J Fluids Engineering 2009 Vol 131 N4

Journal of
Fluids Engineering
Published Monthly by ASME
VOLUME 131 NUMBER 4 APRIL 2009

FLUIDS ENGINEERING DIVISION
Editor
J. KATZ 2009
Assistant to the Editor
L. MURPHY 2009
Associate Editors
M. J. ANDREWS 2009
E. M. BENNETT 2012
S. L. CECCIO 2009
D. DRIKAKIS 2012
P. DURBIN 2012
I. EAMES 2010
C. HAH 2010
T. J. HEINDEL 2011
J. KOMPENHANS 2009
YU-TAI LEE 2009
J. A. LIBURDY 2011
R. MITTAL 2010
T. J. OHERN 2009
N. A. PATANKAR 2011
H. PEERHOSSAINI 2011
U. PIOMELLI 2010
Z. RUSAK 2010
D. SIGINER 2009
M. STREMLER 2012
M. WANG 2011
St. T. WERELEY 2011
Y. ZHOU 2009
RESEARCH PAPERS
Flows in Complex Systems
041101
Considerations on a Mass-Based System Representation

of a Pneumatic Cylinder
M. Brian Thomas and Gary P. Maul
041102
Dynamic Model and Numerical Simulation for Synchronal Rotary

Compressor
Hui Zhou, Zongchang Qu, Hua Yang, and Bingfeng Yu
041103
Evaluation of the Dynamic Transfer Matrix of Cavitating Inducers

by Means of a Simplified Lumped-Parameter Model
Angelo Cervone, Yoshinobu Tsujimoto, and Yutaka Kawata
041104
RANS Analyses of Turbofan Nozzles With Internal Wedge Deflectors

for Noise Reduction
James R. DeBonis
Fundamental Issues and Canonical Flows

041201
Choking Phenomena in a Vortex Flow Passing a Laval Tube: An

Analytical Treatment
Theo Van Holten, Monique Heiligers, and Annemie Jaeken
041202
Pressure Drop in Rectangular Microchannels as Compared With

Theory Based on Arbitrary Cross Section
Mohsen Akbari, David Sinton, and Majid Bahrami
PUBLISHING STAFF
041203
Managing Director, Publishing

P. DI VIETRO
Manager, Journals
C. MCATEER
Production Coordinator
A. HEWITT
Numerical Modeling of Laminar Pulsating Flow in Porous Media

S.-M. Kim and S. M. Ghiaasiaan
041204
Aerodynamic Analysis of a Vehicle Tanker

Ramon Miralbes Buil and Luis Castejon Herrer
041205
Free Surface Model Derived From the Analytical Solution of Stokes

Flow in a Wedge
R. W. Hewson
PUBLICATIONS COMMITTEE
Chair, B. RAVANI
OFFICERS OF THE ASME
President, THOMAS M. BARLOW
Executive Director, THOMAS G. LOUGHLIN
Treasurer, T. D. PESTORIUS
Multiphase Flows
Transactions of the ASME, Journal of Fluids Engineering

ISSN 0098-2202 is published monthly by
The American Society of Mechanical Engineers,
Three Park Avenue, New York, NY 10016.
Periodicals postage paid at New York, NY
and additional mailing offices.
POSTMASTER: Send address changes to Transactions of the
ASME, Journal of Fluids Engineering, c/o THE AMERICAN
SOCIETY OF MECHANICAL ENGINEERS,
22 Law Drive, Box 2300, Fairfield, NJ 07007-2300.
CHANGES OF ADDRESS must be received at Society
headquarters seven weeks before they are to be effective.
Please send old label and new address.
STATEMENT from By-Laws. The Society shall not be
responsible for statements or opinions advanced in papers
or ... printed in its publications B7.1, Par. 3.
COPYRIGHT 2009 by the American Society of
Mechanical Engineers. Authorization to photocopy material for
internal or personal use under those circumstances not falling
within the fair use provisions of the Copyright Act, contact
the Copyright Clearance Center CCC, 222 Rosewood Drive,
Danvers, MA 01923, tel: 978-750-8400, www.copyright.com.
Request for special permission or bulk copying should be
addressed to Reprints/Permission Department.
Canadian Goods & Services Tax Registration #126148048.
041301
Numerical and Experimental Investigations of Steam Condensation

in LP Part of a Large Power Turbine
Wodzimierz Wrblewski, Sawomir Dykas, Andrzej Gardzilewicz,
and Michal Kolovratnik
041302
The Effects of Inlet Geometry and Gas-Liquid Mixing on Two-Phase

Flow in Microchannels
M. Kawaji, K. Mori, and D. Bolintineanu
Techniques and Procedures

041401
A Principle to Generate Flow for Thermal Convective Base Sensors

Thien X. Dinh and Yoshifumi Ogami
Downloaded 03 Jun 2010 to 171.66.16.159. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Contents continued
Journal of Fluids Engineering
Volume 131, Number 4
APRIL 2009
TECHNICAL BRIEFS
044501
Pressurized Flow in a Mesostructured Silica Modified by Silane Groups

Venkata K. Punyamurtula, Aijie Han, and Yu Qiao
The ASME Journal of Fluids Engineering is abstracted and indexed in

the following:
Applied Science & Technology Index, Chemical Abstracts, Chemical Engineering and
Biotechnology Abstracts (Electronic equivalent of Process and Chemical Engineering),
Civil Engineering Abstracts, Computer & Information Systems Abstracts, Corrosion
Abstracts, Current Contents, Ei EncompassLit, Electronics & Communications
Abstracts, Engineered Materials Abstracts, Engineering Index, Environmental
Engineering Abstracts, Environmental Science and Pollution Management, Excerpta
Medica, Fluidex, Index to Scientific Reviews, INSPEC, International Building Services
Abstracts, Mechanical & Transportation Engineering Abstracts, Mechanical Engineering
Abstracts, METADEX (The electronic equivalent of Metals Abstracts and Alloys Index),
Petroleum Abstracts, Process and Chemical Engineering, Referativnyi Zhurnal, Science
Citation Index, SciSearch (The electronic equivalent of Science Citation Index), Shock
and Vibration Digest, Solid State and Superconductivity Abstracts, Theoretical Chemical
Engineering
M. Brian Thomas
Department of Industrial and Manufacturing
Engineering,
Cleveland State University,
2121 Euclid Avenue, SH 221
Cleveland, OH 44115-2214
e-mail: m.thomas.84@csuohio.edu
Gary P. Maul
Department of Industrial, Welding, and Systems
Engineering,
The Ohio State University,
210 Baker Systems,
1971 Neil Avenue,
Columbus, OH 43210
e-mail: maul.1@osu.edu
Considerations on a Mass-Based
System Representation of a
Pneumatic Cylinder
Pneumatic actuators can be advantageous over electromagnetic and hydraulic actuators
in many servo motion applications. The difficulty in their practical use comes from the
highly nonlinear dynamics of the actuator and control valve. Previous works have used
the cylinders position, velocity, and internal pressure as state variables in system models.
This paper replaces pressure in the state model with the mass of gas in each chamber of
the cylinder, giving a better representation of the system dynamics. Under certain circumstances, the total mass of gas in the cylinder may be assumed to be constant. This
allows development of a reduced-order system model. DOI: 10.1115/1.3089533
Keywords: pneumatics, servo motion, control
Introduction and Review
Pneumatic actuators present an alternative to the use of electromagnetic and hydraulic actuators in servo motion applications.
Like electromagnetic actuators i.e., motors, pneumatic actuators
are generally clean and reliable in operation. Like hydraulic actuators, pneumatic actuators may be directly coupled to a payload
without the need for a transmission. Servopneumatics offer high
power-to-weight ratios and can provide a cost benefit as high as
10:1 over competing technologies 1,2. In certain niche applications, pneumatic actuators provide unique advantages over competitive technologies. Unlike electromagnetic actuators, pneumatic actuators are naturally compliant due to the compressibility
of air but are well adapted to wash-down environments. The natural compliance of a pneumatic actuator also makes it advantageous when personnel must be inside the work envelope of the
machine 3.
The principal barrier to the more widespread application of servopneumatic technology in industry has been the problem of control. The dynamics of a pneumatic actuator are highly nonlinear
due to the effects of friction, air compressibility, and time delays
and are compounded by the nonlinear flow characteristics of flow
control valves 3. Linear control methods such as proportionalintegral-derivative PID, while well understood by both engineers
and technicians, perform poorly when coupled to such highly nonlinear systems. With the recent availability of low-cost powerful
microprocessors, computationally expensive nonlinear control
strategies may now be performed in real time. Advanced control
strategies applied to servopneumatics include fuzzy control 4,
sliding mode control 5, model-based adaptive control 1,6, neural networks 7, and pulse-width modulation schemes for solenoid valves 8,9.
The principal contributions of this paper are as follows:
1 Presentation of a mass-based state representation for a
pneumatic cylinder. Previous works use position, velocity,
and the pressure in two chambers of a cylinder as state
variables, as these quantities are readily measured. A massbased representation more closely links the state variables
with the process dynamics.
2 Formal development and verification of the constant mass
assumption. This assumption states that under certain conContributed by the Fluids Engineering Division of ASME for publication in the
JOURNAL OF FLUIDS ENGINEERING. Manuscript received July 25, 2007; final manuscript
received December 22, 2008; published online March 6, 2009. Review conducted by
Joseph Katz.
ditions, the total mass of gas within a pneumatic actuator is

constant. In other words, the instantaneous rate of mass
flow going into one chamber of a pneumatic cylinder under
servo control is equal to the mass flow rate, leaving the
other chamber. This assumption can remain valid even
when the underlying preconditions are not met.
3 A reduced-order system model, based on the constant mass
assumption, for advanced control algorithms. This model
uses three state variablesposition, velocity, and equilibrium positioninstead of four.
The first significant works dealing with the modeling of servopneumatics are two papers by Shearer 10,11, in which he developed a linear model for small motions of a double-rod cylinder
about its midstroke position. While his model is quite restricted in
its practical application, most subsequent researchers have used
Shearers methodology toward model development. Other examples of linear models for pneumatic actuators include Liu and
Bobrow 12, Kunt and Singh 13, and Hamiti et al. 14. Shearers methodology may also be used in developing nonlinear models of pneumatic actuators. A thorough nonlinear model was developed by Richer and Hurmuzlu 15, incorporating not only the
nonlinear dynamics of the cylinder itself but also propagation delays and losses in the air lines connecting valve and cylinder and
the dynamics of the valve. Kawakami et al. 16 compared the
results of linearized and nonlinear models for a pneumatic positioning axis with experimental data, finding that the linearized
model differs significantly from experimental results. More recent
works have focused on the problem of observability for the nonlinear system 8,17.
Servopneumatic systems find applications in both the laboratory and the plant floor. Caldwell et al. 18 discussed the use of
pneumatic actuators in a humanoid robot. The properties of pneumatic muscle actuators make them well suited for this application:
a high power-to-weight ratio, flexibility, and mechanical behavior
similar to those of human muscle tissue. A light-weight internal
combustion engine can power the robot, giving it mobility without
needing an umbilical cord nor heavy banks of batteries. Pneumatics have been considered for a number of more conventional robot
designs. Torque control of individual actuators allows a robot to
have force control at its end effector 19. A survey by Harrison et
al. 20 indicates an industry interest in modular pneumatic robots
for flexible automation applications, particularly assembly, part
handling, and quality control. Finally, this research supported an
effort to incorporate servopneumatic actuators in a food process-
Copyright 2009 by ASME
APRIL 2009, Vol. 131 / 041101-1
Table 2 Simulation system properties, from Ref. 21

Cylinder
Festo DNG-50200-P-A
Servovalve
Position feedback
Flow measurement
Cylinder stroke, L
Cylinder diameter, d
Rod diameter, drod
Payload mass, M
Static friction, Fstat
Dynamic friction, Fdyn
Viscous friction, b
Orientation
Control
Festo MPYE-5a
Gemco Blue Ox 952QD
None
175 mmb
50.0 mm
25 mm
11.0 kgc
21.8 N 4.90 lb
6.9 N 1.55 lb
2.19 N s / m0.125 lbs/ in.
Vertical; rod end down
PI with velocity feed-forward; implemented
through Allen-Bradley M02AE
motor control module
Factory modified to accept a 10 V command signal.

Effective length shortened from 200 mm by an internal position-sensing magnet.
c
Includes estimated mass of piston and rod.
b
Fig. 1 Servopneumatic system
ing environment. Pneumatic actuators were chosen over other

technologies due to their clean operation and compatibility with
equipment wash-down procedures.
2.1 Conventional Representation. The conventional mathematical representation of a pneumatic actuator uses four state
variables x, x, P1, and P2 and three nonlinear differential equations to describe the dynamics of the system. Though previous
works deriving these equations differ on their underlying assumptions, they all use the general approach presented by Shearer in
Ref. 10. This approach begins with the internal energy of volume of gas,
Table 1 Experimental system properties
Cylinder
Servovalve
Position feedback
Flow measurement
Cylinder stroke, L
Cylinder diameter, d
Rod diameter, drod
Payload mass, M
Supply pressure, PS
Orientation
Control
Parker 02.00 HMAU 34A

Festo MPYE-5
Linear potentiometer
TSI 40241 mass flowmeter
304.8 mm 12.0 in.a
50.8 mm 2.0 in.
25.4 mm 1.0 in.
1.6 kg 3.5 lbb
170 kPa 25 psi gauge
Vertical; rod end up or down
Proportional; implemented through LABVIEW
Stroke shortened to accommodate 300 mm length of potentiometer.

Includes estimated mass of piston and rod.
The derivative of the internal energy with respect to time, when

applied to the chambers of a pneumatic cylinder, yields two state
equations,
RT
P 1A 1
1
m
x
A1x + VX,1
A1x + VX,1
RT
P 2A 2
2+
m
x
A2L x + VX,2
A2L x + VX,2
P1 =
System Models
A typical servopneumatic system consists of a pneumatic actuator, one or more control valves to regulate flow, a position sensor,
and a controller. The components may be selected from an extensive list of models and vendors, depending on the particular requirements of the application. This work considers a system consisting of a single-rod, double-acting pneumatic cylinder with a
proportional spool valve controlling the flow of air to and from
the cylinder Fig. 1. Mass flow meters measure flow in and out of
the cylinder. Table 1 details the configuration of the system. Previous work by the authors contains a simulation of a similar system 21; its configuration is detailed in Table 2.
E = cVVT
P2 =
where VX,1 and VX,2 are the excess volumes of gas in either end of
the cylinder and in the plumbing between the cylinder and the
valve. Derivation of Eqs. 2 and 3 is provided in Appendix A.
The third governing equation for the pneumatic actuator addresses
the mechanical dynamics of the system,
x = Mg + P1A1 P2A2 PatmArod bx Ffric/M
The derivation of Eqs. 2 and 3 assumes that the gas undergoes

adiabatic processes in the cylinder. Shearer 10 demonstrated that
for an assumption of isothermal processes, the equation for pressure dynamics is similar, differing only by the factor , the ratio of
specific heats. For dry air at 20 C, this ratio is 1.40. Equation
5 presents the general form for the pressure dynamics of an
isothermal process, which may be compared with the general form
for an adiabatic process, which is presented in Eq. 6,
RT
P
V
P =
m
V
V
RT
P

m
V
P =
V
V
There has been considerable discussion in previous works as to

the best method for modeling a pneumatic cylinder. Shearer
assumed an adiabatic process in Ref. 11; many subsequent researchers have assumed the same, as the heat transfer process is
thought to be significantly slower than the pressure, flow, and
mechanical processes. Exceptions include Pu and Weston 22,
who assumed an isothermal model to aid in their analysis. Richer
and Hurmuzlu 15 used an intermediate term, , which is
bounded by 1.0 and . Based on previous works, these authors
suggest a value of closer to for charging processes and closer
to 1.0 for discharging processes. Using a three-parameter model,
Back and Ohligschlger 23 found a pneumatic cylinder in motion initially behaves adiabatically, but heat flow works to restore
isothermal conditions. Kawakami et al. 16 found that the differ-
041101-2 / Vol. 131, APRIL 2009
Transactions of the ASME
ences between an isothermal and an adiabatic model are not significant in their simulation and recommended using the mathematically less-complicated isothermal model.
The state equationsEqs. 24do not address the compu1
tation of the mass flow rates into each chamber of the cylinder, m
2. These flow rates are dependent on the design, construcand m
tion, and operation of the valve or valves, the upstream and
=
m
C DA
Pup
RT1
2
1
C DA
2/
Pdn
Pup
Pup
RT1
0.6847
where the critical pressure ratio

Pdn
Pup
2
+1
crit
/1
separates the choked and unchoked regimes of flow through the

orifice. For air, the critical pressure ratio is 0.528. The National
Fluid Power Association NFPA presents a less-complicated
model that produces similar results using the same critical pressure ratio as the orifice flow equation 25. Equation 9 uses
English units of measurement, as is the convention in the literature,
QSCFM =
22.48 CV
Pdn
Pdn
Pup PdnPdn
:
Pup
Pup
T
11.22 CV
Pup
Pdn
Pdn
:
Pup
Pup
crit
crit
Less familiar through the literature is the Instrument Society of

America ISA model for compressible flow through a valve. This
model was developed to account for the observation that two
valves with the same flow coefficient Cv can exhibit different flow
characteristics. In addition to the flow coefficient, the ISA model
contains a critical pressure drop ratio, XT. The critical pressure
drop ratio is a reflection of the complexity of the flow path geometry through the valve and changes the point at which choked flow
occurs 26,
QSCFM =
22.67 CV Pup 1
15.11 CV Pup
where
X=
X
3XT
X
:X XT
T
XT
T
Pup Pdn
Pdn
=1
Pup
Pup
:X XT
Pdn
Pup
+1/
Pdn
Pdn
Pup
Pup
crit
Pdn
Pdn
:
Pup
Pup
crit
becomes more attractive than single-orifice models. Using Eq. 7,

one can show that flow through two orifice plates in series as a
function of the total pressure drop is different from that through a
single orifice plate. It is not difficult to find flow conditions in
which one orifice experiences choked flow while the other is unchoked.
2.2 Mass-Based System Representation. Equations 24
describe the conventional state representation of a pneumatic cylinder. An alternate representation for the pneumatic cylinder uses
the mass of gas in each chamber of the cylinder, the cylinder
position, and the cylinder velocity m1, m2, x, and x as state
variables. The derivatives for the mass of gas in either chamber
are found through application of the appropriate flow model of
Eq. 7, Eq. 9, or Eq. 10. The mechanical dynamics of Eq. 4
must be expressed in terms of mass rather than pressure. From the
ideal gas law, pressure in a closed chamber is a function of the
mass of gas in the chamber. Applied to the cylinder,
m1RT
A1x + VX,1
12
m2RT
A2L x + VX,2
13
P1 =
P2 =
The mechanical dynamics of the cylinder may be expressed as
10
11
Equation 10 uses English units of measure, as is the convention

in the literature. Equations 9 and 10 may be converted to mass
flow rates by multiplying the volumetric flow rate by standard
density.
Figure 2 compares the orifice flow model, the NFPA flow
model, and the ISA flow model using three values of XT. Experimental data collected in the course of this research support the use
of the ISA model over the orifice flow modelnote that the ISA
model in Fig. 2 lacks the choked-flow behavior predicted by Eq.
7. If one considers a valve as a series of orifice-type obstructions
to the flow of gas rather than as a single orifice, the ISA model
downstream pressures, and the length and sizing of the plumbing

used in the system. Reference 15 presents a thorough discussion
on the propagation losses and delays that do occur but are often
ignored in systems with short lengths of air line. For the actual
flow through a valve, the literature presents at least three models.
Most researchers use the theoretical model of compressible flow
through an orifice 24, expressed as,
Fig. 2 Valve flow models
APRIL 2009, Vol. 131 / 041101-3
Table 3 Comparative energy rates
Orientation
Incoming flow
Payload direction of travel
Supply pressure
Representative test velocity
Velocities tested
Acceleration
Coulomb friction force
Flow rate source
Potential energy rate, E P = Mgx
Kinetic energy rate, EK = Mx x
Energy rate from Coulomb friction, EF = Ffricx
System studied
System from Ref. 21
Rod end up
Blind end
Up
200 mm/s
201200 mm/s
20 m / s2
N/A
75 SLPM data
3.1 J/s
6.4 J/s
N/A
Rod end down

Rod end
Up
254 mm/s
6.1889 mm/s
1.27 m / s2
6.9 N
136 SLPM simulation
27.4 J/s
3.5 J/s
1.8 J/s
N/A
0.14 J/s
338 J/s
588 J/s
Energy rate from viscous friction, EF = bx2

Incoming gas air energy rate,
d
c PT PAx
Eair = cVVT = m
dt
x =
1
m1RT
m2RT
Mg +
PatmArod bx Ffric
M
x + xT,1 L x + xT,2
14
where xT,1 and xT,2 represent the equivalent cylinder lengths of the
excess volumes in the cylinder and hose,
xT,1 =
VX,1
A1
15
xT,2 =
VX,2
A2
16
The principal benefit of a mass-based system representation is that

it provides a direct link between the dynamics of the control valve
and those of the pneumatic cylinder, as pneumatic servo valves
regulate mass flow rather than pressure. The principal disadvantage of a mass-based representation is that pressure is more
readily measured than flow or mass.
2.3 Constant Mass Assumption. If the sum of the mass of
gas in both chambers of the cylinder is constant, a reduced-order
system model may be derived. Consider a frictionless pneumatic
cylinder moving at a constant velocity, x. Assume that no external
forces act on the cylinder and that constant pressures and temperatures exist in both chambers of the cylinder. While temperature
will change in an adiabatic process, it is assumed here that the
temperature change is not significant 16. Further, assume that
the cylinder does not reach its limits of travel. With these assumptions, the time derivative of the ideal gas law for the blind end of
the cylinder, Eq. 12, becomes
1RT
P1A1x = m
17
2RT
P2A2x = m
18
For the rod end,
The forces acting on the cylinder piston are balanced,

P1A1 = P2A2 + PatmArod
19
Combining Eqs. 1719,

2+e
1=m
m
20
where the error term from atmospheric pressure acting over the
area of the rod, e, is defined as
041101-4 / Vol. 131, APRIL 2009
e=
PatmArodx
RT
21
Using the representative values found in Table 3, the error term e

is found to be 7.6% of the measured mass flow rate. Note that the
error is proportional to velocity. Also note that the cylinder used in
this study has a relatively thick rod, which magnifies the error in
Eq. 21. Supposing that the rod diameter is reduced to 12.7 mm,
the error reduces to 1.9% of the mass flow rate. It is therefore
argued that for many applications, the error term can be ignored,
so that
2
1=m
m
22
Because the mass flow rates are equal in magnitude and opposing
in direction, the sum m1 + m2 must be constant,
m1 + m2 = m
23
This demonstrates that an assumption of a constant mass of gas in

a pneumatic cylinder is valid when the cylinder travels at a constant velocity for many common valve-cylinder configurations. An
energy analysis evaluates the strength of the underlying conditions
for the constant mass assumption. In considering a pneumatic cylinder moving vertically, the rate of energy entering the cylinder as
pressurized gas must equal the rate of energy stored within the
cylinder, plus the energy extracted from the gas through various
means, plus the energy leaving with the air displaced from the
other chamber of the cylinder. For the cylinder under consideration, energy extracted from the gas may a increase the gravitational potential energy of the payload, b increase the kinetic
energy of the payload through acceleration, and c be converted
to heat by overcoming friction in the cylinder.
Table 3 shows the calculated values of these energy rates for
typical experiments conducted in the course of this study. Table 3
also includes values from the earlier study 21, in which the
magnitude of the Coulomb friction was measured. The significance of Table 3 is that the energy rate supplied by the incoming
flow of gas is at least one order of magnitude larger than the
energy rates due to changes in kinetic energy, potential energy,
and overcoming friction. This suggests the constant mass assumption can be a valid system approximation even when the prerequisite conditions are not present.
Numeric simulation, incorporating the effects of friction, pressure losses in the hoses, and external loading on the cylinder,
supports the constant mass assumption. The simulation modeled
the system response to a number of command trajectories, having
Table 4 Experimental trajectory properties

Trajectory
Stroke
Command velocity
Command acceleration
Dwell time
Sampling frequency
Number of samples
Trapezoidal command position, repeating

25275 mm 250 mm total
Variable, from 20 mm/s to 1200 mm/s
Infinite d2x / dt2 =
0.5 s
100 Hz
5 complete cycles, including dwell
Rather, they are associated with the transient spike in the air flow
at the beginning of each cycle, occurring concurrently with the
acceleration of the cylinder.
Table 5 lists the slope of the best-fit lines for the experiments
conducted, and Table 6 lists the R2 values. Figures 5 and 6 chart
these data. While agreement with the constant mass assumption is
good, the general trend is that the error increases as a function of
velocity. Equation 21 predicts this trend, as it contains an error
term proportional to velocity. Also contributing to the error is the
acceleration of the cylinder. As the command velocity increases, a
larger fraction of the duty cycle occurs with the cylinder under
acceleration, violating the preconditions set for the constant mass
assumption.
The constant mass assumption can remain valid even when the
cylinder undergoes stick-slip motion, characterized by short periods of rapid motion with longer periods in which friction prevents
motion. Stick-slip is a common phenomenon in compliant systems
with friction at low command velocities. In Fig. 7a the experimental system is observed to undergo stick-slip behavior on retraction; Fig. 7b shows the constant mass assumption remaining
Fig. 3 Simulated mass flow rates, from Ref. 21, 254 mm/s
command velocity, and 655 kPa supply. a Flow versus time
and b flow versus flow.
velocities from 6 mm/s to 890 mm/s and accelerations to

8.89 m / s2. Figure 3a shows a typical mass flow rate graph from
the simulation as a function of time; Fig. 3b shows the same
data, plotting rod end flow as a function of blind end flow. Regression in Fig. 3b shows that the best-fit line has a slope of
1.103, with an R2 value of 0.987. Similar results were obtained
for all simulations in which the cylinder did not reach either limit
of its stroke.
Experimental verification of the constant mass assumption was
accomplished using two TSI 40241 mass flow sensors Shoreview,
MN capable of measuring air flow to 300 SLPM standard liters
per minute. The configuration of Fig. 1 provided simultaneous
flow measurements for both chambers of the cylinder. Because the
mass flow sensors are unidirectional, they were configured so that
one measured incoming flow to one chamber, while the other
measured outgoing flow from the other. A complete flow history
was obtained by interweaving two data records in which the direction of measurement was reversed. A proportional controller,
implemented through LABVIEW, tracked a repeating trapezoidal
position command signal. Details concerning the trajectories
tested and measurements are given in Table 4.
Figure 4a shows the measured mass flow rates for the cylinder
in the upright position, moving at 200 mm/s and having a nominal
supply pressure of 340 kPa 50 psi gauge. The relationship between the flow rates is apparent. The constant mass assumption
predicts that the flow rate data will fall on a line having a slope of
1. The best-fit line for these data, forced to pass through the
origin, has a slope of 0.9359 and an R2 value of 0.9546, validating the constant mass assumption. While the loops at either end
of Fig. 4b are a visually salient deviation from the constant mass
assumption, they do not represent a significant portion of the data.
Fig. 4 Typical measured mass flow rates, 200 mm/s command

velocity, 340 kPa, and rod end up. a Flow versus time and b
flow versus flow.
APRIL 2009, Vol. 131 / 041101-5
Table 5 Slope of best-fit line, flow from rod end versus flow from blind end of cylinder
Commandvelocity
mm/s
170 kPa,
rod end up
170 kPa,
rod end down
340 kPa,
rod end up
340 kPa,
rod end down
520 kPa,
rod end up
520 kPa,
rod end down
20
50
100
200
300
400
800
1200
0.8857
0.9312
0.9397
0.9013
N/Aa
0.8733
0.8564
0.8561
0.8992
0.9506
0.9431
0.9108
N/Aa
0.8824
0.8770
0.8764
0.9686
0.9976
0.9791
0.9359
0.9076
0.8928
N/Ab
N/Ab
0.9996
1.0117
0.9880
0.9501
0.9233
0.8976
N/Ab
N/Ab
0.9452
1.0082
0.9942
0.9572
0.9338
0.9164
N/Ab
N/Ab
1.0178
1.0172
1.0063
0.9795
0.9443
0.9081
N/Ab
N/Ab
No data collected.
Flow sensors saturated at this combination of pressure and velocity.
valid. The validity of the constant mass assumption with stick-slip

was verified repeatedly in simulation 21, as exemplified in Fig.
8.
2.4 Reduced-Order System Model. The constant mass assumption permits a reduced-order model of the pneumatic cylinder to be developed using position, x, velocity, x, and the cylinders equilibrium position, xeq, as state variables. In a pneumatic
cylinder, equilibrium is established when forces on opposite sides
of the piston face are balanced, P1A1 = P2A2. Following the discussion concerning Eq. 21, error from ignoring the effects of
atmospheric pressure acting over the piston rod has been ignored.
In a mass-based system representation of a closed cylinder with
no leakage and no stiction, static equilibrium is established at
Table 6 R2 values, flow from rod end versus flow from blind
end of cylinder
Command 170 kPa, 170 kPa, 340 kPa, 340 kPa, 520 kPa, 520 kPa,
velocity rod end rod end rod end rod end rod end rod end
down
up
down
up
down
up
mm/s
20
50
100
200
300
400
800
1200
a
0.9465
0.9289
0.9374
0.9215
N/Aa
0.9056
0.9088
0.8927
0.9468
0.9224
0.9201
0.9139
N/Aa
0.9013
0.8754
0.8539
0.9710
0.9674
0.9696
0.9546
0.9409
0.9315
N/Ab
N/Ab
0.9644
0.9678
0.9635
0.9486
0.9392
0.9300
N/Ab
N/Ab
0.9838
0.9808
0.9784
0.9689
0.9612
0.9412
N/Ab
N/Ab
0.9776
0.9784
0.9746
0.9679
0.9473
0.9334
N/Ab
N/Ab
xeq =
m1L + xT,2 m2xT,1

m1 + m2
24
The time derivative of Eq. 24 yields

xeq = L + xT,1 + xT,2
1m 2 m
2m 1
m
m1 + m22
25
For a constant-velocity motion of the cylinder in which the limits

of travel are not reached, the constant mass assumption, stated in
Eqs. 22 and 23, simplifies Eq. 25 to
xeq = L + xT,1 + xT,2
1
m
m
26
The mechanical dynamics of Eq. 14 may also be expressed as a

function of displacement away from equilibrium,
x =
1
Mg kcylx xeq bx PatmArod Ffric
M
27
where kcyl is the nonlinear spring stiffness of the gas in both

chambers of the cylinder. This stiffness may be expressed in terms
of the equilibrium position and is derived in Appendix B. For
small displacements about equilibria not near the limits of travel,
k
cyl
P 1A 1
P 2A 2
+
xeq + xT,1 L xeq + xT,2
28
Table 7 summarizes the state equations for the three models of a

pneumatic cylinder presented in this paper.
For the purposes of simulation, the reduced-order model does
not provide a significant advantage in computational speed, and
No data collected.
Flow sensors saturated at this combination of pressure and velocity.
Fig. 5 Slope of best-fit line through flow data
041101-6 / Vol. 131, APRIL 2009
Fig. 6 Goodness of fit R2 of best-fit line through flow data
Fig. 7 Mass flow rates under stick-slip, 20 mm/s command

velocity, 520 kPa, and rod end up. a Position versus time and
b flow versus flow.
the simplifying assumptions cause it to lose fidelity. Where the

reduced-order model is beneficial is in the control of a servopneumatic system. Adaptive controllers use an internal model of the
plant to update controller gains. The lower-order model lowers the
computational time needed in the controller loop. Previous works
also found that lower-order models can provide better results than
higher-order models due to the destabilizing effects of uncertainties in the higher-order models 27. Similarly, advanced disturbance rejection control ADRC can also benefit from a lowerorder system model. ADRC does not model the system under
control but does require knowledge of the order of the model to be
able to effectively treat unmodeled dynamics as disturbances to
the system 28. The lower-order model is less restrictive to the
ADRC, as it needs less information from the system during operation.
Fig. 8 Simulated mass flow rates under stick-slip, 152 mm/s

peak command velocity, and 340 kPa supply. a Flow versus
time and b flow versus flow.
without external loads or friction and does not reach either limit of
travel. The constant mass assumption was verified by both simulation and experimentation.
Table 7 State equations for pneumatic cylinder models
Conventional model x , x , P1 , P2
RT
P1A1
1
m
x
A1x + VX,1
A1x + VX,1
RT
P2A2
2+
m
x
P2 =
A2L x + VX,2
A2L x + VX,2
1
x = Mg + P1A1 P2A2 PatmArod bx Ffric
M
2 = fvalve
1,m
m
P1 =
Conclusions
Conventional state representation of a pneumatic cylinder uses

four variablesposition, velocity, and two pressuresto model
the system. A model replacing pressure with the mass of gas in
each chamber has been developed. It provides a more direct conceptual link between the command signal to the servovalve and
the dynamics of the cylinder. Consideration of mass flow leads to
the constant mass assumption, which states that for a pneumatic
cylinder under servo control, the instantaneous mass flow rate
going into one chamber of the cylinder is equal to that going out
of the other. It implies the sum of the masses of gas in both
chambers is constant. The underlying assumptions to the constant
mass assumption are that the cylinder moves at a constant velocity
3
4
711
Mass-based model x , x , m1 , m2
x =
1
m1RT
m2RT
Mg +
PatmArod bx
M
x + xT,1 L x + xT,2
14
Ffric
2 = fvalve
1,m
m
711
Reduced-order model x , x , xeq
k x x 1
cyl eq
+ Mg bx PatmArod Ffric
M
M
1
m
xeq = L + xT,1 + xT,2
m
1 = fvalve
m
x =
29
26
711
APRIL 2009, Vol. 131 / 041101-7
1 to the blind end of the cylinder as a function of the command

Fig. 9 Mass flow rate m
voltage and cylinder pressure, supply pressure= 660 kPa
The constant mass assumption permits a reduced-order state

model of the system, in which the mass flow rate input acts on the
cylinders equilibrium position, with classical mechanics describing the cylinders position and velocity. This model is beneficial
for control methodologies such as advanced disturbance rejection
control which require knowledge of the order of the system being
controlled.
A recommended area for future research is in the design of a
pneumatic servovalve. The present paradigm seeks a spool valve
position as a linear function of the command voltage. The valve
used in this study has its own feedback controller to obtain this
behavior. The weakness of this paradigm is that it results in nonlinear flow behavior, exemplified in Fig. 9. This shows the flow
rate into one chamber of the cylinder as a function of both the
static command voltage and the back pressure in the cylinder. An
improved design paradigm for a pneumatic servovalve is to have
the mass flow rate be proportional to the command voltage, thus
removing a significant nonlinearity from the servopneumatic system.
Acknowledgment
The authors wish to acknowledge the Nestl R&D Center, Inc.,
in Marysville, OH, for their support of this project.
Nomenclature
A
b
CD
c p, c V
CV
d
E
F
Fdyn
Fstat
k
L
M
m
area
coefficient of viscous damping
discharge coefficient
specific heats
flow coefficient
diameter
energy
force
Coulomb friction
static friction stiction
stiffness or gain
stroke length
combined mass of payload, piston, and rod
gas mass
041101-8 / Vol. 131, APRIL 2009
P
QSCFM
R
T
V
X
x
Subscripts
1
2
air
atm
crit
cyl
dn
eq
fric
rod
S
spring
T
up
valve
X
pressure
volumetric flow rate, SCFM
gas constant
temperature
volume
pressure drop ratio
cylinder position
ratio of specific heats
density
cylinder blind end

cylinder rod end
air
atmosphere
critical
cylinder
downstream
equilibrium
friction
cylinder rod
supply
spring
equivalent with tubing
upstream
valve
excess
Appendix A
The state equation for a gas, also known as the ideal gas law,
P = RT
A1
The internal energy contained in a homogenous volume of gas,

E = cVVT
A2
In an adiabatic process, the time rate of change in the internal

energy is equal to the rate of energy added to the control volume
by the incoming gas flow, less the rate of work the control volume
performs on the cylinder piston,
d
c PT PV
cVVT = m
dt
A3
P1A1xeq + VX,1 P1 + P1A1xeq + x + VX,1 1

=
+ k1x2
2
1
1
B7
Applying Eq. A1 to Eq. A3,
d cV
c PT
PV + PV = m
dt R
A4
The universal gas constant R, the specific heats of a gas, c P and

cV, and the ratio of specific heats, , are related,
R c P cV
cP
cV
A6
P1A1
P1A1
P1A1 VX,1
xeq +
x +
x
x
x
A1
1
k1x
2
B8
After substitution of Eq. B6,
P1A1 = k1 xeq + xT,1 x

3
2
B9
Substituting Eqs. 12 and 15 into Eq. B9,
A7
Equation A7 is the general form of the pressure dynamics of a

volume of gas. Applied to either chamber of the pneumatic cylinder,
k =
1
m1RT
xeq + xT,1 xeq + xT,1
3
x
2
B10
RT
P 1A 1
1
m
x
A1x + VX,1
A1x + VX,1
A8
A similar analysis may be performed on the rod end of the cylinder. In this case, the deflection is x as it acts in the opposite
direction on chamber 2 than on chamber 1,
RT
P 2A 2
2+
m
x
A2L x + VX,2
A2L x + VX,2
A9
k = F = P2A2
2
x
x
P1 =
P2 =
0 = P 1A 1 +
A5
For air, = 1.4. Expanding Eq. A4, with substitution of Eqs.

A5 and A6,
RT
P

m
V
P =
V
V
Multiplying by 1 / x and eliminating like terms,
where VX,1 and VX,2 are the excess volumes of gas in either end of
the cylinder and in the plumbing between the cylinder and the
valve.
B11
P2A2L xeq + VX,2 P2 + P2A2L xeq x + VX,2

=
1
1
1
+ k2x2
2
Appendix B
Assuming adiabatic compression and expansion of the gas in
the cylinder, the initial equilibrium energy is equal to the displaced internal energy state of the gas plus the potential energy
stored in the compressed gas,
Eeq = Eair + Espring
B1
0 = P 2A 2 +
+
Eeq,1 =
P1A1xeq + VX,1
1
Eair,1 =
P1 + P1A1xeq + x + VX,1
1
B4
The potential energy stored by the gas in the blind end, acting as
a spring, is given by
1
Espring,1 = 2k1x2
B5
where k1 is the effective nonlinear spring rate for the blind end,
defined as the change in force per unit displacement.
k = F = P1A1
1
x
x
Combining Eqs. B1 and B3B5,
B6
B13
k =
2
m2RT
L xeq + xT,2 L xeq + xT,2 +
B3
When the cylinder is displaced from equilibrium, the energy state

of the gas changes,
3
P2A2 = k2 L xeq + xT,2 + x
2
B2
Applying the relationships between the specific energies of a gas

and its gas constant,
P2A2
P2A2
P2A2 VX,2
L xeq
x +
x
x
x
A2
1
k2x
2
The initial energy of the gas in the blind end is

Eeq,1 = cVm1T
B12
3
x
2
B14
B15
The total stiffness of the pneumatic cylinder is the sum of the

stiffness of both chambers,
k = k + k = RT
cyl
1
2
m1
xeq + xT,1 xeq + xT,1
3
x
2
m2
L xeq + xT,2 L xeq + xT,2 +
3
x
2
B16
Equation B16 shows that the stiffness of the pneumatic cylinder

is highly dependent not only on the position of the cylinder, x, but
also on the displacement away from equilibrium x. When the
cylinders equilibrium position is not near the limits of travel,
small displacements in Eq. B16 may be neglected,
APRIL 2009, Vol. 131 / 041101-9
k RT
cyl
m1
m2
+
B17
Equations 12 and 13 may be substituted into Eq. B17 to

express stiffness in terms of cylinder pressures,
k
cyl
P 1A 1
P 2A 2
+
B18
References
1 Bobrow, J. E., and Jabbari, F., 1991, Adaptive Pneumatic Force Actuation and
Position Control, ASME J. Dyn. Syst., Meas., Control, 113, pp. 267272.
2 Richardson, R., Plummer, A. R., and Brown, M. D., 2001, Self-Tuning Control of a Low-Friction Pneumatic Actuator Under the Influence of Gravity,
IEEE Trans. Control Syst. Technol., 92, pp. 330334.
3 Moore, P. R., and Pu, J. S., 1996, Pneumatic Servo Actuator Technology,
IEE Colloq. on Actuator Technology: Current Practice and New Developments
110, pp. 3/13/6.
4 Shih, M., and Ma, M.-A., 1998, Position Control of a Pneumatic Cylinder
Using Fuzzy PWM Control Method, Mechatronics, 8, pp. 241253.
5 Richer, E., and Hurmuzlu, Y., 2000, A High Performance Pneumatic Force
Actuator System: Part II-Nonlinear Controller Design, ASME J. Dyn. Syst.,
Meas., Control, 122, pp. 426434.
6 McDonell, B. W., and Bobrow, J. E., 1993, Adaptive Tracking Control of an
Air Powered Robot Actuator, ASME J. Dyn. Syst., Meas., Control, 115, pp.
427433.
7 Gross, D. C., and Rattan, K. S., 1998, An Adaptive Multilayer Neural Network for Trajectory Control of a Pneumatic Cylinder, IEEE International
Conference on Systems, Man, and Cybernetics, San Diego, CA, Vol. 2, pp.
16621667.
8 Barth, E. J., Zhang, J., and Goldfarb, M., 2003, Control Design for Relative
Staility in a PWM-Controlled Pneumatic System, ASME J. Dyn. Syst.,
Meas., Control, 1253, pp. 504508.
9 Shih, M.-C., and Ma, M.-A., 1998, Position Control of a Pneumatic Cylinder
Using PWM Control Method, Mechatronics, 8, pp. 241253.
10 Shearer, J. L., 1956, Study of Pneumatic Processes in the Continuous Control
of Motion With Compressed AirI, Trans. ASME, 78, pp. 233242.
11 Shearer, J. L., 1956, Study of Pneumatic Processes in the Continuous Control
of Motion With Compressed AirII, Trans. ASME, 78, pp. 243249.
12 Liu, S., and Bobrow, J. E., 1988, An Analysis of a Pneumatic Servo System
and Its Application to a Computer-Controlled Robot, ASME J. Dyn. Syst.,
041101-10 / Vol. 131, APRIL 2009
Meas., Control, 110, pp. 228235.

13 Kunt, C., and Singh, R., 1990, A Linear Time Varying Model for On-Off
Valve Controlled Pneumatic Actuators, ASME J. Dyn. Syst., Meas., Control,
112, pp. 740747.
14 Hamiti, K., Voda-Besanon, A., and Roux-Boisson, H., 1996, Position Control of a Pneumatic Cylinder Under the Influence of Stiction, Control Eng.
Pract., 48, pp. 10791088.
15 Richer, E., and Hurmuzlu, Y., 2000, A High Performance Pneumatic Force
Actuator System: Part INonlinear Mathematical Model, ASME J. Dyn.
Syst., Meas., Control, 122, pp. 416425.
16 Kawakami, Y., Akao, J., Kawai, S., and Machiyama, T., 1988, Some Considerations on the Dynamic Characteristics of Pneumatic Cylinders, J. Fluid
Control, 192, pp. 2236.
17 Bigras, P., 2005, Sliding-Mode Observer as a Time-Variant Estimator for
Control of Pneumatic Systems, ASME J. Dyn. Syst., Meas., Control, 127, pp.
499502.
18 Caldwell, D. G., Badihi, T. D., and Medrano-Cerda, G. A., 1998, Pneumatic
Muscle Actuator Technology a Light Weight Power System for a Humanoid
Robot, IEEE International Conference on Robotics and Automation, Leuven,
Belgium, Vol. 4, pp. 30533058.
19 McDonell, B. W., and Bobrow, J. E., 1998, Modeling, Identification, and
Control of a Pneumatically Actuated Robot, IEEE Trans. Rob. Autom., 145,
pp. 124129.
20 Harrison, R., Weston, R. H., Moore, P. R., and Thatcher, T. W., 1987, A Study
of Application Areas for Modular Robots, Robotica, 5, pp. 217221.
21 Thomas, M. B., 2003, Advanced Servo Control of a Pneumatic Actuator,
Ph.D. thesis, Ohio State University, Columbus, OH.
22 Pu, J. S., and Weston, R. H., 1990, Steady State Analysis of Pneumatic Servo
Drives, Proc. Inst. Mech. Eng., Part C: J. Mech. Eng. Sci., 204, pp. 377387.
23 Back, W., and Ohligschlger, O., 1989, A Model of Heat Transfer in Pneumatic Chambers, J. Fluid Control, 20, pp. 6178.
24 Andersen, B. W., 1976, The Analysis and Design of Pneumatic Systems, Robert E. Kreiger, New York.
25 Hong, I. T., and Tessmann, R. K., 1996, The Dynamic Analysis of Pneumatic
Systems Using HyPneu, International Fluid Power Exposition and Technical
Conference, Chicago, IL.
26 Thomas, J. H., 2000, Proper Valve Size Helps Determine Flow, Control
Engineering Online, http://www.manufacturing.net/ctl/article/CA188679
27 Bobrow, J. E., and Jabbari, F., 1991, Adaptive Pneumatic Force Actuation and
Position Control, ASME J. Dyn. Syst., Meas., Control, 113, pp. 267272.
28 Gao, Z., 2006, Active Disturbance Rejection Control: A Paradigm Shift in
Feedback Control Design, Proceedings of the American Control Conference,
Minneapolis, MN, pp. 23992405.
Dynamic Model and Numerical

Simulation for Synchronal Rotary
Compressor
Hui Zhou
e-mail: zhouhui1125@gmail.com
Zongchang Qu
Hua Yang
Bingfeng Yu
Institute of Compressor,
Xian Jiaotong University,
Xian 710049, China
The synchronal rotary compressor (SRC) has been developed to resolve high friction and
severe wear that usually occur in conventional rotary compressors due to the high relative velocity between the key tribo-pairs. In this study, the working principle and structural characteristics of the SRC are presented first. Then, the kinematic and force models
are established for the key componentscylinder, sliding vane, and rotor. The velocity,
acceleration, and force equations with shaft rotation angle are derived for each component. Based on the established models, numerical simulations are performed for a SRC
prototype. Moreover, experiments are conducted to verify the established models. The
simulated results show that the average relative velocity between the rotor and the cylinder of the present compressor decreases by 8082% compared with that of the conventional rotary compressors with the same size and operating parameters. Moreover, the
average relative velocity between the sliding contact tribo-pairs of the SRC decreases by
9394.3% compared with that of the conventional rotary compressors. In addition, the
simulated results show that the stresses on the sliding vane are greater than those on the
other components. The experimental results indicate that the wear of the side surface of
the sliding vane is more severe than that of the other components. Therefore, special
treatments are needed for the sliding vane in order to improve its reliability. These
findings confirm that the new SRC has lower frictional losses and higher mechanical
efficiency for its advanced structure and working principle. DOI: 10.1115/1.3089534
Keywords: synchronal rotary compressor, dynamic model, motion, forces
Introduction
Rotary type compressors are used more widely than reciprocating compressors in refrigeration and air-conditioning systems because they have advantages such as less components, simpler
structure, better dynamic equilibrium characteristics, and higher
reliability. However, owing to the high relative velocity between
the rotor, the cylinder, and the sliding vane of the conventional
rotary compressors, the frictional loss is high and then limits their
performances. Much work has been devoted to reducing the friction and wear of the conventional rotary compressors.
Jeon and Lee 1 discussed the tribological characteristics of
sliding surfaces between the vane and the flange in the rotary
compressor and conducted experiments to evaluate the effects of
different hard coatings on the compressor performance. Suh et al.
2 and Demas and Polycarpou 3 conducted compressor tribological experiments by using a high tribometer to explore the contact conditions. Oh et al. 4 and Oh and Kim 5 investigated the
friction and wear characteristics of sliding surfaces with various
coatings for a rotary compressor. They conducted experiments
with different working media and pressures in order to find out the
best surface treatment method to improve the tribological properties. Huang and Shiau 6 described a method to improve the
rotary vane compressor performance by employing extended rods
on both edges of each vane and guide slots on both cover plates.
In addition, Huang and Li 7 also established an optimum model
and found the optimal tolerance allocation for a vane rotary compressor. Cai et al. 8 proposed a perfect profile of the vane tip for
a rotary vane compressor to reduce the friction and wear of the
vane. Ooi 9 predicted that a 50% reduction in mechanical loss
Contributed by the Fluids Engineering Division of ASME for publication in the
JOURNAL OF FLUIDS ENGINEERING. Manuscript received October 24, 2007; final manuscript received January 9, 2009; published online March 6, 2009. Assoc. Editor:
Chunill Hah.
can be achieved by employing a multivariable, direct search, constrained optimization technique. Lee and Oh 10 conducted tribological experiments under various operation conditions and proposed an optimum initial surface roughness to prolong the wear
life of sliding surfaces.
The research mentioned above on reducing the friction and
wear is mostly focused on surface treatment and structural optimization for rotary compressors. However, the research cannot
reduce the friction and wear caused by high relative velocity between the key moving parts in the conventional rotary compressors radically. This paper develops an innovative synchronal rotary mechanism, in which the high friction and severe wear
caused by high relative velocity in the conventional rotary compressors are effectively reduced by the method of the cylinder and
the rotor rotating around their own axis synchronously.
Recently, we have performed the research on kinematics characteristics of the vane 11 and leakage model 12 of the SRC.
However, no studies have been published on the whole system
dynamic model of the SRC. The purpose of this study is to investigate the performance of the SRC by establishing the dynamic
model and testing the dynamic performance.
Working Principle and Structural Characteristics
Figure 1 shows the working principle and structural characteristics of the proposed SRC. It can be seen from Fig. 1 that the
machines main components consist of a rotor, a cylinder, a sliding vane, a shaft, and two end covers.
The outside wall of the rotor and the inside wall of the cylinder
are tangent to each other, and they form the working chamber
together. The rotor is driven by the shaft, and it rotates around its
center OR at the angular velocity . The cylinder is driven by the
connecting sliding vane, and it rotates around its own center OC at
the angular velocity C. The sliding vane serves as a connector of
APRIL 2009, Vol. 131 / 041102-1
Fig. 1 Cross-section diagram of a synchronal rotary compressor
the rotor and the cylinder, and it separates the working chamber
into a suction chamber and a compression chamber.
We define the starting point as the tangent point, as the shaft
rotation angle, and as the discharge angle. The typical stages of
the working process are shown in Fig. 1. It can be seen in Fig. 1
that when = 0, the discharge stage is finished. When 0 ,
the compression chamber volume decreases with the increase in
, and the gas is compressed; meanwhile, the suction chamber
volume increases and the gas is sucked in. During this stage, the
suction chamber volume equals the compression chamber volume
when = . When = , the discharge process starts and the gas
is expelled from the compression chamber. When = 2, the discharge is completed; the compression chamber closes and the suction chamber is full of fresh gas again. Thus, one operating cycle
is completed.
Because the relative velocity between the sliding contact tribopairs of the SRC is much smaller than that of the conventional
rotary compressors, the SRC has many advantages such as lower
friction and wear, easier to achieve seals, and lower vibration and
noise level.
3 Kinematic Model for the Sliding Vane and the Cylinder

3.1 Velocity and Acceleration of the Sliding Vane. The sliding vane driven by the rotor exhibits a composite planar motion. It
rotates with the rotor at a uniform speed, and it reciprocates along
the vane slot simultaneously. In order to analyze the kinematics
characteristics of the sliding vane, a rotational coordinate system
ORXY, which rotates synchronously with the rotor, is established.
The velocity triangle of the sliding vane is shown in Fig. 2.
According to the theorem of velocity composition, the absolute
velocity vSG of the barycenter G equals the vector sum of the
convected velocity vSGe and the relative velocity vSGr,
vSG = vSGe + vSGr
The displacement of the barycenter G relative to the rotor center

OR is given as
x = r RS = e cos + R + RS2 e2 sin2 r RS
2
Therefore, the relative velocity of the barycenter G to the rotor
center OR can be expressed as
vSGr = x =
e2 sin cos
dx
= e sin
dt
R + RS2 e2 sin2
The convected velocity of the barycenter G is given by

vSGe = G = e cos + R + RS2 e2 sin2 RS L/2
4
According to the theorem of acceleration composition, the absolute acceleration aSG of point G equals the vector sum of the
convected acceleration aSGe, the relative acceleration aSGr, and the
Coriolis acceleration aSGk,
aSG = aSGe + aSGr + aSGk
where
aSGe = 2G = 2 e cos + R + RS2 e2 sin2 + RS L/2
aSGr = x = d2x/dt2
= 2e cos 2e2cos 2R + RS2 e2 sin2 0.5
+ 1/4e2 sin22R + RS2 e2 sin2 1.5
aSGk = 2 vSGr
3.2 Velocity and Acceleration of the Cylinder. Unlike the
conventional rotary compressors, the cylinder of the SRC is not
fixed but driven by the sliding vane and rotates synchronously
with the rotor around its own center. In order to analyze the kinematics characteristics of the cylinder, a rotational coordinate system OCXY, which rotates synchronously with the cylinder, is
established.
Figure 3 shows the velocity triangle of the cylinder. Because
the hinge joint hole center of the cylinder coincides with the head
center OS of the sliding vane, the absolute velocity vC for the
hinge joint hole center of the cylinder equals the absolute velocity
vS for the point OS on the sliding vane. On the basis of the above
motion model of the sliding vane, the absolute velocity vC at shaft
rotation angle can be expressed as
vC = vS = vSe + vSr
where
Fig. 2 Velocity triangle of the sliding vane
041102-2 / Vol. 131, APRIL 2009
vSe = = e cos + R + RS2 e2 sin2
Fig. 5 Comparison of average relative velocity between the

sliding contact tribo-pairs
Fig. 3 Velocity triangle of the cylinder
vSr = x =
e2 sin cos
dx
= e sin
dt
R + RS2 e2 sin2
Accordingly, the cylinder angular velocity C and angular acceleration C can be expressed as
C = vC/R = vSe2 + vSr2/R

C = dC/dt
C =
3.3 Kinematic Simulation and Analysis. Based on the kinematic models established above, simulations of the relative velocity between the key moving components can be performed. Figure
4 shows the comparison of average relative velocity between the
rotor and the cylinder of the SRC and the conventional rotary
compressors with the same design and operating parameters. In
Fig. 4, it can be seen that the average relative velocity between the
rotor and the cylinder of the SRC decreases by approximately
8082% compared with that of the conventional rotary compressors. As a result, the disadvantages caused by high relative velocity between the rotor and the cylinder such as severe wear, difficulty to control the meshing clearance between the rotor and the
cylinder, and difficulty to achieve seals can be overcome.
For a SRC and a rolling piston compressor RPC, the vane and
the rotor are in sliding contact and the large frictional loss exists
between them. For a rotary vane compressor RVC, the vane tips
and the cylinder are in sliding contact and the large frictional loss
exists between the vane tips and the cylinder. The frictional loss
would decrease if the relative velocity between these sliding contact tribo-pairs could decrease effectively. Figure 5 shows the
comparison of average relative velocity between the sliding contact tribo-pairs of the SRC and the conventional rotary compressors with the same design and operating parameters. It can be seen
in Fig. 5 that the average relative velocity between the sliding
contact tribo-pairs is 0.8 m/s for the SRC, 11.3 m/s for the RPC,
and 13.6 m/s for the RVC. Thus, the average relative velocity
between the sliding contact tribo-pairs of the SRC decreases by
92.894.1% compared with that of the conventional rotary compressors. Therefore, the disadvantages caused by high relative velocity between the sliding contact tribo-pairs such as large frictional loss and severe wear could be overcome completely.
Force Model for the Key Components
The force models established in this research include not only

the principal forces but also the often neglected component weight
and the frictional forces from the oil film. In the models, we
assume the following:
1 The pressure pulsation of the intake and discharge processes is ignored, which means that the intake and discharge pressures are regarded as constants.
2 The effect of oil film thickness change on friction coefficient is neglected; i.e., the friction coefficient is considered
constant.
3 Gas leakage and energy loss in gas flow are ignored.
4.1
Force Model for the Cylinder.
4.1.1 Forces Acting on the Cylinder. Figure 6 shows the

forces acting on the cylinder. The forces are defined to be positive
when their direction is the same as the positive direction of the X
or Y axis of the rotational coordinate system OCXY. The moments
are defined to be positive when they drive the cylinder to rotate in
the same direction with .
Fig. 4 Comparison of average relative velocity between the
rotor and the cylinder
4.1.1.1 Gas force FCg and moment M Cg. The working chamber is separated into a suction chamber and a compression chamber by the sliding vane, and the gas force FCg acting on the cylAPRIL 2009, Vol. 131 / 041102-3
LCSn = OC P = e sin
LCS = OS P = e cos + R + RS2 e2 sin2 + e cos
14
When 0 , the sign of LCSn in Eq. 14 is +; when

2, the sign of LCSn is .
The driving moment M CS to the cylinder by the sliding vane is
given by
M CS = FCSn LCSn + FCS LCS
15
FCSn and FCS can be decomposed along the X and Y directions,

respectively, and the included angle of the force FCS and the X
axis is . The components can be obtained by
FCSnx = FCSn sin = FCSn 1 cos2
FCSny = FCSn cos
FCSx = FCS cos
16
FCSy = FCS sin = FCS 1 cos2

where is defined as
= arccose2 + 2 R + RS2/
Fig. 6 Forces acting on the cylinder
2e e cos + R + RS2 e2 sin2
inder is caused by the pressure difference in the suction chamber

and the compression chamber. As shown in Fig. 3, ORBOC = .
The gas force is given by
FCg = 2HR sin
Pd Ps
2
10
where Ps is the suction chamber pressure at the shaft rotation

angle , which equals the constant suction pressure Ps, and Pd
is the corresponding compression chamber pressure that may be
calculated by
Pd = Ps V/Vdn
11
In Eq. 11, V is the total volume of the working chamber and

Vd is the compression chamber volume at shaft rotation angle
.
The gas force components in X and Y directions can be given
by
FCgx = FCg sin/2
12
FCgy = FCg cos/2
According to the geometrical relationship shown in Fig. 3,

ORB = , can be expressed as
= arccose e cos2 + 2 cos R e sin /2eR

13
2
4.1.1.3 Forces from lubricating oil at the meshing point. Lubricating oil exists in the meshing clearance of the rotor and the
cylinder. Because of the relative motion between the rotor and the
cylinder, there are normal fluid hydrodynamic forces FCMn and
tangential frictional forces FCM from the lubricating oil acting on
the cylinder. FCMn is transmitted to the shaft and causes moment
of flexion, which can be ignored because it is much smaller than
that of the other moments on the shaft.
According to the lubrication theory, the tangential frictional
force FCM and frictional moment M CM at the meshing point can
be obtained by
FCM =
or2H C
o
or2RH C
M CM =
o
18
4.1.1.4 Frictional force from the head of the sliding vane. The
sliding vane rotates in the hinge joint hole of the cylinder and
causes the sliding frictional force FCSf , which is given by
FCSf = f CS FCS = f CS FCSn2 + FCS2
19
4.1.1.5 Weight GC . The weight of the cylinder due to gravity

is given as
G C = m Cg
20
1/2
The gas force crosses the center OC of the cylinder, and thus the
moment caused by gas at the center OC is zero.
4.1.1.2 Driving force FCS and moment M CS from the sliding
vane. The driving force on the cylinder by the sliding vane can be
divided into two components, namely, the normal component FCSn
and the tangential component FCS, which is normal and tangential to the cylinder surface, respectively. The arms of FCSn and
FCS at the cylinder center OC are given as
041102-4 / Vol. 131, APRIL 2009
17
4.1.1.6 Bearing housing opposing force FCB, frictional force

FCBf , and frictional resistance moment M CBf . The gas force, the
driving force by the sliding vane, the frictional forces, and the
weight of the cylinder will be transferred to the cylinder bearing
ultimately. Therefore, the bearing housing opposing force components in X and Y directions can be written as
FCBx = FCgx + FCSx + FCSnx
FCBy = FCgy + FCSy + FCSny + GC
21
Thus the resultant force FCB is given by

end part by the vane slot, and FSRf2 is the corresponding frictional
force.
Both FSR1 and FSR2 are normal to the vane side surface. The
moments at the head center OS of the sliding vane caused by FSR1
and FSR2 are given by
M SR1 = FSR1 LSR1 = FSR1 e cos + R + RS2 e2 sin2 r
27
M SR2 = FSR2 LSR2 = FSR2 L RS
FSRf1 and FSRf2 act along the contact surface of the sliding vane
and the vane slot, and their directions are opposite to the velocity
vSr. FSRf1 and FSRf2 are given by
FSRf1 = f SR FSR1
FSRf2 = f SR FSR2
In Eq. 28, when 0 , the signs of FSRf1 and FSRf2 are .

When 2 the signs are +.
The moments at the point OS by FSRf1 and FSRf2 can be obtained by
Fig. 7 Forces acting on the sliding vane
FCB = FCBx2 + FCBy21/2
22
Accordingly, the frictional force and moment acting on the cylinder are given by
FCBf = f CBFCB
M CBf = FCBf RB
23
4.1.2 Differential Equation of Rotation of the Cylinder. According to the differential equations of rotation of a rigid body
with a fixed axis, the differential equation of rotation of the cylinder can be expressed as
J C C =
M F = M
i
CS
+ M CM + M CBf
24
4.2 Force Model for the Sliding Vane. The forces acting on
the sliding vane are the gas force, the contact forces and frictional
forces from the vane slot, the constrained forces from the cylinder,
and the inertia forces due to the vane motion, etc. Figure 7 shows
all the forces acting on the sliding vane. The forces are assumed to
be positive when they deviate from the rotor, and the moments are
assumed to be positive when they inhibit the sliding vane rotation.
4.2.1.1 Gas force FSg . The pressure difference between the
suction chamber and the compression chamber exerts gas force to
the sliding vane. The gas force is given by
FSg = Pd Ps l H
= Pd Ps e cos + R2 e2 sin2 r H
25
Thus, the moment M Sg at the head center OS of the sliding vane
by the gas force can be given as
26
4.2.1.2 Constrained force FSC by the cylinder. The constrained force FSC on the sliding vane by the cylinder can be
divided into the normal component FSCn and the tangential component FSC. FSCn and FSC are equal and opposite to the driving
forces FCS and FCSn on the cylinder by the sliding vane, respectively.
4.2.1.3 Contact forces FSR1, FSR2 and frictional forces FSRf1,
FSRf2 from the rotor. As shown in Fig. 7, FSR1 is the contact force
caused by the vane slot up margin, and FSRf1 is the corresponding
frictional force. FSR2 is the contact force acting on the sliding vane
M SRf1 = FSRf1 b/2

M SRf2 = FSRf2 b/2
29
4.2.1.4 Frictional force FSCf on the sliding vane head by the

cylinder. This force is equal and opposite to the FCSf on the cylinder.
4.2.1.5 Inertia forces and moments. The inertia forces caused
by the vane motion include the convected inertial force FSe, the
relative inertia force FSr, and the Coriolis inertial force FSk. These
inertia forces cross the barycenter of the sliding vane and can be
written as
FSe = mC aSGe
where JC is the rotary inertia of the cylinder.
M Sg = FSg LSg = FSg R2 e2 sin2 r/2 + RS
28
FSk = mC aSGk
30
FSr = mC aSGr
The corresponding moments caused by the inertia forces may be
expressed as
M Se = 0
M Sk = FSk LSk = FSk L/2 RS
31
M Sr = 0
A set of simultaneous equations including the normal and tangential force balance equations of the sliding vane, the moment
balance equation for the head of the sliding vane, and the differential equation of the cylinder rotation are obtained as
FCSn + FSe + FSr FSR1 f SR FSR2 f SR = 0
FCS + FSa + FSk + FSR1 FSR2 = 0
FSgLSg + FSkLSk + FSR1 LSR1
FSR2 LSR2 FSR1 f SR
32
b
b
FSR2 f SR = 0
2
2
FCSn LCSn + FCS LCS = JCC

This equation set can be expressed in the following form of matrix:
APRIL 2009, Vol. 131 / 041102-5
M RS1 = FRS1 LRS1 = FRS1 r

M RS2 = FRS2 LRS2
= FRS2 e cos + R + RS2 e2 sin2 L + RS
35
M RSf1 = FRSf1 b/2
M RSf2 = FRSf2 b/2
4.3.1.3 Frictional forces from the end covers and forces from
the lubricating oil at the meshing point. There is lubricating oil in
the gap between the rotor and the two cylinder end covers and in
the meshing clearance of the rotor and the cylinder. The relative
motion between the rotor and the cylinder causes viscous frictional forces and moments. Based on the lubrication theory, the
viscous frictional force and frictional moment acting on the rotor
by the oil between the rotor and the two cylinder end covers can
be obtained as
FRCf =
2C r4 rin4
Or rin
M RCf =
f SR
f SR
LCS LCSn
0
LSR1 f SR b/2 LSR1 f SR b/2
FSg FSk
FSe FSr
J C C
FSgLSg FSkLSk
FCSn
FSR2
33
or2HC
O
or3HC
M RM =
O
FSR1
4.3.1.1 Gas force FRg. The gas force acting on the rotor is
caused by the pressure difference between the suction chamber
and the compression chamber. The gas force FRg crosses the rotor
center OR and is given by
34
The moment M Rg created by the gas force is zero because the gas
force crosses the center of the rotor.
4.3.1.2 Reaction forces FRS1, FRS2 and frictional forces FRSf1,
FRSf2 from the sliding vane. These four forces acting on the rotor
from the sliding vane are force couples with the corresponding
FSR1, FSR2, FSRf1, and FSRf2 described previously in the force
model for the sliding vane.
The moments at the rotor center OR caused by these four forces
are given by
37
4.3.1.4 Resistance moment M z . The resistance moment is

composed of all the force moments acting on the rotor and is
given by
M z = M RS1 + M RS2 + M RSf1 + M RSf2 + M RCf + M RM
4.3 Force Model for the Rotor. In the SRC, the rotor rotates
at a uniform angular velocity . The forces and moments acting
on the rotor are shown in Fig. 8, including the gas force, the
reaction forces and frictional forces on the vane slot from the
sliding vane, the frictional forces from the two end covers, the
normal dynamic flow force and the tangential frictional force created by the lubricating oil at the meshing point M of the rotor and
the cylinder, and the driving moment from the shaft.
041102-6 / Vol. 131, APRIL 2009
FRM =
FCS
All the four unknown forces can be obtained by solving the above
matrix. Based on these four forces, all the other unknown forces
acting on the cylinder and the sliding vane could be obtained.
FRg = 2Hr sin Pd Ps

2
36
Likely, the frictional force and moment acting on the rotor by the
oil at the meshing point can be calculated by
Fig. 8 Forces acting on the rotor
C r4 rin4
38
Simulation Results and Discussion
Based on the established models, forces on the cylinder, the

sliding vane, and the rotor can be simulated for a SRC prototype
used in the experiment. We choose 1000 rpm as the rotational
speed. The friction coefficient of each frictional surface depends
on the material, the contact surface roughness, the lubrication, the
relative velocity, etc. In this study, the friction coefficients f SR,
f CS, and f CB are set to be 0.11, 0.12, and 0.008, respectively,
according to the empirical values by Oh et al. 4, Oh and Kim
5, Solzak and Polycarpou 13, and Xu et al. 14 in their tribological experiments.
5.1 Simulation Results for Forces on the Cylinder. Figure 9
shows the simulation results of the forces acting on the cylinder.
From Fig. 9, it can be noted that the gas force and the bearing
force are the principal forces acting on the cylinder and are much
greater than the other forces. They both reach their maximum
values at the shaft rotation angle of 220 deg, where the discharge
process starts. In addition, the bearing loads are composed mainly
of the gas force, and the other forces such as the driving force and
frictional force from the sliding vane can be neglected because
they are much smaller. The simulation results of the cylinder
forces may be helpful to choose proper bearings for the compressor to improve the life and reliability.
5.2 Simulation Results for Forces on the Sliding Vane. Figure 10 shows the simulation results of all the forces acting on the
sliding vane. The forces on the sliding vane are more complex
than those on the other components, for the sliding vane serves as
Fig. 9 Results of forces on the cylinder

Fig. 11 Results of forces on the rotor
a connector that drives the cylinder to rotate and at the same time
is driven by the rotor. In Fig. 10, it can be seen that the principal
forces acting on the sliding vane are the contact forces by the vane
slot, the tangential constrained forces by the cylinder, and the gas
force. The other forces are relatively much smaller, varying over a
range from 38 N to 79 N. Moreover, in Fig. 10, the tangential
constrained force by the cylinder fluctuates from 247.9 N to
237.9 N and changes from negative to positive at the rotation
angle , which not only increases the machine vibration but also
aggravates the fatigue damage of the sliding vane. It is suggested
that the inertia of the cylinder should be reduced in order to mitigate the negative impacts on the sliding vane.
5.3 Simulation Results for Forces on the Rotor. Figure 11
shows the simulation results of the forces acting on the rotor.
From Fig. 11, it can be noted that the gas force is the greatest in
all forces, and its maximum value occurs at the shaft rotation
angle of 220 deg, where the discharge process just starts. All the
other forces are less than 346 N. In addition, it can be seen that the
resistant moment consists mainly of the acting forces by the sliding vane, and the frictional moments are relatively smaller.
Model Verification
6.1 Test Bench. In order to verify the established models and

investigate the general performances of the SRC, a compressor
test bench was set up. On the test bench, the general performances
and the input torque of the SRC were measured. Furthermore, the
investigation of sliding vane wear conditions was carried out.
Figure 12 shows the schematic of the test bench. The test bench
consists of the main parts, such as a second refrigerant calorimeter, a compressor, a condenser, an electronic expansion valve
EEV, a torque meter, and a motor. The structural parameters and
the operating parameters of the prototype used in the experiment
are listed in Table 1.
6.2
Model Verification.
6.2.1 Indirect Verification of the Forces on the Sliding Vane.

The sliding vane is one of the critical components that affect the
compressor leakage and reliability directly. Because of the very
Fig. 10 Results of forces on the vane
Fig. 12 Test bench
APRIL 2009, Vol. 131 / 041102-7
Table 1 Parameters for the SRC prototype

Structural parameters
Diameter of the cylinder m
Diameter of the rotor m
Height of the cylinder m
Thickness of the sliding vane m
Radial length of the sliding vane m
Eccentricity m
Operating parameters
Evaporating temperature k
Suction pressure MPa
Discharge pressure MPa
Subcooling K
Superheat K
Refrigerant
0.120
0.108
0.070
0.008
0.029
0.006
complex forces on the sliding vane, a high speed composite plane

motion of the sliding vane, and little space for the sensor to fit in,
it is very difficult to measure the forces on the sliding vane directly. Thus, the forces on the sliding vane can be verified indirectly by investigating the friction and wear of the sliding vane
used in the experiment.
Figure 13 shows the worn sliding vane after running 72 h in the
compressor test bench. It can be seen that the middle part and the
end part of the side face were severely worn. The simulated results of the forces on the sliding vane show that the forces on the
vane side surface caused by the vane slot up margin within the
rotation angle range of 90290 deg are greater than that within the
other rotation angle range. Moreover, the end part of the sliding
vane is continuously under heavy stress. Therefore, it appears that
the simulated results might agree well with the experimental wear
results of the sliding vane.
6.2.2 Moment Verification. The established models of the SRC
may be verified through measuring the torque input of the SRC.
The comparison between simulated and measured results of the
resistant moment is shown in Fig. 14. It can be seen from Fig. 14
that the simulated result is consistent with the measured result,
which might confirm the correctness of the established models.
The deviation of the average resistant moment between the experimental data and the simulated results is less than 5%. The
measured torque input is smaller than the simulated resistant mo-
ment within the rotation angle range of 190270 deg, and the
former is greater than the latter within the other rotation angle
range. The error between the measured torque input and the simulated resistant moment reaches the maximum value 11 N.m at
the shaft rotation angle of 220 deg, which may be caused by the
buffering of the rotor and the shaft coupling.
Conclusions
The dynamic model has been established, and numerical simulations have been conducted for a novel SRC in order to study its
dynamic characteristics. Motion and stress for its components
have been discussed in detail. From the research, the following
conclusions could be drawn:
a
c
Fig. 13 Wear of the sliding vane
273
0.3
1.5
7
10
R134a
The average relative velocity between the rotor and the

cylinder of the SRC is only 1820% of that of the conventional rotary compressors, and the average relative velocity between the sliding contact tribo-pairs of the SRC
is only 5.97.1% of that of the conventional rotary compressors. Thus, the much lower relative velocity will contribute to resolve those problems caused by high relative
velocity between the rotor and the cylinder, such as severe wear, difficulty to control the mating clearance between the rotor and the cylinder, and difficulty to achieve
seals.
The cylinder has an angular acceleration, which causes
the output torque to fluctuate seriously. The fluctuation
will increase when the shaft rotational speed rises. When
the rotational speed exceeds 3000 rpm, the machine is
found to operate roughly. Therefore, in order to improve
the operation smoothness of the machine at a high rotational speed, the inertia moment of the cylinder is needed
to decrease. Moreover, a flywheel may be useful for the
SRC.
The sliding vane is a key component of the SRC. The
forces on the sliding vane are very complex because the
sliding vane serves as a connector that drives the cylinder
to rotate and at the same time is driven by the rotor.
Therefore, further research should be performed to investigate the fatigue strength and structure optimization of
the sliding vane based on this study. The simulated and
experimental results show that there is severe wear on the
vane surfaces, and then special treatments have to be applied to the sliding vane surfaces to improve the reliability of the SRC.
Acknowledgment
The authors acknowledge the National Natural Science Foundation of China NSFC for funding this research project as part
of the Design for Fluid Mechanism Program Grant No.
50476053.
Nomenclature
Fig. 14 Comparison between simulated and measured results
041102-8 / Vol. 131, APRIL 2009
aSG absolute acceleration for point G of the sliding

vane
aSGe, aSGr, aSGk convected, relative, and Coriolis acceleration for point G of the sliding vane
b thickness of the sliding vane

e eccentricity
f CB friction coefficient of the cylinder bearing
f SR friction coefficient between the vane and the
vane slot
f CS friction coefficient between the vane and the
cylinder
FCB, FCBx, FCBy bearing opposing force and its components
FCBf bearing frictional force
FCg, FCgx, FCgy gas force on the cylinder and its
components
FCMn, FCM normal and tangential forces by oil at the
meshing point
FCS, FCSn, FCS driving force on the cylinder by the sliding
vane and the corresponding normal and
tangential components
FCSnx, FCSny components in X and Y directions of FCSn
FCSf frictional force on the cylinder by head of the
vane
FCSx, FCSy components in X and Y directions of FCS
FSg gas force on the sliding vane
FSC, FSCn normal and tangential constrained force on
vane by the cylinder
FSR1, FSRf1 contact force on the vane from the vane slot
up margin and the corresponding frictional
force
FSR2, FSRf2 contact force on the vane end part from the
vane slot and the corresponding frictional force
FSe, FSr, FSk convected, relative, and Coriolis inertial force
of the vane
FRCf frictional force by oil among the rotor and the
cylinder end covers
FRM frictional force on the rotor from oil at the
meshing point
FRg gas force on the rotor
FRS1, FRS2 reaction forces on rotor from the sliding vane
FRSf1, FRSf2 frictional forces on rotor from the sliding vane
GC weight of the cylinder
H height of the cylinder
JC cylinder rotary inertia
l length, l = RR
L radial length of the vane
LCSn, LCS, LRS1, LRS2,
LSg, LSk, LSR1, LSR2 arms of the corresponding forces
mC mass of the cylinder
M Cg, M CM , M CS, M CBf moments on the cylinder by the
corresponding forces
M RCf , M RM , M RS1, M RS2,
M RSf1, M RSf2, M Rg moments on the rotor created by
the corresponding forces
M Se, M Sg, M Sk, M Sr,
M SR1, M SR2, M SRf1, M SRf2 moments on the sliding vane created by the corresponding forces
M z total resistance moment
Ps suction chamber pressure at shaft rotation
angle
Pd compression chamber pressure at rotation angle
r rotor radius
rin rotor inside radius
R
RB
RS
t
vC
vS, vSe, vSr,
vSG, vSGe, vSGr
V
V d
x
C
C
cylinder radius
cylinder bearing radius
sliding vane radius
time
absolute velocity of the cylinder
absolute, convected, and relative velocity for
point OS of the vane
absolute, convected, and relative velocity
for point G of the vane
total volume of the working chamber
compression chamber volume at rotation angle
displacement
angle, = ORBOC
bow angle of the oil film
shaft rotation angle
included angle between of the FCS with axis
X
angle, = MOCB
length of OSOR
length of ORB
length of ORG
rotor angular velocity
cylinder angular velocity
cylinder angular acceleration
kinematic viscosity of the lubricating oil
oil film thickness
References
1 Jeon, H-.G., and Lee, K-.S., 2007, Friction and Wear of Lubricated Sliding
Surfaces of Coated Vane and Flange in Rotary Compressor With Carbon Dioxide as a Refrigerant, Asian Pacific Conference for Fracture and Strength,
pp. 17851788.
2 Suh, A. Y., Patel, J. J., and Polycarpou, A. A., 2006, Scuffing of Cast Iron and
Al390-T6 Materials Used in Compressor Applications, Wear, 260, pp. 735
744.
3 Demas, N. G., and Polycarpou, A. A., 2006, Tribological Investigation of
Cast Iron Air-Conditioning Compressor Surfaces in CO2 Refrigerant, Tribol.
Lett., 22, pp. 271278.
4 Oh, S.-D., Lee, K.-S., and Kim, J.-S., 2006, Experimental Investigation on
Friction and Wear of Coated Vane Under the Environments of Lubricants and
CO2 as a Refrigerant, Key Eng. Mater., 326328II, pp. 11851188.
5 Oh, S. D., and Kim, J. W., 2004, Friction and Wear Characteristics of TiN
Coated Vane for the Rotary Compressor in a R410a Refrigerant, Tribol.
Trans., 481, pp. 2933.
6 Huang, Y. M., and Shiau, C.-S., 2006, Optimal Tolerance Allocation for a
Sliding Vane Compressor, ASME J. Mech. Des., 128, pp. 98107.
7 Huang, Y. M., and Li, C. L., 2005, The Stress of Sliding Vanes in a Rotary
Compressor, 2005 World Tribology Congress III, pp. 99100.
8 Cai, H., Li, L., and Guo, B., 2005, Research on Tip Profile of Vane for Rotary
Vane Compressor, Fluid Machinery Group-International Conference on Compressors and Their Systems, pp. 215222.
9 Ooi, K. T., 2005, Design Optimization of a Rolling Piston Compressor for
Refrigerators, Appl. Therm. Eng., 25, pp. 813829.
10 Lee, Y.-Ze., and Oh, S.-D., 2003, Friction and Wear of the Rotary Compressor VaneRoller Surfaces for Several Sliding Conditions, Wear, 255, pp.
11681173.
11 Zhou, H., Qu, Z., and Feng, J., 2005, Sliding Vane Kinetic Characteristics
Analysis for a SRC, J. Shanghai Jiaotong Univ., 399, pp. 982984.
12 Zhou, H., Qu, Z., and Yu, B., 2007, Leakage Research in Synchronal Rotary
Gas Compressor, China J. Mech. Eng., 182, pp. 205208.
13 Solzak, T. A., and Polycarpou, A. A., 2006, Tribology of WC/C Coatings for
Use in Oil-Less Piston-Type Compressors, Surf. Coat. Technol., 201, pp.
42604265.
14 Xu, H., Cai, C., Yan, J., Wang, K., and Zhou, S., 2000, Machine Design
Handbook 2, China Machine Press, Beijing, China, Chap. 16, pp. 1623.
APRIL 2009, Vol. 131 / 041102-9
Evaluation of the Dynamic

Transfer Matrix of Cavitating
Inducers by Means of a Simplified
Lumped-Parameter Model
Angelo Cervone
Postdoctoral Fellow
e-mail: angcervone@mbox.me.es.osaka-u.ac.jp
Yoshinobu Tsujimoto
Professor
e-mail: tujimoto@me.es.osaka-u.ac.jp
Engineering Science,
Osaka University,
1-3 Machikaneyama, Toyonaka,
Osaka 560-8531, Japan
The paper will present an analytical model for the evaluation of the pressure and flow
rate oscillations in a given axial inducer test facility. The proposed reduced order model
is based on several simplifying assumptions and takes into account the facility design and
the dynamic properties of the tested inducer. The model has been used for evaluating the
dynamic performance of a prototype of the LE-7 engine liquid oxygen (LOX) inducer, in
tests carried out under given external flow rate excitations. The main results of these
calculations will be shown, including the expected oscillations under a wide range of
operational conditions and the influence of facility design. Calculations showed that the
only way to obtain the two linearly independent test conditions, necessary for evaluating
the inducer transfer matrix, is by changing the facility suction line: Any other changes in
the facility design would result ineffective. Some other important design indications provided by the analytical model will be presented in the paper. DOI: 10.1115/1.3089535
Yutaka Kawata
Professor
Osaka Institute of Technology,
5-6-1 Omiya, Asahi-Ku,
Osaka 535-8505, Japan
e-mail: kawata@med.oit.ac.jp
Introduction
Space rocket turbopumps are one of the most crucial components of all primary propulsion concepts powered by liquid propellant engines. Generally these pumps include an axial inducer,
upstream of the centrifugal stages, in order to improve the suction
performance and reduce the propellant tank pressure and weight.
Severe limitations are associated with the design of high power
density, dynamically stable machines capable of meeting the extremely demanding suction, pumping, and reliability requirements
of space transportation systems 1.
It is well known that many of the flow instabilities acting on
axial inducers are significantly influenced by the so-called dynamic matrix, which relates the pressure and flow oscillations at
the pump inlet with those at the pump outlet 24. The first steps
in the analytical and experimental characterizations of the dynamic matrix date back to the work of Brennen and co-worker in
the 1970s 5,6. However, more recent works have given important contributions by evaluating the previously obtained results
through a careful analysis of the successive experimental and numerical data see, for example, Refs. 7,8.
Conventionally, the dynamics of hydraulic systems is treated in
terms of lumped-parameter models, which assume that the distributed physical effects between two measuring stations can be
represented by lumped constants. This assumption is usually considered valid when the geometrical dimensions of the system are
significantly shorter than the acoustic wavelength at the considered frequency. As a direct consequence of this assumption, the
dynamic matrix of a generic system like an axial inducer can be
written as
pd
Q
d
H11 H12
H21 H22

pu
Q
u
are, respectively, the pressure and flow rate oscilwhere p and Q

lating components, and the subscripts u and d denote, respectively, the flow conditions upstream and downstream of the conContributed by the Fluids Engineering Division of ASME for publication in the
JOURNAL OF FLUIDS ENGINEERING. Manuscript received February 6, 2008; final manuscript received January 19, 2009; published online March 9, 2009. Assoc. Editor:
Steven Ceccio.
sidered system. As a consequence of the well known electrical

analogy, the negative of the real part of H12 is usually denoted as
the system resistance, the negative of the imaginary part of H12
is referred as the system inertance, and the negative of the
imaginary part of H21 is the system compliance 9,2.
It is known that the pump resistance plays a decisive role in its
unstable behavior 2. Considering that at a frequency of 0 Hz it
has the same meaning of the slope of the pump performance
curve, it is easy to understand that a positive value of the resistance is directly connected to surge-mode instabilities. Furthermore, it has been shown that other flow instabilities, such as rotating cavitation, are promoted by particular combinations of
values of the elements of the transfer matrix 3. By this point of
view, cavitation compliance and mass flow gain factor play a decisive role, as clearly shown by a number of analytical and experimental analyses 10,11.
Various experimental activities have also shown that, imposing
a flow oscillation to the system, its behavior can turn from stable
to unstable depending on the value of the frequency of imposed
oscillations 9,2.
The above observations confirm the importance of obtaining a
sufficiently accurate prediction of the dynamic matrix of a cavitating inducer, by both analytical and experimental means. To this
purpose, the present paper will describe a simplified analytical
model for the evaluation of the dynamic transfer matrix of not just
the axial inducer, but the entire test facility in which it is installed.
The model is based on simple preliminary assumptions onedimensional flow, small oscillations, incompressible fluid, and
quasisteady response of the components and can be used for predicting the expected pressure and flow rate oscillations in the facility under a given fluctuation externally imposed to the system.
As a consequence, useful indications can be obtained on how to
design the facility for obtaining the best experimental performance. The inducer transfer matrix, not being the main point on
which the present analytical calculations are focused, is evaluated
by means of a simplified theoretical-empirical model.
In the paper, the proposed model will be applied to the cavitation test facility of the Engineering Science Department at Osaka
University. The test inducer used for the calculations is a model of
the liquid oxygen inducer of the Japanese LE-7 rocket engine.
Previous experimental activities carried out on this inducer 12
APRIL 2009, Vol. 131 / 041103-1
Fig. 1 Schematic of the experimental facility
have shown that several flow instabilities are promoted by the

cavitating pattern on the pump. The experiments, mainly focused
on the characterization of the different flow instabilities acting on
the inducer and their field of existence, lead to the detection of
instabilities including rotating cavitation, asymmetric blade cavitation, and higher order modes. It has also been shown that some
of these instabilities can be reduced or suppressed by changing the
casing geometry: Rotating cavitation and asymmetric cavitation,
as an example, are strongly reduced by means of a J-groove on
the casing.
Experimental Apparatus
A sketch of the cavitation facility of the Engineering Science

Department at Osaka University is shown in Fig. 1. The figure
also shows the main notations, which will be used in the present
paper.
The actual sections where the pressure and flow rate oscillations are measured are denoted by subscripts 1 upstream of the
inducer and 2 downstream of the inducer. The sector of the
facility included between these two points will be referred in the
following as measurement section. A given external flow rate
oscillation denoted by QF is imposed just downstream of the
measurement point 2, by means of a dedicated apparatus. The
suction line between tank and upstream measurement point is
denoted by subscript C, while the discharge line between the
fluctuator and the main tank entrance is denoted by subscript B.
The internal pipe diameter is 203.5 mm in the suction and discharge lines and 155.2 mm inside the measurement section. A
conical expansion is installed downstream the fluctuator, while a
conical contraction is present just after the measurement point 1.
Two valves for the regulation of the flow rate are mounted in the
discharge section.
The test inducer used for the present calculations Fig. 2 is a
model of the liquid oxygen inducer of the Japanese LE-7 rocket
engine. It is a three-bladed axial inducer, with a tip radius rT
= 74.9 mm and a profiled, variable-radius hub the hub/tip radii
ratio is 0.25 at the inlet and 0.51 at the outlet. The inlet tip blade
angle is 7.5 deg and the tip solidity is 1.91.
assumption is not valid in real pumps under cavitating conditions 7,8,11, it could represent a good starting point for
the simplified reduced order analysis presented in this paper. As it will be shown in the following, the results and the
main conclusions drawn by the analyses are not significantly affected by this assumption, especially for what concerns the design of the experiment.
Under the above assumptions, pressure and flow rate can be
written in a complex form, as functions of time, as follows:
pt = p + p eit
e it
+Q
Qt = Q
usually real are the pressure and flow rate
where p and Q
usually complex are the pressure and
steady values, p and Q
flow rate oscillating components, and is the frequency of the
oscillations.
The relevant components for the evaluation of the dynamic matrix are the oscillating ones. Under the above assumptions, the
matrices of all the components of the facility can be easily evaluated as follows.
Model Description
The proposed model is based on the following initial assumptions.

1 The flow is assumed unsteady, quasi-one-dimensional.
2 All oscillations are assumed to be small only first order
terms are taken into account in the equations.
3 The working fluid is assumed incompressible and compliance is therefore considered negligible, except for regions
where cavitation or air volumes are present.
4 The response of all components of the system is assumed
quasisteady. Even if it has been widely shown that this
041103-2 / Vol. 131, APRIL 2009
Fig. 2 The test inducer
3.1 Pipes. With a suitable manipulation of the generic continuity and momentum equations, the dynamic matrix can be written as follows:
H=
1 R iL
1
where the resistance R and the inertance L are given by

R = k
Q
u
A u2
L=
A d2
A u2
Q
u
in which is the density of the working fluid, A is the pipe

section, is the inertial length of the pipe defined as the mean
pipe section divided to pipe length, and k is the loss coefficient
inside the pipe.
3.2
Main Tank. In this case, the dynamic matrix is given by

H=
iC 1
V
p
3.3 Inducer. In the noncavitating case, it can be shown that

the inducer dynamic matrix is very simple, with a similar structure
to that of pipe sections. In this case the inertance L is simply given
by the inertial length of blade passages, while the resistance R can
be evaluated using the momentum equation as follows:
1
1
rT d
+ Q
u
2
Au d
Ad
A u2
in which is the inducer rotating speed, rT is the inducer tip

radius, and d / d is the slope of the inducer performance curve.
In the cavitating case the equations become more complicated.
The driving parameter is the presence of a vapor region on the
blades, the oscillations of which generate a shift between the inlet
and outlet flow oscillations. The evaluation of the volume of cavity formed inside the pump, as a function of the flow coefficient
and the cavitation number , is therefore needed. Elaboration of
the continuity and momentum equations shows that the dynamic
matrix is
H=
1 S + iX R iL
iC
1 i M
where S + iX is the pressure gain factor, R + iL is the pump

impedance, C is the cavitation compliance, and M is the mass
flow gain factor. These parameters are defined as follows:
S + iX = 2
+ CLind2 i
rT
R + iL =
Ad
i M
A u2
Q
rT
u
2
Ad
Ad
C=
2r T2
Q
rT
u
C
Ad
A d2
Q
u
VC
in which VC is the cavity volume and Lind is the inertance of

inducer blade passages.
After having tried different approaches for the evaluation of the
cavity volume VC, including suitable modifications to the
BrennenAcosta model of partially cavitating cascades 13,14, a
simpler empirical-theoretical model will be proposed here. The
choice of this model was led by the consideration that the present
work is not aimed to an accurate evaluation of the inducer dynamic characteristics, but to a characterization of the entire facility, in order to acquire useful theoretical and design indications for
the successive experimental activities.
The cavitation volume has therefore been estimated using a
polynomial relationship and taking into account the experimental
data provided by Brennen for a 10.2 cm inducer 13. Considering
that these data give an estimation of the cavitation compliance on
the inducer at four different values of , the following relation
was used:
+ V C c
in which p and V are the mean pressure and volume of the air at
the top of the tank, and is the specific heat ratio of air.
R=
1
VC
AurT
V C , = a 1 c 4 + a 2 c 3 + a 3 c 2 + a 4 c
where the compliance C is

C=
M=
A d2
where the constants a1, a2, a3, and a4 can be evaluated using
the four experimental values of C provided by Brennen 13.
In the above equation, c is the choked cavitation number of the
BrennenAcosta partially cavitating cascade model i.e., the cavitation number for which the cavity becomes infinitely long 13,
which depends on the inducer geometry and the incidence angle
and, as a consequence, the flow coefficient.
The cavity volume at the choked cavitation number, VCc, is
evaluated assuming the following.
1 At the choked cavitation number c, complete breakdown
occurs i.e., the head rise of the inducer falls to zero.
2 At c, the blade channels are completely filled by bubbly
cavitation, having a void fraction 0, which can be evaluated using the relationship shown by Brennen 5,
=
where b is the tip blade angle and is the amount of

breakdown, estimated using assumption 1.
3 The cavity volume at choked cavitation number is finally
estimated by multiplying the void fraction 0 to the blade
channel volume.
Calculations confirmed that the model can reasonably reproduce the experimental transfer matrix as shown in Ref. 13.
3.4 Characterization of the Complete Facility. Four unknowns have to be evaluated for the solution of the problem: the
pressure and flow oscillations at the inlet measurement point, p1
, and the correspondent oscillations at the outlet measureand Q
1
. The first two equations for their evaluation
ment point, p2 and Q
2
can be obtained by considering the equilibrium inside the measurement section
p2
+ MLind
02 02
21 02 sin2 b
Q
2
HM 11 HM 12
HM 21 HM 22

p1
Q
1
where the transfer matrix HM is given by
+ iLind
HM =
1 Rout iLout
0
1
1 Rin iLin
0
1 S + iX Rind iLind
iCind
1 i M
APRIL 2009, Vol. 131 / 041103-3
Considering the components of the facility between the downstream measurement section and the flow rate fluctuator see Fig.
1, it is possible to write the oscillations immediately after the
fluctuator as
= 1 atm, and air volume in the tank V = 625 liters, it is possible

to obtain the following values for the ratio of the first term to the
second term in the expression of p1:
+Q
=Q
Q
3
2
F
0.255
for
= 1 Hz
0.057
for
= 2 Hz
p3 = p2 RA + iLAQ
2
0.023
for
= 3 Hz
The boundary conditions of the problem can be written by

considering the equilibrium equations in the sector of the facility
outside the measurement section
0.007
for
= 5 Hz
0.002
for
= 10 Hz
p1
Q
1
HB11 HB12
HB21 HB22

p3
Q
3
where the transfer matrix HB is given by

HB =
1 RC iLC
1
iCT 1
1 RB iLB
0
Combining the above expressions, it is possible to obtain the

last two equations needed for the solution of the problem.
The solution of the problem in the noncavitating case is of
particular interest. In this case, taking into account the simpler
structure of the inducer dynamic matrix, the pressure and flow rate
oscillations can be shown to be equal to
RB + iLBQ
F
=Q
=
Q
1
2
Rm + RA + RB + RC + iLm + LA + LB + LC
p1 =
RC + iLCRB + iLBQ
Q
F
F
+
iCT Rm + RA + RB + RC + iLm + LA + LB + LC
p2 =
+ R + iL
+ L R + iL Q
R
Q
C
C
B
B
F
F
+
iCT Rm + RA + RB + RC + iLm + LA + LB + LC
3.5 Backward Calculation. Once the pressure and flow

rate oscillations are known by means of the above equations, they
can be used as input data to simulate the evaluation of the measured transfer matrix HM. In this case, at least two experimental
results obtained under linearly independent conditions are needed
to find the four unknown elements of the matrix. Indicating by
subscripts a and b the two experimental results used for the calculation, it is possible to write

p2a
Q
2a
p2b
Q
2b
HM 11 HM 12
HM 21 HM 22
HM 11 HM 12
HM 21 HM 22

p1a
Q
1a
p1b
Q
1b
which lead to the following system of four equations in four unknowns:
where
Rm = Rin + Rind + Rout
Lm = Lin + Lind + Lout
The first term in the expressions of p1 and p2 is usually very
small due to the large value of the tank compliance CT and can
therefore be neglected, at least for sufficiently high values of the
exciting frequency . As an example, for the Osaka University
facility with LE-7 engine LOX inducer under nominal operating
conditions = 0.078, = 3000 rpm, mean tank pressure p
det T =
It is clear that, for better accuracy in the evaluation of the

elements of HM, the determinant of the matrix T should be as far
as possible from zero. This is obtained by taking two experimental
results under sufficiently independent conditions. In the noncavitating case, using the equations for the pressure and flow rate
oscillations and neglecting the first term in the expression of p1, it
is possible to obtain the following result for the determinant of
matrix T:
2
RBa + iLBaRBb + iLBbRCb RCa + iLCb LCaQ
F
Rm + RAa + RBa + RCa + iLm + LAa + LBa + LCaRm + RAb + RBb + RCb + iLm + LAb + LBb + LCb
It is therefore clear that, if no changes are made to the section C

of the facility i.e., the suction line, between the tank and the
upstream measurement point, the determinant of the matrix T
will be equal to zero whatever other change is made to the other
sections of the facility. It could be shown that this is true also
under cavitating conditions. Furthermore, if calculations are carried out in the case when the fluctuator is installed on the suction
line instead of the discharge line, the opposite result is obtained:
In that case, the only effective changes would be those affecting
the section of the facility where there is no fluctuator, i.e., the
discharge line.
041103-4 / Vol. 131, APRIL 2009
This result provides a first important design indication for the

experiments: The two linearly independent conditions, needed for
the evaluation of the dynamic matrix, can be obtained only by
changes in the suction line. These changes can involve the resistance RC, the inertance LC, or both. All these options will be
evaluated in the following.
Results and Discussion
Two different sets of calculations will be presented here. The

first one allows for the evaluation of the expected oscillations in
the facility under its nominal design, in different operational conTransactions of the ASME
Fig. 3 Amplitude of the upstream pressure oscillations, as a

function of the excitation frequency and the flow coefficient
cavitating conditions, for = 3000 rpm, = 0.1, and amplitude of the flow rate oscillations generated by the fluctuator
equal to 1 l/s
ditions. The second one will be a backward calculation, aimed

to the estimation of the effectiveness of several alternative designs
of the facility for obtaining the second linearly independent experimental condition.
4.1 Nominal Design: Pressure and Flow Rate Oscillations.
The 3D plots from Figs. 36 show the amplitude of pressure and
flow rate fluctuations under cavitating conditions, as functions of
the excitation frequency and the flow coefficient, for = 0.1,
= 3000 rpm, and amplitude of the flow rate oscillations generated
by the fluctuator equal to 1 l/s. Since there is no phase information
in the 3D plots, Figs. 710 show the amplitude and phase of
pressure and flow rate fluctuations under cavitating conditions, as
functions of the excitation frequency, for = 0.1, = 3000 rpm,
= 0.078 inducer nominal operating conditions, and amplitude
of the flow rate oscillations generated by the fluctuator equal to 1
l/s. The phase delay is calculated with respect to the flow rate
oscillations generated by the fluctuator.
Fig. 4 Amplitude of the downstream pressure oscillations, as

a function of the excitation frequency and the flow coefficient
equal to 1 l/s
Fig. 5 Amplitude of the upstream flow rate oscillations, as a

function of the excitation frequency and the flow coefficient
equal to 1 l/s
4.2 Evaluation of Alternative Facility Designs. The effectiveness of different facility designs has been evaluated by means
of the backward calculation. The results provided by the model
have therefore been used as a starting point for the evaluation of
the dynamic matrix elements, in order to compare this evaluation
to the actual nominal values provided by the model. The dynamic
matrix elements have been evaluated at different excitation frequencies and flow coefficients .
In order to better simulate the real experimental conditions expected in the facility, three different artificial error sources have
been intentionally added to the nominal data. The first two error
sources are a random white noise at all the frequencies included
in the spectrum of the measured signal which simulates the typical white noise usually present in laboratory environments and a
defined sinusoidal noise at 60 Hz frequency of electrical disturbance in Japan. The third error source consists in taking only a
finite number of decimals from the numerical solutions provided
by the model, in order to simulate the experimental sensors accuracy.
Fig. 6 Amplitude of the downstream flow rate oscillations, as

a function of the excitation frequency and the flow coefficient
equal to 1 l/s
APRIL 2009, Vol. 131 / 041103-5
Fig. 7 Amplitude and phase of the upstream pressure oscillations, as a function of the excitation frequency cavitating
conditions, for = 3000 rpm, = 0.1, = 0.078, and amplitude
of the flow rate oscillations generated by the fluctuator equal to
1 l/s
Fig. 8 Amplitude and phase of the downstream pressure oscillations, as a function of the excitation frequency cavitating conditions, for = 3000 rpm, = 0.1, = 0.078, and amplitude of the flow rate oscillations generated by the fluctuator
equal to 1 l/s
A fourth important source of errors is given by the determinant

of the evaluation matrix T, which is usually almost singular in
sufficiently realistic design cases.
The simulated experimental signal obtained by this procedure
was then elaborated using the Fourier analysis as it were a real
experimental signal, and the information related to the known
excitation frequency was extracted in terms of amplitude and
phase of the oscillations at that frequency. Finally, these extrapolated data were used for the evaluation of the components of the
transfer matrix, which were compared with the expected ones.
In the proposed calculations, three different alternative facility
designs have been considered, according to the following
nomenclature.
Nominal. The nominal value of the matrix element, as obtained
by the analytical model equations.
New design 6 (ND6) single. For obtaining the second linearly
independent condition, a resistance is added to the suction line of
the facility. The resistance is constituted by a single perforated
plate with 69 holes of 12 mm diameter.
New design 6 (ND6) M-stage. The same as above, but in this
case the resistance is constituted by a multistage perforated plate.
Three plates with holes of 12 mm diameter are considered, having
a different number of holes 58 holes for the first plate, 73 for the
second, and 91 for the third, in order to provide a better distribution of the resistance loss and avoid cavitation in the plates.
New design 7 (ND7). In this case, the inertance of the suction
line of the facility is increased by adding a long pipe section
having a length equal to 6.3 m.
The plots in Figs. 1116 are referred to the case = 0.1 cavitating conditions, = 3000 rpm, with amplitude of the flow rate
oscillations generated by the fluctuator equal to 1 l/s. The determinant of the evaluation matrix T and a comparison between the
nominal and the evaluated elements of the dynamic matrix are

shown for the cases ND6 single, ND6 M-stage, and ND7.
Only the components HM 12 and HM 21 of the dynamic matrix the
most interesting by the point of view of inducer instabilities are
shown. Two cases are considered: fixed flow coefficient
= 0.078 with variable excitation frequency, and fixed excitation
frequency = 5 Hz with variable flow coefficient.
With respect to the flow coefficient, the results are generally
good, with acceptably small errors. With respect to the excitation
frequency, the results are also good, but there is a loss in accuracy
at very low excitation frequencies less than 1 Hz and at higher
excitation frequencies more than 20 Hz, except for ND7, which is
more effective also at higher frequencies. The lower accuracy of
results at low frequencies can be explained considering the very
low value of pressure oscillations in that case see Figs. 7 and 8.
As a consequence, the accuracy of the pressure transducers becomes a significant source of errors. The lower accuracy at higher
frequencies for ND6 cases is explained by considering that, at
higher frequencies, the inductance of the suction line of the facility LC becomes significantly more important than its resistance
RC for the evaluation of the oscillations, and so the ND7 design is
more effective than the ND6 ones.
It is therefore demonstrated that, for most of the frequencies of
actual interest, the simple insertion of a calibrated resistance in the
suction line of the facility between the tank and the upstream
measurement section could provide the second linearly independent condition necessary for the experimental evaluation of the
transfer matrix. However, to obtain accurate results at higher excitation frequencies more than 20 Hz, it should probably be
necessary to add a long pipe section to the suction line ND7
condition in order to increase its inertance.
041103-6 / Vol. 131, APRIL 2009
Fig. 11 Determinant of the evaluation matrix T, as a function

of the excitation frequency , for = 3000 rpm, = 0.1,
= 0.078, and amplitude of the flow rate oscillations generated by
the fluctuator equal to 1 l/s
Fig. 9 Amplitude and phase of the upstream flow rate oscillations, as a function of the excitation frequency cavitating
conditions, for = 3000 rpm, = 0.1, = 0.078, and amplitude
of the flow rate oscillations generated by the fluctuator equal to
1 l/s
Fig. 12 Real and imaginary parts of the element HM12 of the

dynamic matrix, as functions of the excitation frequency , for
= 3000 rpm, = 0.1, = 0.078, and amplitude of the flow rate
oscillations generated by the fluctuator equal to 1 l/s
Fig. 10 Amplitude and phase of the downstream flow rate oscillations, as a function of the excitation frequency cavitating conditions, for = 3000 rpm, = 0.1, = 0.078, and amplitude of the flow rate oscillations generated by the fluctuator
equal to 1 l/s

dynamic matrix, as functions of the excitation frequency , for
= 3000 rpm, = 0.1, = 0.078, and amplitude of the flow rate
oscillations generated by the fluctuator equal to 1 l/s
APRIL 2009, Vol. 131 / 041103-7
Fig. 14 Determinant of the evaluation matrix T, as a function

of the flow coefficient , for = 3000 rpm, = 0.1, = 5 Hz, and
amplitude of the flow rate oscillations generated by the fluctuator equal to 1 l/s
Conclusions
An analytical model for the evaluation of the dynamic matrix of

a cavitating inducer facility has been proposed in the present paper. The model, even if extremely simplified and based on very
strong assumptions, has shown to be a powerful tool for the design of the facility and the preparation of the experimental activity.
In particular, the backward calculation utilization of the model
results as simulated experimental data, after having opportunely
added to them artificial error sources can be used to get extremely useful indications about the effectiveness of different facility designs. Furthermore, the model equations showed that the
only effective design changes for obtaining the second linearly
independent condition, needed for the experimental evaluation of
the dynamic matrix, are those related to the suction line of the
facility or, conversely, the discharge line if the external fluctuations are provided to the suction line.
Future improvements to the model could include, as an example, a better evaluation of the inducer dynamic matrix, based
on more accurate models than the simple empirical-theoretical

dynamic matrix, as functions of flow coefficient , for
= 3000 rpm, = 0.1, = 5 Hz, and amplitude of the flow rate oscillations generated by the fluctuator equal to 1 l/s
041103-8 / Vol. 131, APRIL 2009

dynamic matrix, as functions of flow coefficient , for
= 3000 rpm, = 0.1, = 5 Hz, and amplitude of the flow rate oscillations generated by the fluctuator equal to 1 l/s
model proposed in this paper. As a useful tool to this purpose, an

extensive experimental activity is foreseen during the next
months, in order to measure the pressure and flow rate oscillations
in the facility under several operating conditions and compare the
obtained data to the results of the model.
Acknowledgment
A.C. would like to acknowledge the support of Japan Society
for the Promotion of Science JSPS, which has funded his research program at Osaka University. He also expresses his gratitude to all the colleagues at Pisa University in Italy, and, in particular, Professor Luca dAgostino and Professor Mariano
Andrenucci, for their kind and constant encouragement.
Nomenclature
ai constants in cavity volume equation,
i = 1 4 m3
A pipe section m2
C compliance m4 s2 / kg
k loss coefficient dimensionless
L inertance kg/ m4
M mass flow gain factor s1
p pressure Pa
H generic dynamic matrix
HB dynamic matrix outside the measurement
section
HM dynamic matrix of the measurement section
Q flow rate m3 / s
QF externally imposed flow oscillation m3 / s
R resistance kg/ m4 s
rT inducer tip radius m
S real part of the pressure gain factor
dimensionless
t time s
T characteristic matrix for backward calculation
V volume m3
VC cavity volume m3
X imaginary part of the pressure gain factor s
0 void fraction dimensionless
b inducer tip blade angle rad
head loss at breakdown dimensionless
flow coefficient dimensionless
specific heat ratio dimensionless
inertial length m
head coefficient dimensionless
density kg/ m3
cavitation number dimensionless
choked cavitation number dimensionless
frequency of oscillations Hz
inducer rotating speed rad/s
Superscripts
oscillating component of a given quantity
steady component of a given quantity
Subscripts
a , b linearly independent experimental conditions
A pipework from measurement section to
fluctuator
B discharge line
C suction line
d downstream section generic
in measurement section upstream of inducer
ind inducer
m total measurement section including inducer
out measurement section downstream of inducer
T main tank
u upstream section generic
1 upstream measurement point
2 downstream measurement point
3 point immediately after the fluctuator
References
1 Stripling, L. B., and Acosta, A. J., 1962, Cavitation in TurbopumpsPart 1,
ASME J. Basic Eng., 84, pp. 326338.

2 Kawata, Y., Takata, T., Yasuda, O., and Takeuchi, T., 1988, Measurement of
the Transfer Matrix of a Prototype Multi-Stage Centrifugal Pump, Paper No.
C346/88.
3 Tsujimoto, Y., Kamijo, K., and Yoshida, Y., 1993, A Theoretical Analysis of
Rotating Cavitation in Inducers, ASME J. Fluids Eng., 115, pp. 135141.
4 Tsujimoto, Y., Watanabe, S., and Horiguchi, H., 1998, Linear Analyses of
Cavitation Instabilities of Hydrofoils and Cascades, Proceedings of the USJapan Seminar: Abnormal Flow Phenomena in Turbomachinery, Osaka, Japan.
5 Brennen, C. E., 1978, Bubbly Flow Model for the Dynamic Characteristics of
Cavitating Pumps, J. Fluid Mech., 89, pp. 223240.
6 Ng, S. L., and Brennen, C. E., 1978, Experiments on the Dynamic Behavior
of Cavitating Pumps, ASME J. Fluids Eng., 100, pp. 166176.
7 Otsuka, S., Tsujimoto, Y., Kamijo, K., and Furuya, O., 1996, Frequency Dependence of Mass Flow Gain Factor and Cavitation Compliance of Cavitating
Inducers, ASME J. Fluids Eng., 118, pp. 400408.
8 Rubin, S., 2004, An Interpretation of Transfer Function Data for a Cavitating
Pump, Proceedings of the 40th AIAA/ASME/SAE/ASEE Joint Propulsion
Conference, Fort Lauderdale, FL.
9 Bhattacharyya, A., 1994, Internal Flows and Force Matrices in Axial Flow
Inducers, Ph.D. thesis, California Institute of Technology, Pasadena, CA.
10 Jun, S. I., Tokumasu, T., and Kamijo, K., 2003, Dynamic Response Analysis
of High Pressure Rocket Pumps, Proceedings of CAV2003Fifth International Symposium on Cavitation, Osaka, Japan.
11 Yamamoto, K., and Tsujimoto, Y., 2008, Backflow Vortex Cavitation and Its
Effects on Cavitation Instabilities, Proceedings of the WIMRC Cavitation
Forum, Warwick University, Coventry, UK.
12 Shimiya, N., Fujii, A., Horiguchi, H., Uchiumi, M., Kurokawa, J., and
Tsujimoto, Y., 2006, Suppression of Cavitation Instabilities in an Inducer by
J-Groove, Proceedings of CAV2006Sixth International Symposium on
Cavitation, Wageningen, The Netherlands.
13 Brennen, C. E., 1994, Hydrodynamics of Pumps, Concepts ETI Inc., Norwich,
VT/Oxford University Press, Oxford.
14 Brennen, C. E., and Acosta, A. J., 1973, Theoretical, Quasi-Static Analysis of
Cavitation Compliance in Turbopumps, J. Spacecr. Rockets, 103, pp. 175
179.
APRIL 2009, Vol. 131 / 041103-9
RANS Analyses of Turbofan

Nozzles With Internal Wedge
Deflectors for Noise Reduction
James R. DeBonis1
Inlet and Nozzle Branch,
NASA Glenn Research Center,
Cleveland, OH 44135
e-mail: james.r.debonis@nasa.gov
Computational fluid dynamics (CFD) was used to evaluate the flow field and thrust
performance of a promising concept for reducing the noise at take-off of dual-stream
turbofan nozzles. The concept, offset stream technology, reduces the jet noise observed on
the ground by diverting (offsetting) a portion of the fan flow below the core flow, thickening and lengthening this layer between the high-velocity core flow and the ground
observers. In this study a wedge placed in the internal fan stream is used as the diverter.
Wind, a Reynolds averaged NavierStokes (RANS) code, was used to analyze the flow
field of the exhaust plume and to calculate nozzle performance. Results showed that the
wedge diverts all of the fan flow to the lower side of the nozzle, and the turbulent kinetic
energy on the observer side of the nozzle is reduced. This reduction in turbulent kinetic
energy should correspond to a reduction in noise. However, because all of the fan flow is
diverted, the upper portion of the core flow is exposed to the freestream, and the turbulent
kinetic energy on the upper side of the nozzle is increased, creating an unintended noise
source. The blockage due to the wedge reduces the fan mass flow proportional to its
blockage, and the overall thrust is consequently reduced. The CFD predictions are in
very good agreement with experimental flow field data, demonstrating that RANS CFD
can accurately predict the velocity and turbulent kinetic energy fields. While this initial
design of a large scale wedge nozzle did not meet noise reduction or thrust goals, this
study identified areas for improvement and demonstrated that RANS CFD can be used to
improve the concept. DOI: 10.1115/1.3089536
Introduction
The concept of offsetting the fan stream of a dual flow turbofan

nozzle to reduce jet noise was developed by Papamoschou at the
University of California Irvine UCI 13. The idea is to provide
a thicker layer of fan flow on the lower observer side of the
nozzle. The additional fan flow serves to extend the length of the
fan stream covering or shielding a larger portion of the core
flow on the lower side. This layer of low speed flow reduces the
convective Mach number of the turbulent eddies generated by the
high speed core flow. This reduces these turbulent structures ability to generate strong acoustic waves that travel to the far field on
the lower side of the nozzle. Papamoschous 13 initial concept
was to physically offset the nozzle flow paths. This eccentric
nozzle would be difficult to implement in a real system. He found
that vanes and/or wedges placed in the fan stream can deflect the
fan flow to the lower portion of the jet to achieve the desired flow
field and resulting acoustics in a simple, easy to implement manner.
Papamoschou 13 tested these concepts in a small jet facility
1 in. 2.54 cm fan exit diameter and measured jet noise reductions of over 5.0 dB in the overall sound pressure level. These
impressive results led the National Aeronautics and Space Administration NASA to examine this noise reduction concept at a
larger scale. NASA developed the Offset Stream Technology
OST program to design and test turbofan nozzles with devices to
offset the fan flow. The program consisted of computational fluid
dynamics CFD analysis to screen flow deflector design and
placement and to estimate nozzle performance, design of experiments DOE methodology to reduce the number of experimental
1
Corresponding author.
JOURNAL OF FLUIDS ENGINEERING. Manuscript received March 21, 2008; final manuscript received January 22, 2009; published online March 9, 2009. Assoc. Editor: Zvi
Rusak. Paper presented at the AIAA Aerospace Sciences Meeting, 2008.
test points required, additional small scale testing, and large scale
testing 9.63 in. 24.5 cm fan exit diameter in the NASA Glenn
Nozzle Acoustic Test Rig NATR 4.
There are four other papers from the OST study that form a
complete description of the program. A companion CFD study
examining the vane deflector and s-duct nozzles was performed
by Dippold et al. 5. The experimental results from NATR, acoustics and particle image velocimetry PIVbased velocity measurements in the exhaust plume, were reported by Brown et al.
6, and a comparison of experimental results between UCI and
NASA Glenn was detailed by Zaman et al. 7. The results of the
DOE study were reported by Henderson et al. 8.
The experimental results at NATR indicate that the wedge deflectors do reduce the noise on the lower side of the nozzle at
downstream angles. However, at other angles on the lower side
and at all angles on the sideline and upper side, the noise was
increased. In general, the overall estimated perceived noise level
EPNL increased relative to the baseline due to the higher noise
at the other locations. One configuration, BPR= 5.0, M = 0.0,
showed a small noise reduction of 0.8 EPN dB. Based on measurements of turbulent kinetic energy, Brown et al. 6 concluded
that the wedge designs tested were too aggressive.
Thrust performance is a critical factor in assessing low noise
nozzle concepts. Papamoschou 13 estimated the thrust loss in
nozzles with vane deflectors based on airfoil drag and nozzle flow
deflection 2, and Papamoschou et al. 9 experimentally examined the aerodynamics of a wedge deflector in a simplified configuration. But up to this point, there has been no detailed examination of the performance of these nozzles. The NATR tests did
not include force measurements. A key element of the CFD calculations presented here was to characterize the thrust of these
nozzles and compare them to a baseline nozzle without deflectors.
This work represents the highest fidelity estimates of Offset
Stream nozzle performance to date.
This paper describes the CFD analyses performed to support
the design and testing of turbofan nozzles with the wedge shaped
APRIL 2009, Vol. 131 / 041104-1
Fig. 1 Turbofan nozzle with wedge installed in the fan stream
flow deflectors. The CFD analyses consisted of three parts: 1

pretest predictions to verify the wedge design and its effect on the
flow field, 2 performance calculations to examine the performance penalties due to the noise reduction scheme, and 3 validation of the CFD methodology against experimental data. While
estimates of jet noise cannot be obtained directly from the CFD
analyses, the predicted levels of turbulent kinetic energy can be
used to infer the impact of the wedge deflector on the noise field
at a rudimentary level.
Description of Nozzles and Flow Conditions
The nozzles used in this study were originally designed for

NASA under the Advanced Subsonic Technology program for investigating mixing enhancement devices to reduce jet noise 10.
The baseline nozzles with no mixing devices were used here. For
this study a wedge flow deflector was placed in the fan stream.
The wedge sits completely inside the nozzle, and the base of the
wedge is coincident with the nozzle exit Fig. 1. Three different
nozzles were analyzed: 3BB, 4BB, and 5BB Fig. 2. The 3BB
and 5BB nozzles both have an external nozzle plug in the core
stream and have nozzle bypass ratios BPR of 5 and 8, respectively. These two nozzles share the same fan flow geometry. The
core flow nozzle section was replaced to affect the bypass ratio
change. The 4BB nozzle has an internal plug in the core flow and
has a bypass ratio of 8. This nozzle was studied computationally
but not tested in the experiment. All three nozzles have the same
fan diameter of 9.63 in. 24.5 cm.
The shape of the wedge deflector studied here is the same as the
wedge used by Papamoschou 13 in his initial studies. The
wedge has a half angle of 11 deg and a base width of 3.44 in.
8.74 cm. The upper and lower surfaces of the wedge conform to
the internal contours of the fan stream. The wedge reduces the exit
area of the fan stream by 14%.
It is anticipated that the wedge deflector would be incorporated
as a pair of flaps into the existing structure of the engine. For
underwing mounted engines found on most large transport aircraft, the flaps could be placed in the pylon. For other installations, side mounted or more highly integrated installations, the
flaps could be incorporated into a strut. In either case the flaps
would deploy to form a wedge on take-off and retract for cruise.
The pylons for underwing mounted engines already deflect some
fan flow. However, additional flow deflection is probably necessary to maximize noise reduction.
The calculations performed for the pretest predictions and performance analyses were done using cycle data for a representative
take-off condition. Details of the operating conditions are in Table
1. The experiment was run subsequent to the CFD analyses, and
the data were taken at slightly different conditions. The analyses
were rerun at these conditions for code validation purposes Table
2. For the 3BB validation case, the nozzle was tested with static
freestream conditions. This analysis was run with a freestream
Mach number of 0.05 for numerical stability. Two 5BB cases were
run, one with heated streams and one with cold streams. Both
cases had a freestream Mach number of 0.20. Because the wedge
deflectors will most likely be removed from the flow after takeoff, only that flight condition was analyzed.
Analysis Procedure
3.1 Geometry Definition. The computational grids Fig. 3

were generated using the Gridgen Version 15 software 11. The
database, which provides the geometry for the grid generator, was
taken directly from the computer aided drafting CAD file used to
design and build the experimental model hardware. The grids are
a structured multiblock type, with separate grid blocks represent-
Fig. 2 Schematic drawings of NASA GRC turbofan nozzles
041104-2 / Vol. 131, APRIL 2009
Table 1 Nozzle operating conditions for performance calculations

Fan
Nozzle
3BB
4BB
5BB
Core
Freestream
BPR
NPR
T0
K
U
m/s
NPR
T0
K
U
m/s
p
kPa
T
K
5.0
8.0
8.0
1.83
1.57
1.62
364
347
356
340
290
303
1.68
1.52
1.42
833
844
832
480
437
399
98.6
98.6
98.6
294
294
294
0.29
0.29
0.29
Table 2 Nozzle operating conditions for comparison to experiment

Fan
Nozzle
3BB hot
5BB hot
5BB cold
a
Core
Freestream
BPR
NPR
T0
K
U
m/s
NPR
T0
K
U
m/s
p
kPa
T
K
5.0
8.0
8.0
1.83
1.62
1.42
356
356
298
336
303
239
1.68
1.42
1.86
833
832
298
480
399
311
98.6
98.6
98.6
298
298
298
0.00a
0.20
0.20
CFD analysis used M = 0.05 for numerical stability.
ing the following: 1 the internal primary flow path, 2 the external primary flow path including the plug, 3 the internal secondary flow path, 4 the external secondary flow path, 5 the
nacelle, and 6 the jet plume region Fig. 4. Dimensions of the
grids are given in Table 3. At block interfaces the grids are not
contiguous, and a Roe-based zone coupling algorithm is used to
pass the fluxes between blocks.
The grids for the baseline nozzles with no wedge deflector were
planar, taking advantage of the axisymmetric geometry. A total of
66,734 points were used. The grids for the nozzles with the wedge
deflector were three dimensional: a 180 degree segment, taking
advantage of the symmetry about the xy-plane vertical streamwise plane. A total of 6,072,794 grid points were used.
was modeled using the Menter 13 shear stress transport SST

model. No additional corrections i.e., compressibility corrections
were used in the model.
Grid sequencing in all three computational directions was used
during the flow solution. Sequencing solves the equations on a
reduced grid to speed convergence. The sequenced results are then
interpolated onto the finer grid for continued processing. One
level of sequencing involves removing every other grid point from
the previous level. The solution started at two levels of sequencing
and progressed to the full grid solution. Nozzle force measurements were monitored, and the sequencing was reduced when the
nozzle force converged on that grid level. Nozzle force measurements between the final grid level and the next sequenced level
3.2 Flow Solution. The flow solver used for all the computations was Wind version 5.0 12. Wind is a general purpose Reynolds averaged NavierStokes RANS equation solver. All the
calculations presented here used the same code options. The default numerical scheme was used: implicit time stepping and a
second-order physical Roe scheme for stretched grids. Turbulence
Fig. 4 Grid blocking structure

Table 3 Dimensions of the computational grid
Block
1
2
3
4
5
6
Wedge 3D
Fig. 3 Grid of the 5BB nozzle with wedge installed in the fan
stream
Baseline
Axial
Radial
Azimuthal
Total
73
53
65
45
69
253
45
165
45
109
45
173
91
91
91
91
91
91
298,935
795,795
266,175
446,355
282,555
3,982,979
6,072,794
66,734
APRIL 2009, Vol. 131 / 041104-3
Fig. 5 Streamwise xy-plane contours of u-velocity, u / Uc, for 3BB nozzles
agreed within 1%, indicating that grid convergence was obtained.

Local time stepping was used to achieve a steady state solution.
For the 3BB case at near static ambient conditions M = 0.05,
the solution did not converge in this manner. For this case the
solution was switched to global time stepping, and convergence to
a steady state was achieved.
Results
4.1 Examination of the Flow Fields. The noise benefit of the

offset fan stream concept is predicated on reducing the convective
Mach number of the core stream instabilities by lengthening the
fan stream to cover or shield a larger portion of the high speed
041104-4 / Vol. 131, APRIL 2009
primary flow 14. The increased interface between the primary

and secondary streams reduces the region of large velocity gradients between the core and surrounding flow and thus reduces the
convective Mach number and the strength of the noise sources. In
the context of the CFD solutions presented, the reduced velocity
gradient lowers the resultant turbulent kinetic energy, which can
be related to the noise sources in jets 15. This effect can be seen
in the streamwise xy-plane velocity and turbulent kinetic energy
contours in Figs. 510. Data are presented for all three nozzles
3BB, 4BB, and 5BB, and the results show that all three flow
fields behave similarly.
Velocity contours in the streamwise planes compare the wedge
nozzle to the baseline nozzle Figs. 57. These contours show
that the presence of the wedge directs the core flow upward and
significantly shortens the core streams potential core. The layer of
fan flow on the underside of the jet plume is substantially thicker.
Fig. 8 Streamwise xy-plane contours of turbulent kinetic energy, k / Uc2, for 3BB nozzles
APRIL 2009, Vol. 131 / 041104-5
While it is difficult to quantify, the fan flow does appear to have

lengthened. This can be seen by the presence of large regions of
flow near the fan exit velocity u / Uc = 0.7 as far downstream as
x / D f = 6. A low-velocity region immediately behind the wedge
indicates that a large recirculation region is present there.
Turbulent kinetic energy contours in the streamwise planes are
shown in Figs. 810. On the upper side of the exhaust plume,
levels of k / U2c are increased and the peak level has moved up-
stream, closer to the nozzle exit. On the lower side of the exhaust
plume, the levels of k / U2c have been reduced. This is the desired
effect. A small local peak in turbulence levels is seen directly
behind the wedge, again indicating a separated region and potential noise source.
The movement of the fan flow induced by the wedge deflector
is more easily seen in cross-planes downstream of the nozzle exit
Figs. 1116. The fan flow moves to the lower half of the exhaust
041104-6 / Vol. 131, APRIL 2009
Fig. 11 Cross-plane yz-plane contours of u-velocity, u / Uc, for 3BB nozzles
plume, and at x / D f = 5.0 the fan stream is almost completely below the core flow. The upper portion of the core flow comes in
direct contact with the freestream flow and is not buffered by the
fan flow. This region of large velocity gradients results in high
turbulence levels and will certainly increase the noise on this side
of the nozzle. At x / D f = 3.0 high turbulence levels are present
from the top of the nozzle down to the sideline position 090
deg, measured from the vertical. The turbulent kinetic energy
levels resulting from the core stream interacting directly with the
freestream are on the order of an isolated jet operating at the core

stream conditions and could result in very high noise levels at
sideline and above the nozzle. The lower side has much lower
turbulence levels and should result in lower noise on this side of
the exhaust plume. These conclusions correspond well with the
findings in the experimental noise measurements 6. It appears
that the wedge moved too much fan flow, exposing the core flow
to the freestream. A less aggressive wedge design may improve
the noise characteristics of this device. Lower wedge angles
APRIL 2009, Vol. 131 / 041104-7
Fig. 14 Cross-plane yz-plane contours of turbulent kinetic energy, k / Uc2, for 3BB nozzles
should be investigated to reduce the deflection of the fan flow.

Also, shorter wedges that do not span the entire fan duct height
may prevent the core flow from being completely exposed to the
freestream.
4.2
Nozzle Performance
4.2.1 Calculation Procedure. Nozzle performance was computed using the Wind utility, CFPOST. Stream thrust was com-
puted at the fan and core exits. Pressure and viscous forces were
computed on the remaining surfaces: core nozzle lip, fan nozzle
lip, plug, splitter between fan and core, nacelle, and wedge base.
Nozzle performance will be reported in two ways. First, the
total thrust of the nozzle with wedge deflector will be expressed as
a fraction of the baseline nozzle thrust
F/Fbaseline
041104-8 / Vol. 131, APRIL 2009
Fig. 17 Ratio of wedge nozzle to baseline nozzle performance
Fig. 18 Nozzle performance expressed as a thrust coefficient
Second, it will be expressed in terms of a thrust coefficient. Typically nozzle performance results are reported using the gross
thrust coefficient,
C fg =
Factual
actualUideal
m
Using the actual mass flow in the denominator, C fg, removes the
deviation from the ideal mass flow from the performance measure.
Discharge coefficient is a separate measure for the reduction in
mass flow,
Cd =
actual
m
ideal
m
For subsonic nozzles the mass flow through the nozzle can vary
due to downstream effects, and the thrust coefficient does not give
a true measure of nozzle performance. In these circumstances, the
change in the measured thrust is caused by the change in the
nozzle flow rate and not a change in the exit velocity. This is
especially true for two-stream nozzles where the behavior of one
stream can affect the flow rate of the other. Nozzles with noise
suppressing features can also affect flow rates. Therefore, a thrust
coefficient that accounts for the change in mass flow rate is used
in this paper. It is simply a thrust coefficient based on the ideal
mass flow rather than the actual mass flow. It can be conveniently
written as the product of the gross thrust coefficient and discharge
coefficient,
actual Factual
Factual
m
=
= CdC fg
idealUideal m
ideal m
actualUideal
m
For the cases where the wedge is installed in the nozzle, the ideal
mass flow is computed using the reduced nozzle exit area.
A baseline case, no wedge, was calculated for each nozzle geometry. Performance of the wedge nozzles is referenced to the
baseline and is reported as CdC fg,
CdC fg = CdC fgwedge CdC fgbaseline
4.2.2 Performance Results. The thrust of the wedge nozzles

expressed as a ratio to the baseline nozzle thrust is shown in Fig.
17a. The wedge has created thrust losses between 11% and 13%.
The reason for these losses is clear. Figure 17b shows the loss in
mass flow of wedge nozzles, expressed as a ratio to the baseline
nozzles. Total mass flow reduction also varies between 11% and
13%. This reduction is due entirely to the reduction in fan mass
flow, which is approximately 14% for all three cases. These losses
are a direct result of the blockage created by the wedge 14% of
the fan exit area. For this noise reduction concept to be viable,
the engine would have to be oversized to accommodate the performance penalty at take-off. Reducing the size of the wedge, as
suggested for acoustic reasons, would reduce the thrust penalty.
An alternative concept, placing the wedge external to the fan
nozzle on the splitter, has been proposed and should eliminate this
loss of fan mass flow 16.
APRIL 2009, Vol. 131 / 041104-9
Fig. 19 Force breakout for 5BB baseline and wedge nozzles
Nozzle performance expressed as a thrust coefficient, CdC fg,

for both the baseline and wedge configurations for all three
nozzles is shown in Fig. 18a. For the wedge nozzles, this performance measure uses the ideal conditions based on the actual
exit area and therefore does not include the loss due to blockage.
The external plug nozzles have uniformly higher performance
than the internal plug nozzle. Figure 18b shows the decrement in
performance due to the wedge. The losses due to the wedge vary
Fig. 20 Streamlines for the 5BB nozzle
041104-10 / Vol. 131, APRIL 2009
Fig. 21 Component of the vertical thrust
from about 0.7% to 1.1% and appear to be relatively low, considering the large base area created by the wedge. A detailed breakout of the forces that comprise the thrust calculation is shown for
the 5BB baseline and wedge nozzles in Fig. 19. The figure confirms that the primary loss mechanism is the loss in fan stream
thrust due to the reduced mass flow. The presence of the wedge
reduces the positive force on the nozzle plug and increases the
drag on the nacelle. The base drag on the wedge surface is a
relatively minor component. This is somewhat surprising but can
be explained by the streamlines in Fig. 20. The figure shows that
the flow that separates and recirculates behind the base of the
wedge is the external flow over the nacelle. This also shows the
linkage between the presence of the wedge and the increased nacelle drag Fig. 19. The fan flow, the primary contributor to the
thrust, is turned by the wedge to create the offset stream but does
not separate behind the wedge, which would result in large thrust
losses.
By examining nozzle performance using the CdC fg parameter,
we can see that the losses created by the wedge deflector would be
acceptable if the reduction in mass flow could be removed or
accounted for in the nozzle design. The wedge deflector could be
Fig. 22 Comparison to experiment; u-velocity, u / Uc, for the 3BB nozzle with wedge, hot conditions
APRIL 2009, Vol. 131 / 041104-11
Fig. 23 Comparison to experiment; turbulent kinetic energy, k / Uc2, for the 3BB nozzle with wedge, hot
conditions
a viable noise reduction concept if the wedge size could be significantly reduced or if an external wedge that did not reduce the
fan flow rate was employed.
The wedge induces a downward movement of the fan flow and
a corresponding upward movement of the core flow. The resulting
vertical component of the nozzle thrust is presented in Fig. 21.
The data indicate that the wedge induces a very small vertical
component to the thrust. The external plug nozzles have a smaller
component, about 0.5% of the axial thrust, and the internal plug
has a slightly larger component, approximately 1.5% of the axial
thrust.
The wedge deflector was one of three devices used to offset the
fan stream examined in NASAs OST program. The performance
of the other two devices, vane deflectors and s-duct nozzles, were
also analyzed using a similar CFD method 5. The performance
of all three devices are compared in the Appendix of Ref. 17.
4.3 Comparison to Experiment. Once the experimental test
program was carried out, the CFD analyses were rerun to match
041104-12 / Vol. 131, APRIL 2009
the experimental conditions. The same computational grids, code

version, and code options were used. The experimental data were
withheld until the predictions were complete, and no attempts
were made to adjust code parameters to better match the data.
The nozzles were tested in the NASA Glenn NATR facility 4.
The NATR is a 53 in. 135 cm diameter free jet facility with a
maximum Mach number of 0.30. The nozzles were mounted in
the NATR on the high flow jet exit rig, which provides high pressure and high temperature air to both fan and core nozzle streams.
The experimental data consisted of particle image velocimetry
measurements of instantaneous velocity components in the nozzle
plume 18. The data were measured in planes parallel to the
nozzle exit at distances of 1.0, 1.5, 2.0, 3.0, 5.0, and 7.0 fan
diameters downstream of the fan nozzle exit. A complete description of the experiment including both the PIV shown here and the
acoustic measurements is given by Brown et al. 6. The measured
u-velocity axial component and turbulent kinetic energy were
Fig. 24 Comparison to experiment; u-velocity, u / Uc, for the 5BB nozzle with wedge, hot conditions
compared with the predictions for three wedge nozzle cases listed
in Table 2. Comparisons for the baseline nozzle cases can be
found in Ref. 17.
The 3BB nozzle was tested at conditions similar to the pretest
predictions, but with a static freestream. The CFD analysis used a
small freestream velocity, M = 0.05, to prevent numerical instabilities. Plots showing the comparison between prediction and experiment for both axial velocity and turbulent kinetic energy are
shown in Figs. 22 and 23. Near the fan exit plane at x / D f = 1.0 and
3.0, the proximity of the nozzle plug to the measurement planes
caused a low signal to noise ratio in the experimental data. These
data were removed, resulting in the white regions in the center of
the plots. The white areas at the corners of the experimental data
are due to the orientation of the nozzle relative to the PIV measurement plane. Overall, the agreement between CFD and experiment is very good. The contour plots indicate that the CFD does a
good job of predicting the shape of the nozzle plume and the
movement of diverted fan flow relative to the core. One does see
that the CFD predicts a more diffuse plume shape compared with
experiment. Line plots along the y-axis vertical axis show that
the CFD does a good job of predicting the levels of both velocity
and turbulent kinetic energy. For this case the Wind code predicts
higher peak velocities. This is due to the inclusion of the forward
velocity in the freestream for numerical reasons. This small coflow serves to reduce the nozzle mixing rate 19.
The 5BB nozzle was tested at two different flow conditions.
The quality of the PIV data near the nozzle exit was much better
with only small regions at x / D f = 1.0 where the data were removed. The first case had the same nozzle conditions as the pretest prediction, but the freestream Mach number was 0.20 instead
of 0.29. The comparison is presented in Figs. 24 and 25. For this
case agreement is very good for both velocity and turbulent kinetic energy. The contour plots show similar agreement in plume
shape. The line plots show that there is excellent agreement in the
prediction of both velocity and turbulence levels. The improved
agreement in the levels over the 3BB case is most likely due to the
fact that the CFD analysis matched the experimental freestream
velocity. The second 5BB case was tested at cold nozzle temperatures and a Mach 0.20 freestream. The comparison for this case is
presented in Figs. 26 and 27. Again, for all aspects agreement
APRIL 2009, Vol. 131 / 041104-13
Fig. 25 Comparison to experiment; turbulent kinetic energy, k / Uc2, for the 5BB nozzle with wedge, hot
conditions
between experiment and predictions is very good.

In general for all the cases, very good agreement is achieved
between the CFD and experiment in terms of plume shape, velocity levels, and turbulent kinetic energy levels. The CFD predicts
slightly higher values of velocity and slightly lower values of
turbulent kinetic energy at the downstream locations. This behavior is typical of RANS CFD predictions for jets using twoequation turbulence models 20. The baseline nozzle cases reported in Ref. 17 similarly exhibit very good agreement. The
favorable comparison between the analysis and the experiment is
very encouraging and indicates that this RANS CFD technique
can be used effectively to optimize the wedge nozzle design and
accurately predict the resulting flow fields.
Summary and Conclusions
Detailed Reynolds averaged NavierStokes calculations of turbofan nozzles with wedge deflectors for noise reduction were carried out to support NASAs Offset Stream Technology program.
The nozzles were designed to reduce jet noise at take-off condi041104-14 / Vol. 131, APRIL 2009
tions. The computational fluid dynamics calculations were intended to characterize the exhaust plume aiding in wedge design,
predict nozzle thrust performance, and validate the computational
technique against experimental data.
The predicted flow fields showed that the wedge deflector was
effective at redirecting the fan flow to the underside of the exhaust
plume, effectively reducing noise on this side of the nozzle. The
predictions indicate that the wedge design is too aggressive, deflecting too much flow and exposing the core flow to the
freestream on the top and sides of the plume. Predictions of turbulent kinetic energy suggest that noise levels at the sideline and
upper side of the nozzle may approach those of an isolated jet at
core flow conditions.
Nozzle performance calculations show that the thrust of the
nozzle is significantly reduced relative to an unmodified nozzle.
This thrust loss is caused by a reduction in mass flow of the fan
stream, which is directly proportional to the area reduction of the
fan exit caused by the wedge. Thrust losses imparted by the
wedge not due to the fan flow reduction are relatively small. If the
Fig. 26 Comparison to experiment; u-velocity, u / Uc, for the 5BB nozzle with wedge, cold conditions
loss in mass flow could be removed or accounted for in the nozzle

design, the wedge concept would be viable in terms of performance.
Two of the nozzles considered here were tested at the NASA
Glenn NATR. The calculations were rerun to match the experimental test conditions. Predictions of axial velocity and turbulent
kinetic energy in several downstream planes parallel to the nozzle
exit were compared with experimentally obtained data. The CFD
did a very good job of predicting the offset stream nozzles plume
shape. Also, the agreement between the experiment and predicted
levels of both axial velocity and turbulent kinetic energy was very
good for all nozzle configurations, locations, and flow conditions
considered.
Acoustic testing of the nozzles considered here showed that
they did not achieve the desired noise reduction. The analyses
presented here suggest a smaller, less aggressive wedge diverter
would improve the thrust and noise characteristics. This echoes
the conclusion from the experimental program. Based on the predicted flow fields, performance calculations, and agreement with
experimental data, this study shows that the CFD methodology

used here can be used to improve this nozzle concept.
Nomenclature
k
m
p
u
x,y,z
BPR
Cd
C fg
D
F
M
NPR
T
U
turbulent kinetic energy

mass flow
pressure
axial component of velocity
Cartesian coordinates
bypass ratio
discharge coefficient
thrust coefficient
diameter
nozzle force thrust
Mach number
nozzle pressure ratio
temperature
ideal exit velocity
APRIL 2009, Vol. 131 / 041104-15
Fig. 27 Comparison to experiment; turbulent kinetic energy, k / Uc2, for the 5BB nozzle with wedge, cold
conditions
Subscripts
actual
baseline
ideal
c
f
0
measured quantity
nozzle without flow deflectors
isentropically expanded to freestream condition
core primary
fan secondary
stagnation/total condition
freestream condition
References
1 Papamoschou, D., 2002, Noise Suppression in Moderate-Speed Multistream
Jets, AIAA Paper No. 2002-2557.
2 Papamoschou, D., 2004, New Method for Jet Noise Reduction in Turbofan
Engines, AIAA J., 4211, pp. 22452253.
3 Papamoschou, D., and Nishi, K., 2005, Jet Noise Suppression With Fan Flow
Deflectors in Realistic-Shaped Nozzle, AIAA Paper No. 2005-993.
4 Castner, R., 1994, The Nozzle Acoustic Test Rig; An Acoustic and Aerodynamic Free-Jet Facility, NASA Report No. TM 106495.
5 Dippold, V., Foster, L., and Wiese, M., 2007, Computational Analyses of
Offset Stream Nozzles for Noise Reduction, AIAA Paper No. 2007-3589.
041104-16 / Vol. 131, APRIL 2009
6 Brown, C., Bridges, J., and Henderson, B., 2007, Offset Stream Technology
TestSummary of Results, AIAA Paper No. 2007-3664.
7 Zaman, K., Bridges, J., and Papamoschou, D., 2007, Offset Stream
TechnologyComparison of Results From UC and GRC Experiments, AIAA
Paper No. 2007-438.
8 Henderson, B., Norum, T., and Bridges, J., 2006, An MDOE Assessment of
Nozzle Vanes for High Bypass Ratio Jet Noise Reduction, AIAA Paper No.
2006-2543.
9 Papamoschou, D., Vu, A., and Johnson, A., 2006, Aerodynamics of WedgeShaped Deflectors for Jet Noise Reduction, AIAA Paper No. 2006-3655.
10 Janardan, B. A., Hoff, G. E., Barter, J. W., Martens, S., Gliebe, P. R., Mengle,
V., and Dalton, W. N., 2000, AST Critical Propulsion and Noise Reduction
Technologies for Future Commercial Subsonic Engines Separate-Flow Exhaust System Noise Reduction Concept Evaluation, NASA Report No. CR
2000-210039.
11 Pointwise, Inc., http://www.pointwise.com/gridgen/
12 The NPARC Alliance, http://www.grc.nasa.gov/www/winddocs/
13 Menter, F. R., 1994, Two-Equation Eddy-Viscosity Turbulence Models for
Engineering Applications, AIAA J., 328, pp. 15981605.
14 Papamoschou, D., 2003, A New Method for Jet Noise Reduction in Turbofan
Engines, AIAA Paper No. 2003-1059.
15 Khavaran, A., and Kenzakowski, D. C., 2007, Noise Generation in Hot Jets,
NASA Report No. CR 2007-214924.
16 Papamoschou, D., 2006, Fan Flow Deflection in Simulated Turbofan Exhaust, AIAA J., 4412, pp. 30883097.
17 DeBonis, J., 2008, RANS Analyses of Turbofan Nozzles With Wedge Deflectors for Noise Reduction, AIAA Paper No. 2008-0041.
18 Wernet, M., and Bridges, J., 2002, Application of DPIV to Enhanced Mixing
Heated Nozzle Flows, AIAA Paper No. 2002-0691.
19 Dembowski, M. A., and Georgiadis, N. J., 2002, An Evaluation of Parameters

Influencing Jet Mixing Using the WIND Navier-Stokes Code, NASA Report
No. TM 2002-211727.
20 Georgiadis, N. J., and DeBonis, J. R., 2006, Navier-Stokes Analysis Methods
for Turbulent Jet Flows With Application to Aircraft Exhaust Nozzles, Prog.
Aerosp. Sci., 42, pp. 377418.
APRIL 2009, Vol. 131 / 041104-17
Theo Van Holten
Choking Phenomena in a Vortex

Flow Passing a Laval Tube: An
Analytical Treatment
e-mail: t.vanholten@tudelft.nl
Monique Heiligers
Annemie Jaeken
Department of Design, Integration and Operation
of Aircraft and Rotorcraft,
Faculty of Aerospace Engineering,
Delft University of Technology,
Kluyverweg 1,
2629 HS Delft, The Netherlands
The behavior of a vortex flow through a Laval nozzle was studied in connection with the
purification of natural gas. By creating a vortex and passing it through a Laval nozzle,
the gas will be cooled, and water droplets will form and will be centrifuged out of the
gas. This system is named the Condi-Cyclone. An analytical theory is developed to reveal
the most important phenomena of the flow, to first order accuracy. Experiments have been
performed with a prototype of the Condi-Cyclone. A Euler numerical simulation was
performed, using the geometry of the test channel. This paper presents an analytical
theory for a vortex flow through a Laval nozzle. It will demonstrate that when a vortex is
present the total velocity reaches sonic conditions upstream of the nozzle throat, that the
axial component of the velocity in the nozzle throat is equal to the local speed of sound
and that the mass flow through the Laval nozzle decreases with increasing vortex
strength. The predictions of the analytical theory have been compared with the results of
the experiments and the Euler numerical simulation, and it can be concluded that the
analytical theory describes the main characteristics of the flow very well.
DOI: 10.1115/1.3089532
Keywords: Laval nozzle, vortex flow, natural gas, Condi-Cyclone
Introduction
The behavior of a vortex flow through a Laval nozzle was studied in connection with the purification of natural gas. Water normally present in natural gas needs to be extracted because it can
have damaging effects on the installations and pipelines further
downstream. The steps of dehydration and hydrocarbon removal
are normally achieved in multiple process stages, for example, in
turbo-expanders or with added chemicals, notably glycol or
methanol 1. In addition to requiring large vessels and complex
operation, processes such as glycol regeneration can also produce
toxic atmospheric emissions, including benzene, toluene, ethylene, and xylene 1.
To solve these problems a radical new means to purify natural
gas has been invented: the Condi-Cyclone. By creating a vortex in
a Laval nozzle, the gas will be cooled, and water droplets will
form and will be centrifuged out of the gas. This paper will
present an analytical theory describing the basic process of a vortex flow through a Laval nozzle.
Section 2 will provide a brief explanation of the Condi-Cyclone
and its lay-out. Section 3 will present the actual analytical theory.
To verify the predictions of the analytical theory, both Euler numerical simulation and full scale tests with the Condi-Cyclone
have been performed. The results of the full scale tests will be
presented in Sec. 4; the results of the numerical simulation in Sec.
5. The different results will be compared with the analytical
theory in Sec. 6, and finally the conclusions are presented in Sec.
7.
Condi-Cyclone Lay-Out
The Condi-Cyclone accelerates the natural gas to supersonic

levels and at the same time introduces a strong vortex. By accelerating the gas, the gas will cool down, the water will start to
received January 31, 2008; published online March 6, 2009. Assoc. Editor: Phillip M.
Ligrani. Paper presented at the 45th AIAA Aerospace Sciences Meeting and Exhibit
2007.
condensate, and small water droplets will form. These water droplets will be centrifuged to the wall of the Condi-Cyclone by the
vortex and can then be extracted from the flow. A schematic layout of the Condi-Cyclone is given in Fig. 1.
After the natural gas enters the inlet of the Condi-Cyclone, a
static center-body directs the gas flow past a static array of blades.
This static array of blades deflects the flow and thereby generates
the vortex. This part of the Condi-Cyclone is also referred to as
the subsonic vortex generator SVG. The next step is an acceleration of the flow by means of a Laval nozzle. The flow will
become supersonic, and when the temperature drop is sufficient,
water droplets will start to form. When the water droplets have
grown sufficiently, they are driven toward the nozzle wall by the
centrifugal forces and are extracted from the flow in the recovery
section. When the flow has passed the recovery section, it is
slowed down. The cleaned gas then flows straight into the remaining processes.
Analytical Theory
The purpose of the analytical theory is to reveal the most important phenomena of the flow, to first order accuracy. In the case
of rocket nozzles and the design of supersonic windtunnels, it is
well known that the assumption of quasi-one-dimensional isentropic flow may give a useful first indication of the global flow
properties. For this reason it was attempted to develop a theory for
the Condi-Cyclone flow based on similar assumptions. The validity of such an extremely simplified theory will depend on how
strong viscous effects are or on how much the flow is affected by
local shocks. Comparisons with experiments Sec. 4 are therefore
essential before conclusions can be drawn concerning the validity
of the assumptions. Further insight into details such as the occurrence of local shocks will be provided by Euler numerical simulations Sec. 5. The analytical theory for a vortex flow through a
Laval nozzle will be presented in this section. Among others it
will be shown that the total velocity reaches the local speed of
sound upstream of the throat and that the mass flow is a function
of the vortex strength.
APRIL 2009, Vol. 131 / 041201-1
Fig. 1 Schematic lay-out of the Condi-Cyclone
3.1 Basic Equations and Assumptions. The following basic

assumptions have been made. The flow is considered to be isentropic and a calorically perfect gas is assumed, implying that the
following flow equations can be used:
1 2
Tt
M
=1+
2
T
1 2
pt
M
= 1+
2
p
1 2
t
M
= 1+
2
1 2
M
a = RT = RTt 1 +
2
2
at2
M2 =
at2
1 2
M
1+
2
V2
=
a2
1/1
/1
Fig. 2 Definition of annulus and corresponding velocities
A=
m
Vax
which yields upon differentiation.

4
in which Tt is the total temperature, T is the static temperature,

is the ratio of specific heats, M is the Mach number, pt is the total
pressure, p is the actual pressure, t is the total air density, is the
actual air density, a is the speed of sound, at is the total speed of
sound, and V is the velocity. Furthermore, the flow is considered
to be rotationally symmetric. Quasi-one-dimensional flow is assumed, i.e., it is assumed that the flow properties depend on the
axial coordinate only. For the flow around the central body, this is
an acceptable assumption; the variation in the velocity in the radial direction will indeed be very small in the narrow annular
channel. For the flow downstream of the center-body this assumption will not be valid anymore; therefore in this part of the Laval
nozzle the analytically calculated velocities will merely be treated
as an indication of the average velocities.
The definition of the annulus and the corresponding velocities
are clarified in Fig. 2. Vax is the axial velocity component, parallel
to the axis of the Laval nozzle. Vtan is the tangent velocity due to
the vortex. It is tangent to the mean radius of the annulus r. The
angle between the total velocity V and the axial velocity Vax is ,
and the area of the annulus is A.
The basic equations that will be used for the analytical calculations are conservation of angular momentum and the continuity
equation:
By definition the throat is given by dA / dx = 0. This implies that

dr / dx = 0 and thus, on account of Eq. 6, dVtan / dx = 0. Equation
5 states,
1 2
V
2
Vtan
+ V2ax
1 2
Vtan + V2ax
at2
2
1 dVax
1 d M 2 dVax M 2 dVtan
m
dA
+
=
+
dx
Vax dM 2 Vax dx
Vtan dx
Vax dx
10
Substituting the throat conditions dA / dx = 0 and dVtan / dx = 0

yields,
0=
1 d M 2
1
+
2
dM Vax
Vax
11
According to Eqs. 3 and 4 the derivative d / dM 2 is given by
1 2
1
d
M
= t 1 +
2
2
dM 2
/1
1 2
1
M
= 1+
2
2
1 a2
= 2
2 at
12
The other derivative M 2 / Vax occurring in Eq. 11 is found

using Eqs. 7 and 8:
M2
=
Vax
2Vaxat2
1 2
Vtan + V2ax
at2
2
= 2Vax
at2
a4
13
After substituting Eqs. 12 and 13 into Eq. 11, the following

result is found for the flow in the throat:
14
= VaxA = constant
m
Vax = athroat
Vtanr = constant
I=m
is the mass flow and I is the angular momentum.

in which m
The total Mach number of the flow in the throat section, taking
into account that there is an additional velocity component Vtan,
apparently is larger than unity.
3.2 Throat Conditions. As a preliminary the local Mach

number is expressed as a function of the velocity components Vax
and Vtan. Using Eqs. 1 and 4,
In fact, the Mach number in the throat is uniquely determined by

the value of Vtan, as shown by the following expressions:
at2 a2 = RT
1 2 1 2
Tt
M =
V
1 = a2
2
2
T
so that
041201-2 / Vol. 131, APRIL 2009
M thr 1
2
=
M thr
if Vtan 0
2
2
Vtan
Vtan
+ V2ax
1 2
M thr
=1+ 2 1+
2
2
a
at
15
from which M thr can be made explicit as follows:

= blM blabl cos blAbl = tat cos blAblM bl

m
1+
1 2
M bl
2
+1/21
20
Likewise the angular momentum may be expressed in terms of the

flow condition just downstream of the vortex generator as follows:
at sin blrblM bl 1 +
M blabl sin blrbl = m
I=m
1 2
M bl
2
Fig. 3 Flow angle just downstream of the subsonic vortex

generator and the resulting velocities
1+
2
M thr
=
2
Vtan
at2
2
1 Vtan
2 at2
16
The latter equation again clearly shows that the Mach number in
the throat is larger than unity if a vortex flow is present. Under
such conditions the transition from sub-to supersonic flow occurs
at a position upstream of the throat.
3.3 Mass Flow Through the Channel. Making use of the
property 14 and Eq. 5, the mass flow through the throat section
and therefore the mass flow in every section may be expressed
as follows:
= VaxA = thrathrAthr
m
17
Using Eqs. 1, 3, and 4, the right-hand side of this equation

can be written as a function of M thr only.
= tatAthr 1 +
m
1 2
M thr
2
+1/21
18
If, for example, is taken equal to the value for air = 1.4, the
exponent in Eq. 18 will equal to 3. It can thus be seen that
increasing M thr will lead to a reduced mass flow. In other words,
increasing the vortex strength will throttle the Laval nozzle, as a
consequence of the additional Vtan and the resulting increased
Mach number in the throat. This throttling effect is quite large, as
will be seen later from some experimental results used to determine the validity of the above given theory.
Using the average radius rthr of the flow annulus in the throat
see the earlier remarks about the quasi-one-dimensional approximation underlying the developed theory, Eq. 6 yields for the
conditions in the throat:
Vtan,thr =
I
rthr
m
3.4 Flow Conditions at the Blades of the Vortex

Generator. A second relation between the mass flow and angular
momentum is provided by a consideration of the flow through the
array of blades at the entrance of the channel see Fig. 3, called
the subsonic vortex generator. The blade angle determines the
flow angle bl just downstream of the subsonic vortex generator.
In terms of the flow properties just downstream of the blades,
the mass flow Eq. 5 may be written as
1/2
21
, I, and M bl,
Equations 20 and 21 contain the three variables m
apart from the constants that specify the reservoir- or total- conditions of the gas and the geometry of the channel. The set Eqs.
20 and 21 therefore provides a second relation between mass
flow and angular momentum.
At this point there are two separate relations available between
and I; the first one derived from the throat conditions and the
m
second one given by the inflow conditions. The calculation of the
and I can thus be completed.
values of m
3.5 Computational Scheme to Determine Mass Flow and
Angular Momentum. The two relations between mass flow m

and angular momentum I provided by the two sets of equations
16, 18, and 19 on the one hand, and Eqs. 20 and 21 on the
other hand consist of nonlinear algebraic equations and do not
allow for an explicit solution. The following is an efficient iterative calculation scheme:
Step 1. Guess a value of M thr.
Step 2. The corresponding estimate of the mass flow is given by
Eq. 18, describing the throat condition.
Step 3. Eq. 20 will provide the Mach number just downstream
of the vortex generator. Equation 20 may conveniently be solved
for M bl using a NewtonRaphson procedure.
Step 4. The angular momentum I brought into the flow by the
vortex generator is given by Eq. 21.
Step 5. The tangential velocity in the throat Vtan,thr is known
from Eq. 19
Step 6. An improved estimate for M thr is finally obtained from
Eq. 16, and, if necessary, the iteration can proceed by returning
to step 1.
3.6 Calculation of Flow Conditions at Each Axial Station.
Now the mass flow and the angular momentum have been calculated, and since the flow area Ax and radius rx as functions of
the axial coordinate x are given, the flow conditions at each axial
station can be calculated. The tangent velocity is still given by Eq.
6. The total velocity, axial velocity, and Mach number at each
axial station can be calculated by combining the following four
equations:
19
The set of equations 16, 18, and 19 determines a relation

and the angular momentum I, although
between the mass flow m
an actual calculation will be iterative. In order to complete the
quantification of the flow properties, a second relation between m

and I will be needed.
Vx = Mxax =
Mxat
1+
1 2
M x
2
2
x + V2axx
Vx = Vtan
22
23
Equations 22 and 23 give the following expression for the

axial velocity:
Vaxx =
Mx2at2
2
x
Vtan
1 2
M x
1+
2
24
An expression for the Mach number can be derived from Eq. 9

by substituting Eqs. 24 and 3, as follows:
APRIL 2009, Vol. 131 / 041201-3
Table 1 Input and output of the analytical calculations
Table 2 Experimental results
Analytical results
Experimental results
Input
Output
bl
deg
pt
bar
Tt
K
m
kg/s
m2 / s
Tend nozzle
K
M end nozzle
20
25
30
40
1.5
1.5
1.5
1.5
300
300
300
300
1.07
0.98
0.89
0.73
25.10
28.91
32.27
38.14
180
175
165
155
1.9
1.9
2.0
2.2
t 1 +
1 2
M x
2
1/1
M 2xat2
m
2
x =
Vtan
1 2
Ax
M x
1+
2
25
Solving Eq. 25 will give the Mach number for each axial coordinate: substitution of the Mach number for each axial coordinate
into Eq. 22 will give the total velocity for each axial coordinate.
Finally substitution of the Mach number into Eq. 24 will give
the axial velocity for each axial coordinate.
3.7 Analytical Calculations and Results. Using the analytical theory calculations have been performed on the CondiCyclone test setup that was used during the experiments see Sec.
4. Section 4 will also describe the geometry of the CondiCyclone. Here, the analytical results for four different blade
angles bl are already given see Table 1. The input parameters
are the total pressure and total temperature at the location of the
blades of the subsonic vortex generator. The output parameters are
the mass flow, vortex strength , temperature at the end of the
nozzle Tend nozzle, and Mach number at the end of the nozzle
M end nozzle.
Experiments
Experiments have been performed with a prototype of the

Condi-Cyclone. The purpose of these tests was to obtain an impression about the validity of the above described simplified analytical theory, taking into account the limited scope of such a
theory as stated in the introduction of Sec. 3. Therefore, tests were
run with dry air. Since the mass flow is of primary importance for
the designer, the measurements concentrated on this property.
Mass flow could be determined relatively accurately 2% from
the total and static pressures in the subsonic part of the channel
upstream of the subsonic vortex generator. As a matter of interest
an attempt was done to get an impression of the angular momentum and the variation in the Mach number along the length of the
channel. It should be realized, however, that the latter properties
vary strongly in the radial direction in the region of the throat see
Sec. 5, so that very accurate measurements were not considered
SVG
deg
pt
bar
Tt
K
2%
m
kg/s
20%
m2 / s
20
25
30
40
1.5
1.5
1.5
1.5
300
300
300
300
1.04
0.95
0.887
No result
26
24
31
No result
to be useful; results are therefore not discussed in this paper. This

section will describe the test setup and test conditions, followed
by the experimental results.
4.1 Description of the Experiment. Figure 4 shows the general geometry of the channel. The step in the radial contour of the
center-body has no significance. It was caused by the fact that the
hardware for the test made use of parts of an earlier experiment
with different dimensions. Within the scope of the analytical
and the angular momentum I are detertheory the mass flow m
mined solely by the entrance conditions and the dimension of the
throat. For the purpose of comparison with the analytical theory,
the experimental setup is therefore defined by the following data.
4.1.1 Entrance Geometry. The radius of the center-body at the
station of the vortex generator was 0.23 m, with a channel width
at that station equal to the span of the blades of the vortex generator of 0.01 m.
The blades of the array consisted of simple cambered plates.
Due to the camber the deflection angle of the flow was slightly
larger than the nominal setting, the latter defined by the orientation of the chord. The nominal setting angles were 20 deg, 25 deg,
30 deg, and 40 deg. The effective deflection angles were estimated
to be 2 deg larger in each case.
4.1.2 Throat. The entire channel had a circular cross section.
The narrowest part of the channel located at the 0.5 m station
had a radius of 0.035 m, although boundary layer effects, in principle, slightly affect the effective radius. The effect of viscosity
might show up in the measured mass flow. In order to judge the
magnitude of this effect when compared with the experimental
results, the throat radius of 0.035 m was substituted in the analytical calculations.
4.1.3 Reservoir Conditions. The experiment was fed by pressurized air, which could be considered to be sufficiently clean and
dry for this purpose. Total pressure was pt = 1.5 105 Pa, and total temperature was Tt = 300 K.
4.1.4 Instrumentation. An array of static pressure holes, running from the subsonic part of the channel through the throat to a
station well downstream of the throat section, provided the information needed to determine the mass flow and to obtain an impression about the variation of the average Mach number in the
Fig. 4 General geometry of the channel with axisymmetric mesh. The zero station is located at the end of the center-body;
the throat is located at the 0.5 m station.
041201-4 / Vol. 131, APRIL 2009
Fig. 5 Axisymmetric mesh; zoomed in at the end of the center-body
channel.
A crude check on the magnitude of the swirl was made by
mounting a three-hole pressure tube in the widest part of the channel near the aft end of the central body. This pitot tube could be
rotated in such a way that it was roughly aligned with the flow
direction. From this device the flow angle was derived on the
mean radius of this particular station.
The measured vortex strength was expressed in terms of the
circulation defined as
= 2rVtan = 2rblM blabl sin bl
26
The expected accuracy is not more than 20% due to the uncertainties in the measurement techniques. On account of the quasi-onedimensional assumption underlying the analytical theory, the circulation is related to the angular momentum I by the expression
I=
27
4.2 Experimental Results. Experiments with the CondiCyclone have been performed at the Stork Product Engineering
TM620-Air/Water test facility 2,3. The results of the test for
different blade angle deflections are given in Table 2. With increasing blade deflection angle and thus with increasing vortex
strength, the mass flow through the nozzle decreases. At the highest blade deflection angle 40 deg the test setup started to vibrate
heavily, and the mass flow was severely limited. Because of the
vibrations the test was aborted and therefore no results are available for this blade deflection angle. Either vortex breakdown or
the stalling of the blades of the SVG is thought to be the probable
cause for these vibrations.
Euler Numerical Simulation
The Euler numerical simulation 4 was performed using the

geometry of the test channel. The purpose was to obtain information on the radial variations in the flow, which may be expected to
have occurred during the experiment. Another goal was to check
whether local shocks on the tail of the center-body could have
occurred, thus causing the breakdown of the experiment with the
largest vane angles. For this specific purpose the Euler numerical
simulation was considered adequate.
5.1 Mesh Generation. Meshes have been generated to perform dry air Euler numerical simulations for the Condi-Cyclone
Fig. 6 Tangential velocity horizontal axis as a function of radial distance vertical axis
for several stations
APRIL 2009, Vol. 131 / 041201-5
Table 3 Input and output of the Euler numerical simulation

Numerical results
Input
Output
SVG
deg
M bl
pt
bar
Tt
K
m
kg/s
m2 / s
Tend nozzle
K
M end nozzle
20
25
30
40
0.14
0.13
0.13
0.12
1.5
1.5
1.5
1.5
300
300
300
300
1.12
1.02
0.91
0.69
23.31
27.37
32.68
39.07
186
180
174
168
1.75
1.80
1.85
1.95
SVG with various blade deflection angles. An axisymmetric mesh

was used starting just downstream of the static array of blades
and running through the throat until well downstream of the
throat.
Because an axisymmetric flow is assumed, it was only necessary to generate a two-dimensional mesh. The mesh is a structured
H-type mesh with dimensions of 124 24 cells for the first block
running from just downstream of the blades until halfway of the
center-body. The remainder of the channel consisted of four
blocks of each 108 24 cells. The distribution of mesh points has
been chosen in such a way that the width of the cells approximates
the height of the cells and that a refinement can be found at the
end of the center-body and along the nozzle axis of symmetry see
Fig. 5. Numerical simulations have also been performed with
finer meshes. However mesh convergence was reached, and therefore results for these finer meshes have been omitted. The inflow
boundary conditions, just downstream of the swirl vanes, were
chosen to correspond to the flow conditions determined by the
analytical calculations.
5.2 Results of Numerical Analysis. The main findings of the
numerical simulations are the following: The vortex causes the
flow to reach sonic conditions well upstream of the nozzle throat
in the region of the tail of the center-body. In the nozzle throat the
axial velocity is equal to the local speed of sound, as predicted by
the analytical theory. In the nozzle a vortex structure is found with
a solid body rotation at the axis of symmetry and a 1 over r
behavior toward the nozzle wall see Fig. 6. A severe pressure
drop is present in the region of the solid body rotation.
With increasing blade deflection angle the mass flow decreases,
the vortex strength increases, a higher Mach number at the end of
the nozzle is achieved, and a slightly larger temperature drop occurs see Table 3. With increasing blade deflection angles the
solid body rotation region expands, the supersonic region at the
end of the center-body moves upstream, and the effective cross
section of the channel decreases. At the highest vane angle of 40
deg a flow disturbance at the very end of the center-body was
clear see Fig. 7, showing up as a region of back-flow near the
center-line and possibly involving a small shock. It was therefore
recommended to use in later configurations center-bodies with a
pluglike tail shape.
6 Comparison of Analytical, Numerical, and Experimental Results

The results of the analytical theory, experiments, and numerical
simulation as explained in Secs. 35 will be compared.
6.1 Mass Flow and Vortex Strength. The mass flow and
vortex strength as predicted by the analytical theory, calculated by
the numerical simulation, and measured during the experiments
are plotted as a function of the blade deflection angle in Fig. 8. As
can be seen from this figure, the results correspond reasonably
well, especially when taking into account that the experimentally
obtained values for the vortex strength have an accuracy of only
20%. The measured mass flow does not seem to be affected appreciably by viscosity effects.
6.2 Axial Velocity, Temperature, and Mach Number. Both
the analytical results and the numerical results show that the total
Fig. 7 Calculated back-flow at the end of the center-body
041201-6 / Vol. 131, APRIL 2009
Fig. 8 Mass flow and vortex strength as a function of the blade deflection angle
Fig. 9 Temperature and Mach number at the end of the nozzle as a function of the blade deflection angle
Mach number becomes supersonic well before the throat, just upstream of the end of the center-body. Both results also show that
the axial velocity in the throat is equal to the local speed of sound.
The Mach number and temperature at the end of the nozzle station 1.3 m downstream of the center-body as predicted by the
analytical theory and calculated by the numerical simulation are
plotted in Fig. 9.
Conclusions
The analytical theory developed to describe a vortex flow

through a Laval nozzle predicts that the axial velocity in the throat
will be equal to the local speed of sound and that the total velocity
will be larger than the local speed of sound. The total velocity will
reach M = 1 upstream of the throat. Furthermore, the theory predicts that the mass flow through the nozzle depends on the conditions in the throat and also on the vortex strength. The predictions
from a simplified analytical theory have been compared with the
results of a numerical simulation and experiments and correspond
very well.
Epilogue
After some modifications were applied to the Condi-Cyclone a

change in the end of the center-body and a redesign of the static
array of blades, the concept performed very well during the experiments. The Condi-Cyclone concept has been patented by
Stork Product Engineering and has been further developed by
Twister BV to make the technology commercially available to the
wider industry. A Twister demonstration unit has been operational
in Nigeria since early 2001, successfully performing water and
hydrocarbon separation, while confirming the ability of Twister to
run virtually without supporting systems Twister BV. Earlier
units have been in operation in the Netherlands in Zuiderveen en
Barendrecht since 1997 5.
References
1 Knott, T., 2000, Twist in the Tale, Offshore Engineer, http://
www.oilonline.com/news/features/oe/20000701.Twist_in.67.asp.
2 Mertens, S., 1999, Test Report: Subsonic Vortex Generation 20 & 40 Degrees Turbine Ring, Stork Product Engineering, Amsterdam, The Netherlands, Technical Report No. SPE-CON-RP-102 1/0.
3 Mertens, S., 1999, Test report: Subsonic Vortex Generation 25 & 30 Degrees
Turbine Ring, Stork Product Engineering, Amsterdam, The Netherlands,
Technical Report No. SPE-CON-RP-113 1/0.
4 Schoones, M. M. J., and Van Holten, Th., 1999, Euler Numerical Simulation
of Channel Flow With a Superimposed Vortex, Faculty of Aerospace Engineering, Delft University of Technology, The Netherlands, Technical Report
No. FM&P-99.001.
5 Twister, B. V., http://www.twisterbv.com/
APRIL 2009, Vol. 131 / 041201-7
Mohsen Akbari1
Ph.D. Candidate
Mechatronic System Engineering,
School of Engineering Science,
Simon Fraser University,
Surrey, BC, V3T 0A3, Canada
e-mail: maa59@sfu.ca
David Sinton
Associate Professor
Department of Mechanical Engineering,
University of Victoria,
Victoria, BC, V8W 2Y2, Canada
Majid Bahrami
Assistant Professor
Mem. ASME
Mechatronic System Engineering,
School of Engineering Science,
Simon Fraser University,
Surrey, BC, V3T 0A3, Canada
Pressure Drop in Rectangular

Microchannels as Compared With
Theory Based on Arbitrary Cross
Section
Pressure driven liquid flow through rectangular cross-section microchannels is investigated experimentally. Polydimethylsiloxane microchannels are fabricated using soft lithography. Pressure drop data are used to characterize the friction factor over a range of
aspect ratios from 0.13 to 0.76 and Reynolds number from 1 to 35 with distilled water as
working fluid. Results are compared with the general model developed to predict the fully
developed pressure drop in arbitrary cross-section microchannels. Using available theories, effects of different losses, such as developing region, minor flow contraction and
expansion, and streaming potential on the measured pressure drop, are investigated.
Experimental results compare well with the theory based on the presure drop in channels
of arbitrary cross section. DOI: 10.1115/1.3077143
Introduction
Advances in microfabrication make it possible to build microchannels with micrometer dimensions. Micro- and minichannels
show potential and have been incorporated in a wide variety of
unique, compact, and efficient cooling applications in microelectronic devices 1. These microheat exchangers or heat sinks feature extremely high heat transfer surface area per unit volume
ratios, high heat transfer coefficients, and low thermal resistances.
In biological and life sciences, microchannels are used widely for
analyzing biological materials such as proteins, DNA, cells, embryos, and chemical reagents 2. Various microsystems, such as
microheat sinks, microbiochips, microreactors, and micronozzles
have been developed in recent years 36. Since microchannels
are usually integrated into these microsystems, it is important to
determine the characteristics of the fluid flow in microchannels for
better design of various microflow devices.
In parallel to the recent advances in microfluidic devices, microfabrication techniques have evolved. Some of the important
fabrication techniques include lithography soft and photolithography, lamination, injection molding, hot embossing, laser micromachining, and electrochemical or ultrasonic technologies
2,710. Together with new methods of fabrication, it is possible
to exploit certain fundamental differences between the physical
properties of fluids moving in large channels and those traveling
through micrometer-scale channels 2. In recent years, a large
number of papers have reported pressure drop data for laminar
fully developed flow of liquids in microchannels with various
cross sections 1132. Rectangular cross sections have been extensively studied as they are employed in many applications
1116. However, published results are often inconsistent. Some
authors reported a huge deviation from the conventional theories
and attributed it to an early onset of laminar to turbulent flow
transition or surface phenomena such as surface roughness, electrokinetic forces, viscous heating effects, and microcirculation
near the wall 1226.
Jiang et al. 28 conducted an experimental investigation of
1
JOURNAL OF FLUIDS ENGINEERING. Manuscript received June 11, 2008; final manuscript
received December 5, 2008; published online March 6, 2009. Assoc. Editor: James
A. Liburdy. Paper presented at the ECI International Conference on Heat Transfer
and Fluid Flow in Microscale III, Whistler, BC, Canada, September 2126, 2008.
water flow through different microchannel cross sections including circular, rectangular, trapezoidal, and triangular. The hydraulic
diameter of microchannels varied from 8 m to 42 m. They
collected experimental data with the Reynolds number ranging
from 0.1 to 2, and concluded that there was less influence of the
cross-sectional shape on the microflow in the microchannel and
that the experimental data agreed well with the prediction of the
conventional theory. Baviere et al. 29 performed an experimental study on the water flow through smooth rectangular microchannels. Their channels were made of a silicon engraved substrate anodically bonded to a Pyrex cover. Their results showed
that in smooth microchannels, the friction law is correctly predicted by conventional theories. Judy et al. 15 and Bucci et al.
30 showed that their experimental results were in good agreement with conventional theories in the laminar regime. Also Wu
and Cheng 31, Liu and Garimella 32, and Gao et al. 33
reported good agreement between experimental data with conventional theories.
Low Reynolds numbers characterize many microscale liquid
flows 34. Hence, nonlinear terms in the NavierStokes equation
disappear for fully developed flow, resulting in Poissons equation
2u =
1 dp
,
dz
u = 0 on
where u is the fluid velocity, z is the flow direction, and is the

perimeter of the channel. Exact solution for Eq. 1 in rectangular
cross-section channels can be found in fluid mechanics textbooks
such as Ref. 35. The original analytical solution for the mean
velocity in rectangular channels is in the form of a series, but it
has been shown that using the first term of the series results in
errors less than 0.7%.
Recently, Bahrami et al. 1,36 developed a general model for
prediction of pressure drop in microchannels of arbitrary cross
section. Using the analytical solution of elliptical duct and the
concept of Saint-Venant principal in torsion 37, they showed that
the Poiseuille number, f ReA, is a function of the polar moment
of inertia, I P, area, A, and perimeter of the cross section of the
channel, . Their model showed good agreement with experimental and numerical data for a wide variety of channel cross sections
such as rectangular, trapezoidal, triangular, circular, and moon
shaped.
APRIL 2009, Vol. 131 / 041202-1
However, they 1,36 did not perform any experiment to verify

this model, thus they could not investigate all issues and effects of
different parameters. As a result the goal of the present work is to
independently validate the general model of Bahrami et al. 1,36.
Moreover, effects of different parameters, including microchannel
dimensions, minor and developing region losses, viscous heating,
electroviscous phenomena, and channel deformation on the measured pressure drop, have been discussed. By successfully validating the model, it will be used in more complicated channel geometries such as variable cross-section microchannels.
Theory
Selection of the characteristic length is an arbitrary choice and

will not affect the final solution. However, an appropriate length
scale leads to more consistent results, especially when a general
cross section is considered. A circular duct is fully described with
its diameter, thus the obvious length scale is the diameter or
radius. For noncircular cross sections, the selection is not as
clear; many textbooks and researchers have conventionally chosen
the hydraulic diameter as the characteristic length. Yovanovich
38 introduced the square root of area as a characteristic length
scale for heat conduction and convection problems. Later, Muzychka and Yovanovich 39 proposed the use of A for the fully
developed flow in noncircular ducts. Bahrami et al. 1,37 showed
that the square root of area appears in the solution of fully developed flow in noncircular ducts. They also compared both Dh and
A and observed that using A as the characteristic length scale
results in similar trends in Poiseuille number for microchannels
with a wide variety of cross sections. Therefore, in this study, A
is selected consistently as the length scale throughout the analysis.
According to the model of Bahrami et al. 1, pressure drop of
laminar fully developed flow in arbitrary cross-section microchannels can be obtained from
P = 162QIp
L
A2
where Ip = I p / A is the specific polar momentum of inertia of the

microchannel cross section, = 4W + H is the microchannel
cross-section perimeter, L is the fully developed length, and Q is
the the volumetric flow rate. Substituting for the geometrical parameters in Eq. 2 for microchannels of rectangular cross section
and the definition of the Poiseuille number yields 1
f ReA =
421 + 2
31 +
As can be seen the only parameter effects the Poiseuille number

in rectangular channels is the cross-section aspect ratio, consistent
with previous works 35,40. For rectangular microchannels, two
asymptotes can be recognized, i.e., very narrow rectangular and
square channels
f ReA =
42
f ReA = 13.16;
0
=1
Experimental Setup
3.1 Chemicals and Materials. De-ionized water was used as

the testing liquid. SU-8 photoresist Microchem, Newton, MA
and diacetone-alcohol developer solution Sigma-Aldrich, St.
Louis, MO were used in the making of the positive relief masters
by the procedure outlined below. Polydimethylsiloxane PDMS
casts were prepared by thoroughly mixing the base and curing
agent at a 10:1 ratio as per the manufacturers instructions for the
Sylgard 184 silicon elastomer kit Dow Corning, Midland, MI.
Caution was used to avoid contact between the liquid PDMS and
latex rubber gloves, as this was found to inhibit curing 41.
041202-2 / Vol. 131, APRIL 2009
Fig. 1 Schematic of the soft lithography technique
3.2 Microfabrication. The PDMS/PDMS microchannels

used in this study have been manufactured using the soft lithography technique 42 described by Erickson et al. 41. A schematic of the process is provided in Fig. 1. Photomasks were designed by AUTOCAD software2 and printed with a 3500DPI printer
Island Graphics Ltd., Victoria, BC, Canada. Masters containing
the desired microchannel pattern have been made by spin coating
of SU-8 negative photoresist on a glass slide to the desired thickness. The photoresist film was then hardened through a two-stage
direct contact pre-exposure bake procedure 65 C for 5 min and
95 C for 30 min and exposed to UV light for 100 s through the
mask containing the channel pattern. A two-stage postexposure
bake procedure 65 C for 5 min and 95 C for 30 min was then
used to enhance cross linking in the exposed portion of the film.
The slide was then placed in quiescent developer solution for 10
min to dissolve the unexposed photoresist, leaving a positive relief
containing the microchannel pattern.
Liquid PDMS was then poured over the master and exposed to
vacuum condition 1 h to extract all the bubbles in it and cured at
85 C for 1520 min yielding a negative cast of the microchannel
pattern. An enclosed microchannel was then formed by bonding
the PDMS cast with another piece of PDMS via plasma treatment.
Five microchannels were made with a range of aspect ratios,
0.13 0.76, and tested in this study. Dimensions of the microchannels are shown in Table 1. The channels names in the table
indicate the channel material, cross-sectional shape, and the aspect
ratio i.e., PPR-0.13 corresponds to the PDMS/PDMS chip with
rectangular cross section and the aspect ratio of 0.13. As can be
seen, channels were designed long enough so that fully developed
flow is dominant. Since the microchannel dimensions have a major effect on the friction factor calculation 13,15,33, a destructive careful measurement was performed after the experiments.
Channels were cut carefully at three random cross sections. The
cutting lines were perpendicular to the channel to ensure a 90 deg
viewing angle. Dimensions of the channel were measured by an
image processing method. To do so, a Leica DMI 6000B Leica
Microsystems, Richmond Hill, ON, Canada microscope with a
10, 0.4 N.A. objective was used. Images of the channel cross
2
www.usa.autodesk.com.
Table 1 Dimensions of the fabricated microchannels
Channel
Width, 2W
m
Depth, 2H
m
Length, L
mm
=H/W
LD,max / L
%
PPR-0.13
PPR-0.17
PPR-0.4
PPR-0.6
PPR-0.76
780
581
480
189
134
110
101
192
113
103
50
50
58.8
55.5
50
0.13
0.17
0.4
0.6
0.76
3.47
4.03
5.55
6.75
7.86
section were captured by a high resolution high sensitivity CCD

camera Hamamatsu Orca AG, NJ and imported into a photo
editing software, ADOBE PHOTOSHOP 8.0. Dimensions of the channel were then measured by pixel counting. Size of each pixel was
calibrated using a known dimension provided by the microscope
image acquisition software. Accuracy of this method was found to
be 3.6 m. Due to the microfabrication process, the microchannel cross section has some deviations from the rectangular shape.
Height and width measurements were conducted at different random positions for each cross section and the average value was
determined. This measurement was also performed at several
cross sections. Mean values are reported in Table 1. It was observed that both microchannel depth and width were different
from the expected values in the fabrication process. From our
measurements, channel height, and width deviated less than 4%
and 3% from the designed sizes, respectively. Since smooth surfaces were observed by the image processing technique described
above, the roughness of the channels was estimated to be less than
the accuracy of the image processing method. Hence, the relative
channel roughness channel roughness/channel length scale was
found to be less than 3%. As a result the effect of roughness on the
pressure drop of the fully developed flow can be neglected 27.
3.3 Experimental Procedure. An open loop syringe pump
system, as illustrated in Fig. 2, was designed for this work. A
syringe pump Harvard Apparatus, QC, Canada provided constant flow rate with 0.5% accuracy. A range of Reynolds numbers was covered by changing the volumetric flow rate from
40 l / min to 240 l / min. Water was forced to flow through a
submicron filter Aktreingeselchaf Co., Germany before entering
the channel. To measure the pressure drop, a gauge pressure transducer Omega Inc., Laval, QC, Canada was fixed at the channel
inlet while the channel outlet was opened to the atmosphere. Teflon tubing Scientific Products And Equipment, North York, ON,
Canada was employed to connect the pressure transducer to the
syringe pump and the microchannel. Measured pressure was then
monitored and recorded with a computerized data acquisition sys-
tem LABVIEW 8.5, National Instruments, USA3. The flow was

considered to have reached a steady state condition when the readings of the pressure drop did not change with time. For a given
channel, the measurement of pressure drop was repeated three
times for each flow rate. An arithmetic averaging method 43 was
performed to determine the final results.
Analysis of Experimental Data
Reynolds number was calculated knowing the volumetric flow

rate, Q, and the cross-sectional area, A, from
ReA =
Viscous dissipation effect was neglected since the dimensionless number 4 EcL f ReA / ReA = 0.006 is much smaller than
1. The value of Tref = 1 C was used to calculate the Eckert number, Ec, for water as a working fluid 26. Hence, the properties of
the water was assumed to be constant.
Total measured pressure drop during the experiment, Pmeasured
is
Pmeasured = Pc + Pin + PD + PFD + Pex + 2Pb + Pev
6
where Pc is the pressure loss due to the flow in the connecting
tubes, Pin and Pex are the inlet and exit losses, PD is the
developing region loss, PFD is the pressure drop in the fully
developed region, Pb is the pressure drop due to 90 deg bends,
and Pev is the pressure drop corresponds to the electroviscous
effect. Since fully developed pressure drop is the focus of this
study, right hand side losses except PFD must be subtracted from
the measured pressure drop.
4.1 Connecting Tube Pressure Loss, Pc. The connecting
tube pressure drop includes the losses due to all fittings and the
capillary tube from the transducer to the microchannel inlet. We
measured this loss directly at each flow rate when there was no
microchannel at the end of the tubing. The measurements were
carefully conducted with the conditions identical to the case when
a microchannel was added to the end of the connecting tube to
avoid the effects of hydrostatic pressure.
4.2 Developing Region, PD. Since the viscous boundary
layer inherently grows faster in microchannels than in macroscales, the developing region in most cases is negligible. There are
few references that can be found in literature, in which the effect
of inlet region was considered 40,44,45. Phillips 44 showed
that the length of the hydrodynamic developing region, LD, depends on the aspect ratio of rectangular cross-section microchannels; the higher the aspect ratio, the longer the developing length.
Maximum value of LD can be obtained from Eq. 7
LD =
4 Q
1 + 2
This equation shows that the developing length depends on the

flow rate and the channel aspect ratio. The higher the aspect ratio
and/or flow rate, the higher the developing length. Ratio of the
maximum developing length over the total length of the microchannel is listed in Table 1 for different channels. Obviously this
length occurs in the maximum flow rate. Pressure drop associated
with the entrance region after changing the length scale to A is
46
PD =
where
Fig. 2 Schematic of the test section
f ReAQ
2AA
LD + K
Q2
2A2
K = 0.6796+ 1.2197 + 3.30892 9.59213 + 8.90894
www.ni.com.
APRIL 2009, Vol. 131 / 041202-3
Table 2 Ratio of the developing region pressure drop over the

measured pressure drop of the microchannel
PD / Pmeasured
%
Q
l / min
40
60
80
100
120
240
= 0.13
= 0.17
0.01
0.01
0.01
0.02
0.02
0.03
0.01
0.02
0.02
0.01
0.02
0.05
= 0.4
0.04
0.03
0.04
0.06
0.06
0.10
= 0.6
= 0.76
0.03
0.04
0.06
0.07
0.08
0.20
0.04
0.06
0.08
0.10
0.11
0.24
2.99595.
In Eq. 8, f ReA was calculated based on the measured pressure drop. Table 2 lists the relative pressure loss of the developing
region with respect to the measured pressure drop. As can be seen,
the values are small, less than 0.3%, and can be neglected for the
range of Reynolds numbers studied in this work 1 ReA 35.
For the higher Reynolds numbers Re 100, the developing pressure drop, PD, was found to be less than 2% of the fully developed pressure drop obtained from Eq. 3.
4.3 Minor Losses, Pmin. Other pressure losses associated
with the measured pressure drop are inlet, exit, and bend losses.
These losses are usually obtained from the traditional relationships used in macroscale 15,44,46,47. Phillips 44 showed that
the minor pressure losses can be obtained from
Pmin = Pin + Pex + 2Pb =
Q2
A
Kc + Ke + 2Kb
2A
At
9
where A and At are the channel and connecting tube crosssectional areas, respectively. Kb is the loss coefficient for the
bend, and Kc and Ke represent the contraction and expansion loss
coefficients due to area changes. Phillips 44 recommended Kb to
be approximately 1.2 for a 90 deg bend. Assuming equal crosssectional areas for the channel and connecting tubes and also
maximum possible values for Kc and Ke 48, relative minor
losses with respect to the measured pressure drop are listed in
Table 3. As can be seen, these losses are negligible compared with
the measured pressure drop. For higher Reynolds numbers Re
100, minor pressure drop, Pmin, was found to be less than 5%
of the fully developed pressure drop obtained from Eq. 3.
4.4 Electroviscous Effect, Pev. When a liquid is forced
through a narrow channel under an applied pressure gradient, the
counterions in the diffusive layer of EDL electric double layer
are moving toward the downstream end and a potential gradient is
induced in the flow 49. This so called streaming potential acts to
drive the counterions in the diffuse layer of the EDL to move in

the direction opposite to the pressure-driven flow. The overall result is a reduced flow rate for a constant pressure gradient or
increase in pressure drop for a given flow rate. This phenomenon,
which gives the appearance of an increased viscosity, a, is called
the electroviscous effect 50. The ratio of this apparent viscosity
to the actual fluid viscosity for circular cross-section capillaries
can be found from 50
32 2
a
1+
bD2 eo
where D is the capillary diameter, is the fluid viscosity, b is the

liquid electrical conductivity, and a is the radius of the capillary.
eo = 0r / is the electro-osmotic mobility, where 0 = 8.854
1012 C2 V1 m1 is the permittivity of vacuum, r = 78.3 is the
relative permittivity for water at 25 C 49, and is the zeta
potential of the surface. Approximating the rectangular crosssection microchannel with a capillary of the same cross-sectional
area, b = 5.6 106 S m1 51 and eo = 5.9 108 m2 V1 s1
52 for PDMS, value of a / was found to be less than 1.03 for
the channel sizes in this study.
4.5 Effect of Channel Deformation Under Pressure-Driven
Flow. Effect of channel deformation becomes important since it
can change the microchannel cross-sectional area and as a result
the pressure drop. In all equations used in Secs. 13, we assumed
that the channel cross section did not change under pressuredriven flow. Few works have discussed the importance of bulk
deformation in rectangular microchannels. Holden et al. 53 measured the PDMS bulk deformation under high pressure drops
1 bar using the change in fluorescence intensity. Gervais et al.
54 studied the elastic deformation of PDMS microchannels under imposed flow rates. They demonstrated that the channel deformation can effect the the laminar flow profile and pressure
distribution within the channels with low aspect ratios and high
pressure drops. In a bulged microchannel, measured pressure drop
no longer changes linearly with the flow rate 53. This is because
the microchannel cross section varies due to deformation, and for
a constant flow rate, pressure drop changes. Figure 3 shows the
variation in the channel pressure drop with flow rate for present
work. Experimental results properly follow the linear behavior of
theoretical profiles. Thus one can conclude that the channel deformation has negligible effect on the results in this work. However,
a careful experimental and theoretical investigation is required to
study the effect of the channel deformation on the pressure drop of
moderate and high aspect ratio rectangular cross-section microchannels.
4.6 Uncertainty Analysis. A careful analysis of the experimental uncertainty in this study is critical to the interpretation of
experimental data of Poiseuille number, Po, and exploration of
deviations from the macroscale theory. Neglecting the developing
region, minor losses, electroviscous effects, and considering that
the channel deformation is not significant, the experimental Poiseuille number can be obtained from
Table 3 Ratio of the minor pressure loss over the measured

pressure drop of the microchannel
Pmin / Pmeasured
%
Q
l / min
= 0.13
= 0.17
= 0.4
= 0.6
= 0.76
40
60
80
100
120
240
0.04
0.05
0.07
0.08
0.09
0.16
0.04
0.07
0.10
0.07
0.10
0.24
0.12
0.10
0.14
0.21
0.19
0.35
0.09
0.13
0.17
0.22
0.25
0.51
0.11
0.17
0.23
0.28
0.35
0.70
041202-4 / Vol. 131, APRIL 2009
10
Poexp =
4 A2A
PFD
Q L
11
where PFD = Pmeasured Pc. The uncertainty associated with

Poexp based on the measured variables, and the method provided
in Ref. 43 is then given by
Poexp
Poexp

P
P
L
L
25 A
4 A
Q
Q
2 1/2
12
where values indicate uncertainty associated with each variable
2
2 1/2
+ H
as subscribed, A = 4HW2 + WH21/2 and = 4W

see Appendix for details of calculations. In this equation we
Fig. 3 Channel pressure drop as a function of flow rate. Lines

show the theoretical prediction of pressure drop using Eq. 3
and symbols show the experimental data.
assumed that the fluid properties remained constant. Equation 12

shows that the channel dimensions play an important role in determination of uncertainty by effecting three terms. Also for small
Reynolds numbers, the accuracy of the flow rate becomes more
significant. The uncertainty in the measurement of each parameter
is listed in Table 4. Using Eq. 12 minimum uncertainties for the
samples of PPR-0.13, PPR-0.17, PPR-0.4, PPR-0.6, and PPR-0.76
are 8.7%, 9.3%, 8.9%, 9.3%, and 11.1%, respectively.
Figures 4 and 5 show the comparison between friction factor

obtained from Eq. 3, the solid line, and experimental data for
two typical samples of PPR-0.13 and PPR-0.76, respectively. As
shown, the trend and the values of experimental data are well
predicted by the theoretical model. Consistent with the laminar
flow in channels friction factor decreases with the Reynolds number.
Variation in the Poiseuille number, f ReA, with the Reynolds
number for different channels is shown in Fig. 6. Experimental
data are normalized by the analytical model obtained from Eq. 3.
The solid line shows the analytical model. From Eq. 3, it is clear
that the Poiseuille number does not depend on the Reynolds number and it only varies with the aspect ratio of the channel cross
section. The same trend can be observed in the experimental data.
Large deviations from this trend in the very low Reynolds numbers attributed to the large uncertainties occur due to small flow
rates and pressure drops. The Reynolds number can be a result of
the small cross-sectional area and/or flow rate, with both parameters increasing the uncertainty in the value of the Poiseuille number, see Eq. 12. As can be seen, most of the experimental points
fall within 10% bounds of the analytical model.
Fig. 4 Variation in friction factor with Reynolds number for

sample no. PPR-0.13
Another comparison of the present experimental work with analytical model is illustrated in Fig. 7. Since the Poiseuille number,
f ReA, remains constant for the laminar regime as the Reynolds
number varies, the experimental data for each set were averaged
over the laminar region. The 10% bounds of the model are also
shown in the plot to better demonstrate the agreement between the
data and the model.
As shown in Eq. 3, the Poiseuille number, f ReA, is only a
function of the aspect ratio, which is a geometrical parameter.
This dependency is plotted in Fig. 8. Averaged values over the
studied range of the Reynolds number were used in this plot. It
can be observed that for smaller aspect ratios, the Poiseuille number, f ReA, increases sharply. To better show the trend of the
analytical model, two asymptotes of very narrow and square cross
sections are also included in Fig. 8. As can be seen, when 0,
parallel plate, the Poiseuille number asymptotically approaches
42 / 3.
Table 4 Uncertainty values for measured parameters

Parameter
P
L
W
H
Q
Uncertainty
0.25% of full scale
0.02 mm
3.6 m
3.6 m
0.5%
Fig. 5 Variation in friction factor with Reynolds number for

sample no. PPR-0.76
APRIL 2009, Vol. 131 / 041202-5
Fig. 6 Variation in Poiseuille number, f ReA, with Reynolds

number
It is beneficial to compare the results of present work with the

existing data of rectangular microchannels. Data were collected
from the works of Wu and Cheng 31, Lu and Garimella 32,
and Gao et al. 33. As shown in Fig. 9, the collected data cover a
wide range of the aspect ratio for almost three decades. The solid
line represents the analytical model of Bahrami et al. 1. As can
be seen, the analytical model agrees well with the experimental
data over the wide range of microchannel cross-section aspect
ratio.
Summary and Conclusions
The general model developed by Bahrami et al. 1,36 was

experimentally validated for the case of rectangular cross-section
microchannels. Frictional pressure drop measurements were conducted over a range of Reynolds numbers from 1 to 35 for rectangular cross-section microchannels fabricated with soft lithography method in the aspect ratio range of 0.13 0.76. A careful
measurement of pressure drop and channel dimensions was con-
Fig. 7 Comparison of experimental data of present work with

analytical model of Bahrami et al. 1
041202-6 / Vol. 131, APRIL 2009
Fig. 8 Variation in Poiseuille number, f ReA, with Reynolds

number
ducted, and the pressure drops due to the developing region, minor losses, and electroviscous effect were estimated using available models in literature. Channel deformation due to the pressure
inside the microchannel was found to be negligible since the pressure drop changes linearly with the flow rate. Comparing the results with the general theoretical model shows good agreement
between the model and experimental data.
Uncertainty analysis showed that the measurement of channel
dimensions and flow rate is critical in microscales. Here, microchannel cross-section geometry was determined through processing high quality images of the channel cross section.
According to the analytical model and experimental data of
present work, the Poiseuille number, f ReA, was found to be only
a function of microchannel geometry in the range of Reynolds
numbers studied in this work. For rectangular cross sections, the
channel aspect ratio is the only parameter that affects the Poi-
Fig. 9 Comparison between experimental data of present

study and previous works
seuille number. The analytical model employed in this work will

be applied to other channel geometries such as variable crosssection microchannels in our future work.
W
W
H
H
2 1/2
A4
where A / W = 4H, A / H = 4W, / W = 4, and / H = 4.
Acknowledgment
The authors gratefully acknowledge the financial support of the
Natural Sciences and Engineering Research Council of Canada
NSERC. Also they thank Dr. Viatcheslav Berejnov and Mr. Ali
K. Oskooi for their assistance.
Nomenclature
Greek
A
At
cp
Ec
f
H
Ip
Ip
Kb
Kc
Ke
L
LD
L
Po
Q
ReA
W
microchannel cross-sectional area, m2

tubing cross-sectional area, m2
fluid specific heat, J / kg K
Eckert number, W2 / 2c pTref
Fanning friction factor
channel half height, m
polar momentum of inertia, m4
specific polar momentum of inertia, I p / A2
loss coefficient for bend
flow contraction loss coefficient
flow expansion loss coefficient
channel length, m
developing region length, m
dimensionless length L / A
Poiseuille number, f ReA
volumetric flow rate, m3 / s
Reynolds number, Q / A
channel half width, m
Pb
Pc
PD
Pex
PFD
Pev
Pin
a
eo
bend pressure loss, Pa

connecting tube pressure loss, Pa
developing region pressure loss, Pa
exit pressure loss, Pa
fully developed pressure loss, Pa
pressure loss due to electroviscous effect, Pa
inlet pressure loss, Pa
channel aspect ratio, 2H / 2W
channel cross-section perimeter, m
fluid viscosity, kg/ m s
apparent viscosity, kg/ m s
electro-osmotic mobility, m2 / V s
fluid density, kg/ m3
fluid electrical conductivity, S/m
uncertainty
Appendix: Uncertainty Formulation

To calculate the uncertainty associated with the experimental
measurements the following relation is used 43:
R =

R
i
xi
2 1/2
A1
where R is the uncertainty in results, Rx1 , x2 , . . . , xn, and i is

the uncertainty of the independent variable xi. If the results function, Rx1 , x2 , . . . , xn, takes the form of a product of the independent variables, R = xa11xa22 , . . . , xann, Eq. A1 can be rewritten as
R
=
R

ai
i
xi
2 1/2
A2
The final form of the uncertainty of the Poiseuille number, Eq.

12, is obtained using Eqs. 10 and A2.
To calculate the uncertainty of the channel cross section, A
= 4W H, and perimeter, = 4W + H, Eq. A1 is employed
A =

A
W
W
A
H
H
2 1/2
A3
References
1 Bahrami, M., Yovanovich, M. M., and Culham, J. R., 2006, Pressure Drop of
Laminar, Fully Developed Flow in Microchannels of Arbitrary Cross-Section,
ASME J. Fluids Eng., 128, pp. 10361044.
2 Whitesides, G. M., 2006, The Origins and the Future of Microfluidics, Nature London, 442, pp. 368372.
3 Grushka, E., McCormick, R. M., and Kirkland, J. J., 1989, Effect of Temperature Gradients on the Efficiency of Capillary Zone Electrophoresis Separations, Anal. Chem., 61, pp. 241246.
4 Fletcher, P. D. I., Haswell, S. J., Pombo-Villar, E., Warrington, B. H., Watts, P.,
Wong, S., and Zhang, X., 2002, Micro Reactors: Principles and Applications
in Synthesis, Tetrahedron, 58, pp. 47354757.
5 DeWitt, S., 1999, Microreactors for Chemical Synthesis, Curr. Opin. Chem.
Biol., 33, pp. 350356.
6 Miyake, R., Lammerink, S. J., Elwenspoek, M., and Fluitman, H. J., 1993,
Micro Mixer With Fast Diffusion, MEMS 93, Fort Lauderdale, FL, pp.
248253.
7 Weigl, B. H., Bardell, R. L., and Cabrera, C. R., 2003, Lab-on-a-Chip for
Drug Development, Adv. Drug Delivery Rev., 55, pp. 349377.
8 Becker, H., and Locascio, L. E., 2002, Polymer Microfluidic Devices, Talanta, 56, pp. 267287.
9 Chovan, T., and Guttman, A., 2002, Microfabricated Devices in Biotechnology and Biochemical Processing, Trends Biotechnol., 203, pp. 116122.
10 Ehrfeld, W., 2003, Electrochemistry and Microsystems, Electrochim. Acta,
48, pp. 28572868.
11 Bayraktar, T., and Pidugu, S. B., 2006, Characterization of Liquid Flows in
Microfluidic Systems, Int. J. Heat Mass Transfer, 49, pp. 815824.
12 Peng, X. F., Peterson, G. P., and Wang, B. X., 1994, Frictional Flow Characteristics of Water Flowing Through Rectangular Microchannels, Exp. Heat
Transfer, 7, pp. 24964.
13 Xu, B., Ooi, K. T., and Wong, N. T., 2000, Experimental Investigation of
Flow Friction for Liquid Flow in Microchannels, Int. Commun. Heat Mass
Transfer, 27, pp. 11651176.
14 Ren, C. L., and Li, D., 2004, Electroviscous Effects on Pressure-Driven Flow
of Dilute Electrolyte Solutions in Small Microchannels, J. Colloid Interface
Sci., 274, pp. 319330.
15 Judy, J., Maynes, D., and Webb, B. W., 2002, Characterization of Frictional
Pressure Drop for Liquid Flows Through Microchannels, Int. J. Heat Mass
Transfer, 45, pp. 34773489.
16 Tuckerman, D. B., and Peace, R. F. W., 1981, High-Performance Heat Sinking for VLSI, IEEE Electron Device Lett., 2, pp. 126129.
17 Mala, G. M., and Li, D., 1999, Flow Characteristics of Water in Microtubes,
Int. J. Heat Mass Transfer, 20, pp. 142148.
18 Pfhaler, J., Harley, J., and Bau, H., 1990, Liquid Transport in Micron and
Submicron Channels, Sens. Actuators, A, A21A23, pp. 431434.
19 Pfhaler, J., Harley, J., Bau, H., and Zemel, J. N., 1991, Gas and Liquid Flow
in Small Channels Micromechanical Sensors, Actuators, and Systems, ASME
Dyn. Syst. Control Div., 32, pp. 4960.
20 Urbanek, W., Zemel, J. N., and Bau, H. H., 1993, An Investigation of the
Temperature Dependence of Poiseuille Numbers in Microchannel Flow, J.
Micromech. Microeng., 3, pp. 206208.
21 Qu, W., Mala, G. M., and Li, D., 2000, Pressure-Driven Water Flows in
trapezoidal Silicon Microchannels, Int. J. Heat Mass Transfer, 43, pp. 3925
3936.
22 Papautsky, I., Brazzle, J., Ameel, T., and Frazier, A. B., 1999, Laminar Fluid
Behavior in Microchannels Using Micropolar Fluid Theory, Sens. Actuators,
A, 73, pp. 101108.
23 Ren, C. L., and Li, D., 2005, Improved Understanding of the Effect of Electrical Double Layer on Pressure-Driven Flow in Microchannels, Anal. Chim.
Acta, 531, pp. 1523.
24 Weilin, Q., Mala, H. M., and Li, D., 2000, Pressure-Driven Water Flows in
Trapezoidal Silicon Microchannels, Int. J. Heat Mass Transfer, 43, pp. 353
364.
25 Guo, Z., and Li, Z., 2003, Size Effect on Microscale Single-Phase Flow and
Heat Transfer, Int. J. Heat Mass Transfer, 46, pp. 149159.
26 Morini, G. L., 2005, Viscous Heating in Liquid Flows in Microchannels, Int.
J. Heat Mass Transfer, 48, pp. 36373647.
27 Bahrami, M., Yovanovich, M. M., and Culham, J. R., 2006, Pressure Drop of
Laminar, Fully Developed Flow in Rough Microtubes, ASME J. Fluids Eng.,
128, pp. 632637.
28 Jiang, X. N., Zhou, Z. Y., Huang, X. Y., and Liu, C. Y., 1997, Laminar Flow
Through Microchannels Used for Microscale Cooling Systems, IEEE/CPMT
Electronic Packaging Technology Conference, pp. 119122.
29 Baviere, R., Ayela, F., Le Person, S., and Favre-Marinet, M., 2005, Experimental Characterization of Water Flow Through Smooth Rectangular Microchannels, Phys. Fluids, 17, p. 098105.
30 Bucci, A., Celata, G. P., Cumo, M., Serra, E., and Zummo, G., 2003, Water
Single-Phase Fluid Flow and Heat Transfer in Capillary Tubes, International
Conference on Microchannels and Minichannels, Vol. 1, pp. 319326, ASME
Paper No. 1037.
APRIL 2009, Vol. 131 / 041202-7
31 Wu, H. Y., and Cheng, P., 2003, Friction Factors in Smooth Trapezoidal
Silicon Microchannels With Different Aspect Ratios, Int. J. Heat Mass Transfer, 46, pp. 25192525.
32 Liu, D., and Garimella, S., 2004, Investigation of Liquid Flow in Microchannels, J. Thermophys. Heat Transfer, 18, pp. 6572.
33 Gao, P., Le Person, S., and Favre-Marinet, M., 2002, Scale Effects on Hydrodynamics and Heat Transfer in Two-Dimensional Mini and Microchannels,
Int. J. Therm. Sci., 41, pp. 10171027.
34 Squires, T. M., and Quake, S. R., 2005, Micro Fluidics: Fluid Physics at
Nanoliter Scale, Rev. Mod. Phys., 77, pp. 9771026.
35 White, F. M., 1974, Viscous Fluid Flow, McGraw-Hill, New York, Chap. 3.
36 Bahrami, M., Yovanovich, M. M., and Culham, J. R., 2007, A Novel Solution
for Pressure Drop in Singly Connected Microchannels, Int. J. Heat Mass
Transfer, 50, pp. 24922502.
37 Timoshenko, S. P., and Goodier, J. N., 1970, Theory of Elasticity, McGrawHill, New York, Chap. 10.
38 Yovanovich, M. M., 1974, A General Expression for Predicting Conduction
Shape Factors, AIAA Prog. in Astro. and Aeronautics: Thermophysics and
Space Craft Control, Vol. 35, R. G. Hering, ed., MIT Press, Cambridge, MA,
pp. 265291.
39 Muzychka, Y. S., and Yovanovich, M. M., 2002, Laminar Flow Friction and
Heat Transfer in Non-Circular Ducts and Channels Part 1: Hydrodynamic
Problem, Proceedings of Compact Heat Exchangers, A Festschrift on the 60th
Birthday of Ramesh K. Shah, Grenoble, France, pp. 123130.
40 Shah, R. K., and London, A. L., 1978, Laminar Flow Forced Convection in
Ducts, Supplement to Advances in Heat Transfer, Academic, New York.
41 Erickson, D., Sinton, D., and Li, D., 2003, Joule Heating and Heat Transfer in
Polydimethylsiloxane Microfluidic Systems, Lab Chip, 3, pp. 141149.
42 McDonald, J. C., Duffy, D. C., Anderson, J. R., Chiu, D. T., Wu, H., Schueller,
O. J. A., and Whiteside, G., 2000, Fabrication of Microfluidic Systems in
041202-8 / Vol. 131, APRIL 2009
Polydimethylsiloxane, Electrophoresis, 21, pp. 2740.

43 Holman, J. P., 2001, Experimental Methods for Engineering, 7th ed., McGrawHill, New York, Chap. 3.
44 Phillips, R. J., 1990, Microchannel Heat Sinks, Advances in Thermal Modeling
of Electronic Components and Systems, Hemisphere, New York, Chap. 3.
45 Sharp, K. V., Adrian, R. J., Santiago, J. G., and Molho, J. I., 2005, Liquid
Flows in Microchannels, MEMS Hand Book: Introduction and Fundamentals,
2nd ed., Taylor & Francis, New York.
46 Kandlikar, S. G., Garimella, S., Li, D., Colin, S., and King, M. R., 2006, Heat
Transfer and Fluid Flow in Minichannels and Microchannels, Elsevier, Oxford.
47 Steinke, M. E., and Kandlikar, S. G., 2006, Single-Phase Liquid Friction
Factors in Microchannels, Int. J. Therm. Sci., 45, pp. 10731083.
48 Kays, W. M., and London, A. L., 1984, Compact Heat Exchangers, McGrawHill, New York.
49 Probstein, R. F., 1994, Physicochemical Hydrodynamics, 2nd ed., Wiley, New
York.
50 Masliyah, J. H., and Bhattacharjee, S., 2006, Electrokinetic and Colloid Transport Phenomena, Wiley, Englewood Cliffs, NJ.
51 Lide, D., and Kehiaian, H. V., 1994, CRC Handbook of Thermophysical and
Thermochemical Data, 1st ed., CRC Press, Florida.
52 Lee, J. S. H., Hu, Y., and Li, D., 2005, Electrokinetic Concentration Gradient
Generation Using a Converging-Diverging Microchannel, Anal. Chim. Acta,
543, pp. 99108.
53 Holden, M. A., Kumar, S., Beskok, A., and Cremer, P. S., 2003, Microfluidic
Diffusion Diluter: Bulging of PDMS Microchannels Under Pressure Driven
Flow, J. Micromech. Microeng., 13, pp. 412418.
54 Gervais, T., El-Ali, J., Gunther, A., and Jensen, K., 2006, Flow-Induced Deformation of Shallow Microfluidic Channels, Lab Chip, 6, pp. 500507.
Numerical Modeling of Laminar

Pulsating Flow in Porous Media
S.-M. Kim1,2
e-mail: kim698@purdue.edu
S. M. Ghiaasiaan1
e-mail: mghiaasiaan@gatech.edu
G. W. Woodruff School of Mechanical
Engineering,
Georgia Institute of Technology,
Atlanta, GA 30332-0405
The laminar pulsating flow through porous media was numerically studied. Twodimensional flows in systems composed of a number of unit cells of generic porous
structures were simulated using a computational fluid mechanics tool, with sinusoidal
variations in flow with time as the boundary condition. The porous media were periodic
arrays of square cylinders. Detailed numerical data for the porosity ranging from 0.64 to
0.84, with flow pulsation frequencies of 2064 Hz were obtained. Based on these numerical data, the instantaneous as well as the cycle-average permeability and Forchheimer
coefficients, to be used in the standard unsteady volume-averaged momentum conservation equation for flow in porous media, were derived. It was found that the cycle-average
permeability coefficients were nearly the same as those for steady flow, but the cycleaverage Forchheimer coefficients were significantly larger than those for steady flow and
were sensitive to the flow oscillation frequency. Significant phase lags were observed
between the volume-averaged velocity and the pressure waves. The phase difference
between pressure and velocity waves, which is important for pulse tube cryocooling,
depended strongly on porosity and the mean-flow Reynolds number.
DOI: 10.1115/1.3089541
Keywords: porous media, pulsating flow, permeability, Forchheimer coefficient, phase
shift
Introduction
Pulsating and periodic flows in porous media are poorly understood. An important area of application of such flow is in the
Stirling and pulse tube cryocooler systems. In these systems, microporous structures constitute cold and warm heat exchangers,
and more importantly a regenerator, all of which are subject to a
periodic flow of a gas usually helium.
The fluid-solid interactions in these microporous structures, in
particular, the regenerator, are important with respect to the performance of the cryocooler system. The frictional losses, in particular, adversely affect the overall system performance. A number
of experimental investigations dealing with friction in commonlyused cryocooler regenerator material have been reported in the
recent past 16. These investigations have all dealt with the
macroscopic aspects of friction in porous media subject to periodic flow and have shown that, except at very low frequencies, the
closure parameters that represent friction between the fluid and
solid are different in steady and periodic flows. However, no systematic study of pore-level phenomena during periodic or pulsating flows through porous media has been reported in the past. It is
thus the objective of this investigation to study the pore-level
phenomena in a pulsating flow through a generic two-dimensional
porous medium by numerical analysis, and from there examine
the effects of flow pulsations on the closure parameters that are
encountered in the standard volume-averaged momentum equation for porous media. We have chosen the pulsating flow as an
intermediate step toward the more difficult problem of periodic
flow.
Since pore-level analysis in general is not feasible for the usual
design and analysis calculations for flow through porous media,
model equations are often applied instead. The most widely used
1
Present address at Boiling and Two-Phase Flow Laboratory, School of Mechanical Engineering, Purdue University, West Lafayette, IN 47907-1288
JOURNAL OF FLUIDS ENGINEERING. Manuscript received August 8, 2008; final manuscript received January 5, 2009; published online March 9, 2009. Assoc. Editor:
Neelesh A. Patankar.
2
method for the derivation of tractable model equations for porous

media is volume averaging, where surface integrals and integrals
involving dispersion and tortuosity are replaced with tractable
model terms 711. The model equations derived in this way will
include terms with unknown coefficients that must be found by
experiments or numerical simulations aimed at pore-level analysis. For the pore-level analysis, the exact pore structure is needed,
however, which is often complex, nonuniform, and even unknown. Nevertheless, valuable information about the flow in porous media can be obtained by direct pore-level simulation of
idealized generic porous media composed of periodicallyconfigured circular or square cylinders. Published investigations
in this area include Refs. 1218.
These investigations have all addressed steady flow, however, it
appears that transient flows in porous media and, in particular,
pulsating or periodic flow have thus far not been systematically
studied by numerical simulation of pore-level flow and heat/mass
transport phenomena.
Geometric Configuration
For the numerical simulation of flow through porous media,

two-dimensional periodic arrays of cylindrical or rectangular cylinders are usually assumed 1218. In this study, the porous medium is simulated by an array of square cylinders. Figure 1 depicts
the unit cells and dimensions of all the configurations considered.
The flow everywhere is from left to right along the x coordinate.
The geometric dimensions have been chosen such that porosity
ranging from 0.64 to 0.84 is covered, where evidently see Fig. 2
= 1 D/L2
Figure 2 displays the generic computational domain, which is

composed of six unit cells in series. Past studies have shown that
the end effects typically extend for a few unit cells in porous
media, and an array of several unit cells is an adequate computational domain 9,12. The forthcoming simulation results confirm
that by using six unit cells in series and investigating in detail the
flow field in the fifth unit cell, the end effects are effectively
eliminated in the simulations.
APRIL 2009, Vol. 131 / 041203-1
Fig. 1 Different porous structure geometries showing a unit cell of continuous porous
structures
Figure 3 shows a sample grid system for the porous structure

with = 0.84, where each unit cell is represented with a 20 40
grid structure. Since the location of the inlet and outlet can affect
the convergence and the accuracy of the simulations, following
several numerical tests, it was found out that the inlet and outlet
should be separated from the computational domain by buffer
zone lengths of 2L and 7L, respectively. With these separation
lengths the simulation results became effectively independent of
the inlet and outlet locations.
u
1
+ u u = P + 2u
t
where properties all belong to the fluid.

The macroscopic-level conservation equations will be the
commonly-used model equations for flow in porous media
1921. The superficial volume-average velocity is defined as
u =
Governing Equations
Two separate sets of conservation equations are solved for the

system shown in Fig. 3. The NavierStokes equations are solved
for detailed pore-level simulations. These will be referred to as the
microscopic-level equations. The results obtained with the
microscopic-level equations will be compared with the solution of
volume-average porous media-type equations. The latter equations
will be referred to as the macroscopic-level equations.
Assuming incompressible and constant-property fluid, the
microscopic-level equations will be the following:
u=0
1
V
udV
Vf
where V and V f denote the averaging control volume and the fluid
volume contained within the averaging control volume, respectively see Fig. 2. The intrinsic volume-average fluid velocity is
defined as
u f =
1
Vf
udV
Vf
Clearly, = V f / V = u / u f . The macroscopic-level momentum

conservation equation will be as follows 17,20,21:
Fig. 2 Computational domain with boundary conditions
041203-2 / Vol. 131, APRIL 2009
only on the geometry of the porous medium, and the Forchheimer

effect can be negligible when the flow rate becomes vanishingly
small.
For the porous structures shown in Fig. 3, the x coordinate is a
principle direction. As a result, when a pressure gradient in the x
direction is imposed, the flow on a macroscopic scale will also be
in the x direction only. With respect to a coordinate system that
and B are
coincides with the principle directions, the tensors K
both diagonal. Thus, Eq. 6 can be cast for flow in x direction as
ux f
2ux f
ux f
P f
ux f
=
+ ux f
+
x
x2
t
x
Kxx
2Bxxux f ux f
where Kxx and Bxx are the x components of the diagonal tensors K
and B, respectively. In case of steady macroscopic flow in the x

direction, Eq. 7 leads to the Forchheimer-extended Darcy equation.
0=
P f
ux + Bxx,stuxux
x
Kxx,st
This equation can be written in nondimensional form as
L2
+ Bxx,stL
Kxx,st ReL
where
uL
10
P f L
x ux2
11
ReL =
and
Fig. 3 Sample grid system for = 0.84
u f
u f
+ u f u f = P f + 2u f
t
K
2B u f u f
u f will be referred to as the Darcy term, which

where / K
accounts for the viscous effect; 2u f is the Brinkman term,
which is important only in the interface between the porous medium and other solid boundaries, such as a wall or an adjacent
fluid region, where the effective viscosity e of the porous medium is assumed as e ; and 2B u f u f is referred to as
the Forchheimer term, which accounts for the inertial effects, and
becomes dominant when u f is large. Under macroscopically
fully developed flow, the Darcy and Forchheimer terms depend
Numerical Solution Method
The microscopic-level equations are solved using the FLUENT 6.3

computational fluid dynamics CFD code 22. The density-based
NavierStokes coupled solution algorithm of FLUENT, which includes low-Mach number preconditioning, artificial compressibility, and dual time-stepping for the solution of unsteady incompressible flow, solves the governing equations of continuity and
momentum simultaneously 22,23. The first-order upwind discretization scheme is used to treat the convection terms. For the
diffusion terms, the central difference scheme is used.
The two-dimensional mesh as depicted in Fig. 3 was created
by using the GAMBIT 2.2 software 24. Detailed numerical data
representing the oscillating velocity and pressure variations are
obtained from the calculation of the microscopic-level governing
equations. Then the volume-averaged velocities and pressures of
each unit cell are used to numerically calculate the instantaneous
as well as the cycle-average permeability and Forchheimer coefficients, as described below. For this purpose, a program was developed in C. This program solves the unsteady volumeaveraged momentum equation implicitly by using the finitedifference method.
To obtain the volume-averaged velocities and pressures at each
unit cell of the generic porous structures, a user defined function
UDF was coded in the C programming language. The numerical simulations are conducted with air as the working fluid,
assuming = 1.225 kg/ m3 and = 1.7894 105 kg/ m s. The
sinusoidal pulsating area-averaged velocity at inlet boundary, to
be discussed later, was also implemented by a UDF of FLUENT 6.3
22.
Equation 7 is numerically solved for the macroscopic-level
analysis, using the backward in time and central in space BTCS
finite-difference scheme, where a first-order backwardAPRIL 2009, Vol. 131 / 041203-3
differencing for the time derivative and a second-order central

differencing for the spatial derivative are used. Thus, for unit cell
i,
f,n
f,n
f,n
f,n
ui+1
ui1
uif,n uif,n1
Pi1
Pi+1
=
+ uif,n
2x
t
2x
f,n
f,n
ui+1
2uif,n + ui1
u f,n 2Bxx,instt
2
x
Kxx,instt i
uif,nuif,n + Ot,x2
12
where subscript n and superscript i are time step and grid point
indices, respectively. Note that u in the above equation implies ux,
and the subscript x has been left out for convenience.
The calculation procedure for extracting the permeability and
Forchheimer coefficients from the microscopic-level solutions for
pulsating flow is as follows. We note that at the limit of u f 0,
the Forchheimer term in Eq. 7 or Eq. 12 can be neglected.
Thus, for each of the configurations in Fig. 1, the following steps
are followed:
1 Obtain uif,n and Pif,n for i = 1 6 and during n equallyspaced time snapshots.
2 Solve the following equation for the time snapshots covering two pulsation cycles for the low-flow simulations, and
thereby obtain Kxx,instt.
f,n
f,n
ui+1
ui1
uif,n uif,n1
=
+ uif,n
2x
t
f,n
f,n
f,n
f,n
ui+1
2uif,n + ui1
Pi1
Pi+1
+
2
x
2x
u f,n
Kxx,instt i
13
3 Obtain the cycle-average permeability coefficient Kxx,avg.

4 Solve Eq. 12 for the time snapshots covering two pulsation cycles for the high-flow simulations by treating
Kxx,instt as known, and thereby obtain Bxx,instt.
5 Obtain the cycle-average Forchheimer coefficient Bxx,avg.
The cycle-average parameters are calculated from the following:
Kxx,avg = f
Fig. 4 Steady-state streamline patterns for a ReL = 0.11 and

b ReL = 560 in the case of = 0.75
to+1/f
Kxx,insttdt
14
Bxx,insttdt
15
selected for the calculation of the instantaneous permeability and

Forchheimer coefficients, respectively. The velocity at the inlet of
the computational domain in pulsating flow simulations is
Uin = Um1 + a sin2 ft,
a = 0.4
16
Steady Flow
For CFD simulations, a convergence criterion of 106 was applied to the residuals of the continuity and momentum equations.
Grid independence was examined by using two grid systems with
20 40 and 40 80 per unit cell. The two simulations showed
nearly identical velocity and pressure distributions for the pairs of
simulations with the maximum difference between the volumeaveraged pressure gradients within 1%. For calculation efficiency,
a grid system with 20 40 per unit cell was chosen for all the
subsequent simulations.
Figure 4 represents the typical steady-state streamline patterns.
The streamline patterns are similar for all the unit cells. The flow
details in the unit cells 4, 5, and 6 are virtually identical, indicating a fully developed state. It is seen that for the low Reynolds
number flow, a pair of small vortices appears between the adjacent
square rods. For the high Reynolds number flow, the position of
the vortex center in the cavity is shifted slightly downstream.
The pressure drop characteristics along the flow direction are
presented in terms of the nondimensional pressure gradient in
Fig. 5. The nondimensional pressure gradient undergoes a large
change in the entrance region and reaches an almost constant
value along the flow direction. At very low Reynolds numbers
ReL 1, it is observed that the hydrodynamically fully developed regime is reached just after the second unit cell. However, in
case of high Reynolds numbers, the pressure gradient still varies
slightly from the third to fourth unit cells, and the change in
pressure gradient from one unit cell to the next is decreased as the
fully developed condition is reached further downstream. Similar
to
Bxx,avg = f
to+1/f
to
A similar but simpler procedure is used for the calculation of Kxx

and Bxx under steady-flow conditions see Ref. 25 for details.
The boundary conditions for the simulations are now discussed.
The inlet boundary is located at a distance 2L upstream from the
inlet to the first unit cell, where the velocity has a uniform profile.
The no-slip boundary condition is applied on the wall. Referring
to Fig. 2, in view of symmetry with respect to the x-axis, only half
of the physical domain that represents the six unit cells is considered, and at the top and bottom faces of a unit cell, the normal
gradients of all variables are prescribed as zero. The outflow
boundary is located at a distance 7L downstream from the exit of
the last unit cell, where a zero-gradient is imposed for all dependent variables.
The ReL ranging from 0.1 to 1000 is used for the steady-flow
simulations. The converged steady-state results are used as the
initial values for the relevant unsteady calculations. For the pulsating flow simulations, the mean Reynolds numbers based on the
mean superficial Darcy velocity Um of Rem,L = 0.11 and 560 are
041203-4 / Vol. 131, APRIL 2009
Fig. 5 Nondimensional pressure gradient along the flow direction with Reynolds number for the steady flow
Table 1 The permeability and Forchheimer coefficients for

steady and pulsating flows
L2
Kxx,st
L2
Kxx,avg
Bxx,stL
Bxx,avgL
0.64
0.6975
0.75
0.75a
0.75b
0.7975
0.84
153
105
75
75
75
55
41
159
113
79
80
80
59
47
0.071
0.058
0.049
0.049
0.049
0.041
0.036
0.121
0.108
0.096
0.078
0.088
0.085
0.077
20 Hz.
64 Hz,
Fig. 6 Nondimensional pressure gradient as a function of Reynolds number for steady flow
trends have also been observed experimentally and numerically by

many researchers, including Coulaud et al. 12, Benarji et al.
26, Raju and Narasimhan 27, and Sahiti et al. 28. It was
found that the change in nondimensional pressure gradient between the fourth and fifth unit cells is 0.04% and 1.7% for ReL
= 35 and 980, respectively. Therefore, the flow after the fourth unit
cell is considered to be fully developed, and the following results
and observations are based on the fifth unit cell.
Figure 6 shows the variation in dimensionless pressure gradient
with ReL, while the dependence of the Forchheimer term on ReL
is displayed in Fig. 7. Evidently, for ReL 200, the Forchheimer
term approaches a constant that depends on and, as expected, it
vanishes as ReL 0.
The permeability and Forchheimer terms for steady flow are
shown in Table 1. The table also contains parameter values for
pulsating flow, which will be discussed later. Nakayama et al. 16
performed numerical simulations for steady flow in a generic porous medium similar to Fig. 1c. Our calculations show
L2 / Kxx,st = 75 for = 0.75, which agrees well with L2 / Kxx,st = 76
reported by Nakayama et al. 16. However, we obtained Bxx,stL
= 0.049, which is one-quarter of Bxx,stL = 0.2 reported in Ref. 16.
Pulsating Flow
The pulsating flow CFD simulations are started from an initial

condition represented by the aforementioned steady-flow simulations and are continued until steady-pulsating flow conditions are
achieved. Steady-pulsating flow is assumed when all calculated
Fig. 7 The Forchheimer term as a function of Reynolds number for steady flow
quantities at t and t + T are identical for points 16 in Fig. 3.

Similar to steady-flow simulations, two grid systems with 20
40 and 40 80 nodes per unit cell were generated, and grid
independence was investigated by comparing the volumeaveraged pressure gradients. The maximum difference in volumeaveraged pressure gradient between the two grid systems was
2.8% at the first unit cell. Additionally, the effect of the time step
size was studied by using three time step sizes t = 2.5 104 s
T / 100, 1.25 104 s T / 200, and 6.25 105 s T / 400, in
simulations where the period of oscillation T was 0.025 s. Figure
8 compares the instantaneous area-averaged velocities at inlet and
outlet for = 0.75 with a grid size of 20 40 per unit cell, for the
aforementioned three time step sizes. Evidently, there is essentially no difference among the model predictions for the three time
steps. Therefore, a time step size of 1.25 104T / 200s and a
grid size of 20 40 per unit cell are used in the forthcoming
calculations.
Figure 9 shows the instantaneous velocities and pressures along
the flow direction for a typical run. Although the calculations are
continued for 0.375 s, the cyclic steady state was obtained after
only 7 cycles, or 0.175 s. The convergence criterion was the relative error in the oscillating velocity component, which was required to be smaller than 103.
Figures 10 and 11 display typical instantaneous velocities and
pressures for steady-pulsating flow condition for Rem,L = 0.11 and
Rem,L = 560, respectively. Although the pulsating inlet velocity is
sinusoidal, the calculated velocity and pressure waves are not exactly sinusoidal. Furthermore, the amplitudes of velocity and pressure waves are attenuated along the flow direction, and a phase
shift develops between instantaneous velocity and pressure waves.
For the low pulsating Reynolds number flow, the amplitudes of
Fig. 8 Time independence test for the base case of = 0.75

with the grid size of 20 40 per unit cell
APRIL 2009, Vol. 131 / 041203-5
Fig. 9 Convergence check for = 0.75 with 20 40 nodes per unit cell for
a velocity and b pressure
instantaneous pressure waves along the flow direction decrease

notably; whereas for high Rem,L cases, the reduction in instantaneous pressure amplitudes is insignificant along the flow direc-
tion, and the waves are flattened with decreasing porosity. The
results also show that the cycle-average instantaneous velocities at
the measured points increase as the porosity decreases, and as the
Fig. 10 Variation in the instantaneous a velocity and b

pressure waves along the flow direction for Rem,L = 0.11 and
= 0.7975
Fig. 11 Variation in the instantaneous a velocity and b

pressure waves along the flow direction for Rem,L = 560 and
= 0.84
041203-6 / Vol. 131, APRIL 2009
Fig. 12 Phase shifts V, P, and VP in terms of porosities

for pulsating low Reynolds number flow Rem,L = 0.11
porosity decreases the differences in cycle-average instantaneous

pressures between adjacent two unit cells increase.
Figures 12 and 13 display the velocity phase shift V, pressure
phase shift P, and the phase difference between velocity and
pressure VP at low and high Reynolds numbers, respectively.
The aforementioned phase shifts are defined as follows:
V = V,pt6 V,pt1/5
17
P = P,pt6 P,pt1/5
18
VP = VP,pt1 + VP,pt6/2
19
where subscripts 1 and 6 refer to points 1 and 6 in Fig. 3; V and

P represent the phase angles associated with the velocity and
pressure waves; and VP = V P. The phase shifts V and P
thus represent the average phase lags, per unit cell, which occur to
the velocity and pressure, respectively, and VP is the average
phase shift between velocity and pressure. As noted, at low Rem,L,
V, P, and VP are all relatively insensitive to both frequency and . However, for high Rem,L Fig. 13, the aforementioned phase shift parameters all depend on and frequency. The
dependence of V and VP on is not monotonic, and for the
40 Hz frequency their maxima occur at = 0.75 and 0.7975, respectively. The parameters V and P for the low Rem,L are
significantly larger than those for the high Rem,L cases.
Fig. 13 Phase shifts V, P, and VP in terms of porosities

for pulsating high Reynolds number flow Rem,L = 560
Fig. 14 Variation in the instantaneous permeability coefficients for different porosities f = 40 Hz calculated from Rem,L
= 0.11
The permeability and Forchheimer terms are now discussed.

Figures 14 and 15 display the instantaneous permeability and
Forchheimer terms. These terms were calculated following the
procedure described in Sec. 4. As noted, for the instantaneous
permeability coefficient, the amplitude and the cycle-average
value of permeability coefficient increase with decreasing porosity. An interesting observation in Fig. 15 is that during a portion of
each cycle, the values of instantaneous Forchheimer coefficient
were negative, since the left hand side of the unsteady volumeaveraged momentum equation Eq. 7, which presents the local
and convective acceleration terms, is less than the pressure drop
term in the same equation. With increasing porosity, the portion of
the cycle in which the value of instantaneous Forchheimer coefficient is negative is extended. The negative values for the Forchheimer coefficient are, of course, a result of the phase difference
between the pressure and velocity waves. It should also be emphasized that the form of the volume-averaged porous-media momentum equation is theoretically plausible only for quasisteady
i.e., relatively slow transient processes 9. For fast transients,
theory would indicate that additional higher-order terms may be
needed in the equation. Inclusion of such additional higher-order
terms is not feasible at this time, however, since such terms would
require additional unknown closure relations and parameters. The
standard DarcyForchheimer volume-averaged equations are
therefore generally used in the analysis of transient processes.
Fig. 15 Variation in the instantaneous Forchheimer coefficients for different porosities f = 40 Hz calculated from Rem,L
= 0.11 and 560
APRIL 2009, Vol. 131 / 041203-7
Fig. 16 Comparison of the permeability coefficients between

the steady and pulsating flow for different porosities
The cycle-average permeability and Forchheimer terms are

shown in Figs. 16 and 17, respectively, and their numerical values
are provided in Table 1. In both figures, model predictions are
shown for f = 40 Hz for all porosities of interest. For = 0.75,
however, model predictions for f = 20 Hz and 60 Hz are also depicted. Furthermore, in both figures the parameters representing
steady-flow conditions are also shown for comparison. The permeability parameter, as noted in Fig. 16, is a strong function of
porosity and monotonically decreases as porosity is increased.
More importantly, furthermore, the cycle-average permeability parameter for pulsating flow is only slightly larger than the permeability parameter for steady flow or, equivalently, the permeability for pulsating flow is only slightly larger than the permeability
for steady flow. On the other hand, as noted in Fig. 17, the cycleaverage Forchheimer coefficients decrease with increasing porosity and are about two times larger than those for steady flow at the
same porosity. It is also noted in Fig. 17 that the Forchheimer
parameter depends on the frequency, and its dependence on frequency is not monotonic. Our simulation results, although relatively limited in scope and parametric range, thus suggest that the
application of the DarcyForchheimer model equations with
steady-flow parameters to pulsating flow is, in general, incorrect.
While the permeability is likely to be approximately the same for
steady and pulsating flow conditions, the Forchheimer coefficient
can have significantly different values for steady-flow and pulsating flow conditions. More comprehensive investigations are recommended for better understanding the trends and dependencies
in pulsating flow.
Fig. 17 Comparison of the Forchheimer coefficients between

the steady and pulsating flow for different porosities
041203-8 / Vol. 131, APRIL 2009
Conclusions
We have investigated, by numerical simulation, the momentum

transfer parameters associated with pulsating laminar flow
through generic porous media made of periodic square cylinders
patterned on a rectangular pitch, for flow pulsation frequencies of
2064 Hz and the porosity ranging from 0.640.84. Detailed numerical data representing the oscillating two-dimensional velocity
and pressure variations were obtained by solving the Navier
Stokes using the CFD code FLUENT 6.3. These numerical data
were then utilized for the calculation of various volume-averaged
properties to be used in the standard Darcy-extended Forchheimer
momentum equation. Instantaneous and cycle-average parameters
were both extracted and displayed. The results indicated that the
application of the Darcy-extended Forchheimer momentum equation with coefficients representing steady flow to pulsating flow
conditions is only adequate at very low-flow low frequency conditions and is otherwise incorrect. Within the range of parameters
studied in this investigation, the pulsating flow cycle-average permeabilities were a strong function of porosity, but they were only
slightly larger than the permeabilities for steady flows. The cycleaverage Forchheimer coefficients for pulsating flows strongly depended on frequency, however, and were significantly larger than
steady-flow Forchheimer coefficients.
The results also showed that the velocity and pressure phase
shifts are sensitive to the mean-flow Reynolds number. The velocity phase shifts V and pressure phase shifts P for low meanflow Reynolds numbers Rem,L = 0.11 were significantly larger
than those for the high-flow Reynolds number Rem,L = 560 flows.
However, the phase shifts between velocity and pressure VP for
Rem,L = 560 were several times larger than those for the Rem,L
= 0.11 case.
Nomenclature
a normalized amplitude of velocity
A constant
B Forchheimer coefficient 1/m
Bxx,avg cycle-average Forchheimer coefficient m2
Bxx,inst instantaneous Forchheimer coefficient along
the x direction for pulsating flow m2
Bxx,st Forchheimer coefficient along the x direction
for steady flow m2
B Forchheimer coefficient tensor 1/m
D side of square rod m
f frequency Hz
I identity tensor
K permeability coefficient m2
Kxx,avg cycle-average permeability coefficient m2
Kxx,inst instantaneous permeability coefficient along the
x direction for pulsating flow m2
Kxx,st permeability coefficient along the x direction
for steady flow m2
K permeability tensor m2
L length of unit cell m
P static pressure N / m2
P f intrinsic volume-average fluid static pressure
over V f N / m2
ReL Reynolds number based on a unit cell length L
in steady flow
Rem,l Reynolds number based on mean velocity and
unit cell length pulsating flow
t time s
T period of pulsation s
u, U velocity m/s
u superficial volume-average fluid velocity m/s
u f intrinsic volume-average fluid velocity m/s
V averaging control volume m3
V f fluid volume in an averaging control volume

m3
t time step size s
Greek Letters

V
P
VP

porosity of medium
molecular viscosity kg/ m s
kinematic viscosity m2 / s
phase angle deg
density kg/ m3
phase shift of velocity deg
phase shift of pressure deg
phase shift between velocity and pressure deg
nondimensional pressure gradient
Subscripts
avg
e
exit
i
in
inst
m
st
x, xx
cycle-average value
effective
exit
grid point index
inlet
instantaneous value
mean value
steady state
x direction
Superscripts
f fluid
n time step index
= tensor
References
1 Ju, Y. L., Wang, C., and Zhou, Y., 1998, Numerical Simulation and Experimental Verification of the Oscillating Flow in Pulse Tube Refrigerator, Cryogenics, 38, pp. 169176.
2 Hsu, C. T., Fu, H. L., and Cheng, P., 1999, On Pressure Velocity Correlation
of Steady and Oscillating Flows in Regenerators Made of Wire-Screens,
ASME J. Fluids Eng., 121, pp. 5256.
3 Nam, K., and Jeong, S., 2002, Experimental Study on the Regenerator Under
Actual Operating Conditions, Adv. Cryog. Eng., 47, pp. 977984.
4 Jeong, S., Nam, K., and Jung, J., 2002, Regenerator Characterization Under
Oscillating Flow and Pulsating Pressure, Cryocoolers, 12, pp. 531537.
5 Cha, J. S., Ghiaaisiaan, S. M., Desai, P. V., Harvey, J. P., and Kirkconnell, C.
S., 2006, Multi-Dimensional Flow Effects in Pulse Tube Refrigerators,
Cryogenics, 46, pp. 658665.
6 Clearman, W. M., Cha, J. S., Ghiaasiaan, S. M., and Kirkconnell, C. S., 2008,
Anisotropic Steady-Flow Hydrodynamic Parameters of Microporous Media
Applied to Pulse Tube and Stirling Cryocooler Regenerator, Cryogenics, 48,
pp. 112121.
7 Whitaker, S., 1973, The Transport Equations for Multi-Phase Systems,
Chem. Eng. Sci., 28, pp. 139147.

8 Gray, W. G., and ONeill, K., 1976, On the General Equation for Flow in
Porous Media and Their Reduction to Darcys Law, Water Resour. Res., 12,
pp. 148154.
9 Kaviany, M., 1995, Principles of Heat Transfer in Porous Media, 2nd ed.,
Springer-Verlag, Berlin.
10 Quintard, M., Kaviany, M., and Whitaker, S., 1997, Two-Medium Treatment
of Heat Transfer in Porous Media: Numerical Results for Effective Properties,
Adv. Water Resour., 20, pp. 7794.
11 Moyne, C., Didierjean, S., Amaral Souto, H. P., and da Silveira, O. T., 2000,
Thermal Dispersion in Porous Media: One Equation Model, Int. J. Heat
Mass Transfer, 43, pp. 38533867.
12 Coulaud, O., Morel, P., and Caltagiorne, J. P., 1988, Numerical Modeling of
Nonlinear Effects in Laminar Flow Through a Porous Medium, J. Fluid
Mech., 190, pp. 393407.
13 Kuwahara, F., Nakayama, A., and Koyama, H., 1996, A Numerical Study of
Thermal Dispersion in Porous Media, ASME J. Heat Transfer, 118, pp. 756
761.
14 Kuwahara, F., Shirota, M., and Nakayama, A., 2001, A Numerical Study of
Interfacial Convective Heat Transfer Coefficient in Two-Energy Equation
Model for Convection in Porous Media, Int. J. Heat Mass Transfer, 44, pp.
11531159.
15 Kuwahara, F., Yamane, T., and Nakayama, A., 2006, Large Eddy Simulation
of Turbulent Flow in Porous Media, Int. Commun. Heat Mass Transfer, 33,
pp. 411418.
16 Nakayama, A., Kuwahara, F., Umemoto, T., and Hayashi, T., 2002, Heat and
Fluid Flow Within an Anisotropic Porous Medium, ASME J. Heat Transfer,
1244, pp. 746753.
17 Nakayama, A., Kuwahara, F., and Kodama, Y., 2006, An Equation for Thermal Dispersion Flux Transport and Its Mathematical Modeling for Heat Transfer and Fluid Flow in a Porous Medium, J. Fluid Mech., 563, pp. 8196.
18 Nakayama, A., Kuwahara, F., and Sano, Y., 2007, Concept of Equivalent
Diameter for Heat and Fluid Flow in Porous Media, AIChE J., 53, pp. 732
736.
19 Vafai, K., and Tien, C. L., 1981, Boundary and Inertia Effects on Flow and
Heat Transfer in Porous Media, Int. J. Heat Mass Transfer, 24, pp. 195203.
20 Vafai, K., and Amiri, A., 1998, Non-Darcian Effects in Confined Forced
Convective Flows, Transport Phenomena in Porous Media, D. B. Ingham and
I. Pope, eds., Pergamon, New York.
21 Hsu, C. T., and Cheng, P., 1990, Thermal Dispersion in a Porous Medium,
Int. J. Heat Mass Transfer, 33, pp. 15871597.
22 FLUENT Inc., 2006, Fluent 6 User Manual.
23 Weiss, J. M., and Smith, W. A., 1995, Preconditioning Applied to Variable
and Constant Density Flows, AIAA J., 33, pp. 20502057.
24 FLUENT Inc., 2006, GAMBIT 2 User Manual.
25 Kim, S.-M., 2008, Numerical Investigation on Laminar Pulsating Flow
Through Porous Media, MS thesis, Georgia Institute of Technology, Atlanta,
GA.
26 Benarji, N., Balaji, C., and Venkateshan, S. P., 2008, Unsteady Fluid Flow
and Heat Transfer Over a Bank of Flat Tubes, Heat Mass Transfer, 44, pp.
445461.
27 Raju, K. S., and Narasimhan, A., 2007, Porous Medium Interconnector Effects on the Thermohydraulics of Near-Compact Heat Exchangers Treated as
Porous Media, ASME J. Heat Transfer, 129, pp. 273281.
28 Sahiti, N., Lemouedda, A., Stojkovic, D., Durst, F., and Franz, E., 2006, Performance Comparison of Pin Fin In-Duct Flow Arrays With Various Pin CrossSections, Appl. Therm. Eng., 26, pp. 11761192.
APRIL 2009, Vol. 131 / 041203-9
Aerodynamic Analysis of a
Vehicle Tanker
Ramon Miralbes Buil
Luis Castejon Herrer
Research Group in Vehicles and
Road Safety (VEHIVIAL),
CPS,
University of Zaragoza,
Zaragoza 50012, Spain
The aim of this article is the presentation of a series of aerodynamic improvements for
semitrailer tankers, which reduce the aerodynamic resistance of these vehicles, and,
consequently, result in a positive impact on fuel consumption, which is substantially
reduced (up to 11%). To make the analysis the computational fluid dynamics (CFD)
methodology, using FLUENT, has been used since it allows simulating some geometries
and modifications of the geometry without making physical prototypes that considerably
increase the time and the economical resources needed. Three improvements are studied:
the aerodynamic front, the undercarriage skirt, and the final box adaptor. First they are
studied in isolation, so that the independent contribution of each improvement can be
appreciated, while helping in the selection of the most convenient one. With the aerodynamic front the drag coefficient has a reduction of 6.13%, with the underskirt 9.6%, and
with the boat tail 7.72%. Finally, all the improvements are jointly examined, resulting in
a decrease of up to 23% in aerodynamic drag coefficient. DOI: 10.1115/1.3077135
Keywords: aerodynamic, undercarriage skirts, tank, vehicle tanker, nozzle, boat-tails,
drag, penetration, consumption, vehicle, articulated
Introduction
From its inception, aerodynamics has been a field linked to

aerospatial engineering. Nonetheless, the first study in aerodynamics appeared in 1934 and was applied to an automobile, the
Chrysler Airflow. Nowadays, factors such as the increase in the
price of fuel, the dwindling of renewable energy sources, the
greenhouse effect, etc., have made an impact on the automobile
industry and encouraged the production of vehicles, which are less
polluting and are more energy-efficient. In turn, this has led to
modifications in vehicle-design criteria, as well as to the development and application of techniques suited to such purposes. Aerodynamics, which allows the optimization of vehicles by merely
varying their external design, changing the superficial finish, and
diminishing tolerance between component parts, is one such design strategy.
That is why in recent years the external appearance of vehicles
has undergone great changes with the purpose of reducing both
the transversal area and the aerodynamic penetration coefficient,
replacing the straight lines of yesteryear with new and sinuous
shapes and more stylized profiles 1. These developments are the
result of the application of strict efficiency and economic profitability criteria, since approximately half of the energy used by a
vehicle is used to counteract the resistance of the atmosphere. This
in itself would suffice to prove the importance of aerodynamics,
which is the only way of overcoming this invisible obstacle. Over
the past 30 years, the average penetration coefficient of automobiles has gone down by 40%, an improvement that has brought
about a 19% reduction in fuel consumption. However, similarly
significant developments in relation to the aerodynamics of larger
sized trailers have not taken place, because the principal developments studied in the past few years were on trucks. These are
essential for the future development of more environmentallyfriendly vehicles; should improvements be carried out, they could
entail a feasible 15% 2 reduction in fuel consumption.
Therefore, 20% improvements in a vehicles aerodynamic penetration coefficient can lead to savings of up to 10% in annual
JOURNAL OF FLUIDS ENGINEERING. Manuscript received September 1, 2008; final manuscript received December 5, 2008; published online March 11, 2009. Assoc. Editor:
Zvi Rusak.
expenditure, to savings of up to 12,000 l in fuel, and to the lowering of carbon dioxide emissions by almost 30 tons 3.
It should also be kept in mind that aerodynamic improvements
not only influence consumption and the power required by vehicles; they also increase their stability and decrease the effects of
aerodynamics on other vehicles traveling on the same road, as
well as diminishing splashing and improving the vehicles resistance to lateral winds, etc.
Therefore, an aerodynamic analysis of tankers was considered
essential in order to verify how certain elements improve the aerodynamics of the whole and thus contribute to the vehicles overall
performance.
New aerodynamic elements applicable to vehicle tankers and
which considerably reduce the vehicles aerodynamic resistance
were also developed.
Main Aerodynamic Improvements for Semitrailers
Before undertaking a detailed study of the aerodynamic improvements implemented, the vehicles energy budget has to be
calculated using Clarks equation 4, as follows:
P = 2 Cx At V3 + M G V + M G V sin
1
This equation is a simplified view of the energies required to

move a vehicle, since neither the performance of the transmissions
or the motors heat losses are taken into account, which nonetheless are extremely useful data when it comes to assessing the
contribution of each factor to the power required to set a vehicle
in motion.
As may be appreciated, aerodynamic power depends on the
cubing of speed, whereas the dependence of the other two equations is linear.
That is why aerodynamic improvements are truly effective at
high speeds, whereas at low speeds they may even result in an
increase in the vehicles maximum allowed weight PMA M,
which could be counterproductive.
Therefore, aerodynamic improvements are specially recommended for vehicles that travel regularly at high speeds, whereas
they are not recommendable for vehicles that perform distribution
tasks within cities or generally circulate at low speeds.
APRIL 2009, Vol. 131 / 041204-1
Fig. 3 Aerodynamic fairings
Fig. 1 Wind-averaged drag values for an articulated vehicle

with a square box 4
If we look closely at the previous equation we see that the main

factors that can be modified are the vehicles transversal area At,
the aerodynamic coefficient Cx, and the tire friction coefficient
with low consumption tires.
Regarding the vehicles transversal area, we may notice that the
width of an articulated vehicle designed for the transport of commodities is determined both by the width of the tractor and by its
height. However, the use of tankers with a low level chassis diminishes the overall height and allows the use of tractors with a
lower height or smaller superior spoilers.
As far as the aerodynamic resistance coefficient is concerned,
several aerodynamic improvements will lower this coefficient.
Among these, a distinction must be made between those which
apply to the tractor and those which apply to the semitrailer,
which is responsible for up to 70% of aerodynamic losses see
Fig. 1. These losses could be reduced by up to 40% by using the
available technology and would equate to a 28% reduction in the
aerodynamic resistance coefficient. Figure 1 shows the high pressure areas of an articulated vehicle associated with high aerodynamic losses.
If we manage to lower the effects of pressure on each of these
areas, the overall aerodynamics of the vehicle will therefore improve considerably.
It should be pointed out that latter-day aerodynamic improvements are applied mainly to square box 5 vehicles and therefore
have to be adapted for their use with tankers. Among these, we
will be examining boat-tails, aerodynamic undercarriage skirts
and aerodynamic fairings.
2.1 Boat-Tails . Boat-tails see Fig. 2 prevent the detachment
of the outer layer of the rear end of the box by preventing pressure
gradients that would generate vorticity.
There are several possible boat-tail configurations. One is the
use of rear plates up to 4% aerodynamic improvement. Another
is the application of round beam sections up to 5% 6 to the rear
area of the box, although the use of one type of boat-tail or the
other depends on the type of box, the maximum size of the vehicle, etc.
2.2 Aerodynamic Fairings. These aerodynamic elements are
fixed on the upper front area of the semitrailer see Fig. 3 and
redirect the air flow that appears in that area toward the box,
keeping it out of the king-pin cavity and thus involving an aerodynamic improvement of up to 3% 7.
2.3 Aerodynamic Undercarriage Skirt. This part prevents
the entry of air in the lower part of the vehicle, both in the tractor
area and the semitrailer area, decreasing air flow in these areas,
which due to the wheels and several parts of the vehicles lower
area would otherwise generate a turbulent flow that would adversely affect the vehicles aerodynamics. A decrease of up to 7%
8 can be achieved this way see Fig. 4.
3 Methodology for Aerodynamic Calculation Using

CFD
A commercial aerodynamic program was used for the aerodynamic calculation of the vehicle and its possible modifications.
The following methodology was used for modeling, discretization, and vehicle calculation.
3.1 Computer Aided Design (CAD) Modeling. In order to
proceed with the analysis, the first step is to perform CAD modeling of the vehicles outer area and remove or simplify certain
elements, such as the vehicles rearview mirrors, turn signals,
small external accessories, sleeves, and other auxiliary elements,
as the modeling of these elements would involve an excessively
complex and detailed finite element model 9,10.
In our particular case, a CAD model faithfully reflecting the
external dimensions of both the vehicle and the tractor was designed, and therefore we chose to use the main dimensions of a
tanker supplied by CryoEnergy S. A. Tarrega, Spain 11, to
which a medium range tractor was added see Figs. 5 and 6.
This cab was used as a basis, removing all the optional aerody-
Fig. 2 Boat-tails
041204-2 / Vol. 131, APRIL 2009
Fig. 4 Aerodynamic undercarriage skirt
namic accessories that may come with it: spoiler, lateral air flow
streamliners, and bumper spoiler. These elements were not used,
as we only intended to affect the semitrailer. The reason for not
using them is that since these elements are cab accessories, not all
Fig. 5 Baseline equipment trailer
Fig. 6 Baseline equipment trailer
cabs employ them, and also that, due to the fact that they have
been studied and optimized in depth, these elements have many
different configurations and shapes.
The air inlet of the vehicles radiator was also left out of the
modeling, because taking this factor into account would have required a very elaborate model of the vehicle, which would, in
turn, have involved numerical complications that would have
made calculation unfeasible. Besides, the contribution of this part
to the global model is relatively small and is not under study here
9,10,12.
As far as the tanker is concerned, the turn signals, sleeves, and
tubes of the lower area of the tanker, etc., were removed.
The several elements, which comprise the tankers lower area,
were not removed, since this is one of the areas under study,
despite the fact that this entails an increase both in the complexity
of the model and in the number of elements that comprise it.
In relation to the tire area, the wheel pitch and the empty spaces
within it, along with its dimensions, were maintained in order to
correctly model the contribution of these elements to the vehicle,
since the tire area significantly affects the vehicles overall aerodynamics and is one of the areas subject to study 9,10,12. It may
be worth pointing out that a symmetrical model was used to reduce the necessary computational cost. Consequently, only half
the vehicle was modeled.
3.2 Discretization Through Volumetric Elements. In order
to perform modeling using volumetric elements, the vehicles
CAD was taken as the starting point and introduced into a volume,
which is to simulate the air on the outside of the vehicle. The
following volumein keeping with the recommendations of other
authors 9,10,12 and those of the programwas considered for
this purpose 12.
In theory, there should be a distance of 2000 mm in the air
volume of a vehicle similar to the one selected between the front
part of the vehicle and the air inlet, and a distance of 10,000 mm
between the end of the vehicle and the air outlet in order to perfectly collect the slipstream generated by the vehicle. Regarding
both lateral and superior distances, these should be at least 4000
mm in relation to the outermost tangent surface.
The dimensions of this volume are such that the air circulating
in the areas furthest from the vehicle is not altered by the latter,
which implies that the imposition of contour conditions on the
outer area of the volume will not be altered by the vehicle, as this
would involve a modification of the models air flows and, consequently, an error in the results.
APRIL 2009, Vol. 131 / 041204-3
Fig. 7 Mesh of the baseline equipment trailer rear zone
The next step was to extract the vehicles volume from the total
volume, with which we could obtain a negative image of the
vehicle with regards to how discretization will be carried out
To do so, a tetrahedral mesh with lineal Lagrange elements was
chosen, because it allows moderately simple vehicle discretization
and an optimum degree of approximation in relation to computational cost.
First of all, discretization of the outer surfaces was performed
both on the inside and on the outside. The size of these surface
elements depends on the area in which discretization is carried out
9,10,12, on its magnitude and on previous experience, although
it was attempted to reduce the number of elements to the minimum, with the proviso that they be as equilateral as possible in
order to ensure convergence and reduce time and calculation requirements. A more elaborate discretization was performed in
those areas where greater speed gradients were expected, allowing
for improved simulation of air flow in those areas: the wheel
pitch, the front of the vehicle, the area between the cab, and the
tank and the rear part of the vehicle, where detachment of the
outer layer will occur.
After the discretization of the surfaces was performed, boundary layers 13 were imposed. This procedure allows us to per-
form fine meshing of the zones immediately surrounding the surfaces of the vehicle in order to faithfully model the vehicles outer
layer. To this end, an initial element size of 6 mm and a boundary
layer of three elements with a mesh growth factor of 1.2 9,10,12
was chosen. The reason for building such a mesh was to try to
correctly approximate the outer layer, despite the fact that this
entails a substantial increase in the number of elements and an
increase in computational cost; however, this is justifiable and was
necessary for the correct modeling of the outer layer of fluid.
Following this, discretization of the remaining air volume was
carried out; in order to do so, meshing of the outer area with a
larger size element 1000 mm was performed, since attached flow
will be preponderant in this area and there will be no flow alterations or speed gradients. In order to perform discretization, a size
function was used, allowing control of the growth size of the
elements from the vehicle to the outside area. To achieve this, a
maximum element size of 1000 mm and a 1.2 growth factorthe
maximum recommended by the programwas imposed. In this
way, discretization with the most equilateral surfaces possible can
be achieved see Figs. 7 and 8.
Fig. 8 Mesh of the baseline equipment trailer middle zone and tractor
041204-4 / Vol. 131, APRIL 2009
Fig. 9 Flow around a semitrailer tanker side view
3.3 Aerodynamic Calculation. In order to perform aerodynamic calculation, several tasks have to be carried out prior to the
calculation proper:
3.3.1 Definition of Solution Options. A segregated solution
modelrecommended for this type of calculationswas chosen
in order to lower computational requirements. This model uses a
double precision calculation recommended for aerodynamic calculation of vehicles 14,15 for a stationary steady model based
on node calculation and is recommended by the program in order
to achieve better results with this type of calculation.
3.3.2 Definition of the Turbulence Model. A turbulence model
based on k- type realizable Reynolds average Navier-Stokes
RANS equations 1416 and a wall treatment with nonequilibrated flow were chosen 17. The choice of this turbulence model
was mainly due to the fact that nowadays it is the most commonly
used one for aerodynamic calculations and because it also offers
an optimum degree of approximation in regard to computational
requirements.
3.3.3 Definition of Contour Conditions. Since there are several
areas, different contour conditions were considered for each of
them. An air flow at a constant speed 30 m/s in a tangential
direction 7 with very low turbulence 0.1% turbulence intensity
was introduced for the entry boundary, which means that any flow
that there may be will be laminar. Regarding the upper, lateral,
and rear areas, these were simulated as being open to the outside,
with a pressure of 1 atm and an air flow of 30 m/s in the relevant
direction.
The ground was modeled as a smooth surface traveling at the
same speed as the fluid 30 m/s 18, and the wheels were considered smooth and rotational surfaces, coherent with the speed of
the vehicle 17.
Symmetry was imposed on the interior wall and the vehicles
walls were considered fixed and smooth walls without friction.
This was due to the fact that the contribution of friction between
the fluid and the surface in relation to the global model represents
less than 10% of the resistant contribution and that the implementation of this contribution would entail a substantial increase in
computational cost and might even have lead to divergent results.
Nonetheless, it would be useful to establish, very precisely and by
means of experiments, the friction coefficient of each of the models superficial finish types in order to establish their contribution
to the aerodynamic resistance, which would not be substantially
modified by variations in the vehicles geometry, since the latter
depends mainly on the choice of superficial finishes.
Thus, simulation of the movement of the vehicle is possible, but
from the relative viewpoint of the tanker.
3.3.4 Selection of Options for Solving the Equations. A
13 type resolution algorithm was chosen to solve the
SIMPLE-C
pressure-speed fields, which is the appropriate resolution scheme

for discretizations with elements not aligned with the flow 19. In
addition, standard discretization schemes were used for pressure
solution and second order schemes were used for calculating momentum and turbulence. These are the parameters recommended
by the program for this type of calculations.
It should be pointed out that in order to improve the models
convergence, first order equations were used for calculating momentums and turbulences during the first 300 iterations and that
the second order scheme was implemented subsequently.
4 Adaptation of Aerodynamic Improvements to the

Tankers
In this paper, only those aerodynamic improvements directly
applicable to cryogenic tanks will be analyzed; and therefore, despite the fact that they improve the aerodynamics of the tank area,
the improvements applied to the tractor will not be taken into
account in this study.
Before proceeding with the analysis, it should be pointed out
that aerodynamic resistance coefficients for latter-day articulated
vehicles can be extracted from article 20, and that these can be
used to verify the results obtained in the calculation of the original
vehicle. Thus, the aerodynamic resistance coefficient of a vehicle
with the characteristics of the one we are dealing with is roughly
0.69.
As a starting point for the analysis of this type of vehicles, the
way in which air flow covers the outer surface of the vehicle must
be understood qualitatively. In order to do so, we will set out from
the previous experience existing regarding automobiles 2022.
On the basis of this information it may be presumed that in a
tanker-type articulated vehicle 23, there will be a flow of particles like the one shown in Fig. 9. Therefore, just like in a car,
there will be an area at the front where particles will stagnate due
to their collision with the frontal area, although since our frontal
area is much larger and its aerodynamic shape is much less stylized than a cars, and also because this area is perpendicular to the
flow, this area will be much larger. There will also be an area near
the king-pin cavity in which the outer boundary will become detached due to the existence of the said cavity, whereas the particles entering the cavity will collide with its rear part, creating a
turbulent zone and, presumably, low longitudinal speed and high
vorticity.
Since the external geometry of the tank has no protuberances,
projections, cavities, etc., it is likely that coupling will recur behind the cavity of the king-pin, and therefore, that there will be
attached flow in this area.
In the vehicles rear area, where the surface ends, flow separation will recur, presumably with low longitudinal speed, and a
turbulent slipstream will be generated behind the vehicle.
APRIL 2009, Vol. 131 / 041204-5
Table 1 Total resistant forces and aerodynamic resistance coefficients of all the cases considered
The existence of turbulent flow in the lower part of the vehicle

may be presumed due to the large amount of elements and projections in that area. Besides, the presence of the wheels and their
turning will generate turbulence and additional flow detachment.
When we speak of null speed we assume a viewpoint relative to
the vehicle.
Once the flow was theoretically established, the next step was a
numerical verification of the hypotheses set forth; therefore, numerical calculation and postprocessing of the original configuration of the articulated vehicle was carried out.
Total forces and aerodynamic coefficients for all configurations
are reflected in Table 1.
The global aerodynamic resistance coefficient is 0.618, which is
within the established range for this type of vehicles 24. This
coefficient is somewhat high, because a tractor without aerodynamic elementswhich involve considerable aerodynamic
penalizationwas used. This is also appreciable in relation to the
contribution of each part of the articulated vehicle to the global
resistance coefficient: 55.4% in the tractor and 44.6% in the body.
It should also be pointed out that the bodys contribution to the
vehicles average total drag is inferior to the 6575% range indicated in Fig. 1. This is because this figure established the contribution for a square box semitrailer, the aerodynamic coefficient of
which is usually 5% superior to that of a semitrailer tanker 18
and due to the use of a vehicle with aerodynamic improvements in
the tractor.
Once the correctness of the aerodynamic resistance coefficient
and the resistant forces of the vehicle have been established, the
next step is to analyze the several areas, which make up the vehicle so as to verify in which areas flow detachment occurs and
turbulent flow predominates. In order to do so, we analyzed pressure maps high pressures indicate vorticity and turbulent flow,
speed maps low relative speeds indicate vorticity and turbulent
041204-6 / Vol. 131, APRIL 2009
flow, turbulence energy maps, and particle flow maps draft lines
were analyzed in order to perform this analysis, as well as an
analysis of the flow of particles in motion, which was not included
here due to its being a moving image.
It must be pointed out that in order to keep the article relatively
short, only the more representative images of the vehicle were
included.
As shown in Fig. 10 and predicted earlier, the low speed areas
are the lower part of the vehicle, the king-pin cavity area and the
slipstream of the tank; i.e., these are the areas with greater turbulence and higher pressures. This figure shows the longitudinal
speed of the vehicle. It shows the wake in the rear part of the
vehicle and indicates a high aerodynamical resistance. It also
shows, in the bottom part of the vehicle past the tractor, a zone
with low velocity and the same in the wheel zone of the tanker.
This indicates high aerodynamical losses too. Analyzing Figs. 11
and 12 that show the k turbulence parameter and then the zones
with high aerodynamical losses, these match with the zones previously mentioned and with Fig. 9, so these zones should be
modified to improve the aerodynamics. These zones also have
whirls that indicate turbulence see Fig. 10 and 13 and cause
aerodynamical losses, usually because the geometry of the vehicle
is discontinuous, and the attached flow changes to a turbulence
one. In Fig. 10 there is a zero velocity zone in the gap between the
tractor and the trailer and it matches with a high losses zone see
Figs. 11 and 15.
We also notice in Figs. 10 and 13 a stagnation zone in the front
area that is one of the zones with high aerodynamical losses of
the tractors, the detachment of the outer layer above the cab in
the vehicles rear part and the lower part of the tractor, as well as
coupling of the flow in the upper part of the tank all was
predicted in Fig. 9.
Fig. 10 Longitudinal speed of the original configuration at the vehicles plane of symmetry
Moreover, we may appreciate high fluid speeds which are

completely normal on the edges of the tractor see Figs. 14 and
15 as a result of the latters stagnation.
If we perform a cut, which is parallel to the plane of symmetry
and near the lateral surface of the cab, we can appreciate the effect
of the tires on the speed and the increase in turbulence due to this
phenomenon see Fig. 16 and an increase in the aerodynamical
losses, and the effect of the change in geometry on the bottom part
of the trailer near the tractor
If we perform a cut on a horizontal plane located 2 m above the
road we can verify the flows and speeds shown in Fig. 5 and we
can also obtain flows, turbulence, speed, etc.
Fig. 11
A close look at Fig. 13 reveals that the flow of particles is very

similar to the one proposed and that the areas where the k turbulence parameter is highly correlated with those areas in which
high aerodynamic losses were predicted.
The analysis of the tire area, where turbulent flow predominates, as evidenced in Fig. 20, is also worth paying special attention to. A detachment of the outer layer in the areas near the
wheels may also be appreciated, as well as in the lower part of the
vehicle, the latter presumably due to the discontinuity between the
tractor and the semitrailer.
Once the analysis was carried out, the conclusion was that the
k turbulence parameter of the configuration original at the vehicles plane of symmetry
Fig. 12
k turbulence parameter on a horizontal plane 2 m above the road
APRIL 2009, Vol. 131 / 041204-7
Fig. 13 Flow lines and associated speeds of the original configuration at the vehicles plane of
symmetry for the cab and rear areas
previously established flow hypotheses were correct, and consequently that modification of the king-pin cavities and the wheels
would be necessary in order to reduce the vehicles aerodynamic
losses. Regarding the slipstream area of the vehicle, no initial
modifications had to be undertaken, since the aerodynamic profile
and the level of turbulence are correct and flow detachment cannot
be avoided in that area; the configuration with the box in the rear
will require modification, since it generates greater flow detachment and more turbulence, which ought to be prevented.
It is worth mentioning that depending on the positioning of the

box in the rear or the front part there are two different tanker
configurations. Each of them will be studied separately and certain
aerodynamic improvements will have to be adapted for each configuration. Due to the shape of the tank itself, some of the aerodynamic improvements proposed for vehicles with a square box
are inefficient for this type of semitrailer. The use of boat-tails,
aerodynamic fairings, and rear vorticity generators, for example,
would require adaptation. Due to the geometry, the use of vortex
Fig. 14 Longitudinal speed of the original configuration at a plane parallel to the plane of
symmetry and in the tire area
Fig. 15 Longitudinal speed of the original configuration at a plane parallel to the plane of
symmetry tangential to the tank
041204-8 / Vol. 131, APRIL 2009
Fig. 16
k turbulence parameter on a horizontal plane in the tire area
traps in the lower front part of our vehicle is also impossible, and
therefore, this improvement cannot be implemented either.
The main aerodynamic elements subject to study will be undercarriage skirts. An aerodynamic modification based on the boattail will also be designed for the vehicle with a rear box configuration in order to prevent the detachment of the outer layer of this
zone, which we will call aerodynamic box adaptor; on the
other hand, an aerodynamic improvement for the vehicles front
areain order to prevent the generation of vorticity and the detachment of the outer layer of the king-pin cavitywill be designed on the basis of a modification of the aerodynamic fairings
and the lateral air flow streamliners, but will be applied to the tank
and will be named aerodynamic front.
4.1 Undercarriage Skirts. At present, all newly built semitrailers have to implement lateral protection measures in compliance with EEC guideline 98/297. This directive regulates the use
of lateral protectors. The use of these protective elements, in their
horizontal spar configurationthe most widely used one
nowadaysleads to aerodynamic penalization, since they favor
detachment of the outer layer due to their discontinuities with the
tank.
Furthermore, the cavities of the wheel area allow air to penetrate this area and therefore generate an aerodynamic penalization, which increases in proportion to the increase in the air flow
running through that area; therefore, the use of aerodynamic undercarriage skirts will be implemented to reduce the air flow laterally penetrating the area.
Consequently, an aerodynamic improvement was suggested in
order to decrease the aerodynamic penalization caused by these
elements.
The aerodynamic improvement in question is based on the reduction of free edges and geometries capable of generating a detachment of the outer layer in the area between the tractor and the
wheels of the vehicle, since in this area there are many elements
that generate vorticity and aerodynamic losses.
To this end, we suggest a continuous surface going from the
front area of the semitrailer to the rear area, comprising of three
clearly differentiated areas, and additionally, made out of different
materials.
1 The area between the tractor and the parking holder A:
this area is to be made of plastic materials, since it will not
be subjected to important mechanical stress. Its function is
to avoid air penetrating the lower area of the tank from the
lateral areas, mainly issuing from the tractor.
2 The area between the parking holder and the wheels B:
this part was designed according to EEC directive 98/297
and allows the use of said area as a lateral protector in
compliance with previous specifications.
3 The area of the semitrailers wheels C: this part covers
both the cavities located above the vehicles wheelguards
and those between the semitrailers wheels, as these cavities
generate vorticity and aerodynamic losses.
We can see the aerodynamic improvements in Fig. 17.
When we analyze the results obtained we will see that k turbulence is reduced in the lower part of the vehicle see Figs. 16 and
18. This indicates a decrease in turbulence in the area, and therefore a decrease in aerodynamic losses.
The trajectories of particles Fig. 19 follow the direction of the
flow and very small amounts of fluid penetrate the vehicles lower
area, as might have been expected, while the curved part in area A
performs its task, preventing air flow in that area Fig. 20. This
indicates that the turbulence flow first has decreases and then
aerodynamical losses. This is because the underskirt avoids abrupt
geometry changes and then the whirls that are generated are lower
see Figs. 18 and 20 versus Fig. 11.
On the other hand, we may see that the final curved shape of
area C of the undercarriage skirt redirects flow and reduces slipstream in that area see Fig. 20.
Regarding the forces acting on the vehicle see Table 1 and the
aerodynamic resistance coefficient, we see that the vehicles aerodynamic coefficient is 0.56, i.e., 9% less, which means that the
aerodynamic benefit of this part is considerable and has no impact
in terms of weight increase approximately 77 kg
4.2 The Aerodynamic Front. The king-pin area is a zone in
which detachment of the outer layer and high vorticity occur, as
shown in Fig. 11. Consequently, it is an area with high aerodynamic losses, and therefore we suggested an aerodynamic improvement in order to reduce the king-pin cavity and redirect the
air in that area to prevent whirlwinds and detachments of the outer
layer.
The aerodynamic front was designed so as not to interfere with
the vehicles maneuvers or with access to the area when it comes
to connecting it to the tractor. With this in mind, geometry such as
Fig. 17 Aerodynamic undercarriage skirt in a configuration with the box in the lower part versus linebase one
APRIL 2009, Vol. 131 / 041204-9
Fig. 18
k turbulence parameter in the aerodynamic undercarriage skirt area
the one which appears in Figs. 21 and 22 was suggested.

This geometry reduces the size of the king-pin cavity both in its
lower and upper parts, and also performs the same function as the
lateral air flow streamliners of the tractor, preventing air from
penetrating the cavity and redirecting it to the outside area.
It is worth pointing out that the aerodynamic front can be made
out of low density materials such as aluminum or plastic, as it is
not an element subject to mechanical stress. Therefore, given an
appropriate selection of materials and low thickness, the weight
penalization for including these materials will be inferior to the
aerodynamic improvement obtained. The approximate weight of

the part when built in PVC with an approximate thickness of 3
mm will be 42.3 kg.
After analyzing the results obtained for this configuration, we
appreciate a reduction in the k turbulence parameter of the aerodynamic front Fig. 23 as opposed to Fig. 11 in relation to the
initial parameter, which indicates a decrease in turbulence in the
area and therefore an aerodynamic improvement, as expected.
This is because the gap between the tractor and the trailer has
Fig. 19 Speed in the aerodynamic undercarriage skirt area
Fig. 20 Detail of areas A right and C left of the aerodynamic undercarriage skirt
041204-10 / Vol. 131, APRIL 2009
Fig. 24 Flow lines at the plane of symmetry and on a vertical

plane 2 m above the ground in the king-pin cavity area for a
vehicle with an aerodynamic front
Fig. 21 Vehicle tanker with an aerodynamic front versus linebase one
been reduced and the new geometry reduces geometry changes.

Likewise, if we analyze both the path lines and the speed of the
vehicle Fig. 24 as opposed to Fig. 14, we appreciate a decrease
in turbulent vortices in the king-pin cavity area because the geometry changes are lower.
In relation to the models forces and aerodynamic coefficients, a
6.2% force reduction, besides a substantial improvement of the
aerodynamic resistance coefficient, which now has a value of
0.58, Table 1 may be pointed out.
4.3 The Aerodynamic Box Adaptor. The vehicle being analyzed has two different box configurations in which the elements
for manipulating the load are placed and which allows loading and
unloading operations. One configuration has the box in the rear
part and the other one has the box in the lower part of the vehicle.
The disadvantage of the configuration with the box in the rear
part of the vehicle is that it causes disturbances in air flow lines,
increases vorticity, and leads to detachment of the outer layer, as

can be appreciated in Figs. 25 and 26 that indicate zones with high
aerodynamical losses. This is due to the fact that this configuration
involves cavities between the box and the rear part of the tank,
and thus the aerodynamic improvement of the tanker is due to the
shape of its rear bottom part, which reduces the vorticity of the
vehicles slipstream as well as the detachment of the outer layer.
In the following images we may appreciate the increase in vorticity due to the presence of the box and an increase in the slipstream
generated behind the vehicle see Fig. 27. We may also appreciate an increase in the k turbulence parameter in relation to the
model with the box in the lower part Fig. 11 as opposed to Fig.
25.
As might have been expected, this rear box configuration considerably worsens the aerodynamic coefficient, which reaches
0.657, i.e., 6.25% more.
The use of aerodynamic elements between the box and the rear
part of the tank in order to allow for transition between these
zones is suggested. To this end, curved shapes that prevent the
Fig. 22 Vehicle tanker with an aerodynamic front versus linebase one. Top view.
Fig. 23 k turbulence parameter on the plane of symmetry and on a vertical plane 2 m above
the ground in the king-pin area for a vehicle with an aerodynamic front
APRIL 2009, Vol. 131 / 041204-11
Fig. 25 k turbulence parameter on a horizontal plane 2 m above the ground and at the plane of
symmetry for a rear box configuration without improvements
detachment of the outer layer of the said area and minimize overall aerodynamic losses were used.
Since these elements are not subject to significant mechanical
stress, they can also be made out of plastic, which minimizes the
total weight contribution 57 kg in 3 mm PVC
As a result, the vehicles slipstream Figs. 25 and 26 as opposed

to Fig. 29 is reduced.
If we pay attention to the movement of particles, we will notice
that flow detachment is not as sharp in the vehicles rear area Fig.
28, and therefore an aerodynamic improvement might be ex-
Fig. 26 k turbulence parameter at the plane of symmetry for a rear box configuration without improvements
Fig. 27 Final box adaptor versus baseline one
041204-12 / Vol. 131, APRIL 2009
Fig. 28 Detail of the flow lines at the plane of symmetry in a vehicle with final box adaptor
pected in this area.

The k turbulence parameter decreases in the rear part of the
vehicle Fig. 26 as opposed to Fig. 29.
We can appreciate a 7.6% reduction in the forces and aerodynamic resistance coefficients Table 1 and a 0.606 aerodynamic
coefficient in relation to the rear box configuration. The lower part
of the vehicle shows an aerodynamic improvement of 3% due to
the fact that the rear part is more aerodynamic than the front,
although this reduction does not compensate for the need to
change the geometry of that area in the vehicle with lower box
configuration see Fig. 30.
Fig. 29
4.4 Global Contribution of all the Aerodynamic Improvements Shown. To complete the analysis, an aerodynamic calculation of both box configurations including all the aerodynamic improvements at our disposal was carried out in order to verify the
overall contribution of these improvements to the vehicle.
To this end, all previous aerodynamic improvements were
adapted with the purpose of jointly integrating them with the vehicle and achieving a perfect union of the different parts in order
to avoid detachment of the outer layer and make the vehicle even
more aerodynamic.
Consequently, these were the geometries chosen.
k turbulence parameter at the plane of symmetry in a vehicle with final box adaptor
Fig. 30 Front area of a vehicle with all the aerodynamic improvements, regardless of box configuration and equipped with a
rear box in the rear area
APRIL 2009, Vol. 131 / 041204-13
Fig. 31 Speed at the plane of symmetry in a vehicle with lower box configuration and all the
improvements
If we examine the results obtained for the lower box configuration, we find a 0.52 aerodynamic coefficient, i.e., 15.85% less,
due both to the contribution of the undercarriage skirt and that of
the aerodynamic front, as well as to the joint integration of both
elements. This can be appreciated in Figs. 3133, in which we
may notice the reduction of the k vorticity parameter in the undercarriage skirt area and the king-pin cavity. We also notice the
movement of particles, which tends to produce less vortices and,
consequently, to generate less turbulence. Thus the aerodynamic
improvements are expected to prove effective see Table 1.
The implementation of all the aerodynamic improvements for
the rear box configuration yields a 0.505 aerodynamic coefficient,
i.e., 23% less in relation to the original rear box configuration, as
might have been expected see Fig. 34. This aerodynamic coefficient is noticeably inferior to that of the vehicle with the lower
box configuration including all the aerodynamic improvements,
due to the fact that the configuration used for the rear box is more
aerodynamic, although the improvement obtained amounted only
to 3%.
As far as turbulence is concerned, and as a result of the aerodynamic improvements implemented in these areas, we notice a
decrease in the value of the k turbulence parameter in the slipstream, the wheel area and the king-pin cavity Fig. 33 versus Fig.
14.
We may also notice, in Fig. 35 versus Fig. 13, that there are less
vortices in the flow lines and that flow is redirected.
Figures 33 and 36 versus Fig. 13 shows that the turbulence in
the bottom zone and in the wheel zone of the trailer has been
reduced, because the underskirt decreases the geometry changes
and redirects the air flow properly see Figs. 32 and 35 versus Fig.
13.
Economical Analysis
All the improvements that have been used in the cryogenic

tanker trailer can be used in any type of tanker trailer without the
need to make any modification, so the analysis that has been done
is for all types of tankers. The benefit of these improvements is
clear regardless of the vehicle tanker, but the percentage can
change depending on the geometry of the tanker. In this paper a
standard tanker has been used.
To know if the improvements are viable it is essential to make
an economic analysis of them in the US market, so we have used
the NADA 21,22,25 database and the Truck Inventory Use and
Survey 26. They indicate that in 2003 there were, in USA,
85,174,800 trailers and out of them 34,056,134 were tanker trail-
Fig. 32 Flow lines on a vertical plane 1 m above the ground in a vehicle with lower box
configuration and all the improvements
041204-14 / Vol. 131, APRIL 2009
Fig. 33 k turbulence parameter at the plane of symmetry in a vehicle with a lower box and the
improvements
ers. For the cost analysis a price of 2.1 $/gal 11/09/2008 has
been used and a vehicle consumption of 16 gal/100 miles and
100.000 miles/year for a standard vehicle. This indicates an annual consumption of 16,000 gal and a cost of 33,600 $/year. As
for the improvements, the repercussion of them on the consumer
depends on some factors average speed, driving style, orography,

etc., but can vary from 10% for slow trailers to 50% for trailers
that make railway routs. 40% has been used. The potential costs
have been estimated see Table 2.
The economic results show that with low cost less than $1100
Fig. 34 Speed at the plane of symmetry in a vehicle with rear box configuration and all the
improvements
Fig. 35 Flow lines on a vertical plane 1 m above the ground in a vehicle with a rear box and all
the improvements
APRIL 2009, Vol. 131 / 041204-15
Fig. 36 k turbulence parameter on a vertical plane 1 m above the ground in a vehicle with a
rear box and improvements
Table 2 Economical analysis of the improvements
the fuel consumption is cut down and the annual saving can be
$2100, so the return on investment ROI is approximately 0.5
years and the economic gain is clearly large. all economic results
are in Table 2. There are also environmental benefits with the
CO2 emissions being reduced by 2.73 Tm during a year.
Conclusions
Following a detailed analysis of the results obtained and the

methodology implemented, the conclusion reached was that CFD
programs are very useful tools for performing aerodynamic modifications in vehicles at a very low cost, and that they allow the
achievement of satisfactory results in a short period of time.
There is a wide field of research and a number of improvements
to be explored as far as aerodynamic improvements in articulated
tanker-type semitrailers are concerned. This is due to the fact that
latter-day vehicles have few or no aerodynamic improvements
owing to the small amount of research carried out in this transport
sector. Therefore, bearing in mind the results obtained and the
future prospects of reducing fuel consumption in vehicles, this
field of research would seem most promising.
It must be pointed out that the coefficient of aerodynamic resistance can be considerably lowered without modifying the dimensions of the elements subject to mechanical stress by adding
external aerodynamic elements that do not entail substantial
weight penalization, such as those included in this chapter. Thus
the vehicles presently being used can implement these improvements to reduce fuel consumption and improve energy efficiency
without having to resort to far-reaching modifications.
In relation to the aerodynamic improvements introduced, it is
worth pointing out that the aerodynamic skirt entails the greatest
aerodynamic gain of all, and involves a 9% improvement in the
aerodynamic coefficient, by preventing the flow of air toward the
lower part of the vehicle, where turbulence is predominant.
041204-16 / Vol. 131, APRIL 2009
The contribution of the remaining aerodynamic improvements

is also relevant, and their use is always advantageous see Table
1, although the undercarriage skirt is preferable as an initial aerodynamic improvement. Should the others be implemented as well,
they ought to be integrated with the use of the skirt. The integrated
use of all these improvements can lead to an aerodynamic improvement of up to 23%, which would measure up to an improvement of up to 11% in fuel consumption. Although this comes
close to the 15% expounded in Sec. 1, it does not include improvements in the tractor or those improvements due to modifications in the shape of the resistant elements, which may make up
for that 4% difference.
Last but not the least relating to the economic aspect, the aerodynamical improvements have a ROI of 0.5 years and the economic benefit is clearly substantial.
References
1 Morelli, A., 2000, A New Aerodynamic Approach to Advanced Automobile
Basic Shapes, SAE Technical Paper No. 2000-01-0491, p. 104.
2 Wood, R. M., and Bauer, S. X. S., 2003, Simple and Low Cost Aerodynamic
Drag Reduction Devices for Tractor-Trailer Trucks, SAE Paper No. 2003-013377.
3 Bachman, L. J., Erb, A., and Bynum, C. L., 2005, Effect of Single Wide Tires
and Trailer Aerodynamics on Fuel Economy and NOx Emissions of Class 8
Line-Haul Tractor , SAE Paper No. 05CV45.
4 Rutledge, W. H., Polansky, G. F., and Clark, E.L., 1988, Aerodynamic Design
and Performance of a Bent-Axis Geometry Vehicle, J. Spacecr. Rockets,
254, pp. 257262.
5 Wood, R. M., 2004, Advanced Aerodynamic Trailer Technology.
6 2007, FreightWings web page, http://www.freightwing.com
7 2007, Nosecones web page, http://www.nosecone.com
8 Graham, S., and Bigatel, P., 2001, Freight Wing Trailer Aerodynamics, Technical Report No. JA-132.
9 Kleber, A., 2005, Simulation of Air Flow Around an OPEL Astra Vehicle
With Fluent, International Technical Development Center, Adam, Opel AG,
Journal Articles by FLUENT Software users, www.fluent.com/solutions/articles/
ja132.pdf
10 Saiz, V., 2005, Estudi i Simulaci Aerodinmica del Roadster FS1, MS

thesis, Escola Tcnica dEnginyers Industrials de Barcelona, Barcelona, Spain.
11 ROS ROCA INDOX CryoEnergy, http://www.cryoenergy.net
12 Canet, J. M., Bugeda, G., and Oate, E., 1991, Simulacin Numrica de la
Aerodinmica de Vehculos, Centro Internacional de Mtodos Numricos en
Ingeniera, Barcelona, Spain.
13 2004, FLUENT 6.2 Documentation, Users Guide, Fluent, Inc.
14 Parneix, S., Durbin, P. A., and Behnia, M., 1998, Computation of 3-D Turbulent Boundary Layer Using theV2 Model, Flow, Turbul. Combust., 60, pp.
1946.
15 Menter, F. R., 1994, Two-Equation Eddy-Viscosity Turbulence Models for
Engineering Applications, AIAA J., 328, pp. 15981605.
16 Davidson, L., 2003, An Introduction to Turbulence Models, Chalmers University of Technology, Reston, VA.
17 Durbin, P. A., 1991, Near-Wall Turbulence Closure Modelling Without
Damping Functions, Theor. Comput. Fluid Dyn., 3, pp. 113.
18 Mercker, E., and Knape, H. W., 1989, Ground Simulation With Moving Belt
19
20
21
22
23
24
25
26
and Tangential Blowing for Full-Scale Automotive Testing in a Wind Tunnel,

SAE Paper No. 890367.
Skea, A. F., Bullen, P. R., and Qiao, J., 1998, The Use of CFD to Predict the
Air Flow Around a Rotating Wheel, MIRA, International Symposium on
Automotives Technology and Automation, Dsseldorf, Germany.
Wong, J. Y., 1993, Theory of Ground Vehicles, 2nd ed., Wiley, Ottawa, Canada.
Carr, G. W., 1983, Potential for Aerodynamic Drag Reduction in Car Design, Int. J. Veh. Des., pp. 5152.
Carr, G. W., 1983, Potential for Aerodynamic Drag Reduction in Car Design, Int. J. Veh. Des., pp. 176196.
Heisler, H., 2002, Advanced Vehicle Technology, 2nd ed., Butterworth
Heinemann, Oxford, pp. 584617.
Comisin Nacional para el Ahorro de Energa de Mxico Mexico D.C.,
Mexico, 2007, www.conae.gob.mx
National Automobile Dealers Association, http://www.nada.org/
U.S. Census Bureau, Truck Inventory and Use Survey, http://
www.census.gov/svsd/www/tiusview.html
Dr. Ramon Miralbes Buil obtained a degree in Industrial Engineering from Zaragoza University in 2005 (Spain), with specialization
in Vehicle and Machinery Design. From this time on, he continued his Ph.D. studies in the area of New Materials for Road Transport
in this university. In February 2008 he finished his Ph.D. studies with the thesis: Flow-Termo-Mechanical Analysis of Cryogenic
Tanker Vehicles and is working on several research projects involving vehicle strength and rigidity performance simulation by the
FEM. He also works as an Associate Professor in the area of Design and Graphic Expression at the University of Zaragoza.
Dr. Luis Castejon Herrer graduated with a degree in Mechanical Engineering (University of Zaragoza, Spain, 1992) and a Ph.D. in
Mechanical Engineering (University of Zaragoza, Spain, 1998). Dr. Luis Castejon Herrer is a full Professor in Transportation Engineering at the University of Zaragoza. Since 1991, he has been involved in several projects in the areas of Vehicle Dynamics and New
Materials for vehicles, buses, railways, and ground transportation. He has written over 120 papers in technical journals and international congresses and 9 books, published in Europe. He has been a researcher in the European BRITE-EURAM Project MULTEXCOMP and he has collaborated in other European projects such as SEABUS and AFICOSS. He has participated in several R + D
Spanish projects. He developed his Ph.D. thesis on the simulation of roll over of metallic and composite buses by means of FEM,
obtaining the Doctoral Special Award of the University of Zaragoza.
APRIL 2009, Vol. 131 / 041204-17
R. W. Hewson1
School of Mechanical Engineering,
University of Leeds,
Woodhouse Lane,
Leeds, LS2 9JT UK
e-mail: r.w.hewson@leeds.ac.uk
Free Surface Model Derived From

the Analytical Solution of Stokes
Flow in a Wedge
The formation of a thin liquid film onto a moving substrate is a commonly encountered
industrial process, and one that is encountered in lubrication, oil extraction processes,
and coating flows. The formation of such a film is analyzed via the analytical Stokes flow
solution for the flow in a wedge bounded on one side by a free surface and on the other
by a moving surface. The full solution is obtained by numerically integrating a set of
ordinary differential equations from far downstream, in the region of the final film thickness. The results show excellent agreement with the results obtained by the Bretherton
equation, the Ruschak equation, the Coyne and Elrod model, and a two-dimensional free
surface finite element simulation of the problem. DOI: 10.1115/1.3089540
Introduction
The deposition of a thin liquid film is commonly encountered in

a number of tribological and coating applications 13 and involves the deposition of the thin film onto a moving substrate as it
sweeps from a fluid filled gap, as shown in Fig. 1. As such there
have been a number of analyses of the problem, most notably that
of Bretherton 4 see also Ref. 5 who, using a similar method
of matched asymptotic expansion as Landau and Levich 6, determined the residual fluid deposited onto a moving substrate by,
h0
= 1.337Ca2/3,
r
Ca 0,102
where h0 is the residual film thickness, r is the constant meniscus radius, and Ca is the capillary number given by Ca= U /
is the fluid viscosity, U is the substrate speed, and is the interfacial surface tension.
Ruschak 7 used the finite element method FEM to obtain an
empirical expression governing the residual film thickness to gap
ratio for higher capillary numbers than those for which the
Bretherton expression is valid.
h0
= 0.54Ca,
h =
Ca 102,101
where h= is the initial fluid filled gap. One limitation of Ruschaks equation is that although it predicts the residual film thickness accurately, the nonconstant meniscus radius means that Eq.
2 is not able to calculate the local pressure at the point of film
formation, an important parameter, and one that is often used as a
pressure boundary condition in a range of different coating and
lubrication problems. This is noted by Thompson et al. 8 who
compared the finite element method with the lubrication approximation for reservoir fed rigid roll coating.
Coyne and Elrod 9 derived a set of four ordinary differential
equations ODEs to describe the process of film formation by
solving the NavierStokes equations along the free surface. The
model assumes a quadratic velocity profile to describe the flow
perpendicular to the free surface. One limitation of the model is
the stiffness of the set of equations around the free surface stagnation point occurring at 3H. Despite this, the method has been
used by some investigators as a free surface boundary condition
1
received January 15, 2009; published online March 11, 2009. Assoc. Editor: Malcolm
J. Andrews.
for lubrication 3 via a function fitted to the data in Coyne and

Elrods paper 9 as well as for coating problems 2,7.
The cavitation boundary condition can have a significant effect
on the pressure profile within the lubricated geometry studied
10. In addition to the film splitting model described here, there
are a number of other models that can be applied to lubrication
flows to describe the film splitting process, these include the
SwiftSteiber condition 11 and the PrandtlHopkins condition
12,13 u = du / dx = 0 as examined for the case of roll coating by
Savage 14.
Hewson et al. 15 solved the film forming problem as a boundary value problem BVP for shear thinning fluids obeying both
the power law and the Ellis fluid models. The results for the Newtonian fluids power index n = 1 compared well with those of the
Coyne and Elrod model and the Bretherton and Ruschak equations within their ranges of applicability. The model described
here is an attempt to provide a solution to the film forming problem, in a similar way to that of Coyne and Elrod i.e., the numerical solution of ODEs using equations that are computationally
more stable, in doing so the film forming process has been described for a range of capillary numbers and contact angles. The
primary difference between the model presented here and that of
Coyne and Elrod is that the flow described here is based on the
solution of Stokes flow in a wedge type geometry rather than as a
quadratic velocity profile perpendicular to the free surface as is
consistent with the lubrication approximation.
Model Formulation
The problem is solved using an analysis similar to that of the

long wave approximation but with the local flow solution based
on the Stokes solution for the flow in a wedge shaped geometry.
For such an assumption to be strictly valid, the free surface radius
of curvature must be greater than the residual film thickness
16in addition to the work of Snoeijer the analysis here uses a
similar method to that of Cox 17,18. This is valid in the region
where the free surface is close to the moving substratei.e., in the
region where viscous forces are greatest 4. In undertaking the
analysis the variation in film thickness in the region where the free
surface is close to the moving substrate is taken into account in
determining the free surface profile.
The solution of Stokes flow in a wedgelike geometry with one
free surface, one moving surface, and a net flux is formulated in
polar coordinates and solved in terms of the stream function , as
follows:
APRIL 2009, Vol. 131 / 041205-1
Free surface
Phase 1
Film
Phase 2
(r, )
U
Fig. 1 Illustration of the fluid film forming process
4 = 0
Moving Substrate
where ur = 1 / r / and u = / r are the velocity components

in the radial and normal directions, respectively. The problem geometry and local coordinate system employed is shown in Fig. 2.
All lengths x , r have been made nondimensional by dividing by
the residual film thickness H, by making the velocities nondimensional by dividing by the substrate velocity the flow rate q
within the fluid domain is equal to one, and the capillary number
defined earlier is recovered. The solution of Eq. 3 for the flow in
a wedge requires the following boundary conditions to be applied:
=0
=1
at = 0
at =
Flux = 1
Fig. 2 Definition of the local coordinate system r , , the local
tangent to the free surface = 0, and the moving substrate
=
1 =
at =
1 2
=0
r2 2
at = 0
In order to solve Eq. 3 subject to the boundary conditions

47, two simpler solutions were sought, with the final solution
being the sum of these solutions. The first of these is for the flow
in a wedge with a free surface and a stationary surface with a net
flux. For this problem the following solution is obtained:
For the case of a free surface, a moving surface, and zero net flux,
the following equation is obtained:
1
=1
r
2 cos2 sin cos

2 cos2 sin cos
2 = r
cos sin cos sin

sin cos
For more information of the solution of viscous flow in a wedge

the reader is referred to the work of Moffatt 19 or the more
recent derivation of Paggi and Carpinteri 20. The general solution of Eq. 3 subject to boundary conditions 47 is the sum
= 1 + 2due to the linear nature of the stream function equation and the corresponding addition of the boundary conditions.
The sum of the two stream functions is shown graphically in Fig.
3. From the solution of the stream function , the pressure gradient along the free surface can be obtained at the free surface
= 0.
p 2r cos3 4r sin cos2 + 2r 4 sin cos 4 + 2r sin

=
r
r3 2 cos3 sin + 22 cos2 2 + cos2 cos4
By applying this pressure gradient equation and those describing

the meniscus curvature and surface tension p = / Ca, where is
Stationary surface, = , = 1,
the meniscus curvature = d / ds at the free surface, the free

surface profile can be obtained and the problem defined as a set of
Free surface,
= 0, = 0,
1 2
r 2 2 = 0.
Free surface,
= 0, = 0,
1 2
r 2 2 = 0.
1
r
(a)
= 0.
Moving surface, = , = 0,
(b)
10
Free surface,
= 0, = 0,
1 2
r 2 2 = 0.
1
r
= 1.
Moving surface, = , = 1,
1
r
= 1.
(c)
Fig. 3 Graphical representation of the addition of 1 and 2: note the free surface stagnation point predicted by the
addition of the two stream functions 1 and 2
041205-2 / Vol. 131, APRIL 2009
dx
= cos
ds
(a)
(b)
Fig. 4 COMSOL MULTIPHYSICS finite element implementation of the
free surface problem
four first order ODEs with respect to distance along the free
surface s, one of which is Eq. 10 p / r = dp / ds, the other
three are as follows:
d
= Cap
ds
11
dh
= sin
ds
12
13
where h is the film thickness perpendicular to the substrate, and x

is the distance along the substrate from the initial condition. Equation 13 is not required in order to close the problem but provides
information on the distance downstream from the origin, allowing
the meniscus shape to be determined. In addition to the four ODEs
the identity r = h / sin is also used. The set of four ODEs were
solved using MATLABs general ODE solver ode45 and were integrated from downstream, in the final film region to an angle of
= 90 deg however different limits of can be used if
requiredas determined by the contact angle. It should also be
noted that it is possible to solve the set of equations as a BVP with
the option of using a further equation dq / ds = 0 describing the
flux if this is to be solved for. For this approach the reader is
referred to Ref. 15. The initial conditions used are characteristic
of the final film thickness with a perturbation applied to the free
surface angle = 105 rad. These initial conditions allow stable
perturbation independent solutions to be obtained. The model detailed here has the advantage over the widely used analysis of
Coyne and Elrod 9, in that the instability in obtaining their solution is not present in this analysis. As the solution is obtained
80
20
Model Free Surface

Constant radius
Pressure
FE Free Surface
70
60
15
50
10
40
30
20
0
10
0
10
20
(a)
100
80
60
40
20
10
30
(b)
25
20
15
10
x
3
6
5
2.5
1.5
1
0.5
0
0
0.5
2
3
(c)
12
10
6
x
(d)
Fig. 5 Free surface meniscus shape, pressure distribution and constant radius at = 90 deg, as obtained by the analytical
model Eqs. 1013
APRIL 2009, Vol. 131 / 041205-3
using numerical integration rather than by solving the problem as

a BVP as in the analysis of Hewson et al., the model is memory
and time efficient and allows the model to be applied to a wide
range of coating type flows.
10
ODE
Bretherton
Ruschak
CE
FEM
10
3
10
041205-4 / Vol. 131, APRIL 2009
10
(a)
10
Capillary number, Ca
10
0.9
0.8
Figure 5 shows the free surface profile for a range of different

capillary numbers; also plotted is a constant radius based on the
curvature at = 90 deg. It is interesting to note that as the capillary number decreases then as would be expected, the meniscus
shape follows the constant radius more closely and that the greater
the viscous forces, the greater the proportion of the meniscus that
is not constant in radius. For capillary numbers of Ca= 0.001 and
Ca= 0.01, the finite element and model results are almost identical
the finite element results line is obscured by the free surface line
obtained from the model. For a capillary number of Ca= 0.1 there
is still good agreement between the model and the finite element
results, with the model capturing the divergence from the constant
radius reasonably well. However, for a capillary number of Ca
= 1 the results diverge significantly.
Perhaps the results of greatest interest are those of the final film
thickness and the meniscus curvaturewhich allows the local
pressure to be determined and can be applied as a boundary condition in a range of different lubrication analyses. Here a comparison is made with the results of the Bretherton equation, Ruschaks
equation, and the numerical results of Coyne and Elrod as well as
the FEM.
Figure 6a shows how the residual film thickness varies with
capillary number. It can be seen that these results are in excellent
agreement with one another. Comparing the radius of curvature to
film thickness ratio at = 90 deg Fig. 6b shows general agreement for capillary numbers below Ca= 0.1 if a constant meniscus
radius is assumed for Ruschaks model a constant radius of curvature of unity would result due to the choice of scaling employed. It can be seen that both the model described here and that
of Coyne and Elrod both represent the free surface curvature for
low capillary numbers reasonably well. The model described here
more closely follows the finite element results for capillary numbers below Ca 0.1 however the Coyne and Elrod model is more
accurate as the capillary number increases further.
The reducing accuracy of the model for higher capillary numbers is consistent with the observation that the analysis is valid for
the location where the free surface radius of curvature is large
relative to the film thickness. This is in agreement with the analysis of Bretherton 4, however unlike the analysis of Bretherton
the divergence from a constant radius is accounted for, alluding to
10
(d/ds)/(C1)
2.1 FEM. The FEM was used to compare with the model
results. The simulation was undertaken in COMSOL MULTIPHYSICS
using the arbitrary Lagrangian Eulerian ALE method. The fluid
was modeled as Stokes flow with the free surface curvature obtained by calculating the rate of change with distance along the
free surface of the angle the free surface makes with the moving
substrate an implementation that can be achieved in COMSOL by
defining the free surface angle as an additional variable along the
boundary. The simulation used to obtain the results had
27,733DOF and solution times were of the order of 10 min; the
mesh and an example solution are shown in Fig. 4. The velocity
inlet boundary condition on the left hand side of the domain is
quadratic, consistent with the lubrication approximation. The flow
rate across this boundary was varied until the angle the free surface made with the symmetry boundary condition along the top
of the domain was 90 deg. The main results extracted from the
FEM simulations where the final film thickness and the free surface profile. To facilitate the analysis the final film thickness was
solved for with the gap h= kept fixed, this is accounted for in
Sec. 3 were the results have been appropriately scaled.
0.7
0.6
0.5
ODE
CE
FEM
0.4
3
10
(b)
10
10
Capillary number, Ca
10
Fig. 6 Results of the model, compared with Bretherton equation, Ruschak equation, and Coyne and Elrod CE results; for
a gap, h=0 at = 90 deg and b scaled meniscus radius of
curvature d / ds / h= 1
the possibility that the analysis may be able to predict the film
profile at higher capillary numberssomething already shown by
examining the free surface profile data.
Conclusions
The model detailed here provides an accurate and reasonably

efficient means of calculating the free surface shape of a fluid film
drawn from a gap. The method is more stable than that developed
by Coyne and Elrod and allows greater capillary numbers to be
investigated than the simpler relationships developed by Bretherton and Ruschak. It should be noted that there is some additional
computational cost associated with the solution, when compared
with the elegant solution of Bretherton so a compromise between
computational cost and accuracy at moderately high capillary
numbers must be borne in mind when deciding on the free surface
model to adopt. However the model outlined here provides an
extension to higher capillary numbers compared with the Bretherton equation, with the results showing that a capillary number of
Ca= 0.1 is a realistic upper limit for the model.
Acknowledgment
It is a pleasure to acknowledge Yeaw Chu Lee for his interesting discussions on Stokes flow.
References
1 Greener, J., and Middleman, S., 1979, Theoretical and Experimental Studies
2
3
4
5
6
7
8
9
10
11
of the Fluid Dynamics of a Two-Roll Coater, Ind. Eng. Chem. Fundam.,

181, pp. 3541.
Benkreira, H., Edwards, M. F., and Wilkinson, W. L., 1982, Ribbing Instability in the Roll Coating of Newtonian Fluids, Plastics and Rubber Processing
and Applications, 2, pp. 137144.
Priest, M., Dowson, D., and Taylor, C. M., 2000, Theoretical Modelling of
Cavitation in Piston Ring Lubrication, Proc. Inst. Mech. Eng., Part C: J.
Mech. Eng. Sci., 2143, pp. 435447.
Bretherton, F., 1961, The Motion of Long Bubbles in Tubes, J. Fluid Mech.,
10, pp. 166168.
Park, C. W., and Homsy, G. M., 1984, Two-Phase Displacement in Hele
Shaw Cells: Theory, J. Fluid Mech., 139, pp. 291308.
Landau, L., and Levich, B., 1942, Dragging of a Liquid by a Moving Plate,
Acta Physicochim. URSS, 17, pp. 4254.
Ruschak, K. J., 1982, Boundary-Conditions at a Liquid Air Interface in Lubrication Flows, J. Fluid Mech., 119, pp. 107120.
Thompson, H. M., Kapur, N., Gaskell, P. H., Summers, J. L., and Abbott, S. J.,
2001, A Theoretical and Experimental Investigation of Reservoir-Fed, RigidRoll Coating, Chem. Eng. Sci., 5615, pp. 46274641.
Coyne, J. C., and Elrod, H. G., 1970, Conditions for the Rupture of a Lubricating Film. Part 1: Theoretical Model, ASME J. Lubr. Technol., 92, pp.
451456.
Cameron, A., 1979, Basic Lubrication Theory, Longman, Greens, New York.
Swift, H. W., 1932, The Stability of Lubricating Films in Journal Bearings,
Proc. Inst. of Civ. Eng. UK, 233, pp. 267288.

12 Prandtl, L., 1928, Motion of Fluids With Very Little Viscosity, Technical
Memorandum of the National Advisory Committee, NACA.
13 Hopkins, M. R., 1957, Viscous Flow Between Rotating Cylinders and a Sheet
Moving Between Them, Br. J. Appl. Phys., 8, pp. 442444.
14 Savage, M. D., 1982, Mathematical-Models for Coating Processes, J. Fluid
Mech., 117, pp. 443455.
15 Hewson, R. W., Kapur, N., and Gaskell, P. H., 2005, Film Formation Model
of Shear Thinning Power Law Fluids Using Lubrication Analysis, European
Coating Symposium, Bradford, West Yorkshire.
16 Snoeijer, J. H., 2006, Free-Surface Flows With Large Slopes: Beyond Lubrication Theory, Phys. Fluids, 182, pp. 021701.
17 Cox, R. G., 1986, The Dynamics of the Spreading of Liquids on a SolidSurface. Part 1. Viscous-Flow, J. Fluid Mech., 168, pp. 169194.
18 Boender, W., Chesters, A. K., and Vanderzanden, A. J. J., 1991, An Approximate Analytical Solution of the Hydrodynamic Problem Associated With an
Advancing Liquid Gas Contact Line, Int. J. Multiphase Flow, 175, pp.
661676.
19 Moffatt, H. K., 1964, Viscous and Resistive Eddies Near a Sharp Corner, J.
Fluid Mech., 181, pp. 118.
20 Paggi, M., and Carpinteri, A., 2008, On the Stress Singularities at Multimaterial Interfaces and Related Analogies With Fluid Dynamics and Diffusion,
Appl. Mech. Rev., 612, pp. 020801.
APRIL 2009, Vol. 131 / 041205-5
Wodzimierz Wrblewski
e-mail: wlodzimierz.wroblewski@polsl.pl
Sawomir Dykas
Institute of Power Engineering and
Turbomachinery,
Silesian University of Technology,
Gliwice, Poland
Andrzej Gardzilewicz
Institute of Fluid-Flow Machinery,
Polish Academy of Science,
Gdansk, Poland
e-mail: gar@imp.gda.pl
Michal Kolovratnik
Czech Technical University in Prague (CTU),
Prague, Czech Republic
e-mail: kolovrat@fsid.cvut.cz
Numerical and Experimental

Investigations of Steam
Condensation in LP Part of a
Large Power Turbine
This paper presents the experimental investigations of steam flow with condensation in
the blading system of the low-pressure (LP) part of a 360 MW turbine. To this end,
special probes were used, which provided flow visualization opportunities including localization of the front of condensation, determining distributions of pressure, temperature, velocity, and flow angle in the inter-row gaps, measurements of water droplet concentration and sizes. The measurements have proved that the condensation process in the
LP turbine might be of heterogeneous nature, depending on the concentration of chemical
impurities in steam. The measurement results constituted the basis for computational fluid
dynamics (CFD) flow calculations, which were performed using the time-dependent 3D
Reynolds averaged NavierStokes equations coupled with two-equation turbulence model
(k- SST) and additional conservation equations for the liquid phase. The set of governing equations has been closed by a local real gas equation of state. The condensation
phenomena were modeled on the basis of the classical nucleation theory. The heterogeneous condensation model on the insoluble and soluble impurities was implemented into
presented CFD code. The system of governing equations was solved by means of a finite
volume method on a multiblock structured grid. The obtained numerical results and
experimental data were compared and discussed. DOI: 10.1115/1.3089544
Introduction
The flow of steam through the blades of a low-pressure LP

turbine results in condensation that unfavorably influences efficiency of power generation. The homogeneous condensation in
steam flow occurs when there are no foreign nuclei. In real conditions, for steam power cycle, the steam always contains some
impurities of both insoluble and soluble characters. Therefore, in
real turbine, the wetness appears as a result of a mixed
homogeneous/heterogeneous condensation or pure heterogeneous
condensation, everything depends on the quantity of foreign nuclei.
The lack of experimental data for the steam flow with heterogeneous condensation and new data for the flow with pure homogeneous condensation limit the progress in numerical modeling of
the wet steam flows. It is very difficult to reproduce the real conditions in large output turbine in experiment; it refers to the flow
conditions and steam quality, the number, and the sort of the impurities. The impurities may change the mechanism of the liquid
phase generation and may intensify the corrosion and erosion processes. Many operational experiences in the power plants e.g.,
Refs. 1,2 and authors own observations confirm that facts.
In the feedwater cycle of power plant, different sources of impurities may be identified: mechanical, chemical, or electrochemical ones. Particles that exist in steam are soluble and insoluble
compounds salts and alkalis. For the parameter range occurring
in water-steam cycle, these impurities can be divided into groups
with and without phase change. In the first group, sodium chloride
NaCl is the most important one, taking into consideration condensation process, which dissolves just above the saturation line
of pure steam. As regards condensation on soluble particles, it is
JOURNAL OF FLUIDS ENGINEERING. Manuscript received October 30, 2007; final manuscript received October 6, 2008; published online March 9, 2009. Assoc. Editor:
Chunill Hah.
necessary to take into account the size change of the particles and
the solution concentration in the droplets. The condensation model
requires knowledge of many physical and chemical properties of
the salt solution in steam. It is often very hard to find enough
information in the references, as well as from the feedwater analysis in the power plant.
Two processes have to be considered during steam expansion in
the turbine: homogeneous and heterogeneous condensations.
Steam impurities, necessary to initiate heterogeneous condensation, occur in the form of suspension of particles, which are insoluble or soluble in water. The chemical and physical properties
of a suspension influence the size of the droplets nucleated on the
particles and the nucleation process.
It is possible to model various types of condensation processes,
i.e., homogeneous and heterogeneous ones on soluble and insoluble particles. The necessary step in the nonequilibrium wet
steam flow modeling is the validation of the numerical models
against the experimental results. The development of numerical
methods for wet steam flow calculations has to allow for real gas
properties, which, in fact, is the fundamental one for correct calculations of the condensation process. The correct calculation of
the steam flow for low pressures close to the saturation line and
for high pressures needs to implement the real gas equation of
state into numerical algorithm. The relation between thermal parameters of state is nonlinear within these ranges and does not
follow the ideal equation of state. Moreover, in the case of wet
steam, a two-phase flow model has to be considered.
The developed steam condensing flow models have been validated against experimental data for simple 2D test cases. Usually,
they are the Laval nozzles in which 1D flow form dominates. All
available experimental investigations relate to the homogeneous
condensation, whereas in real turbine the heterogeneous condensation is predominant. Therefore, validation of the numerical algorithm for modeling the flow in real turbine stages by means of
such test cases is not easy; it has been discussed in Refs. 3,4.
The comparison of the numerical modeling and measurements
APRIL 2009, Vol. 131 / 041301-1
of the flow in real turbine of the large output is not often published. The crucial value of such verification comes down to the
assessment of flow conditions for real LP steam turbine stages by
means of the CFD code.
This paper presents the results of experimental and numerical
investigations of 3D flows through the last two stages of the LP
part of a 360 MW turbine. The problems of condensation in a LP
steam turbine has been presented and discussed also, e.g., by
Hesler et al. 5, Bakhtar and Heaton 6, and Petr et al. 7.
Measurements of thermodynamic parameters, the size, and concentration of water droplets together with CFD results have been
described below. The presented work was performed by research
teams of the Institute of Fluid-Flow Machinery from Gdansk PL,
the Czech Technical University CTU of Prague, and the Silesian
University of Technology PL within the framework of the Research Project No. 3T10B04726 sponsored by the State Committee for Scientific Research in Poland.
Physical and Mathematical Model
2.1 Governing Equations. All results presented in this paper

were obtained by means of the in-house CFD code. The code is
based on the time-dependent 3D Reynolds averaged Navier
Stokes RANS equations formulated for the vapor/water mixture,
which are coupled with a two-equation turbulence model k-
SST model, and with additional mass conservation equations for
the liquid phase: two for homogeneous condensation,
y hom y homu y homv y homw

+
+
+
t
x
y
z
4
2 drhom
= l r3Jhom + 3nhomrhom
3
dt
nhom nhomu nhomv nhomw

+
+
+
= Jhom
t
x
y
z
and one for heterogeneous condensation:
y het y hetu y hetv y hetw

2 drhet
+
+
+
= 4lnhetrhet
t
x
y
z
dt
2
The droplet radius rhet in the heterogeneous condensation is calculated from the relation
rhet = r3p +
3 y het
4 lnhet
1/3
h = hv1 y + h1y
where AT and BT are defined as

AT = a0 + a1T + a2T2
BT = b0 + b1T + b2T2
The coefficients ai , bii = 0 , . . . , 2 of the polynomials AT and
BT are the functions of temperature only, and they can be found
from an approximation of thermodynamic properties of steam calculated using IAPWS-IF97 9. It is easily to notice that this
equation is represented by a relatively uncomplicated approximate
surface.
The equation of state 5 is called the local real gas equation
of state because, due to the simple mathematical form, it can be
applied locally within the limited parameter range only.
Expressions for the specific enthalpy, specific entropy, and
other properties such as specific heat capacities and speed of
sound can be determined from the standard thermodynamic
relations.
The concept of the local real gas equation of state can be applied to determine the thermodynamic properties of any real gas,
in the case when the simple mathematical form is needed.
2.3 Nucleation and Droplet Growth Model. The homogeneous condensation phenomenon is modeled on the ground of the
classical nucleation theory of Volmer, Frenkel 10, and Zeldovich
and continuous droplet growth model of Gyarmathy 11.
The nucleation rate, i.e., number of supercritical droplets produced per unit mass of vapor per unit time, is calculated from the
relation derived by the assumption of thermodynamic equilibrium
between critical droplets and vapor. Additionally, the nonisothermal correction factor C proposed by Kantrowitz 12 has been
used because the isothermal model assumption does not apply to
the vapor:
Jhom = C
1/2
mv3/2
v2
4r2
exp
3kTv
l
where coefficient C is calculated from the relation
C= 1+2
1 hv hl hv hl 1
2
+ 1 RTv
RTv
and mv is the mass of a water molecule and is the correction

factor in the calculations assumed= 1.
The radius of critical clusters r for the applied real gas EOS 5
has a form that differs from the known relation for ideal gas:
r =
2
lfpv fps pv ps
where
4
= v/1 y
The value of the nonequilibrium wetness fraction y is calculated
as the sum of the homogeneous and heterogeneous wetness obtained from the conservation equations for the liquid phase 1 and
2.
041301-2 / Vol. 131, APRIL 2009
p
BT
= ZT, = AT +
RT
where r p is the radius of the particle on which the droplet is

growing.
For the two-phase nonequilibrium flow, it has been assumed
that the volume occupied by droplets is negligibly small. The
interaction between droplets is not taken into account in the
model. The heat exchange between the liquid phase and the solid
boundary, and the velocity slip between vapor and the liquid
phase are neglected. The nonslip assumption is much less significant than the effects of the thermal nonequilibrium 8. The conservation equations are formulated for the two-phase mixture with
the specific parameters calculated with the following relations:
s = sv1 y + s1y
2.2 Equation of State. For wet steam flow calculations, the

set of ten governing equations has to be closed by a real gas
equation of state EOS. The formulation of the real gas EOS is
compound even in case when simplified model has been used. The
idea of applied real gas EOS consisted in creation of equation of
state with as simple mathematical form as possible, but simultaneously very accurate. The simple mathematical form can be very
accurate, but only within the small parameters range.
The mathematical form of used real gas EOS is similar to the
virial equation of state with one virial coefficient.
b
b
fp = ln p + c ln
2
2
c
b
c
1
b
1+
b = ATRT
Fig. 1 Location of probes in the blade system of a 360 MW turbine LP part of Belchatow
Power Plant: meridional section a, cross section b, and photo of measurement stand
c
c = ATRT2 + 4pBTRT1/2
Equation 7 has an ideal gas form since A 1 and B 0 .
The further behavior of the critical droplets can be described by
suitable droplet growth law. The size of the droplets for the vapor
under low pressure is much smaller than mean free path of the
vapor molecules. Therefore, the growth of the droplets should be
governed by considering molecular and macroscopic transport
processes HertzKnudsen model. Difficulties to choose the condensation and accommodation coefficients make the application
of the HertzKnudsen model very difficult to calculate. This problem can be avoided by using Gyarmathys droplet growth model,
which takes into account diffusion of vapor molecules through the
surrounding vapor, as well as heat and mass transfer, and the
influence of the capillarity:
r r Ts Tv
v
dr 1
2
=
dt l 1 + 3.18 Kn
r hv hl
In the case of heterogeneous condensation model, the nucleation

process is neglected. The droplet growth on the particle impurities, which are assumed to be spherical with the given initial mean
radius and concentration, is modeled according to the same droplet growth law 8.
The model of heterogeneous condensation on soluble particles
is based on the work of Gorbunov and Hamilton 13. In the
model, the influence of the NaCl salt is taken into account.
Knowledge of specific physical and chemical parameters of the
salt solution in steam is necessary for process definition. This
information is usually inaccessible due to the lack of experimental
research in this field. Physical model includes supplementary parameters, such as molality of salt solution M l and the activity of
water aw:
Ml =
1000ns
nwM ww
aw = 1
aM,
i
i
l
M l M l max
i=1
aw = 1
aM
i
i
l max,
i=1
M l M l max
10
where M ww is the molar mass of water and for NaCl is M l max

= 6.109 mol/ kg.
The
coefficients
are
a1 = 3.2975479
a2 = 2.081262 104 kg2 / mol2,
and
a3
102 kg/ mol,
= 1.85425 104 kg3 / mol3. The activity of water is used to
specify the saturated pressure for the aqueous solution necessary
in Eqs. 7 and 8.
The surface tension for a water solution of soluble particles is
calculated from the relation according to Gorbunov and Hamilton
13:
= oT + BM l
11
where the value of surface tension 0 is for M l = 0 mol/ kg pure

steam. Constant B depends on the solution for NaCl it equals
1.62 103 N kg/ m mol.
Numerical Method
The numerical calculations were carried out using the in-house

CFD code TRACOFLOW. The system of governing equations was
solved on a multiblock structured grid with the use of the finite
volume method integrated in time with the explicit RungeKutta
method. In time integration, the fractional step method was used
to split the equations into an adiabatic and a diabatic part in order
to apply different time steps for the flow and condensation calculations. The Monotone Upstream-centered Schemes for Conservation Laws MUSCL technique was implemented to get a third
order accuracy in space. For modeling the stator-rotor interaction,
a mixed-out technique was used. A detailed description of the
numerical method can be found in Refs. 14,15, where the validation of the CFD code against laboratory tests has been presented.
Experimental Investigations in the LP Turbine
4.1 Thermal and Flow Measurements. The instrumentation

installed at the 18 K360 turbine and used for thermal and flow
measurements is shown in Fig. 1.
Figure 1 shows the location of special measuring probes in the
spaces between the last stage rows. These probes allowed a measuring pressure and temperature distributions, and evaluating velocity magnitudes and angles; b visualizing the flow and localizing the beginning of condensation; and c taking samples of
APRIL 2009, Vol. 131 / 041301-3
Fig. 2 Plate probe for thermal and flow measurements a including the optical system b
early condensate for chemical analyses.

The probes can be mounted during nonstop operation into special seats prepared during an overhaul of the turbine. The measuring probe, presented in Fig. 2, can be classified as a disk probe.
This type is not sensitive to water films that form on it 16. The
probe tip consists of four five Pitot tubes inclined toward the
flow at four various pitch angles every 20 deg from 20 up to
+40 60. It allows us to determine the streamline angles in meridional plane. It is also possible to measure the total pressure,
which is taken as the highest value of all four readings. Two holes
have been made on the side surface of disk to determine the yaw
angle. The holes are also used to determine the static pressure.
The whole tip of probe the head and the tubes is lined inside and
covered outside with Teflon to avoid watering of the probe and to
assure effective blowthrough and extraction of deposited water.
Smooth regular flowthrough is assured at joints of tubes of different diameters. Owing to this, the probe disk was miniaturized to a
diameter of less 20 mm. The diameter of the probe tubes is less
than 1 mm. In order to perform the blowthrough of the probe, a
cutoff valve has been installed, which links all tubes with atmosphere. When the blowthrough is switched off and the measurement starts, the complete leak tightness is secured. It is vital that
the probe is placed properly with respect to the flow. Therefore,
turbine geometry is studied and the probe is aligned carefully
prior to the measurement. Visualization is based on a special telescope. The probe is equipped with an optical system adjusted for
microphotography purposes, see Fig. 2.
1000 W halogen lighting is a part of the system to enable the
flow visualization. TV camera or an Olympus system can be utilized here. A special wiper cleans the prismatic eyepiece if watered. Before each measurement, the measuring probes were calibrated at the laboratory rigs. This, along with the stabilization of
the genset operating parameters, provided a high accuracy of the
measurements, in which temperature was assessed with an uncertainty equal to 0.5 K, and pressurewith that equal to 0.5%.
The uncertainty in evaluating flow inclination angles was equal to
2 deg in the circumferential plane and 10 deg in the meridional plane. The measuring procedure was described in detail by
Gardzilewicz and Marcinkowski 17.
The investigations were done for two characteristic loads of
300 MW and 360 MW.
041301-4 / Vol. 131, APRIL 2009
Fig. 3 Measurement radial parameters at the inlet fourth stage

a and behind the last stage b. Circumferential static pressure at the root stage behind the turbine c.
During thermal and flow measurements, sections between the

three last stages were radially traversed probes 13, as shown in
Fig. 1. These measurements were completed by measurements in
circumferential direction behind last stage probes 4 and 5.
The unsymmetrical flow behind the turbine checked additionally by static pressure measurements at six points of circumference distributed each 60 deg.
A complete set of the measured results can be found in Ref.
18.
Selected results of these measurements, which were a database
for numerical calculations, are shown in Fig. 3 concerned power
of 360 MW. They include all measurement parameters at the inlet
to the fourth stage and distribution of static pressure along the
radial and circumferential directions behind last stage. Other necessary data from the complete set are presented in the next paragraph in order to compare them with calculation results.
It is noteworthy that the thermal measurements made it possible
to determine the course of the expansion line in the wet steam
area. Average enthalpy values required for this purpose were obtained from additional heat balance calculations and mass flow
rate known from experiment 17. The shape of the expansion line
for one of the reported experimental case made with the liquid
phase measurements 360 MW, case 2 is shown in Fig. 4. In this
figure, an area is marked in which the condensation begins. This
bly be explained by the heterogeneous nature of the steam condensation process at the presence of other chemical compounds.
The presence of chemical inclusions was confirmed by pH conductivity tests and chemical analyses of the first condensate and
condenser water samples. This is presented in Table 1. Noteworthy are high numbers of Na and Cl ions, which, according to
investigations by Steltz et al. 19 and Svoboda et al. 20, accelerate the condensation process.
Fig. 4 Expansion lines in LP part of a 360 MW turbine
area, detected by the optical system of the probe, is situated behind the third stage, below the saturation line, which could possi kg s1 case 1
Table 1 Calculated mass flow rates m
Adiabatic
Homogeneous
condensation
Homo-/heterogeneous
condensation
Heterogeneous
condensation
103.5
102.2
102.1
101.8
4.2 Measurement of Liquid Phase Characteristics. Measuring the liquid-phase structure in the LP part was a special part of
360 MW turbine tests. The optical measurements require a special
measuring device. Therefore, the CTU team was invited, who has
long-time experience with this problem. Measurements of the
droplet size distribution and steam wetness were carried out by an
extinction probe, developed at CTU 7. The probe measures the
multispectral extinction of light, i.e., attenuation I / Io, caused by
the light scattering on droplets, at the series of light wavelengths.
The universal measuring system with an extinction probe was
used for measuring the multispectral light extinction. The probe
measures the transmittance I / Ioi, caused by the light scattering
on droplets, of the light beam transmitted through the wet steam at
the series of light wavelengths ii = 1 , . . .. The layout of the measuring system is shown in Fig. 5a, while the measuring head of
the extinction probe with the 25 mm diameter is presented in Fig.
5b.
The distribution function FD of the droplet sizes, together
with the wetness fraction y, is then found by means of computer
analysis using Mies theory of light scattering on droplets of the
measured transmittance data I / Io. In principal, the problem leads
to the solution of the system of the first kind Fredholm integral
equations.
The structure of fine droplets at the measurement site was
evaluated from the measurement data I / Io by means of the CTU
inversion method using the well-known exponential law:
1
Io
ln
l
I
Dmax
ED/ i,mFDD2dD
12
Dmin
where i means the discrete wavelength.

The Sauter mean droplet diameter and volume concentration of
droplets could be evaluated then from Eqs. 13 and 14, respectively.
Fig. 5 Measuring system of CTU extinction probe a and head of extinction probe b
APRIL 2009, Vol. 131 / 041301-5
Numerical Results
The presented experimental data were the basis for the CFD
calculations and the validation tests. A numerical analysis was
performed for the geometry of the last two stages of the five-stage
LP cylinder of a 360 MW steam turbine 3000 rpm. The analyzed
last two stages operate in the conditions under saturation line.
Two different measurements were performed in the power plant
for the loads 300 MW case 1 and 360 MW case 2. In case 1,
only flow parameters were measured. In case 2, the liquid phase
parameters were included in the experimental study.
Fig. 6 Variation in transmittance measured at the blade midspan behind the L-0, L-1, and L-2 stages
D32 =
Dmax
FDD3dD
Dmin
13
Dmax
FDD2dD
Dmin
n=
Dmax
14
FDdD
Dmin
For measured values of static pressure p at the measuring levels,

the corresponding steam wetness was determined according to the
equation
y=
l
1+
6 v
1
Dmax
3
FDD dD
Dmin
15
where v and l are assumed as saturation parameters.

Measuring the multispectral extinction was carried out at the
exit of the last three stages of the LP part of the steam turbine
always in several traverse positions along the blade height. Let us
denote the fifth the last, fourth, and the third stages of turbine by
L-0, L-1, and L-2, respectively.
The measuring slot of the probe was oriented according to the
outlet yaw angles measured by IF-FM Gdansk and checked using
I / Iomin of the optical extinction probe.
The primary result of the extinction measurements in the turbine is the dependence of the transmittance I / Io on the light wavelength for each measuring position. Figure 6 shows an example
of the measured variation I / Io = f at the blade midspan behind
the L-0, L-1, and L-2 stages.
As it can be seen from Fig. 6 L-2 under nominal operating
conditions of the turbine, it is probable that nucleation might start
at the L-2 stage. It is to be emphasized that the liquid phase at L-2
was observed at the blade midspan only. In addition, the occurrence of the liquid phase was not observed in repeated measurements. It confirms that the onset of the nucleation process can
move along the LP flow path in accordance with the possible
operation variation in the turbine inlet steam parameters. Behind
the L-1 and L-0 stages, liquid phase was observed in all measuring points along the blade height.
According to Eqs. 1215, the liquid phase parameters are
calculated. The Sauter mean diameter D32 of droplets was evaluated according to Eq. 13 from measured transmittance data. For
evaluation of the corresponding steam wetness, according to Eq.
15, the values of static pressure measured by the IF-FM Gdansk
were used. The obtained variation in steam wetness fine droplets
only along the blade height with traverse coordinate is also
shown in Fig. 12. The uncertainty in estimation of the steam wetness and diameter D32 values is about 10%.
041301-6 / Vol. 131, APRIL 2009
5.1 Test Case 1: Without Liquid Phase Measurements. The

measurements in this case were made without a liquid phase
probe, and only thermal and flow parameters could be compared.
The turbine operated with a lower load and high condenser back
pressure. Boundary conditions at the inlet were taken from the
measurements and were assumed as constant along the blade
length. The total temperature was t0 = 103 C, and the total pressure p0 = 87 kPa. The flow angles were assumed as the linear
distribution in the radial direction: angle changed from 62.4 deg
to 80.4 deg, and from 6 deg to 36 deg. This distribution was
taken as a linear approximation of the measurement values to
eliminate high changes between the measured values especially in
the tip region.
Calculations were performed using the adiabatic and diabatic
flow models. In the diabatic flow model, the liquid phase occurs
due to the condensation process. The case called adiabatic flow
model takes only into account gas phase, and steam under saturation line is assumed as the supercooled steam. This is a nonphysical flow model without phase change and steam parameters are
determined by local real gas equation of state. The solution obtained from adiabatic flow model is used as initial condition for
diabatic flow model calculations.
Due to the lack of information regarding the steam quality, the
type of the condensation process was unknown. Theoretically,
three types of flows are available: with pure homogeneous condensation when the steam is clean, with homoheterogeneous condensation for small quantity of steam, and finally with heterogeneous one when the condensation on the impurities is
predominant. In all cases, the radius of impurities was r p
= 108 m and the concentrations at the inlet were nhet
= 1013 kg1 and nhet = 1016 kg1. Impurities regarded as insoluble
were taken into considerations in all calculations. There were simplifications in the geometry of the flow path. The bleed between
stages was not taken into account and the tip clearances were
neglected.
The calculated static pressure and flow angle distributions along
the blade were compared with measurements in Fig. 7. The static
pressure in the gap between stages calculated for condensation
models is higher than in the experiment, but the discrepancy is not
significant. The flow angle distribution obtained using the heterogeneous condensation model is similar to the experimental results.
The flow angles calculated with the homogeneous and heterohomogeneous models are in this case about 10 deg higher. At the
outlet, the static pressure distributions in all cases are similar but
in outlet swirl angle only for the diabatic one are relatively close
to the experimental data. In the measurements at the outlet, the
high influence of the leakage flow at the tip is visible and this is
the reason of discrepancies in this region. It should be admitted
that the static pressure as the boundary condition at the outlet was
assumed for calculations in the cross section located downstream
from the measurement plane. It was necessary due to the too short
distance between the trailing edge and the probe, not acceptable
for the calculations.
The wetness fraction plots in the midspan for different condensation models are presented in Fig. 8. In this figure, the condensation onset can be easily recognized.
In the case of pure heterogeneous condensation, the liquid
phase in the steam flow occurs earlierin the end part of the
Fig. 7 Static pressure and flow angle distributions: a between the last two stages and b at the outlet
stator of penultimate stage Fig. 8. The liquid phase appears not

as rapid as in the homogeneous case. It is the reason that the
thermodynamic losses are smaller. Figure 9 presents the radial
distribution of circumferentially averaged wetness fraction between the stages and at the outlet. At the outlet, the wetness fraction is smaller for the homogeneous case but between the stages
the values of wetness fraction are similar for homogeneous and
mixed homo-/heterogeneous condensation models.
The mass flow rates were calculated in all cases and gathered in
Table 1. The results obtained using the condensation models are
similar and less than in the adiabatic flow model and also are less
about 5% than the value estimated from the measurement data at
the inlet to the penultimate stage.
This test case proves that the assumption of impurities gives
better results. The fact that the condensation starts earlier in the
turbine than in the case with homogeneous model only was confirmed visually during the experiments. This tendency was observed in calculations only for pure heterogeneous condensation
Figs. 8 and 9.
5.2 Test Case 2: With Liquid Phase Measurements. The
boundary conditions at the inlet and outlet for TRACOFLOW CFD
code were determined from the measurements for a 360 MW load,
assuming that the inlet steam was superheated. For these calculations, the presence of the chemically soluble impurities was assumed. Based on the chemical analyses of the condensate samples
in the condenser Table 2, the NaCl salt concentration was assumed to be close to 2 ppb parts per billion. It corresponds to
particle concentration nhet = 2 1014 kg1, and particle radius of 2

nm when clusters with 100 molecules are assumed. The similar
structure of the clusters was also considered in Ref. 21. The
static outlet pressure of pout = 14.3 kPa was assumed at the midspan, as the circumferentially averaged value. The outlet conditions were assumed in the plane located downstream from the
measurement plane.
Similar to the previous case, the rotor tip clearance was not
modeled. The liquid phase separation in the stages was not modeled as well.
The computational domains for RANS calculations were discretized by means of the structural multiblock grid eight blocks
for each blade row. The total number of grid nodes was above
600,000. The Reynolds number calculated with the use of the total
parameters at the inlet, the inlet speed of sound, and blade chord
was Re= 1.7 106.
The spanwise distributions of parameters were calculated on the
basis of circumferentially mass averaged quantities at the inlet, in
the gap between stages and at the outlet. For assumed concentration of salt impurities, the heterogeneous condensation was predominant in the flow. The calculated mass flow rate was 102 kg/s,
that is, about 3% less than that calculated from measured data at
the turbine outlet. The main cause of this discrepancy could be the
circumferential outlet pressure variation, which was not taken into
account in the steady-state calculations.
The distributions of the absolute angle and static pressure between the stages and at the stage outlet are presented in Fig. 10.
APRIL 2009, Vol. 131 / 041301-7
Fig. 8 Wetness fraction at the midspan section: a homogeneous, b homoheterogeneous, and c heterogeneous condensations
The numerical results are compared with the experimental data.

The calculated flow angles at the outlet coincide relatively well
with the measurement data Fig. 10b. The comparison of calculated parameters with experiment between the stages is shown
in Fig. 10a. The absolute outlet angle is within the range between 60 deg and 90 deg without taking into account endwall
boundary layers. The maximum discrepancy between the calcu-
lated and measured values is twice as big as the measurement

error. The calculated distribution of static pressure is consistent
with the experiment.
To show the wet steam formation, the blade-to-blade cross sections of stages at the midspan are presented in Fig. 11. For the
assumed soluble impurities in the steam, the condensation process
shows homo- and heterogeneous patterns, but the second one is
Fig. 9 Wetness fraction distributions: a between the last two stages and b at the outlet
Table 2 Results of chemical analyses of condensate samples
Notation
Unit
Condensate behind the LP part
First condensate
041301-8 / Vol. 131, APRIL 2009
pH
Na+
K+
NH+4
Cl
SO4
SiO2
Fe
88.4
6.86.9
g / kg
23
180
g / kg
5
40
g / kg
60
g / kg
2
130
g / kg
12
90
g / kg
23
900
g / kg
78
1200
Fig. 10 Flow angle and static pressure distribution: a between the last two stages and b at the outlet
predominant. The liquid phase generated in the flow forms in

6570% on the salt particles. The beginning of the condensation
process is in the stator row of the first calculated stage Fig.

12a. The proper conditions for the homogeneous condensation
Fig. 11 Wetness fraction contours at midspan section: global wetness fraction contours homogeneous and heterogeneous a and wetness fraction
contours generated only due to the heterogeneous condensation b
APRIL 2009, Vol. 131 / 041301-9
Fig. 12 Wetness fraction and mean droplet Sauter diameter: a between the last two stages and b at the outlet
are in the rotor row of the first stage as can be deduced from the
wetness contours in Figs. 11a and 11b. The contours are shown
for the global wetness fraction obtained due to the homogeneous
and heterogeneous condensation processes and for the wetness
fraction generated only on the soluble particles.
The spanwise distributions of the wetness fractions and droplet
radii in the gap between stages and at the stage outlet are shown
together with experimental data in Fig. 12. The relatively good
agreement in the wetness fraction distributions is observed below
the midsection at the outlet. The calculated wetness fraction between the stages is about 30% smaller than that estimated from
the experimental data. This difference is high in the midpart of the
blade and decreases in the tip sections. The discrepancies at the
outlet are smaller than those between the stages. This can be explained by fact that at the outlet the equilibrium in the wet steam
was achieved. The simplifications in condensation models and the
accuracy of the real-scale measurements bear on the discrepancies
in the flow parameters between the calculations and experiment.
In Fig. 12, the spanwise distributions of the Sauter mean diameters are presented. From calculations, the volume averaged diameter is obtained; therefore, the Sauter diameter was calculated on
the basis of droplet size spectrum known from experiment.
The calculated droplet radii show a more even distribution and
are confined within the same range in the both sections.
041301-10 / Vol. 131, APRIL 2009
Conclusions
The reported investigation of steam flow through the LP part of

a 360 MW turbine suggests that the heterogeneous condensation
might be a dominating process for the considered concentration of
impurities NaCl. This could be the reason why supercooling is
not observed in the turbine, and the condensation starts practically
close to the saturation line. The optical extinction test reveals that
the water droplets recorded in the turbine flow can be as large as
300500 nm in diameter. The comparison of the flow field parameters obtained by means of TRACOFLOW in-house CFD code and
from experimental data shows relatively good correspondence and
confirms proper modeling of the real gas properties and condensation processes in the steam flowing through the stages.
The applied CFD tool has been before thoroughly validated
against laboratory experiments. It has been possible to validate the
mathematical model under conditions present in the real turbine as
well.
It should be emphasized that it was the first step in the comprehensive analysis of the condensation phenomena using numerical
and experimental research made for the working steam turbine of
large output.
Acknowledgment
The authors would like to thank the Polish Ministry of Science
and Information Technology for the financial support of the Research Project No. PB 1431/T10/2004/26 and the management of
the Belchatow Power Plant for the consent for onsite measurements.
Nomenclature
aw
D
D32
E
F
h
Io
I
J
k
l
L
m
m
Ml
Mw
n
p
R
r
r
s
T
u,v,w
y
Z
x,y,z
activity of water in solution

droplet diameter, m
Sauter mean droplet diameter, m
Mie extinction function
distribution function of diameter D, m3
specific enthalpy, J kg1
incident light intensity no droplets
light intensity
nucleation rate, m3 s1
Boltzmann constant, J K1
light path length in wet steam, m
blade height, m
relative refraction index, mass, kg
mass flow rate, kg s1
molality, mol kg1
molal mass, kg kmol1
number of droplets, kg1
pressure, Pa
individual gas constant, J kg1 K1
radius, m
critical radius, m
specific entropy, J kg1 K1
temperature, K
velocity components in Cartesian coordinates,
m s1
wetness fraction, kg kg1
compressibility coefficient
Cartesian coordinates, m
inclination angle, deg
circumferential angle, deg
wavelength of light, m, thermal conductivity,
W m1 K1
density, density of the mixture, kg m3
isentropic exponent
specific volume, m3 kg1
surface tension, N m1
0
hom
het
v
total parameter
homogeneous
heterogeneous
vapor phase
Subscripts
l liquid phase
s saturation value
w water
References
1 Keller, H., 1974, Erosionkorrosion an Nassdampfturbinen, VGB Kraftwerktechnik, 54, pp. 292295.
2 Jonas, O., 1985, Steam Turbine Corrosion, Mat. Perf., 242, pp. 918.
3 Bakhtar, F., White, A. J., and Mashmoushy, H., 2005, Theoretical Treatments
of Two-Dimensional, Two-Phase Flows of Steam and Comparison With Cascade Measurements, Proc. Inst. Mech. Eng., Part C: J. Mech. Eng. Sci., 219,
pp. 13351355.
4 Wrblewski, W., Dykas, S., and Gepert-Niedobecka, A., 2007, Condensing
Steam Flow ComputationsValidation Problems, Seventh European Conference on Turbomachinery, Conference Proceedings, Athens, K. D. Papailiou, F.
Martelli, and M. Manna, eds., pp. 841848.
5 Hesler, S., Shuster, L., and McCloskey, T., 1998, Optical Probe of Measurement of Steam Wetness Friction in LP Turbine, IPJC II, ASME, New York,
PWR-Vol. 33.
6 Bakhtar, F., and Heaton, A. V., 2005, Effects of Wake Chopping on Droplet
Sizes in Steam Turbines, J. Mech. Eng. Sci., 219C12, pp. 13571367.
7 Petr, V., Kolovratnk, M., and Hanzal, V., 2003, Instrumentation and Tests on
Droplet Nucleation in LP Steam Turbines, Power Plant Chemistry, 57, pp.
389395.
8 White, A. J., Young, J. B., and Walters, P. T., 1996, Experimental Validation
of Condensing Flow Theory for a Stationary Cascade of Steam Turbine
Blade, Philos. Trans. R. Soc. London, Ser. A, 354, pp. 5988.
9 Wagner, W., Cooper, J. R., Dittmann, A., Kijima, J., Kretzschmarr, H.-J.,
Kruse, A., Mare, R., Oguchi, K., Sato, H., Stcker, I., ifner, O., Takaishi, Y.,
Tanishita, I., Trbenbach, J., and Willkommen, Th., 2000, The IAPWS Industrial Formulation 1997 for the Thermodynamic Properties of Water and
Steam, Trans. ASME: J. Eng. Gas Turbines Power, 1221, pp. 150181.
10 Frenkel, J., 1946, Kinetic Theory of Liquids, Oxford University Press, New
York.
11 Gyarmathy, G., 1960, Grundlagen einer Theorie der Nassdampfturbine,
Ph.D. thesis, ETH, Zrich.
12 Kantrowitz, A., 1951, Nucleation in Very Rapid Vapor Expansions, J. Chem.
Phys., 19, pp. 10971100.
13 Gorbunov, B., and Hamilton, R., 1997, Water Nucleation on Aerosol Particles
Containing Both Soluble and Insoluble Substances, J. Aerosol Sci., 282, pp.
239248.
14 Dykas, S., 2001, Numerical Calculation of the Steam Condensing Flow,
Task Quarterly, Scientific Bulletin of Academic Computer Centre in Gdansk,
54, pp. 495519.
15 Wrblewski, W., Dykas, S., and Gepert, A., 2006, Modeling of Steam Flow
With Heterogeneous Condensation, Monografia 95, Wydawnictwo Politechniki Slskiej, Gliwice, in Polish.
16 Lagun, V. P., Simoyu, L. L., and Frumin, Yu. Z., 1975, Full-Scale Tests of the
Exhaust Hoods of a High-Capacity Steam Turbine, Teploenergetika Moscow,
Russ. Fed., 222, pp. 3135.
17 Gardzilewicz, A., and Marcinkowski, S., 1995, Diagnosis of LP Steam Turbine Prospects of Measuring Techniques, Proceedings of the ASME Joint
Power Generation Conference, Vol. 3, pp. 349358.
18 Gardzilewicz, A., Marcinkowski, S., Karwacki, J., and Kurant, B., 2006, Results of Measurements of Condensing Steam Flow in LP Part of Turbine 18 K
in Belchatow Power Plant, IF-FM Report Nos. 5681 and 6157.
19 Steltz, W., Lindsay, P., and Lee, W., 1983, The Verification of Concentrated
Impurities in Low Pressure Steam Turbine, ASME Winter Meeting, WA.
20 Svododa, R., Phlug, H.-D., and Warnecke, T., 2003,Investigation into the
Composition of the Early Condensate in Steam Turbines, Power Plant Chemistry GmbH, Neulusshei.
21 Petr, V., and Kolovratnik, M., 2001, Heterogeneous Effects in the Droplet
Nucleation Process in LP Steam Turbines, Proceedings of the Fourth European Conference on Turbomachinery, Florence, Italy, pp. 783792.
APRIL 2009, Vol. 131 / 041301-11
M. Kawaji1
Department of Chemical Engineering and Applied
Chemistry,
University of Toronto,
200 College Street,
Toronto, ON, M5S 3E5, Canada
K. Mori
Osaka Electro-Communication University,
18-8 Hatsu-cho,
Neyagawa, Osaka, Japan 572-8530
D. Bolintineanu
Department of Chemical Engineering and Applied
Chemistry,
University of Toronto,
200 College Street,
Toronto, ON, M5S 3E5, Canada
The Effects of Inlet Geometry and

Gas-Liquid Mixing on Two-Phase
Flow in Microchannels
The effects of gas-liquid inlet geometry and mixing method on adiabatic gas-liquid twophase flow in a microchannel of 100 m diameter have been investigated using a
T-junction inlet with the same internal diameter as the microchannel. Two-phase flow
patterns, void fraction, and friction pressure drop data obtained with the T-junction inlet
were found to be significantly different from those obtained previously with a reducer
inlet. For the T-junction inlet, the two-phase flow patterns in the microchannel were
predominantly intermittent flows with short gas and liquid plugs/slugs flowing with nearly
equal velocities. The void fraction data then conformed nearly to that of a homogeneous
flow model, and the two-phase friction multiplier data could be described by the
LockhartMartinelli correlation applicable to larger channels. However, when a reducer
inlet was used previously and the diameter of the inlet section was much larger than that
of the microchannel, an intermittent flow of long gas slugs separated by long liquid slugs
became prevalent and the void fraction decreased to values far below the homogeneous
void fraction. The differences in the two-phase flow characteristics between a T-junction
inlet and reducer inlet were attributed to the differences in the gas bubble/slug generation
mechanisms. DOI: 10.1115/1.3089543
Introduction
Few two-phase flow studies are yet available in the literature

that involved a microchannel diameter of 100 m or less as noted
in recent reviews by Serizawa 1 and Thome 2. The available
results for microchannels have shown certain differences from
those for conventional channels: bubbly and droplet flows may
not always occur in microchannels, the void fraction data in microchannels may deviate from the Armand correlation 3 and the
two-phase friction pressure drop in microchannels could be predicted by conventional models.
The available results also indicate that the adiabatic two-phase
flow characteristics in the same microchannel such as flow patterns, void fraction, and friction pressure drop can vary significantly depending on the microchannel inlet configuration and the
gas-liquid mixing method, unlike in minichannels and conventional channels.
Stanley et al. 4 investigated two-phase pressure drop and heat
transfer in the square and rectangular aluminum channels of hydraulic diameters Dh ranging from 56 m to 256 m. They
mixed water with argon, helium, or nitrogen gas upstream of the
1
microchannel and injected the two-phase mixture from a 4 in.
tube into the microchannel test section, so the inlet was a reducer
1
with sudden reduction in diameter from 4 in. to 56 256 m.
They reported that the conventional semi-empirical correlations
for two-phase pressure drop overpredicted their microchannel data
substantially.
Air-water two-phase flow experiments were conducted in
25 m and 100 m circular microchannels connected to a
reducer-type inlet by Serizawa and co-workers 5,6. Five flow
patterns were observed: dispersed bubbly flow, gas slug flow,
liquid-ring flow, liquid lump flow, and liquid droplet flow. The
occurrence of liquid-ring flow and liquid lump flow clearly indi1
JOURNAL OF FLUIDS ENGINEERING. Manuscript received October 8, 2007, final manuscript received January 19, 2009; published online March 11, 2009. Assoc. Editor:
Theodore Heindel. Paper presented at the Third International Conference on Microchannels and Minichannels ICMM2005, Toronto, ON, Canada, June 1315, 2005.
cated that there are differences in the flow patterns between microchannels of D 100 m and minichannels of D 1 mm.
Kawaji and co-workers 7,8 investigated the two-phase flow of
nitrogen gas and water through horizontal circular microchannels
of 50 250 m diameter. The gas and liquid were mixed in a
reducer inlet where the diameter changed from 750 m down to
the microchannel diameter. They reported significant differences
in the flow pattern maps and void fraction data from those previously described for minichannels with diameters above 1 mm.
In particular, they reported the existence of flow patterns unique to
microchannels liquid-ring flow and serpentinelike gas core flow.
Furthermore, the time-average void fraction data deviated from
Armands correlation 3, which is normally applicable to minichannels with diameters 1 mm. This was in contrast with the
void fraction data of Serizawa et al. 6 for a 20 m circular
microchannel, which could be well described by Armands type
correlation.
Chung et al. 9 later found that the two-phase flow characteristics in a 96 m square channel and a 100 m circular channel
connected to the same reducer inlet used by Kawahara et al. 7
were quite similar and independent of the channel cross section
for a mixture of nitrogen gas and water. Furthermore, Kawahara et
al. 10 found only a moderate effect of fluid properties such as
liquid viscosity and surface tension on the two-phase flow characteristics in a microchannel.
Besides the reducer-type inlet, several different inlet designs
and gas-liquid mixing methods have been used to inject a twophase mixture into a microchannel. A simple T-junction has also
been popular particularly for microfluidic applications as used by
Thorsen et al. 11, Okushima et al. 12, Gnther et al. 13, Xu et
al. 14, Garstecki et al. 15, and Ide et al. 16, among others. In
many cases, the T-junction diameter was the same as or close to
the microchannel diameter, and its use was intended to produce
monodisperse microbubbles. On the other hand, a much larger
1/8 in. diameter T-junction was used by Xiong and Chung 17
upstream of a rectangular microchannel. The gas was injected into
the liquid stream through a 200 m diameter needle, which was
inserted from the branch of the T-junction so that the needle tip
would be positioned in the middle of the T-junction.
APRIL 2009, Vol. 131 / 041302-1
Fig. 1 Experimental apparatus
In recent experiments, Yue et al. 18 used a Y-junction connected to square microchannels with 200 m and 400 m hydraulic diameters to absorb CO2 gas into a water stream. Another
inlet geometry used in the past involved a flow focusing method
in a cross junction by Gan-Calvo and Gordillo 19, Cubaud
and Ho 20, and Garstecki et al. 21, where the gas flowed
straight through the main line into the microchannel and liquid
was injected from the sides through two branches at equal flow
rates. This way, the gas stream could be pinched off to generate
discrete small gas bubbles, short plugs, or long slugs depending
on the gas and liquid injection rates. Another variation of the
T-junction inlet was used by Waelchli and von Rohr 22, who
injected gas and liquid streams from the opposite ends of the main
channel. The two phases mixed at the T-junction where several
50 m diameter round pins were placed to serve as a static mixer,
and then the mixture flowed into the branch line connected to a
microchannel.
As reviewed above, different types of inlet and gas-liquid mixing methods have been used in adiabatic two-phase flow experiments in microchannels, but no systematic comparisons of the
two-phase flow characteristics have been carried out. Some of the
results in the literature have shown significant differences, but the
diameters of the microchannels were not the same, so the effects
of the inlet geometry and mixing method cannot be readily isolated from those of the microchannel diameter. Hence, the objective of this work was to investigate the effect of the inlet geometry
using the same microchannel of 100 m diameter and identify
the key reasons for differences in the two-phase flow characteristics observed in the microchannel. The same experimental apparatus and approach used by Kawahara et al. 7 and Chung and
Kawaji 8 have been used again with the same microchannel but
different inlet designs and gas-liquid mixing configurations.
Since there is no phase change heat transfer involved in this
work, the results are not directly applicable to heat exchangers.
Most relevant applications would be in microreactors in which gas
and liquid streams are contacted in a microchannel to effect gas
absorption with or without a chemical reaction. In the design of
such microreactors, the gas-liquid inlet and mixing method can be
chosen from a T-junction, Y-junction, co-axial reducer, cross junction, and others. Thus, it would be useful to understand better how
the two-phase flow characteristics are affected by the inlet geometry and gas-liquid mixing method.
Fig. 2
Gas-liquid mixing at the microchannel inlet
For the microchannel inlet, a Valco Nanovolume four-way

cross connector with an inner bore diameter of 100 m was used
to directly connect the microchannel test section to the gas and
liquid injection lines as shown in Fig. 2. The fourth line was
connected to a pressure transducer to measure the test section inlet
pressure, so the inlet was effectively a T-junction. To study the
effect of gas-liquid injection schemes for the T-junction inlet, two
configurations were tested.
1 Configuration 1 ConFig 1: liquid injected into the main
channel and gas injected into the branch
2 Configuration 2 ConFig 2: gas injected into the main
channel and liquid injected into the branch.
In both configurations, mixing of the gas and liquid phases took
place inside the 100 m diameter junction immediately before
the microchannel test section. Thus, there was no change in the
tubing diameter between the gas-liquid mixing section and the
microchannel test section.
Figure 3 illustrates the reducer inlet used previously by Kawahara et al. 7 and Chung and Kawaji 8 to mix the gas and liquid
phases upstream of a microchannel. Liquid was injected through a
500 m ID tube, while gas was injected through a 200 m annular gap. The two streams mixed by flowing collectively into a
500 m ID tube that was connected to a 250 m ID T-junction
for pressure measurement just upstream of the microchannel test
section. The two-phase mixture thus flowed through 750 m and
500 m diameter tube sections before entering the 250 m diameter T-junction and finally into the 100 m diameter test section.
A circular microchannel fused silica capillary tubing with standard polyimide coating from Polymicro Technologies, Phoenix,
AZ with an inner diameter of 100 m used by Kawahara et al.
7 was again used in the present experiments. The total length of
the test section from the gas-liquid mixing point to the exit of the
microchannel was 65 mm. A sufficiently large length-to-diameter
ratio L / D = 650 was used to diminish the entrance and exit ef-
Experimental Apparatus and Data Analysis
Details of the present experiments are explained briefly. The

reader can refer to Kawahara et al. 7 and Chung and Kawaji 8
for additional information not given here. Figure 1 illustrates the
experimental apparatus used. De-ionized water was driven
through the horizontal microchannel by a pneumatic-type pump
after being mixed with nitrogen gas from a pressurized cylinder.
041302-2 / Vol. 131, APRIL 2009
Fig. 3 Reducer inlet and mixer used by Kawahara et al. 7 and

Chung and Kawaji 8
Table 1 Measurement uncertainties

Parameters
Mass flow rate of liquid, mL
Mass flow rate of gas, mG
Superficial liquid velocity, jL
Superficial gas velocity, jG
Void fraction,
Total two-phase pressure gradient, P / Z
Two-phase frictional pressure gradient, P f / ZTP
Two-phase friction multiplier, L2
Uncertainty range
0.1% to 1%
1% to 10%
2.1% to 3%
3% to 12%
0.05
1% to 4%
2% to 7%
2% to 7%
fects. The window for flow visualization was located as far back
as possible 46.1 mm from the gas-liquid mixing point on the
microchannel to achieve a well-developed flow.
Three pressure transducers PX303 series from Omega Engineering
Stamford,
CT
covering
different
ranges
3450 8.6 kPa, 690 1.7 kPa, and 210 0.52 kPa were used
to measure the pressure drop between the microchannel inlet and
outlet. The liquid was discharged freely from the test section, so
that the pressure was atmospheric and temperature ambient at the
microchannel outlet.
The liquid flow rate was determined by using a load cell
50.97 0.255 g, model ELJ-0.5N from Entran Devices Fairfield, NJ to weigh the discharged liquid collected over time. The
gas flow rate was measured using three mass flow meters model
179A-23012 from MKS Instruments Andover, MA, and models
8172-0411 and 8172-0451 from Matheson Gas Products Montgomeryville, PA to encompass a wide range of conditions
1 0.01 sccm, 10 0.1 sccm, and 50 0.5 sccm. The analog
signals for all of the pressure, temperature, and mass flow rate
readings were sampled at 10 Hz by a 16-bit data acquisition system model NI 6035E from National Instruments Austin, TX.
Background illumination to see the flow patterns was generated
by a cold lamp model DCR III from Schott-Fostec Auburn,
NY located behind the test section. The view field was magnified by attaching a 20 microscope objective lens model 378802-2 from Mitsutoyo America Corporation Aurora, IL to the
video camera. A monochrome CCD camera model TM-1040
from Pulnix America Sunnyvale, CA with a resolution of
1024H 1024V pixels, a maximum frame rate of 30 fps, and
a shutter speed of 1/16,000 s was used for image capturing. The
problem of optical distortion was dealt with earlier by Kawahara
et al. 7. They elected to not correct for optical distortion so that
the images of the gas-liquid interfacial structure near the channel
wall could be enlarged and better examined. Likewise, optical
distortion is not addressed here for the same reason.
The previous experiments covered a range of 0.01 jL
5.77 m / s for the superficial liquid velocity and 0.02 jG,avg
73 m / s for the average superficial gas velocity. The new experiments covered slightly smaller ranges: 0.027 jL 2.05 m / s
and 0.1 jG,avg 50 m / s. The values of dimensionless parameters calculated for the previous experiments are the Bond number
0.000083 Bo 0.0095, the superficial Weber number for liquid 0.00057 WeLS 116 and gas 0.0000020 WeGS 12,
the superficial Reynolds number for liquid 2 ReLS 2300 and
gas 0.4 ReGS 751, and the capillary number 0.00013 CaL
0.070. These values indicate the relative magnitudes of the
forces
involved
inertia surface tension viscous force
gravity.
2.1 Measurement Uncertainties and Data Analyses. The
measurement uncertainties estimated for the present experiments
are summarized in Table 1. A propagation of error analysis 23
was performed to estimate the uncertainties in the derived quantities such as superficial velocities, void fraction, and pressure
gradients.
Fig. 4 Two-phase flow patterns observed in a microchannel

with a micro-T-junction inlet
Initially, experiments were conducted with de-ionized water to

calculate the single-phase friction factor from the frictional pressure drop. The computed friction factor was compared with the
conventional value for laminar flow in a pipe. The two-phase frictional pressure drop was calculated by subtracting estimates of the
pressure drop due to inlet contraction and flow acceleration from
the total pressure drop measured in the horizontal channel. To
estimate the pressure drop attributed to the acceleration of the
flow, the void fraction at the inlet and outlet of the test section was
estimated based on the homogeneous flow model.
The experimental results on the two-phase flow patterns, void

fraction, and two-phase friction pressure drop obtained with a
T-junction inlet will be presented and compared with those reported previously by Kawahara et al. 7 for a reducer inlet to
determine the effects of inlet geometry and gas-liquid mixing
method.
3.1 Two-Phase Flow Patterns. The two-phase flow patterns
obtained in the present experiments are found to be quite different
from those observed in our previous work. Typical examples of
the flow patterns seen in the present experiments for both Configurations 1 and 2 are shown in Fig. 4. Short gas bubbles or plugs
such as those shown in Figs. 4a and 4b were seldom observed
in the previous experiments using a series of reducer sections
between the gas-liquid mixer and the microchannel test section.
Instead, slug flows with a long gas core shown in Fig. 4c and
surrounded by a thin, thick, or wavy liquid film were predominant. Thus, the flow patterns observed in the present experiments
were classified differently from the previous experiments and according to the following definitions.
1 Bubbly flow: the presence of spherical bubbles whose diameters are less than the diameter of the microchannel
2 Plug flow: the presence of gas plugs that are shorter than
the width of the viewing window 250 m and are
separated by liquid bridges
APRIL 2009, Vol. 131 / 041302-3
Fig. 6 Sequence of images showing gas plug generation at a

micro-T-junction from Ref. 24
Fig. 5 Two-phase flow pattern maps for a micro-T-junction inlet a ConFig. 1 and b ConFig. 2, and c a reducer inlet used
by Kawahara et al. 7 an open circle indicates the occasional
appearance of a ring-film flow pattern
3 Slug flow: the presence of gas slugs that are longer than the
viewing window
4 Semi-annular flow: long gas core surrounded by a liquid
film and interrupted by infrequent liquid slugs
5 Ring flow: 10% or more of the images for a given run have
a ring-shaped liquid film surrounding the gas core
The two-phase flow pattern maps were then constructed based
on the above definitions for the two T-junction inlet configurations
as shown in Fig. 5. At low superficial gas velocities jG 1 m / s,
bubbly flow, plug flow, and slug flow patterns appeared at progressively lower liquid flow rates. At moderate to high superficial
gas velocities jG 2 m / s, slug and semi-annular flow patterns
were observed. An open circle in Figs. 5a and 5b indicates
the occasional appearance of a ring-film flow pattern.
The difference in the gas and liquid injection configurations for
the micro-T-junction inlet had only a small effect on the flow
pattern map as clear from a comparison of Fig. 5a and 5b.
However, these results for the micro-T-junction inlet were significantly different from the map previously obtained for a reducer
inlet case by Kawahara et al. 7 and shown in Fig. 5c. When the
inlet section consisted of a series of reducer sections from
750 m to 100 m Fig. 2, the predominant flow pattern ob041302-4 / Vol. 131, APRIL 2009
served was a slug flow with a long gas core surrounded by a thin,
thick or ring-shaped film. Depending on the frequency of the appearance of different liquid film shapes, the flow patterns were
defined as slug-ring, ring-slug, and multiple flows Fig. 5c.
Although a semi-annular flow pattern was observed in both
micro-T-junction and reducer inlet experiments, a strong difference existed between them in the appearance of short gas bubbles
and plugs. This is due to the fact that small gas bubbles are readily
created at the T-junction inlet as reported in previous studies. The
mechanism of small gas bubble generation at the micro-T-junction
can be readily illustrated as shown in Fig. 6. Such a mechanism
had been previously investigated for the ConFig. 1 case liquid
injected into the main channel while gas is injected into the
branch, Fig. 6a by Shibata et al. 24. Short gas plugs were
observed to be generated as a result of interfacial instability or
pinch-off at the corner of the T-junction as shown in Figs.
6b6e.
Although no flow visualization was conducted for ConFig. 2, a
similar interfacial instability or pinching off of a gas bubble or
plug at the micro-T-junction can be conjectured. Thus, regardless
of which phase is injected into the main channel or the branch, the
use of a T-junction to mix and inject the gas-liquid mixture into a
microchannel can result in two-phase flow patterns that are very
different from those observed in the reducer inlet case. The larger
inlet cross section D = 750 m for the reducer inlet used by
Kawahara et al. 7 would allow larger gas bubbles to form upstream of the microchannel test section. When these gas bubbles
flow into the microchannel with a much smaller diameter D
= 100 m, the gas slug length would increase by more than 50
times, as illustrated in Fig. 7. Thus, for example, if the diameter of
the gas bubble formed in the inlet section were 2 mm, then it
would form a 100 mm long gas slug in the microchannel. This can
explain why Kawahara et al. 7 obtained mostly long gas core
flows with a reducer inlet rather than short gas plug flows.
The present flow pattern results obtained with a T-junction inlet
are now compared with the previous results obtained with other
types of inlet as shown in Fig. 8. Only the bubbly-to-plug or slug
and slug-to-annular or semi-annular flow pattern transition
boundaries are indicated. Although the transition boundaries are
represented by thin lines, the flow patterns did not change as
sharply as suggested by those lines. Nevertheless, the present
Fig. 7 Formation of a long gas slug from a bubble in a reducer inlet not to scale
T-junction inlet results are in general agreement with previous

studies for both bubbly-to-slug and slug-to-annular flow transition
boundaries.
It is noted in Fig. 8 that unlike the reducer inlet results obtained
by Kawahara et al. 7, the flow pattern transition boundaries obtained by Serizawa et al. 6 are in agreement with the T-junction
inlet results. This may be attributed to the small diameter of the
gas injection nozzle D = 40 m used within the 300 m diameter reducer inlet section, which was connected to a 25 m diameter microchannel. The size of the gas bubbles generated could
have been much smaller than the internal diameter of the reducer
due to the small size of the gas injection nozzle and the effect of
co-axially flowing liquid, so the phenomenon illustrated in Fig. 7
may not have occurred in the experiments of Serizawa et al. 6.
3.2 Void Fraction. The differences in the two-phase flow patterns obtained in microchannels with a T-junction inlet and reducer inlet can also cause large differences in the void fraction as
discussed below. To estimate the void fraction in the 100 m
microchannel, about 300 video images were analyzed for each
run. The instantaneous void fraction was estimated for plug flow
images such as those shown in Fig. 4 by measuring the length of
a gas bubble or plug and calculating the volume by assuming a
cylindrical shape with semispherical caps. The instantaneous void
fraction was then computed as the ratio of the gas volume to the
total channel volume within the viewing window. The images
were assigned a void fraction of zero when they showed only the
liquid flowing alone in the entire view window or a void fraction
of unity when they showed a continuous gas core with a thinsmooth liquid film or ring-shaped liquid film. The time-averaged
void fraction was then obtained by averaging the instantaneous
void fraction values from about 300 images. The effect of optical
distortion on the above-mentioned method of measuring the void
fraction, particularly for long gas core flows with ring-shaped
films and thick liquid films, was previously addressed by Kawahara et al. 7.
The time-averaged void fraction data for the T-junction inlet
are plotted against the volumetric quality =JG / JG + JL in Figs.
9 and 10 for Configurations 1 and 2, respectively. The homogeneous flow model = and the Armand correlation 3 for narrow channels recommended by Ali et al. 25 are also shown in
the figures, respectively. Ali et al. 25 recommended using a correlation of the form = 0.8 originally developed by Armand 3
for narrow rectangular channels with DH 1 mm.
Both Configurations 1 and 2 data show a linear relationship
between the void fraction and volumetric quality, consistent with
the homogeneous flow model and Armands correlation, but unlike the nonlinear relationship obtained by Kawahara et al. 7 for
the reducer inlet case and given by
=
0.030.5
1 0.970.5
Equation 1 indicates the void fraction to increase slowly with

increasing volumetric quality and then exponentially after
0.8. This trend implies a strong departure from homogeneous
flow and indicates a large velocity slip between the liquid and gas
phases.
Thus, as the inlet geometry was changed from a reducer to
T-junction, the two-phase flow patterns in the microchannel
changed and this also strongly affected the time-average void fraction. For example, a plug flow pattern with alternating passages of
Fig. 8 Comparison of flow pattern transition boundaries
APRIL 2009, Vol. 131 / 041302-5
Fig. 9 Void fraction data for Configuration 1 of the T-junction

inlet; homogeneous model = , Armand correlation = 0.833, and ---- Eq. 1
short gas and liquid plugs as shown in Fig. 4 would lead to a slip
ratio close to unity, which is representative of a homogeneous
flow. Since such a plug flow pattern was not observed in the
reducer inlet case, the void fraction deviated strongly from the
homogeneous flow model in the experiments of Kawahara et al.
7. The present data show that the inlet geometry and the method
of gas-liquid mixing can drastically change the time-average void
fraction for the same microchannel at given superficial gas and
liquid velocities.
3.3 Friction Pressure Drop. The two-phase friction pressure
drop was evaluated from the measured total pressure drop by subtracting the acceleration pressure drop. The data were correlated
using the LockhartMartinelli correlation 26, which was developed based on a separated flow assumption. This correlation uses
a two-phase friction multiplier L2 defined as follows:

P f
Z
= L2
TP

P f
Z
2
L
where P f / ZL is the frictional pressure drop when liquid is

assumed to flow alone in the channel.
The two-phase friction multiplier has been correlated in terms
of the LockhartMartinelli parameter X given by
X2 =
P f /ZL
P f /ZG
where P f / ZG is the frictional pressure drop when the gas is

assumed to flow alone in the channel. A widely used correlation to
calculate the friction multiplier is that proposed by Chisholm and
Laird 27.
Fig. 10 Void fraction data for Configuration 2 of the T-junction

inlet; homogeneous model = , Armand correlation = 0.833, ---- Eq. 1
041302-6 / Vol. 131, APRIL 2009
Fig. 11 Two-phase friction multiplier data for a ConFig. 1 and

b ConFig. 2 of the T-Junction Inlet
L2 = 1 +
C 1
+
X X2
where the coefficient C is a constant ranging in value from 5 to 20

for larger channels, depending on whether the liquid and gas flows
are laminar or turbulent. In conventional channels and minichannels, a strong mass velocity effect has been observed in the twophase friction multiplier data 25, because the flow can change
from laminar to turbulent as the total mass velocity is increased.
The C-value in Eq. 4 is also known to decrease as the channels
hydraulic diameter is reduced 28,29.
The two-phase friction multiplier data obtained with a
T-junction inlet in the present experiments are compared with the
LockhartMartinelli correlation in Fig. 11. Also shown are the
curves given by Eq. 4 with C = 5 and the best fit values of C
= 2.0 and 2.5 for the present experiments, as well as C = 0.24 for
the reducer inlet experiments obtained by Kawahara et al. 7.
Overall, the LockhartMartinelli model with properly chosen values of C can be used to predict the two-phase friction multiplier
data reasonably well for the T-junction inlet experiments. Switching the gas and liquid injection lines in the T-junction Configurations 1 and 2 is seen to have only a minor effect.
The data from the T-junction inlet experiments for both Configurations 1 and 2 can be represented by different C-value curves
depending on the range of the LockhartMartinelli parameter C
= 2.0 or 2.5 for X 2, and C = 5 for X 3. These C-values are
more than ten times greater than the value of C=0.24, obtained
for the reducer inlet by Kawahara et al. 7. Although not shown
here, only small deviations from the C = 0.24 curve were seen for
the reducer inlet data for comparable ranges of gas and liquid flow
rates tested 7. This is likely to be due to a greater variety of flow
patterns observed in the T-junction inlet experiments described
earlier. Thus, as in the void fraction data, the two-phase friction
pressure drop could be significantly different between the

T-junction and the reducer inlet cases indicating the strong effect
of the inlet geometry.
References
Conclusion
The effects of inlet geometry and gas-liquid injection method

on adiabatic two-phase flow in a 100 m microchannel have
been investigated experimentally. Nitrogen gas and water were
injected into a micro-T-junction with the same diameter as the
microchannel connected downstream. The two-phase flow patterns observed have been found to differ strongly from those obtained in a previous study using a reducer inlet and a 750 m
diameter mixing section. For the micro-T-junction inlet, plug and
slug flows with alternating short gas plugs and liquid slugs were
observed, which were rarely observed in the reducer inlet case.
Due to the obvious differences in the flow patterns observed,
the void fraction data for the micro-T-junction inlet case agreed
with the homogeneous flow model, unlike the data obtained with
a reducer inlet with a 750 m diameter mixing section. The twophase friction multiplier data could be well correlated in terms of
the LockhartMartinelli parameter but the value of the constant C
in Chisholms correlation was much larger than the value obtained
for the reducer inlet case.
Since the time-average void fraction and consequently interfacial area and friction pressure drop would strongly affect the
performance of microchannel devices such as microreactors for
direct contacting of gas and liquid streams, the dimensions and
type of a gas-liquid mixer to be used at the microchannel inlet
would have to be carefully considered in the design process.
Acknowledgment
This work was supported by a Discovery research grant from
the Natural Sciences and Engineering Research Council of
Canada. K. Mori would like to thank Osaka ElectroCommunication University for a research leave grant, which allowed a one-year stay at the University of Toronto.
Nomenclature
Bo
C
Ca
D
g
j
L
m
P
Re
u
We
X
Z
Greek Symbols

P

L2

Subscripts
f
G
L
M
S superficial
TP two-phase
Bond number
coefficient in Eq. 4
capillary number
tube inner diameter m
gravitational acceleration m / s2
superficial velocity m/s
length m
mass flow rate kg/s
pressure Pa
Reynolds number
axial velocity m/s
Weber number
LockhartMartinelli parameter
flow direction coordinate m
volumetric quality
void fraction
pressure drop
dynamic viscosity Pa s
two-phase friction multiplier
density kg/ m3
surface tension N/m
friction
gas phase
liquid phase
mixing section
1 Serizawa, A., 2006, Gas-Liquid Two-phase Flow in Microchannels, Multiphase Flow Handbook, C. T. Crowe, ed., CRC Press, Boca Raton, FL.
2 Thome, J. R., 2006, State-of-the-Art Overview of Boiling and Two-Phase
Flows in Microchannels, Heat Transfer Eng., 279, pp. 419.
3 Armand, A. A., 1946, The Resistance during the Movement of a Two-phase
System in Horizontal Pipes, Izv. Vses. Teplotekh. Inst. AERE-Lib/Trans.
828, 1, pp. 1623.
4 Stanley, R. S., Barron, R. F., and Ameel, T. A., 1997, Two-Phase Flow in
Microchannels, Micro-Electro-Mechanical Systems (MEMS), ASME, DSCVol. 62/HTD-Vol. 354, New York, pp. 143152.
5 Serizawa, A., and Feng, Z. P., 2001, Two-Phase Flow in Micro-Channels,
Proceedings of the Fourth International Conference on Multiphase Flow, New
Orleans, LA, May 27Jun. 1.
6 Serizawa, A., Feng, Z., and Kawara, Z., 2002, Two-Phase Flow in Microchannels, Exp. Therm. Fluid Sci., 2667, pp. 703714.
7 Kawahara, A., Chung, P. M.-Y., and Kawaji, M., 2002, Investigation of TwoPhase Flow Pattern, Void Fraction and Pressure Drop in a Microchannel, Int.
J. Multiphase Flow, 289, pp. 14111435.
8 Chung, P. M. Y., and Kawaji, M., 2004, The Effect of Channel Diameter on
Adiabatic Two-Phase Flow Characteristics in Microchannels, Int. J. Multiphase Flow, 3078, pp. 735761.
9 Chung, P. M.-Y., Kawaji, M., Kawahara, A., and Shibata, Y., 2004, TwoPhase Flow Through Square and Circular MicrochannelsEffect of Channel
Geometry, ASME J. Fluids Eng., 1264, pp. 546552.
10 Kawahara, A., Kawaji, M., Chung, P. M.-Y., Sadatomi, M., and Okayama, K.,
2005, Effects of Channel Diameter and Liquid Property on Void Fraction in
Adiabatic Two-Phase Flow Through Microchannels, Heat Transfer Eng.,
263, pp. 1319.
11 Thorsen, T., Roberts, R. W., Arnold, F. H., and Quake, S. R., 2001, Dynamic
Pattern Formation in a Vesicle-Generating Microfluidic Device, Phys. Rev.
Lett., 86, pp. 41634166.
12 Okushima, S., Nishisako, T., Torii, T., and Higuchi, T., 2004, Controlled
Production of Monodisperse Double Emulsions by Two-Step Droplet Breakup
in Microfluidic Devices, Langmuir, 20, pp. 99059908.
13 Gnther, A., Khan, S. A., Thalmann, M., Trachsel, F., and Jensen, K. F., 2004,
Transport and Reaction in Microscale Segmented GasLiquid Flow, Lab
Chip, 4, pp. 278286.
14 Xu, J. H., Li, S. W., Wang, Y. J., and Luo, G. S., 2006, Controllable GasLiquid Two Phase Flow Patterns and Monodispersive Microbubbles in a Microfluidic T-Junction Device, Appl. Phys. Lett., 88, pp. 133506.
15 Garstecki, P., Fuerstman, M. J., Stone, H. A., and Whitesides, G. M., 2006,
Formation of Droplets and Bubbles in a Microfluidic T-JunctionScaling
and Mechanism of Break-Up, Lab Chip, 6, pp. 437446.
16 Ide, H., Kimura, R., and Kawaji, M., 2007, Optical Measurement of Void
Fraction and Bubble Size Distributions in a Microchannel, Heat Transfer
Eng., 288, pp. 713719.
17 Xiong, R., and Chung, J. N., 2007, An Experimental Study of the Size Effect
on Adiabatic Gas-Liquid Two-phase Flow Patterns and Void Fraction in Microchannels, Phys. Fluids, 19, p. 033301.
18 Yue, J., Chen, G. W., and Yuan, Q., 2004, Pressure Drops of Single and
Two-Phase Flows Through T-Type Microchannel Mixers, Chem. Eng. J.,
102, pp. 1124.
19 Gan-Calvo, A. M., and Gordillo, J. M., 2001, Perfectly Monodisperse Microbubbling by Capillary Flow Focusing, Phys. Rev. Lett., 87, p. 274501.
20 Cubaud, T., and Ho, C. M., 2004, Transport of Bubbles in Square Microchannels, Phys. Fluids, 16, pp. 45754585.
21 Garstecki, P., Stone, H. A., and Whitesides, G. M., 2005, Mechanism for
Flow-Rate Controlled Breakup in Confined Geometries: A Route to Monodisperse Emulsions, Phys. Rev. Lett., 94, p. 164501.
22 Waelchli, S., and von Rohr, P. R., 2006, Two-Phase Flow Characteristics in
GasLiquid Microreactors, Int. J. Multiph. Flow, 327, pp. 791806.
23 Kline, S. J., and McClintock, F. A., 1953, Describing Uncertainties in Single
Sample Experiments, Mech. Eng. Am. Soc. Mech. Eng., 75, pp. 38.
24 Shibata, Y., Ikeda, K., Hanri, T., Chung, P. M. Y., and Kawaji, M., 2003,
Study on Two-Phase Flow Through a T-Junction, Proceedings of the Fourth
ASME-JSME Joint Fluids Engineering Conference, Honolulu, Hawaii, Jul.
611, Paper No. FDSM2003-45385.
25 Ali, M. I., Sadatomi, M., and Kawaji, M., 1993, Two-Phase Flow in Narrow
Channels Between Two Flat Plates, Can. J. Chem. Eng., 715, pp. 657666.
26 Lockhart, R. W., and Martinelli, R. C., 1949, Proposed Correlation of Data
for Isothermal Two-Phase, Two-Component Flow in Pipes, Chem. Eng. Prog.,
451, pp. 3948.
27 Chisholm, D., and Laird, A. D. K., 1958, Two-Phase Flow in Rough Tubes,
Trans. ASME, 802, pp. 276286.
28 Mishima, K., and Hibiki, T., 1996, Some Characteristics of Air-Water TwoPhase Flow in Small Diameter Vertical Tubes, Int. J. Multiphase Flow, 224,
pp. 703712.
29 Kawaji, M., 1999, Fluid Mechanics Aspects of Two-phase Flow, Handbook
of Phase Change: Boiling and Condensation, Taylor & Francis, London, Chap.
9, pp. 205259.
APRIL 2009, Vol. 131 / 041302-7
Thien X. Dinh
e-mail: thien@cfd.ritsumei.ac.jp
Yoshifumi Ogami
Ritsumeikan University,
1-1-1 Nojihigashi,
Kusatsu, Shiga 525-8577, Japan
A Principle to Generate Flow for

Thermal Convective Base
Sensors
This paper presents a thin millimeter-scaled device that can generate a closed flow within
itself with a velocity of the order of a few m/s. The device comprises a piezoelectric pump
with a PZT membrane, housing chamber, and a closed network channel connected to the
housing chamber through a specific throat. We investigate the device by computational
fluid dynamics. This device is used to produce several free jet flows depending on the
structure of the network channel. In this study, four jet flows comprising two perpendicular pairs of flows are demonstrated. If the PZT membrane vibrates within a suitable
range, the self-similarity of the axial velocity (along the jet direction) to the cross distances scaled by the half-widths of the jet is observed for a certain range of axial
distance. Each jet flow can bend almost freely in three dimensions. The two remaining
flow components are small as compared to the axial component. The device potentially
has wide applications in flow-based sensors. DOI: 10.1115/1.3089538
Keywords: microfluidics, micropumps, piezoelectrics, numerical simulation, free jet flows
Introduction
Microfluidics is a concept that is used to manipulate flow in

devices whose dimensions range from millimeters to micrometers.
Microfluidics has attracted attention for several reasons in recent
years. The first reason is the development of microelectromechanical systems MEMS, which simplifies the fabrication of
miniaturized devices. The second reason is the numbers of fields
that are related to microfluidics, such as convection-based sensors,
analytical chemistry, and lab-on-a-chip.
The flow is usually a vital aspect in the functioning of microfluidic devices. For example, biological samples must be moved
through the components of miniature assay systems 1,2 and a
coolant must be forced through microheat exchangers 3. Another
great application of flow in microfluidics is the Coriolis effectbased sensors e.g., gyroscope in which jet flow is required. The
flow in this type of sensors is usually generated by a selfcontained valveless micropump. The advantage of self-contained
valveless micropump Coriolis effect-based sensors is the absence
of moving parts, with exception of the diaphragm. Therefore, it
reduces the risk of damage or fragility as an incoherent of mechanical sensors. Furthermore, the convective sensors with valveless pump have quick respond performance. One of the greatest
difficulties of the Coriolis effect-based sensors is forming jet flow,
which can freely vibrate in space. In the recent applications for
gyroscopes 46, only one jet flow was created. This makes the
measurement ability of the gyroscopes limited in two components
of angular rate. In two later works, MEMS technique was applied
to fabricate the devices. The jet flow can bend freely on one
plane. Therefore, they are single axis sensors. The former gyroscope is dual axes sensing because the jet can move in threedimensional space. However, it is not compatible with MEMS
process, except for the fabrication of hotwires. In these works, the
flow properties have not been discussed in detail.
In this paper, we present a piezoelectrically actuated microfluidic device that can create a closed flow with four jet flows. The
JOURNAL OF FLUIDS ENGINEERING. Manuscript received June 25, 2008; final manuscript
received January 14, 2009; published online March 11, 2009. Assoc. Editor: James A.
Liburdy. Paper presented at the 2007 ASME International Mechanical Engineering
Congress IMECF2007, Seattle, WA, November 1016, 2007.
structure of the device is compatible to planar layer-by-layer geometry that is feasibly produced by lithography-based microfabrication techniques. The working principle and some primitive
results are presented. All velocity components are examined.
However, the axial component is mostly studied, since it is the
most important flow characteristic.
The Device and Working Principle
The sealed device comprises the pump chamber, the housing

chamber, and a network channel on a substrate. The pump chamber and the network channel are connected through a housing
chamber and the throats Fig. 1. The network channel includes
the drive-channel and the main chamber in which the jet flow is
created. One end of the main chamber is connected to the outermost channel through the nozzle, and the other end directly
matches the back-channels. The depth and length of the main
chamber are 1.5 mm and 3 mm, respectively. The main nozzle,
whose width and depth are 0.3 mm and 0.5 mm, respectively, is
located at a distance of 4 mm from the center of the housing
chamber. The widths of the other channels are typically 0.5 mm,
and their depths are the same as that of the main chamber. On the
top surface of the pump chamber, a thin piezoelectric layer is
bonded to vibrate this surface. The device is cylindrical and has a
diameter and depth of 10 mm and 2 mm, respectively.
The working principle of the device relies on the behavior of
the flow through the throat, as shown in Figs. 2a and 2b. The
length and thickness of the arrows qualitatively represent the flow
rates and their directions in the designated channel. In the pump
phase, the fluid in the housing chamber can be easily pushed
straight into the drive-channel than into the back-channels, which
are perpendicular to the flow. In the suction phase, because of the
geometry of the throat, the back flow from the drive-channel
slows down and becomes comparable to those from the backchannels. Therefore, the former flow is interrupted by the two
latter perpendicular flows. Additionally, the back flow from the
drive-channel experiences the diffuser-nozzle effect, wherein the
resistance of the nozzle is higher than that of the diffuser. Due to
this effect at the throat, a jet flow is created in the main chamber
through the nozzle Fig. 1.
APRIL 2009, Vol. 131 / 041401-1
discretization equations are advanced with time by a second-order

time stepping method. The semi-implicit method for pressurelinked equation SIMPLE scheme is used as the velocity-pressure
coupling algorithm. The general version of the algebraic multigrid
method AMG is used to solve the Poisson equation. The simulations were performed on RedHat 7.3 OS on a Pentium III CPU
with 4 Gbyte memory.
Usually, the PZT diaphragm is excited by an alternating circuit
with frequency f. Therefore, the time step should be determined
according to this frequency by the following expression:
t =
Fig. 1 Structure design of the device
The Governing Equations and Numerical Setup
The flow is generated by the vibration of a PZT diaphragm in a

sealed packed device. Therefore, the fluid inside the device is
compressed or extended due to the oscillation of the diaphragm.
To cover these effects, the governing equations of the motion of a
compressible Newtonian fluid are used as follows:
+ u = 0
t
u
+ u u = p + u
t
Here, u and p denote the velocity vector and pressure of the

flow field; and are the dynamic viscosity and density of the
fluid, respectively. Since the working fluid is air, the relation between pressure and density can follow the state equation of an
ideal gas, p = RaT / M w, where Ra is the universal gas constant,
M w is the molecular weight, and T is temperature. To close the
governing equations, the solution of temperature is needed by
solving the energy equation
c pT
+ u c pT = T
t
where N is an arbitrary integer. For the exact reproduction of the

flow behavior in 1 cycle of the PZT diaphragm oscillation, N is
selected to be greater than 20.
The grid sensitivity or simulation accuracy was assessed in advance by several additional simulations on a refined double mesh.
The solution for the mass flow rate in the jet nozzle is compared
for the cases of coarse and fine meshes. If the solutions for
the coarse and fine grids differ by less than 1%, the coarse grid
system is selected for the simulation. A mesh with 300,000 grid
points is sufficient to accurately simulate the flow.
The reference fluid properties, i.e., ideal air, are as follows:
temperature= 298 K, molecular weight M w = 28.96 g / mol, viscosity = 1.789 105 Pa s, thermal conductivity = 2.42
103 w / m K, and specific heat c p = 1006.43 J / kg K. The estimated Knudsen number, Kn= F / L, where F 100 nm is the
mean free path of the air and is less than 103. Hence, the continuum hypothesis and nonslip condition at the walls are applicable.
The PZT diaphragm is assigned as the inlet surface condition
for the simulation model of the device. Assuming uniform force
distribution on the membrane, which is mounted at its perimeter,
the displacement at a distance r from the center of the membrane
is given by 7
In Eq. 3, and c p are the thermal conductivity and specific

heat coefficient of the fluid, respectively.
In our simulation, the change of temperature is insignificant
i.e., 0.5 K; thus the density varies mainly with the change of
pressure due to the deformation of the PZT diaphragm.
Because of the symmetry flow feature, only one-eighth of the
device is simulated. The computational domain is discretized into
a structural mesh grid system by using GAMBIT, a commercial
software developed by ANSYS, Inc. The mesh grids are clustered
for regions where a high velocity gradient exists, i.e., near the
walls, throat, and main nozzle. Then, FLUENT 6.3 ANSYS, Inc. is
employed to obtain approximate values for the flow variables in
this grid system. A second-order accuracy scheme is used for the
convection and diffusion terms of each governing equation. The
1
fN
r,t = t 1
r
R
2 2
where the displacement of the membrane center t is a function

of time depending on the applied electric power and R denotes the
radius of the membrane. Since the applied power varies sinusoidally with time at frequency f, we obtain
t = 0 sin2 ft
Because 0 of the order of tens of microns is small as compared to the depth of the pump chamber 300 m, it is not
necessary to use the deformation mesh in the simulation. The local
rate of deformation of the diaphragm is converted to local velocity
applied on the inlet surface by taking derivative of Eq. 5 respective with time as
vr,t = 2 f0 1
r
R
cos2 ft
A commercial PZT membrane usually comprises LiNbO3 with

a natural frequency of 7 MHz. Therefore, in the present simulation, the frequency is selected as 7 MHz. This implies that the
membrane functions at its resonance frequency.
Fig. 2 Working principle of the device
041401-2 / Vol. 131, APRIL 2009
In order to facilitate the discussion, a Cartesian coordinate system is designated with the origin located at the center of the
nozzle outlet. The x-, y-, and z-axes correspond to the axial, horizontal, and vertical directions, respectively, as shown in Fig. 1.
The corresponding velocity components along the x-, y-, and
z-directions are denoted as u, v, and w.
Figures 3a and 3b shows the instantaneous vector velocity
field at the throat for the discharge and suction phases, respecTransactions of the ASME
Fig. 3 Flow behavior at the throat in the discharge phase a and suction phase b
tively. The vectors are plotted with constant length for simplification. The magnitude of the vector velocity is represented by the
color bar. The highest and lowest values are represented by red
and blue, respectively. The working principle of the device is
clearly observed. In the discharge phase, there is a ring-shaped
circulatory flow its cross sections are indicated by characters A
and B in the front and side views, as shown in Fig. 3a around
the throat. It intercepts the flow from the throat to the backchannels. The flow from the housing chamber goes straight to the
drive-channel. In the suction phase, the flows from the backchannels are comparable to that from the drive-channel at the
throat. Then, the former slows down the fluid from the drivechannel to the housing chamber. Moreover, the secondary flow
from the wall of the drive-channel indicated by the arrows in Fig.
3b also contributes to slowing down the flow from this channel
to the housing chamber. Consequently, only a small portion of the
flow in the drive-channel, which was created in the pump phase, is
sucked back into the housing chamber. The other major portion
still flows in the drive-channel. We can observe this effect by the
division of the vector velocity field into opposite directions at the
throat, as shown in Fig. 3b.
Figure 4 shows the variation in the spatial-average axial velocity at the nozzle outlet, UJ, against the dimensionless time t f.
We observe that the flow attains the steady state after 4 cycles of
vibration of the membrane. It is also observed that UJ completes 1
cycle during the dimensionless time t f = 1. In other words, the
flow in the main chamber oscillates at the same frequency as the
PZT membrane. It should be noted that the PZT vibrates at a very
high frequency. Therefore, it is expected that the flow is almost in
the steady state for the applications time scales.
Fig. 4 Variation in the spatial-average axial velocity at the outlet of the nozzle against time. The maximum amplitude of the
PZT membrane is 10 m
Hereafter, the characteristics of one jet flow are considered, and

all the mentioned quantities are the time averaged values. In our
present simulation, the time averaged values are computed in the
period of 20 t f 60. This ensures that the flow in the main
chamber has reached steady state, and that time is long enough for
taking averaged values.
The measured profile of the axial velocity ux , y = 0 , z along
the z-direction at different axial locations, x, is plotted in the lefthand side of Fig. 5 for the case of 0 = 15 m. Similarly, the
right-hand side of Fig. 5 plots the profiles of the axial velocity
ux , y , z = 0 against the y-distance from the axis of the jet flow for
the same working conditions. From Fig. 5, it is evident that the
peaks of the profile decrease with an increase in the axial distance
from the nozzle, x. This implies that the jet decays with an increase in the axial distance from its source. The tails of these
profiles have negative values. This shows that there is a circulatory flow in the main chamber.
Often, a central characteristic of free jet flows is self-similarity.
Therefore, roughly, the self-similar property of a jet flow can be
used to measure its freedom in space, which is an important feature in Coriolis effect-based applications. As shown in Fig. 5,
there is a secondary flow from the near wall of the main chamber
to the jet flow. Therefore, we seek the self-similar property of the
axial flow as a turbulent jet instead of a laminar jet.
In terms of the axial velocity field ux , y , z, the centerline velocity is
ucx = ux,0,0
The jets half-width along the y- and z-directions, y 1/2x and
Fig. 5 Profiles of ux , 0 , z and ux , y , 0 against z and y, respectively, for 0 = 15 m
APRIL 2009, Vol. 131 / 041401-3
Fig. 6 Axial velocity of the jet against cross-distance for different amplitude vibrations of the membrane and different axial
distance, x. The left-hand side of each figure shows a plot of
ux , 0 , z / ucx against z / z1/2x and the right-hand side shows
the variation in ux , y , 0 / ucx against y / y1/2x.
z1/2x, respectively, are defined as the transverse distance from

the jet axis to the location where the mean axial velocity is half of
the centerline value ux , 0 , 0 as 8
ux,y 1/2x,0 = 1/2ucx
ux,0,z1/2x = 1/2ucx
Figures 6a6c show the plots of the profiles ux , 0 , z / ucx

against z / z1/2x on the left-hand side and those of ux , y , 0 / ucx
against y / y 1/2x on the right-hand side for 0 = 5, 10, and
15 m, respectively. It is interesting to note that the profiles on
both sides of Figs. 6a6c collapse into a single curve for a
range of x. The tails of the profiles are slightly deviated. The
important conclusion is that the axial velocity profile exhibits the
self-similar property. It is found that the secondary flow and backchannel flows affect the profiles of ux , 0 , z / ucx against
z / z1/2x, as shown in the left-hand side of Figs. 6a6c. This is
clearly observed for the case when 0 = 15 m left-hand side of
Fig. 6c. Therefore, the profiles of ux , 0 , z / ucx against
z / z1/2x near the back-channels x 2 mm or nozzle x
0.5 mm are no longer self-similar. It is observed that for 0
= 5 15 m, the profiles of ux , 0 , z / ucx against z / z1/2x are
adequately self-similar when x ranges from 0.5 mm to 2 mm. In
contrast, only the effect of the back-channel flows affects the profiles of ux , y , 0 / ucx against y / y 1/2x, which breaks the selfsimilarity of these profiles for x 2 mm right-hand side of Figs.
6a6c. The self-similar region for the profiles of
ux , y , 0 / ucx against y / y 1/2x is 0.2 mm x 2 mm shown
in right-hand side of Figs. 6a6c. Furthermore, it is observed
from Figs. 6a6c that self-similarity is assured within twice
the jets half-width region i.e., y / y 1/2x 2 and z / z1/2x 2. In
the region of 0.5 mm x 2 mm, the profiles of ux , y , 0 / ucx
against y / y 1/2x and those of ux , 0 , z / ucx against z / z1/2x are
symmetrical to each other. In other words, the flow is similar to an
axisymmetric flow. From these discussions, the jet flow is expected to be free in three dimensions in the range of 0.5 mm
x 2 mm.
It should be noted that the self-similar profiles of the axial
velocity correspond to the cross-distance scaled by the jets halfwidth. This implies that the jet flow created in the main chamber
is more similar to a turbulent jet flow than to a laminar jet flow.
Theoretically, the profiles of the axial velocity of a laminar jet
041401-4 / Vol. 131, APRIL 2009
Fig. 7 Axial velocity of the jet against cross-distance for different amplitude vibrations of the membrane at x = 1.5 mm. The
left-hand side of each figure shows a plot of ux , 0 , z / ucx
against z / z1/2x and the right-hand side shows the variation in
ux , y , 0 / ucx against y / y1/2x. The approximate curve is plotted by Eq. 10.
flow are self-similar to the momentum width i.e., J / u2c x1/2,

where J denotes the surface integral of the momentum of the jet at
its outlet.
Figure 6d plots the axial velocity profiles in the same manner
as Figs. 6a6c, with the exception that 0 = 20 m. This
shows that the self-similar profiles are no longer observed for the
entire range of axial distance. This is because the effects of the
secondary and back-channel flows expand throughout the main
chamber. These effects destroy the free property of the jet. Therefore, the flows cannot be self-similar. It is noteworthy that selfsimilarity in the main chamber requires a specific condition for the
amplitude of vibration of the PZT membrane and occurs for a
certain range of distance from the nozzle. In Coriolis effect-based
sensor applications, a flow that is less affected by the wall of the
channel is preferable. Therefore, the recommended 0 in our
present studied device should be around 10 m. A general conclusion here is that amplitude of the vibration of the PZT membrane is an important factor for studying the device sensitivity.
Figure 7 plots the axial velocity profiles at x = 1.5 mm middle
of the self-similar region in the same manner as Fig. 6 for the
cases of 0 = 5, 10, and 15 m. They also collapse onto a single
curve. In other words, the self-similar profile is not affected by the
jets velocity. The solid lines represent the approximate curves
obtained by the following equation 8:
f = 1 + S22
10
where S is computed by the nonlinear least squares method and its

value is 0.411 for both sides of Fig. 7. is a dimensionless variable representing y / y 1/2x or z / z1/2x. Figure 7 shows that in the
range of y / y 1/2x 1.25 and z / z1/2x 1.25, the velocity distribution can be evaluated by Eq. 10.
In order to complete the description of the axial velocity, the
axial variations in the center axial velocity ucx and the jets
half-widths y 1/2x and z1/2x must be determined in the selfsimilar region i.e., 0.5 mm x 2 mm.
For practical applications, the center axial velocity ucx should
be relatively measured with the center velocity of the diaphragm,
i.e., 20 f. It is found that ucx / 20 f varies linearly with the
axial distance in the self-similar region, as plotted in Fig. 8. A
lower amplitude vibration of the membrane generates a higher
slope coefficient absolute value because the flow magnitude decreases more rapidly for decreasing 0.
A feature of the current design compatible with MEMS photolithography fabrication on a silicon wafer is that the ratios of the
main chamber depth to the nozzle depth and that of the main
chamber width to the nozzle width are different. This results in the
Fig. 8 Variation in the center velocity against axial distance in

the self-similar region
difference between y 1/2x and z1/2x, as shown in Fig. 9. The

half-widths of the flow in the depth, z1/2x, appear to remain
unchanged with axial distance, x, and their values are approximately 0.2. Figure 9 also shows that the jet spreads along the
y-direction. The spread is approximately linear with a rate,
dy 1/2 / dx, of around 0.450.5.
The other flow components, vx , y , z and wx , y , z, along the
y- and z-directions, respectively, are also examined in the selfsimilar region i.e., 0.5 mm x 2 mm. Similar to the examination of the axial flow, vx , y , z / ucx and wx , y , z / ucx are
plotted as functions of y / y 1/2x and z / z1/2x, respectively, as
shown in Fig. 10. The self-similar properties are not observed in
these cross section velocity components. The cross profiles,
vx , y , z / ucx against z / z1/2x and wx , y , z / ucx against
y / y 1/2x, approach zero. Physically, a circulatory flow is created
in the bracelet shape in the main chamber encompassing the jet
flow, as shown in Fig. 11, which plots the mean velocity vector on
the horizontal and vertical planes. This also explains why the tails
of the axial velocity profiles against y / y 1/2x or z / z1/2x are not
self-similar. It is observed in Fig. 10 that vx , y , z and wx , y , z
are of the same order of magnitude and are very smallless than
ucx by a factor of 25as shown in Fig. 10. This implies that a
passive scalar e.g., temperature is mainly transported by axial
flow.
Fig. 9 Variation of half-widths against axial distance in the

self-similar region. Black symbols z1/2x, white symbols
y1/2x, circle 0 = 5 m, upward-facing triangle 0 = 10 m,
and downward-facing triangle 0 = 15 m.
Fig. 10 Variation of cross-stream velocity components with

radial distance in the self-similar region. The left-hand side of
each figure shows plots of v or wx , 0 , z / ucx against z / z1/2x
and the right-hand side shows the variation of v or
wx , y , 0 / ucx against y / y1/2x.
Conclusions
A simple device that is compatible with planar layer-by-layer

geometries and can create several jet flows within is proposed and
numerically studied by computational fluid dynamics. In this
study, four jet flows are independently formed. The principle of
the device is presented.
The axial velocity component is self-similar with a cross-stream
nondimensionlized by the half-widths of the jet flow if the amplitude of vibration of the PZT membrane lies in a suitable range.
This implies that the spread of the jet flow along two perpendicular directions is similar to a turbulent free jet flow. The self-similar
region covers almost the entire main chamber, except in areas
very close to the nozzle or the back-channels.
In the self-similar region, the jet flow is expected to vibrate
freely in three dimensions.
Fig. 11 Time averaged velocity fields on two perpendicular

planes containing the axis of the jet flow in the main chamber
APRIL 2009, Vol. 131 / 041401-5
The two remaining components of the flow are small as compared to the axial component.
Furthermore, the characteristics of the centerline axial velocity,
as well as the half-widths of the jet, should be addressed in a
general way. The device has been investigated only for one specific dimension. Since the geometry, throat, and nozzle may contribute significantly in optimizing the operation of the device,
these factors should be carefully considered in the future.
References
1 Woolley, A. T., Hadley, D., Landre, P., deMello, A. J., Mathies, R. A., and
Northrup, M. A., 1996, Functional Integration of PCR Amplification and
Capillary Electrophoresis in a Microfabricated DNA Analysis Device, Anal.
Chem., 68, pp. 40814087.
2 Taylor, M. T., Nguyen, P., Ching, J., and Petersen, K. E., 2003, Simulation of
041401-6 / Vol. 131, APRIL 2009
4
5
6
7
8
Microfluidic Pumping in a Genomic DNA Blood-Processing Cassette, J. Micromech. Microeng., 13, pp. 201208.
Zhang, L., Koo, J. M., Jiang, L., Asheghi, M., Goodson, K. E., Santiago, J. G.,
and Kenny, T. W., 2002, Measurements and Modeling of Two-Phase Flow in
Microchannels With Nearly Constant Heat Flux Boundary Conditions, J. Microelectromech. Syst., 11, pp. 1219.
Dau, V. T., Dao, V. D., Shiozawa, T., Kumagai, H., and Sugiyama, S., 2005,
A Dual Axis Gas Gyroscope Utilizing Low Doped Silicon Thermistor, 18th
IEEE MEMS, pp. 626629.
Zhou, J., Yan, G., Zhu, Y., Xiao, Z., and Fan, J., 2005, Design and Fabrication
of a Microfluidic Angular Rate Sensor, 18th IEEE International Conference
on Micro Electro Mechanical Systems MEMS, pp. 363366.
Saito, K., Komatsu, G., and Kato, K., 2001, An Analysis of Fluid Behavior on
Gas Angular Rate Sensor, JSME Int. J., Ser. C, 67, pp. 135140.
Bassi, A., Gennar, F., and Symonds, P. S., 2003, Anomalous Elastic-Plastic
Respond to Short Pulse Loading of Circular Plates, Int. J. Impact Eng., 28,
pp. 6591.
Pope, S. B., 2000, Turbulent Flows, Cambridge University, Cambridge.
Pressurized Flow in a Mesostructured

Silica Modified by Silane Groups
Venkata K. Punyamurtula
Aijie Han
Yu Qiao1
e-mail: yqiao@ucsd.edu
Department of Structural Engineering,
University of California, San Diego,
La Jolla, CA 92093-0085
By applying a quasihydrostatic pressure, water or electrolyte solution can be compressed into a surface treated MSU-H mesoporous silica. Based on the pressure-volume curves, thermodynamic
and kinetic characteristics of the pressurized flow are analyzed.
For pure water based system, continuum theory explains the testing data quite well but fails to capture the rate effect. For electrolyte solution based system, the classic interface theory breaks
down, probably due to the unique ion behaviors in the
nanoenvironment. DOI: 10.1115/1.3089542
Introduction
Liquid behavior in nanoenvironment has been an important research topic for more than a decade 1,2, which shed much light
on fundamental mechanisms governing solid-liquid interactions at
small length and time scales. The studies in this area have been
immensely useful to developing better techniques for drug delivery, micro-/nanotransportation, micro-/nanofabrication, biosensing, etc.
Over the years, energy dissipation associated with nanofluidic
behaviors has drawn considerable attention 3,4. The initial idea
was based on the assumption that, as a liquid flows in nanopores
and the internal friction takes place across the large pore surface,
the energy dissipation capacity can be ultrahigh. Such a system
may have great applicability in advanced dampers, bumpers, and
armors. For instance, in an experiment performed by Li 5, bending moments were applied on a hydrophilic nanoporous silica
block soaked by water. As water flowed from compressive parts to
tensile parts, there was a detectable damping effect. However, the
energy dissipation efficiency was lower than the expected level.
Recently, an improved concept has been investigated by a few
research teams 611, including the authors of this paper, who
replaced the hydrophilic nanoporous solids by hydrophobic ones.
Due to the capillary effect, as a hydrophobic nanoporous material
is immersed in water, the nanopores remain empty. Once a high
pressure is applied on the liquid phase, pressure induced infiltration PII would occur, accompanied by conversion of a significant amount of external work to excess solid-liquid interfacial
energy. In a number of nanoporous systems, the excess solidliquid interfacial energy cannot be converted back to the hydrostatic pressure as the external loading is removed; that is, the
sorption isotherm curves are hysteretic, which results in ultrahigh
energy dissipation efficiency larger than a few J/g, much better
than many conventional protective materials 12. A similar design concept can be extended to nonpolar liquids using lyophobic
1
JOURNAL OF FLUIDS ENGINEERING. Manuscript received February 23, 2007; final manuscript received March 11, 2008; published online March 6, 2009. Assoc. Editor: Ali
Beskok.
nanoporous materials, which has led to an even higher energy

density 13,14. In these investigations, it has been noticed that
increasing degree of hydrophobicity, decreasing pore size, and
adding electrolytes can be of beneficial effects to improving system performance.
Experimental Procedure
In the current study, we investigated a MSU-H type nanoporous

silica that was commonly used as adsorbent and catalyst support.
It was of a hexagonal framework, small pore size, and relative
large specific pore volume, ideal for energy absorption applications. The network material was obtained from Sigma-Aldrich St.
Louis, MO CAT No. 643637. The as-received material was in
powder form. The particle size was 20 80 m. According to a
gas absorption analysis using a Tristar-3000 analyzer, the average
pore size was 7.1 nm, the specific pore volume was 0.9 cm3 / g,
and the specific pore surface area was 510 m2 / g. Figure 1 shows
the BarrettJoynerHalenda BJH analysis result of nanopore
volume distribution. Its pore surface was hydrophilic, and therefore could not be directly used for PII testing. In order to modify
the pore surface structure, the silica particles were first dried at
100 C in a vacuum oven for 24 h and then refluxed in a silane
agent solution in a hot mantle. The solution was prepared by injecting 1 part of chlorotrimethylsilane into 40 parts of dry toluene.
Prior to the refluxing, the mixture had been gently stirred at room
temperature for 10 min. The refluxing time ranged from 3 h to 24
h. During this process, the chlorotrimethylsilane molecules diffused into the nanopores and formed hydrophobic O SiCH33
groups at nanopore walls, changing the surface wettability 15.
The treated sample was vacuum filtered, washed with dry toluene,
and dried at 100 C for 24 h in vacuum.
Figure 2 depicts the PII experimental setup, which consisted of
0.2 g of surface treated silica particles immersed in 7 g of liquid,
a steel cylinder, and a piston. A compressive load was applied on
the liquid phase through the piston by an Instron machine, forcing
the piston into the cylinder at a constant rate. The liquid phase was
either de-ionized water or saturated aqueous solution of sodium
chloride. The compression rate, V, was 120 mm/min. The quasihydrostatic pressure in the liquid, p, was calculated as P / A0, with
P being the axial load and A0 = 286 mm2 being the cross-sectional
area of the cylinder. The piston displacement, p, was also measured, based on which the liquid volume change was calculated as
V0 = p A0. As the pressure exceeded 3040 MPa, the piston was
moved out of the cylinder at the same rate. Figures 3 and 4 show
the typical sorption isotherm curves. The resolutions of load and
displacement measurements were 250 N and 20 m, respectively.
Without the surface treatment, the silica is hydrophilic and thus

the nanoporous particles can be infiltrated by water. Under this
condition, as shown by the dashed curve in Fig. 3, no infiltration
can be observed in the PII experiment. With the surface treatment,
as shown by the solid curves, infiltration plateaus are formed. The
infiltration plateau is defined as the middle portion of the loading
part of sorption isotherm where the slope is much smaller than
that of the linear compression sections. The surface treated nanoporous silica particles are initially filled by entrapped air. As the
pressure is sufficiently high 4.5 MPa and 8.1 MPa for silicas
surface treated for 3 h and 24 h, respectively, liquid starts to enter
the largest nanopores. As a result, the system volume decreases
rapidly. As the pressure rises, smaller nanopores are involved in
the PII process, during which the entrapped gas molecules are
dissolved by the confined liquid. Note that the interactions among
gas and liquid molecules in a nanopore can be fundamentally
different from that in a bulk phase, as demonstrated by the molecular dynamics simulation in Ref. 16. As the nanopore wall
confines the molecular motion, there must be a sufficiently large
gap between the gas molecule and the solid surface; otherwise,
APRIL 2009, Vol. 131 / 044501-1
Pressure (MPa)
Sodium Chloride
Solution Based System
Pure Water
Based System
Specific System Volume Change (cm3/g)
Fig. 1 The BJH analysis result of the nanopore volume

distribution
the liquid molecules would be blocked. When the pressure is lowered, since gas phase nucleation may be energetically unfavorable,
the confined liquid does not come out of the nanopores 17; that
is, a decrease in solid-liquid interfacial energy associated with the
expansion of a nanobubble can be lower than the energy required
for evaporating molecules from liquid phase to vapor phase,
which causes a significant hysteresis in sorption isotherm. The
enclosed area of the loading-unloading cycle indicates the dissipated energy Table 1.
According to Fig. 3, the infiltration volume of the silica surface
treated for 3 h, which can be measured by the width of infiltration
Fig. 2 A schematic of the testing system
Fig. 4 The effect of addition of sodium chloride. The silica is

surface treated for 3 h.
plateau, is about 0.7 cm3 / g, smaller than the BJH measurement

result of nanopore volume of untreated silica. Note that the molecular size of O ClCH33 is around 0.26 nm. As the nanopore
surface is covered by a surface group layer, the effective pore size
should be reduced from 7.1 nm to 6.6 nm, and the effective pore
volume would decrease by about 15%, from 0.9 cm3 / g to
0.76 cm3 / g, quite close to the infiltration volume. The effective
solid-liquid interfacial tension, according to the classic Youngs
equation, can be calculated as = pin r / 2 = 13 mJ/ m2, with pin
being taken as 7.9 MPa and r = 3.3 nm.
As the surface treatment time, tr, is increased to 24 h, the infiltration pressure largely rises to 12.4 MPa, while the infiltration
volume decreases considerably from 0.7 cm3 / g to 0.4 cm3 / g,
even though the nanoporous structure does not change. This phenomenon may be attributed to the formation of dense layers of
silane groups at nanopore surfaces. When the surface coverage is
saturated, in a prolonged surface treatment process, chlorotrimethylsilane molecules would continue to diffuse into the nanopores.
Consequently, the pore volume decreases. The reduction in accessible pore volume from 0.7 cm3 / g to 0.4 cm3 / g is equivalent to
a change in effective nanopore radius from 3.3 nm to 2.5 nm. With
the same solid-liquid interfacial tension of 13 mJ/ m2, the infiltration pressure would increase from 7.9 MPa to 10.5 MPa, which is
quite close to the values shown in Fig. 3. That is, the large variation in infiltration pressure and volume associated with the increase in treatment time are caused by the decrease in empty
nanopore space.
If the liquid motion in nanopores could be described by a circular Poiseuille flow, as the infiltrating liquid molecules move
along the pore wall, an additional pressure must be applied to
overcome the viscosity effect. For a zero-order approximation, the
pressure increase can be assessed as dp = 32 v dz / D2 18,
where = 103 Pa s is the liquid viscosity, v is the inflow rate, dz
is the pore depth, and D = 6.6 nm is the effective pore diameter
for the silica sample that is treated for 3 h. In the current study, the
cross-sectional area of the testing cylinder, A0, is 286 mm2, and
therefore it takes about 30 s to fill the nanopores. The average
silica particle size is about 50 m. Thus, the inflow rate, v, is
Table 1 Energy absorption capacity of surface treated MSU-H
silica J/g
Surface treatment time h
Experimental condition
Fig. 3 Typical sorption isotherm curves of pure water based

system. The specific system volume change is defined as the
system volume change normalized by the mass of the nanoporous silica.
044501-2 / Vol. 131, APRIL 2009
In pure water
v = 1 mm/ min In saturated NaCl solution
v = 20 mm/ min in pure water
12
24
5.1
6.8
5.2
5.6
7.0
5.6
3.1
3.6
3.2
around 1 m / s, and the characteristic value of dz is about

10 m. Consequently, when the piston speed changes from 1
mm/min to 20 mm/min, the sorption isotherm curve should shift
along the pressure axis by a few MPa, which was not observed in
the PII experiment. When the loading rate varied in this range, the
infiltration plateaus were nearly identical. Clearly, in the confining
nanoenvironment, the no-slip boundary condition is no longer
valid, and the definition of liquid viscosity breaks down, which
explains why the internal-friction-based energy dissipation
mechanism did not have the expected energy density 5.
As sodium chloride is added in the liquid phase, as shown in
Fig. 4, while the infiltration volume does not vary, the infiltration
pressure becomes much higher. The pin increase is about 2.2 MPa,
which corresponds to a large variation in effective solid-liquid
interfacial tension of about 8.6 mJ/ m2, 66% of that of pure water.
Note that the surface tension of saturated sodium chloride solution
is larger than that of pure water by less than 10% 19. The large
increase in the effective interface tension should be related to the
confinement effect of pore walls on ion distribution. In a nanopore, normal to the pore surface, there is no enough space for the
formation of a double layer. The bulk phase may not exist. Therefore, the effective ion concentration is higher than that at a large
solid surface; that is, the ion structure is affected by the opposite
nanopore surface, causing a squeezing effect, which can lead to
a higher liquid composition sensitivity. The details of the fundamental mechanisms are still under investigation.
Conclusion
To summarize, through a pressure induced infiltration experiment on a surface treated MSU-H nanoporous silica, the liquid
motion in nanopores are examined. For the system under investigation, the classic capillary theory can be employed to relate the
infiltration pressure to the nanopore size. However, the continuum
theory does not explain the low sensitivity of sorption isotherm to
loading rate. It also fails to capture the large variation in infiltration pressure caused by addition of sodium chloride.
Acknowledgment
This work was supported by the Army Research Office under
Grant No. W911NF-05-1-0288, for which the authors are grateful
to Dr. D.M. Stepp.
References
1 Sansom, M. S. P., and Biggin, P. C., 2001, Biophysics: Water at the Nanoscale, Nature London, 414, pp. 156158.
2 Majumder, M., Chopra, N., Andrews, R., and Hinds, B. J., 2005, Experimental Observation of Enhanced Liquid Flow Through Aligned Carbon Nanotube
Membranes, Nature London, 438, pp. 4446.
3 Borman, V. D., Grekhov, A. M., and Troyan, V. I., 2000, Investigation of the
Percolation Transition in a Nonwetting Liquid-Nanoporous Medium System,
J. Exp. Theor. Phys., 91, pp. 170181.
4 Fadeev, A. Y., and Eroshenko, V. A., 1997, Study of Penetration of Water Into
Hydrophobized Porous Silicas, J. Colloid Interface Sci., 187, pp. 275282.
5 Li, J. C. M., 2000, Damping of Water Infiltrated Nanoporous Glass, J. Alloys
Compd., 310, pp. 2429.
6 Martin, T., Lefevre, B., Brunel, D., Galarneau, A., Di Renzo, F., Fajula, F.,
Gobin, P. F., Quinson, J. F., and Vigier, G., 2002, Dissipative Water Intrusion
in Hydrophobic MCM-41 Type Materials, Chem. Commun. Cambridge,
2002, pp. 2425.
7 Borman, V. D., Belogorlov, A. A., Grekhov, A. M., Lisichkin, G. V., Tronin, V.
N., and Troyan, V. I., 2005, The Percolation Transition in Filling a Nanoporous Body by a Nonwetting Liquid, J. Exp. Theor. Phys., 100, pp. 385
397.
8 Surani, F. B., Kong, X., and Qiao, Y., 2005, Two-Staged Sorption Isotherm of
a Nanoporous Energy Absorption System, Appl. Phys. Lett., 87, p. 251906.
9 Surani, F. B., Kong, X., Panchal, D. B., and Qiao, Y., 2005, Energy Absorption of a Nanoporous System Subjected to Dynamic Loadings, Appl. Phys.
Lett., 87, p. 163111.
10 Kong, X., and Qiao, Y., 2005, Improvement of Recoverability of a Nanoporous Energy Absorption System by Using Chemical Admixture, Appl.
Phys. Lett., 86, p. 151919.
11 Kong, X., Surani, F. B., and Qiao, Y., 2005, Effects of Addition of Ethanol on
the Infiltration Pressure of a Mesoporous Silica, J. Mater. Res., 20, pp. 1042
1045.
12 Lu, G., and Yu, T., 2003, Energy Absorption of Structures and Materials, CRC,
Boca Raton, FL.
13 Punyamurtula, V. K., and Qiao, Y., 2007, Hysteresis of Sorption Isotherm of
a Multiwall Carbon Nanotube in Paraxylene, Mater. Res. Innovations, 11, pp.
3739.
14 Punyamurtula, V. K., Han, A., and Qiao, Y., 2007, Damping Properties of
Nanoporous Carbon-Cyclohexane Mixtures, Adv. Eng. Mater., 9, pp. 209
211.
15 Lim, M. H., and Stein, A., 1999, Comparative Studies of Grafting and Direct
Syntheses of Inorganic-Organic Hybrid Mesoporous Materials, Chem. Mater.,
11, pp. 32853295.
16 Qiao, Y., Cao, G., and Chen, X., 2007, Effects of Gas Molecules on Nanofluidic Behaviors, J. Am. Chem. Soc., 129, pp. 23552359.
17 Han, A., Kong, X., and Qiao, Y., 2006, Pressure Induced Infiltration in Nanopores, J. Appl. Phys., 100, p. 014308.
18 Fay, J. A., 1994, Introduction to Fluid Mechanics, MIT, Cambridge, MA.
19 Hartland, S., 2004, Surface and Interface Tension, CRC, Boca Raton, FL.
APRIL 2009, Vol. 131 / 044501-3

J Fluids Engineering 2009 Vol 131 N4

Uploaded by

Copyright:

Available Formats

J Fluids Engineering 2009 Vol 131 N4

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

J Fluids Engineering 2009 Vol 131 N4

Uploaded by

Copyright:

Available Formats

Journal of

VOLUME 131 NUMBER 4 APRIL 2009

Considerations on a Mass-Based System Representation

Dynamic Model and Numerical Simulation for Synchronal Rotary

Evaluation of the Dynamic Transfer Matrix of Cavitating Inducers

RANS Analyses of Turbofan Nozzles With Internal Wedge Deflectors

Fundamental Issues and Canonical Flows

Choking Phenomena in a Vortex Flow Passing a Laval Tube: An

Pressure Drop in Rectangular Microchannels as Compared With

Managing Director, Publishing

Numerical Modeling of Laminar Pulsating Flow in Porous Media

Aerodynamic Analysis of a Vehicle Tanker

Free Surface Model Derived From the Analytical Solution of Stokes

Transactions of the ASME, Journal of Fluids Engineering

Numerical and Experimental Investigations of Steam Condensation

The Effects of Inlet Geometry and Gas-Liquid Mixing on Two-Phase

Techniques and Procedures

A Principle to Generate Flow for Thermal Convective Base Sensors

Volume 131, Number 4

Pressurized Flow in a Mesostructured Silica Modified by Silane Groups

The ASME Journal of Fluids Engineering is abstracted and indexed in

Introduction and Review

Journal of Fluids Engineering

ditions, the total mass of gas within a pneumatic actuator is

Copyright 2009 by ASME

APRIL 2009, Vol. 131 / 041101-1

Table 2 Simulation system properties, from Ref. 21

Factory modified to accept a 10 V command signal.

Fig. 1 Servopneumatic system

ing environment. Pneumatic actuators were chosen over other

Parker 02.00 HMAU 34A

Stroke shortened to accommodate 300 mm length of potentiometer.

The derivative of the internal energy with respect to time, when

The derivation of Eqs. 2 and 3 assumes that the gas undergoes

There has been considerable discussion in previous works as to

041101-2 / Vol. 131, APRIL 2009

Transactions of the ASME

where the critical pressure ratio

separates the choked and unchoked regimes of flow through the

Less familiar through the literature is the Instrument Society of

becomes more attractive than single-orifice models. Using Eq. 7,

The mechanical dynamics of the cylinder may be expressed as

Equation 10 uses English units of measure, as is the convention

downstream pressures, and the length and sizing of the plumbing

Fig. 2 Valve flow models

APRIL 2009, Vol. 131 / 041101-3

Table 3 Comparative energy rates

System from Ref. 21

Rod end down

Energy rate from viscous friction, EF = bx2

The principal benefit of a mass-based system representation is that

For the rod end,

The forces acting on the cylinder piston are balanced,

Combining Eqs. 1719,

Using the representative values found in Table 3, the error term e

This demonstrates that an assumption of a constant mass of gas in

Table 4 Experimental trajectory properties

Trapezoidal command position, repeating

velocities from 6 mm/s to 890 mm/s and accelerations to