You are on page 1of 15

Wear, 136 (1990) 313 - 327 313

FRACTAL CHARACTERIZATION AND SIMULATION OF ROUGH


SURFACES

A. MAJUMDAR and C. L. TIEN


Department of Mechanical Engineering, University of California, Berkeley, CA 94720
(U.S.A.)
(Received December 30, 1988; revised July 24, 1989; accepted August 29, 1989)

Summary

Roughness measurements on a variety of machined steel surfaces and a


textured magnetic thin-film disk have shown that their topographies are
multiscale and random. The power spectrum of each of these surfaces fol-
lows a power law within the length scales considered. This spectral behavior
implies that when the surface is repeatedly magnified, statistically similar
images of the surface keep appearing. In this paper the fractal dimension is
identified as an intrinsic property of such a multiscale structure and the
Weierstrass-Mandelbrot (W-M) fractal function is used to introduce a new
and simple method of roughness characterization.
The power spectra of the stainless steel surface profiles coincide at high
frequencies and correspond to a fractal dimension of 1.5. It is speculated
that this coincidence occurs at small length scales because the surface remains
unprocessed at such scales. Surface processing, such as grinding or lapping,
reduces the power at lower frequencies up to a certain corner frequency,
higher than which all surfaces behave as unprocessed ones.
The W-M function is also used to simulate deterministically both
brownian and non-brownian rough surfaces which exhibit statistical resem-
blance to real surfaces.

1. Introduction

The geometric structure of rough surfaces influences a multitude of


physical phenomena which are relevant to many engineering problems such as
contact resistance, friction, wear, sealing, pool boiling and wave scattering.
Conventionally, the deviation of a surface from its mean plane is assumed to
be a random process, for which statistical parameters such as variances of
the height, the slope and the curvature are used for characterization [l].
However, it has been observed that surface topography is a non-stationary
random process [Z], implying that the variance of the height distribution is
related to the length of the sample and hence is not unique for a particular

0043-1648/90/$3.50 0 Elsevier Sequoia/Printed in The Netherlands


314

surface. Rough surfaces are also known to exhibit the feature of geometric
self-similarity and self-affinity [ 3, 4 1, by which similar appearances of the
surface are seen under various degrees of magnification. Since increasing
amounts of detail in the roughness are observed at decreasing length scales
the concepts of slope and curvature, which inherently assume the smooth-
ness of a surface, cannot be defined. So the variances of slope and curvature
depend strongly on the resolution of the roughness-measuring instrument or
some other form of filter and are therefore not unique [ 51.
In contemporary literature such a large number of characterization
parameters occur that the term “parameters rash” is aptly used [6]. This
results from the use of inst~ment~ependent ch~acterization p~amete~
which show different values for the same surface. Therefore it is neces-
sary to characterize rough surfaces by intrinsic parameters which are inde-
pendent of all scales of roughness. This suggests that the fractal dimension
[ 31, which is invariant with length scales and is closely linked to the concept
of geometric self-similarity, is an intrinsic property and should therefore be
used for surface characterization.
In this paper a new characterization method is introduced which uses
the Weierstrass-Mandelbrot (W-M) fractal function to obtain the fractal
dimension and the new characterization parameters. These are determined
by comparing the spectrum of surface profiles with that of the W-M func-
tion. The simplicity of the method is illustra~d by comparing the spectra
of stylus-measured surface profiles of different stainless steel samples with
the W-M spectrum. Using these parameters, the W-M function is employed
to simulate deterministically both brownian and non-brownian surfaces.

2. Fractal characterization

The self-similarity or self-affinity of rough surfaces implies that for a


decreasing unit of measurement the surface area of the rough surface, which
is a measure for two-dimensional objects, tends to infinity whereas the
volume, a three-d~ension~ measure, approaches zero. Therefore, the
Hausdorff or fractal dimension, fl + 1, of rough surfaces is a fraction between
2 and 3 131. In this paper homogeneous and isotropic surfaces are consid-
ered. The homogeneity of a surface implies that the statistical distribution
of the randomness in height is independent of the location on the surface.
The property of isotropy relates to the invariance of the distribution under
rotation of the coordinate axes and reflection in any plane. For such a sur-
face the profile, of dimension D, along a straight line at any location and in
any arbitrary direction is a statistically valid representation of the surface.

2.1. Theory
The profile of a rough surface z(x), typically obtained from stylus mea-
surements, is assumed to be continuous even at the smallest scales. Although
this assumption breaks down at atomic scales, for engineering studies the
315

continuum is assumed to exist down to the limit of a zero-length scale.


Therefore, since ever-finer levels of detail appear under repeated magnifica-
tion, the tangent at any point cannot be defined. Thus the profile has the
mathematical properties of being continuous everywhere but non-differen-
tiable at all points. Surface profiles are also known to be self-affine in rough-
ness structure [3, 41. The Weierstrass-Mandelbrot (W-M) fractal function
[3, 7, 81 satisfies the properties of continuity, non-differentiability and self-
affinity, and is therefore used to characterize and simulate such profiles. The
W-M function has a fractal dimension D, between 1 and 2, and is given by

O” cos 2nynx
z(x) = AcD - ‘) 2 l<D<2 Y>l
f2 -D)n
n= “,
where A is a scaling constant introduced in this study and whose significance
is elaborated on later.
Although the W-M function is similar to a conventional Fourier series
the difference lies in the relation between the frequency modes of their
respective spectra. For a Fourier series the discrete frequencies increase in
arithmetic progression, n, 2n, 3rz . . ., as multiples of a basic frequency. The
presence of harmonics in a Fourier series makes the phases of the several
modes coincide at certain nodes thereby giving a periodic or a non-random
appearance. In contrast, the frequencies 7” of the W-M function form a
spectrum ranging from ~“1 to infinity in geometric progression. The param-
eter y of the W-M function determines the density of the spectrum and the
relative phase differences between the spectral modes. Since the phases in a
real surface profile are random, a simulation of such a profile by the W-M
function requires that the phases of the frequency modes do not coincide at
any value of x and thereby appear random. This is achieved by avoiding the
point x = 0, where the phases coincide, and choosing 7 = 1.5 such that its
powers, which form the frequency spectrum, are not multiples of a basic
frequency.
The parameter n1 corresponds to the low cut-off frequency of the
profile. Since surfaces are non-stationary random processes [2] the lowest
cut-off frequency depends on the length L of the sample and is given by
ynl = l/L.
The W-M function has interesting mathematical properties which are
relevant to the present study. The series for z(x) converges whereas that for
dz/& diverges, implying that it is non-differentiable at all points. The func-
tion also satisfies a self-affinity relation [ 81
2(7X) = y(2-%(3c) (2)
so that when x is magnified by a factor y, the height z is magnified by yC2-D).
The autocorrelation R(T) of the W-M function is
AZ{“- 1) cos 2?Yy”T
jz(x)z(x
+r)dx
m

R(T) = Jim_ f = ----2-- I: Y(4_2D)n (3)


0 ” = “,
316

The power spectrum g(o), which is the Fourier transform of the autocorre-
lation, follows as

St0 - 7”)
$4 - 2D)n (4)

where 6(x) is the Dirac-delta function. The discrete W-M power spectrum
can be approximated by a continuous spectrum which is given as [8]

(5)

This power law behavior is the essence of a fractal characterization of rough


surfaces. Since

S(yw) = y-‘” -=%J(o)

the power law spectrum follows a self-affinity relation.


When eqn. (5) is compared with the power spectrum of a surface, the
dimension D is related to the slope of the spectrum on a log-log plot. The
constant A is a c~aructeristic length scale of a surface which is i~var~~t with
respect to all frequencies of roughness and determines the position of the
spectrum along the power axis. In this new ch~acte~zation method both A
and D are independent of the roughness scales of the surface and are there-
fore intrinsic parameters. It is interesting to note that A is closely related to
the parameter, topothesy, of a surface which was introduced in previous
studies [ 2, 8, 91.
The constants of the W-M function, A, D and nl, form a complete and
fundamental set of scale-independent parameters to characterize an isotropic
rough surface. If these parameters are known, then the roughness structure
at any length scale I can be determined from the W-M function to be of the
form z a: 1(21”). In order to analyze any surface phenomenon, the physics
has to be studied at the length scales of the phenomenon. Every physical
phenomenon has one or more length scales associated with it, For example,
the wavelength of light is the length scale for light scattering by a rough
surface. Since the present characterization method provides the roughness
structure at any arbitrary length scale, it forms the basis of a fundamental
approach to the study of interactions of surface roughness with any physical
phenomenon.
In contrast with conventional statistical parameters, such as the vari-
ances of height mo, slope mz and curvature m4, the new set of parameters is
independent of the roughness length scales and of the resolution of the mea-
suring instruments. For stylus measurements the high frequency cut-off
oh usually depends on the radius of the stylus or some form of instrument
filter whereas the low frequency cut-off o1 is a function of the length of
the sample. The relation between the new and the conventional set of param-
317

eters can be shown to be


*h A?@ - 1) 1 1 1
<(2)2>= ma =
s S(w) do =
2 In r (4 -20) 1 ~~~~~~~ - o$4-2D)
(6)

AZ@ - 1)
02S(w) do = 21ny (2,12)(w$“-2)-~~2D-2))

(7)

((32)=m,=
j+W4S(o)dm= ‘;;’ &(‘+2D-‘$D) (8)
*1

where () denotes the spatial average over x. If it is assumed that o& 9 ol, it
can be seen that the variation with the conventional statistical parameters
follows as m, 0: u1-(4-2DD), m, 0: w$2D-2) and m4 a tihZD. In contrast with
the fractal parameters, these statistical parameters do not provide any infor-
mation about the roughness structure at any length scale, which would be
needed to study a surface phenomenon.

2.2. Application to contact mechanics


The theory of elastic contact between rough surfaces [lo, 111 depends
strongly on the parameter a = (mom4)/mz2 which determines the asperity
density of a rough surface [ 11. In terms of the new characterization param-
eters and assuming that ah B wl, it follows that

It can be observed that for D < 2, cx depends on the instrument-dependent


ah and is therefore not unique. It is well known that the determination of
wh is fraught with difficulties since the length scale beyond which contact
mechanics is unaffected is still unknown [ 111.
The Greenwood-Williamson (G-W) model [lo] for elastic contact is
based on hertzian 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 a radius equal to the
mean radius of asperity curvature. It has been shown in this study that the
mean curvature is a function of the instrument resolution and is therefore
not unique.
Since the parameters in the G-W model are instrument dependent, they
may cause errors in predictions of elastic contact. Therefore it is necessary to
develop a new theory which would use the multiscale nature of a rough
surface. A possible method would be first to determine the size distribution
of contact spots at the contact interface. It has been shown [12] that the
total number N of contact spots of area greater than a follows as N@ > a) a
a-D’2, which is same as the distribution of islands on the earth [ 131. If each
318

spot, of length scale E = a”‘, is assumed to be formed by hertzian contact,


then the load required to create a spot of area a can be determined. There-
fore, for a given contact area A,, the total load can be obtained by adding up
the load on individual asperities, in conjunction with the size distribution.
Moreover, the size distribution can be divided into two regimes - contact
spots in elastic and in plastic contact, whereby an elastic-plastic contact
theory can also be developed. For a contact spot of size 1 = l/w, the W-M
function gives the roughness structure, which is deformed by hertzian
contact, as
A’” - I) 1 1
z(x) = - cos 23wx - --<xx---
w(2 -D) 4 4
Therefore the radius of curvature p of the tip of each asperity at x = 0 is
1
’ = (2x)zWo~@‘- 1) (11)
So the curvature and the deformation of the asperities are functions of the
size of the contact spot. Since the size distribution of contact spots is known,
the total load on the surface can be obtained.

3. Exper~ental details

The application of this new surface characterization method was demon-


strated on roughness measurements obtained from a variety of processed
stainless steel surfaces and a textured magnetic thin-film rigid disk. The dif-
ference in the formation of these two types of surfaces was that for the
stainless steel surfaces, material was removed by polishing it, whereas the
thin-film disk was formed by depositing material by a sputtering process.
Roughness measurements were performed on both these types of surfaces to
find the differences in their roughness structure and to determine whether
they produced fractal spectra.
The profiles of the stainless steel surfaces were measured with a contact
stylus profiler. This ins~ment used a diamond stylus of radius 2.5 pm and
had a vertical resolution of 5 A. The scan lengths ranged from 50 pm to
30 mm with each scan having 300 - 1000 evenly spaced data points. For each
profile the scan length L and the resolution of the data are provided in
Table 1. The power spectra for the profiles were obtained using a Fast
Fourier Transform routine and are shown in Fig. 1.
The topography of the textured thin-film disk, the two-dimensional
power spectrum and the averaged power spectra in the x and the y directions
are shown in Fig. Z(a), Fig. 2(b) and Fig. 2(c) respectively. The topography
of the textured thin-film disk was measured with a non-contact optical pro-
filer [14], which used a wavelength of 653.1 X 10Wg m. The r.m.s. height of
the surface was found to be 3.45 X lOa m, The scan lengths in the x and the
y directions were 240 X 10e6 m and 232 X 10m6 m respectively, with each
319

TABLE 1
Specimens of stainless steel surfaces used for experiments

Specimen Surface Length Resolution Variance A1 DI 0,

processing (mm) 0-W x 10” X 10-9 x 10-4


techniquea (m2) (m) W’)

1 Lapped (4) 4.6 5.0 0.716 12.445 1.882 7.32


2 Lapped (8) 4.0 5.0 6.027 1.254 1.701 4.88
3 Ground (8) 5.0 5.0 8.483 0.119 1.579 4.88
4 Ground (16) 5.0 5.0 74.944 - - -
5 Shape 10.0 10.0 30.639 1.599 1.752 1.95
turned (32)

aNumbers in parenthesis correspond to average roughness (microinches) of the surface, as


specified for General Electric Roughness Specimen (Cat. No. 8651831Gl), which indicate
the degree of surface finish.

scan containing 256 digitized points. The averaging of the power spectra was
done to eliminate the noise in the spectra, which were seen as in the single-
profile spectra of the stainless steel samples.

3.1. Power spectrum


The power spectrum of a surface quantifies the multiscale roughness
features present on it. Figure 1 shows the surface spectra of the stainless
steel samples, in which the amplitude of the roughness decreases at smaller
length scales. Since the spec~m is tetrad at the Nyquist frequency,
the roughness features at smaller length scales remain undetected. Figures 1
and 2 show that the spectra of all the surfaces follow power laws of the form

(12)
Such a power law suggests some scaling relation between the roughness struc-
tures at different length scales.
The power spectra of the stainless steel surface profiles have two distinct
spectral regions, as indicated in Fig. 1. At high frequencies, it is interesting
to note that the power spectra of all the profiles coincide and follow the
same power law of 1 /w2. This corresponds to a brownian process [ 43 with a
fractal dimension of 1.5. Each surface-processing technique reduces the
amplitudes of lower frequencies and is characterized by a different power
law in this spectral region. Brown and Scholz [15] also observed similar
behavior in spectra of surfaces of different materials. The corner created
at the junction of the two spectral regions is represented by a corner fre-
quency w,.
The physical explanation of such spectral behavior is that every surface-
processing technique, such as grinding or lapping, has a finite frequency of
influence beyond which the surface remains unaffected and behaves as an
unprocessed surface. Examples of the scale of influence include the
320

lo-‘*

IO+

z
10-16
s
u,
ci
g lo-‘*

12

1o-Z0

4.88 7.32
I I I Ld
102 103 104 ld 106

(4 FREQUENCY,w [m-l)

1”

10-14

z
lo-16
5
G

“j
* 10-18

&

IO-

~~

1.95
1 L
10-a
10’ ld ld 104 ld d
@I ERFQuENcY, 0 @PI-‘)
Fig. 1. Power spectra of the profiles of stainless steel surfaces: (a} specimens 1, 2 and 3;
(b) specimens 4 and 5.

radius of a turning tool or the grain size of a grinding wheel. This hypothesis
suggests that surfaces with finer finish would have higher comer frequencies.
For a processed surface, the spectrum needs to be characterized only up to
the corner frequency, since beyond o, it is unique for all the stainless steel
surfaces.
The topography of the thin-film rigid disk, shown in Fig, Z(a), shows
textured grooves running along the x axis, corresponding to the circumferen-
tial direction of the disk. These grooves appear quite periodic along the y
axis which corresponds to the radial direction. Such anisotropy can be quali-
321

(a) Distance [x10%] -----fx (b ) Spatial frequency [ 10-3m-1]

10-22
-

z
-5 K+-

ul

d
3 10-M-
2
0 x-diction
10-z v y direction
t
5.25
I I I
lO-=
ld lti ld

(c) FREQUENCY. 0 [m-‘I

Fig. 2. Roughness measurements of a smooth magnetic thin-film rigid disk: (a) topo-
graphy; (b) two-dimensional power spectrum; (c) averaged power spectra in the x and y
directions.

tatively seen in the two-dimensional spectrum in Fig. 2(b) where the bias in
the spectrum is along the y axis. Since the deep grooves are periodic in the y
direction, the y spectrum has higher power than the x spectrum as seen in
Fig. 2(c). It was observed that the x spectrum had two spectral regions. In
the range 7.59 X lo3 (m-l) < o < 5.25 X lo4 (m-l), the spectrum followed
a power law of S(0) 0: w-1-W which corresponds to a fractal dimension of
D, = 1.53. In the second region, i.e. 5.25 X lo4 (m-l) < o < 5.13 X lo5
(m-l), the spec t rum followed a power law of S(o) a w-1*42corresponding to
a fractal dimension of D2 = 1.79. In the spectral region 6.07 X lo4 (m-l) <
w < 5.13 X lo5 (m-l), the y spectrum varied as S(o) 0: o-2*51 which corre-
sponds to a fractal dimension of D = 1.25. In the low frequency range, the
averaged y spectrum oscillated and did not follow any power law owing to
the presence of textured grooves running along the x axis. Since fractal char-
acterization relies on power law spectral behavior and the presence of all
322

frequency modes, it could not be applied in this spectral region where the
periodicity of the grooves produced only a few spikes in the spectrum.

3.2. Characterization parameters


Although the characterization technique has so far been developed for
a single power law, the two spectral regions of the stainless steel specimens
can be easily characterized by splitting the W-M function into two parts,
given as
“c cos 27Tyx cos 27Ty”x
z(x) = A,@-‘) z + A*(% - 1) (13)
+2 -D,)n c
?I=*, n=?lc $2 -D,)n
1<DI,D2<2 Y>l
where y”c is equal to the corner frequency. The continuous power spec-
trum for this function is
AI 2(D, - 1) 1
S(w) = y”‘< w<y”c
2 In y & -20,)

A+ - 1) 1 (14)
= y”c< w<*
w(S -2D,)
2 In y
The complete characterization of such a profile requires the values of
AI, A2, D1, Up, length L and 0,. It was observed that specimen 4 had a
single power law spectrum for the length scales considered and behaved as
an unprocessed surface. The dimension D2 was determined from the average
slope of this spectrum, and found to be equal to 1.5. The value of A2 was
obtained by equating the experimental variance of the profile to that of the
W-M function. Since the bandwidth of the spectrum, o1 < o < oh, was
provided by the length L = l/q of the sample and the Nyquist frequency,
the variance of the W-M function was predicted from R(0) and by the
approximate spectral method of eqn. (6). The first method yielded A2 =
0.98 X lo-” m whereas the spectral method gave A2 = 1.19 X 10-i’ m. For
profiles of the other specimens the corner frequency o, was determined in
the following way. Starting from the lowest frequency and proceeding to-
wards the highest, a least-square analysis was performed on the log S(w)-
log w data for groups of 50 digitized frequencies. Such an analysis provided
the average slope of each group which determined the dimension in that
spectral window. The corner frequency was located when the slope of the
frequency group suddenly decreased as the window was scanned along the
spectrum. The value of D, was obtained by performing a least-square anal-
ysis between the lowest and the corner frequencies. Since the values of A2,
D1, D2, L and o, were known, the value of AI was determined by comparing
the variance of the real profile with that obtained from R(0) of the split
W-M function. Since A2 and D2 were the same for all the profiles, only the
values of AI, D, and w, are listed in Table 1. It can be observed that with
increasingly fine surface finish the value of w, increased as expected.
323

The fractal dimensions of the thin-film rigid disk, in different directions


and spectral regions, were obtained by performing a least-square analysis on
the log-log data. For the x spectrum this analysis provided the values of
Al = 2.18 X lo-l4 (m) and AZ = 1.82-l’ (m) corresponding to D, = 1.53 and
Dz = 1.79 respectively, for the two spectral regions.

4. Fractal simulation

Simulations of surfaces is a subject of current research in diverse fields


such as geophysics, tribology and computer graphics. Such simulations not
only substitute surface measurements by artificially creating surfaces, but
they also facilitate the simulation of surface-related random phenomena
which may be difficult to control experimentally.
After the characterization parameters are determined, the simulation of
surface profiles follows directly by the use of the parameters in the W-M
function. Previous studies have created artificial surfaces by the methods of
tectonic plate movements [3, 161, brownian motion as independent cuts
[ 171, random midpoint displacement [17] and the Weierstrass-Mandelbrot
function [17]. The W-M function provides a means to obtain information
from the spectra of real surface profiles and produce artificial profiles deter-
ministically. The advantage of the W-M method is that the characterization
and the simulation can be done by the same mathematical model.

4.1. Profile simulation


Roughness measurements on processed surfaces show that rough sur-
faces are composed of both brownian and non-brownian spectra at various
length scales. Using parameters from Table 1 in eqn. (13), simulations of
such profiles are shown in Fig. 3. It is apparent that there is considerable
statistical resemblance between the real and fractal profiles. It should be
noted that although the fractal profile appears random, it is generated by a
deterministic function. This results from randomizing the phases of
the frequency modes by the choice of 7 = 1.5 and avoiding the point x = 0.
Since rough surfaces are random and are known to have a gaussian
probability distribution it is necessary to check statistically the simulation
of the W-M profiles for a dimension of 1.5. Figure 4 shows a comparison of
the probability density function of an artificially generated profile with that
of a gaussian distribution. The close agreement between the theory and
simulation validates the use of this procedure to simulate rough surfaces.

4.2. Surface simulation


When a profilometer is traversed on rough surfaces, the stylus typically
rides over the shoulders of the asperities and rarely goes over the summit or into
the deepest valley. Therefore spectra of surface profiles do not contain
complete information about the surface. For isotropic surfaces, Nayak [18]
324

-0.61 I I I
0 1 2 3 4
(a) Horizontal Distance, x x103 [m I

(b) Horizontal Distance, x x103 [m]


Fig. 3. Comparison of surface profiles and fractal simulations of (a) specimen 1 and (b)
specimen 5.

v
I t v
d
f v
9 v
0

7
v I / 1 /?_
-3 -2 -1 0 1 2 3

NONDIMENSIONAL HEIGHT, I I&


Fig. 4. Comparison of the probability density function of a fractal profile with a gaussian
distribution.
325

established the relation between the spectra of a surface and its profile along
an arbitrary direction. Since the spectrum of a profile is given by eqn. (5) it
can be shown that the surface spectrum S’(o) is

(5 - 2D)(7 - 20) AztD- ‘) E(D)


SS(w) = (15)
f 2 In y UC6-2D)

where E(D) is an integral

n/2
E(D) = j- {cosC5-W 8 - COS(‘-~~) 0) de (16)
0

For a brownian surface E(1.5) = 7r/16.


The spectrum in eqn. (15) suggests that the equation for surface simula-
tion should have the form

z(x,y) = F(D)A(D- 1) g ‘OS 2Tyn{x+ “;;‘;; 2sY”b + g(‘))


(17)
?l=ll,

l<D<2 Y>l
where F(D) is given by

2 In ~(5 - 2D)(7 - 2D)E(D) 1’2


F(D) = (18)
i 71

It should be noted that D is the dimension of a profile whereas the dimen-


sion of the surface is D, = D + 1. The function g(u) is introduced to ran-
domize the phases such that for each value of y the profile along x has a
different phase and vice versa. Since the W-M function produces signals of
random appearance it is used for phase randomization in the following form:

- cos 27ry”a
g(‘) = ’ c. ,+2-D),, (1%
“=?a1
Equation (19) produces signals of a gaussian distribution with standard
deviation a. If the random phases should vary between T and -71, or -l/2 <
g(a) < l/2, within a 2a limit of the gaussian distribution, then 20 = l/2. This
yields
1 +’ -D)(n I- 1)
c= 2 x 2112 {+4-2D) - 1)‘/2 (20)

Figure 5 shows a simulation of a brownian surface produced by eqn. (17).


Thus a surface of random appearance is produced by a purely deterministic
function.
For simulating a surface-related random phenomenon on such an srti-
ficial surface, the high frequency limit in eqn. (17) is determined by the
326

D = 2.50

Fig. 5. Fractal simulation of a surface of dimension 2.5: parameters used were nl = 0,


y = 1.5 and n = 0, 1, . 5.

length scale of that phenomenon. Examples include the wavelength of light


for scattering or the size of a gas molecule for adsorption.

5. Conclusions

Roughness measurements on a variety of stainless steel surfaces and a


textured magnetic thin-film rigid disk show that their power spectra follow
power laws within the length scales considered. This suggests that the rough-
ness structure can be characterized by fractal geometry.
This study introduces a new method of characterizing rough surfaces.
The non-stationarity and self-similarity of random rough surfaces suggests
that the conventional characterization parameters are not unique because the
variances of height, slope and curvature depend on the length of the sample
and the resolution of the measuring instrument. The new method uses the
Weierstrass-Mandelbrot fractal function to obtain scale-independent fractal
parameters to characterize rough surfaces. This provides the roughness struc-
ture at any length scale and therefore forms the basis for studying any sur-
face phenomenon at its relevant length scales.
The spectra of machined stainless steel surfaces follow a typical pattern.
At high frequencies surface processing does not affect the surface structure.
At such scales the spectra of all surfaces coincide thus determining a unique
fractal dimension of 1.5, corresponding to a brownian process. The influence
of surface processing is felt only for roughness features of frequencies smaller
than a certain comer frequency in the spectrum. In this low frequency
region, each processing technique produces a different type of spectral
behavior.
Using the experimentally determined fractal parameters, the artificial
profiles generated by the W-M function show a striking statistical resem-
327

blance to real profiles. The advantages of such a surface simulation are (a)
fractal profiles are not restricted by instrument-dependent filtering; (b) stan-
dardizing the characterizing parameters for different surface processing tech-
niques will replace the need for extensive experimentation by the creation
of artificial profiles; (c) they can be used to simulate surface-related random
phenomena and (d) the deterministic mathematical representation of a
random surface will facilitate the development of deterministic models for
analyzing such phenomena.

Acknowledgments

We thank Professor D. B. Bogy and E. R. Kral for providing the experi-


mental facilities.

References

1 P. R. Nayak, Random process model of rough surfaces, J. Lubr. Technol., 93 (1971)


398 - 407.
2 R. S. Sayles and T. R. Thomas, Surface topography as a non-stationary random
process, Nature (London), 271 (1978) 431 - 434.
3 B. B. Mandelbrot, The Fractal Geometry of Nature, W. H. Freeman, New York, 1982.
4 H. 0. Peitgen and D. Saupe, The Science of Fractal Images, Springer, New York, 1988.
5 T. R. Thomas, Defining the microtopography of surfaces in thermal contact, Wear,
79 (1982) 73 - 82.
6 J. I. McCool, Relating profile instrument measurements to the functional perfor-
mance of rough surfaces, J. Tribot., 109 (1987) 264 - 270.
7 A. N. Singh, The theory and construction of non-differentiable functions, in Squaring
the Circle and Other Monographs, Chelsea Publishing, 1953.
8 M. V. Berry and Z. V. Lewis, On the Weierstrass-Mandelbrot fractal function, Proc.
R. Sot. London, Ser. A, 370 (1980) 459 - 484.
9 D. L. Jordan, R. C. Hollins and E. Jakeman, Measurement and characterization of
multiscale surfaces, Wear, 109 (1986) 127 - 134.
10 J. A. Greenwood and J. B. P. Williamson, Contact of nominally flat surfaces, Proc. R.
Sot. London, Ser. A, 29.5 (1966) 300 - 319.
11 J. I. McCool, Comparison of models for the contact of rough surfaces, Wear, IO7
(1986) 37 - 60.
12 A. Majumdar, Fractal surfaces and their applications to surface phenomena, Ph.D.
Thesis, University of California, Berkeley, 1989.
13 B. B. Mandelbrot, Stochastic models for the earth’s relief, the shape and the fractal
dimension of the coastlines, and the number-area rule for islands, Proc. Nat. Acad.
Sci. U.S.A., 10 (1975) 3825 - 3828.
14 J. C. Wyant, C. L. Koliopoulos, B. Bhushan and D. Basila, Development of a three-
dimensional non-contact digital optical profiler, J. TriboZ., 108 (1986) 1 - 8.
15 S. R. Brown and C. H. Scholz, Closure of random elastic surfaces in contact, J. Geo-
phys. Res., 90, B7 (1985) 5531 - 5545.
16 J. J. Gagnepain and C. Roques-Carmes, Fractal approach to two-dimensional and
three-dimensional surface roughness, Wear, 109 (1986) 119 - 126.
17 R. F. Voss, Random fractal forgeries, in R. A. Earnshaw (ed.), Fundamental Algo-
rithms for Computer Graphics, 1985, pp. 805 - 835.
18 P. R. Nayak, Some aspects of surface roughness measurements, Wear, 26 (1973)
165 - 174.

You might also like