Three-dimensional rarefied flows in rotating helical channels

Young-Kyu Hwang and Joong-Sik Heo

Citation: Journal of Vacuum Science & Technology A 19, 662 (2001); doi: 10.1116/1.1350979
View online: http://dx.doi.org/10.1116/1.1350979
View Table of Contents: http://scitation.aip.org/content/avs/journal/jvsta/19/2?ver=pdfcov
Published by the AVS: Science & Technology of Materials, Interfaces, and Processing

Articles you may be interested in

Three-dimensional Lagrangian transport phenomena in unsteady laminar flows driven by a rotating sphere
Phys. Fluids 25, 093602 (2013); 10.1063/1.4819901

Gaseous Slip Flow in ThreeDimensional Uniform Rectangular Microchannel

AIP Conf. Proc. 1333, 742 (2011); 10.1063/1.3562735

Direct simulation of rarefied gas flows in rotating spiral channels

J. Vac. Sci. Technol. A 20, 906 (2002); 10.1116/1.1472418

Numerical and experimental investigations of channel flows in a disk-type drag pump

AIP Conf. Proc. 585, 903 (2001); 10.1063/1.1407655

Molecular transition and slip flows in rotating helical channels of drag pump
AIP Conf. Proc. 585, 893 (2001); 10.1063/1.1407653

Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13
 Three-dimensional rarefied flows in rotating helical channels
Young-Kyu Hwanga) and Joong-Sik Heob)
School of Mechanical Engineering, Sungkyunkwan University, 300 Chunchun-dong, Jangan-ku, Suwon,
440-746, South Korea
共Received 27 July 2000; accepted 8 January 2001兲
Numerical simulations of rarefied gas flows in pumping channels of a helical-type drag pump
共HTDP兲 are carried out by using the direct simulation Monte Carlo 共DSMC兲 method. Performance
curves obtained by the DSMC method in the slip flow regime are compared with those obtained by
the Navier–Stokes equations with slip boundary conditions. Satisfactory agreement between the two
results is achieved. Also, the previously developed diffusion model 关J. Vac. Sci. Technol. A 13,
2592 共1995兲兴 is modified to predict the performance of a HTDP. Experimental data are presented in
the pressure range of 0.02–4 Torr. The numerical results obtained by the modified diffusion model
are in good agreement with the experimental data over the entire pressure range from the molecular
to the slip flow. © 2001 American Vacuum Society. 关DOI: 10.1116/1.1350979兴

I. INTRODUCTION and good agreement between numerical and experimental

A turbo-molecular pump 共TMP兲 operates efficiently at
very low gas density. Under the condition, it has high com-
pression ratio and large pumping speed. Today, molecular
drag stages are widely used in conjunction with a TMP in
order to improve the pumping performance of it.1 In the
present study, the pumping performance of a helical-type
drag pump 共HTDP兲 is numerically and experimentally ana-
lyzed.
Nanbu et al.2 numerically studied the molecular transition
and slip flows in pumping channels of a HTDP by using the
direct simulation Monte Carlo 共DSMC兲 method. They found
that most of the pressure rise occurs near the outlet of the
channel. Panos, Antoniou and Valamontes3 defined the sepa-
ration pressure between the viscous and the free molecular
state, and they extended Gaede’s equation, which was used
for the helical single-groove drag pump in the free molecular
flow, to the helical multi-groove drag pump in both the vis-
cous and free molecular states. Sawada and Sugiyama4 ana-
lyzed the pumping mechanism by solving the Navier–Stokes
共NS兲 equations with first-order slip boundary conditions
共SBC兲. Valamontes and co-workers5 developed a HTDP,
which is able to compress directly to the atmosphere, by
examining the behavior of the viscosity coefficient. How-
ever, the previous studies2–5 were limited to the quasi three-
dimensional 共Q3D兲 model of Couette flow. Namely, the
three-dimensional 共3D兲 rotating helical channels were con-
sidered as a rectangular groove facing a wall, which is lin-
early moving along the groove; see Fig. 1.
In the recent study of Niu6 and Cheng et al.,7 the rotating
helical channels were three-dimensionally modeled to inves-
tigate the curvature effect of channels, and a laminar com-
pressible NS equation with first-order SBC was used. Jou
et al.8 developed a new HTDP, which works in the outlet
pressure range up to 1000 Pa. Pumping performance of the
pump was predicted by using the DSMC and NS methods,
a兲
Professor; electronic mail: ykhwang@yurim.skku.ac.kr
b兲
Research associate.
results was shown.
The purpose of the present study is to understand the
characteristics of the molecular transition flow field existing
within rotating channels of a HTDP. The DSMC method
using the 3D model is employed to investigate the effect of
centrifugal and Coriolis forces on the pumping performance.
The DSMC results provide information on the velocity and
density fields. Also, the formulation of second-order SBC is
utilized in the NS method. A particular attention is paid to
matching the DSMC method with the solutions obtained by
the NS method in the slip flow regime. Our experimental
results are presented in the pressure range of 0.02–4 Torr
and are compared with numerical ones.
For a Gaede-type drag pump 共see Fig. 2兲, a simple ana-
lytical diffusion model of rarefied channel flow was devel-
oped by Helmer and Levi.9 Predictions of their model agree
well with the published Knudsen’s data for capillary leaks in
the molecular transition flow, in addition to giving a good
account of the Knudsen minimum. In this study, the diffu-
sion model of Helmer and Levi is modified to predict the
performance of a HTDP.
II. Q3D AND 3D MODELS
Helical channels 共Q3D兲 are cut in the axial direction and
expanded in the circumferential direction, as seen in Fig. 3.
Two different kinds of rotors are illustrated in Figs. 3共a兲 and
1共b兲, respectively. In these figures, ␣ is the helix angle, w the
channel width, b the ridge width, h the channel depth, and ␦
the clearance between rotor and stator. The Knudsen number
is defined as Kn⫽␭/h, where ␭ is the mean free path of
molecules. The geometrical dimensions of channels are
given in Table I. The channel depth of rotor 1 is 4 mm, while
that of rotor 2 is continuously varied from 8 mm 共at inlet兲 to
4 mm 共at outlet兲. Also, the channel shape of rotor 2 consists
of two parts, i.e., ␣⫽40° for the upper part of channel and
␣⫽15° for the lower part of channel, respectively.
We present the 3D model to show the effect of the cen-
trifugal and Coriolis forces on the pumping performance.
Figures 4共a兲 and 4共b兲 illustrate the computational grid of the
3D channel for rotors 1 and 2, respectively.

662 J. Vac. Sci. Technol. A 19„2…, MarÕApr 2001 0734-2101Õ2001Õ19„2…Õ662Õ11Õ$18.00 ©2001 American Vacuum Society 662

Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13
 663 Y.-K. Hwang and J.-S. Heo: Three-dimensional rarefied flows 663

FIG. 1. Geometry of HTDP: 共a兲 HTDP; 共b兲 simplified

model of a HTDP.

III. COMPUTATIONAL METHODS field is sampled every five time steps during 20 000 time
steps after reaching the steady state. At the steady state of the
A. DSMC simulations
simulation, the total number of simulated particles in the
The DSMC method is a particle-based numerical tech- flow domain is about 2⫻105 .
nique pioneered by Bird.10 It computes trajectories of a large In the case of the 3D model, it is necessary to calculate
number of particles and calculates macroscopic quantities by trajectories of particles on the rotating frame of reference.
sampling particle properties. A computational cell network
representing physical space is first set up for the requirement
of the selection of collision pairs and the sampling of micro-
scopic flow properties. A rectangular grid system is used in
the Q3D model, as seen in Fig. 5. In the case of the 3D
model, the computational grid consists of deformed hexahe-
dral cells, i.e., the four comers of a cell face may not be
coplanar. Therefore, each cell is subdivided into five tetrahe-
dral subcells, which have four well-defined triangular faces,
to enable the tracking of the particles from one cell to an-
other; see Fig. 4. In most of our calculations, the number of
cells are 40⫻20⫻10 in the x, y, and z directions, respec-
tively. More grid points are placed near the outlet region of
the channel to handle large pressure gradients.
Both wall and gas particle passing through the pumping
channels are assumed to be in an isothermal 共300 K兲 state. In
the present simulation, the gas is nitrogen. The null-collision
technique of Koura11 is adopted to determine the number of
collision pairs selected within a cell, together with the vari-
able soft sphere model of Koura and Matsumoto.12 The flow

FIG. 2. Gaede-type molecular drag pump: 共a兲 cross section of a commercial

pump containing Gaede-type drag stage; 共b兲 model of a Gaede pump. FIG. 3. Geometry of HTDP 共Q3D model兲: 共a兲 rotor 1; 共b兲 rotor 2.

JVST A - Vacuum, Surfaces, and Films

Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13
 664 Y.-K. Hwang and J.-S. Heo: Three-dimensional rarefied flows 664

TABLE I. Geometrical dimensions of helical-type rotors.

Rotor 1 2

Rotor diameter D 168.6 mm 168.6 mm

Rotor axial L⬘ 132.0 mm L 1 ⫽66.0 mm
length L 2 ⫽66.0 mm
Helix angle ␣ 15° ␣ 1 ⫽40°
␣ 2 ⫽15°
Channel width w 13.6 mm ␻ 1 ⫽42.2 mm FIG. 5. Computational cells for Q3D model.
␻ 2 ⫽13.6 mm
Ridge width b 5.4 mm b 1 ⫽5.4 mm
b 2 ⫽5.4 mm along the x axis, c x , is independent of the frame of refer-
Channel depth h 4.9 mm h 1 ⫽8.9 mm ence, i.e., c x is the same for both moving and nonmoving
h 2 ⫽4.9 mm
Clearance ␦ 0.9 mm 0.9 mm
frames of reference.
No. of grooves Nb 7 7 At the end of the simulation, the dimensionless pumping
Rotational speed rpm 24 000 24 000 efficiency w, pumping speed S(l/s), and throughput
Q(Torr•l/s) are calculated by
N 12⫺N 21
w⫽ , 共2兲
Thus the fourth-order Runge-Kutta algorithm is used to inte- N inlet

S⫽w•A 1 • 冑8RT/ ␲ •K 共 s 1 兲 /4.0,

grate the following equation including the Coriolis and cen-
共3兲
trifugal forces:
where
a r ⫽⫺ ␻ ⫻ 共 ␻ ⫻r mol兲 ⫺2 ␻ ⫻V mol 共1兲
K 共 s 1 兲 ⫽exp共 ⫺s 21 兲 ⫹ 冑␲ s 1 关 1.0⫹erf共 s 1 兲兴 ,
in which a r is the relative acceleration, r mol and V mol are the
trajectory and the relative velocity of particles, and ␻ is the s 1 ⫽u 1 / 冑2RT
angular velocity of the rotor. From Eq. 共1兲 we can find po-
sitions and velocities of particles at any time on the y⫺z and
plane; see Fig. 4. The component of the particle velocity Q⫽ P 1 •S 共4兲
in which N 12 共or N 21) is the number of particles to be trans-
mitted through the channel from the inlet 共or outlet兲 to the
outlet 共or inlet兲, and N inlet is the number of particles coming
from the inlet during sampling time.2 The quantity u 1 is the
x component of the bulk velocity crossing the open bound-
ary.

B. N-S simulations
In a molecular transition regime, the failure of the NS
solutions stems from the continuum assumption of the NS
equations, which can be derived from the Boltzmann equa-
tion based on the Chapman–Enskog expansion of the veloc-
ity distribution function f. Naturally, the NS equations break
down in the molecular transition regime. Therefore the con-
ventional no-slip boundary conditions are necessary to refor-
mulate.
In the slip flow regime, it has generally been recognized
that the layer of gas adjacent to the wall is no longer at rest
but has a finite slip velocity. Thus the continuum assumption
is not valid near the wall. Sawada and Sugiyama4 showed
that the NS equations can be solved numerically for a HTDP
in the slip flow regime if the boundary conditions are modi-
fied to include a slip velocity at the wall. The slip velocity on
the wall surfaces is given by

u s ⫽u w ⫹
2⫺F
F
Kn
⳵u
⳵n 冉 冊 w
, 共5兲

FIG. 4. Computational grid system of 3D helical channel: 共a兲 rotor 1; 共b兲 where F is the accommodation coefficient, u s and u w refer to
rotor 2. the slip velocity and the wall velocity, respectively. In the

J. Vac. Sci. Technol. A, Vol. 19, No. 2, MarÕApr 2001

Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13
 665 Y.-K. Hwang and J.-S. Heo: Three-dimensional rarefied flows 665

FIG. 7. Comparison of present numerical results with Nanbu et al.2 at clear-

ance ␦⫽0.88 mm and outlet pressure P 2 ⫽0.3 Torr.
FIG. 6. Schematic diagram of experimental apparatus.
In a similar way, the diffusion coefficient D v for a viscous
flow is derived as

above, n is the nondimensional coordinate perpendicular to C 2 共 w⫹h 兲

the channel wall that is normalized by the channel depth h,
and subscript w indicates that the variable is evaluated at the
wall. For continuum flow, Kn→0, Eq. 共5兲 reduces to the
familiar no-slip condition.
In the present study, the formulation of second-order SBC
is utilized in the NS method. For the second-order modifica-
tion of Eq. 共5兲, the following slip velocity can be derived in
the case of F⫽1:
D v ⫽D m , 共10兲
24C 1 ␭
where C 2 is a correction factor.13
Helmer and Levi9 derived a simple diffusion model for a
molecular transition flow in circular tubes of a Gaede-type
drag pump. They assumed that a pumping leak exists only at
the stripper and that the average surface velocity in the strip-
per equals u d , which is the average surface velocity in the
pumping channel. Then their leak term, q⫽n 2 u d A l /A p ,

u s ⫽u w ⫹Kn 冉 冊
⳵u
⳵n w
⫹ 冉 冊
Kn2 ⳵ 2 u
2 ⳵n2 w
. 共6兲
where A l and A p are the stripper and pumping channel area,
is reasonable in the case of the Gaede-type drag pump.
In the present case of a HTDP, as seen in Fig. 1, the
pumping leak continuously exists from the inlet to the outlet
C. Diffusion model along the clearance ␦ between the rotor and the stator. Thus
A diffusion velocity v , which is proportional to the gra- the leak term is modified as follows:
dient of the number density n, can be represented by a fol-
lowing diffusion equation: q⫽ 冕 0
L
nu d ␦ bds, 共11兲
1 dn
v ⫽⫺D , 共7兲
n ds
where D is the diffusion coefficient, and s is the coordinate
along the channel; see Fig. 1共b兲.
In a free molecular flow, the diffusion coefficient D m can
be derived as follows. For a tube of rectangular cross section
having a depth h and a width w, the throughput Q is ex-
pressed in terms of the pressure gradient d P/ds

Q⫽C 1 冉 冊
8 RT
3 2␲
1/2
w 2h 2
w⫹h
⫺ 冉 冊
dP
ds
, 共8兲

in which C 1 is a correction factor.13 Dividing Eq. 共8兲 by the

product of pressure and cross sectional area, PA, we obtain
Q 1 dn
⫽ v m ⫽⫺D m 共9兲
PA n ds
in which v m is the diffusion velocity for a free molecular FIG. 8. Normalized pressure distribution P/ P 1 along pumping channel
flow, D m ⫽2C 1 v 0 wh/ 兵 3(w⫹h) 其 , and v 0 ⫽ 冑8RT/ ␲ . length x/L at P 1 ⫽0.023 and P 2 ⫽0.1 Torr.

JVST A - Vacuum, Surfaces, and Films

Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13
 666 Y.-K. Hwang and J.-S. Heo: Three-dimensional rarefied flows 666

FIG. 9. Flow field of 3D helical channel for rotor 1 at P 1 ⫽0.023 and P 2

⫽0.1 Torr: 共a兲 velocity vectors; 共b兲 density contours.

where b⫽ ␦ sin ␣/Ap and u d ␦ ⫽uL ⬘ / 兵 2(L ⬘ ⫹ ␦ ) 其 .

Let y be the inverse Knudsen number 1/Kn. Since 1/n␭ FIG. 10. Flow field of Q3D helical channel for rotor 1 at P 1 ⫽0.023 and
equals the collision cross section ␴, we have P 2 ⫽0.1 Torr: 共a兲 velocity vectors; 共b兲 density contours.

1 h
y⫽ ⫽ ⫽n ␴ h. 共12兲

冉 冊冉 冊
Kn ␭
Following Helmer and Levi,9 if u d is balanced by the
diffusion velocity, then Eq. 共7兲 becomes v0 h 再
u d L 3 共 1⫹h/w 兲
2C 1 冎
⫽C 3 y 2 1⫺
1
K
⫹ 1⫹C 3
1⫹C 24 Z̄ 2
2

q
u d ⫺ ⫽ 共 D m f m ⫹D v f v 兲
1 dn
, 共13兲 1 冉 冊 y2
K
⫹1 共 y 2 ⫺C 4 Z̄ 兲

冉 冊
n n ds ⫻ ln
1⫹C 4 Z̄ y2
where f m ⫽␭/(␭⫹h) and f v ⫽h/(␭⫹h)⫽1⫺ f m are the mo- 共 y 2 ⫹1 兲 ⫺C 4 Z̄
K

冉 冊冉 冊
lecular and viscous weighting functions, respectively.
From the combination of Eqs. 共11兲, 共12兲, and 共13兲, we y2 y2
⫹1 ⫺C 4 Z̄
have 1⫺C 4 Z̄ K K
⫹C 3
冉 冊
ln ,
1 y 2
dy 2 共 y 2 ⫹1 兲共 y 2 ⫺C 4 Z̄ 兲

再 冎
⫹C 3 共15兲
u d 3 共 w⫹h 兲 1⫹y 1⫹y ds

冉 冊
⫽ , 共14兲
v 0 2C 1 wh u d ␦ 兰 L0 nds
␴ hn 1⫺b where C 4 ⫽bu d ␦ /u d and Z̄⫽ 兰 L0 yds. If the geometrical pa-
ud n
rameters, such as L/h and h/w, are given, we can compute
in which C 3 ⫽C 2 (w/h⫹1)/24C 1 . the compression ratio K(⫽y 2 /y 1 ⫽ P 2 / P 1 ) with a root-
We integrate from s⫽0 to L, and y 1 to y 2 finding algorithm.

J. Vac. Sci. Technol. A, Vol. 19, No. 2, MarÕApr 2001

Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13
 667 Y.-K. Hwang and J.-S. Heo: Three-dimensional rarefied flows
FIG. 11. Flow field of 3D helical channel for rotor 2 at P 1 ⫽0.023 and P 2
⫽0.1 Torr: 共a兲 velocity vectors; 共b兲 density contours.
IV. EXPERIMENTAL APPARATUS
Experiments on the pumping characteristics of a HTDP
are carried out by using the test pump to verify the present
numerical results. Dimensions of the helical-grooved rotor of
the test pump are listed in Table I. The rotor is 168.6 mm in
diameter and 132 mm in axial length. It has seven threaded
rectangular grooves. A schematic diagram of the experimen-
tal apparatus is shown in Fig. 6. This apparatus mainly con-
sists of two parts: the test pump, and instruments that facili-
tate pressure measurements and flow supply into the test
pump. The test pump is connected to a two-stage oil rotary
pump 共970 l/min兲. The pressure in the high-vacuum side is
measured with a Pirani (7.6⫻10⫺4 ⬃7.6 Torr) and Penning
(7.6⫻10⫺8 ⬃7.6⫻10⫺3 Torr) gauge, and the pressure in the
fore-vacuum side is measured with a Pirani gauge. Experi-
ments are performed by varying the outlet pressure P 2 in the
range of 0.02 Torr⭐ P 2 ⭐4 Torr. The rotational speed of the
rotor is 24 000 rpm and is controlled by a frequency con-
JVST A - Vacuum, Surfaces, and Films
Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13
667
FIG. 12. Flow field of 3D helical channel along the rotational axis for rotor
1 at P 1 ⫽0.023 and P 2 ⫽0.1 Torr: 共a兲 velocity vectors; 共b兲 density contours.
verter. Test gas (N2 ) is supplied through a mass flow con-
troller from a regulated high-pressure cylinder.
V. RESULTS AND DISCUSSION
To validate our computer program, the present numerical
results are compared with the previously known numerical
and experimental ones obtained by Nanbu et al.2 The perfor-
mance curves are shown in Fig. 7. The computations are
conducted by employing the DSMC and NS methods by
changing the inlet pressure P 1 for the fixed outlet pressure
P 2 ⫽0.3 Torr. The present DSMC results for the Q3D model
agree well with those of Nanbu et al.2 But the NS results
deviate from those of the DSMC method. The throughput Q
 668 Y.-K. Hwang and J.-S. Heo: Three-dimensional rarefied flows
FIG. 13. Flow field of Q3D helical channel along the rotational axis for rotor
1 at P 1 ⫽0.023 and P 2 ⫽0.1 Torr: 共a兲 velocity vectors; 共b兲 density contours.
increases linearly as the pressure difference ⌬ P(⫽ P 2 ⫺ P 1 )
decreases. The DSMC results for the Q3D model underesti-
mate by as much as 20% in the throughput at ⌬ P
⫽0.1 Torr compared to the experimental data. This discrep-
ancy is partly related to a motion between the rotor and cas-
ing wall, which is assumed to be linear in the Q3D model.
Also, the DSMC results for the 3D model are included in
Fig. 7. In the case of the Q3D model, only the moving effect
of the casing wall on the performance exists; however, in the
3D model, Coriolis and centrifugal forces are dominant ef-
fects along with the curvature of channels. In the present
study, these effects on the pumping performance are consid-
ered. In particular, trajectories of particles are calculated by
integrating Eq. 共1兲. Consequently, the DSMC results using
the 3D model give the best correspondence to the experimen-
tal data.
The variations of pressure along the helical channel at
P 1 ⫽0.023 and P 2 ⫽0.1 Torr are shown in Fig. 8, in which
the pressure and the length are normalized by the inlet pres-
sure P 1 and the axial length of the helical rotor L, respec-
tively. In the case of rotor 1, the pressure is continuously
increased with flow through the channel, and most of the
pressure rise occurs near the outlet. Especially, it can be seen
that the 3D model leads to a sharper rise in pressure near the
outlet compared to the Q3D model. In the case of rotor 2, the
pressure profile becomes flat in the middle of the lower part
of the channel.
To investigate the detailed flow structure, the velocity and
J. Vac. Sci. Technol. A, Vol. 19, No. 2, MarÕApr 2001
Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13
668
FIG. 14. Flow field of 3D helical channel along the rotational axis for rotor
2 at P 1 ⫽0.023 and P 2 ⫽0.1 Torr: 共a兲 velocity vectors; 共b兲 density contours.
density fields for rotors 1 and 2 are illustrated in Figs. 9–11.
The pressure conditions are P 1 ⫽0.023 and P 2 ⫽0.1 Torr.
The velocity data are sampled after the flow field reaches
steady state and are expressed in dimensionless form
c/ 冑2RT, where c and 冑2RT are the mean molecular veloc-
ity and the most probable molecular speed, respectively. Ve-
locity vectors show a backstreaming of particles near the
outlet. For rotor 1, as seen in Figs. 9 and 10, density contours
indicate that the gas density increases rapidly very near the
outlet. This trend is also observed in Fig. 8.
Flow fields on the y⫺z plane are presented in Figs. 12–
14. We present 4 velocity and density profiles in each figure,
respectively, to see the evolution of flow fields along the
rotational axis. In these figures, the numbers 햲, 햳, 햴, and
햵 represent the y⫺z plane corresponding to the axial loca-
tion at x⫽L/4, L/2, 3L/4, and L, respectively. The density
increases along the channel from the inlet to the outlet. Thus
the local Knudsen number Kn decreases with flow. From
these figures, we can see the effect of Kn on the velocity and
 669 Y.-K. Hwang and J.-S. Heo: Three-dimensional rarefied flows
FIG. 15. Inverse Knudsen number effects on conductance through pumping
channel from molecular to slip flow regime.
density fields. Density gradients in the pressure side 共P-S兲 of
flow channel are larger than those in the suction side 共S-S兲,
as seen in Figs. 12共b兲, 13共b兲, and 14共b兲. The velocity fields
are likely to be two-dimensional Couette flow. For rotor 2, a
region of small velocities appears at section 햲 共i.e., x
⫽L/4) and 햳 共i.e., x⫽L/2), respectively.
Figure 15 shows the evolution of the conductance,
C/C fm , versus the inverse Kn, based on the channel depth
and the average pressure P a ⫽( P 1 ⫹ P 2 )/2. Here, C fm is the
free molecular conductance. We obtain a pronounced Knud-
sen minimum of the conductance at Kn⬇1.
The variations of P 1 as a function of Q at P 2 ⫽0.8 and 1.0
Torr are shown in Figs. 16 and 17 for rotors 1 and 2, respec-
tively. The value of P 1 becomes higher as Q increases, and it
depends on P 2 . Comparison between the experimental data
and the DSMC results shows good agreement.
The effect of the outlet pressure on the inlet pressure at
zero-flow condition is shown in Fig. 18 for rotor 1, and the
comparison between the experimental data and the numerical
results is also made. The good agreement between the ex-
perimental data and the DSMC results is shown in this fig-
ure. This clearly reveals the ability of the DSMC method to
FIG. 16. Inlet pressure vs throughput of rotor 1 at P 2 ⫽0.8 and 1.0 Torr.
JVST A - Vacuum, Surfaces, and Films
Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13
669
FIG. 17. Inlet pressure vs throughput of rotor 2 at P 2 ⫽0.8 and 1.0 Torr.
predict the molecular transition flows. Especially, the 3D
model predicts well the experimental data compared to the
Q3D model.
Comparison between the experimental data and the NS
results also shows good agreement in the range of P 2
⭓2 Torr 共in this region, Kn⫽␭ 2 /h, based on the mean free
path ␭ 2 at the outlet and the channel depth h, is in the order
of 10⫺3 ). But the significant discrepancy between the two
results is seen in the range of P 2 ⭐1.14 Torr, i.e., Kn
⭓0.01.
Consideration is now given to the effect of SBC on the
NS results. It is seen that the results obtained by the NS
method with second-order SBC are nearly the same as those
with first-order SBC in the range of P 2 ⬎1 Torr and that the
differences between the two results appear noticeably for
P 2 ⬍0.1 Torr. Although the NS method with second-order
SBC fails to predict quantitatively the experimental data, it
seems that the results obtained by second-order SBC agree
qualitatively with the experimental data compared to those
obtained by first-order SBC.
FIG. 18. Effect of outlet pressure on inlet pressure at zero throughput.
 670 Y.-K. Hwang and J.-S. Heo: Three-dimensional rarefied flows 670

FIG. 19. Performance comparison of rotor 1 with rotor 2.

In general, for higher density flows, since the DSMC

method ceases to be computationally viable, the NS method
is usually much more efficient than the DSMC method.
Thus, in the present analysis of flows which involve both
molecular and slip flow regimes, it is reasonable to apply
both methods. From Fig. 18, it is apparent that the slip flow
analysis of the NS method is inappropriate for Kn⭓0.01,
regardless of boundary conditions. Even so, a smooth transi-
tion in the flow description between the molecular transition
and slip flow regimes can be seen in this figure. In addition,
the NS method simplifies the simulation and saves computa-
tional effort in predicting the performance in the slip flow
regime. The present calculations have been performed by
using double precision on a Silicon Graphics O2 共MIPS
R10000 processor兲 workstation with 128 Mbytes of memory.
FIG. 21. Knudsen number distribution along pumping channel of rotor 1 at
The computation time required for the NS simulation is
h/w⫽0.36 and L/h⫽50: 共a兲 P 2 ⫽15 Torr; 共b兲 P 2 ⫽0.1 Torr.
about 40 s. In contrast, the DSMC simulation takes approxi-
mately 60 h of run time to obtain a converged solution.
The effect of the outlet pressure on the inlet pressure at
zero-flow condition for rotor 2 is illustrated in Fig. 19. For also included in this figure. The NS results presented here are
comparison, the experimental data for rotor 1 in Fig. 18 are those obtained by second-order SBC. It is seen that the maxi-
mum attainable inlet pressure is about 6⫻10⫺5 Torr for rotor
1 and 4.3⫻10⫺5 Torr for rotor 2, respectively. As indicated
in Fig. 18 for rotor 1, the NS results for rotor 2 also deviate
from the experimental data, except the range of P 2
⭓2 Torr. On the other hand, as expected, the DSMC results
give good correspondence to the experimental data.
In this study, the diffusion model of Helmer and Levi9 is
modified to predict the performance of a HTDP. The results
are presented in Figs. 20–24.
In Fig. 20, we repeat Fig. 18, including the numerical
results obtained by the modified diffusion model. Clearly we
can get a good match of the diffusion model with the experi-
mental data throughout the entire pressure range from the
molecular to the slip flow.
The local inverse Knudsen number distribution along the
channel at h/w⫽0.36 and L/h⫽50 is depicted in Fig. 21,
which shows the degree of gas rarefaction from the inlet to
the outlet. The local Knudsen number is based on the chan-
FIG. 20. Inlet vs outlet pressure at zero throughput of rotor 1. nel depth. Figure 21共a兲 is for the P 2 ⫽15 Torr, which is the

J. Vac. Sci. Technol. A, Vol. 19, No. 2, MarÕApr 2001

Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13
 671 Y.-K. Hwang and J.-S. Heo: Three-dimensional rarefied flows
FIG. 22. Compression ratio vs outlet pressure.
case of slip flow. It is seen that the variation of Kn is linear
in this regime. On the other hand, the Kn distribution is
exponential in the molecular transition regime, as shown in
Fig. 21共b兲.
The compression ratio K as a function of P 2 is shown in
Figs. 22–24 for various parameters of the helical channel. In
all cases, the rotational speed is 24 000 rpm. For a Gaede-
type drag pump, Helmer and Levi9 showed that the channel
length has no effect on the compression ratio due to the
pumping leak at the stripper, whereas a longer channel ex-
tends the pump performance to higher outlet pressure. In our
case of a HTDP, however, both K and P 2 increase with the
channel length, as seen in Fig. 22.
Also, Helmer and Levi9 pointed out that the leak-limited
value of the compression ratio results in the very sharp cor-
ner in transition from the molecular to the viscous flow. By
introducing the aperture effect of a circular cylinder, they
obtained the corrected numerical results corresponding to the
FIG. 23. Compression ratio vs outlet pressure.
JVST A - Vacuum, Surfaces, and Films
Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13
671
FIG. 24. Compression ratio vs outlet pressure.
experimental data. As seen in Figs. 22–24, the sharp corner
in transition does not exist for a HTDP.
VI. CONCLUDING REMARKS
Numerical and experimental investigations have been
made for the molecular transition and slip flows in the rotat-
ing helical channels of a HTDP. The computations were con-
ducted by employing the NS and DSMC methods. Experi-
ments were performed by varying the outlet pressure P 2 in
the range of 0.02 Torr⭐ P 2 ⭐4 Torr. The present DSMC and
NS results agree well with the experimental data for Kn
⭐0.01. By using both numerical methods, we obtained a
smooth transition in the flow description between the mo-
lecular transition and slip flow regimes.
The DSMC results for the 3D model were compared with
those for the Q3D model. Of the two models, the present 3D
model gives more accurate results than does the Q3D model.
In the NS method, the formulation of second-order SBC was
utilized. It was found that the results obtained by second-
order SBC agree qualitatively with the experimental data
compared to those obtained by first-order SBC. Velocity and
pressure fields in a pumping channel were obtained by the
DSMC simulations. Velocity vectors show a backstreaming
of particles near the outlet, and the gas density increases
rapidly very near the outlet.
The diffusion model of Helmer and Levi was modified
and successfully applied to the analysis of the performance
of a HTDP. The results indicate that both the compression
ratio and the outlet pressure increase with the channel length.
ACKNOWLEDGMENT
This work was supported by the Brain Korea 21 Project.
1
M. H. Hablanian, in Vacuum Science and Technology: Pioneers of the
20th Century 共AIP, New York, 1994兲.
2
K. Nanbu, H. Kubota, S. Igarashi, C. Urano, and H. Enosawa, Trans. Jpn.
Soc. Mech. Eng., Ser. B 57, 172 共1991兲.
 672 Y.-K. Hwang and J.-S. Heo: Three-dimensional rarefied flows 672

3 8
C. N. Panos, A. G. Antoniou, and S. E. Valamontes, Vacuum 45, 841 R. Y. Jou, H. P. Cheng, Y. W. Chang, F. Z. Chen, and M. Iwane, J. Vac.
共1994兲. Sci. Technol. A 18, 1016 共2000兲.
4
T. Sawada and W. Sugiyama, J. Vac. Sci. Technol. A 17, 2069 共1999兲. 9
J. C. Helmer and G. Levi, J. Vac. Sci. Technol. A 13, 2592 共1995兲.
5 10
S. E. Valamontes, C. N. Panos, and E. S. Valamontes, Vacuum 54, 421 G. A. Bird, Molecular Gas Dynamics and the Direct Simulation of Gas
共1999兲. Flows 共Clarendon, Oxford, 1994兲.
6
Y. Y. Niu, Numer. Heat Transfer, Part A 36, 75 共1999兲. 11
K. Koura, Phys. Fluids 29, 3509 共1986兲.
7
H. P. Cheng, R. Y. Jou, F. Z. Chen, Y. W. Chang, M. Iwane, and T. 12
K. Koura and H. Matsumoto, Phys. Fluids 4, 1083 共1992兲.
Hanaoka, J. Vac. Sci. Technol. A 18, 543 共2000兲. 13
A. Roth, Vacuum Technology 共Elsevier, New York, 1990兲.

J. Vac. Sci. Technol. A, Vol. 19, No. 2, MarÕApr 2001

Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 131.111.185.72 On: Wed, 03 Dec 2014 18:44:13