the fabrication also the use of silicon based microreactors for
the production of aniline (hydrogenation of nitrobenzene) was analyzed.
It can be shown, that the fraction of of the lifetime of a microreactor
used for effective aniline production is in the range of 7%. Conventional
batch reactors are in the order of 50%. Not taken into consideration are
the chemicals and facilities for the reactor fabrication. Obviously eco-
logical reasons do not legitimate the use of microreactors for the aniline
production.

9.2 Outlook
Further research work advancing the present project is proposed, con-
tinuing the path to chemical engineering applications.

9.2.1 Reactions with supercritical solvents in mi-


croreactors

The combination of the enhanced heat and mass transfer properties in


micro-scale reactors and the unique fluid properties of supercritical re-
action media, including elimination of gas/liquid phase boundaries, pro-
vide unique and unparalleled synergies for novel exploitation in micro
reaction engineering. The challenging opportunity to develop novel envi-
ronmentally beneficial and inherently safe catalytic reaction systems by
combining the beneficial properties of micro-scale structures and super-
critical fluids is the main rationale of the proposed project.
Research in microreaction engineering mostly focuses on exothermic re-
actions at near-ambient pressures to demonstrate better temperature
control at the micro-scale. High pressure reactions in microreactors are
rarely reported. The pressure management in micro-scale reactors, due
9.2 Outlook 161

to small dimensions, is easier to realize than in macro-scale setups. In


addition, the high surface-area-to-volume ratio facilitates better control
of temperature in the case of exothermic reactions. These are perhaps
the most compelling reasons to investigate reactions with supercritical
solvents in microreactors. From a manufacturing perspective, microfab-
rication allows the integration of heaters and sensors directly into the
reactor and rather easy tailoring of designs for special applications. For
fast and high throughput analysis e.g. for catalyst libraries, microre-
actors are an alternative solution to common laboratory analysis equip-
ment. The proposed project may thus also contribute to the development
of high-throughput analysis techniques following the idea of "Lab-on-a-
chip". Current commercially available reactors for use with supercritical
solvents are typically in the milliliter range. The proposed microreactor
assembly is at least an order of magnitude smaller.
As a model reaction, the Pd-catalyzed cyclohexene hydrogenation to cy-
clohexane in supercritical carbon dioxide is suggested. Measurements of
the operating parameters p, T and flow rate can be obtained in situ. Ex
situ, Gas Chromatography (GC) and Mass Spectroscopy (MS) can be
used for subsequent qualitative and quantitative information on prod-
uct composition, respectively. Systematic investigations of the effects of
the operating parameters p, T, flow rate on the performance parame-
ter turnover frequency, temperature control and catalyst stability of the
model reaction system in the fabricated microreactor will be addressed.
A comparison of the experimentally obtained data with numerical simu-
lations is proposed. The modelling of the microreactor (fabricated suit-
able for supercritical reaction conditions, ~ 150 bar and 150C) can
be conducted with the finite element method (FEM) based software
FEMLAB (Comsol Inc.). This software package allows solving multi-
physics problems (coupled differential equations) such as combined ther-
mal and structural analysis. Firstly, the mechanical behavior of the mi-
croreactor under reaction conditions is simulated. Material properties
like E-modulus, thermal conductivity, thermal expansion coefficient and
heat capacity and their thermal dependencies have a significant influence
on the stress and strain distribution inside the microreactor. Therefore,
the numerical simulation plays in important role for the dimensioning
and the continually improving of the reactor design.
162 9. Concluding remarks and outlook

HPLC
Pump
/f
I
I
Substrate

onIlne sample
Analysl. method.: Preparation
Thermocouples
SubsequenlAnalysis
Pressure SenSOl'&
lIlPIV)

Figure 9.1: Schematic of a possible experimental setup for the


supercritical hydrogenation of cyclohexene.

9.2.2 Design of a micromixer for industrial relevant


reactions

Gas-liquid (catalytic) reactions require a high mixing quality to ensure


small diffusion paths. The use of meandering channel layouts was shown
to affect the internal recirculation motion without changing the over-
all flow behavior. For a further increa..'le in interfacial area geometrical
adaptations of the microreactor are required. Several mixing possibilities
have to be considered. The influence of the micromixer on the pressure
drop, the bubble length (intermittent flow), the bubble stability, and the
mass transport are investigated.
9.2 Outlook 163

To minimize the experimental cost and to properly predict the quality


of any mixer design, the process is planned to be modelled numerically.
Objective of the proposed research project is to experimentally and nu-
merically determine the optimum micromixer design, weighing up the
additional pressure drop and the achieved increase in surface-to-volume
ratio depending on fluid properties, channel material, and channel ge-
ometry.
164

References

Adrian, R. J., Christensen, K. T., and Liu, Z. C. (2000). Analysis and


interpretation of instantaneous turbulent velocity fields. Experiments
in Fluids, 29(3):275-290.
Adrian, R. J. (1991). Particle-imaging techniques for experimental fiuid-
mechanics. Annual Review of Fluid Mechanics, 23:261-304.
Armand, A. and Treschew, S. (1946). Izv. Vsesou Teplotek Inst., 1(16).
Baker, O. (1954). Simultaneous flow of oil and gas. Oil Gas J., 53:184-
195.
Born, M. and Wolf, E. (1997). Principles of optics. Oxford:Pergamon
Press.
Chawla, J. M. (1968). Friction pressure drop in 2-phase flows of liq-
uid/gas mixtures. Chemie Ingenieur Technik, 40(13):670.
Chisholm, D. (1967). A theoretical basis for lockhart-martinelli cor-
relation for 2-phase flow. International Journal of Heat and Mass
Transfer, 10(12):1767.
Chong, M. S., Perry, A. E., and Cantwell, B. J. (1990). A general clas-
sification of 3-dimensional flow-fields. Physics of Fluids a-Fluid Dy-
namics, 2(5):765-777.
Coleman, J. W. and Garimella, S. (1999). Characterization of two-phase
flow patterns in small diameter round and rectangular tubes. Inter-
national Journal of Heat and Mass Transfer, 42(15):2869-2881.
References 165

Coleman, J. W. and Garimella, S. (2003). Two-phase flow regimes in


round, square and rectangular tubes during condensation of refrigerant
r134a. International Journal of Refrigeration-Revue Internationale Du
Froid, 26(1):117-128.
Coleman, J. W. and Krause, P. E. (2004). Two phase pressure losses of
r134a in microchannel tube headers with large free flow area ratios.
Experimental Thermal and Fluid Science, 28(2-3):123-130.
Dallman, D., Hilgenstock, A., Riedelbanh, S., Schulte-Werning, B., and
Vollmers, H. (1991). On the footprints of three-dimensional separated
vortex flows around blunt bodies. AGARD CP-494.
de Mas, N., Gunther, A., Schmidt, M. A., and Jensen, K. F. (2003).
Microfabricated multiphase reactors for the selective direct fluorina-
tion of aromatics. Industrial and Engineering Chemistry Research,
42(4):698-710.
Ehrfeld, W., Hessel, V., and Lowe, H. (2000). Microreactors. New Tech-
nology for modem chemistry. Wiley-VCH.
Fairbank, W. M., Hansch, T. W., and Schawlow, A. L. (1975). Absolute
measurement of very low sodium-vapor densities using laser resonance
fluorescence. Journal of the Optical Society of America, 65(2):199-204.

Figueras, F. and Coq, B. (2001). Hydrogenation and hydrogenolysis


of nitro-, nitroso-, azo-, azoxy- and other nitrogen-containing com-
pounds on palladium. Journal of Molecular Catalysis a-Chemical,
173(1-2):223-230.
Friedel, L. (1978). Pressure-drop during gas-vapor-liquid flow in pipes.
Chemie Ingenieur Technik, 50(3):167-180.
Gad-el Hak, M. (1999). The fluid mechanics of microdevices - the free-
man scholar lecture. Journal of Fluids Engineering-Transactions of
the Asme, 121(1):5-33.
Gad-EI-Hak, M. (2004). Transport phenomena in microdevices.
Zeitschrijt Fur Angewandte Mathematik Und Mechanik, 84(7):494-
498.
166 References

Gallardo, B. S., Gupta, V. K., Eagerton, F. D., Jong, L. I., Craig,


V. S., Shah, R. R., and Abbott, N. L. (1999). Electrochemical prin-
ciples for active control of liquids on submillimeter scales. Science,
283(5398):57-60.
Gelder, E. A., Jackson, S. D., and Lok, C. M. (2002). A study of ni-
trobenzene hydrogenation over palladium/carbon catalysts. Catalysis
Letters, 84(3-4):205-208.
Gravesen, P., Branebjerg, J., and Jensen, O. S. (1993). Microfluidics - a
review. Journal of Micromechanics and Microengineering, 3:168-182.
Hassa, C., Paul, P. H., and Hanson, R. K. (1987). Laser-induced fluores-
cence nlOdulation techniques for velocity-measurements in gas-flows.
Experiments in Fluids, 5(4):240--246.
Holler, V., Wegricht, D., Yuranov, I., Kiwi-Minsker, L., and Renken, A.
(2000). Three-phase nitrobenzene hydrogenation over supported glass
fiber catalysts: Reaction kinetics study. Chemical Engineering and
Technology, 23(3):251-255.
Huq, R. (1998). An analytical two-phase flow void fraction prediction
method. AIAA/ASME 5th Joint Thermophysics and Heat Transfer
Conf., Paper No. 90-1738.
Inoue, S. and Spring, K. (1997). Video microscopy, 2nd ed. Ox-
ford:Plenum Press.
Jahnisch, K., Baerns, M., Hessel, V., Ehrfeld, W., Haverkamp, V., Lowe,
H., Wille, C., and Guber, A. (2000). Direct fluorination of toluene us-
ing elemental fluorine in gas/liquid microreactors. Journal of Fluorine
Chemistry, 105(1):117-128.
Jensen, K. F. (1999). Microchemical systems: Status, challenges, and
opportunities. Aiche Journal, 45(10):2051-2054.
Jensen, K. F. (2001). Microreaction engineering - is small better? Chem-
ical Engineering Science, 56(2):293-303.
References 167

Kawahara, A., Chung, P. M. Y., and Kawaji, M. (2002). Investiga-


tion of two-phase flow pattern, void fraction and pressure drop in a
microchannel. International Journal of Multiphase Flow, 28(9):1411-
1435.
Kestenbaum, H., de Oliveira, A. L., Schmidt, W., Schuth, F., Ehrfeld,
W., Gebauer, K., Lowe, H., Richter, T., Lebiedz, D., Untiedt, 1., and
Zuchner, H. (2002). Silver-catalyzed oxidation of ethylene to ethylene
oxide in a microreaction system. Industrial and Engineering Chemistry
Research, 41(4):710-719.
Kychakoff, G., Howe, R. D., Hanson, R. K., and McDaniel, J. C. (1982).
Quantitative visualization of combustion species in a plane. Applied
Optics, 21(18):3225-3227.
Lanzillotto, A., Leu, T., Amabile, M., Samtaney, R., and Wildes, R.
(1997). A study of structure and motion in fluidic microsystems. AIAA
Paper 97-1790.
Lee, H. J. and Lee, S. Y. (2001). Pressure drop correlations for two-
phase flow within horizontal rectangular channels with small heights.
International Journal of Multiphase Flow, 27(5):783-796.
Lockhart, R. W. and Martinelli, R. C. (1949). Proposed correlation
of data for isothermal 2-phase, 2-component flow in pipes. Chemical
Engineering Progress, 45(1):39-48.
Losey, M. W., Jackman, R. J., Firebaugh, S. L., Schmidt, M. A., and
Jensen, K. F. (2002). Design and fabrication of microfluidic devices
for multiphase mixing and reaction. Journal of Microelectromechanical
Systems, 11(6):709-717.
Losey, M. W., Schmidt, M. A., and Jensen, K. F. (2001). Microfabri-
cated multiphase packed-bed reactors: Characterization of mass trans-
fer and reactions. Industrial and Engineering Chemistry Research,
40(12):2555-2562.
Lumley, J. L. (1967). Similarity and turbulent energy spectrum. Physics
of Fluids, 10(4):855.
168 References

Mandhane, J. M., Gregory, G. A., and Aziz, K. (1975). Critical eval-


uation of holdup prediction methods for gas-liquid flow in horizontal
pipes. Journal of Petroleum Technology, 27(AUG):1017-1026.
Massena, W. (1960). Steam-water critical flow using the separated flow
model. Hanford Atomic Products Operation Report HW-65739.
Meinhart, C. D., Prasad, A. K., and Adrian, R. J. (1993). A parallel
digital processor system for particle image velocimetry. Measurement
Science and Technology, 4(5):619-626.
Meinhart, C. D., Wereley, S. T., and Santiago, J. G. (1999). Piv measure-
ments of a microchannel flow. Experiments in Fluids, 27(5):414-419.
Meinhart, C. D., Wereley, S. T., and Santiago, J. G. (2000). A piv algo-
rithm for estimating time-averaged velocity fields. Journal of Fluids
Engineering- Transactions of the Asme, 122(2):285-289.
Mishima, K. and Hibiki, T. (1996). Some characteristics of air-water
two-phase flow in small diameter vertical tubes. International Journal
of Multiphase Flow, 22(4):703-712.
Muller-Steinhagen, H. and Heck, K. (1986). A simple friction pressure-
drop correlation for 2-phase flow in pipes. Chemical Engineering and
Processing, 20(6):297-308.
Prasad, A. K., Adrian, R. J., Landreth, C. C., and Offutt, P. W. (1992).
Effect of resolution on the speed and accuracy of particle image ve-
locimetry interrogation. Experiments in Fluids, 13(2-3):105-116.
Rode, C. V., Vaidya, M. J., Jaganathan, R., and Chaudhari, R. V. (2001).
Hydrogenation of nitrobenzene to p-aminophenol in a four-phase reac-
tor: reaction kinetics and mass transfer effects. Chemical Engineering
Science, 56(4):1299-1304.
Santiago, J. G., Wereley, S. T., Meinhart, C. D., Beebe, D. J., and
Adrian, R. J. (1998). A particle image velocimetry system for mi-
crofluidics. Experiments in Fluids, 25(4):316-319.
References 169

Senkan, S., Krantz, K., Ozturk, S., Zengin, V., and Onal, 1. (1999). High-
throughput testing of heterogeneous catalyst libraries using array mi-
croreactors and mass spectrometry. Angewandte Chemie-International
Edition, 38(18):2794-2799.

Serizawa, A., Feng, Z. P., and Kawara, Z. (2002). Two-phase flow in


microchannels. Experimental Thermal and Fluid Science, 26(6-7):703-
714.
Sirovich, L. (1987). TUrbulence and the dynamics of coherent structures
.1. coherent structures. Quarterly of Applied Mathematics, 45(3):561-
571.
Stampanoni, M., Borchert, G. L., Abela, R., Patterson, B., Vermeulen,
D., Ruegsegger, P., and Wyss, P. (2002a). An x-ray tomographic
microscope with submicron resolution. Acta Physica Polonica B,
33(1):463-469..
Stampanoni, M., Borchert, G., Wyss, P., Abela, R., Patterson, B., Hunt,
S., Vermeulen, D., and Ruegsegger, P. (2002b). High resolution x-
ray detector for synchrotron-based microtomography. Nuclear Instru-
ments and Methods in Physics Research Section a-Accelerators Spec-
trometers Detectors and Associated Equipment, 491(1-2):291-301.

Stone, H. A. and Kim, S. (2001). Microfluidics: Basic issues, applications,


and challenges" Aiche Journal, 47(6):1250-1254.
Stone, H. A., Stroock, A. D., and Ajdari, A. (2004). Engineering flows
in small devices: Microfluidics toward a lab-on-a-chip. Annual Review
of Fluid Mechanics, 36:381-411.
Taitel, Y. and Dukler, A. E. (1976). Model for predicting flow regime
transitions in horizontal and near horizontal ga..<:;-liquid flow. Aiche
Journal, 22(1):47-55.
TeGrotenhuis, W. E., Cameron, R. J., Butcher, M. G., Martin, P. M.,
and Wegeng, R. S. (1999). Microchannel devices for efficient contacting
of liquids in solvent extraction. Separation Science and Technology,
34(6-7):951-974.
170 References

'lriplett, K. A., Ghiaasiaan, S. M., Abdel-Khalik, S. I., LeMouel, A., and


McCord, B. N. (1999a). Gas-liquid two-phase flow in microchannels
- part ii: void fraction and pressure drop. International Journal of
Multiphase Flo'!il, 25(3):395-410.
'lriplett, K. A., Ghiaasiaan, S. M., Abdel-Khalik, S. I., and Sadowski,
D. L. (1999b). Gas-liquid two-phase flow in microchannels - part i:
two-phase flow patterns. International Journal of Multiphase Flow,
25(3):377-394.
Veser, G. (2001). Experimental and theoretical investigation of h-2 oxi-
dation in a high-temperature catalytic microreactor. Chemical Engi-
neering Science, 56( 4):1265~ 1273.
Wallis, G. B. (1969). Annular 2-phase flow .i. a simple theory. Mechanical
Engineering, 91(10):73.
Wang, H. Z., Iovenitti, P., Harvey, E., and Masood, S. (2002). Optimiz-
ing layout of obstacles for enhanced mixing m microchannels. Smart
Materials and Structures, 11(5):662~667.
Wang, H. Z., Iovenitti, P., Harvey, E., and Masood, S. (2003). Numer-
ical investigation of mixing in microchannels with patterned grooves.
Journal of Micromechanics and Microengineering, 13(6):801~808.
Yeong, K. K., Gavriilidis, A., Zapf, R., and Hessel, V. (2004). Exper-
imental studies of nitrobenzene hydrogenation in a microstructured
falling film reactor. Chemical Engineering Science, 59(16):3491~3494.
Yu, X. B., Wang, M. H., and Li, H. X. (2000). Study on the nitrobenzene
hydrogenation over a pd-b/si02 amorphous catalyst. Applied Catalysis
a-General, 202(1):17~22.
Zhao, T. S. and Bi, Q. C. (2001). Co-current air-water two-phase flow
patterns in vertical triangular microchannels. International Journal
of Multiphase Flow, 27(5):765~782.
Zhou, J. G., Adrian, R. J., and Balachandar, S. (1996). Autogeneration
of near-wall vortical structures in channel flow. Physics of Fluids,
8(1) :288~290.
References 171

Zhou, J., Adrian, R. J., Balachandar, S., and Kendall, T. M. (1999).


Mechanisms for generating coherent packets of hairpin vortices in
channel flow. Journal of Fluid Mechanics, 387:353-396.
Zlokarnik, M. (1983). Model scale-up in chemical-engineering. Chemie
Ingenieur Technik, 55(5):363-372.
172

Curriculum Vitae

Severin Walchli
Date of birth: June 13th 1974
Place of birth: Brugg (AG), Switzerland
Citizen of: Brittnau (AG), Switzerland

04/2001-06/2005 Doctoral student (Ph.D.) and teaching assistant,


Institute of Process Engineering, Swiss Federal Insti-
tute of Technology (ETH), Ziirich, Switzerland
(advisors: Prof. Dr. Ph. Rudolf von Rohr, Prof. Dr.
Th. Rosgen)
10/1995-02/2001 Graduate studies in Mechanical Engineering,
Swiss Federal Institute of Technology (ETH) ,
Switzerland
academic degree: Dipl. Masch.-Ing. ETH
12/1998-07/1999 Research project (Semesterarbeit),
Michigan State University, East Lansing, Michigan,
USA
(advisor: Prof. Dr. M.M. Koochesfahani)
02/1994-08/1994' Obligatory military service, Swiss army
04/1990-02/1994 Alte Kantonsschule Aarau (AG)
academic degree: Matura, Typus C
04/1987-03/1990 Bezirksschule Windisch (AG)
03/1981-03/1986 Primarschule Hausen (AG)

Zurich, 2005