Role of Fractal Geometry in

Roughness Characterization and

Contact Mechanics of Surfaces
A proper characterization of the multiscale topography of rough surfaces is very
crucial for understanding several tribological phenomena. Although the multiscale
nature of rough surfaces warrants a scale-independent characterization, conven-
tional techniques use scale-dependent statistical parameters such as the variances of
A. Majumdar height, slope and curvature which are shown to be functions of the surface
Graduate Student. magnification. Roughness measurements on surfaces of magnetic tape, smooth and
textured magnetic thin film rigid disks, and machined stainless steel surfaces show
B. Bhushan1 that their spectra follow a power law behavior. Profiles of such surfaces are,
therefore, statistically self-affine which implies that when repeatedly magnified, in-
Fellow ASME. creasing details of roughness emerge and appear similar to the original profile. This
paper uses fractal geometry to characterize the multiscale self-affine topography by
Department of Mechanical Engineering, scale-independent parameters such as the fractal dimension. These parameters are
University of California at Berkeley, obtained from the spectra of surface profiles. It was observed that surface process-
Berkeley, CA 94720
ing techniques which produce deterministic texture on the surface result in non-
fractal structure whereas those producing random texture yield fractal surfaces. For
the magnetic tape surface, statistical parameters such as the r.m.s. peak height and
curvature and the mean slope, which are needed in elastic contact models, are found
to be scale-dependent. The imperfect contact between two rough surfaces is com-
posed of a large number of contact spots of different sizes. The fractal representa-
tion of surfaces shows that the size-distribution of the multiscale contact spots
follows a power law and is characterized by the fractal dimension of the surface. The
surface spectra and the spot size-distribution follow power laws over several decades
of length scales. Therefore, the fractal approach has the potential to predict the
behavior of a surface phenomenon at a particular length scale from the observations
at other length scales.

1. Introduction
The topography of rough surfaces influences a number of the curvature are used for characterization (Nayak, 1971). The
tribological phenomena such as friction, wear and lubrication. theories of elastic (Greenwood and Williamson, 1966) and
It is well known that the geometric structure of rough surfaces plastic (Nayak, 1973a; Gupta and Cook, 1972; Bhushan,
is random and that roughness features are found at a large 1989a) contacts between rough surfaces use these parameters
number of length scales between the length of the sample and for their analysis. However, it has been found that the
atomic scales. To study the mechanisms of any tribological variances of slope and curvature depend strongly on the
phenomenon it is absolutely necessary to characterize such resolution of the roughness-measuring instrument or any
multiscale rough surfaces and to know the structure at all other form of filter and are hence not unique (Thomas, 1982a;
length scales relevant to the phenomenon. This study em- Bhushan et al., 1988). It is also known that surface
phasizes the role of fractal geometry in understanding the ef- topography is a nonstationary random process for which the
fect of surface roughness in tribology. variance of the height distribution is related to the length of
Conventionally, the deviation of a surface from its mean the sample (Sayles and Thomas, 1978). Consequently, in-
plane is assumed to be a random process for which statistical struments with different resolutions and scan lengths yield dif-
parameters such as the variances of the height, the slope and ferent values of these statistical parameters for the same sur-
face. The conventional methods of characterization are
therefore fraught with inconsistencies which give rise to the
On sabbatical from IBM Research Division, Almaden Research Center, San term "parameter rash" (McCool, 1987) commonly used in
Jose, CA 95120. contemporary literature. The underlying problem with the
Contributed by the Tribology Division of THE AMERICAN SOCIETY or conventional methods is that although rough surfaces contain
MECHANICAL ENGINEERS and presented at the STLE/ASME Joint Tribology roughness at a large number of length scales, the characteriza-
Conference, Fort Lauderdale Fla., October 16-19,1989. Manuscript received by
the Tribology Division April 3, 1989. Paper No. 89-Trib-20. tion parameters depend only on a few particular length scales,

Journal of Tribology APRIL 1990, Vol. 112/205

Copyright © 1990 by ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 such as the instrument resolution or the sample length. A
logical solution to this problem is to use scale-invariant
parameters to characterize rough surfaces.
-^^^
In this study, roughness measurements on a variety of sur- ^N
faces show that the power spectra of the surface profiles
follow power laws. This suggests that when a surface is
magnified appropriately, the magnified image looks very
similar to the original surface. This property can be
mathematically described by the concepts of self-similarity
and self-affinity. The fractal dimension, which forms the
essence of fractal geometry, is both scale-invariant and is
closely linked to the concepts of self-similarity and self-
affinity (Mandelbrot, 1982). It is therefore natural to use frac- 102 103
tal geometry to characterize rough surfaces and provide the UNIT OF MEASUREMENT, e [km]
geometric structure at all length scales. The possible applica- Fig. 1 Increase in the lengths of coastlines for decreasing units of
tions of fractal geometry to tribology was explored recently measurement. Data from Mandelbrot (1967).
(Ling, 1988). Length measurements of profiles of engineering
surfaces show that at macroscales, such surfaces are not frac-
tal (Ling, 1987). However, the present paper shows that the
spectra, and not the length, of profiles suggest that in some timeter or a micrometer scale is used. However, a multitude of
cases of surface processing, the surfaces produced are factal objects found in nature (Mandelbrot, 1982), rough surfaces
even at macroscales of about 100/xm or even few millimeters, for example, appear disordered and irregular for which the
whereas in other cases they are non-fractal. The influence of measures of length, area and volume are scale-dependent. This
processing techniques on the fractal or non-fractal structure is suggests that the dimensions of such objects cannot be in-
also examined. The goals of this paper are to provide a basic tegers. A generalized concept of a dimension and the origins
understanding of fractal geometry, to take a fundamental of fractal geometry are now discussed.
look at the structure of rough surfaces, to indicate a method Fractal geometry was founded by Mandelbrot (1967) when
of using fractal geometry to characterize the roughness, he showed that for decreasing unit of measurement, the length
to apply the characterization to understand some surface of a natural coastline does not converge but, instead, increases
phenomena and finally to discuss the potential applications of monotonically. On plotting the length L as a function of the
fractal geometry to tribology. unit of measurement e on a log-log plot, he found a simple
relation of the form L — e1'0 as shown in Fig. 1. He finally
made an interesting conclusion that the real number D
2. Fractal Geometry associated with every coastline is the dimension of the
Euclidean geometry describes ordered objects such as coastline. This study marked the origins of fractal geometry
points, curves, surfaces and cubes using dimensions of 0, 1, 2, which has now found numerous applications in characterizing
and 3, respectively. Associated with each dimension is a and describing disordered phenomena in science and engineer-
measure of the object such as the length of a line, the area of a ing (Mandelbrot, 1982; Feder, 1988; Peitgen and Saupe,
surface and the volume of a cube. These measures are in- 1988). In order to understand the relevance of this new field of
variant with respect to the unit of measurement implying that mathematics to rough surfaces, it is necessary to provide a for-
the length of a line remains independent of whether a cen- mal background of fractal geometry.

Nomenclature

A area [m2]
B fractional Brownian motion;
also exponent in number-area
rule S power in a spectrum [m3] a = standard deviation of heights,
D = dimension t time [s] Vrn^ [m]
G = surface characterization x distance [m]; also exponent in
parameter [m] equation (35) a' = standard deviation of slope,
H = exponent in fractional Brown- height [m] •4m2
ian motion; also hardness of ZP = mean height of asperity peaks
material [N/m 2 ] a = surface parameter used for a" = standard deviation of cur-
l,L length [m] elastic contact theory vature, Vrri4 [m _ 1 ]
m0 zeroth moment of power length-scaling factor
spectrum, variance of height scaling ratio determining the w = frequency, l / / [ m _ 1 ]
[m2] spectral density
second moment of power unit of measurement [m] Subscripts
m0
spectrum, variance of slope surface parameter described a = referring to apparent area
fourth moment of power in equation (24) c = referring to a cavity
spectrum, variance of cur- mean curvature of peaks h = referring to high frequency
vature [m~ 2 ] [m- 1 ] limit
M measure [mD] thermal conductivity / = referring to low frequency
N number [Wm-'K-1] limit
Pr mean pressure, [N/m 2 ] number of contact spots L = referring to largest spot or
P load, [N] thermal resistance [W ~' K] cavity
resolution of surface measur- parameters for elastic contact p = referring to asperity peaks
ing instrument [m] theory, equation (18) / = referring to the total area

206/Vol. 112, APRIL 1990 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Hausdorff Dimension. The length of a line is measured by
breaking the line into small units of length e and then adding LINE
the number of units in the form
L = Ee> (1)
Similarly the area of surface is measured by breaking up the ' 1/m I I I I I
surface into small squares of size e x e and then adding the
number of units as M A G N I F I C A T I O N = m, N = m
2
A = Ze (2) (a)
It should be noted that in equations (1) and (2) the exponents 1
and 2 correspond to the dimensions of the objects. The unique
property of these measures, length and area, is that they are
independent of the unit of measurement e and in the limit e—0
these measures remain finite and non-zero. This concept of
Euclidean measure and dimension can be generalized to the SURFACE
form
M= LeD (3)
where Mis the measure and D is a real number. The exponent
D is the dimension of an object if it makes the measure M in-
dependent of the unit of measurement e in the limit of e—0.
This is a simple definition of the Hausdorff dimension [ 1/m
whereas a more rigorous one can be found in Mandelbrot
(1982). Contrary to common understanding, this generaliza- 1/m
tion allows the dimension of an object to take noninteger
values. M A G N I F I C A T I O N = m, N = m 2
If, in the above argument, it is assumed that an object is
broken into N equal parts then equation (3) can be written as
M=NeD. Since the measure is invariant with the unit of
measurement it follows that N~e~D. Now if the length of an
object, of dimension D, is evaluated, then the length would
vary as L = Ne1 ~ e 1 _ £ > , as was observed for the lengths of the GENERALIZATION
coastlines. It can be easily seen that only when D=\, the
length will be independent of e. N = mP
Self-Similarity. The generalized concepts of measure and ^ logiV
dimension are fundamental to the issue of self-similarity. Con-
sider a one-dimensional line of unit length as shown in Fig. 2a. D = ——
Each segment of the line, of size 1/m, is similar to the whole log m
line and needs a magnification of m to be an exact replica of Fig. 2 Self-similarity in Euclidean objects of integer dimension (a) line
the whole line. Since the length of the line remains indepen- (b) surface and its generalization to any noninteger dimension
dent of \/m, it follows that the number of units is N~m. Now
consider a square in Fig. 2b which has a side of unit length.
Each small square of side 1/m is similar to the whole square that a tangent cannot be drawn at any point and therefore the
and needs a magnification of m to be an exact replica of the curve cannot be differentiated. Secondly, the curve is exactly
whole square. However, the number of small squares in the self-similar. This is because if a small portion of the curve is
whole is N~m2. In general, for an object of dimension D, it appropriately magnified, it will be an exact replica of the
follows from the above argument that whole Koch curve. Thirdly, although the curve contains
roughness at a large number of scales, the dimension of the
N~mD (4) curve remains constant at all scales. This scale-in variance of
the dimension is an important property which will be utilized
Hence the dimension of the object can be obtained as
to characterize rough surfaces. An example of a self-similar
logTV object found in nature is the coastline of an island. Although
D=-^— (5)
log/K these objects are not exactly self-similar, they are statistically
self-similar. Statistical self-similarity implies that the prob-
This definition of dimension, which is based on the self-
ability distribution of a small part of an object will be con-
similarity of an object, is called the similarity dimension
gruent with the probability distribution of the whole object if
(Mandelbrot, 1982). To perceive what an object of a
the small part is magnified appropriately. However, not all
noninteger dimension looks like, follow the recursive con-
fractal objects are self-similar. This brings us to the more
struction in Fig. 3 which yields the Koch curve of dimension
general concept of self-affinity.
1.26. The first step in this construction breaks a straight line
into three parts and replaces the middle portion by two Self-Affinity. The definition of self-similarity is based on
segments of equal lengths. In the subsequent stages each the property of equal magnification in all directions as
straight segment is broken into three parts and the middle por- described in Fig. 2b. However, there are many objects in
tion of each segment is replaced by two parts. If this recursion nature which have unequal scaling in different directions. For
is done infinite times then the Koch curve is obtained. This example, consider the function B(t) which represents the x-
curve has some unique mathematical properties which are im- location of a particle in Brownian motion as a function of
portant to this study. Firstly, the curve is continuous but it is time, t, as shown in Fig. 4. If this curve is repeatedly
not differentiable anywhere. The nondifferentiability arises magnified, more and more details keep appearing, which sug-
from the fact that if the curve is repeatedly magnified, more gests that it is in some sense similar to the Koch curve.
and more details of the curve keep appearing. This implies However, since the time t and the location B are two different

Journal of Tribology APRIL 1990, Vol. 112/207

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 B(0

Fig. 4 Plot of the x-location versus time for a particle in Brownian

motion

variance of the probability distribution under the rotation of

the coordinate axes and reflection on any plane. The
homogeneity of a surface implies that the probability distribu-
tion of the heights is independent of the location on the sur-
face. The profile, zix), of such a surface along a straight line
and in any arbitrary direction is of dimension
D = Ds-\ (Mandelbrot, 1982) and is a statistically valid
representation of the surface. Such a profile is typically ob-
tained by stylus measurements or by optical techniques.
Weierstrass-Mandelbrot Function. It is assumed that the
function zix) is continuous at all points. Although this
log TV assumption breaks down at atomic scales, for engineering
D = log m
= 1.26 studies the continuum is assumed in the limiting case of a zero
length scale. It is observed that if the profile of a surface is
Fig. 3 Recursive construction ol a Koch curve of dimension D = 1.26 repeatedly magnified, more and more roughness keeps ap-
pearing and the profile is never smooth (Thomas, 1982b; Ma-
physical entities, they usually scale differently. Therefore the
jumdar, 1989a). Therefore, the function zix) is nondifferen-
function Bit) is not self-similar but self-affine. This can be
tiable. In addition, when the profile is magnified, there is a
quantitatively shown by the statistics of the Brownian func-
striking resemblance in the probability distributions of the
tion which follow the relation (Mandelbrot, 1985)
magnified and the original surface. Therefore, in all generality
B(yt) = yl/2B(t) Assuming 5(0) = 0 (6) the function zix) is self-affine. The Weierstrass-Mandelbrot
where B is the probability distribution of B. Equation (6) (W-M) function satisfies all these mathematical properties and
shows that if time t scales by y then the position of the particle was therefore used for characterization of the function z(x)
B scales by yxn. Since the time and the position scale dif- (Majumdar and Tien, 1989b). This function has a fractal
ferently, the Brownian function B{t) is statistically self-affine. dimension, D, and is given as
One important example of self-affine fractals is the
generalization of Brownian motion to the whole class of func- (8)
tions called fractional Brownian motion or fBm (Mandelbrot
and van Ness, 1968). The statistics of this generalized Brown-
ian motion follow the relation where G is a scaling constant and the frequency modes y" cor-
respond to the reciprocal of the wavelength of roughness as
<(fi(t1)-B{t2))2>-\t1-t2\ 2H (7) y" = l/\„.
where < > implies temporal average. Brownian motion is a Properties of W-M Function. The interesting mathe-
special case of this function when H= 1/2. The dimension of matical properties of the W-M function are that the series for
self-affine fractals cannot be obtained from equation (5) z(x) converges whereas that for dz/dx diverges, which means
which is based on the self-similarity of an object. Instead, that it is non-differentiable (Berry and Lewis, 1980). The W-M
Voss (1985) showed that the dimension is related to the function also follows the relation
parameter H by the relation D = E+\-H where E is equal to
the number of arguments of the function B. In the above ex- Z(yx) = yV~Dh(x) (9)
ample since t is the only argument of B, the dimension of the which makes it a self-affine function since the scaling for x
function is D = 2-H. Mandelbrot (1985) showed that the and z are unequal.
lengths of self-affine fractal curves do not follow the relation Although the W-M function appears similar to a Fourier
L ~ e1 ~D for all values of e and therefore the dimension of self- series the difference is that in a Fourier series, the frequencies
affine curves cannot be obtained by measuring their lengths. increase in an arithmetic progression whereas in equation (8)
Instead, the dimension of self-affine functions can be ob- they increase in a geometric progression. In an arithmetic pro-
tained from their power spectra. gression, n, 2n, 3« , it is likely that the phases of some of
Fractional Brownian motion is very important to the study the frequency modes will coincide at some points which will
of rough surfaces because, as will be shown later, the profiles make the function appear non-random. The phases of the fre-
of rough surfaces are self-affine and follow the statistics of quency modes of zix) are randomized by choosing a
equation (7) with the time substituted by a spatial coordinate. nohinteger y and taking its powers to form a geometric series.
With this mathematical background of fractal geometry, a Therefore, if y is chosen to be a fraction, then the phases of
scale-invariant characterization of rough surfaces is now the frequency modes in the W-M function will never match
presented. and the profile of zix) will appear random. Since y also deter-
mines the density of the spectrum it was shown that a choice of
3. Fractal Characterization of Rough Surfaces 7 = 1.5 provides both the phase randomization and high spec-
Consider an isotropic and homogeneous rough surface of tral density (Majumdar and Tien, 1989b).
dimension Ds. The property of isotropy relates to the in- The parameter nx corresponds to the low cut-off frequency

208/Vol. 112, APRIL 1990 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 of the profile. Since a surface is a non-stationary random Case Study. If the dimension of the surface profile is
process (Sayles and Thomas, 1978), the low cut-off frequency chosen to be the limiting case of D = 2, the spectrum behaves
depends on the length L of the sample by the relation, as S(to)~ 1/co. If it is assumed that coh> >oi/ then the r.m.s.
7 = 1//,.
height varies as In(&>A/a>/), the r.m.s. slope varies as uh and the
r.m.s. curvature varies as co2,. Now if the surface profile is
Power Spectrum. Since z(x) is composed of a superposi- magnified by a length-scaling factor |8 > 1, it increases the
tion of infinite frequency modes, it is a multiscale function. spatial resolution and decreases the sample length and
The multiscale nature of this function is displayed by the transforms o>/ and oih to (8w/ and 0coh, respectively. Therefore,
power spectrum of the function which provides the amplitude one would predict that the r.m.s. height would remain cons-
of the roughness at all length scales. The power spectrum of tant, the r.m.s. slope would increase as (3 and the r.m.s. cur-
the W-M function is given as (Berry and Lewis, 1980) vature would-increase as /32. Figure 5 shows such trends for the
. G 2 < fl -" 1 r.m.s. curvature and r.m.s. slope for a magnetic tape surface
(Bhushan et al., 1988). The trend for r.m.s. height shows a
power law of fl~0A suggesting a fractal dimension of 1.9 for
The power spectrum is the most fundamental quantity of such the surface profile. The longitudinal direction of the tape cor-
a multiscale function because most statistical parameters can responds to the x axis whereas the transverse direction cor-
be obtained from the spectrum. Some of the parameters which responds to the y axis.
are relevant to this study are
Application to Elastic Contact Theory. The theory of
elastic contact between rough surfaces (Greenwood and
< (*(*,) - z(x2))2 > = [ S(a>)(e""<*i -*2> - l)du> Williamson, 1966; McCool, 1986) is based on the statistical
J — 00 parameter a = {m0m^)/m\ which determines the asperity den-
sity of a rough surface (Nayak, 1971). In terms of equations
G2(fl_1> C2D-3W
; (12)-(14) and assuming coA > >OJ, it follows that
( 4 - 2 D ) I n 7 ~r v( 2 £ > - 3" ')"s "h / 2„ lx!-x 2 l< 4 - 2fl » (11)
pwA QV.D-1) l {D-D2 (4-2D)
(16)
2
<(Z) > =m0= S(co)dco = — £>(2-£>) ( — )
Jw/ 21n7 (4-2Z>)
It can be seen that for D<2, the parameter a depends on the
(12) instrument-dependent uh and is therefore not unique.
4 4 25 The Greenwood-Williamson (G-W) model is based on Hert-
L / ^ ) " a,,' - '/ zian contact between a smooth plane and a large number of
spheres, the centers of which are distributed normally about
the mean plane. All the spheres are assumed to be of the same
size with the radius equal to the mean radius of asperity cur-
o>2S(a>)dw = - (4 - >-cof - ) 2B 2 D 2)
(13)
vature. It has been shown that the mean radius of curvature is
CMP-1) i a function of the resolution of the instrument and therefore
21n7 (2D-2)
not unique (Bhushan et al., 1988).
d2z \ 2 Whitehouse and Phillips (1978) assumed a surface to be an
<
(-^") > isotropic random process and expressed the parameters of
tribological interest in terms of three statistical quantities
,WA anD-D j
2D _ , .2D\ (14) which are the standard deviation a = \frrT0, <j>, and </>2. The
= w,
parameters 4>i and <j>2 are related to the autocorrelation func-
tion R(j) as </>! =R(r)/R(0) and <j>2=R(2r)/R(0), where r is the
<Z(X,MAT2)> resolution of the instrument. It should be noted that the high-
frequency limit, o)h, is related to the resolution of the instru-
Ql(D-l) 2 £ > 3
won « • ( - )T, I (4-2D)
= mn — T(2D - 3)sin 1 xx (15)
2(4-2D)ln7
200.0
1 1 i i <
A comparison of equation (11) with equation (7) shows that & r.m.s. x -curvature ° i
00.0
since H=2-D, the W-M function is a self-affine fractional O r.m.s. y -curvature * s7
$ r.m.s. peak curvature V (
Brownian motion.
50.0 _ A mean peak curvature
/c
Equations (12)-(14) provide the variances of the height, the A r.m.s. x -slope
V r.m.s. y -slope ^
slope and the curvature respectively, between the frequencies a mean peak slope
oil and uh. These parameters are conventionally used for sur- *\X D r.m.s. height «>
face characterization and are typically obtained from stylus © r.m.s. peak height 1
measurements of surface profiles. For such measurements the g IO.O
low frequency limit, ut, corresponds to the length of the sam-
+/ vp' -
ple whereas the high frequency limit, uh, corresponds to the
Nyquist frequency which is related to the resolution of the in- X 4> ^^' ^s&
^f
strument. Equations (12)-(14) show that these statistical
parameters are functions of the two length scales and their p-o.i
dependence involves the fractal dimension of the surface. The HEIGHT e ^
variances of the height, slope and the curvature alone provide ;
1 1 i *•
little information about the multiscale structure of the surface. Of 2.0 6.0 S.O 1 0 . 0
In addition, however, if the length of the sample and the
resolution of the instrument are known, equations (12)-(14) MAGNIFICATION, p
can be used to predict the fractal dimension D and the Fig. 5 Variation of r.m.s. height, r.m.s. peak height, r.m.s. slope, mean
peak slope, r.m.s. curvature, r.m.s. peak curvature and mean peak cur-
parameter G which characterize the roughness at all length vature with magnification /3 for a magnetic tape surface. Data obtained
scales. from Bhushan et al. (1988).

Journal of Tribology APRIL 1990, Vol. 112/209

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 ment by the relation r=l/(2wh). Assuming wh>>u>h the Table 1 Scale-dependence of statistical surface parameters
autocorrelation function of equation (15) can be written as on the instrument length scales, co, and coA, and \<D<2.
R(j) = a2(\-C{w,T)^2D^ (17) Parameters Dependence on co,, and to,
where C=r(2Z>-3)sin(2Z>-3)7i72. Using equation (17), <j>{ (D-2)
and 4>2 can be found to be
(fl-2)
-2D) <">/•
*, = 1 - C 4m2, ap (P-i)
\2whJ "A
(4-2D) mA, ap, K <4
4>2 = 1 - C (18)
(—)
All three parameters, a, 4>l a n d <j>2, are functions of co, a n d co,, T o study the mechanisms of contact between two rough sur-
corresponding to the two length scales. Also note that since 0 , faces it is necessary to know the structure of roughness at all
and </>2 a r e functions of the ratio u/oo/,, they remain invariant length scales a n d the size-distribution of contact spots at the
under scaling of the surface by a factor /3. In terms of these contact interface. T h e former is provided by equation (25) a n d
quantities, the parameters of tribological interest such as the the size-distribution will be discussed later.
mean peak height Zp, the standard deviation of the peak
heights op, the mean slope a'p, the mean peak curvature R and Obtaining Characterization Parameters From Ex-
the standard deviation of the peak curvature a'p are found to periments. The parameters G a n d D characterize the W - M
be (Bhushan, 1984a) function a n d its spectrum in equation (10). W h e n the W - M
spectrum of equation (10) is compared to the power spectrum
„ ff[(l-0,)/T],/2 of a real surface profile, the dimension D is related to the slope
(19) of the spectrum on a log-log plot. T h e constant G determines
2i?
the location of the spectrum along the power axis and is a
characteristic length scale of a surface. T h e value of G can be
••!+- (20) obtained by equating the variance from equation (12), within a
( # 2irr)tan(7)Tr) 4TT?)2
given spectral bandwidth, t o that of the actual profile. Note
that both G and D are independent of co a n d therefore form a
^ = 2 ^ - — - ) (21) fundamental set of scale-invariant parameters to characterize
isotropic rough surfaces.
2co2,g(3-4</>1+</>2) Figure 6a shows the topography of a magnetic tape surface,
(22)
ijWl-*,)]"2 of r.m.s. height 17.1 nm at 200X magnification, which was
measured by a noncontact optical profilometer (Wyant et al.,
/ qffP;Vy __4^(3
4^(3-40!+02) 1986) with a light wavelength of 652.1nm. The details of the
\ (7 / 7TI magnetic tape are available in Bhushan et al. (1988). T h e
isotropy or anisotropy of the topography can be qualitatively
2[(3-4<£ 1+ tf> 2 )(l-</> 2 )]" 2 3-40,+02 observed in the two-dimensional power spectrum, as shown in
[8TT(1 - * , ) -
iT Fig. 6b. It can be seen that the spectrum is biased in the x and
the y directions, a n d is therefore anisotropic. This bias in the
(23) spectrum is due t o the anisotropy of unidirectionally rolling
the tape during fabrication, which is described in Bhushan
where i\ is (1989b). T h e averaged power spectra of the profiles in the x
and the y directions are shown in Fig. 6c. Although there are
1 _1/3-4^1+02\ 1/2
two distinct spectral regions, the power law behavior of the
(24)
spectra is seen in both these regions. In the low frequency
spectral region, the spectra behave as co -1 - 56 which cor-
responds to a fractal dimension of Z>=1.72, according to
Case Study. If the fractal dimension is chosen to be the
equation (10). In the high frequency region, the spectra behave
limiting case of D = 2 as was done earlier, then under a
as co - 3 - 3 7 . Equation (10) suggests that a profile is fractal only
magnification by the length-scaling factor j3, the tribological
when the power law of its spectrum falls in the range between
parameters vary as op~@, K ~ / 3 2 , op~(}2 whereas ap remains
co - 3 and co~', corresponding to D= 1 and D = 2, respectively.
constant. T h e trends for the peak curvatures a n d peak slope
If a spectrum follows a power law o>~" with n > 3 , then the in-
are experimentally observed from the data obtained for
tegral for the variance of slope, j™co2S(a>)cfcu, converges to a
magnetic tape surfaces by Bhushan et al. (1988), as shown in
finite value. Therefore if the slope of a profile is finite, it can
Fig. 5. The trend for r . m . s . peak height follows a power law of
be concluded that the dimension is D= 1. However, a dimen-
/ 3 - 0 - 1 suggesting a fractal dimension of 1.9 for the surface
sion of one does not imply that the surface is perfectly flat at
profile.
the length scales considered. T h e spectra suggest that the sur-
For both the Nayak a n d the Whitehouse-Phillips models, face consists of roughness structure at all scales. But the steep
the variations of the statistical parameters with respect to the decent of spectra implies that the contribution of the small
two bounds on the length scales, co, a n d coA, are provided in roughness structures to the area of the surface is negligible.
Table 1. Since these parameters are instrument-dependent, Therefore, although the area of the surface will still increase as
they m a y cause errors in the prediction of elastic contact. A the unit of measurement is reduced, the area will not increase
possible solution t o this problem is to develop a new theory to infinity but will converge to a finite quantity. It should be
which would use the multiscale structure of a small asperity noted that although the contribution of the smaller roughness
resting on big asperity which rests on an even bigger asperity in features to the surface area is negligible, their influence on a
a self-affine fashion. The W - M function provides the structure surface p h e n o m e n o n need not be insignificant. It will be worth
of roughness at any length scale / in the form examining whether the fractal or nonfractal behavior exists at
2-KX other length scales. Nevertheless, it is interesting t o note that
z=QiD-l)p-D)cos. where l/2<x<l/2 (25) in both cases, the spectra behave as power laws, a result which

210/Vol.112,APRIL1990 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 .m.s. height = 59.9nm height - 2.77nm

Distance (microns)
£0 120 IBB 240
(a)
Distance (microns)
X
(a) ~~

-2500 -1250 0 1250 2S00

Spatial frequency (1/mm)
-51E -258 0 25S 51E
S p a t i a l frequency (1/mm)
(10

10 13
3 '

£ 10"'
o
D x -direction
v y -direction

104 105 10"

104 105 106
FREQUENCY, CO [m~']
FREQUENCY, CO [m _1 ]
(c) (=)
Fig. 6 Roughness measurements on a magnetic tape surface (a) Fig. 7 Roughness measurements on a smooth magnetic thin film rigid
topography (b) two-dimensional power spectrum (c) averaged power disk (a) topography (b) two-dimensional power spectrum (c) averaged
spectra in x and y directions power spectra in x and y directions

suggests some scaling relation between the roughness structure textured channels along the x-direction which appear quite
at different length scales. periodic in the ^-direction, which correspond to the cir-
The surface topography of a smooth magnetic thin film cumferential and radial directions, respectively. The
rigid disk was obtained by the optical profiler and is shown in anisotropy is also seen in the two-dimensional spectrum of
Fig. la. The wavelength of light used was 653. \nm and the Fig. 8£>. The averaged spectra follow power laws of co~2-38
disk used was of type C as described by Bhushan and Doerner (D=1.31) in the ^-direction and CJ~3-52 (D=l) in the y-
(1989c). The topography appears random and isotropic, as direction, as shown in Fig 8c. Due to the presence of textured
can be qualitatively seen in the two-dimensional power spec- grooves along the x-direction, the .y-spectrum contains more
trum in Fig. lb. The circumferential direction corresponds to power than the x-spectrum. However, the trends in the y-
the x axis whereas the radial direction corresponds to the y, spectrum suggests that at higher frequencies, undetected by
axis. The spectra in the x and the y directions coincide and the optical profiler, the spectra in the two directions would
follow a power law of co-2-24 corresponding to a fractal dimen- meet. Although the roughness in the two directions follow a
sion of 1.38 at the length scales shown in Fig. 1c. In com- fractal and non-fractal structure, the spectra of both are seen
parison to a smooth disk, Fig. 8 shows the measurements of a to follow power laws, as observed for the previous case.
textured magnetic thin film rigid disk which was optically The profiles and the spectra of lapped and ground stainless
measured by a wavelength of 653.1M/M. The disk used was of surfaces are shown in Figs. 9a and 9b, respectively. The
type Bl as described by Bhushan and Doerner (1989c). The roughness measurements were made by traversing a diamond
topography in Fig. 8a is strongly anisotropic and consists of stylus, of radius 2.5/iW, in contact with the surface. The in-

Journal of Tribology APRIL 1990, Vol. 112/211

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 r.m.s. h e i g h t = 6.91nm _ 0-6
o
"
•S
0.3
0.0
-0.3

" -o.et
v^^^
Fractal Simulation

(a)
xz 0.6

••'Vi/VyiA'^^

Horizontal Distance, x {mm)

0 ' G0 120 IBS 240
Distance (microtis)
(a)

(*)

" 0 1 2 3 4 5
Horizontal Distance, x (mm)
-51E -259 0 259 51E
Spatial frequency (1/mm)

(=)

„ 10"'

1 10" 2 3

B LAPPED STAINLESS STEEL

D J-direction
g i<r" A GHOUND STAINLESS STEEL

o v v -direction
102 103 104 105

FREQUENCY, a [m~

Fig. 9 Comparison of fractal and real profiles of (a) lapped stainless

steel surface (b) ground stainless steel surface, (c) Power spectra of real
profiles of (a) and (b).
FREQUENCY, ra [m ']
(=)
Fig. 8 Roughness measurements on a textured magnetic thin film
rigid disk (a) topography (b) two-dimensional power spectrum (c) ave- wheel or the tool radius for turning. Beyond this frequency of
raged power spectra in x and y directions influence, machining does not affect the spectrum and the sur-
face behaves as an unprocessed one. It is in this unprocessed
spectral region that the power spectra of the two surfaces coin-
cide. The parameters D and G in the machined spectral region
strument had a vertical resolution of 5 Angstroms and scan of a variety of stainless steel surfaces have been provided by
lengths of 4mm to 5mm, with each scan having 800 to 1000 Majumdar and Tien (1989b). It is observed that machining
evenly spaced data points. Figure 9c shows that in the high- flattens a power spectrum and thus increases the fractal
frequency region, the spectra of both the stainless steel dimension.
samples coincided and followed a power law of to -2 (D= 1.5) To answer the question as to why are some surfaces fractal
with G = 0.98xl0-'°m (Majumdar and Tien, 1989b). The and some nonfractal, it is first necessary to address the ques-
dimension of 1.5 corresponds to a Brownian process. The tions—how are surfaces formed and how does surface
coincidence of the spectra at high frequencies has been seen processing change the roughness structure? Most surfaces
not only for stainless steel samples but also for other materials used in engineering are initially formed by either fracture of a
such as glass and sandstone (Brown and Scholz, 1985). solid (Mandelbrot et al., 1984; Termonia and Meakin, 1986),
However, at lower frequencies the spectra of the lapped sur- deposition of materials (Meakin, 1987) or solidification of a li-
face profile diverges. The reason for this lies in the type of quid (Vicsek, 1986). It has been shown that the surfaces form-
machining used to polish the surface. Machining reduces the ed by these natural and random processes produce fractal
amplitude of roughness and therefore the power in the spec- structures. However, when a surface is processed, in some
trum. However, machining can reduce the power only till a cases it produces a fractal structure whereas in others it does
particular frequency in the spectrum which is determined by not, as is evident in Figs. 6c-9c. The nonfractal spectra are
the type of the tool used. Examples of the smallest length scale observed for the high frequency region of the tape surface in
affected by machining include the grain size of a grinding Fig. 6c, and in the j>-direction of the textured disk surface in

212/Vol. 112, APRIL 1990 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Fig 8c. It should be noted both these surfaces are strongly are not limited by an instrument resolution. Since the
anisotropic and the bias in their spectra results from the bias in parameters G and D remain constant at high frequencies, as
the processing technique - directional rolling for the tape sur- indicated in Fig. 9c, such simulations provide details of
face and directional grooves for the textured disk. However, roughness at higher frequencies which would remain
when the processing technique is random and unbiased in undetected by experiments. Surface simulations also help in
direction at the length scales considered, such as the free simulating surface phenomena at the length scales of the
abrasive slurry process for the smooth disk (Bhushan, 1989a) phenomena which may be difficult to observe or control
and lapping/grinding for the stainless steel surface, the spectra experimentally.
are seen to be fractal in nature. The fractal structure is also
observed for the x-spectra of the textured disk. This is because 4. Distribution of Contact Spots
the bias in this spectrum is in the /-direction whereas in other
directions, artificial texturing is not present. Although When two surfaces come in contact, the presence of surface
grinding is an anisotropic process at macroscales, the ran- roughness causes an imperfect contact at their interface. Due
domness and isotropy in the distribution of the grains at the to the multiscale nature of rough surfaces, the imperfect con-
microscale makes the surface isotropic at such scales. It is tact is composed of a large number of contact spots of all
probably for this reason that the ground and lapped stainless length scales. To understand the mechanics of static and
surfaces are fractal at the length scales considered in Fig. 9c. dynamic contact, which are responsible for phenomena of
In summary, it can be concluded that natural surfaces are contact conductance, friction, wear and lubrication, it is very
formed by random phenomena and are observed to have a crucial to have a knowlege of the size-distribution of contact
fractal structure. For processed surfaces, the measurements of spots.
Figs. 6-9 suggest that when the processing technique is ran- Number-Area Rule for Islands. The contact between two
dom and directionally unbiased within a range of length rough surfaces can be modeled as one surface contacting a
scales, the surfaces remain fractal in this range. However, smooth plane. To study the distribution of contact spots, it is
when the processing produces deterministically biased texture therefore sufficient to inspect the horizontal cross-sections of
on the surface, the surface is observed to be nonfractal within a rough surface at different heights. Figure 10 shows horizon-
the range of length scales affected by the texturing. It is in- tal cuts of a surface, which was simulated by the W-M func-
teresting to note that in all cases, fractal and non-fractal, the tion, at different heights above the mean plane (Majumdar
spectra of the profiles follow power laws. This observation and Tien, 1989c). It can be seen that a large number of contact
needs to be effectively used in modeling any tribological spots of different sizes coexist at the contact plane and are
phenomenon. spread randomly over the surface. Such a scenario could be
Although the effects of wear of a surface on the fractal visualized as a distribution of islands spread over the earth's
characteristics have not been experimentally observed, some surface where the dark regions are the projected areas of the
speculations can be made from the background developed so islands and the blank region is the ocean. Mandelbrot (1975,
far. When two surfaces are sliding against each other, the run- 1982), while studying the geomorphology of the earth, showed
in wear which occurs is a globally random process on a that the total number of islands of areas greater than a par-
macroscale. Therefore, run-in wear is expected to have the ticular area, a, follows the power law
propensity to make a surface fractal. However, on a
microscale the localized scratches and dents form deter- N(A>a)~a~B where B = D/2 (26)
ministic texture on the surface. Therefore, at microscales a and D = Ds—l is the fractal dimension of coastlines. The
worn surface is expected to be nonfractal. equality in this equation can be invoked by using aL to be the
area of the largest island to yield
Simulation. Besides characterizing a rough surface, the
W-M function can also be used to simulate a surface profile. N(A>a) = aBLa-B (27)
The characterization parameters obtained from the ex- It will be shown later that the contact spots of engineering sur-
perimental spectra can be used in the W-M function to faces also follow this distribution. From equation (27) the
generate artificial profiles. Figures 9a,b show typical simula- number of islands of area lying between a and a + da can be
tions of a lapped and a ground surface, respectively. It can be obtained as
seen that there is a striking statistical resemblance be-
tween the fractal and real profiles. Although the fractal profile n{a)da=-dN=BaBLa~^B+l)da (28)
was generated by a purely deterministic function, the phase The total area of all the islands, At, can now be found to be
randomization technique makes it appear totally random. By
appropriately transforming a profile spectrum to a surface B D
spectrum (Nayak, 1973b), fractal surfaces can also be ar- A,= \ n(a)ada = (29)
\-B 2-D
tificially created by a generalization of the W-M function in
the x and y directions (Majumdar and Tien, 1989b). The ad- There are several features of the power-law distribution of
vantages of such simulations are that unlike experiments, they equation (27) which are worth discussing. Firstly, the distribu-

Fig. 10 Horizontal cross-sections in the x-y plane of a simulated rough

surface at various heights above the mean plane

Journal of Tribology APRIL1990, Vol. 112/213

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Table 2 Scale-dependence of the contact parameters 10"
predicted by elastic theory, on the instrument length scales o>; MAGNETIC TAPE
and w,, for \<D<2. «^ -0.56
a 38MPa
Contact parameter Dependence on uA and «/ i o 3 -- ^\^" V 69KPa

Total contact area, A,

/OM ^ «
2 g 10* -
Number of contact spots, £ .f-^uf s
..£1/2 -1.47 IP

Mean real pressure, pr

-
o\
9 MAGNETIC RIGID DISK
•\

tion displays the multiscale nature of contact spots since it is 10" 1 1 1

not restricted by a smallest length scale. Secondly, the 10° 101 102 103 I0 4
distribution is fully determined by the area of the largest island AREA, a [Jim 2 ]
aL and the fractal dimension of the surface. Since the earth's
surface is a magnified image of an engineering surface, it is Fig. 11 Size-distributions of contact spots for a magnetic tape surface
natural to expect the distribution of islands on the earth's sur- under two different loads and a smooth magnetic thin film rigid disk.
Data obtained from Bhushan (1985).
face to be the same as that of contact spots for an engineering
surface. The only difference lies in the area of the largest
island which determines the complete distribution in equation
(27). Thirdly, although the number of infinitesimally small where ac is the area of a cavity and aLc is the area of the largest
spots is infinitely large, their contribution to the total area of cavity. Since the sum of the total areas of the contact spots
contact is negligible. Therefore, the total area of contact, An and the cavities should be equal to the apparent area of the
can be written in terms of aL as shown in equation (29). This surface, Aa, it follows that
equation also indicates that the knowledge of A, and the
dimension of a surface fully determines the size-distribution of D
(aLc + aL) (34)
contact spots given by equation (27). 2-D
Conventional theories of elastic contact between rough sur- Besides the size-distribution of cavities, the knowledge of
faces, predict the total area of contact At, the number of con- the cavity depths is also important for the prediction of
tact spots £ and the mean real pressure/),, as (Bhushan, 1984b; lubricant-retention or wear-debris capacity. Equation (25)
Bhushan and Dugger, 1989b) provides a relation between the roughness height z and the
3.2P length scale of roughness /. Using the analogy of an asperity
Ar (30) and a cavity, the depth of a cavity can be found to be of the
E\lapk order of z - G ^ - ^ a ' 2 " ^ 2 .
2.64P The size-distribution of contact spots and cavities presented
(31) in this paper facilitates the study of contact phenomena at all
{op/k)E\lapK length scales. It therefore provides a more fundamental ap-
proach to understand problems involving microcontacts.
pr~0.32E^7n (32)
Size-Distributions of Contact Spots. Although the idea of
where P is the load on the surface and E is the composite using results from the geomorphology of the earth to predict
elastic modulus. Table 1 shows that when uh>>wh the size-distribution of contact spots seems lucrative it is necessary
dependence of ap and K on the length scales are ap~cdf{2~D') to verify it experimentally. Bhushan (1985, 1989b) used an op-
and K~aif. From this, the total predicted area varies as tical technique to observe the contact of a smooth glass plane
j4 r -~wj 2 ~ fl)/2 /«j? /2 . This shows that if wh or the resolution of with a magnetic tape surface and a smooth magnetic thin film
the measuring instrument is increased the predicted total area rigid disk, under different load conditions. The images of the
would decrease. The reason for this ambiguous result lies in contact plane were digitized and the numbers and sizes of con-
the fact that the predicted total area varies as square root of tact spots were calculated. The results of the size distribution
the mean radius of curvature of the asperities. Therefore, under nominal pressures of 69 kPa and 38 MPa are depicted in
when the resolution of instrument increases, smaller asperity Fig. 11 and have several interesting features worth discussing.
peaks are detected which decrease the value of the mean radius For the magnetic tape surface, it is observed that although
of curvature. The scale-dependence of the other contact the size-distribution of the contact spots does follow a power
parameters is shown in Table 2. Note that since all the contact law, the exponent of area is not universal at all length scales.
parameters depend on the surface measuring instrument the Since the change in the exponent corresponds to a change in
elastic contact theories do not have unique predictive dimension, it is conjectured that this corresponds a similar
capabilities. This justifies the suggestion that a new contact change at the corner frequency of the spectrum. For the
theory, based on scale-independent parameters, needs to be smooth thin film rigid disk, the power law behavior was
developed. observed although the corner in the distribution was not
In some applications such as contact conductance, lubrica- observed.
tion and wear, the cavity size-distribution is also important. If The size-distribution of equation (27) is fully determined by
a surface is turned upside down then a mountain becomes a the area of the largest island. Since the number of the largest
cavity and vice versa. Therefore, it is expected that if the island is always unity, the distribution is fixed along the
probability distribution of a surface is symmetric, the sizes of number axis and can move only along the area axis when
cavities will follow a distribution similar to equation (27) and graphically plotted. This is exactly what is seen in Fig. 11.
is given as When the load on the magnetic tape surface is increased from
69 kPa to 38 MPa, the power law behavior of size-distribution
Nc{Ac*ac)=(-f-) (33) does not change at all but the distribution moves horizontally

214/Vol. 112, APRIL 1990 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 along the area axis. This feature is clearly seen by the move- D

ment of the corner in the power law. Such an observation con- ha ( P \ —

firms the idea of using the size-distribution of equation (27) ~AD)\HA-) 2 (37)
for contact spots.
At smaller length scales the size-distribution follows the This has the same form as equation (35) with a' replaced by a
relation N(A>a)~a~0-54 or a~0-56 suggesting that the dimen- function of the dimension f(D). The network analysis shows
sion of the surface profile corresponds to 1.08 or 1.12. that the exponent x, which was previously obtained empirical-
However, at larger length scales the exponent of - 1.47 does ly, is related to the spectrum of the rough surface and is equal
not correspond to any value of dimension lying between 1 and to D/2. Since D varies between 1 and 2 the exponent is ex-
2. It should be noted that the limits of the dimension D = 1 and pected to vary between 0.5 and 1. The reason it lies between
D = 2 correspond to spectral variations of 1/co3 and 1/co, 0.85 and 0.99 is due to surface machining which increase the
respectively. A 1/co spectrum therefore corresponds to an area value of dimension by flattening the spectra, as shown in Fig.
exponent of - 1 in the size-distribution. Since the observed ex- 9c.
ponent is - 1.47, it suggests that the spectrum must be of the
form 1/co" where « < 1 . Since the exponent of - 1 . 4 7 cannot 5. Conclusions and Future Applications
be explained by the number-area rule it is necessary to in- Roughness measurements on surfaces of magnetic tape, tex-
vestigate the origins of such a rule. The number-area rule for tured and smooth magnetic thin film rigid disks, and lapped
islands was first obtained from empirical correlations by Kor- and ground stainless steel surfaces show that when the surface
cak (1938). Mandelbrot (1982) theoretically proved that the processing technique is random and directionally unbiased
area-exponent obtained from correlations is the fractal dimen- within a range of length scales, the surface produced is fractal
sion of coastlines. However, there seems to be no relation bet- within that range. Processes which produce deterministic tex-
ween the surface spectrum and the area-exponent in equation ture within a range of scales, result in non-fractal surfaces
(26). Since surface profiles are fractional Brownian motions within that range. However, the spectra of both fractal and
which are characterized by their spectra, an investigation in non-fractal surfaces follow power laws which suggest scaling
the relation between the surface spectrum and the area- relations between the roughness structures at different length
exponent is called for. scales.
In summary, the idea of using a power law of the form This study shows that when the power law spectra of surface
N(A>a)~a~B is verified by the experiments and will be profiles fall within the fractal range, the multiscale roughness
useful in analyzing contact mechanics. However, not all ex- structure can be characterized by fractal geometry. While con-
ponents obtained from the experiments can be related to the ventional methods of characterization use statistical
fractal dimension. This warrants an investigation into the rela- parameters which depend on either the resolution of the
tion between the spectral exponent and the area-exponent of measuring instrument or the length of the sample, the present
the size-distribution. method uses scale-independent fractal parameters. In contrast
with conventional statistical parameters, the fractal
Application to Contact Conductance. When a com-
parameters provide the structure of roughness at any length
pressive load is applied between two nominally flat surfaces,
scale and can therefore facilitate the study of a surface
the imperfect contact at the interface imposes a resistance to
phenomenon at all its relevant length scales. In addition, the
the flow of heat and electricity. The classical relation between
size-distribution of contact spots is also characterized by the
the conductance h and the load P is given as (Yovanovich,
fractal representation by a power law number-area rule.
1987)
However, the relation between the spectral exponent and the
ha / P \x exponent for the number-area rule needs to be investigated.
= o'\ ) (35) The conventional theories of elastic contact use statistical
X \HAJ parameters which are scale-dependent. This suggests that new
contact theories, which would depend on scale-independent
where a is the standard deviation of the surface, X is the con- parameters and would have unique predictive capabilities,
ductivity, a' is the mean slope, His the hardness of the softer should be developed.
material, Aa is the smaller of the apparent areas of the two The spectra of the surfaces and the size-distribution of con-
surfaces and x is an exponent obtained from empirical correla- tact spots are seen to follow power laws for several decades of
tions of experimental observations. length scales. This creates a potential for the fractal approach
This paper has shown that the mean slope a' is an to form a link between the micro- and macroscale phenomena
instrument-dependent parameter which is not unique for a which enables it to predict what happens at some length scale
surface and has so far been erroneously used. The other source from what is observed at other scales.
of uncertainity lies in the exponent x which has been ex-
perimentally shown to lie between 0.85 and 0.99. To question
the validity of equation (35) Majumdar and Tien (1989c) Acknowledgments
recently developed a network model for contact conductance The authors would like to thank Professor C. L. Tien for
which considered the contact spots to form a network of con- supporting A. Majumdar during this study, and Professor D.
ductance paths. This model uses the power law distribution of Bogy and E. Krai for providing the experimental facilities.
spots to obtain the effective resistance of all spots of any par-
ticular size and to construct a series-parallel network. For a
length scale /, the roughness is of the form z~l2~D. References
Therefore, the resistance p of a basic roughness element is Berry, M. V., and Lewis, Z. V., 1980, "On the Weierstrass-Mandelbrot Frac-
tal Function," Proceedings of the Royal Society of London, Vol. A 370, pp.
459-484.
Bhushan, B., 1984a, "Prediction of Surface Parameters in Magnetic Media,"
Wear, Vol. 95, pp. 19-27.
Bhushan, B., 1984b, "Analysis of the Real Area of Contact between a
Hence the conductance of the basic element is of the form Polymeric Magnetic Medium and a Rigid Surface," ASME JOURNAL OF
TRIBOLOGY, Vol. 106, pp. 26-34.
h ~ aDn. The complete network of resistances was formed by a Bhushan, B., 1985, "The Real Area of Contact in Polymeric Magnetic Media-
linear combination of the basic resistive element and the total II: Experimental Data Analysis," ASLE Transactions, Vol. 28, pp. 181-197.
contact conductance was found to be Bhushan, B., Wyant, J. C , and Meiling, J., 1988, " A New Three-

Joumal of Tribology APRIL 1990, Vol. 112/215

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Dimensional Non-Contact Digital Optical Profiler," Wear, Vol. 122, pp. Mandelbrot, B. B., Passoja, D. E., and Paullay, A. J., 1984, "Fractal
301-312. Character of Fracture Surfaces of Metals," Nature, Vol. 308, pp. 1571-1572.
Bhushan, B., 1989a, Tribology and Mechanics of Magnetic Storage Devices, Mandelbrot, B. B., 1985, "Self-Affine Fractals and Fractal Dimension,"
Springer-Verlag, New York. Physica Scripta, Vol. 32, pp. 257-260.
Bhushan, B., and Dugger, M. T., 1989b, "Real Contact Area Measurements McCool, J. I., 1986, "Comparison of Models for the Contact of Rough Sur-
on Magnetic Rigid Disks," submitted for publication to Wear. faces," Wear, Vol. 107, pp. 37-60.
Bhushan, B., and Doerner, M. F . , 1989c, "Role of Mechanical Properties McCool, J. I., 1987, "Relating Profile Instrument Measurements to the Func-
and Surface Texture in the Real Area of Contact of Magnetic Rigid Disks," tional Performances of Rough Surfaces," ASME JOURNAL OF TRIBOLOGY, Vol.
JOURNAL OF TRIBOLOGY, in press. 109, pp. 264-270.
Brown, S. R., and Scholz, C. H., 1985, "Closure, of Random Elastic Surfaces
in Contact," Journal of Geophysical Research, Vol. 90, No. B7, pp. 5531-5545. Meakin, P., 1987, "Fractal Scaling in Thin Film Condensation and Material
Feder, J., 1988, Fractals, Plenum Press, New York. Surfaces," CRC Critical Reviews in Solid State and Materials Science, Vol. 13,
Gupta, P. K., and Cook, N. H., 1972, "Junction Deformation Modes for pp. 143-189.
Asperities in Sliding Interactions," Wear, Vol. 20, pp. 73-87. Nayak, P. R., 1971, "Random Process Model of Rough Surfaces," ASME
Greenwood, J. A., and Williamson,,J. B. P., 1966, "Contact of Nominally JOURNAL OF LUBRICATION TECHNOLOGY, Vol. 93, pp. 398-407.
Flat Surfaces," Proceedings of the Royal Society of London, Vol. A 295, pp. Nayak, P. R., 1973a, "Random Process Model of Rough Surfaces in Plastic
300-319. Contact," Wear, Vol. 26, pp. 305-333.
Korcak, J., 1938, "Deux types fondamentaux de distribution statistique," Nayak, P. R., 1973b, "Some Aspects of Surface Roughness Measurements,"
Bulletin de I'Institut International de Statistique, III, pp. 295-299. Wear, Vol. 26, pp. 165-174.
Ling, F. F., 1987, "Scaling Law for Contoured Length of Engineering Sur- Peitgen, H. O., and Saupe, D., 1988, The Science of Fractal /mages,
faces," Journal of Applied Physics, Vol. 62, pp. 2570-2572. Springer-Verlag, New York.
Ling, F. F., 1988, "The Possible Role of Fractal Geometry in Tribology," Sayles, R. S., and Thomas, T. R., 1978, "Surface Topography as a Nonsta-
presented at the STLE-ASME Tribology Conference, to appear in Tribology tionary Random Process," Nature, Vol. 271, pp. 431-434.
Transactions, Vol. 32, 1989. Termonia, Y., and Meakin, P., 1986, "Formation of Fractal Cracks in a
Majumdar, A., 1989a, "Fractal Surfaces and their Applications to Surface Kinetic Fracture Model," Nature, Vol. 320, pp. 429-431.
Phenomena," Ph.D. thesis, Department of Mechanical Engineering, University Thomas, T. R., 1982a, "Defining the Microtopography of Surfaces in Ther-
of California, Berkeley. mal Contact," Wear, Vol. 79, pp. 73-82.
Majumdar, A., and Tien, C. L., 1989b, "Fractal Characterization and
Simulation of Rough Surfaces," submitted for publication to Wear. Thomas, T. R., 1982b, Rough Surfaces, Longman Inc., New York.
Majumdar, A., and Tien, C. L., 1989c, "Fractal Network Model for Contact Vicsec, T., 1986, "Formation of Solidification Patterns in Aggregation
Conductance," to be presented at the ASME/AIChE 26th National Heat Models," in Fractals in Physics, eds. L. Pietronero and E. Tossati, pp. 247-250.
Transfer Conference, Philadelphia, Aug. 6-9. Voss, R. F., 1985, "Random Fractal Forgeries," in Fundamental Algorithms
Mandelbrot, B. B., 1967, "How Long is the Coast of Britain? Statistical Self- for Computer Graphics, ed. R. A. Earnshaw, pp. 805-835.
Similarity and Fractional Dimension," Science, Vol. 155, pp. 636-638. Whitehouse, D. J., and Phillips, M. J., 1978, "Discrete Properties of Ran-
Mandelbrot, B. B., and van Ness, J. W., 1968, "Fractional Brownian Mo- dom Surfaces," Philosophical Transactions of Royal Society of London, Vol. A
tions, Fractional Noises and Applications," SIAM Review, Vol. 10, pp. 290, pp. 267-298.
422-437. Wyant, J. C , Koliopoulos, C. L., Bhushan, B., and Basila, D., 1986,
Mandelbrot, B. B., 1975, "Stochastic Models for the Earth's Relief, the "Development of a Three-Dimensional Noncontact Digital Optical Profiler,"
Shape and the Fractal Dimension of the Coastlines, and the Number-Area Rule ASME JOURNAL OF TRIBOLOGY, Vol. 108, pp. 1-8.
for Islands," Proceedings of the National Academy of Science, USA, Vol. 72, Yovanovich, M. M., 1987, "Theory and Applications of Constriction and
pp. 3825-3828. Spreading Resistance Concepts for Microelectronic Thermal Management,"
Mandelbrot, B. B., 1982, The Fractal Geometry of Nature, W. H. Freeman, Keynote Address, Proceedings of the International Symposium on Cooling
New York. Technology for Electronic Equipment, Honolulu.

216/Vol. 112, APRIL 1990 Transactions of the ASME

Downloaded From: http://tribology.asmedigitalcollection.asme.org/ on 01/17/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use