Tribology Letters 4 (1998) 1–35 1

Perspectives

Contact mechanics of rough surfaces in tribology: multiple asperity

contact
Bharat Bhushan
Computer Microtribology and Contamination Laboratory, Department of Mechanical Engineering, The Ohio State University, Columbus,
OH 43210-1107, USA

Received 16 June 1997; accepted 3 September 1997

Contact modeling of two rough surfaces under normal approach and with relative motion is carried out to predict real area of contact
and surface and subsurface stresses affecting friction and wear of an interface. When two macroscopically flat bodies with microroughness
come in contact, the contact occurs at multiple asperities of arbitrary shapes, and varying sizes and heights. Deformation at the asperity
contacts can be either elastic and/or elastic–plastic. If a thin liquid film is present at the interface, attractive meniscus forces may
affect friction and wear. Historically, statistical models have been used to predict contact parameters, and these generally require many
assumptions about asperity geometry and height distributions. With the advent of computer technology, numerical contact models of
3-D rough surfaces have been developed, particularly in the past decade, which can simulate digitized rough surfaces with no assumptions
concerning the roughness distribution. In this article, a comprehensive review of modeling of multiple-asperity contacts in dry and wet
conditions is presented. Contact models for homogeneous and layered, elastic and elastic–plastic solids with and without tangential
loading are presented. The models reviewed in this paper fall into two groups: (a) analytical solutions for surfaces with well-defined
height distributions and asperity geometry and (b) numerical solutions for real surfaces with asperities of arbitrary shape and varying
size and height distributions. Implications of these models in friction and wear studies are discussed.
Keywords: contact mechanics, tribology, asperity contact, surface deformation, contact area, rough surfaces

1. Introduction The figure shows that roughness is found at scales ranging

from millimeter to nanometer (atomic) scales. Although
Solid surfaces, irrespective of their method of formation, surface roughness is intrinsic, measured roughness is ex-
contain irregularities or deviations from a prescribed geo- trinsic. Instruments using different sampling intervals mea-
metrical form. The high points on the surfaces are referred sure features with different length scales. It can be con-
to as asperities, peaks, summits, or hills and the low points cluded that a surface is composed of a large number of
are referred to as valleys. When two nominally flat surfaces length scales of roughness that are superimposed on each
are placed in contact, surface roughness causes contact to other [4–6]. Therefore, “It is not that different asperities
occur at discrete contact spots (junctions). Deformation oc- come in different sizes but it is that one asperity comes in
curs in the region of the contact spots, establishing stresses different sizes”. Distribution of size and shape of asper-
that oppose the applied load. The sum of the areas of all ities is dependent on the measurement technique. When
the contact spots constitutes the real (true) area of con- the sampling interval at which the surface is examined is
tact, and for most materials with applied load, this will be reduced, the number of asperities detected and their cur-
only a small fraction of the apparent (nominal) area of con- vature appear to rise without a limit down to the atomic
tact (that which would occur if the surfaces were perfectly scale. This means that an asperity is not a “definite object”.
smooth). The real area of contact is a function of the surface Attempts have been made to identify a correct sampling in-
topography, material properties and interfacial loading con- terval which yields the relevant number of summits and the
ditions. The proximity of the asperities results in adhesive relevant curvature for a particular application (functional
contacts caused by interatomic interactions. When two sur- filtering) [7,8]. An asperity relevant for contact mechan-
faces move relative to each other, the adhesion of these as- ics is defined as that which makes a contact in a particular
perities and other sources of surface interactions contribute application (contacting asperity). As the two surfaces are
to friction force. Repeated surface interactions and surface brought in contact, nanoscale asperities are the first to come
and subsurface stresses, developed at the interface, result in into contact, instantly deforming plastically, and merging to
the formation of wear particles and eventual failure. Thus, form micro- to macrocontacts, which upon further applica-
the contact modeling of two rough surfaces is of consider- tion of load may deform by elastic or elastic–plastic defor-
able interest in the study of friction and wear [1–3]. mation [9]. Thus, size and number of contacting asperities
Roughness maps of a ceramic substrate measured using are dependent on the interface conditions. Both friction
three different measuring instruments with different sam- and wear are strongly influenced by the degree of plastic
pling intervals (lateral resolutions) are shown in figure 1. interaction of the contacting asperities.

 J.C. Baltzer AG, Science Publishers

2 B. Bhushan / Contact mechanics of rough surfaces

ure 2. Therefore, predictions of the contact models based

on conventional roughness parameters may not be unique
to a pair of rough surfaces. Majumdar and Bhushan [4],
Bhushan [6] and Ganti and Bhushan [5] have shown that the
roughness of engineering surfaces can be characterized by
fractal geometry. Majumdar and Bhushan [23] and Bhushan
and Majumdar [24] developed a new fractal theory of elas-
tic and plastic contact between two rough surfaces which
uses fractal parameters for surface characterization. (Also
see [6].)
Although the statistical models mentioned earlier so far
can predict important trends in the effect of surface proper-
ties on the real area of contact, their usefulness is very lim-
ited because of over-simplified assumptions about asperity
geometry and height distributions, the difficulty in determi-
nation of statistical roughness parameters, and the neglect-
ing of interactions between adjacent asperities. With the
Figure 1. Surface roughness profiles of a glass-ceramic substrate measured advent of computer technology, a measured surface profile
using an atomic force microscope (lateral resolution ∼ 15 nm), stylus can be digitized and used for computer simulation [25–28].
profiler (SP) P2 with a stylus tip of 0.2 µm radius (lateral resolution
Digital maps of pairs of different surfaces can be brought
∼ 0.2 µm), and Alpha-step 200 (α-200) with a stylus tip of 5 µm radius
(lateral resolution ∼ 2 µm), and noncontact optical profiler (NOP) (lateral together to simulate contact inside the computer and con-
resolution ∼ 1 µm) [29]. tours of contacts can be predicted. In computer simula-
tions, the resulting contour maps can be analyzed to give
Modeling of the contact of rough surfaces is difficult contact parameters for various interplanar separations of the
and has been treated in many papers using a number of ap- rough surfaces. The common approach has been to convert
proaches over the past three decades. The difficulty in the the complex stress-deformation analytical expressions to a
development of a theoretical model is that the surface is a system of linear algebraic equations, which are generally
random structure and may be anisotropic, so that stochas- solved, based on iterative numerical methods, with a com-
tic models must be used. The classical statistical model puter. Most of these analyses do not require assumptions
for a combination of elastic and plastic contacts between of surface isotropy, asperity shape, or distribution of asper-
rough surfaces is that of Greenwood and Williamson [10], ity heights, slopes and curvatures. However, one still has
which assumes that surfaces are composed of hemispheri- to select scan size and lateral resolution of the instrument
cally tipped asperities of uniform radius of curvature, with relevant for the interface problem on hand [29].
their heights following a gaussian distribution about a mean Several numerical techniques have been developed and
plane. Their model was modified by Whitehouse and Ar- used to solve contact problems of rough surfaces numeri-
chard [11] and Onions and Archard [12]. Nayak [13], Bush cally. Finite element methods [30] have been used to solve
et al. [14,15], Gibson [16] and McCool [17] developed plane-strain elastic contact problems with only a few cylin-
an elastic contact model that treated asperities as ellipti- drical asperities in contact [31]. For 3-D rough surface
cal paraboloids with randomly oriented elliptical contact contact problems with many asperities of arbritary shape,
areas. Slight modifications to these models have been pre- requirements of large number of mesh elements make the
sented in several papers (e.g., see [7,18–21]). Kotwal and finite element approach unfeasible. One of the techniques
Bhushan [22] developed a statistical model for nongaussian provides a deterministic solution to stresses and areas for
surfaces. A nongaussian surface may exhibit skewness the approach of two real rough surfaces [32–38]. This tech-
and kurtosis. Skewness represents asymmetric spread and nique takes full account of the interaction of deformation
kurtosis represents peakedness of the distribution [3]. Al- from all contact points and predicts the contact geometry
though these statistical models result in simple relation- of real surfaces under loading. It provides useful infor-
ships, the assumptions are unrealistic. mation on the contact pressure, number of contacts, their
A surface with gaussian distribution of heights and ex- sizes and distributions, and the spacing between contacts.
ponential autocorrelation function can be characterized by For contacts with a moderate number of contact points,
two parameters: the standard deviation of surface heights both elastic and elastic–plastic analyses of two rough sur-
(σ or Rq or rms) and correlation length (β ∗ ) [11]. Other faces can be carried out [35,38,39]. Sayles and co-workers
parameters that are commonly measured are peak-to-mean (e.g., [35]) have used an iterative process with a conven-
distance (Rp ) and peak-to-valley (P–V) distance. Due to tional matrix inversion technique to solve the contact prob-
the multiscale nature of surfaces, it is found that the surface lem for contact pressure and contact area. However, with
roughness parameters depend strongly on the resolution of an increase in contact points of rough surfaces, the so-called
the measuring instrument or any other form of filter, and influence matrix will become very large, and possible ill-
hence are not unique for a surface; for an example see fig- posed. The large, ill-posed influence matrix will not only
 B. Bhushan / Contact mechanics of rough surfaces 3

Figure 2. Scale dependence of standard deviation of surface heights (σ), peak-to-mean (Rp ), peak-to-valley (P–V) distance and correlation length (β ∗ )
for a glass-ceramic substrate, measured using AFM, stylus profiler (SP-P2) and noncontact optical profiler (NOP) [29].

require a great deal of time for the conventional matrix the 3-D rough surface contact problem with a larger number
inversion method to complete a large number of iteration of contact points.
cycles, but also, in some cases, cause serious nonconverg- Subsurface stress fields and the effect of friction on the
ing problems. contact stresses and contact area has been analyzed by Bai-
Instead of using the conventional matrix inversion ap- ley and Sayles [42] and Yu and Bhushan [43]. Merriman
proach, Tian and Bhushan [38] utilized the approach on the and Kannel [44] and Cole and Sayles [45] developed a
basis of the variational principle. The use of this principle 2-D model to analyze the contact of layered rough bodies.
leads to a standard quadratic mathematical programming A 3-D model of layered rough bodies has not been reported
problem after an infinite-to-finite dimension transformation. to date. Oden and Martins [46] and Martins et al. [47] have
In the variational method, the real area of contact and con- presented a continuum model of sliding of two rough sur-
tact pressure distributions are those that minimize the total faces using the finite element method.
complementary potential energy. The use of the variational In the presence of a thin liquid film with a small con-
principle for elastic contact problems was first introduced in tact angle (wetting characteristics), such as a lubricant or
the early 1970s [40,41], but the numerical difficulty of using an adsorbed water layer at the contact interface, curved
a standard simplex-type algorithm to search for the mini- menisci form around contacting and noncontacting asperi-
mum of the potential energy of contact with a large number ties. The attractive meniscus force arises from the negative
of contact points has kept the variational approach from Laplace pressure inside the curved meniscus as a result of
general applications in engineering practice. The varia- surface tension [48–50]. This instrinsic attractive force may
tional approach developed by Tian and Bhushan [38] uses a result in high static friction (stiction), kinetic friction and
direct quadratic mathematical programming method which wear. This problem is particularly important in the com-
guarantees the uniqueness of the solution of rough surface puter data storage industry [3,51]. Several models have
contact problems. It is also able to significantly reduce been developed to predict meniscus forces developed at a
computing time since it does not need an additional itera- wet interface using statistical approach [52–55]. Kotwal
tion process which has been required by the conventional and Bhushan [55] have extended statistical models to study
matrix inversion technique, and makes it feasible to solve nongaussian surfaces with and without a liquid film present
4 B. Bhushan / Contact mechanics of rough surfaces

at the interface. More recently Tian and Bhushan [56] de-

veloped a numerical model that uses digitized real rough
surface maps without any arbitary assumption of the shape
and size of asperities. The model accounts the effect of sur-
face roughness, thin liquid film, environment and operating
conditions. Bhushan and Chilamakuri [57] and Chilamakuri
and Bhushan [58] used this numerical model to study inter-
action of nongaussian surfaces with skewness and kurtosis.
It is evident that substantial on-going contributions, par-
ticularly in the last decade, are being made in the modeling
of multiple-asperity contacts under both dry and wet condi-
tions. In this perspective article, we present various contact
models for two rough surfaces in static and sliding contact.
Homogeneous and layered elastic and elastic–plastic solids,
with and without a liquid film present at the interface, will
be considered. Emphasis in this paper has been on the re- Figure 4. Normal contact pressure profile beneath the rigid spherical
sults and implications of the contact models. The paper is indenter along the x axis for a layered elastic half-space with different
values of E1 /E2 and µ = 0 and when the radius of contact (at E1 /E2 )
primarily intended for a newcomer or individual not very a0 = h [61].
familiar with this field. The paper starts out with a short
review of modeling of single asperity contact followed by a 2.1. Elastic contact
comprehensive review of the modeling of multiple asperity
contacts under both dry and wet conditions. We first consider a generalized case of a spherical body
in a sliding contact with a layered elastic half-space with
a layer of thickness h, figure 3 [60,61]. In the following
results, Poisson’s ratio was taken as 0.3 for both layer and
2. Single asperity contact
substrate. Poisson’s ratio has little effect on stresses. Plas-
tic deformation will be incorporated later. Various stress
If the asperity tip is assumed to be spherical then the profiles on and beneath the contacting interface for homo-
contact problem reduces to a sphere in contact with a flat geneous and layered elastic solids in static and sliding con-
surface. A comprehensive review of modeling of single as- tacts are presented in figures 4 to 6. Figure 4 presents
perity contact has been presented by Bhushan [59]. Here, the pressure profile beneath the indenter at various E1 /E2 .
we summarize important results of contact stresses and con- The pressure is normalized by p0 , which is the maximum
tact areas for homogeneous and layered elastic and elastic– pressure under the center of indenter for a homogeneous
plastic solids in both static and sliding contacts. These medium (E1 /E2 ) when the radius of contact a0 = h. Note
results are of interest to better understand the modeling of that the so-called hertzian pressure distribution is parabolic
multiple asperity contacts. with a maximum at the center of contact. The radius of the
circular contact zone decreases and the maximum pressure
increases with an increase in the value of E1 /E2 . Fig-
ure 5 presents the three nonzero stress components σz , σx
and τxz , as a function of depth in the layer and substrate
under the center of the indenter (x = y = 0) for a coeffi-
cient of friction µ = 0.25 and for various values of E1 /E2 .
Note that σx drops off rapidly through the body (along the
z-axis) (at z/a0 ∼ 1.3 for homogeneous solids) whereas
the σz drops off slowly (at z/a0 > 3 for homogeneous
and nonhomogeneous solids). Maximum values of these
stresses increase with an increase in friction and E1 /E2 .
The stress component σx is tensile at the interface for a
stiffer layer, which is significant for cracks at the base
of the layer and orthogonal to the interface. The shear
stress component is also enhanced by the stiffer layer, al-
though, it decays rapidly into the bulk. High interfacial
shear stress adversely affects the adhesion of the layer to
the substrate. For layers that are more compliant than the
substrate, both the maximum value of normal stresses and
Figure 3. Schematic of spherical rigid body in a sliding contact with a the interfacial shear stress are reduced. Figure 6 shows the
layered elastic half-space. normal stress profile on the contacting surface along the
 B. Bhushan / Contact mechanics of rough surfaces 5

Figure 6. Variation of normal stress on the surface beneath the rigid

spherical indenter along the x-axis for a indenter acting on a layered
elastic half-space with different values of E1 /E2 and µ and when the
radius of contact (at E1 = E2 ) a0 = h [61].
Figure 5. Normal and shear stresses under the center of indenter along the
z-axis for a rigid spherical indenter acting on a layered elastic half-space
with different values of E1 /E2 when µ = 0.25 and when the radius of leading edge of the contact area (x = a) and maximally
contact (at E1 /E2 ) a0 = h [61]. tensile at the trailing edge (x = −a). A stiffer layer and
friction can thus degrade the brittle failure characteristics of
x-axis. Note that for a homogeneous solid, the normal a layered medium, whereas a more compliant layer can be
stress (in all radial directions) is tensile outside the loaded beneficial. Based on the normal and shear stresses at the in-
circle. It reaches its maximum value at the edge of the terface and maximum tensile normal stresses at the surface,
circular contact. This is the maximum tensile stress occur- layers that are more compliant (e.g., solid lubricants – Ag,
ring anywhere and it is held responsible for the ring cracks MoS2 , graphite) than the substrate for multilayered body
which are observed to form when brittle materials such as are preferred. However, high wear resistance may require
glass are pressed into contact [62]. Normal stresses are harder (which are generally stiffer) layers. Low values of
compressive inside the contact. A stiffer layer and fric- coefficient of friction are also preferable.
tion both increase the maximum tensile and compressive The results presented so far are for the case of a0 = h
stresses on the surface. As the coefficient of friction in- with various values of E1 /E2 . For values of a0  h, the
creases, σx becomes unsymmetrical, compressive at the maximum stresses in the layer are similar to the homoge-
6 B. Bhushan / Contact mechanics of rough surfaces

1/2
Figure 7. Contours of constant normalized von Mises stresses J2 /p0 , for a rigid spherical indenter acting on a layered elastic half-space at
E1 = E2 , E1 = 2E2 , and E1 = 0.5E2 when the radius of contact (at E1 /E2 ) a0 = h [61].

neous case with modulus E1 . Similarly, for a0  h, the the difference in the predictions of the two criteria is not
stress field is dominated by the substrate. In the transition large.
zone (∼ 0.5h < a0 < 6h), the stress field depends strongly According to von Mises yield criterion, the contact pres-
on the value of E1 /E2 . sure, p0 , at which yielding in the layer (in the case of
layered solid) and substrate occurs, is found by equating
2.2. Limits of elastic deformation the square root of the invariant of the stress deviator ten-
1/2
sor (J2 ) to the yield stress in simple shear k, von Mises
As the normal load between the two contacting bod- stress. For static contact, maximum Hertz contact pres-
1/2
ies is applied, they initially deform elastically according to sure to initiate yield is 2.8k. Contour plots of J2 /p0 are
their Young’s moduli of elasticity. As the load is increased, shown in figure 7 for µ = 0, 0.25 and 0.50 for various
one of the two bodies starts to deform plastically. As the values of E1 /E2 . For µ = 0, the figure shows symmetrical
normal load is further increased, the plastic zone grows un- contours, and the maximum shear stress occurs beneath the
til the whole of the material surrounding the contact has surface on the axis of symmetry, the z-axis (figure 3). As
undergone plastic deformation. Metals, alloys and some µ is increased, the region of maximum von Mises stress
nonmetals and brittle materials deform predominantly by moves from a subsurface location towards the surface and
“plastic shear” or “slip” in which one plane of atoms slides becomes more intense; yield occurs at the surface when
over the next adjacent plane. The load at which the plastic µ exceeds about 0.3. With respect to the center of contact,
flow or yield begins is related to the yield point of the softer the maximum von Mises shear stress location moves in the
material in a simple tension or pure shear test through an direction of friction force acting on the body (or in the op-
appropriate yield criterion. Two of the yield criteria most posite direction of sliding velocity of the body). Contour
commonly employed for most ductile materials as well as plots of E1 /E2 = 2 show higher von Mises stresses with
sometimes for brittle materials are Tresca’s maximum shear significant discontinuities occurring at the interface. The
stress criterion and von Mises shear strain energy criterion. presence of a layer increases the von Mises stress in the
The von Mises criterion usually fits the experimental data layer. For a more compliant layer case of E1 /E2 = 0.5,
of metallic specimens better than other theories. However, von Mises stresses are lower than in the case of a nonlay-
 B. Bhushan / Contact mechanics of rough surfaces 7

with E/Y = 685. Figure 10 √ shows the evolution of nor-

malized von Mises stress ( 3J2 /Y ) contours in the half-
space. Y is the initial yield stress in simple tension. As
the load increases, the elastic core developed over the sub-
surface plastic zone diminishes and eventually disappears
while the maximum shear stress moves radially outward
from the center and toward the surface. At loads on the
order of 300 times the load required to initiate yielding,
the material goes through full plasticity. The depth of the
plastic zone is about two times the contact radius at the
load. The authors reported that the mean contact pressure
at full plasticity is about 2.5 times the yield stress with the
contact diameter being about 10 times that of the contact
diameter at the time yielding is initiated.
Kral et al. [64,65] carried out finite element analysis of
a rigid sphere against an elastic–plastic layered medium.
They reported that a stiffer and harder layer increases the
load for inception of yielding for both layer and the sub-
strate. However, as reported earlier, a stiffer layer results in
Figure 8. Effect of coefficient of friction on the maximum Hertz contact large tensile stresses on the layer surface and shear stresses
pressure for yield (von Mises). at the layer–substrate interface, which may result in the
formation of cracks and debonding of the interface.

3. Multiple-asperity dry contacts

3.1. Analysis of identical asperities

We first consider the contact between a smooth plane and

a nominally flat surface covered with a number of spheri-
cal asperities of similar radius and the same height z, rela-
tive to the reference plane, figure 11. As the surfaces are
loaded together, the total displacement (normal approach)
δ is equal to (z − d), where d is the current separation of
the smooth surface and the reference plane of the rough
Figure 9. For the case of slider in a sliding contact with a layered solid, surface. Each asperity is deformed equally, and carries the
schematic of plastic zone and maximum tensile and maximum shear stress same normal load, Wi , so that for N asperities the total
location in the lower body. Note that stress distributions for the upper body load W will be N Wi . For each asperity, the load, Wi , and
will be that of the lower body with x and z replaced with −x and −z, the area of contact, Ai , are known from the Hertz analysis.
respectively.
Thus, if R is the radius of all identical asperities,
ered homogeneous body and, further, only a mild disconti- 4E ∗ 1/2 3/2
nuity occurs at the interface. The maximum Hertz pressure Wi = R δ , (1a)
3
at which yield will occur, is presented as a function of co-
efficient of friction for a homogeneous solid in figure 8. and
Based on figures 6 and 7, a schematic of the plastic
Ai = πRδ, (1b)
zone and maximum tensile stress location are schematically
shown in figure 9. Note that with brittle materials, the where E ∗ is the composite or effective modulus. The total
appearance of tensile stresses is more important than the load is
value of the maximum shear responsible for yield.
4 N E ∗ Ai
3/2
W = . (2)
2.3. Elastic–plastic contact 3π 3/2 R
Load is related to the total real area of contact, A (= N Ai )
Kral et al. [63] carried out a detailed finite element from equations (1) and (2) as
analysis of a rigid sphere in contact with an elastic–plastic
strain-hardening half-space. Loads and distances were nor- 4E ∗ A3/2
malized by the values at the onset of plastic deformation W = , (3)
3π 1/2 N 1/2 R
8 B. Bhushan / Contact mechanics of rough surfaces

Figure 10. Contours of normalized von Mises stress in the region 0 6 r/ay 6 12, 0 6 z/ay 6 12 for loads W/Wy equal to (a) 15.5, (b) 100.6,
(c) 200, and (d) 300 for a rigid sphere acting on an elastic–plastic (strain hardening exponent n = 0.3) half-space. Contour number 4 represents the
boundary of plastic zone [63].

Figure 11. Schematic of contact of a regular patterned rough surface against a smooth plane surface.

where The contact area is circular, having a radius a, and the

contact pressure is elliptical with a maximum pressure at
1 (1 − ν12 ) (1 −ν22 )
∗
= + , the center of the contact. The mean (pm ) and maximum
E E1 E2 contact pressures (p0 ) are
E1 and E2 and ν1 and ν2 are the Young’s moduli of elastic-
 1/3
ity and Poisson’s ratios, respectively, of the mating bodies 1 2 Wi 4E ∗ δ 1/2 16W E ∗2
pm = p0 = = = . (4)
and 2. Equation (1b) indicates that the surface just outside 3 Ai 3πR1/2 9π 3 N R2
the footprint is displaced in such a way that the real area
of contact is exactly half of the area 2πRδ which would be The propensity for yielding is governed by the Tresca’s
obtained by plastic flattening of the spheres. Equation (3) maximum shear stress criterion or von Mises shear strain
shows that, for this particular model, the real area of con- energy criterion. Yielding is initiated if the maximum shear
tact is proportional to the two-thirds power of the applied stress (which occurs at the subsurface in the Hertz problem
normal load, when the deformation is elastic. considered here) exceeds the shear shear strength. For a
 B. Bhushan / Contact mechanics of rough surfaces 9

typical case using both yield criteria, this condition is sat- a constant radius on each surface, Rp1 and Rp2 , and the
isfied, composite radius (Rp ) could be assigned to the rough sur-
face; (3) that their heights vary randomly; and (4) that most
pm ∼ 1.07Y , (5a)
engineering surfaces have a gaussian distribution of peak
where Y is the yield strength in simple tension of the softer heights. From Williamson et al. [69], only surfaces pre-
material [59]. As the contact pressure is increased further, pared by cumulative processes (such as peening and spark
the plastic zone grows until the whole of the material sur- erosion, where the final shape of each small region is the
rounding the asperity undergoes plastic deformation. The cumulative result of a large number of discrete local events)
condition for full plasticity for an elastic–perfectly plastic have a guassian height distribution. Single-point processes
material occurs, (such as turning or shaping) and extreme-value processes
(such as grinding and milling) lead to only an approximate
pm ∼ 2.8Y , Tabor [66] (5b) gaussian distribution. The assumption of peak radii being
∼ 2.5Y , Kral et al. [63]. constant is clearly not valid.
Greenwood and Tripp [10] have treated the contact of
Based on a number of numerical analyses and experimental
two rough surfaces instead of the one rough surface against
measurements of indentation of half-space of elastic–plastic
a flat surface treated by G&W. For the case of two rough
materials, hardness (H) is related to Y as [67]
surfaces in contact with the pairs of asperities not aligned
H ∼ 2.8Y. (6) and the usual contact between the shoulders of the two hills,
they found that, for a gaussian peak-height distribution, the
Therefore, yielding is initiated when specification of asperity shape and the locations of asper-
pm ∼ H/3. (5c) ities on one or both surfaces are unimportant. Therefore,
although they have assumed the asperity tips to be spherical
for numerical simplicity, this will not affect real area cal-
From equations (4) and (5c), the critical load beyond culations. Also, they showed that the contact of two rough
which plastic deformation occurs is given by surfaces could be reduced to an equivalent, single, rough
Wcrit π 3 R2
surface with a plane. O’Callaghan and Cameron [71] and
∼ H3 . (7) Francis [18] also considered a case in which both surfaces
N 48E ∗2
can be rough and asperities need not contact at their tops.
It is general practice to introduce a factor of safety to ac- They concluded that the contact of two rough surfaces is
count for the fluctuation of the hardness measurement and negligibly different from the contact of a smooth and an
all the uncertainties involved in design. The factor includes equivalent rough surface.
any dynamic effect during asperity contacts. The value of The equivalent rough surface is defined as one whose
a factor of safety is normally chosen between 2 and 3 [68]. asperity-peak curvature, 1/Rp , is the sum of the curvatures
If the load exceeds the critical load, the softer material of two rough surfaces,
of the contacting bodies deforms plastically. If the material
deforms plastically at the interface, at full plasticity each 1/Rp = 1/Rp1 + 1/Rp2 , (9)
asperity contact can be thought of going through an inden-
tation process. For elastic–perfectly plastic materials (with and by elementary statistics, if the peak-height distributions
no work hardening), the flow pressure under full plasticity of two rough surfaces follow independent random distrib-
is found to be almost independent of the load. In this case, ution (not necessarily gaussian) with standard deviations
    of σp1 and σp2 , the distribution of the equivalent rough sur-
Wi W face has a standard deviation σp (e.g., [72]),
= = H, (8)
Ai p A p

2 2 1/2
σp = σp1 + σp2 . (10)
thus the real area of contact is proportional to the load.
Equations (9) and (10) are valid when the two surfaces are
3.2. Statistical analysis of contacts independent, as is likely when two surfaces are prepared
separately. However, when the surfaces have slid together,
If the two rough surfaces, both nominally flat, come in this assumption may be violated. If so, the expression for
contact until their reference planes (taken to pass through σp must be modified by a covariance term.
the mean of the peak height distribution) are separated by
a distance h, then there will be contact at those asperities 3.2.1. Elastic contact
whose total heights, (z1 + z2 ), are greater than h. The For elastic contacts under static conditions or under dy-
contacts can be either elastic or plastic. Greenwood and namic conditions with no frictional stresses present at the
Williamson [10] (G&W) analyzed a rough surface against contact, using G&W’s assumptions, we can calculate the
a smooth surface. They assumed that (1) the rough surface apparent pressure, pa ; mean real pressure, pr ; (elastic) real
is covered with a large number of asperities, which, at least area of contact, Are ; number of contact spots, n; and mean
near their summit, are spherical; (2) asperity summits have asperity real area of contact as a function of separation, d.
10 B. Bhushan / Contact mechanics of rough surfaces

Figure 12. Schematic representation of the contact between a rough surface and a smooth surface.

Based on the G&W analysis, we consider the contact where D, the dimensionless separation, is d/σp ; η is the
between a plane and a nominally flat surface covered with density of asperity summits per unit area (N/Aa ) on a sur-
a large number of spherically tipped asperities of the same face with smaller density; and Fm (D) is a parabolic cylin-
radius Rp and with their peak heights described by a prob- der function given by
ability density function of p(z), figure 12. Z∞
If the two surfaces come together until their reference
Fm (D) = (s − D)m p∗ (s) ds, (17)
planes are separated by a distance d, then there will be
D
contact at any asperity whose height was originally greater
than d. Thus, the probability of making contact at any where p∗ (s) is the standardized peak-height-probability
given asperity of height z, is density function, in which the height distribution has been
Z∞ scaled to make its standard deviation unity. For the case of
P (z > d) = p(z) dz, (11a) peak height distribution following a gaussian height distri-
bution,
d
  Z∞
1

and if there are N asperities in all, the expected number Fm (D) = (s − D)m exp − s2 /2 ds
of contacts will be (2π)1/2
Z∞  
D

n = N p(z) dz. (11b) m! 
  1 

= exp −D 2
/4 U m + ,D
d
(2π)1/2 2
for m > 0. (18)
Also, since δ = z − d, the total real area of contact is
Z∞ The values of U are listed in Abramowitz and Stegun [73].
Are = πN Rp (z − d)p(z) dz. (11c) A short table of functions Fm (D) is also given by Green-
wood and Tripp [70].
d
Note that equations (12) to (16) hold for all surface dis-
Similarly, we find the expected total load is tributions. However, the assumption of two rough surfaces
Z∞ being the same as one equivalent rough surface with a plane
4
W = pr Are = pa Aa = NE ∗
Rp1/2 (z − d)3/2 p(z) dz. is valid only for surfaces having a gaussian distribution.
3 A simple relationship exists for an exponential height
d
(11d) distribution,
It is convenient to introduce nondimensional variables. The p∗ (s) = exp(−s). (19)
relationships in form of dimensionless variables are pre-
For this case, functions Fm (D) are just m! exp(−D), and
sented here, we have,
4  4 
pa
1/2
(ηRp σp )E ∗ (σp /Rp )1/2
=
3
F3/2 (D), (12) pr = (1.5!)E ∗ σp /Rp , (20a)
3π
pr  4   3π  1
= F3/2 (D)/F1 (D), (13) pa Aa
E ∗ (σp /Rp )1/2 3π Are = ∗
, (20b)
 3π  4 (1.5!) E (σp /Rp )1/2

1/2
Are E ∗ σp /Rp /pa Aa = F1 (D)/F3/2 (D), (14) 3 1 pa Aa
4 4 n= , (20c)

1/2 4 (1.5!) Rp σp E ∗ (σp /Rp )1/2
nRp σp E ∗ σp /Rp /pa Aa = F0 (D)/ F3/2 (D), (15)

3 Are
Are /n Rp σp = πF1 (D)/F0 (D), (16) = πRp σp . (20d)
n
 B. Bhushan / Contact mechanics of rough surfaces 11

Table 1
For this surface, Onions and Archard [12] expressed the
Interplanar separation, mean real pressure, real area of contact, number
of contact spots, and mean asperity real area of contact for elastic con- peak heights, curvatures and asperity density in terms of σ
tacts [19]. and β ∗ . In their contact model, they did not assume that
peak heights follow a gaussian distribution but follow a
interplanar D = 1.40[log(0.57/Pa )]0.65
separation distribution derived from the assumed gaussian distribution
pr of surface heights; second, peak radii are not constant and
mean real = 0.42Pa0.04 ∼ 0.32 have a distribution which is dependent upon the height.
E ∗ (σp /Rp )1/2
pressure
They reported that the distribution of peak heights is not
Are quite gaussian and that the peak curvature of the higher
real area = 2.40Pa0.96 ∼ 3.20Pa
Aa [ηRp σp ] peaks tended to have higher values than those at lower
of contact
n
levels. Based on their contact model,
number of = 1.21Pa0.88 ∼ 2.64Pa = 0.5 (D → 0)
ηAa pa Aa
contact spots Are ∝ . (23)
E (σ/β ∗ )
∗
Are /n
mean asperity real = 2.00Pa0.08 ∼ 1.21
area of contact
Rp σp By comparison of contact analyses by Whitehouse and Ar-
chard and Greenwood and Williamson, it is observed that
Pa = pa /(ηRp σp )E ∗ (σp /Rp )1/2 6 0.57. σp is proportional to σ, Rp is proportional to β ∗2 /σp and
η is proportional to 1/β ∗2 . Therefore, the three parameters
We note that the real area of contact and the number used in G&W analysis should be related by the equation
of contacts are both proportional to load, even though the for a given type of finishing operation,
asperities are deforming elastically. The real contact pres-
sure and mean asperity real area of contact are independent σp Rp η = constant. (24)
of load. For other distributions, such a simple relationship
Nayak [74] analyzed the surface roughness following the
will not apply.
pioneering work in statistical geometry of the oceanogra-
For a gaussian distribution, D vs. pa is obtained from
pher Longuet-Higgins [75,76]. He has shown that a random
equation (12). Then, with the help of this relationship
and isotropic surface with a gaussian height distribution can
and equations (13) to (16), the relationships between pa
be completely characterized by the three spectral moments
and pr , Are , n, and Are /n are obtained in a dimension-
of the profiles: M0 , M2 and M4 known as the zeroth, sec-
less form. Next, the data are fitted to a power form us-
ond and fourth spectral moments. M0 , M2 and M4 are
ing a least-squares fit. Approximate direct relationships
equivalent to the variance of the distributions of profile
of D, pr , Are , n and Are /n with pa are presented in ta-
heights (σ2 ), slopes (σ0 ) and curvature (σ00 ), respectively.
2 2
ble 1 [19]. From equations in table 1, note that D is a very
He obtained expressions of the distribution of peak height
weak function of pa ; pr and Are /n are practically indepen-
and peak curvature in terms of a parameter α, defined as
dent of pa ; and Are and n are approximately proportional
to pa . An important relationship for the real area of contact M0 M4
in the elastic regime is listed here, α= , (25)
M22
3.2pa Aa which defines the width of the power spectrum density of
Are ∼ . (21)
E ∗ (σp /Rp )1/2 the surface roughness, i.e., range of wave lengths encoun-
This model is defined by three parameters: σp , Rp and η. tered in it. For an isotropic, gaussian surface, α > 1.5.
Whitehouse and Archard [11] regarded the profile of a ran- If the surface is gaussian, the surface slope in an arbitary
dom surface as a random signal represented by a height direction is essentially a difference between two correlated
distribution and an autocorrelation function. They showed gaussian variables and is therefore also gaussian distrib-
that all features of a surface with gaussian distribution of uted. Likewise the second derivative of the surface in any
heights and an exponential autocorrelation function could general direction is gaussian distributed.
be represented by two parameters: σ and β ∗ . For an expo- Given the three spectral moment values M0 , M2 and
nential autocorrelation function, M4 for each of the two rough surfaces, the corresponding
  values for the equivalent rough surface are computed as
τ respective sums, i.e.,
C(τ ) = exp − ∗ , (22)
β
(M0 )e = (M0 )1 + (M0 )2 , (26a)
where 1/β ∗ is the decay rate of function. The correlation
length can be defined as C(τ ) = 0.1 (when τ = 2.3β ∗ ) or (M2 )e = (M2 )1 + (M2 )2 , (26b)
C(τ ) = 1/e (when τ = β ∗ ). Both forms are considered to (M4 )e = (M4 )1 + (M4 )2 . (26c)
be a measure of the length at which simple readings become
statistically independent of one another. Whitehouse and Nayak [13] and Bush et al. [14] carried out the con-
Archard took β ∗ as the correlation length given by C(τ ) = tact analysis by modeling the rough surfaces as isotropic,
1/e. gaussian surfaces in terms of spectral moments.
12 B. Bhushan / Contact mechanics of rough surfaces

Bush et al. [15], Gibson [16] and McCool [17] used Based on the G&W approach, Onions and Archard [12]
Nayak microgeometry assumptions to develop an elastic also defined a plasticity index based on σ and β ∗ as,
contact model for anisotropic surfaces. The asperities were E∗  σ 
represented as elliptical paraboloids with random principal ψ= . (31)
H β∗
axis orientation and aspect ratio of the grains. They de-
veloped expressions for contact area in the terms of five They found that if ψ > 0.45, plastic flow occurs even at
surface parameters – M0 , and M2 and M4 along the grains trivial loads, and if ψ < 0.25 plastic flow is most unlikely.
and across the grains. Kotwal and Bhushan [22] carried One of the benefits of Onions and Archard’s analysis is
out contact analysis of nongaussian surfaces with skewness that plasticity index and contact parameters are expressed
and kurtosis. in terms of surface parameters, which are most easily mea-
Using Greenwood and Williamson’s contact analysis, sured. Based on Whitehouse and Archard [11] and White-
Fuller and Tabor [77] developed an analysis for adhesion house and Phillips [79], σ and Rp can be expressed in terms
of two rough elastic surfaces in contact. They related work of σ and β ∗ . All roughness parameters and their relation-
of adhesion to a reduction in the free energy of the system ships are a function of sampling interval [29,80].
resulting from reduction in surface energy as proposed by We note that in a sliding contact with friction present at
Johnson et al. [78]. They found that the relative adhesion the interface, the maximum shear stress would be larger and
force between the two contacting solids depends on sev- would occur nearer the surface (section 2). Therefore, the
eral parameters such as Young’s moduli, surface roughness contacts would become plastic at lower values of ψ. In ad-
parameters (σp and Rp ) and work of adhesion. dition, in a multilayered solid for a fixed value of substrate
Young’s modulus, the stresses increase for larger values in
3.2.2. Limit of elastic deformation the Young’s modulus of the overcoat (section 2), implying
For a random surface with asperities with gaussian that the contact becomes plastic at lower values of ψ and
height distribution and constant radii, the normal approach vice versa. If the contact radius is much greater than the
in terms of pm from equation (4) is overcoat thickness, the effect of overcoat Young’s modulus
would be negligible and the yielding would be dominated
9π 2 2 Rp by the modulus and hardness of the substrate [81]. Further,
δ= p . (27) during sliding, polishing of the asperity generally occurs,
16 m E ∗2
which results into a smoother surface and an increase in
From equations (5), (6) and (27), the critical value of δ for the real area of contact [3]. During sliding, instantaneous
the asperity necessary to initiate subsurface plastic flow is roughness should be used.
 2 For calculations of the real area of contact, E, Y , and
H H should be measured at a strain rate corresponding to
δp ∼ Rp . (28)
E∗ the loading and unloading of the asperity contacts. During
sliding, the asperities are loaded and unloaded periodically
Greenwood and Williamson [10] defined a plasticity in-
in a time corresponding to that taken for a moving asperity
dex ψ as the square root of the inverse of δp normalized
to traverse its contact diameter. Therefore, the strain rate
with σp as
involved in the loading cycle (which determines the area
 1/2   1/2 of contact) can be estimated as the sliding velocity divided
σp E∗ σp by the diameter of an asperity contact.
ψ= = . (29)
δp H Rp Finally, in the case of materials that creep to a marked
extent such as polymers, the real area of contact will in-
This index is indicative of the degree of plasticity. For
crease with time of application of the load [3].
plastic contact, the real area of contact is given as

pa Aa 3.2.3. Optimization of mechanical properties and surface

Arp = . (30) roughness parameters
H
To minimize friction and wear in a machine for given
Assuming Arp /Are = 0.02 as the criterion for the onset of operating conditions, the fraction of real area of contact to
a significant degree of plasticity, it was found that if the apparent area of contact and the real contact pressure should
ψ < 0.6, the deformation is largely elastic and if ψ > 1, be as low as possible. The real area of contact versus the
surface deformation is largely plastic. Note that probability plasticity indices given from equations (21) or (23) and (30)
of plastic flow is virtually independent of the load and solely are plotted in figure 13. An examination of figure 13 shows
a function of the plasticity index as long as the asperities that the plastic contact results in a minimum contact area.
continue to deform independently. The index depends on However, repeated plastic contact would lead to an unde-
both the mechanical properties and the surface roughness of sirable permanent deformation and smoothening resulting
the contacting surfaces. Slight modifications to the elastic– in elastic contacts (and higher real area of contact). Wear
plastic models have been presented in several papers (e.g., is more probable when asperities touch plastically than in
see [7,18–21]). pure elastic contacts. Therefore, it is desirable to design
 B. Bhushan / Contact mechanics of rough surfaces 13

Figure 14. Qualitative description of statistical self-affinity for a surface

profile [4].

A unique property of rough surfaces is that if a sur-

face is repeatedly magnified, increasing details of rough-
ness are observed right down to the nanoscale. In addition,
the roughness at all magnifications appears quite similar in
structure as qualitatively shown in figure 14. Such behav-
ior can be characterized by fractal geometry [4–6]. The
Figure 13. Influence of plasticity index on the real area of contact [19].
main conclusions from these studies are that a fractal char-
acterization of surface roughness is scale-independent and
machine components in the elastic contact regime and ψ provides information of the roughness structure at all the
close to the elastic contact limit (ψ ∼ 0.6) or E ∗ (σp /Rp )1/2 length scales that exhibit the fractal behavior. Based on this
to be as high as possible. Intuitively, we can explain the observation, Majumdar and Bhushan [23] and Bhushan and
fact that E ∗ (σp /Rp )1/2 has to be higher for lower real area Majumdar [24] developed a new fractal theory of contact
of contact by the following observation. The asperities with between two rough surfaces.
high E and low Rp produce high contact stresses and result Structure function and power spectrum of a self-affine
in lower Are for a given load. In addition, high σp allows fractal surface follow a power law and can be written as
contact with fewer asperities and again produces high con- (Ganti and Bhushan model) [5]
tact stresses and results in lower Are for a given load.
In principle, one can produce a well-defined (rather than S(τ ) = Cη (2D−3)τ (4−2D) , (32)
statistical) roughness with few tall asperities of the same
height and a small radius of curvature on a very smooth
surface to minimize contact. Selection of number of as- c1 η (2D−3)
P (ω) = , (33a)
perities and radii of curvature of summits should be made ω (5−2D)
such that the mean real pressure is just below the yield and
strength of the softer material [68,82]. An advantage of Γ(5 − 2D) sin[π(2 − D)]
this approach is that a well-defined localized roughness is c1 = C. (33b)
2π
produced and unwanted roughness is eliminated which re- Fractal analysis allows the characterization of surface
sults into better mechanical durability. However, creation roughness by two parameters D and C, which are instru-
of localized roughness may be an impractical approach be- ment independent and unique for each surface. D (gener-
cause mechanical properties of the asperity tips and load ally ranging from 1 to 2) primarily relates to distributions
variations in an application may be uncontrolled, making it of different frequencies in the surface profile and C to the
difficult to select optimal shape and number of asperities. amplitude of variations at all frequencies. η is the lateral
resolution of the measuring instrument, τ is the size of the
3.3. Fractal analysis of contacts increment (distance) and ω is the frequency of the rough-
ness. Note that if S(τ ) or P (ω) are plotted as a function
Due to multiscale nature of surfaces, it is found that the of τ or ω, respectively, on a log–log plot, then the power
surface roughness parameters depend strongly on the reso- law behavior would result in a straight line. The slope of
lution of the roughness measuring instrument or any other line is related to D and the location of the spectrum along
form of filter, hence not unique for a surface. Therefore, the power axis is related to C.
the predictions of the contact models based on conventional A fractal model of elastic–plastic contacts [23] has been
roughness parameters may not be unique to a pair of rough developed to predict whether contacts experience elastic or
surfaces. However, if a rough surface is characterized in plastic deformation, and to predict the real area of con-
a way such that the structural information of roughness tact and statistical distribution of contact points. Based on
at all scales is retained (possible with modern roughness the fractal model of elastic–plastic contact, whether con-
measurement tools – atomic force microscope), then it will tacts undergo elastic or plastic deformation is determined
be more logical to use such a characterization in a con- by a critical area that is a function of D, C, hardness, and
tact theory. In order to develop such a contact theory, it is modulus of elasticity of the mating surfaces. If the contact
first necessary to quantify the multiscale nature of surface spot is smaller than the critical area, it undergoes plastic
roughness. deformation and large spots undergo elastic deformations.
14 B. Bhushan / Contact mechanics of rough surfaces

Figure 15. Schematic of local asperity deformation during contact of a

rough surface; upper profile measured by an optical profiler and lower
profile measured by AFM. Typical dimensions are shown for a polished,
magnetic thin-film rigid disk against a flat magnetic slider surface [9].

The critical contact area of inception of plastic deformation

for a magnetic thin-film disk was reported by Majumdar
and Bhushan [23] to be about 10−27 m2 – so small that
all contact spots can be assumed to be elastic at moderate
loads.
The question remains as to how large spots become elas-
tic when they must have initially been plastic spots. A pos-
sible explanation is shown in figure 15 [9]. As two sur-
faces touch, the nanoasperities (detected by AFM type of
instruments) first coming into contact have smaller radii of
curvature and are therefore plastically deformed instantly,
Figure 16. Schematic of (a) two 3-D rough surfaces in contact, and (b) cor-
and the contact area increases. When load is increased,
responding contact areas.
nanoasperities in the contact merge, and the load is sup-
ported by elastic deformation of the large-scale asperities study plane-strain and frictionless elastic contact problems
or microasperities (detected by optical profiler type of in- with only few cylindrical asperities with constant radius.
struments). For 3-D rough surface contact problems with many as-
Majumdar and Bhushan [23], Bhushan and Majum- perities of arbitary shape, the requirement of large num-
dar [24], and Bhushan [6] have reported relationships for ber of mesh elements makes the finite element approach
cumulative size distribution of the contact spots, portions unfeasible. One of the techniques provides a determinis-
of the real area of contact in elastic and plastic deformation tic solution to stresses and areas for the approach of two
modes, and the load–area relationships. 3-D real rough surfaces, figure 16 [32–38]. This technique
takes full account of the interaction of deformation from
3.4. Numerical 3-D contact models and computer all contact points and predicts contact geometry of real sur-
simulations faces under loading. It provides useful information on the
contact pressure, number of contacts, their sizes and dis-
Williamson [25], Gupta and Cook [26] and Bhushan and tributions, and the spacing between contacts. For contacts
Cook [27] digitized measured profiles of contacting sur- with a moderate number of contact points, both elastic and
faces and used them for computer simulation. For elas- elastic–plastic analyses of two rough surfaces can be car-
tic deformation at individual contacts, the contact pressure ried out [35,38,39]. These analyses develop the physical
and contact area were approximated by the Hertz analysis. model of two rough surfaces in elastic contact. As shown
For plastic deformation, Gupta and Cook [26] analyzed the in figure 16, the real area of contact between the two bodies
contacts using the plane-strain slip line field solution. In occurs at the tips of highest asperities. The contact area is
their analysis, they assumed that peaks of asperities were a small fraction of the surface areas of the contacting bod-
spherical in shape, which is not valid. Finite element meth- ies. Therefore, we can assume that the asperity contacts of
ods [30] have been used to solve elastic plane-strain con- each body occur on an elastic half-space. Another assump-
tact problems. Komvopoulos and Choi [31] used the fi- tion used in this study is that the area of individual contact
nite element method (commercial package ABAQUS) to is much smaller than the radii of curvature of contacting
 B. Bhushan / Contact mechanics of rough surfaces 15

The contacting interface region is discretized into a mesh

of small elements (2a×2b) over the domain of entire contact
area as shown schematically in figure 17 and a piecewise
interpolation of contact pressure distribution can be used
inside each element. The various Lagrange polynomials
(a) can be used for piecewise interpolation of contact pressure
inside each element. However, when a mesh element is
small enough, the contact pressure at each element can be
treated as a constant. In this case, equation (34) reduces to
ZZ
1 p(x0 , y 0 ) dx0 dy 0
(uz )l = p
πE ∗ (x − x0 )2 + (y − y 0 )2
Ω

M  ZZ 
1 X dx0 dy 0
= p pk
πE ∗ (x − x0 )2 + (y − y 0 )2
k=1 Ωk

X
M
= Ckl pk , (35a)
k=1

where (uz )l is the surface displacement at the center of the

rectangular element of 2a × 2b, Ωk is the sub-domain of
contact area of a single element, M is the total number of
intitial contact points (from a total of N 2 elements) where
the geometrical interference occurs based on relative rigid
body movement of the two rough surfaces, k and l are the
two indices representing the contact pressure location and
surface displacement location, respectively, and Ckl is the
(b) element of the so called “influence” matrix and based on
Figure 17. Discretization of the contacting rough surfaces (a) 2-D profile
Love [85] is given as
and (b) top view of discretized surface [38].
(  p 
1 (y + b) + (y + b)2 + (x + a)2
Ckl = (x + a) ln p
asperities. This allows the use of the linear theory of elas- πE ∗ (y − b) + (y − b)2 + (x + a)2
ticity as well as the approximation of plane surface around  p 
(x + a) + (y + b)2 + (x + a)2
the real contact area. Sayles and co-workers (e.g., [34,35]) + (y + b) ln p
have used an iterative process with a conventional matrix (x − a) + (y + b)2 + (x − a)2
inversion technique to solve the contact problem for contact  p 
(y − b) + (y − b)2 + (x − a)2
pressure and contact area. The matrix inversion approach + (x − a) ln p
starts with the classical approach for finding the relation- (y + b) + (y + b)2 + (x − a)2
ship between the contact pressure p(x, y) and surface dis-  p )
(x − a) + (y − b)2 + (x − a)2
placement uz (x, y) due to Boussinesq [83] and Cerruti [84] + (y − b) ln p . (36)
who made use of the theory of potential. Their theory (x + a) + (y − b)2 + (x + a)2
is described by Love [85]. The Boussinesq–Cerruti theory Equation (35a) can also be expressed in a matrix form as
(commonly referred to as Boussinesq theory) gives the nor-
mal surface displacement at a general surface point (x, y) u = C · p, (35b)
due to a concentrated contact (load) at some other point on where C, u and p are the influence matrix, displacement
the surface of a semi-infinite body. If the displacement of vector and contact pressure vector, respectively.
a surface point is required due to a number of concentrated This equation, adjusted by putting the pressure as unity,
contact points over area Ω, the total displacement is is used to form the influence coefficient of any contacting
ZZ point on any other point on the surface. The total deflection
1 − ν2 p(x0 , y 0 ) dx0 dy 0
uz (x, y) = p , (34) at any point can therefore be calculated by summing each
πE (x − x0 )2 + (y − y 0 )2 element of pressure times its influence coefficient:
Ω
X
where the Ω is the domain of the contact surface on which Total deflection of a point = elemental pressures ×
surface forces act. For two bodies in contact, the compos-
influence coefficients.
ite surface displacement uz (x, y) is given by equation (34)
after replacing E/(1 − ν 2 ) by the composite Young’s mod- Thus, if the pressures on each contacting element are
ulus E ∗ . known, the deflection of any point on the surface and hence
16 B. Bhushan / Contact mechanics of rough surfaces

deformed geometry can easily be found. However, the el-

emental pressures are unknown, an iteration process has to
be used to solve the equation (35). For a given rigid body
normal approach between two rough surfaces, the surface
displacements at the contact surface are determined from
the geometrical interference of the two bodies. Then, equa-
tion (35) is used to solve for the pressure distribution inside
the contact region. Those contact elements with negative
(tensile) contact pressure are excluded for the next iteration
since tensile force is not permitted at the contact surface.
The deformations over the entire contact surface are then
computed using equation (35) to check that no geometrical
interference occurs outside the chosen contact region. The
iteration is repeated until convergence occurs to a set of sur-
(a)
face elements with positive or zero contact pressure. These
surface elements with positive contact pressure define the
real area of contact, and the total applied load is the sum
of the contact pressure at those surface elements. One of
a major difficulty for contact problems with a large num-
ber of contact points is that the storage of influence matrix
usually requires a large amount of computer memory since
for M contact node points, the total number of influence
matrix elements will be M 2 . Ren and Lee [37] developed
a moving grid method to reduce the size of the computer
storage for storing a large influence matrix.
For a large number of contact points, the influence ma-
trix may become very large, and possibly ill-posed. The
large, ill-posed influence matrix will not only require a great
deal of time for the conventional matrix inversion method
to complete a large number of iteration cycles, but also, in (b)
some cases, cause serious nonconverging problems. Instead
of using the conventional matrix inversion approach, Tian Figure 18. (a) Definition of strain energy and complementary energy, and
(b) relationship between elastic strain energy and internal complementary
and Bhushan [38] utilized the approach on the basis of vari-
energy for a linear elastic or a linear elastic–perfectly plastic material [38].
ational principle. The use of the variational principle leads
to a standard quadratic mathematical programming prob- plementary energy is shown in figure 18(a). The minimum
lem after an infinite-to-finite dimension transformation. In total complementary potential energy principle was used
the variational method, the real area of contact and contact in the analysis developed by Tian and Bhushan [38] since
pressure distributions are those which minimize the total it is more convenient to work in terms of unknown con-
complementary potential energy. The variational approach tact pressure [41] for elastic contact problems. For the
uses a direct quadratic mathematical programming method minimum total complementary potential energy principle,
which guarantees the uniqueness of the solution of rough the variational theory states that of all possible equilibrium
surface contact problems. It is also able to significantly stress fields for a solid subjected to prescribed loadings
reduce computing time since it does not need an additional and boundary displacements, the true one corresponding to
iteration process which is required by the conventional ma- a compatible strain field, renders the total complementary
trix inversion technique, and makes it feasible to solve the energy stationary [86]. For rough surface contact, when
3-D rough surface contact problem with a large number of the pressure distribution and the real area of contact are
contact points. The variational approach will be described
unknown, the problem then becomes to find the minimum
next.
value of an integral equation that minimizes the total com-
Besides differential equations, the contact problem can
plementary potential energy of the contacting system.
also be governed by minimum potential energy theory. The
For an elastically deformed body, the total complemen-
boundary value problem of differential equations for a me-
tary potential energy is given by [86]
chanics system is equivalent to the problem of solving a Z Z
minimum value of integral equation, i.e., a variational prob- V = U0 (σij ) dD − u∗i Ti dΩ,
∗ ∗
(37)
lem. In the variational approach, two minimum energy
D Ω
principles, total elastic strain energy and total complemen-
tary potential energy, can be used for solving mechanics where V ∗ is the total complementary potential energy, U0∗ is
problems [86]. The definition of strain energy and com- the internal complementary energy density, D is the domain
 B. Bhushan / Contact mechanics of rough surfaces 17

of deformed body, Ω is the domain of the contact surface on The total prescribed displacement of the two contacting
which the surface forces, Ti , act. Ti , σij and u∗i are surface bodies, u∗zl , in equation (41) can be determined using the
forces, stresses and prescribed displacements, respectively. following geometrical interference criterion:
For a contact problem of two rough surfaces, the total
u∗z + f (x, y) − δ = 0 (within contact area),
complementary potential energy is given by
Z u∗z + f (x, y) − δ > 0 (outside contact area),
V ∗ = UE∗ − p(u∗z1 + u∗z2 ) dΩ where δ is the rigid-body movement under applied load,
Ω and f (x, y) is the initital separation of the two contact sur-
Z faces and equals
= UE∗ − pu∗z dΩ, (38)
f (x, y) = z1 (x, y) − z2 (x, y) .
Ω
Here we assume that at the initial position, there is no
where UE∗ is the internal complementary energy of the two contact between the two surfaces. Thus, within the contact
stressed bodies, p is the contact pressure, u∗z1 and u∗z2 are area, the total prescribed displacement of the two contacting
the prescribed displacements of the two contacting bodies bodies, u∗z , in equation (35) equals
identified as 1 and 2, respectively, and u∗z is the total pre-  
scribed displacement of the two contacting bodies inside the u∗z = δ − z1 (x, y) − z2 (x, y) . (42)
assumed contact zone based on geometrical interference.
The total complementary potential energy given in equa-
For linear elastic materials, the internal complementary
tion (41) is a standard quadratic function of the constant
energy UE∗ is numerically equal to the elastic strain en-
pressure, and can be solved using mathematical program-
ergy UE as shown in figure 18(b) and can be expressed in ming methods once all possible contact areas are included
terms of surface forces and displacements by [67] in equation (41). Since no tension (forces in the direction
Z Z opposite to motion of contacting bodies) is allowed at the
1 1
UE = p(uz1 + uz2 ) dΩ = puz dΩ, (39) contact interface, the search for a minimum value of the
2 2
Ω Ω equation (41) is restricted by

where uz is the composite surface displacement inside the pk > 0, k = 1, . . . , M , (43)

contact zone to be calculated, and equals total contact de-
where M is the total number of initial contact points. We
formation of two contacting bodies.
note that the number of final contact points which satisfy
Combining equations (38) and (39), the total comple-
the restriction, pk > 0, would be smaller than M .
mentary potential energy for the two contacting bodies be-
One way to find the minimum value of equation (41)
comes
under the restriction of equation (43) is to use a standard
ZZ ZZ
∗ 1 simplex-type algorithm [87]. Examples of using the stan-
V = puz dx dy − pu∗z dx dy. (40) dard simplex-type algorithm for elastic contact problems
2
Ω Ω can be found in some earlier studies [40,41]. However,
the use of the simplex-type algorithm is limited to an opti-
Now, to relate uz with p, Boussinesq’s solution can be
mization problem which only contains a smaller number of
used to relate contact pressure with the surface displacement
variables. For the search of feasible solution in each step
for normal distributed load on the surface of a semi-infinite
of simplex algorithm is rather tedious and time consum-
body given earlier in equations (35) and (36).
ing when the number of variables is large. In the model
After substituting uz in equation (40) by equation (35),
developed by Tian and Bhushan [38] a direct quadratic
the contact problem of finding a minimum value of the total
mathematical programming method is used to obtain the
complementary potential energy now redues to solving the
minimum value of complementary potential energy. To re-
minimum value problem of the integral equation (40) in
duce the required storage space for the influence matrix,
terms of contact pressure p(x, y).
a space-reduced matrix storage was developed to compress
In order to obtain an approximate value of minimum
the entire influence matrix elements in a space-reduced ma-
complementary potential energy, the integral equation (40)
trix.
for the total complementary potential energy and equa-
Tian and Bhushan [38] developed a computer program
tion (35) relating surface displacement to the surface pres-
to obtain the minimum value of the total complementary
sure are discretized into a mesh of small elements over the potential energy of equation (41) with restrictions of equa-
entire contact area (figure 17), as described earlier. tion (43). In the program, the 3-D surface profiles of the
After discretization of equations (35) and (40), the total two surfaces are initially read in. Contacting surfaces are
complementary potential energy further reduces to discretized into small elements corresponding to the dif-
! ferent points of surface heights. For a given rigid body
1X X X
M M M
∗ approach (or load) between two rough surfaces, the total
V = pl Ckl pk − pl uzl . (41)
2 surface displacement at the surface of real contact is equal
l=1 k=1 l=1
18 B. Bhushan / Contact mechanics of rough surfaces

Figure 20. Schematic of a half-space subjected to both surface normal and

tangential forces over an arbitary area Ω. Concentrated normal loading
and friction force on the area Ω is shown.

once the local contact pressure exceeds the hardness of the

softer material. An expression for the total complimentary
potential energy during plastic deformation is developed by
Tian and Bhushan [38] and its minimum value is searched
to obtain real area of contact and contact pressures.
Subsurface stress fields and the effect of friction on
stresses have been analyzed by Bailey and Sayles [42], Lee
and Ren [89] and Yu and Bhushan [43]. For friction effects,
Figure 19. Flow chart of the computer program for contact analysis of
two rough surfaces [38].
the contact pressure and contact locations (contact points)
are considered not to be altered by the presence of surface
to the interference of two contacting bodies. The elements friction. This assumption of no effect on contact pressure is
with finite displacement are included in the formulation. strictly true only when the two bodies are smooth or have
An influence matrix is constructed to relate contact pres- the same elastic contacts. In the case of a rough-on-rough
sure to the given displacements dependent on the location contact, asperity interactions may invalidate this assump-
of the pressure element and contact points and an expression tion, although for the case of a smooth-on-rough contact
for the total complementary potential energy involving dis- with approximately equal moduli, it is considered a rea-
placement and pressure is developed. The minimum value sonable assumption. For the case of friction present at the
of the total complementary potential energy is obtained us- interface, tangential force at each contact is equal to its
ing direct quadratic mathematical programming technique. contact pressure multiplied by the coefficient of friction.
The real area of contact and contact pressure distribution Subsurface stress calculations are based on analyses for
are those which minimize the total complementary potential the stresses produced in a semi-infinite solid by surface
energy. The analysis solves for contacts with positive pres- normal forces [85] and tangential forces [90] over a rectan-
sure. Figure 19 shows the flow chart of the main program. gular element or patch. For analytical solutions to stresses
The main feature of the this program is that the correspond- and displacements, also see [91]. At each point, the stress
ing contact pressure for a given rigid body approach can be
is calculated by integrating the influences of surface nor-
obtained directly from a single minimization process, thus
mal and tangential forces over the entire nominal contact
eliminating the additional iteration cycles in the conven-
area, figure 20. As presented earlier, the nominal area is
tional matrix inversion methods described earlier. By com-
divided into rectangular elements (or patches), the stress
parison, the elimination of these iteration cycles effectively
speeds up the entire process several times as compared to being simply the sum of the effects of each element. Ana-
conventional matrix inversion methods. lytical expressions for stresses at any arbitary point inside
So far, only elastic contact models have been presented. the half-space due to a uniformly distributed load on a rec-
These models have been extended to elastic–plastic contact tangular element are used to obtain the stress components
analyses for the case in which material is assumed to de- for all subsurface locations.
form as an elastic–perfectly plastic material [35,38,39,88]. Merriman and Kannel [44] and Cole and Sayles [45]
In these analyses, it is assumed that the regime of plastic developed a 2-D model to analyze the contact of layered
deformation is confined to a very small area and does not elastic bodies (cylinder on a flat body) with real rough sur-
significantly alter the geometry of the elastically deformed faces. They presented the contact pressure distributions for
contact surface. Outside the plastic regime, the relationship layers of varying thicknesses and elastic moduli. They re-
between contact pressure and elastic deformation (equa- ported that a soft surface layer will reduce the magnitude
tion (35)) still holds. A contact point deforms plastically of peak contact pressures due to roughness.
 B. Bhushan / Contact mechanics of rough surfaces 19

(a) (a)

(b)
Figure 21. (a) Undeformed and deformed surface profiles of rough sphere
and contact pressures normalized with the maximum hertzian pressure (b)
plotted along y = z = 0 line and (b) 3-D contact pressures map for a
rough sphere in contact with a smooth flat surface or a smooth sphere 1/2
Figure 22. Lines of constant normalized von Mises stresses (J2 /p0 )
(hertzian) [89]. plotted on the y = 0 plane for (a) a perfectly smooth sphere in contact
with smooth flat surface and (b) a rough sphere in contact with a smooth
3.4.1. Simulation of cylindrical or spherical surface on flat surface in frictionless contact [89].
itself or on a nominally flat surface
A number of investigators have investigated contact in the von Mises shear strain energy criterion, yielding will
1/2
a two-dimensional (cylinder on itself or on a flat surface) occur when the von Mises stress J2 reaches the yield
or three-dimensional (sphere on itself or on a flat surface) stress in simple tension. Figures 22 and 23 show the lines
1/2
fashion to simulate contacts in machine elements such as of constant normalized von Mises stresses (J2 /p0 ) for
gears, cam – cam follower, and rolling element bearing smooth and rough spheres, with and without interfacial
components [34,37–39,42,88,89,92]. Roughness is gener- friction. Note that most of the high stresses are concen-
ally placed on one of the two surfaces and another sur- trated near the top. Further down the surface, the stress
face is assumed to be smooth. We present here typi- field becomes similar to that of the smooth surface solu-
cal results of a spherical surface in sliding contact with tion. The stresses in friction case (figure 23) are larger
a rigid flat surface, based on Lee and Ren [89]. Fig- than in the frictionless case (figure 22). For the friction
ure 21 shows undeformed and deformed surface profiles case, the location of the maximum stress has moved closer
of a rough sphere and contact pressure profiles of smooth to the surface and it is no longer directly below the cen-
(hertzian) and rough surfaces. Note that the pressure pro- ter of the hertzian conjunction. Figure 23(c) shows the
file for the rough surface has many pressure peaks with von Mises stresses magnified at the surface for sliding
magnitudes far greater than that of the smooth surface con- contact of the rough surface. The peak values are much
tact. The highest pressure peaks are found near the center greater than the maximum stress (0.37) of the smooth sur-
of the hertzian contact. For ductile materials, based on face.
20 B. Bhushan / Contact mechanics of rough surfaces

(a) (a)

(b)
(b)
Figure 24. Normalized tensile principal stress field for (a) a perfectly
smooth sphere in contact with a smooth flat surface and (b) a rough sphere
in contact with a smooth flat surface, in sliding contact with a coefficient
of friction of 0.25 [89].

Tensile radial stresses makes brittle materials suscepti-

ble to ring cracks. Therefore, for brittle materials, tensile
stress may be more important than maximum shear stress
(von Mises stress) at the subsurface. Figure 24 shows
3-D plots of the normalized tensile principal stresses for
smooth and rough surfaces in sliding contact. For a smooth
surface, the maximum value is located at the x = −a
and y = 0. For frictionless contact, the stress field is
symmetrical and surrounds the hertzian conjunction. As
the friction increases, the stress field becomes asymmetri-
cal and the maximum grows larger. For the rough surface,
(c) there are many peaks which far exceed the maximum value
1/2 of the smooth surface. The largest tensile stresses occur
Figure 23. Lines of constant normalized von Mises stresses (J2 /p0 )
plotted on the y = 0 plane for (a) perfectly smooth sphere in contact with near the asperity contacts.
smooth flat surface, and (b) a rough sphere in contact with a smooth flat Ren and Lee [93] showed that the asperity aspect ra-
surface, and (c) stresses close to the surface in the case (b), for the case tio has little effect on the real area of contact and contact
of sliding contact with a coefficent of friction of 0.25 [89]. pressures.
 B. Bhushan / Contact mechanics of rough surfaces 21

tire surface. Yu and Bhushan [43] calculated surface and

subsurface stresses. Figure 26 shows the contact pressure
map at a nominal pressure of 32.8 kPa, typical of magnetic
head-disk interface. Contact pressure at the asperities is
high, compared to the rest of the contact region. Under the
applied load, a small number of contact spots (6) are ob-
tained and the deformation is elastic for ceramic to ceramic
contacts. Contours of von Mises stresses on the surface
and subsurface (y = 11 µm) on and close to the maximum
von Mises stress plane for frictionless and frictional con-
tacts are shown in figure 27. Maximum von Mises stress at
both frictionless and frictional contacts occurs very close to
the surface. The tensile principal stress distributions at and
near the location of maximum von Mises stress, are shown
in figure 28. A pulse of high stress appears
√ on the sur-
Figure 25. Computer generated surface profile: standard deviation of face (z = 0) at the location of maximum J2 , x = 16.5 µm
surface heights (σ) = 1 nm, correlation length (β ∗ ) = 0.5 µm, and total and y = 11 µm. The effect of friction is to increase the
scan size = 20 × 20 µm2 . magnitude of the maximum stress but not to change the
distribution.
Figure 29 shows the computed results of the real area
of contact under various applied nominal pressures. As
expected, both the real area of contact and the number of
contact points increase with nominal pressure. The bear-
ing area of the contact surface, corresponding to the same
geometrical interference is given by slicing off the surface
for a given interference. Tian and Bhushan [38] showed
that the bearing area greatly overestimates the real area
of contact because elastic deformation is not included in
the bearing area calculations. The relationship between the
maximum contact pressure and the real area of contact and
applied nominal pressure (load) for two rough surfaces is
shown in figure 30. It can be seen that the real area of con-
tact is linearly proportional to the applied pressure when
Figure 26. Contact pressure map of a computer generated rough surface
the pressure is relatively small. This relationship between
(σ = 1 nm and β ∗ = 0.5 µm) on a rigid smooth flat surface (E ∗ =
100 GPa) at a nominal pressure of 32.8 kPa [43]. the applied normal pressure and the real area of contact is
consistent with what is commonly observed in engineering
3.4.2. Simulation of two nominally flat surfaces practice, i.e., the friction force is linearly proportional to
Tian and Bhushan [38], Bhushan and Chilamakuri [57], the applied pressure. Figure 30 also shows the maximum
Poon and Bhushan [94,95], Yu and Bhushan [43] and Chil- contact pressure under different applied loads. Contrary to
amakuri and Bhushan [58] used their numerical model to common belief, we note that the contact pressure increases
predict the real area of contact and contact stresses be- with applied nominal pressure. Figure 31 shows the con-
tween a rigid flat and computer generated or measured tact pressure maps and contours of von Mises stresses at
rough surfaces. A digital filter technique combined with the surface and subsurface (y = 6.5 µm) for two surfaces
a fast Fourier transform (FFT) developed by Hu and Ton- with σ = 3 nm and β ∗ = 0.1 µm, at a nominal pres-
der [96] was used to generate 3-D gaussian and nongaussian sure of 32.8 MPa (1000 × of nominal pressure used in
rough surfaces with given standard deviation of surface previous figures). The real area of contact is low and the
heights and correlation lengths. A typical computer gen- contact pressure and von Mises stresses are high for the
erated, 3-D rough surface map is shown in figure 25. The surface with σ = 3.0 nm as compared to other surface
heights of the surface map consist of 256 × 256 data points, with σ = 1.0 nm. For the surface with σ = 3.0 nm, contact
and the profile data file has the similar data format of a occurs at 1570 contact points, among which contact pres-
measured surface map from an atomic force microscope or sure at the surface and von Mises stress very close to the
a noncontact optical profiler. An exponential autocorrela- surface are very high at six points and plastic deformation
tion function was used to generate the rough surface. The occurs. Since the maximum von Mises stress occurs very
standard deviation of surface heights of the rough surface close to the surface, plastic deformation and consequently
in figure 25 is 1 nm, the correlation length at which the wear will occur on and close the surface.
autocorrelation value drops to 0.1 of that at origin is 0.5 Poon and Bhushan [97] developed computer generated
µm, and the total scan size is 20 × 20 µm2 for the en- gaussian surfaces with various values at σ and β ∗ . For
22 B. Bhushan / Contact mechanics of rough surfaces

(a) (b)
Figure 27. Contours of von Mises stresses (a) on the surface and (b) and (c) in the subsurface (y = 11 µm) at µ = 0, 0.25 and 0.50, for the case of
a rough surface (σ = 1 nm, β ∗ = 0.5 µm) on a rigid smooth flat surface (E ∗ = 100 GPa) at a nominal pressure of 32.8 kPa. The contour levels are
natural log values of the calculated stresses expressed in kPa [43].
 B. Bhushan / Contact mechanics of rough surfaces 23

(c)
Figure 27. Continued.

Figure 29. 2-D images of real area of contact between a computer gen-
erated rough surface (σ = 1 nm, β ∗ = 0.5 µm) on a rigid smooth flat
surface (E ∗ = 100 GPa) at three different nominal pressures [43].

these surfaces, in the elastic contact regime, they calculated

real area of contact and contact pressure. They found that
contact area is inversely proportional to, and the contact
Figure 28. Distributions of tensile principal stresses on the surface along
x- and y-axes at and near the location of maximum von Mises stress
pressure is proportional to σ/β ∗ , as proposed by Onions
(x = 16.5 µm, y = 11 µm) for the case of a rough surface (σ = 1 nm, and Archard [12]. Therefore, plastic deformation can be
β ∗ = 0.5 µm) on a rigid smooth flat surface (E ∗ = 100 GPa) at a avoided by decreasing σ or increasing β ∗ to promote bet-
nominal pressure of 32.8 kPa [43]. ter interface durability. Yu and Bhushan [43] studied the
24 B. Bhushan / Contact mechanics of rough surfaces

predominantly submicron details and increasing τ will re-

sult in a loss of high frequency information. As τ increases,
high frequency details of the original profile gradually dis-
appear resulting in higher β ∗ . σ is the vertical parameter
and is not sensitive to τ but generally increases with L.
β ∗ is the spatial parameter affected by both τ and L.
Next, the effect of sampling interval on contact area and
contact pressure is shown in figure 33. For both surfaces,
contact area decreases and contact pressure increases as the
sampling interval decreases. The contact pressure of the
GC surface is always higher than that of the Ni–P surface.
As the sampling interval becomes smaller, the maximum
contact pressure can approach the hardness value. The sub-
surface stress with low interfacial friction is always higher
than the surface stress and plastic deformation is expected
to occur before surface stress reaches the hardness.
From figure 33, the asperity contact pressure of both sur-
faces reaches the hardness of the softer material as τ de-
creases. This implies that the asperity contact of real sur-
faces starts from plastic deformation, the scale of which
depends on different surface topographical structures. The
GC surface has a larger scale of plastic deformation be-
cause it has higher frequency structures, compared to the
Ni–P surface. Thus, roughness distribution plays a key role
in plastic deformation on asperity contacts.

3.5. Dynamic friction modeling

In dynamic friction models, a constant value of coef-

Figure 30. Predicted maximum contact pressure and real area of contact as ficient of friction is assumed and perturbations under dy-
a function of applied nominal pressure for the case of two rough surfaces namic conditions are investigated [46,47,98,99]. Oden and
on a rigid smooth flat surface (E ∗ = 100 GPa) [43]. Martins have presented continuum models for solving a
large class of dynamic frictional phenomena and compu-
effect of σ, β ∗ , and friction on subsurface stresses. Increas- tation methods for their analyses. Their particular inter-
ing σ and friction, and decreasing β ∗ generally increased ests were to adequately predict stick–slip motion, fric-
the magnitude of stresses. tional damping in structural dynamics, and sliding resis-
Poon and Bhushan [94,95] studied the effect of high fre- tance. Through experimental observations and theoretical
quency roughness (sampling resolution of roughness mea- analyses, they identified the key features that the interface
surement) on contact pressure and contact area. They mea- models should exhibit and developed interface models to
sured rough surfaces of Ni–P coated Al–Mg (Ni–P) and reflect these features. Their considerations of the normal
glass-ceramic (GC) substrates used for construction of mag- deformability of the interface asperities, nonlinear bound-
netic disks, by an AFM using different scan sizes L × L ary effect, and generalized friction law are different from
with L = 64, 32, 16, 8, 4, 1, 0.5, 0.25 and 0.1 µm, each the classical approaches of the statistical description of the
containing 256 × 256 points. Therefore, the sampling inter- interface geometry and normal contact, nonpenetration uni-
val τ decreases linearly with L. The variation of σ and β ∗ lateral contact condition, and Coulomb’s friction law. In
with scan size for the Ni–P and GC surfaces is shown in their interface models, as shown in figure 34, a continu-
figure 32. From the curves of σ with L, σ of each surface ous material body β is considered to be in contact with
initially increases with L and approaches a constant value another material β1 over a contact surface Γc ⊂ ∂β. u is
when L reaches about 16 µm. This indicates that both denoted the displacement of particles X ∈ β relative to
surfaces contain a long-wavelength limit of about 16 µm. a fixed reference configuration which is located at the un-
Therefore β ∗ , which represents the wavelength structure deformed body. It is assumed that the contact surface Γc
of a random surface, should approach a constant value as has an exterior normal vector n and that the actual inter-
L reaches 16 µm. But figure 32 shows that β ∗ increases face (asperities, oxide film, gas, work-hardened material,
with L. The increase is due to increasing τ as L increases. etc.) is initially of thickness t0 . The initial gap between
As τ increases, the high frequencies in the surface will ap- Γc and the other contact surface is defined as the distance
pear as a lower frequency in the digitized signal [7] and er- and denoted as g, which is measured along a line normal
ror in sampling occurs. The surfaces being studied contain to Γc between the highest asperities of these surfaces in
 B. Bhushan / Contact mechanics of rough surfaces 25

(a)

(b)

(c)
Figure 31. (a) Contact pressure maps and (b) contours of von Mises stresses on the surface and (c) in the subsurface at y = 6.5 µm at a nominal
pressure of 32.8 MPa for two rough surfaces on a rigid smooth flat surface (E ∗ = 100 GPa) [43].

the reference configuration. The interface thickness after components of the Cauchy stress tensor at a particle X in β,
deformation is denoted as t and the actual displacement of then the normal stress developed in Γc is
Γc in the direction of n is un = u · n. Thus, the approach
of the contact surfaces is σn = σij ni nj . (45)

δ = t0 − t = (un − g)+ . (44) The relationship between σn and δ is obtained by plot-

ting the variation in the normal force versus the penetrating
This definition is valid for a rigid body β1 bounded by an approach as obtained from experimental results, which ap-
ideally flat surface. Applications of this model to cases pears to be a power-law relation as given by
of deformable and rough surfaces have been carried out
by Kikuchi and Oden [100]. Let σij denote the cartesian −σn = cn (un − g)mn
+ , (46)
26 B. Bhushan / Contact mechanics of rough surfaces

Figure 33. Variation of mean contact pressure (pm ), maximum contact

Figure 32. Variation of standard deviation of surface heights and correla- pressure (p0 ) and contact area with sampling interval for Ni–P and GC sur-
tion length with scan size [93]. faces [93].

where cn and mn are material parameters determined from

the establishment of existence and uniqueness results un-
interface compression experiments. If there is friction
present on the contact interfaces, the following friction law der appropriate hypotheses on the regularity and smallness
applies, as σT denotes the tangential stress on the contact of the data, particularly the smallness of the coefficient of
surface: friction for prescribed external loads.
Martins et al. [47] continued Oden and Martins’ work [46]
un 6 g ⇒ σT = 0, to study small speed sliding phenomena. They reviewed the
un > g ⇒ |σT | 6 cT (un − g)m T
+ , or relationship between static and kinetic friction, examined
|σT | < cT (un − g)+ if ùT = U , or
mT the role played by the dynamic properties of experimental
apparatus, and studied the importance of normal degree-of-
|σT | = cT (un − g)m
+
T
freedom under frictional sliding. The time and dissipate
if ∃λ > 0, ùT − U = −λσT , (47) effects are included in the constitutive equation and fric-
where cT and mT are material constants depending on in- tional law as given respectively in the following:
terface properties, ùT is the tangential velocity of material
particles on Γc , and U is the prescribed tangential velocity σn = −cn δ mn + bn δ tn δ,t , (51)
of β1 . This friction law, equation (47), is a generalization −σT ∈ cT δ mT
sgn(v), (52)
of Coulomb’s friction law and allows for possible devia-
tions from Amontons’ laws, i.e., a possible dependence of where the subscript t denotes differentiation with respect to
the friction coefficient µ on the normal stress according to time and sgn(v) denotes the multivalued relation
µ = C|σn |α , (48) (
1 if ζ > 0,
if the normal dissipate effects are negligible (denoted as ζ ∈ R → sgn(ζ) = [−1, 1] if ζ = 0, (53)
m /m −1 if ζ < 0.
bn ≈ 0), where α = mT /mn − 1 and C = cT /cn T n . If
mT = mn (and bn ≈ 0) the usual Coulomb’s law of friction They found that contrary to classical interpretations, no dis-
is recovered with µ = cT /cn . tinction between the static and kinetic coefficient of friction
Rabier et al. [101] further expressed Oden and Mar- is required for a body to have an accelerated motion. Their
tins’ [46] phenomenological laws of contact interfaces in other conclusion is that the experimentally observed depen-
the following forms for elastostatic problems: dence of static coefficient of friction on the time of station-
σn = −cn δ mn , (49) ary contact, distinctions between the static and kinetic co-
efficient of friction, and dependence of the latter coefficient
σT = cT δ m Á (if friction is present), (50)
on sliding speed at small speeds are not necessarily intrinsic
where Á is a unit tangential vector field parallel to the properties of dry metallic interfaces. Physically small exter-
sliding velocity. They formulated a class of contact prob- nal perturbations and geometric and dynamic properties of
lems in elastostatics according to the governing laws of an experimental apparatus may lead to these experimental
equations (49) and (50), and showed Oden and Martins’ observations even when the coefficient of friction is con-
models did provide a mathematical framework that admits stant and independent of sliding speed or time of contact.
 B. Bhushan / Contact mechanics of rough surfaces 27

Figure 34. Initial gap, normal displacement and penetrating approach at the contact surface Γc [46].

4. Multiple-asperity wet contacts where µr is the true coefficient of friction in the ab-
sence of meniscus, and smaller than the measured value
With the presence of a thin liquid film with a small of µ = F/W . The sum of W and Fm is the total normal
contact angle (wetting characteristics) such as a lubricant load. Fm is the meniscus force in the normal direction, and
or an adsorbed water layer at the contact interface, curved Fv is the viscous force in the sliding direction. The friction
menisci form around contacting and noncontacting asperi- force (µr W ) depends on the material properties and surface
ties due to surface energy effects. The attractive meniscus topography, while Fm depends on the roughness parameters
force arises from the negative Laplace pressure inside the as well as the type of liquid and its film thickness.
curved meniscus as a result of surface tension [48–50]. The For static friction calculations, the viscous effect can
product of this pressure difference and the immersed sur- be neglected; therefore, the measured coefficient of static
face area of the asperity is an attractive (adhesive) force, friction µs including the effect of the meniscus force is
referred to as the meniscus force. This intrinsic attractive given by
force may result in high static friction (stiction), kinetic  
friction and wear. This problem is particularly important in F Fm
µs = = µr 1 + , (55)
the computer data storage industry [3,51]. W W
In order to calculate the meniscus force, different menis- where µr Fm is the friction force due to liquid-mediated
cus geometries and assumptions have been used by various adhesion.
authors. Figure 35 presents the existing models for calcula-
tion of the meniscus force between a sphere and a flat, either
4.1. Statistical analysis of contacts
with a liquid droplet (cases A and B) or with a uniform liq-
uid film (case C). In the figure, fm is the force due to a
Several models have been developed to predict menis-
single meniscus along the normal direction. For multiple
cus forces developed at a wet interface using a statistical
menisci, the total meniscus force is determined by summing
approach [52–55,22]. A schematic of a rough surface in
the meniscus forces over all menisci formed at the interface.
contact with a smooth surface with a continuous liquid film
However, the contact statistics needs to be determined.
on the smooth surface, is shown in figure 36. Note that both
The total normal force on the wet interface is the exter-
contacting and near-contacting asperities wetted by the liq-
nally applied normal force plus the meniscus force. There-
uid film contribute to the total meniscus force. The Green-
fore, during sliding, the force required to inititiate or sustain
wood and Williamson [10] approach described earlier, is
sliding is equal to the sum of the intrinsic (true) friction
used to model the contact. The peak heights are assumed
force, Fi , and the stiction force, Fs , the latter being a com-
to follow a gaussian distribution function and peak radii are
bination of the friction force due to the meniscus and the
assumed to be constant. In general, given the peak-height
viscous effects [3],
distribution function p(z), the mean peak radius (Rp ), the
F = Fi + Fs = µr (W + Fm ) + Fv , (54) thickness of liquid film (h), the liquid surface tension (γ),
28 B. Bhushan / Contact mechanics of rough surfaces

Figure 35. Different meniscus geometries and formulas for calculation of the meniscus force for a single asperity in contact with liquid of surface
tension γ. In cases A and B, the contact angle θ is assumed to be the same for the liquid in contact with either surface; if the contact angles are
different, cos θ should be replaced by (cos θ1 + cos θ2 )/2. In case C, θ is the contact angle of the liquid with the sphere. Note that all meniscus forces
are attractive and directed normal to the meniscus contact.

and the contact angle for the liquid in contact with the than a molecular layer thick, Fm is zero since no meniscus
rough surface (θ), the total meniscus force (Fm ) at the slid- can be formed and the problem reduces to dry contact.
ing interface is obtained by summing up the meniscus forces Gao et al. [54] showed that the meniscus force increases
from all individual contacting and noncontacting asperities as a function of liquid film thickness (h). For a given
that form menisci over the nominal contact area, shown in film thickness, the meniscus force decreases with an in-
figure 36(a), crease in standard deviation of peak heights (σp ) and it in-
Z∞ Z∞ creases with an increase of radii of peaks (Rp ) and number
Fm = fm N0 p(z) dz = 2πRp γ(1 + cos θ)N0 p(z) dz, of peaks (N0 ), figure 37. Gao et al. [54] and Chilamakuri
and Bhushan [82] produced the design curves to provide a
d−h d−h
(56) relationship between critical σp and Rp for minimum stic-
where N0 is the total number of peaks in the nominal con- tion and no plastic deformation.
tact area. The interplanar separation, d, is determined from Meniscus force analyses discussed so far consider only
what happens at equilibrium. When a body first comes to
Z∞
4 rest on another body, in the presence of a liquid film, the in-
W + Fm = E ∗ Rp1/2 N0 (z − d)3/2 p(z) dz. (57) terface is not in equilibrium. In the early stages of meniscus
3
d formation, meniscus force increases as a result of flow of
An iterative numerical approach is used to solve equa- liquid into the low pressure region created inside the curved
tions (56) and (57). liquid–air interface causing them to grow until the Laplace
It is evident that the maximum meniscus force can be pressure is sufficiently reduced to match the disjoining pres-
obtained by setting h very large so that the integral in equa- sure of the liquid remaining on the surface. For the entire
tion (56) approaches its maximum value of unity. Therefore interface to reach equilibrium, liquid from the wet body
the maximum possible mensicus force is surrounding the other body has to flow into the interface
until the capillary pressure equals the disjoining pressure
Fmax = 2πRp γ(1 + cos θ)N0 , (58)
of the liquid film in the rest of the wet body. Bhushan
regardless of the distribution function of peak heights. Con- et al. [55] developed a kinetic meniscus model to predict
versely, when the film thickness h is very small, i.e., less the time dependence of static friction. In this analysis, the
 B. Bhushan / Contact mechanics of rough surfaces 29

(a)

(b)
Figure 36. (a) Schematic for a rough surface in contact with a flat surface
with a liquid film, and (b) schematic of contact area and meniscus area in
a contacting asperity.

liquid flow at the contacting and near contacting asperities

is calculated based on Laplace pressure. The liquid menis-
cus at equilibrium is obtained. The meniscus growth with
time is used in this analysis to calculate the time dependent
static friction. They found that the relative meniscus force
reached a maximum at a rest time called the “equilibrium”
time. The equilibrium time is found to increase with an
increase in liquid film thickness, liquid viscosity and aver-
age asperity radius. Growth rate of the relative meniscus
force is inversely proportional to the liquid viscosity. This
model explains the time dependent stiction encountered in
magnetic disk drives [3].
Kotwal and Bhushan [22] developed a statistical model
for nongaussian surfaces with skewness and kurtosis. They
used a Pearson system of frequency curves, based on the
method of moments, to obtain analytical expressions for the Figure 37. Ratio of the meniscus force to applied load (Fm /W ) as a
probability density functions for distributions with various function of water film thickness at different σp , Rp and N for magnetic
skewness and kurtosis values. These probability density head–disk interface [54].
functions were then used to calculate the real area of con-
tact, contact pressure and meniscus forces. They reported 4.2. Numerical 3-D contact models
that nongaussian surfaces with a range of positive skewness
(between 0.3 and 0.7) and a high kurtosis (greater than 5) A first numerical meniscus model was developed by Tian
significantly lower the real area of contact and meniscus and Bhushan [56]. They were first to modify the classical
forces and these surfaces are somewhat insensitive to the meniscus theory of a single asperity (sphere on a flat) con-
liquid film as far as magnitude of meniscus force is con- tact presented earlier, to include the effect of multi-asperity
cerned. Futher discussion will be presented in the next contacts with a pre-existing ultra-thin liquid film during
section. contact of two rough surfaces. They modified the numerical
30 B. Bhushan / Contact mechanics of rough surfaces

contact model based on the variational approach presented

earlier, to take into account the effect of meniscus forces.
For a meniscus formed around a pair of contacting asper-
ities, the pressure difference across any meniscus surface,
independence of asperity sizes and shapes, can be deter-
mined by the Laplace equation [49]
 
1 1
∆p = γ + , (59) (a)
r1 r2
where ∆p is the pressure difference between the pressures
across the liquid and air interface and is commonly referred
to as the Laplace pressure (or capillary pressure) in the
liquid; and 1/r1 and 1/r2 are the meniscus curvatures along
the mutually orthogonal planes. If |r1 |  |r2 | (note that this
condition is always satisfied for contacting asperities at the
contact interface of rough surfaces where asperity height is
several orders of magnitude smaller than asperity radius),
then equation (59) becomes

∆p = γ/r1 . (60)
(b)
The meniscus force (normal force) due to the pressure dif- Figure 38. (a) A schematic diagram of a rough surface in contact with a
ference is then given by smooth surface with a liquid film, and (b) a schematic diagram of contact
ZZ ZZ area, wetted area and cross cut area at mean meniscus height for meniscus
1 bridges around contacting and noncontacting asperities [56].
Fm = ∆p(x, y) dΩ = γ dΩ, (61)
r1
Ω Ω
where s is the mean meniscus height for an irregular as-
where Ω is the projected area of meniscus enclave which in- perity. We note that, for rough surfaces, Ω is a function
tersects the upper contacting asperity at meniscus height, s. of mean meniscus height. For a typical spherical con-
For multiple isolated menisci scattered over the whole con- tacting asperity, a simple relation exists between Ω and
tact interface, Ω should be the sum of the projected area of s (Ω ≈ 2πRs) [50], and equation (62) reduces to the classi-
each meniscus enclave. cal expression. However, for randomly distributed menisci
To solve equation (61), we need to know both the menis- around irregular contacting asperities, no relationship be-
cus radius at different locations and the projected area of the tween Ω and s is available.
meniscus enclave. These two parameters are in turn func- Figure 38 shows some concepts for selecting the pro-
tions of the shape (which determines meniscus radius) and jected meniscus area. (In figure 38(a), for convenience, we
the volume of the meniscus. The determination of the exact put the thin film of liquid on the smooth surface instead
shape and volume of the meniscus in the thin film situation of on the rough surface. This is because the two situations
shown in figure 36 is not tractable. Tian and Bhushan [56] are numerically equivalent.) Menisci will form around the
adopted a relatively simplified approach as follows. If the contacting and the noncontacting asperities which touch the
effective section of a meniscus is approximated as a section liquid. We refer to the areas where asperities touch the liq-
of arc (which is true if both gravitation and van der Waals uid as wetted areas (as shown in figure 38(a)). To determine
forces are negligible), then the relationship between the ef- the projected meniscus area, Ω(s), we first obtain the cross
fective radius of the meniscus and the meniscus height is cut area of the rough surface at a given mean meniscus
given by height, s. The cross cut area at the mean meniscus height
s is then grouped into individual area islands. There is no
r≈ , (62) geometrical link among individual islands. Those islands
1 + cos θ
which overlap the wetted area are selected as the projected
where r is the effective radius of the meniscus. If we meniscus area, Ω(s)i , and those islands which do not over-
further assume that the variation of meniscus height around lap the wetted area are discarded. The aim of this step is to
an irregular asperity is small (this is probably the case), eliminate those areas which do not touch the liquid; there-
then the meniscus force in equation (61) is reduced to fore, no meniscus forms around nonwetted asperities. The
ZZ ZZ total projected meniscus area is equal to
1 1
Fm = γ dΩ ≈ γ(1 + cos θ) dΩ X X

Ω
r s
Ω
Am = Ω(s)i = A∗i
c − Ar ,
i
(64)
i i
Ω
≈ γ(1 + cos θ) , (63) where Am is the total projected meniscus area, A∗i
s c is the
 B. Bhushan / Contact mechanics of rough surfaces 31

(a)
Figure 39. Contact area and meniscus area for the case of computer
generated rough surface (σ = 1 nm, β ∗ = 0.5 µm) in contact with a
smooth, flat surface with a composite elastic modulus of 100 GPa and
with a water film (γ = 73 dynes/cm, θ = 60◦ ) thickness of 1 nm and
meniscus height of 1 nm [97].

area of the ith island which contains wetted asperities from

the cross cut area at the mean meniscus height, Air is the
real area of contact located inside the ith island. Finally,
the meniscus force is given by equation (63) replacing Ω
by Am ,
γ X

Fm ≈ (1 + cos θ) A∗i
c − Ar .
i
(65)
s i

Tian and Bhushan [56] developed a computer program to

select the individual meniscus island and to compute the
total projected meniscus area according to the method just
described. Two parameters, meniscus height and liquid film
thickness, are required for the analyses, which can be ob-
tained for a given relative humidity. Water film thickness
for a given relative humidity can be measured. For capil- (b)
lary condensation from a humid environment, the meniscus Figure 40. (a) The effect of relative humidity on the relative meniscus
radius at equilibrium, the so-called Kelvin radius rK , is re- force for a glass-ceramic disk substrate in contact with a smooth sur-
lated to relative humidity (relative vapor pressure), by the face [56]. (b) The effect of water film thickness and surface roughness
Kelvin equation [49], on the relative meniscus force for computer generated gaussian surfaces
(correlation distance β ∗ = 0.5 µm) in contact with a smooth surface. The
γV dotted line define the critical film thickness for different σ [97].
rK = , (66)
RT log RH
is first analyzed. In the next step, a liquid film of known
where RH is the relative humidity in fraction, V is the mo- mean thickness is introduced over the deformed rough sur-
lar volume, R is the gas constant and T is the absolute faces. Wetted areas are determined by selecting the areas
temperature. γV /(RT ) = 0.54 nm for water at 20 ◦ C. For where asperities of both contacting surfaces touch the liq-
RH = 1, rK = ∞. For a given meniscus radius, equa- uid. The total projected meniscus area is determined by
tion (62) is then used to calculate the meniscus height. The selecting those area islands of cross cut area at a given
adsorbed water film thickness for a given relative humidity mean meniscus height that overlap the wetted area. The
can be obtained from an empirical relation [102], meniscus force is then calculated using equation (65). Fig-
  ure 39 shows representative contact area and meniscus area
h = h1 (RH) + h2 exp α(RH − 1) , (67)
maps for a computer generated rough surface in contact
where h1 = 0.3 nm, h2 = 0.5 nm, α = 20 and RH is rela- with a smooth surface in the presence of water film. As
tive humidity fraction ranging from 0 to 1 for carbon-coated expected, the meniscus area is larger than the contact area
disk surface. For a given liquid film thickness, meniscus and meniscus force is three times that of the normal force.
height is a function of liquid film thickness and needs to Tian and Bhushan [56] and Poon and Bhushan [97]
be estimated. used the model to predict the effects of relative humid-
In the numerical wet model developed by Tian and ity and liquid film thickness on the meniscus force (also
Bhushan [56], elastic–plastic dry contact of rough surfaces see Bhushan [51]). Effect of relative humidity on an inter-
32 B. Bhushan / Contact mechanics of rough surfaces

(a)

(a)

(b)
Figure 41. Probability density functions for surfaces with (a) different
skewness and (b) different kurtosis values [57].

face is shown in figure 40(a). Effect of liquid film thick-

ness and interface roughness on computer generated rough
surfaces in contact with a smooth surface is shown in fig-
ure 40(b). An increase in either relative humidity or liquid
film thickness increases the liquid present at the interface.
The thicker a liquid film, the more asperities touch the liq- (b)
uid surface and the menisci form on a larger number of Figure 42. (a) Fractional real area of contact and relative meniscus force
asperities. In addition, with a thicker film, larger volume as a function of skewness and kurtosis at various nominal pressures, and
of liquid is present around the asperities resulting in greater (b) relative meniscus force as a function of h/σ for different skewness
amount of meniscus volume accumulated at the contact in- and kurtosis values, for an interface in the presence of perfluoropolyether
film (γ = 25 dynes/cm, θ = 10◦ ) [57].
terface and greater meniscus height. These effects lead to
larger meniscus forces. There is a critical film thickness for and 0.2 at low pressure and about 0.2 at higher pressures
a surface with given roughness, above which the meniscus results in lowest real area of contact and meniscus force.
force increases rapidly. The critical film thickness is on the Contact area and meniscus force decrease with an increase
order of 3/4 of the liquid film thickness. in the kurtosis. Fewer peaks present on a surface with
Bhushan and Chilamakuri [57,58] used the numerical positive skewness or high kurtosis can explain the trends.
model to study contact between a flat rigid surface and Figure 42(b) shows the variation of relative meniscus force
a nongaussian rough surface with skewness and kurtosis. with h/σ ratio for different skewness and kurtosis values.
Nongaussian surfaces may exhibit skewness (Sk) and kur- Note that the sensitivity of h/σ to the meniscus force de-
tosis (K). Rough surfaces with different skewness and kur- creases at a range of positive skewness of 0 to 0.2, and
tosis were generated using the 2-D digital filter technique, kurtosis of about five or larger is optimum.
referenced earlier. Figure 41 shows the probability density
functions of surfaces with various skewness and kurtosis
values. Figure 42(a) shows the effect of skewness and kur- 5. Concluding remarks
tosis on the fractional real area of contact (Ar /Aa , where Aa
is the apparent area) and relative meniscus force (Fm /W ) at When two nominally flat surfaces are placed in contact,
different nominal pressures. A positive skewness between 0 surface roughness causes contact to occur at discrete con-
 B. Bhushan / Contact mechanics of rough surfaces 33

tact spots. The sum of the areas of all the contact spots [3] B. Bhushan, Tribology and Mechanics of Magnetic Storage De-
constitutes the real (true) area of contact. Deformation oc- vices, 2nd Ed. (Springer, New York, 1996).
curs in the region of the contact spots, establishing stresses [4] A. Majumdar and B. Bhushan, Role of fractal geometry in rough-
ness characterization and contact mechanics of rough surfaces,
that oppose the applied load. Relative sliding introduces a ASME Journal of Tribology 112 (1990) 205–216.
tangential force (referred to as friction force) at the con- [5] S. Ganti and B. Bhushan, Generalized fractal analysis and its ap-
tact interface. The mode of surface deformation is either plications to engineering surfaces, Wear 180 (1995) 17–34.
elastic or elastic–plastic. Operation of the interface at high [6] B. Bhushan, Handbook of Micro/Nanotribology (CRC Press, Boca
relative humidities introduces a water film at the interface. Raton, FL, 1995).
[7] T.R. Thomas, Rough Surfaces (Longman, London, 1982).
In addition, a liquid film may be deliberately applied at the [8] D.J. Whitehouse, Handbook of Surface Metrology (Institute of
interface to reduce friction and wear. A liquid film at the Physics Publishing, Bristol, 1994).
interface results in attractive meniscus forces that can also [9] B. Bhushan and G.S. Blackman, Atomic force microscopy of mag-
increase friction and wear. netic rigid disks and sliders and its applications to tribology, ASME
A rough surface is a random system and stochastic mod- Journal of Tribology 113 (1991) 452–457.
[10] J.A. Greenwood and J.B.P. Williamson, Contact of nominally flat
els are used to characterize and model the contact of two surfaces, Proc. Roy. Soc. Lond. A 295 (1966) 300–319.
rough surfaces. The classical statistical model for a combi- [11] D.J. Whitehouse and J.F. Archard, The properties of random surface
nation of elastic and elastic–plastic contacts between rough of significance in their contact, Proc. Roy. Soc. Lond. A 316 (1970)
surfaces is that of Greenwood and Williamson [10] which 97–121.
uses significant and unrealistic assumptions. Nevertheless [12] R.A. Onions and J.F. Archard, The contact of surfaces having a
random structure, J. Phys. D.: Appl. Phys. 6 (1973) 289–304.
the model allows one to identify important roughness para- [13] P.R. Nayak, Random process model of rough surfaces in plastic
meters and mechanical properties that govern contact me- contact, Wear 26 (1973) 305–333.
chanics. The basic model has been the subject of much [14] A.W. Bush, R.D. Gibson and T.R. Thomas, The elastic contact of
research in the last three decades and many refinements a rough surface, Wear 35 (1975) 87–111.
(rather insignificant) have been proposed. The statistical [15] A.W. Bush, R.D. Gibson and G.P. Keogh, Strongly aniostropic
rough surfaces, ASME Journal of Lubrication Technology 101
models have been extended to take into account the effect (1979) 15–20.
of a liquid film present at the interface in wet contacts [3]. [16] R.D. Gibson, The surface as a random process, Rough Surfaces,
With the advent of computer technology, numerical mod- ed. T.R. Thomas (Longman, London, 1982).
els have been developed, particularly in the past decade, [17] J.I. McCool, Predicting microfracture in ceramics via a microcon-
that can analyze the contact of two rough surfaces with mea- tact model, ASME Journal of Tribology 108 (1986) 380–386.
[18] H.A. Francis, Application of spherical indentation mechanics to
sured 3-D roughness maps and mechanical properties. An reversible and irreversible contact between rough surfaces, Wear
important numerical model was developed by Sayles and 45 (1977) 221–269.
co-workers (e.g., [35]), which uses an iterative process with [19] B. Bhushan, Analysis of the real area of contact between a poly-
a conventional matrix inversion technique to solve the con- meric magnetic medium and a rigid surface, ASME Journal of
tact problem for contact pressure and contact area. To han- Tribology 106 (1984) 26–34.
[20] J.I. McCool, Comparison of models for the contact of rough sur-
dle a large number of contact points, Tian and Bhushan [38] faces, Wear 107 (1986) 37–60.
developed a numerical model based on the variational prin- [21] W.R. Chang, I. Etsion and D.B. Bogy, An elastic–plastic model
ciple. This model is very powerful as it can handle a large for the contact of rough surfaces, ASME Journal of Tribology 109
number of contact points (tens of thousands). A first nu- (1987) 257–263.
merical contact model for wet interface has been developed [22] C.A. Kotwal and B. Bhushan, Contact analysis of nongaussian sur-
faces for minimum static and kinetic friction and wear, Tribol.
by Tian and Bhushan [56]. The effect of friction can rather Trans. 39 (1996) 890–898.
easily be included in these models. Multilayered structures [23] A. Majumdar and B. Bhushan, Fractal model of elastic–plastic
in 3-D models have not yet been analyzed. Sliding and contact between rough surfaces, ASME Journal of Tribology 113
failure criteria have not been used much in the numerical (1991) 1–11.
models to predict interface failure. [24] B. Bhushan and A. Majumdar, Elastic–plastic contact model of
bifractal surfaces, Wear 153 (1992) 53–64.
These models have been used to study the effect of [25] J.B.P. Williamson, The microtopography of surfaces, Proc. Instn.
roughness distribution and mechanical properties on the Mech. Engrs. 182 Part 3k (1967/68) 21–30.
real area of contact, surface and subsurface stresses and [26] P.K. Gupta and N.H. Cook, Junction deformation models for asper-
meniscus forces (in the case of wet contacts). The nu- ities in sliding interaction, Wear 20 (1972) 73–108.
merical models are now extensively used to design texture [27] B. Bhushan and N.H. Cook, On the correlation between friction
coefficients and adhesion stresses, ASME J. Eng. and Mat. Technol.
for magnetic disks and sliders in the computer data storage 97 (1975) 285–287.
industry [3,51]. [28] R.S. Sayles and T.R. Thomas, Computer simulation of the contact-
ing rough surfaces, Wear 49 (1978) 273–296.
[29] C.Y. Poon and B. Bhushan, Comparison of surface roughness mea-
References surements by stylus profiler, AFM and non-contact optical profiler,
Wear 190 (1995) 76–88.
[1] F.P. Bowden and D. Tabor, The Friction and Lubrication of Solids, [30] F.A. Akyuz and M. Merwin, Solution of nonlinear problems of
Part I (Clarendon Press, Oxford, 1950). elastoplasticity by finite element method, AIAA J. 6 (1968) 1825–
[2] F.P. Bowden and D. Tabor, The Friction and Lubrication of Solids, 1831.
Part II (Clarendon Press, Oxford, 1964). [31] K. Komvopoulos and D.H. Choi, Elastic finite element analysis
34 B. Bhushan / Contact mechanics of rough surfaces

of multi-asperity contact, ASME Journal of Tribology 114 (1992) [55] B. Bhushan, C.A. Kotwal and S.K. Chilamakuri, Kinetic meniscus
823–831. model for prediction of rest stiction, ASME Journal of Tribology
[32] H.A. Francis, The accuracy of plane-strain models for the elastic 119 (1997) (in press).
contact of three-dimensional rough surfaces, Wear 85 (1983) 239– [56] X. Tian and B. Bhushan, The micro-meniscus effect of a thin liquid
256. film on the static friction of rough surface contact, J. Phys. D: Appl.
[33] W.T. Lai and H.S. Cheng, Computer simulation of elastic rough Phys. 29 (1996) 163–178.
contacts, ASLE Trans. 28 (1985) 172–180. [57] B. Bhushan and S. Chilamakuri, Nongaussian surface rough-
[34] M.N. Webster and R.S. Sayles, A numerical model for the elas- ness distribution of magnetic media for minimum friction/stiction,
tic frictionless contact of real rough surfaces, ASME Journal of J. Appl. Phys. 79 (1996) 5794–5996.
Tribology 108 (1986) 314–320. [58] S.K. Chilamakuri and B. Bhushan, Contact analysis of nongaussian
[35] M.A. West and R.S. Sayles, A 3-dimensional method of studying 3- random surfaces, Proc. Instn. Mech. Engrs., Part J: J. Eng. Tribol.
body contact geometry and stress on real rough surfaces, in: Proc. (1997) (in press).
14th Leeds-Lyon Symposium on Tribology, Tribology Series 12 (El- [59] B. Bhushan, Contact mechanics of rough surfaces in tribology:
sevier, Amsterdam, 1987) pp. 195–200. single asperity contact, Appl. Mech. Rev. 49 (1996) 275–298.
[36] A.A. Lubrecht and E. Ioannides, A fast solution of the dry con- [60] G.M. Hamilton and L.E. Goodman, Stress field created by a circular
tact problem and associated subsurface stress field using multilevel sliding contact, ASME Journal of Applied Mechanics 33 (1966)
techniques, ASME Journal of Tribology 113 (1991) 128–133. 371–376.
[37] N. Ren and S.C. Lee, Contact simulation of three-dimensional [61] T.C. O’Sullivan and R.B. King, Sliding contact stress field due to
rough surfaces using moving grid method, ASME Journal of Tri- a spherical indenter on a layered elastic half-space, ASME Journal
bology 115 (1993) 597–601. of Tribology 110 (1988) 235–240.
[38] X. Tian and B. Bhushan, A numerical three-dimensional model [62] B. Lawn, Fracture of Brittle Solids (Cambridge Univ. Press, Cam-
for the contact of rough surfaces by variational principle, ASME bridge, 1993).
Journal of Tribology 118 (1996) 33–42. [63] E.R. Kral, K. Komvopoulos and D.B. Bogy, Elastic–plastic finite
[39] C.Y. Poon and R.S. Sayles, Numerical contact model of a smooth element analysis of repeated indentation of a half-space by a rigid
ball on an anisotropic rough surface, ASME Journal of Tribology sphere, ASME Journal of Applied Mechanics 60 (1993) 829–841.
116 (1994) 194–201. [64] E.R. Kral, K. Komvopoulos and D.B. Bogy, Finite element analysis
[40] T.F. Conry and A. Seireg, A mathematical programming method of repeated indentation of an elastic–plastic layered medium by
for design of elastic bodies in contact, ASME Journal of Applied a rigid sphere, Part I: surface results, ASME Journal of Applied
Mechanics 38 (1971) 387–392. Mechanics 62 (1995) 20–28.
[41] J.J. Kalker and Y. van Randen, A minimum principle for frictionless [65] E.R. Kral, K. Komvopoulos and D.B. Bogy, Finite element analysis
elastic contact with application to non-hertzian half-space contact of repeated indentation of an elastic–plastic layered medium by a
problems, J. Eng. Math. 6 (1972) 193–206. rigid sphere, Part II: subsurface results, ASME Journal of Applied
[42] D.M. Bailey and R.S. Sayles, Effect of roughness and sliding fric- Mechanics 62 (1995) 29–42.
tion on contact stresses, ASME Journal of Tribology 113 (1991) [66] D. Tabor, The Hardness of Metals (Clarendon Press, Oxford, 1951).
729–738. [67] K.L. Johnson, Contact Mechanics (Cambridge University Press,
[43] M.H. Yu and B. Bhushan, Contact analysis of three-dimensional Cambridge, 1985).
rough surfaces under frictionless and frictional contact, Wear 200 [68] B. Bhushan and X. Tian, Contact analysis of regular patterned rough
(1996) 265–280. surfaces in magnetic recording, ASME J. Electronic Packaging 117
[44] T. Merriman and J. Kannel, Analyses of the role of surface rough- (1995) 26–33.
ness on contact stresses between elastic cylinders with and without [69] J.B.P. Williamson, J. Pullen and R.T. Hunt, The shape of solid
soft surface coating, ASME Journal of Tribology 111 (1989) 87–94. surfaces, Surface Mechanics (ASME, New York, 1970) p. 24.
[45] S.J. Cole and R.S. Sayles, A numerical model for the contact of [70] J.A. Greenwood and J.H. Tripp, The contact of two nominally flat
layered elastic bodies with real rough surfaces, ASME Journal of rough surfaces, Proc. Instn. Mech. Engrs. 185 (1970–71) 625–633.
Tribology 114 (1992) 334–340. [71] M. O’Callaghan and M.A. Cameron, Static contact under load be-
[46] J.T. Oden and J.A.C. Martins, Models and computational meth- tween nominally flat surfaces in which deformation is purely elastic,
ods for dynamic friction phenomana, Computer Methods in Appl. Wear 36 (1976) 76–97.
Mech. and Eng. 52 (1985) 527–634. [72] I.N. Gibra, Probability and Statistical Inference for Scientists and
[47] J.A.C. Martins, J.T. Oden and F.M.F. Simoes, A study of static and Engineers (Prentice-Hall, Englewood Cliffs, NJ, 1973).
kinetic friction, Int. J. Engr. Sci. 28 (1990) 29–92. [73] M. Abramowitz and I.A. Stegun, Handbook of Mathematical Func-
[48] B.V. Derjaguin, N.V. Chuarev and V.M. Muller, Surfaces Forces tions (Dover, NY, 1965) pp. 685–720.
(Consultants Bureau, New York, 1987). [74] P.R. Nayak, Random process model of rough surfaces, ASME
[49] A.W. Adamson, Physical Chemistry of Surfaces (Wiley, New York, J. Lub. Tech. 93 (1971) 398–407.
1990). [75] M.S. Longuet-Higgins, The statistical analysis of a random, moving
[50] J.N. Israelachvili, Intermolecular and Surface Forces, 2nd Ed. surface, Phil. Trans. R. Soc. Lond. A 249 (1957) 321–387.
(Academic Press, San Diego, CA, 1992). [76] M.S. Longuet-Higgins, Statistical properties of an isotropic random
[51] B. Bhushan, Methodology for roughness measurement and contact surface, Phil. Mag. R. Soc. Lond. A 250 (1957) 157–174.
analysis for optimization of interface roughness, IEEE Trans. Magn. [77] K.N.G. Fuller and D. Tabor, The effect of surface roughness on
32 (1996) 1819–1825. the adhesion of elastic solids, Proc. Roy. Soc. Lond. A 345 (1975)
[52] Y. Li and F.E. Talke, A model for the effect of humidity on stiction 327–342.
of the head/disk interface, in: Tribology and Mechanics of Magnetic [78] K.L. Johnson, K. Kendall and A.D. Roberts, Surface energy and
Storage Systems, ed. B. Bhushan (SP-29, STLE, Park Ridge, Ill, the contact of elastic solids, Proc. Roy. Soc. Lond. A 324 (1971)
1990) pp. 79–84. 301–313.
[53] H. Tian and T. Matsudaira, The role of relative humidity, surface [79] D.J. Whitehouse and M.J. Phillips, Discrete properties of random
roughness and liquid build-up on static friction behavior of the surfaces, Proc. Roy. Soc. Lond. A 290 (1978) 267–298.
head/disk interface, ASME Journal of Tribology 115 (1993) 28–35. [80] C.Y. Poon and B. Bhushan, Surface roughness analysis of glass-
[54] C. Gao, X. Tian and B. Bhushan, A meniscus model for optimiza- ceramic substrates and finished magnetic disks, and Ni–P coated
tion of texturing and liquid lubrication of magnetic thin film rigid Al–Mg and glass substrates, Wear 190 (1995) 89–109.
disks, Tribol. Trans. 38 (1995) 201–212. [81] B. Bhushan and M.F. Doerner, Role of mechanical properties and
 B. Bhushan / Contact mechanics of rough surfaces 35

surface texture in the real area of contact of magnetic rigid disks, [93] N. Ren and S.C. Lee, The effects of surface roughness and topog-
ASME Journal of Tribology 111 (1989) 452–458. raphy on the contact behavior of elastic bodies, ASME Journal of
[82] S.K. Chilamakuri and B. Bhushan, Optimization of asperities for Tribology 116 (1994) 804–810.
laser-textured magnetic disk surfaces, Tribol. Trans. 40 (1997) 303– [94] C.Y. Poon and B. Bhushan, Rough surface contact analysis and its
311. relation to plastic deformation at the head disk interface, J. Appl.
[83] J. Boussinesq, Application des Potentials a l’Etude de l’Equilibre Phys. 79 (1996) 5799–5801.
et du Mouvement des Solides Elastiques (Gauthier-Villars, Paris, [95] C.Y. Poon and B. Bhushan, Nano-asperity contact analysis and sur-
France (1885). face optimization for magnetic head slider/disk contact, Wear 202
[84] V. Cerruti, Roma Acc. Lincei, Mem. fis. mat. (1882). (1996) 83–98.
[85] A.E.H. Love, The stress produced in a semi-infinite solid by pres- [96] Y.Z. Hu and K. Tonder, Simulation of 3-D random surface by 2-D
sure on part of the boundary, Proc. Roy. Soc. Lond. A 228 (1929) digital filter and Fourier analysis, Int. J. of Mach. Tool Manufact.
377–420. 32 (1992) 82–90.
[86] T.H. Richards, Energy Methods in Stress Analysis (Ellis Horwood, [97] C.Y. Poon and B. Bhushan, Numerical contact and stiction analyses
Chichester, 1977). of gaussian isotropic surfaces for magnetic head slider/disk contact,
[87] P. Wolfe, The simplex method for quadratic programming, Econo- Wear 202 (1996) 68–82.
metrica 27 (1959) 382–395. [98] D.P. Hess and A. Soom, Normal and angular motions at a rough
[88] S.C. Lee and N. Ren, Behavior of elastic–plastic rough surface con- planar contact during sliding with friction, ASME Journal of Tri-
tacts as affected by surface topography, load and material hardness, bology 114 (1992) 567–578.
Tribol. Trans. 39 (1996) 67–74. [99] A.A. Polycarpou and A. Soom, A two-dimensional mixed friction
[89] S.C. Lee and N. Ren, The subsurface stress field created by three- model for a lubricated line contact, ASME Journal of Tribology
dimensionally rough bodies in contact with traction, Tribol. Trans. 118 (1996) 183–189.
37 (1994) 615–621. [100] N. Kikuchi and J.T. Oden, Contact Problems in Elasticity (SIAM,
[90] N. Ahmadi, L.M. Keer, T. Mura and V. Vithoontien, The interior Philadelphia, 1985).
stress field caused by tangential loading of a rectangular patch on [101] P. Rabier, J.A.C. Martins and J.T. Oden, Existence and local unique-
an elastic half space, ASME Journal of Tribology 109 (1987) 627– ness of solutions to contact problems in elasticity with nonlinear
629. friction laws, Int. J. Engng. Sci. 24 (1986) 1755–1768.
[91] J.J. Kalker, Numerical calculation of the elatic field in a half-space, [102] B. Bhushan and Z. Zhao, Friction/stiction and wear studies of mag-
Communications in Appl. Numerical Method 2 (1986) 401–410. netic thin-film disks with two perfluoropolyether lubricants, IEEE
[92] C.Y. Poon and R.S. Sayles, Contact analysis of a smooth ball on an Trans. Magn. 31 (1997) 918–925.
anisotropic rough surface, ASME Journal of Tribology 116 (1994)
850–859.