Beruflich Dokumente
Kultur Dokumente
Three-Dimensional Bristle
Cesare Guardino
Bending in Brush Seals
CD-adapco Group,
London Office, This paper presents a new method for predicting the three-dimensional (3D) bending
200 Shepherd Bush Road, behavior of bristles in brush seals. The model builds on and addresses shortcomings in an
London W6 7NY, UK earlier two-dimensional bending model. The work was motivated by the need to develop
e-mail: cesare@uk.cd-adapco.com a general 3D solid mechanical model that can ultimately be incorporated into CFD
models of flow and heat transfer through brush seals. The iterative method considered
here, which is based on linear beam-bending theory, allows relatively large numbers of
John W. Chew bristles to be considered with arbitrary imposed aerodynamic forces. Bristle-to-bristle
School of Engineering, contacts and deflections are considered, as well as shaft and backing ring contacts. The
University of Surrey, method also allows arbitrary initial bristle packs and lay-angles to be considered, as well
Guildford, as periodic conditions so that the model represents a sector of a brush seal. Other
Surrey GU2 7XH, UK physically important features, such as the so-called 3D-splay and inclined prop effects,
e-mail: j.chew@surrey.ac.uk are also taken into account. The method described here has been incorporated into a new
computer code called SUBSIS (Surrey University Brush Seal Iterative Simulator). Ex-
ample results from this code are presented that show the bending behavior of initially
hexagonally packed brush seals under model imposed pressure loads acting on the bristle
tips. The effects of rotor incursions into the bristle pack, increase of the pressure load, and
changes in the lay-angle and Young’s modulus are also shown. The results illustrate the
expected bending behavior observed in real brush seals. Procedures for coupling SUBSIS
with CFD models are also currently under investigation. 关DOI: 10.1115/1.1850943兴
Introduction porous CFD brush model by Chen et al. 关6兴, and the model was
calibrated using experimental data. Aksit and Tichy 关7兴 used a
Conventional brush seals comprise of a dense pack of bristles commercially available FEM package in attempting to build a
held between a narrow front plate and a backing ring, as shown in three-dimensional 共3D兲 model of a brush seal. Results show hys-
Fig. 1. Over the past 20 years, these advanced seals have emerged teresis effects when rotor interference is relieved during pressure
to be a very promising technology for gas-path sealing applica- load, leaving the bristles ‘‘hung-up’’ due to frictional forces. How-
tions in turbo-engines. However, despite much experimentation ever, no results showing bristle-to-bristle contacts are shown. A
and the fact that brush seals have successfully operated in jet commercial FEM code was also used more recently by Crudging-
engines, their behavior is far from being fully understood, and ton and Bowsher 关8,9兴. Their results, which included bristle-bristle
advances are still required if brush seals are to be more generally contacts and friction, showed that hysteresis and blow-down ef-
fects are important in determining the brush seal characteristics.
used. Wear, which occurs as bristle tips contact the rotating parts,
This paper describes an alternative efficient 3D bristle-bending
is a particular problem. There are, therefore, considerable poten- model incorporating bristle-to-bristle, backing ring, and rotor con-
tial benefits in better understanding and possibly controlling the tacts. The method and algorithm, which are specifically designed
factors that influence wear. This study is aimed at improving the for brush seal applications, have been coded into a new computer
modeling of brush seals and, hence, giving further insight and program, which can be run stand-alone or as a ‘‘plug-in’’ for a
predictive capability for their behavior. CFD code. Circumferential periodicity conditions are imposed in
Previous investigations have established that brush seals offer order for the model to represent a sector of a brush seal, and
improved sealing performance as compared to existing labyrinth arbitrary initial packing arrangements can be considered. Rotor
seals. Useful data, including leakage rates, pressure distributions, incursions into a perfectly packed periodic brush seal 共i.e., hex-
agonal packing with no initial spacing between bristles and the
forces on the rotor, and stiffness measurements, have been ob-
backing ring兲 would cause displacements in the axial direction, as
tained from both CFD and experimental studies 共see, for example, the volume of the pack cannot be reduced further. This 3D-splay
Refs. 关1–3兴兲. Prediction of the effects of bristle bending has also effect is also included in the model. At present the model neglects
been carried out by Hendricks 关4兴 using a ‘‘bulk flow model.’’ the effects of static and kinetic friction due to the backing ring
These results suggest that abrupt flow and pressure changes occur and/or rotor surface, although these effects are important and
as the rotor clearance changes, but they do not take bristle-to- should be addressed in later versions. Bristle bending predictions
bristle contacts into account. These latter effects have been con- are presented here for a typical brush seal geometry, and qualita-
sidered in the two-dimensional 共2D兲 bending model of Chew et al. tive comparisons with previous results are discussed. Note that,
关5兴. Friction due to the backing ring was later included into this although the brush seal is modeled here as a linear bristle pack,
model by Turner et al. 关3兴. This model was later coupled with a rather than a circumferential one, lengths in the width 共X兲 and
height 共Z兲 directions are referred to as the circumferential and
Contributed by the International Gas Turbine Institute 共IGTI兲 of THE AMERICAN
radial directions, respectively, as shown in Fig. 1.
SOCIETY OF MECHANICAL ENGINEERS for publication in the ASME JOURNAL OF
ENGINEERING FOR GAS TURBINES AND POWER. Paper presented at the Interna- Basic Theory and Notation
tional Gas Turbine and Aeroengine Congress and Exhibition, Vienna, Austria, June
13–17, 2004, Paper No. 2004-GT-53176. Manuscript received by IGTI, October 1, Bristles are identified with the coordinates (X i, j,k ,Y i, j,k ,Z i, j,k )
2003; final revision, March 1, 2004. IGTI Review Chair: A. J. Strazisar. of points along their centerline, and each bristle is subdivided into
Journal of Engineering for Gas Turbines and Power JULY 2005, Vol. 127 Õ 583
Copyright © 2005 by ASME
冦
⫺ for 0⭐ k ⬍ m ,
2 6
k,m ⫽ (6)
k m2 m3
K⫺1 elements, each of length ⌬ ⫽L/(K⫺1). The distance from ⫺ for m ⭐ k ⭐L
the fixed bristle root is denoted by k ⫽(k⫺1)⌬ . A bristle- 2 6
oriented coordinate system (x,y,z) is also considered, and b i, j,k It can also be shown that in order to move a point b i, j,k by an
denotes the point k along the centerline of the bristle with fixed amount s i, j,k , a force of magnitude r i, j,k acting at this point suf-
root at position (x i, j ,y i, j ). This coordinate system is related to the fices, where
global Cartesian coordinate system (X,Y ,Z) using the transforma-
tion matrix 共see Fig. 2兲 Bs i, j,k
r i, j,k ⫽ (7)
冉 冊冉 冊冉 冊
k,k
X i, j,k cos ⫺sin x i, j,k
⫽ (1) The deflection angles ␣ i, j,k and  i, j,k are defined as
Z i, j,k sin cos z i, j,k
共note that Y ⫽y). Here, z is related to the local variable by tan ␣ i, j,k ⫽ 冉 冊
d
d i, j,k
, tan  i, j,k ⫽ 冉 冊
d
d i, j,k
(8)
再
tor兲. This will displace a point such that:
1 if i⫹m⬎I
a. it just contacts the backing ring, rotor, or another bristle, or
b. it moves until the reaction force at the point has been re- ⌼⫽ ⫺1 if i⫹m⬍I (16)
duced to zero. 0 otherwise
The algorithm can be summarized into the following basic steps, The angle i, j,k,m,n shown in Fig. 4 can now be found using the
some of which will be decribed further in the proceeding sections. following equation:
The complete algorithm and the procedures for determining which
reaction forces to correct are presented in more detail in Ref. 关11兴. Ui, j,k "Vi, j,k,m,n ⫽ 兩 Ui, j,k 兩兩 Vi, j,k,m,n 兩 cos共 i, j,k,m,n ⫹ /2兲 (17)
1. Compute the total forces 共i.e., aerodynamic plus reaction It is also useful to define si, j,k,m,n as the vector from the reference
forces兲 on all the bristles and determine their deflections. to a target bristle in local bristle coordinates, i.e.,
2. Calculate resulting deflection, orientation and deformed lay- si, j,k,m,n ⫽ 共 x i tar , j tar ,k ⫺x i, j,k ⫹⌼C cos 兲 i⫹ 共 y i tar , j tar ,k ⫺y i, j,k 兲 j
angles. Apply periodicity conditions.
3. Determine the required corrections to all bristle-bristle, rotor (18)
and backing ring reaction forces for all bristles. We now define Ci, j,k,m,n to be the reaction force acting on a
4. Compute residuals and determine which reaction forces to reference bristle centerline point b i, j,k from a target point
adjust by comparing all the computed reactions forces b i tar , j tar ,k . According to Newton’s third law, an equal and opposite
throughout the brush seal. reaction force must also be imposed on the target bristle b i tar , j tar ,k
5. Scan for and adjust the bristle pair with the greatest reaction
by setting
force surface-angle error 共i.e., bristles for which the reaction
forces do not point exactly normal to the corresponding Ci tar , j tar ,k,⫺m,⫺n ⫽⫺Ci, j,k,m,n (19)
bristle surfaces兲.
6. Apply Newton’s Third Law via Eq. 共19兲. If contact occurs with a target bristle point b i tar , j tar ,k 共i.e., if
7. Go back to Step 1 and repeat until either the maximum of i, j,k,m,n ⬎0), then it can be shown that the corrective reaction
all the residuals is smaller than a given tolerance or until force ci, j,k,m,n required to just separate the bristles is given by
the calculated reaction forces and displacements have
converged. Bi, j,k,m,n
ci, j,k,m,n ⫽⫺ ŝi, j,k,m,n (20)
2 k,k
Bristle-Bristle Interactions. SUBSIS scans bristle pairs by set-
ting the current bristle as a reference bristle 共i.e., i⫽i ref and j Without the 3D-splay effect, the interference i, j,k,m,n is calculated
⫽ j ref) and identifying target bristles by setting using
i tar⫽i ref⫹m, j tar⫽ j ref⫹n (13) i, j,k,m,n ⫽2R⫺ 兩 si, j,k,m,n 兩 (21)
With 3D-splay effects taken into account, it can be shown using
where the integers m and n range from ⫺⌳ to ⌳. Assigning ⌳⫽1
geometric considerations that
means only immediate bristle neighbors are scanned. For ⌳⬎1,
interference of non-neighboring bristles can be considered, but i, j,k,m,n ⫽ 共 i,refj,k,m,n ⫹ itartar , j tar ,k,i, j ⫺ 兩 Wi, j,k,m,n 兩 兲 cos i, j,k,m,n
with an increase in computing time. In order to allow all bristles (22)
to interact with each other, we can set ⌳⫽max(I,J)⫺1. We also
apply periodicity conditions in the circumferential directions if i tar where 共see Fig. 5兲
is out of range, i.e.,
a i, j,k b i, j,k
i tar⫽ 再 i ref⫹m⫹I
i ref⫹m⫺I
if
if
i tar⬍1
i tar⬎I
(14)
⫽
冑 a i, j,k ⫹ 共 b i, j,k ⫺a i,2 j,k 兲 cos2
2 2
(24)
It is convenient to define the vector Vi, j,k,m,n joining two points on
neighboring bristles as i, j,k,m,n ⫽ i, j,k,m,n ⫺ i, j,k (25)
Journal of Engineering for Gas Turbines and Power JULY 2005, Vol. 127 Õ 585
ci, j,k,m,n ⫽⫺
B
k,k
冉
min s i, j,k,m,n ,
兩 Ci,oldj,k,m,n 兩 k,k old
B
冊
R̂i, j,k,m,n
reaction force on the bristles due to interactions with the rotor is
given by 共see Fig. 9兲
(29) r i, j,K
j,K ⫽
ri,rotor 兵 sin共 ␣ i, j,K ⫹ 兲 i⫹sin  i, j,K j其 (34)
where the negative sign indicates the correction force acts in the sin i, j,K
direction to reduce the magnitude of the reaction force. Finally,
where
the total reaction force at b i, j,k is computed by summing the con-
tributions due to all the scanned target bristles
⌳ ⌳
r i, j,K ⫽
B
K,K
冉
min 兩 i, j,K 兩 ,
j,K K,K sin
R i,radial
B
2
冊 (35)
j,k ⫽
Ri,bristles 兺 兺
m⫽⫺⌳ n⫽⫺⌳
Ci, j,k,m,n (30) Assuming frictionless contact, the corrective reaction force ni, j,K
on the rotor points entirely in the radial direction and can be
It should be noted that the computed reaction force may not lie computed using
normal to the bristle surfaces 共due to other 3D bristle deflections
occuring within the bristle pack兲. In order to reduce this error, the 兩 ri,rotor
j,K 兩
ni, j,K ⫽ K (36)
reaction force must be realigned normal to the bristle surface, but sin i, j,K
with its magnitude kept constant as shown in Fig. 8. The net effect
of this rotation is to displace the point b i, j,k by an amount
兩 ⌬ci, j,k,m,n 兩 , where
兩 ⌬Ci, j,k,m,n 兩 ⫽ 兩 Ci,new
j,k,m,n ⫺Ci, j,k,m,n 兩 ⭐2 兩 Ci, j,k,m,n 兩
old old
(31)
Fig. 7 Free bristle-point movement in a Äconst plane Fig. 9 Rotor interference, 3D global view
r i, j,k ⫽
B
k,k
冉
min 兩 i, j,K 兩 ,
兩 Ri,brj,K 兩 k,k
B
冊 (39)
model with no lay-angle, backing ring, or rotor 共the geometrical
parameters are the same as those in Table 1, but with ⫽0 deg
and no periodicity conditions兲. One of the corner bristles is sub-
Notice that the i component, which is much smaller than the j jected to a concentrated external force F1,1,K ⫽2⫻10⫺3 I⫹2
component, is due to the so-called inclined 3D-prop effect. This ⫻10⫺3 J applied at the tip, causing it to bend and split the brush
occurs only if the reaction force is located at 共or very close to兲 the into two parts. For this simulation, the bristle search parameter ⌳
backing ring edge 共i.e., if k ⬇ edge) and implies that the bristles was set as 7 in order to allow all bristles to interact with each
are bent downward toward the rotor and in the circumferential other. Although no comparsion with analytical solutions was pos-
direction. Otherwise, the reaction force is directed entirely in the sible, due to the highly nonlinear solution, this case demonstrates
axial direction. the capability of the main bristle-to-bristle contact routines to deal
with complicated cases.
SUBSIS „Surrey University Brush Seal Iterative Simulator….
The 3D method described here has been implemented into a new
FORTRAN 90 code called SUBSIS 共Ref. 关12兴兲. This code is capable Results
of handling arbitrary initial brush seal geometries, operating con- Example results using SUBSIS are shown here for brush seals
ditions, and imposed forces. These forces can either be obtained with parameters that are representative of those used in Refs.
from CFD calculations, experimental results or simple theoretical 关2,3,5兴 共see Table 1兲. The value of the Young’s modulus is repre-
models 共the latter has been employed here兲. All target bristles, the sentative of a cobalt-based alloy known as Haynes-25 共see Ref.
backing ring and rotor are considered as generic ‘‘objects’’ with 关14兴兲. Table 2 presents a summary of the assumed boundary and
which a reference bristle can interact. This allowed the code to be operating conditions. Only converged solutions are shown in this
written in a modular fashion, which can later be expanded to paper for which all the displacement residuals are much smaller
allow additional features to be included. SUBSIS has also been 共typically by a factor of 10⫺3 ) than the local point deflections.
developed with the ultimate aim of being coupled with CFD or
other CAD packages. In order to allow this, a generic method of
importing/exporting aerodynamic forces and current bristle coor-
dinates to/from SUBSIS was included in the design specification.
A quantitave comparison of the SUBSIS results with experimen-
tal data was not possible at this stage due to the idealized condi-
tions, although the results do show the expected behavior. How-
ever, comparison with theoretical results have been obtained for a
single bristle in contact with the backing ring or rotor. In all cases
the comparison with the theoretical solutions was accurate to
within the round-off error. The code has also been tested on Win-
dows®, Linux, SGI 共IRIX 64-bit兲, and SUN operating systems.
The main bristle-to-bristle contact routines have also been vali-
dated for the case of a single row of four bristles with no initial
gap between them 共see Fig. 11兲. The length L and bristle diameter
2R were taken as 9.30 mm and 0.0760 mm, respectively. The first
bristle is subjected to a concentrated force F of 10⫺3 N at its tip in Fig. 12 Incursion of a single bristle into a hexagonally-packed
the plane of the bristle row. This results in a uniform deflection of brush
Journal of Engineering for Gas Turbines and Power JULY 2005, Vol. 127 Õ 587
再再 冎
ward the upstream bristles, which help to ‘‘hold up’’ the down-
2␥ stream bristles. Further work is suggested to study this effect. This
共 1⫺ ␥ 兲 ⫹ Y Q axial J if k⫽K typical SUBSIS calculation took approximately 50 min to complete
Qi, j,k ⫽ W i, j,k average (41)
using a 1.0 GHz processor with the Red Hat Linux operating
0 otherwise system. 共For comparison, Ref. 关9兴 report typical computing times
In the above equation, Q axial of 6 –20 h on a Pentium 4 PC for a 9⫻5 bristle pack simulation
average represents the averaged axial force
per bristle. This was determined using the following simple rela- obtained using commercial finite-element software.兲
tionship based on the pressure gradient ⌬p across the seal and the
cross-sectional area CH between the backing ring and rotor for
the segment of the brush seal under consideration:
CH⌬p
average⫽
Q axial (42)
IJ
The parameter ␥ can be specified by the user, and for all of the
solutions shown here, this was taken as 0.2. Although not shown,
the numerical results were found to be almost insensitive to this
parameter. No external circumferential forces are considered in
the results shown here.
Figure 13 shows the tip positions for a converged solution in
which there is no pressure load, but in which the rotor position
was gradually moved into the pack 共in 10,000 iterations兲 from no
initial interference, up to a final incursion of 0.20 mm. For this
case, the minimum initial spacing ␦ between the bristles was taken
as 0.0 mm 共i.e., the brush was perfectly packed兲. It is evident that
the displacements of the bristles in the axial upstream direction
Fig. 14 3D bristle deflections „⌬pÄ0.0 bar, Z rotorÄ0.20 mm,
are entirely due to the 3D-splay effect. Figure 14 shows a 3D view
Ä45 deg, ␦Ä0.0 mm…
of the bristle deflections.
Figures 15–17 show the solution for the case of a brush seal in
which both the pressure load and rotor interference were gradually
increased from zero up to their final values of 1.0 bar and 0.20
mm in 50,000 iterations. The numerical results show the expected
bending behavior of the bristles. Figure 18 shows that the radial
force on the rotor is not uniform 共which may be partly responsible
for the uneven wear typically observed on rotors兲. These results
also suggest a ‘‘mechanical transmission’’ of reaction forces to-
Fig. 18 Radial force on rotor „⌬pÄ1.0 bar, Z rotorÄ0.20 mm, Fig. 20 Effect of increasing rotor interference Z rotor „⌬p
Ä45 deg… Ä4.0 bar, Ä45 deg…
Journal of Engineering for Gas Turbines and Power JULY 2005, Vol. 127 Õ 589
Nomenclature
a, b ⫽ elliptical semi-major and minor axes
B ⫽ bending stiffness (⫽EM)
c ⫽ bristle-bristle corrective reaction force
C ⫽ bristle-bristle actual reaction force
C ⫽ circumferential length of brush seal model
E ⫽ modulus of elasticity
axial
F br ⫽ axial reaction force on backing ring
radial
Fig. 21 Effect of increasing the lay-angle „⌬pÄ1.0 bar, F br ⫽ radial reaction force on backing ring
radial
Z rotorÄ0.20 mm… F rotor ⫽ radial reaction force on rotor
F ⫽ total force
H ⫽ backing ring overhang height
i, j, k ⫽ unit vectors in the x, y and z directions
Conclusion I, J, K ⫽ unit vectors in the X, Y and Z directions
i, j, k ⫽ grid point indices in (X,Y ,Z) coordinates
A new predictive capability for the 3D bending of bristles in
I, J, K ⫽ maximum value of indices in i, j, k, respectively
brush seals has been developed and tested. Results using this
L ⫽ actual length of bristles
model have been obtained that are qualitatively consistent with
M ⫽ second moment of area
the expected 3D bending behavior of brush seals. Calculated val-
n ⫽ corrective radial force on rotor
ues for the reaction forces on the rotor and backing ring are ob-
Q ⫽ aerodynamic force
tained and compared for different pressure loads and rotor incur-
sions into the bristle pack. The 3D-splay effect is also highlighted Q axial
average ⫽ averaged model axial aerodynamic force
in the results as an important effect that must be taken into ac- r ⫽ correction force
count. The results show that the classical approximation to the s ⫽ reference-to-target vector in a ⫽const plane
radial force on the rotor due to bristle interactions 共see Eq. 共43兲兲 R ⫽ total reaction force
may not yield reliable estimates for brush seals operating at high R ⫽ radius of bristles
pressure loads or for seals with relatively small lay-angles or U ⫽ vector aligned with bristle center-line
Young’s modulus. Possible hysteresis due to the deformation his- V ⫽ reference-to-target vector in a Z⫽const plane
tory of the bristles has been observed but not reported in the W ⫽ reference-to-target vector in Z⫽const plane
current results, requiring further analysis. W ⫽ axial length of brush seal model
Further work could involve incorporating frictional effects be- x, y, z ⫽ bristle-oriented coordinates
tween the bristles and the backing ring or shaft surfaces. As shown X, Y, Z ⫽ circumferential, axial and radial coordinates
by Crudgington and Bowsher 关8,9兴, these effects are important ␣,  ⫽ deflection angles in x and y-directions
and should be addressed. Additional geometrical effects due to ␥ ⫽ ratio used in Eq. 共41兲
brush seals of small rotor radii may also need to be considered. ␦ ⫽ initial gap clearance between bristles and backing
Conservation of bristle length, which is not satisfied using stan- ring
dard linear beam-bending theory, may also need to be considered ⌬p ⫽ pressure difference across seal
for cases with relatively large bristle deflections. The ultimate , ⫽ bristle deflections in the x and y-directions
⫽ bristle orientation angle 共see Eq. 共11兲兲
⫽ local interference
⫽ ellipse center-to-surface length
⫽ 3D bristle interaction angle
⌳ ⫽ target bristle search parameter 共integer⭓1兲
⫽ deformed lay-angle 共see Eq. 共10兲兲
⫽ coordinate along bristle length 共⫽0 at fixed bristle
end兲
⫽ angle between reaction force and target point
⫽ beam deformation tensor
⫽ lay angle 共⫽0 deg for radially aligned bristles兲
⫽ ellipse center-to-surface angle
⌼ ⫽ periodicity parameter 共see Eq. 共16兲兲
⫽ reference-to-target angle in a Z⫽const plane
⫽ reference-to-target angle in a ⫽const plane
⍀ ⫽ relaxation parameter
b i, j,k ⫽ center-line point on bristle (i, j) at position k .
br ⫽ backing ring
edge ⫽ edge 共overhanging兲 part of backing ring
init ⫽ initial 共unperturbed兲 value
old ⫽ value at previous iteration
Fig. 22 Effect of varying the Young’s modulus E „⌬p ref ⫽ reference bristle
Ä1.0 bar, Z rotorÄ0.20 mm, Ä45 deg… rotor ⫽ rotor surface
Journal of Engineering for Gas Turbines and Power JULY 2005, Vol. 127 Õ 591