J. L.

Teale
Research Assistant.2
An Evaluation of the Average Flow
A. 0. Lebeck
Associate Professor,
Model [ 1 ] for Surface Roughness
Effects in Lubrication1
The University of New Mexico,
Albuquerque, New Mexico

The average flow model presented by Patir and Cheng [1] is evaluated. First, it is shown
that the choice of grid used in the average flow model influences the results. The results
presented are different from those given by Patir and Cheng. Second, it is shown that the
introduction of two-dimensional flow greatly reduces the effect of roughness on flow. Re-
sults based on one-dimensional flow cannot be relied upon for two-dimensional problems.
Finally, some average flow factors are given for truncated rough surfaces. These can be
applied to partially worn surfaces. The most important conclusion reached is that an even
closer examination of the average flow concept is needed before the results can be applied
with confidence to lubrication problems.

Introduction

The analysis of surface roughness effects on lubrication has gained factors are altered by the choice of numerical representation of contact
considerable attention in recent years since it is widely recognized that shape and size and boundary conditions.
surface roughness can alter the solution for pressure and leakage in This work is intended to clarify the issues created by the non-
hydrodynamic bearings from that given by classical theory such as uniqueness of flow factors. In some cases enhancements of the flow
described by Cameron [2]. Several concepts of analysis have evolved, model have been suggested, such as the corner point method and
the goal of each being to derive an average Reynolds equation gov- truncated roughness distribution described later. Recommendations
erning expected pressure in rough bearings such that equations rely are made for further work to explain these issues, particularly in the
on surface roughness statistics rather than a specific topography. area of numerical analysis. It is concluded that the boundary depen-
The first concept is based on averaging the Reynolds equation prior dence issue limits the applicability of the average flow model.
to solution and relies on stochastic methods. Equations of this type
have been derived by Christensen and Tonder [3], [4], [5j. A second R e v i e w of t h e A v e r a g e F l o w M o d e l
concept involves perturbation techniques such as used by Elrod [6]. The average Reynolds equation as developed by Cheng and Patir
However, the equations derived using the first concept are limited [1] is derived by considering mean flow through the rectangular
to specific roughness configurations and using the second concept are control volume shown in Fig. 1. This volume has base area Ax Ay
limited to noncontacting situations. which is small relative to the bearing dimensions but contains nu-
A third concept as introduced by Tzeng and Saibel [7] involves merous asperities and possible asperity contacts. Surfaces bounding
averaging the results of solutions of the Reynolds equation. Recently, this volume have motion U; in the x direction and a height H(x, y)
Patir and Cheng [1] developed an average flow model based on aver- which can be represented as composed of a nominal part h and a
aging results of numerical solutions to the Reynolds equation. In this random component b for each surface. Referring to Fig. 1, total film
technique flow factors are defined which are applied as correction thickness is given by
terms to smooth bearing equations. While this method overcomes
some limitations of the other concepts, it will be shown herein that H = h + o1 + 82 if h + «i + 02 > 0
the quantitative results for flow factors presented by Patir and Cheng # = 0 if h + 5! + 82<0 CD
[1] are not unique. Certain fundamental questions are raised which Assuming that the lubricant is incompressible and isothermal, then
at the present time cast some doubt on the validity of this otherwise the volumetric flows per unit width are expressed as:
very useful concept. The results herein demonstrate that the flow
H3ap + / u 1 ± m H
12/j.dx \ 2 /
1
This research sponsored in part by the Office of Naval Research. (2)
2
Author currently located at IBM General Products Division, Tucson, Ariz. H3 dp
Contributed by the Lubrication Division of THE AMERICAN SOCIETY OF <Jy:
12ndy
MECHANICAL ENGINEERS and presented at the ASME-ASLE Lubrication
Conference, Dayton, Ohio, October 16-18,1979. Manuscript received at ASME Now the expected value of these flows can be approximated by the
Headquarters, July 27,1979. Paper No. 79-Lub-37. average flows in the sample control volume:

360 / VOL. 102, JULY 1980 Transactions of the ASME

Copyright © 1980 by ASME
Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 y+A
1 -U2 d<t>s , M
E(qx) a: qx
Ay f qxdy (3)
2 dx dt
(6)

; qydx
E{qy) =* qy D i s c u s s i o n of t h e A v e r a g e F l o w H y p o t h e s i s
Ax Jx
Patir and Cheng assert that the pressure flow factors 0X, (j>y com-
where the bars designate average quantities and the expected value pare flow in a rough bearing to that predicted by smooth bearing
E is defined in conjunction with the distribution function / of the equations. The shear flow factor </>s is purported to account for addi-
variable: tional flow in rough bearings due to sliding that is presumably net
"fluid carried in the valleys" [8]. Furthermore, it is stated that the flow
E(x) = f_lxfix)dx (4) factors depend only upon the statistics of the rough surfaces.
Now, considering the Patir-Cheng hypothesis in greater detail,
At this point Patir and Cheng hypothesize that the expected unit rearrangement of equations (2), (3), and (5) results in
flows can be expanded as follows:
fefes j l J l + \J2\ fUt - u 2 ^ E\m ,ap
^l = 0 x h 3
^ ^ - 6/to-(Ui - U 2 )0 8
d p
dx dx
dp' , M?(p)
dp E\H3 (7)
Qy= - -
dy] dy
12^ dy It is acknowledged that Patir and Cheng have not expressed their
These relations define pressure flow factors <f>x, 4>y and the shear flow hypothesis in this form explicitly, but this is the only reasonable in-
factor ips- Quantity a is the combined RMS roughness of the surfaces, terpretation of flow factor definitions.
i.e., a2 = <TI + a\. Although there is some support for this hypothesis [9], equations
After performing a mean flow balance on the control volume, ex- (7) do remain strictly a hypothesis and must be verified as such. As
pressions for average flow (5) are substituted to obtain the average the results will later show, the first of equations (7) appears to be valid
Reynolds equation proposed by Patir and Cheng: based on one-dimensional flow. However, the introduction of two-
_d_ Uxh3dp\ _d_ Uyh* dpi _ IVI + U2\ dff dimensional flow leads to flow factor values which differ from those
derived on the basis of one-dimensional flow. The explanation for this
dx \ 12/u dx] dy \ 12/i fa] \ 2 ) dx difference may simply be that the hypothesis above is not generally
valid.

q + -r-2- Ay S o l u t i o n of F l o w F a c t o r s
M
ASPERITY y ay In the Patir-Cheng method, average flow in a small bearing segment
NTACTSy
\ is analyzed to determine flow factors. Figure 2 is an illustration of the

NO FLOW AT CONTACTS-
<

8 D aq P(X,L y ) J
-^^ x V x AX

0
,k
* 6
P(0,y) QP ^ P(Lx.y)

P °
P (X.O)

Fig. 1 Control volume for average flow analysis showing edge view of surface #— Lx »
roughness functions. Dashed lines are nominal geometries from which
roughness is measured Fig. 2 Model to analyze flow factors

.Nomenclature.
E = expectancy operator qx, qy = leakage components ix = viscosity
/ = density function r = truncation level \x, Xy = delay lengths
h = nominal film thickness function t = time Xo.5 = 50 percent delay length
H = total film thickness function U = velocity a = standard deviation
Lx, Ly = dimensions of bearing segment x,y = Cartesian coordinates 0S = shear flow factor
p = pressure 7 = surface pattern parameter 4>x, <t>y = pressure flow factors
8 = surface roughness function $ s = shear flow factor

Journal of Lubrication Technology JULY 1980, VOL. 102 / 361

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 model used to simulate rough bearing lubrication. Although real
surface roughness data could be used to define gap function H(x, y), r 1
a more general method is to numerically generate roughness functions, ! 1 , v
1 W 1 W 1 *• 1
5i, ^2 for each surface. In this manner the roughness will have a con- 1 1 1 ._, 1
trolled distribution of asperity heights and autocorrelation function. 1
Film thickness is solved using Equations (1) and the requirement that
H > 0 implicitly enforces zero flow at contact points. 1 1 1 1 1 1
Local pressure p(x, y) is governed by the Reynolds equation for
L
y 1 1 T [ T 1
rough surfaces best described by Godet and Berthe [10]:
i i i i i i
1 r 1 r--'-r 1
1
n ' 0 ' 0 ! n ' n '
dx \ ox; dy \ dy/ 1 1 1 1 1 1

= 6MU1 + U 2 ) ^ + 6 M ( U 2 - U 1 ) ^ ^ ^ (8) I r D T
Pi,
' 1 v'
-1
1
1
-- r
1
•" r
,
1
-p—i
HIXTI.I
1
dx dx Ay .--I
-X-.
^nx,i
This equation presumes that Eulerian coordinate transformation on
the time derivatives of the roughness is valid:
»
•• 1 , 2 (9)
"4 Lx >
dt ' dx'
The boundary conditions are selected to inhibit all y leakage: — BOUNDS PRESSURE CONTROL VOLUME
— BOUNDS REGION OF CONSTANT FILM
p(0, y) = pa THICKNESS H
p(Lx, y) = Pb PRESSURE NODE
dp H
i,j
0 (10)
dy Fig. 3 Grid network for corner point method

Flow factor 4>y cannot be evaluated with these boundary conditions

since qy = 0 is enforced. It is pointed out by Patir that <j>y can be found
from <j>x according to the following:
J =N
<t>yW(r,y) = <t>x(.h/a,lft) (11)
J=N,
where y is the surface pattern parameter developed by Peklenik [11].
This parameter is the ratio of the x to y 50 percent correlation lengths

X.5x
(12)
^.6y

It will be assumed for simplicity that 7 i = 72 = 7 for each surface.

Furthermore the nominal film thickness h is constant.
Patir postulates that shear flow factor 4>s depends upon the indi-
vidual surface statistics in the following manner:

2
^ f $,(/»/„, 72) ~ R *.Wc, 71) (13)

Hence for a situation of 71 = 72 = 7 , equation (7) may be rewritten

as it applies to the bearing segment under consideration:
I=N„
6JU
(Ui - U2)(<rl - crf)*s (14)
dx dx
BOUNDS PRESSURE CONTROL VOLUME
Assuming that <fc and $ s are independent, these factors are solved
by numerically evaluating the left side of (14) for the simulation
BOUNDS REGION OF CONSTANT FILM
THICKNESS H
problem posed. Several sets of different but statistically identical
surfaces are generated to evaluate average flow in order to obtain PRESSURE NODE
confidence intervals for the flow factors. For each set of surfaces two X HY|
sets of boundary conditions are required to solve unknowns 4>x and
$ s . Patir employs the following: + HXi
Fig. 4 Grid network tor Patir method
(a) Ui = U 2 = U r , Pa^Pb' Hs*Ei
S
h (Pb ~ Pa) \ dx
U. However, new surfaces must be created to examine the dependence
(b) U2=-Ui=-
2 of flow factors on pattern parameter 7.

Pa = Pb => * s = • H3 — (15) Numerical Aspects

6/*U,(ffl-.o-f) \ dx
Two numerical methods of evaluating (Ha dp/dx) were used in
this work, the difference between these two being the finite difference
Rather than regenerate the artificial surfaces for each value of h, representation of the rough bearing segment. The first technique,
the same set of surfaces may be used to obtain smooth curve results. termed the "corner point method," is sketched in Fig. 3, This grid is

362 / VOL. 102, JULY 1980 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 based on an idea by Castelli [12] wherein roughness values are gen- ferent values for pressure flow factors. The confidence intervals of
erated equidistant from the four neighboring pressure nodes. Fig. 4 the corner point and Patir method results do not overlap. The solution
represents the difference network believed to be used by Patir [8]. In for shear flow factor $ s (hi a, 1) is presented in Fig. 6. Again it observed
the "Patir method" random roughness is generated between all ad- that the difference in solutions resulting from the two methods is
jacent pairs of pressure nodes. statistically significant.
Considering first the pressure flow factors for isotropic surfaces (7 The difference in the results obtained using the two different
= 1) having Gaussian distribution of asperity heights, results for methods is believed to be due to the fact that the corner point inter-
pressure factor 4>x(hla, 1) using the above mentioned methods are pretation offers less resistance to flow than the Patir method. This
plotted in Fig. 5. Also shown for reference are Patir and Cheng's characteristic is believed to be fundamental to the numerical methods
published results [1] and those obtained by Elrod [6] using pertur- and a detailed discussion is given in reference [13]. The point to be
bation techniques. The validity of the Patir-Cheng solutions based made is that the flow factors do not seem to have a unique solution
on the method used by those authors is not in question since they fall based on results presented above. A case can be made that because
within the 90 percent confidence intervals of the Patir method results the corner point element blocks leakage in both directions at contact
obtained herein. The Patir method results also concur with the Elrod points, then it is a more realistic representation than the Patir method
solutions which apply in the non-contacting regime hi a > 3. However, where flow is blocked in one direction only by contact.
it is evident from the corner point method solution that a slightly A second possible cause for the difference between the results re-
different interpretation of the flow problem yields significantly dif- lates to asperity size. The results in figure 5 were all based on a 26 X
26 grid with a 50 percent correlation length of two grid units. Results
shown in reference [13] show that as the correlation length of the grid
size is increased, the value <j>x based on the corner point method de-
•4 CORNER POINT METHOD crease whereas those based on the Patir method stay about the same.
1
9 0 % CONFIDENCE INTERVAL Thus for larger correlation lengths the difference between the two
methods becomes smaller.
i - PATIR METHOD
U
9 0 % CONFIDENCE INTERVAL The question that remains is which of the methods is appropriate.
Further studies are needed using larger grid and asperity sizes to
1.0 determine whether the differences are due to the statistical nature
of the problem or a real difference between the representations of the
asperities.

1.2
f CORNER POINT METHOD
1
9 0 % CONFIDENCE INTERVAL
0.5 1.0
§ PATIR METHOD
90_% CONFIDENCE INTERVAL
PATIR ELROD — PATIR
RESULT RESULT RESULTS
0.6
</>s
0 1 2 3 4 5 6 04
h
cr
Fig. 5 4>x for isotropic, Gaussian-symmetric surfaces 0.2

Fig. 6 $ „ for isotropic, Gaussian-symmetric surfaces

Fig. 7 tpx for nonisotropic, Gaussian-symmetric surfaces Fig. 8 $ „ for nonisotropic, Gaussian-symmetric surfaces

Journal of Lubrication Technology JULY 1980, VOL. 102 / 363

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Results for non-isotropic flow factors using the corner point method p(0,0)=pi
are offered in Figs. 7 and 8 as alternatives to the Patir results.
p(0, Ly) = p 2 (16)
Independence of Flow Factors p(LX,Ly) = P 3
To establish the independence of cfe and <£s as implied by Equation
(7), the nominal pressure gradient Z>p/dx = Pb ~ pJLx and the rel- p(Lx, 0) = p 4
ative velocity Ui — Ua = Us are varied. Results of this study are pre- Pressure is specified over the remainder of each boundary as a linear
sented in Fig. 9 for hia = 2 and 7 = 1. Five values of the nominal function between corners of the segment (Fig. 2). This configuration
pressure flow to shear flow ratio / were chosen. The extreme cases / results in non-zero nominal y leakage. For a case of pure rolling (U\
= 0 (dp/dx = 0) and / = °°(US = 0) correspond to pure sliding and = U2) and dp/dy = 0 (pi = p , Ps = pd the independence of <f> and
2 x
pure rolling, respectively. Equation (14) plots as a line which is 4> would require that if> remain unchanged in value from previously
x x
bounded by 90 percent confidence interval regions in each case. It is reported results.
concluded that <fc and <3>s are indeed independent as evidenced by Using the corner point method it is found that the solution to this
the fact that a unique solution for flow factors is obtained. The shaded problem gives <j> (1,1) = 1.14 compared to a value of 0.85 given by Fig.
x
region common to all cases corresponds to the previously reported 5. Thus, allowing local side leakage, even if net qy = 0, affects the so-
values for (j>x, $ s . lution for <px. Table 1 gives results for pressure flow factors using
The same conclusion is not reached for the independence of <j>x and several values of dp/dy, while in Table 2 7 is varied. It is concluded
<py on boundary conditions. Independence of these factors is tested from Table 1 that the magnitude of applied pressure gradients does
by considering a different set of boundary conditions from those used not affect results. The validity of equation (11) is verified by the re-
previously: sults in Table 2. But the fact remains that flow resistance using the
above boundary conditions is everywhere lower {t$>x higher) than
predicted in Figs. 5 and 7 where side leakage is maintained at zero.
One explanation for this behavior is that when flow is obstructed
by a contact and local side leakage is allowed, flow simply turns rather
than passing around the obstacle. The significance of the above results
is that restricting edge flow, as was done by Patir and Cheng, can lead
to incorrect flow factors or perhaps the hypothesis of equations (7)
is invalid.

Roughness Distribution
All results presented to this point have assumed that the distri-
bution of asperity heights is Gaussian. While this is not a limitation
of the average flow model, there are some aspects of the analysis which
are of questionable validity when applied to Gaussian surfaces. The
use of this distribution requires that asperities are elastically deformed
at contact points and spring back to their original shapes subsequent
to deformation. While Patir has assumed that deformations are small,
it is unrealistic to assume that deformation rates will be small.
The point above implies that the Eulerian coordinate transfor-
"0 0.2 0.4 0.6 0.8 1.0 mation used in equation (9) would not be correct because of an ad-
ditional 8(t) due to elastic deformation. Functional description of this
Fig. 9 Independence of <j>x and $ ,
additional term would add undesirable complexity to the flow model

Table 1 Testing the boundary dependence of <j>xand

<!>y 7 = 1
Corner point method
hh
: / dy
dx/
CO <t>x = 1 . 1 4 0 0 0.9707 0.9693
(fry = — —
3 0X = 1.1381 0.9650 0.9656
<t>y = 1.0616 0.9091 0.9354
2 0X = 1.1372 0.9621 0.9638
<t>y = 1.0683 0.9262 0.9464
1 <t>x = 1.1138 0.9536 0.9583
<t>y = 1.0751 0.9433 0.9575

Table 2 Boundary dependence of <f>x, $ y for

nonisotropic surfaces dp/ dx/ dp/ 6y = 1
Corner point method
hh
7 1 1.5 2 3
1/3 4>x = 0.6686 0.6548 0.7025 0.8026
<t>y = 1.6289 1.3373 1.2176 1.1139
1 0* = 1.1138 — 0.9536 0.9583
(j>y = 1.0751 — * 0.9433 0.9575
3 <j>x = 1.6356 1.3414 1.2209 1.1158
Fig. 10 Roughness density functions (a) Gaussian-symmetric (b) Trun-
<t>y = 0.8199 0.7569 0.7759 0.8430
cated

364 / VOL. 102, JULY 1980 Transactions o! the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 as well as reduce the general applicability to various bearing materials. Results for <t>s indicate that the dependence of 0S on <n, cr2 as pro-
This difficulty can be circumvented by considering the truncated posed by equation (13) is no longer valid. Previous results indicated
Gaussian distribution which is shown compared to the symmetric that for o\ = 02 no shear flow would develop. However, for r ^ . 5 there
distribution in Pig. 10. exists excess "valley volume" on one surface resulting in net fluid
For bearing materials that have been "run in" the asperities are transport consistent with Patir's description of <j>s. Thus $ s rather
often sheared at a significant value oihla. In such cases the truncated than $ s is presented in Fig. 13.
distribution is a useful description of the surface character. The The remaining Figs. 14 and 15 are included to demonstrate the
mating of two such surfaces is sketched in Fig. 11 and the truncation effect of non-isotropy on truncated roughness flow factors. These
level r accounts for the differences in truncation levels. It is evident solutions may have a wider applicability than the Gaussian symmetric
that deformation rates can now be assumed small because the distribution results. It is also believed that they may be more valid
asperities do not spring back. Hence the transformation in equation as hla -* 0 because deformation is not a factor. The results in Figs.
(9) becomes valid at least for regions of continuous film thickness 12 to 15 are based on qy = 0 so that these results are subjected to the
gradient. limitations of accuracy imposed by boundary condition depen-
The pressure flow factor <j>x for isotropic truncated surfaces having dence.
cri = 02 is given in Fig. 12. Values of r from 0 to 0.5 are given since for
ffi = <T2, (t>x(h/(T, 7, r) = 4>x(h/o, y, 1 — r). As r increases to .5 flow Conclusions
decreases because contact area is maximum at r = .5. It is also true The average flow model is a valuable and stimulating contribution
that Hs is greater for this distribution than for the symmetric distri- to the analysis of surface roughness effects in hydrodynamic lubri-
bution because deformable asperities which would decrease IP have cation. However, the work presented herein indicates that the model
been truncated (dotted lines in Fig. 12). Comparing these results to at present may not be as broadly applicable as its authors have indi-
those in Fig. 5 for the corner point method for Gaussian symmetric cated. The major purpose of this review is to point out certain dis-
distribution shows that 4>x is generally larger. crepancies and to emphasize that more work is needed to resolve these
issues:
1. The selection of finite difference grid alters the results from
the average flow model significantly. Implicit in this selection is the
contact shape and characteristics. It is thought that the corner point
method used herein is a more realistic representation of the rough
surface lubrication. The fact that a different physical interpretation
of the contact situation alters the flow factor suggests that more work
is needed.
2. Since the corner point method intersects the results for pressure
flow factor obtained by Patir and by Elrod in the noncontacting re-
gime, there is reason to believe that this technique is valid. It is also
thought that the corner point method is only one of possibly several
valid models of the rough surface lubrication. This method is char-
acterized by a lower resistance to flow than the Patir method so that
Fig. 11 Profile of truncated distribution surface pressure flow factors are larger.
3. Results of Tables 1 and 2 indicate that the pressure flow factors
are boundary condition dependent. Suppression of the local and total
2.0 y leakage in the model to solve for flow factors as was done by Patir
CORNER POINT METHOD and Cheng leads to different results than when these conditions are
r=0.0 not imposed. Results based upon more general boundary conditions
give higher pressure flow factors than the Patir model due to lower
flow resistance.
4. It is thought that the truncated distribution of asperity heights
_ 0.2 o
is more realistic than the deformable rough surface concept, at least
I.5 for cases where sliding occurs. For this type of distribution it is evident
that both pressure and shear flow factors are dependent on the indi-
vidual surface statistics. Truncated surfaces are characterized by less

0.5

CORNER POINT METHOD

I.O
0.4
= 0.5

0.3

n ^ C ,
0.5
£°-5x.

2 3 4 5 6
h

Fig. 12 <K for isotropic, truncated surfaces Fig. 13 0s for isotropic, truncated surfaces

Journal of Lubrication Technology JULY 1980, VOL. 102 / 365

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 3.0 i l 1 1 1
CORNER POINT METHOD CORNER POINT METHOD
r =0.0
r =0.5
°-7=cr|=0.5
o-f . < r | . 0 . 5
6
\
2.0 - 3
-

\ \
<£x
Y--\
-C^A-
1.0
1/3
-B--'""^
1/6

Fig. 15 4>s for nonlsotroplc, truncated surfaces

1 1 i i l

0 1 2 3 4 5
h Rough Surfaces," Proc. Instn. Mech. Engrs., Tribology Group 184, Part 1, Vol.
ar 55,1969-70, p. 1013.
4 Christensen, H. and Tonder, K., "The Hydrodynamic Lubrication of
Fig. 14 <px for nonlsolropic, truncated surfaces Rough Bearing Surfaces of Finite Width," AMSE JOURNAL OF LUBRICATION
TECHNOLOGY, Vol. 93,1971, p. 324.
5 Christensen, H.," A Theory of Mixed Lubrication," Proc. Instn. Mech.
Engrs., Tribology Group, Vol. 186,1972, p. 421.
3
resistance to flow due in part to less contact and also because E{H ) 6 Elrod, H. G., "A General Theory for Laminar Lubrication with Reynolds
is greater than for Gaussian-symmetric surfaces. Roughness," Report for Department of Machine Design, Technical University
of Denmark, 1977
5. The independence of <f>x and 0S as implied by equation (7) is 7 Tzeng, S. T. and Saibel, E., "Surface Roughness Effect on Slider Bearing
verified at least for bearing segments with zero flow at the edges. More Lubrication," ASLE Trans., Vol. 10,1967, p. 334.
work is required to determine the applicability of this expression to 8 Patir, N., "Effects of Surface Roughness on Partial Film Lubrication
general two-dimensional lubrication problems. Using an Average Flow Model Based on Numerical Simulation," Dissertation
at Northwestern University, Evanston, 111., June 1978.
Continuing work with the average flow model will undoubtedly 9 Chow, L. S. H. and Cheng, H. A., "The Effect of Surface Roughness on
provide even more insight to the lubrication of rough surfaces. It is the Average Film Thickness Between Lubricated Rollers," ASME JOURNAL
recommended that the above aspects of the problem be investigated OF LUBRICATION TECHNOLOGY, Jan. 1976, pp. 117-124.
further. 10 Godet, M. and Berthe, D-, "A More General Form of Reynolds Equa-
tion—Application to Rough Surfaces," Wear, Vol. 27,1973, pp. 345-357.
11 Peklenik, J., "New Developments in Surface Characterization and
References Measurement by Means of Random Process Analysis," Proc. Instn. Mech.
1 Patir, N., and Cheng, H. S., "An Average Flow Model for Determining Engrs., Vol. 182, Part 3K, 1967-1968, p. 108.
Effects of Three-Dimensional Roughness on Partial Hydrodynamic Lubrica- 12 Castelli, V. and Pirvics, J., "Review of Numerical Methods in Gas
tion," ASME JOURNAL OF LUBRICATION TECHNOLOGY, Vol. 100, Jan. 1978, Bearing Film Analysis," ASME JOURNAL OF LUBRICATION TECHNOLOGY,
p. 2. Vol. 90, Oct. 1968, p. 777.
2 Cameron, A., The Principles of Lubrication, Longmans, Green and Co., 13 Teale, J. L., "Surface Roughness Effects in Hydrodynamic Lubrication,"
Ltd., London, 1966. Master's thesis, The University of New Mexico, Albuquerque, New Mexico,
3 Christensen, H., "Stochastic Models for Hydrodynamic Lubrication of Aug. 1978.

-DISCUSSION.

K. T0nd©r2 Taking the Berthe-Godet version of Reynolds' equation (equation

(8) of paper) it is seen t h a t p and, thereby, dp/dx are given as the sum
The authors are to be congratulated for their effort to settle a dif- of individual contributions due to (C/2 + U\)dh/dx, to (U2 - Ui)d(8i
ficult and much-disputed problem in lubrication. The present paper — d2)/dx and to dp/dx, assuming dp/dy to be zero.
is a valuable contribution. Hence, by averaging, the mean x-flow may be written
The discusser feels, however, that a few points of the work require
some comments.
First, the form of the Patir-Cheng results, (equation (7) of paper) qx = (L7 2 + UJH ~ W ldpus | dpud | dppx\
can hardly be said to represent a hypothesis only. This may be seen 2 12/x \ dx dx dx j
from the following. where the suffixes refer to t / 2 + Ui, U2 - Ux and dp/dx, respec-
The expression for x-flow is: tively.
IU2 + Ui\ „ H* dp The component dpud/dx must be proportional to fi(U2 - U\), so
Qx for dimensional reasons this flow contribution may be written as (U2
I 12M; <dx
— Ui)h/2fi, where f\ is a nondimensionaMunction depending on ge-
2
Institute for Maskindeler University of Trondheim, Norwegian Institute geometry only. Similarly dppx/dx is proportional to dp/dx so the
of Technology, 7034 Trondheim-NTH, Norway. corresponding flow component may be written

366 / VOL. 102, JULY 1980 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use