You are on page 1of 17

Proceedings of the Institution of Mechanical Engineers http://pme.sagepub.

com/

The Contact of Two Nominally Flat Rough Surfaces


J. A. Greenwood and J. H. Tripp Proceedings of the Institution of Mechanical Engineers 1970 185: 625 DOI: 10.1243/PIME_PROC_1970_185_069_02 The online version of this article can be found at: http://pme.sagepub.com/content/185/1/625

Published by:
http://www.sagepublications.com

On behalf of:

Institution of Mechanical Engineers

Additional services and information for Proceedings of the Institution of Mechanical Engineers can be found at: Email Alerts: http://pme.sagepub.com/cgi/alerts Subscriptions: http://pme.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav Citations: http://pme.sagepub.com/content/185/1/625.refs.html

>> Version of Record - Jun 1, 1970 What is This?

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

625

TRI BOLOGY GROUP

THE CONTACT OF TWO NOMINALLY FLAT ROUGH SURFACES


J. A. Greenwood, PhD*

J. H, Tripp, P h D t

Most models of surface contact consider the surface roughness to be on one of the contacting surfaces only. The authors give a general theory of contact between two rough plane surfaces. They show that the important results of the previous models are unaffected: in particular, the load and the area of contact remain almost proportional, independently of the detailed mechanical and geometrical properties of the asperities. Further, a single-rough-surface model can always be found which will predict the same laws as a given two-rough-surface model, although the required model may be unrealistic. I t does not seem possible to deduce the asperity shape or deformation mode from the load-compliance curve.

INTRODUCTION

THE STUDY of friction, wear and thermal and electrical contact between solids has relied heavily on the idea that the real area of contact is extremely small. A newcomer to the subject might think it a truism that to understand these contact areas one must first understand the topography of the individual surfaces but, in practice, the classic texts by Holm (I)* and by Bowden and Tabor (2) use only the qualitative property that surfaces are rough. The reason is, of course, that, in the specific case of ideal plastic flow at the contact spots, one can calculate the total area of contact without a knowledge of the surface topography. It is easy to show that contact between surfaces is rarely perfectly, or even predominantly, elastic. It does not, of course, follow that ideal plastic flow is the rule. If the asperities on the surfaces have rounded summits, then individual contacts will be analogous to contacts between a ball and a plane, as in a Brine11 hardness test; but in this case (3) the load has to be increased about 200-fold to take the contact from the elastic limit to the fully plastic state. Even more important is the idea, introduced by Archard (4), that plastic flow involves permanent deformation and so cannot be repeated indefinitely; the contact geometry will gradually change until elastic contact becomes possible. In addition, when we wish to predict sliding temperatures (9, thermal or electrical contact or resistance (6)(7), a knowledge of the total area of contact
This paper is p a i s h e d for written discussion. The M S . was received on 3rd October 1968 and accepted for publication on 30th July 1970. 23 * The University of Camhidge, University Engineering Department, Trumpington Street, Cambridge, C B 2 1PZ. t Department of Physics, University of Connecticut. References are given in Appendix 4.
Proc lnstn Mech Engrs 1970-71

is inadequate; we need to know something of the number and sizes of individual contacts. However, the surfaces of sliding solids are as complex as the surface of the earth and, indeed, geometrically rather similar to it. This appears to rule out any exact treatment of contact and instead we have to rely on models which, it is hoped, represent the essential features of the surface roughness. Fortunately, there are properties which are reasonably independent of the details of the model so that its fidelity is not of the first importance. Interest in models has been enhanced by the recent introduction of quantitative measurements of surface topography obtained by analysis of the output of a stylus profilometer. The analogue system used by Tallian et al. (8) has many advantages but digitizing the signal and recording it on paper-tape for subsequent analysis on a digital computer has proved more popular (~)-(II). The one simplifying feature of surface roughness so far found is that, although the height of a particular asperity is random, the distribution of the asperity heights is rather close to Gaussian especially for ground or gritblasted surfaces (8)-(13). Perhaps more important, the height distribution on a worn surface can be as shown in Fig. 1. This would usually be regarded as non-Gaussian but in contact the effect is Gaussian since the height of parts of the surface which do not touch is irrelevant. It is not yet known how cQmmon such behaviour is because systems for acquiring data of this sort readily are fairly recent, but other examples have already been found (13) (14). If we can add to this characteristic distribution of
O r normal but since this term has lost the meaning ideal and acquired the meaning usual) prefer to avoid it. we
VOI 1a5 48/71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

626

J. A. GREENWOOD AND J. H.TRIPP

x
xx

99
e

t
X

.
X
0

95 c
OI v
(Y L

80X
X .

s ?
a

& 50w
X

X.

contact between two such surfaces. The pairs of asperities will not be aligned so that the usual contact will be between the shoulders of the two hills. Let the distribution of asperity heights be +(z), and let the shape of each asperity with respect to its summit be y = f ( p ) , where p and y are the horizontal and vertical distances from the summit. Consider a particular asperity on the first surface, with height z1and all the asperities of height z, on the second surface; a typical one will be a horizontal distance r away (see Fig. 2). Since the asperities are identical, any contact must lie symmetrically between their centres and, if they are convex, the contact will be centred a distance r/2 from the individual centres, at which point the heights above the respective mean planes will be
z -f(r/2) ; z -f(r/2) 1 2 For contact, the separation d of the mean planes must be less than the sum of these heights, i.e.

x
y x 09

O1-x

XI.

x
I

asperity heights a characteristic asperity shape, we would have a model surface which we could use to predict some properties of real surfaces. Archard (10) has recently suggested that surface features of different scales may be essential in a realistic model. However, the object of this paper is to attempt to settle a particular controversy which seems unlikely to depend on such questions, i.e. whether the contact problem can be simplified by assuming only one of the contacting surfaces to be rough or whether treating both surfaces as rough introduces new features. While the only other work in which both surfaces were rough (15)gave results for one specific case, we shall treat several cases and present some general conclusions. It must be emphasized that this is a purely theoretical paper; the analysis is based on measured properties of surfaces but the predictions of the theory have not been verified experimentally.
GENERALTHEORY

d < z1+z2-2f(r/2) and we may define the interference at the contact as w = zl+z2-2f(r/2)-d During loading, the area of contact A and the vertical force P will depend only on the interference w and the misalignment r, i.e. A = A(w, r ) ; P = P(w, r ) (During unloading they may involve also the maximum interference achieved : we shall not consider this case.) The number of asperities with heights in the range z, to z,+dz2 situated between r and r+dr from the first asperity will be 7 1 . 2 dr+(z,) dz, ~ where is the surface density of asperity peaks on either surface, and so the expected value of the force on the first asperity due to the second surface will be

I*, s,

2 d Y w , r)+(zz)r dr dz2

where the r integration will, in principle, be limited by the extent of the surfaces but in practice may be taken from 0 to 00. (While all the asperities on the second surface are considered in the integral, very few make any contribution since w will be negative and P(w, r ) will vanish, but see Appendix 1.)

A rough surface is represented by an array of identical asperities differing only in their heights above a reference plane, which is conveniently taken to pass through the mean of the height distribution. We shall consider the
Proc lnstn Mech Engrs 1970-71

R_efeerenC_eplane I

dli

- - - - - --

Fig. 2. Contact of t w o rough surfaces. Since all asperities are assumed t o have the same shape, contact will first occur midway between their centres
V d 185 48/71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

THE CONTACT OF TWO NOMINALLY FLAT ROUGH SURFACES

627

If the apparent area of contact is a2 there will be d ~ + ( z ,dz, asperities on the first surface with heights ) between z1 and zl+dzl. Thus, the expected total force between the surfaces will be

of rough surfaces covered with cones has no plausible equivalent (although suitable cusp-like asperities could be invented).
SPECIFIC MODELS

p ( 4 = 2 T 2 d J1

J-, P(w, r>#(z1)4(zz)r dr dz1 dzz s,

The force between a pair of asperities with given heights does not depend on the individual heights z1and z2 but only on their sum (z, +z,), and therefore only the statistical distribution of the sum of pairs of heights can affect p ( d ) . The distribution of the sum can be found from the individual distributions + z and +(z2) by the standard (J processes of mathematical statistics. This may be simple, or not, but for our purposes it is convenient to ignore this step and start from the distribution of the sums of heights as the basic distribution. Thus, we set z = z1+z2, and assume it has a distribution #o(z).The expected total force is then

We assume first that the surfaces are covered with paraboloidal asperities (which in the usual Hertzian approximation includes spheres and radially symmetrical ellipsoids as well). The asperity shape is then y = f ( p ) = p2/2/3 where ,!3 is the radius of curvature at the peak. Thus 2f(r/2) = r2/4/3. We consider two modes of deformation which we shall call elastic and plastic: there seems no justification for the mathematical complexities which more exact solutions would involve.

p(d) = 2 7 r ~ ~ d J P(w, ~ ) + ~ ( z ) r dr dz
I T

(1)

Now, w = z - d - 2 f ( r / 2 ) = wp-2f(r/2) where w, = zl+z,-d is the overlap between the peaks of the pair of asperities considered : because of misalignment the peaks may overlap without there being any contact. If we define the result of the r-integration as

Elastic deformation The Hertzian solution for the contact of two paraboloids applies here only approximately because of the misalignment, which means that the normal force is not acting vertically and there must also be a tangential component. However, the slopes on real surfaces are so small that the errors will be small and we shall ignore them. We assume the asperity deformation laws to be exactly

Po(w,)= 27r7 Jo- P(wp-22f(r/2),r)r dr


then

(2)

where the term 2/8 i s - h e sum of the asperity curvatures and E is the usual camposite elastic modulus, given by 1 E
-=1--v12

El

+- E,
JOm

1--v22

In the same way, if we write

A,(w,) = 2 ~ JOm A(wp-2f (r/2),r)r dr 7


the expected total area of contact is

(4)

Then from equation ( 2 ) the average force-displacement law is

pO(wp) =

(i) g)
VTE

(w~-$)~

r dr

If we analyse the contact between a rough surface and a plane in a similar way, taking the laws describing the deformation of a single contact to be P = Po(w),A = Ao(w), we arrive at equations (3) and (5) precisely. The difference, then, between a single rough surface and both surfaces being rough is simply (a) the replacement of the actual single-contact laws by generalized laws, found by integrating the actual deformation laws over all possible radial misalignments and (b) the replacement of the distribution of asperity heights by the distribution of sums of pairs of heights. Therefore, there is nothing essentially different about models in which both surfaces are rough. However, the behaviour may sometimes be more readily explained.ononehypothesis than on the other. For example, the contact between a spherical asperity and a plane gives A cc w; between a cone and a plane A cc w2. Higher powers seem unlikely. The effect of radial averaging (see below) is to lead to A oc wp2,A oc wp4respectively, so that a single rough surface covered with conical asperities behaves like a pair of rough surfaces covered with spheres, but a pair
Proc lnstn Mech Engrs 1970-71

where ( ), a Macauley bracket (16)~ denotes that the term is set to zero if the quantity enclosed is negative. Thus

PO(wp) (162/2/15)7r~E/i?~~ = (w,)~/ Then the total load, using equation (3), is
.

(6)

P(d)= (162/2/15)7rq2Ep32a2
and putting z
= sa

Irn
d

(~--d)~~+,(z) dz

we get

where

F,(u) = Jum (s-u)~+*(s)ds

and $*(s) is the standardized height distribution so that +*(s) ds = #(z)dz. The expression for P(d)may, in fact, be obtained more neatly without introducing Po(w,) by changing the order of integration in equation (1); but the parallel with singlerough-surface models is then lost.
Vol 185 48/71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

628

J. A. GREENWOOD AND J. H. TRIPP

Similarly, we find the area

A is given by

while the expected number of contacts is

so we have expressed P, A and N in terms of integrals of the distribution of asperity heights. We can compare these results in several ways with those for contact of a single rough surface with a plane. In the first place, we can consider an equivalent single rough surface, defined as one whose asperity curvature is the sum of the two curvatures and whose height distribution is that of the sum of the heights of the original pair. In fact, the behaviour of the equivalent surface is exactly that of the original pair with their asperities aligned (non-genuine two-rough-surface models of this type were discussed in (17))~so this comparison is a study of the effect of permitting misalignment. We can also compare the behaviour with that of one of the original rough surfaces touching a plane. We divide the force equations by the nominal area a?to get the nominal pressurep = P / d . For the first case, by substituting Po(w) = (4/3)E/3e12~32equation (3), we get in

--_-_
-.-.-

NOMINAL PRESSURE-kgf/cm2 Both surfaces rough, ? J ~ U= 0.05 (equation (11)). Equivalent surface against a plane (equation (10)). _ Single rough surface against a plane (equation (12)).

Fig. 3. Relation between load and separation

while when both surfaces are rough we obtain from equation ( 6 ) :

value 1000 kgf/cm2. The results are shown in Fig. 3. The curve for two rough surfaces differs from that for the equivalent surface against a plane only in that the separation is reduced-a not unexpected effect of permitting misalignment of the asperities. The slopes of the log(1oad)separation curves, which are all that can be detected experimentally, remain unchanged. On the other hand when one of the rough surfaces is loaded against a plane, the slope is much less-a much more significant change. If we make similar calculations for the expected area of contact as a function of separation, we can then plot the area as a function of load (Fig. 4). The points for a single rough surface and for two rough surfaces fall on the same, almost straight, line. Thus, again, we obtain the basic statistical law of contact, that the area is proportional to the load: and we can think of an elastic hardness as in (10)

At this point a specific distribution of asperity heights must be chosen. We shall use the Gaussian distribution because experimental work suggests that actual surface distributions approximate well to it. A short table of the functions F,(h) for this case is given in Appendix 2. By elementary statistics, if the height distribution on each surface is Gaussian with standard deviation ul, the distribution of sums of heights is also Gaussian, with standard deviation u = u12/2. Therefore, considering one of the same surfaces touching a plane, we find, as in equation (lo),

T o compare these three cases we need to choose a value for = 7#7u12/2. Experiments with the equipment described by Greenwood and Williamson (9) suggest that $ul is reasonably constant with a value of 0-03-0-05, so we shall assume $0 = 0.05. The value of K is less important, but to get definite scales we select the plausible
Proc lnstn Mech Engrs 1970-71

04
0

1 10 NOMINAL PRESSURE -kgf/crn2

100

Both surfaces rough.

---- Assuming a constant contact pressure equal to 0 . 3 E \/(.I&.


Fig. 4. Relation between area of contact and load
Vol 185 48/71

x Equivalent surface against a plane.

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

THE CONTACT O F TWO NOMINALLY FLAT R O U G H SURFACES

629

equal to 0.3Ef2/(o/~,) terms of the equivalent properties in for the pair of surfaces. I n terms of the parameters for a single surface, the 'elastic hardness' becomes approximately 05Ez/(01//3).

Plastic deformation It may be shown that the intersection of two equal paraboloids

is an ellipse whose horizontal projection is a circle of radius

where r is the distance between the paraboloid axes. The interference at the centre of the circle is w = ~ ~ + z ~ - d - ( r / 2 ) ~ / the radius is just (/3w)"" so 2 ~ , and the area vpw. While the actual force will be inclined to the vertical, it seems reasonable to ignore the horizontal component and take the vertical component to be proportional to the horizontal projection of the contact area, i.e. P = vj3H < w > where H is a mean pressure related to the hardness. The integration is similar to that for elastic deformation, and we find

Conical asperities Now consider conical asperities whose sides have a slope a. It can be shown that geometrically a pair of these, with axes separated by r and a peak overlap w, will intersect in an ellipse whose horizontal projection has semi-axes. +wpcot a, + l / ( W , 2 cot2 a - F ) provided w, > r cot a. The physical deformation law is unknown, so we shall take the area of contact to be this projected ellipse, and the vertical force to be the area multiplied by an effective hardness H : i.e. TH P = - - w p cot a . ( w p 2cot2 0 1 - r ~ ) ~ ' ~ 4 This may be written in terms of the interference w = w, cot a - r , but there is no advantage in doing so. Carrying -out the radial averaging process, we find TH - - w p cot a.(wp2cot2 a-r2)1/2rdr
= 3 T 2 7 H ~ p 4 a. cot4

= &T'~H cot4 a . wP4

r2

cos 0 . sin

e. cos 0 d0

Thus, again assuming a Gaussian height distribution, we find


r m 1
1

The result for aligned asperities is In order to compare this result with the previous ones, we need to harmonize the two models of surface roughness. We might argue that a paraboloidal asperity of height h and radius of curvature /3 is 'equivalent' to a cone inscribed in it (M. El Refaie, private communication) which leads to tan a = l/(h/28) l/(u/28). Assuming this, equation (14) becomes
N

Load-separation curves according to equations (13) and (14) are shown in Fig. 5. They &revery similar to those already given for elastic contact.

which is compared with equation (1 1) in Fig. 5.


DISCUSSION

NOMINAL PRESSURE,p/Hor p/(O3E;l(cr/p))

Elastic spheres. flastic spheres. - - - - - - Profilometric' theory, avoiding the concept of asperities ( 6 . 1) -. .-. .- Equivalent surface with plastic spheres.
Fig. 5. Independence o load-separation relation and f deformation model
Proc lnstn Mech Engrs 1970-71

---- P h i c cones.
-.-.-

It appears from Fig. 5 that, as long as the height distribution is Gaussian, the mode of deformation, asperity shape, and whether the asperities are on one or both surfaces are all unimportant. In all cases the area of real contact is almost proportional to the load and the load-compliance curve is almost straight on a log (load)-linear compliance plot, with a slope such that a movement of one standard deviation of the joint height distribution gives a load increase by a factor of about 50. This corresponds very roughly with a movement equal to the c.1.a. of each surface. The analysis requires a single type of asperity on both surfaces and, to this extent, it is unrealistic. Examination of the theory shows that the height distribution on the two surfaces need not be the same; only the joint
Vol 185 48/71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

630

J. A. GREENWOOD AND J. H. TRIPE'

height distribution is important. However, it seems unlikely that a pair of surfaces with similar asperity shapes but different height distributions would occur in practice. Of course height distributions other than the Gaussian often occur and the general theory can immediately be applied to them. For example, Zhuravlev (IS),in possibly the earliest proof that the real area of contact could be almost proportional to the load even though individual contacts behaved elastically, used a model of aligned, spherical asperities, with a linear height distribution 2 0 <z <L I.(+ = 5 @--I, The distribution of sums of heights may be shown to be 2 +,,(z)= - ( ~ L - - z ) ~ for L < z < 2L 3L4 and a more complicated expression for 0 < z < L (Fig. 6). The radially averaged law for the contact of elastic spheres has already been found (equation (6))and, combining the two, we obtain the generalization of Zhuravlev's result for non-aligned asperities : 2L 2 ( ~ - - d ) ~ / ~( ~ L - z ) ~ dz P ( d ) = -*16d2 ,12E'/33/2,d 15 d 3L4 for d > L i.e. P(d) K ( 2 ~ - d ) ~ 3 / ~

Similarly A(d)cc(2L-d)6 and so zc P"12'13. c For aligned asperities we omit the radial averaging and obtain 2 2 dz P ( d ) = -.d 2 ~ E ' / 3 ' I 2 o l . L(L~ - d ) ~ / ~ . -(-~ L - z ) ~ 3L4 3

i.e.

P"(d)cc (2L-d)11/2 A"cc P'O'" and similarly A(d)cc(2L-d)';

Thus the effect of misalignment is small, but leads to a slight improvement in the proportionality compared to Zhuravlev's result. Ling (IS) used the simple rectangular height distribution 1 -L<z<L +(z) = -, 2L The distribution of sums of heights is

(see Fig. 6b). If we combine this with a model of conical asperities deforming as in the previous section, we get 1 P ( d ) = - rr2q2dHcot4 ~ 1 C ' ( z - d ) ~ ( 2 L - z ) 6 dz (z > L ) i.e.

P"(d) = -

1 2 d H ~ ~ .(2L-d)6 CI t 4 180 7T

'

Proc lnstn Mech Engrs 1970-71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

-+

(d > L )
a Gaussian. b 'Cubic'. c Triangular. d Inverted 'cubic'.
Fig. 6. The point of first contact: whenever the height distribution falls gently t o zero, the point of first contact will be subject t o a high degree of scatter. The asperity height distributions on a single surface corresponding t o the joint height distributions are shown on the left. The zone dotted is the inter-quartile range of the highest asperity on a surface with I O O O O asperities

(16)

which is close to the previous result, although the model is very different. On the other hand, the same height distribution with spherical asperities deforming plastically gives P"cc (2L-d)4, a much lower dependence. Thus we can construct alniost any power law we like (with index > 1 ) by a suitable combination of asperity shape and distribution law. The actual deformation mode is the least important part. For example, Ling suggested that 45" conical asperities undergo a process of compression followed by the tip shearing off and the force building up again from zero. This still leads to a radially
Vol 18548171

T H E CONTACT OF TWO NOMINALLY FLAT ROUGH SURFACES

631

averaged law P o x wP4and, in fact, the constant of proportionality is rather close to the result found above for simple compression. Thus, the final result is indistinguishable from equation (16) with a = 45". Actual measurements of asperity height distributions do not support either of the above forms which, indeed, were postulated long before adequate surface data were available. They are important, however, in indicating what may be expected from a non-Gaussian distribution. In particular, approximate proportionality can be obtained between contact area and load, provided the number of asperities at a given height does not increase with height (Fig. 6c). The effect of both surfaces being rough is to increase tolerance of non-Gaussian distributions. The load-compliance curve shows more variations, but it is impossible to work back from a given load-compliance curve and deduce a mode of deformation. Ip practice, it may also be difficult to determine experimentally exactly what the load-compliance relation is. Ling (IS) found that his experimental results were of the form P (L-d)" with II = 7-8. But when a typical set of his results is replotted on log (load)-linear compliance scales (Fig. 7), they could be interpreted as resulting from a Gaussian ciistribution. The scatter of the experimental points might well mask a slight curvature: the slope of the curve implies a standard deviation of the joint height distribution equal to 10 pin, in good agreement with the quoted c.1.a. of 7 pin for each of the surfaces involved.
N

The point of first contact We have found two types of load-compliance law: a power law form in which the point of first contact is taken as origin, and an approximately logarithmic form in which the point of first contact does not appear and, indeed, does

not exist: small but non-zero loads are predicted whatever the separation. The practical experimenter may dismiss this but it indicates a very practical problem. A more general question is whether there is a finite possibility of an asperity 0.01 in high on a ground surface of 100 pin c.l.a., or whether the possibility is exactly zero. (For a Gaussian distribution the probability would be 10- looo, so the distinction is a fine one.) Some of the argument arises from a confusion between a sample and a population: the highest asperity on an actual surface will certainly have a finite height but so will the largest of a sample of 10 000 asperities from a Gaussian population. In fact, the expected greatest height will be about 4a above the mean, the same as the maximum possible height for the cubic distribution of Fig. 6b. A point often overlooked, however, is that the highest member of a sample from afinite distribution, such as the cubic distribution, is also subject to sampling variations. In fact, we find that the interquartile range (19)for the highest of a sample of, 10 000 from the cubic distribution-the range in which the probability of finding the highest is one half-is from 3 . 4 9 ~ 3 . 6 6 ~ Appendix 3). The corresponding interto (see ~ quartile range for the Gaussian is 3 . 6 8 to 4.04~.The only important difference is the extent of the spread; twice as great or the Gaussian as for the cubic. The only distributions for which the spread is very different are those 'with a sharp edge', i.e. those like the rectangular distribution for which the probability density falls abruptly to zero at a particular height. Distributions of this kind have not been found experimentally*. Thus we can say confidently that the highest asperity on a surface, or the separation at which contact first occurs, are poor reference points. Load-compliance curves based on the point of first contact may be expected to show scatter over their full range introduced by their origin and the more sensitive the apparatus so that contact is detected earlier, the more will be the scatter.
CONCLUSION

LOAD-kg Fig. 7. Replot of Ling's experimental results (Fig. 9 of (15)) Ling's original criticism of single-rough-surface models was based on the discrepancy between experimental and theoretical curves on a log-log plot. On a log-linear plot, the discrepancy does not appear and it seems that it arises from an invalid extrapolation t o zero load t o find the origin for the log-log plot
Proc lnstn Mech Engrs 1970-71

Any model of contact between surfaces, both of which are assumed to be rough, can be simulated by a model in which only one surface is rough. Models with both surfaces rough, which take into account experimental findings on surface topography, give results indistinguishable from single-rough-surface models. Nothing can be deduced from the load-compliance curve about the asperity shape or mode of deformation; the one clear prediction which can be made is that the scale of the separation is determined by the surface roughness, so that a load increase by a factor of about 50 will reduce the separation by one standard deviation of the joint roughness, or by 1.4 x the standard deviation of the individual roughness in the case of two similar random surfaces.

Even if they did exist, combining them to form the joint height distribution will eliminate the 'edge' so that the point of Jirst contact will be subject to scatter.
Vol 185 48/71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

632

J. A. GREENWOOD AND J. H. TRIPP

ACKNOWLEDGEMENT

APPENDIX 3
EXTREME V A L U E D I S T R I B U T I O N S

The authors wish to acknowledge the helpful advice and encouragement of Dr J. B. P. Williamson, Director of the Burndy Research Division in Norwalk, Connecticut, in whose laboratory most of the work described here was done.
APPENDIX 1
THE LIMITATIONS O F ASPERITY MODELS

Equation (9) states that the real area of contact is proportional to F2(d/a).When the load is increased sufficiently, the mean planes will overlap and d becomes negative. We can show that F2x x 2 when x is large and negative, suggesting that the real area of contact may be larger than the apparent area ! This, of course, reflects a basic failure of asperity models: the cross-sectional area of an individual asperity becomes indefinitely large as the distance from its summit increases, because it is difficult to introduce mathematically the condition that the asperity must be bounded by its neighbours. I n a similar way, the number of contacts becomes infinite as d + 00 when in fact it should not exceed ( 7 ~ 2 and )~ should begin to decrease as distinct contacts merge. The infinite value occurs because the model used implies contacts with asperities whose summits lie outside the area . d considered. These problems are unimportant provided only positive separations are considered. It is only the highest parts of a surface which can be modelled by isolated asperities of simple shapes; lower down the individual hills begin to link up just as mountains do on the surface of the earth. The profilometric approach (17)~ despite its limitations, seems to offer the best way around this problem.
N

Let p be the probability of an individual asperity height exceeding a particular value z. Then the probability that in a sample with n members all will lie below z is (1 -p)". A probability distribution is usually described by its mean and standard deviation, but it may equally be described by its median and inter-quartile range, where the quartiles and the median are the values such that +,3and of the members of the population lie below them. Unlike the mean and standard deviation, they are readily found for extreme value distributions: in fact we have simply to solve the equations (1-p(z))" = X where
X = 0.25, 050, or 0.75

We are usually interested in the case of p small and n large, in which case (1-p)" e-np and

p(z)

- -n A h

For example, the distribution of the highest of a sample with lo4 members will have a median and quartiles given by

plower quartile = - 10 - In (0.25) = 1.386x 10pmedian 0.6931 x lo-*; pupper = quartile = 0.2877 x
T o complete the analysis we have to specify the parent distribution from which the sample is drawn. For example, for the cubic height distribution used in the text, we have

APPENDIX 2
I N T E G R A L S OF THE G A U S S I A N D I S T R I B U T I O N

(u-h)" e-uz/2 du d(2n) h These were calculated as described in (9). Useful properties are: for h large and positive

F"(h) = -

Srn

for heights in the range L < z < 2L. If we measure heights from the mean p = 2L/3 and change from the range L to the standard deviation u = L/3 this becomes

Thus, for the sample of lo4 asperities, the median is where

-&(4-g)4
while for h large and negative F"(h) (--h)" The recurrence relation F,, ,(h) = nF,- ,(h) -hF,(h) can be used to build up the higher functions. i.e.
0

0-6931x

z' - = 4-(486x0.6931 x 10-4)1/4= 3.5716

and the quartiles are 3.49060 and 3.65610.

0.0

05 .
1.0 1 3 2.0 2.5 3.0 36 4.0

0500 00 0.308 54 0.158 65 0.066 81 0.022 75 0.006 21 0.001 35 0.000 23 0.000 03

0,411 09 0-225 34 0.104 15 0.039 88 0.012 48 0.003 16 0.000 64 0.000 10 0.000 01

0.398 94 0.197 80 0.083 32 0.029 31 0.008 49 0.002 00 0.00038 0.000 06

0.430 02 0.195 20 0.075 67

0300 00 0.209 64 0.075 34 0.022 85 0005 77 0.001 20 0.000 20 0.000 03

0.616 64 0.240 40 0.080 56 0-022 86 0.005 42 0.001 06 OQOO 17 0.000 02 0.000 00

Proc lnstn Mech Engrs 1970-71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

THE CONTACT OF TWO NOMINALLY FLAT ROUGH SURFACES

633

APPENDIX 4
REFERENCES

HOLM, R. Electric contacts handbook 1958 (Springer,


Berlin). F. D. BOWDEN, P. and TABOR, Friction and lubrication of solids 1954 (Oxford University Press). TABOR, The hardness of metals 1951 (Oxford University D. Press). ARCHARD,F. Elastic deformation and the laws of friction, J. Proc. R. SOC. 1957 243, 190. A ARCHARD, J. F. The temperature of rubbing surfaces, Wear 1959 2, 438. HOLM, R. On metallic contact resistance, W i n . Verbfl. Siemens-Werk 1929 7 217. , GREENWOOD, Constriction resistance and the real area J. A. of contact, Br. 3 appl. Phys. 1966 17, 1621. . TALLIAN, E., CHIU, Y . P., HUTTENLOCHER, T. D. F., KAMENSHINE, A., SIBLEY, B. and SINDLINGER, E. J. L. N. Lubricant films in rolling contact of rough surfaces, A S L E , Trans. 1964 7 109. GREENWOOD,A. and WILLIAMSON, B. P. Contact of J. J. nominally flat surfaces, Proc. R. SOC. 1966 295, 300. A WHITEHOUSE, J. and ARCHARD,F. Properties of random D. J.

surfaces of significance in their contact, Proc. R. SOC. . A 1970 316, 97. (11) THOMAS,R. and PROBERT,D. Establishment of conT. S. tact parameters from surface profiles, J . Phys. D: Appl. Phys. 1977 3 277. , (12) BICKEL, Some fundamental problems in the measurement E. of surfaces, Proc. Int. Conf.Prod. Eng. Res. 1963 (Am. SOC. Mech. Engrs). (13) OSTVIK, R. and CHRISTENSEN, H. Changes in surface topography with running-in, Proc. Instn mech. Engrs 196869 183 (Pt 3P), 57. (14) WILLIAMSON, B. P., PULLEN, and HUNT,R. J. The J. J. shape of solid surfaces, Surface Mechanics, ASME 1969. (IS) LING, F. On asperity distributions of metallic surfaces, F. 3. Appl. Phys. 1958 29, 1168. [Note also corrections given 81 in discussion of (17),p. 8 . (16)RYDER, H. Strength of materials 1969 (Macmillan). G. (17) GREENWOOD,A. On the area of contact between a rough J. surface and a flat, J. Lubric. Technol., Trans. A m . SOC. mech. Engrs 1967 1, 81. (IS) ZHURAVLEV,A. Onthe physical basis of the AmontonsV. Coulomb law of friction, Zh. tekh. Fiz. 1940 1 , 1447. 0 (19) GUMBEL, J. Statistics of extremes 1958 (Columbia UniE. versity Press, New York).

Proc lnstn Mech Engrs 1970-71

Vol 185 48/71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

D243

Discussion
R. M. Baul Member and W. Scott Member The authors have stressed the Gaussian model regarding the description of surfaces since this is probably the most frequently encountered. They mention, however, that the height distribution for a worn surface, which was found to be non-Gaussian, can be treated as Gaussian when considering only that portion which is relevant in contact. Our own studies on abrasive surfaces have resulted in analysis of the outer asperities of a grinding wheel surface. Because of wheel porosity, a continuous touching profile cannot be obtained when applying stylus techniques to these surfaces. We have found the height distribution of the asperities to be non-Gaussian. A typical trace is shown in Fig. 8a. Tests with a transducer other than a stylus would indicate that an approximation of the whole surface is Gaussian; this is confirmed by other workers (20) (21) who applied replica techniques. We seem to have the opposite effect from the authors, i.e. a non-Gaussian part of what is normally taken as a Gaussian distribution. Theoretically, this should not happen but the difference

may be due to the increased resolution when examining only the field of interest. This is in contrast to the authors lack of resolution in this region (Fig. 1). Is it possible that other surfaces which are termed as Gaussian may be found to be non-Gaussian if only the outer relevant asperities are considered ? When surface asperities are deformed plastically the displaced material may affect the original height distribution. It is normal to consider only truncated asperities in assessing the contact area in conjunction with the approach of the surfaces. Would the authors consider the discrepancy due to this factor significant ?
REFERENCES

PEKLENIK, Contribution to the correlation theory for the J. grinding process, Trans. Am. SOC. mech. Engrs 1964 B86, 85. (21) STRALKOWSKI, Wu, S.M. and DEVOR, E. CharacC. M., R. terisation of grinding wheel profiles by autoregressivemoving average models, Inr. J. Mach. Tool. Des. Res. 1969 9 145. ,
(20)

> k

i dl
a
a
a
0

M. M. El Refaie Member The authors remark that in the asperities the contact usually takes place at the shoulders rather than at the tips is an important proposition. It actually throws light on the requirements of a more generalized solution to elastic contact problems. It is a remote possibility that both surfaces of the asperities will have to be aligned so that tip-to-tip contact will take place. I n addition to this, perfectly smooth surfaces do not usually exist; therefore, surface parameters relating to tip dimensions are of no consequence. (A perfectly smooth surface is either composed of one asperity of infinite radii of curvature or an infinite number of asperities with infinitesimally small radii of curvatures.) Surfaces in profile can be fully defined by either of the two following definitions.

(1) An all height probability distribution function in conjunction with a suitable auto-correlation function (10)
I

HEIGHT
U

MAX

HEIGHT

(24-

a Probability density function. b Cumulative distribution function. Fig. 8. Probability functions for 20A80-K5-VBE grinding wheel with 0.001 in stylus penetration
Proc lnstn Mech Engrs 1970-71

(2) Peaks and valleys probability height distribution function in conjunction with a suitable geometry para2) meter (22) ( 3 .

The two definitions are obviously in parallel. The all height probability distribution can be obtained in terms of
Vol 185 48/71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

D244

DISCUSSION ON J. A. GREENWOOD AND J. H. TRIPP

the geometry parameters and the peaks height distribution; some assumptions can make this also apply to the auto-correlation function (22). The authors have used definition ( 2 ) in their analysis. However, from our studies of surface models (23) we have not restricted the model to an infinite probability distribution and it does seem that the analysis presented in this paper can be generalized to include height distributions with finite cut-offs. The disadvantages of an infinite distribution of peaks heights can be seen from the following.
Nominal dimensions of surfaces When the peaks height probability distribution is infinite (Gaussian) and the asperities have a realistic finite geometry, the nominal dimensions of the surfaces are infinite. Apparent contact area It appears that A =

H. A. Francis London A measure of the quality of any asperity model of a single surface is how closely its bearing area curve conforms to the true bearing area curve given by the surface height distribution. Similarly, the total projected area of intersection of two rough surfaces as given by an asperity model may be compared with the bearing area curve of the algebraic sum z,(x,y) of the two surfaces, since the surfaces overlap wherever z,(x,y) > f, where f is the separation of their mean planes. For many real surfaces, both peak and surface height distributions are Gaussian (14); I limit this discussion to such surfaces. For clarity, a slightly different notation will be used for certain quantities. Surface height and peak height above the mean surface level are denoted by z, z9;subscripts 1, 2, s refer to the two real surfaces and the Imaginary sum surface; u,~, u, upl,ups are the standard deviations of , , the random variables zl, z p l ,zps,respectively. For the z,, authors model we may write

lmco 2 m dr =

wp = z p 1 + z p a - s

= (Zpl-Zpl)f(Zpz-G)

Therefore, the expected total load and the expected total real area of contact (equations (3) and (4)) are infinite. According to the statistical analysis of surface models (23), the length of the profile is determined by the height distribution and the geometry of the asperities. Therefore, the upper limits of integration in equations (2)-(5) should be related to the surface parameters, including the nominal dimensions.
The area asperity density In all previous work of the authors and myself (24), 7 has been assumed to be uniform in a plane parallel to the surface. The validity of this assumption is now questionable. I would expect a higher asperity to occupy a larger projected area in a plane parallel to the surface than in the neighbouring small asperities. The effect of this is to introduce 7 as an increasing function of r into equations (2)-(5). This function will be related to the asperity dimensions and peak height distributions, by virtue of equation (2). In conclusion, it does seem that missing out information about the relationship between the peaks height distribution, the geometry of peaks and the nominal dimensions of the surfaces resulted in the load-approach, and the load-area of contact relationships, being independent of the mechanical and geometrical properties of the asperities.
REFERENCES

-P-(zpl+G)I (17) Now, the authors Gaussian probability density distribution cbo(t)is symmetric about t = 0. Therefore, its argument must be the variable ( z p l - z p l ) + ( z p 2 - ~ , and the authors separation variable d is given by
d = S-(Z,,+z,,) . . . * Since zl, are mutually independent, we have z2
(18)

: , a = u.,12+u,z2 . . (19) It is shown elsewhere (25) that for two statistically identical Gaussian surfaces,

= %l/%l = u p z / % z (20) which, with equation (19), gives ups = 2ll2uP1; upstherefore has the same value as the authors u.
fJps/%s

Thus, for identically Gaussian surfaces we have, for the authors non-aligned asperity model (equation (13))
= 4.r2(71B1a,1)Z~z(d/o,,) and for aligned asperities (equation (14))

. .

(21)

(22) We wish to compare these two equations with the bearing area curve of the sum surface (itself Gaussian) given by its cumulative height distribution function
~ ~ 7 1 s 1 ~ P 1 ~ ~ 1 ~ ~* / o p s ~

ad =

(23) When surface 2 is smooth, the asperity model gives

I/& Fo(f/uzs) . . . =

EL REFAIE, M. Aecto-correlation analysis of a statistical model of surface projile. (To be published.) (23) HALLING, and EL REFAIE, J. M. A statistical model for engineering surfaces, Tribology Convention 1971 60 (Institution of Mechanical Engineers, London). (24) EL REFAIE, M. A numerical solution for the problem of contact of a sphere and a rough surface, Departmental Report, University of Salford, No. USME/T/4/68.
(22)

Z / d = 2.r(71P,(JPl)Fdd/.,,) * (24) to be compared with the actual bearing area curve of surface 1, A/d = Fo(f/u,,) . . * (25) From equations (18)-(20) we may write, for identically rough surfaces, d S - = 212A+p-. . . * (26)
02,

UPS

Proc lnstn Mech Engrs 1970-71

Vol 185 48/71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

THE CONTACT OF TWO NOMINALLY FLAT ROUGH SURFACES

D245

and, for surface 2 smooth

where h = zPl/u,, and p = uPl/uz1.Thus, given values for h and p, expressions (21)-(23) can all be plotted as functions of d/u,,, and expressions (24) and (25) as functions of dla,,. We will use the values h = 1.14, p = 0.89 measured for a bead-blasted surface for which both zland zpl were strongly Gaussian (14). In Fig. 9a and b, equations (21)(23) and (24)-(25) are plotted in the normalized forms (Ai,Ql)/~o(212h+p~/~P~)Y i , Q l ) / ~ o ( ~ + p ~ / a , , ) , (A respectively, to facilitate comparison of the functions. Equations (21), (22) and (24) are plotted as shaded areas corresponding to the range of experimental values 0.03 ~ l P l u P l Q 0.05.Fig. 9a shows that above the mean peak height (d > 0) the asperity model roughly simulates the true

021

S d _- X+PG -

. . .

(27)

bearing area curve of this Gaussian surface. Fig. 9b shows that as long as the mean peak levels do not overlap (d >O), the non-aligned asperity model is a reasonable approximation to the geometric intersection of two such beadblasted surfaces, whereas the less realistic aligned model is not. Whether the statistical parameters A, p are constants for all Gaussian surfaces remains to be seen. I n any event, the results for this bead-blasted surface do support the authors model.
REFERENCE

(25) FRANCIS, A. H.

Predicting contact area using the algebraic sum of two surfaces. (To be published.)

<

W l LL

SE PA RAT I0 N, d/vp I

- - - - Asperity model (equation (24)) 0.03 < 7, 8, upl Q 0.05.


True bearing area (equation (25)). a One surface smooth.

V. E. Gough Fellow I n the Introduction, the authors say that they treat several cases and present some general conclusions. I n the opening paragraph in the section on General theory, the authors appear to consider the contact of two surfaces from the same population of asperities. I n the section on Elastic deformation they permit the introduction of different values of E for the two surfaces. There are some important classes of pairs of contacting materials in which E, Y and asperity characteristics are specific to each material and different for the two materials; e.g. metal and plastic or metal and long chain highly elastic polymer, or roadstone and long chain highly elastic polymer. How would widely different asperity characteristics influence the findings ? For instance, which would be taken as the smooth surface, the metal or the polymer? Would the other retain its asperity distribution modified for height and material moduli, or must the asperity distribution be computed from those of the two materials ? P. K. Gupta Latham, New York It is true that the analysis for two rough surfaces is similar to that for a rough and smooth surface. This is derived directly from the statistical descriptions of the .surfaces. As the authors have pointed out, it is the distribution of the sum of the peak heights on the two surfaces which determines the probability of contact. Now if all the contacting asperities are geometrically similar, the analysis in closed form integrals, as proposed by the authors, is directly applicable to either the problem of a smooth and rough surface or the interactions of two rough surfaces. Recent work (26) has shown that the radius of curvature at the peaks has a skewed distribution and the data obtained are best fitted by a log-normal function. However, the statistical correlation between heights and radii is negligible. Thus, for the analysis of interacting surfaces, peaks of all heights may be assumed to have identical radii of curvature distribution. Hence, spherical asperities of varying heights and radii are assumed. The introduction of this fact into the analysis results in the replacement of simple closed form integrals by slightly more complicated numerical integrals, which are easily solved by the
Vol 185 48/71

- Non-aligned asperity model (equation (21)). 0.03 < 7, j, 3 - - - - Aligned asperity model (equation (22)). J u , ~ < 0.05.
Sum surface bearing area (equation (23)).

b Both surfaces rough.


Fig. 9. Discrepancy among bearing area curves of various models
Proc lnstn Mech Engrs 1970-71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

D246

DISCUSSION ON J. A. GREENWOOD AND J. H. TRIPP

use of modern high-speed computing machines. Furthermore, the simple elastic model may be replaced by plastic deformation models; also, surfaces may be subjected to a sliding interaction. Such a general analysis is quite similar in nature for interaction of a pair of rough surfaces and the problem of a rough and smooth surface. The basic assumption of this analysis is that if the surface topography of each of the mating surfaces is described by statistical distributions of peak heights and radii, as determined experimentally by the analysis of surface profiles than with a known junction deformation model, the contact problem of mating surfaces is solved. Using Hertz solution and elastic contacts and Green's slip-line field solution for plastic contacts, the proportionality between the real contact area and normal load is explained in terms of the size distribution of microcontacts. It is found that for the various pairs of surfaces analysed, the density of microcontacts is proportional to 0-91 power of the normal load and the average radius of a microcontact varies as 0.045 power of the load. The similarity between the problems of a pair of rough surfaces and a rough and smooth surface seems quite obvious and the topographic analysis proposed by the authors and extended by the discusser, gives quite interesting results. The next major question concerns the representation of an actual surface by a single surface profile. It is therefore necessary to establish a relationship between surface statistics and statistics of surface profiles. This will indeed strengthen the present topographic models for mating rough surfaces.
REFERENCE

mathematics, it will be assumed that the friction force is given by pP", which will be an overestimate of the effect. The asperity deformation law is assumed to be
p" = 4 E'p,112w"312 -

. . .

(29)

From the geometry shown in Fig. 10, equation (29) becomes

Expanding equation (30) and using equation (2) there results


P ~ ( w ,= 2777 - E'pe1I2 ) 3
X

E 2777 - E'/3,lI2 /o4x( '


3

[(1 -$)+$ - r dr &)I :j)3'2 +$)r dr (I


(1
Wp--

(wp-f)3'2

(31) (32)

Integration is simplified by letting r2 = x, which reduces equation (32) to a form to be found in (29). The result is Po(wP)= - ~ 7 E ' / 3 ~ ' ~ 15
16d2

. .

(33)

(26) GUPTA, K. P.

Topographic analysis of friction between a pair of rough surfaces Sc.D. Thesis, Massachusetts Institute of Technology, 1970.

Substituting equation (33) into equation (3) and integrating using the easy to handle exponential distribution of heights (see (9)), gives

J. J. Kauzlarich Charlottesville, Virginia

The authors have added an important paper to a series on the subject (8) (9). Since this paper is so closely related to the prior literature, I wish to add a comment about the printed discussion (17) concerning the bearing area law. Halling and El Refaie (23) recently pointed out that Abbott and Firestone (27) correctly set the bearing area to apparent area ratio equal to the total land length at heighty to total length ratio. I n the field of metallography a similar law is used (28), where, in the case of an aggregate material Volume ratio = area ratio = length ratio (28) As a part of the development in the section on elastic deformation, the authors assume that slopes on real surfaces of contact (asperities) are so small that the errors may be ignored. I n their model, they apply macroscopic Hertz theory to a microscopic event of asperity contact but neglect the frictional force, It is interesting to consider the effect of friction on their results. Fig. 10 includes a vertical contribution of load due to friction. To simplify the
Proc lnstn Mech Engrs 1970-71

[(z)

+0-491p ,(!] &3)

(34)

For the bead-blasted materials investigated by Thomas and Probert (II), d/.IBvaries from 0.07 to 0.4. Letting p = 0.5, where (5/2)! = 3,323, the correction term in equation (34) will increase the pressure by 3-15 per cent. A more realistic assumption for frictional force will decrease this correction term.

Fig. 10. Contact of two hemispherical asperities

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

THE CONTACT OF TWO NOMINALLY FLAT ROUGH SURFACES

D247

A similar analysis for the real area of contact to apparent area ratio based on (the authors assume 8 = 0) A = Tp,wcos2 e
results in

. . -

(35)

Perhaps it might be concluded from this and numerous similar findings already reported, that most of the surface descriptions presently used are adequate. Nevertheless, it is unfortunate, although probably natural, that the whole field still does not have something which resembles a common description.
REFERENCES

The term -u/P is the correction, and represents a decrease in real area of contact by 0.5-185 per cent for bead-blasted surfaces. This analysis does not alter the major conclusions of the authors.
REFERENCES

(30) MIKIC, B. Analyticalstudies of contact of nominally flat B. surfaces; effect of previous loading, 3. Lubric. Technology, mech. Engrs. (To be published.) Trans. Am. SOC. (31) MIKIC, and ROCA,R. On elastic deformation of rough B. surfaces in contact. (To be published.)

(27) ABBOTT,E. J. and FIRESTONE,A. Specifying surface F. quality, Mech. Engng 1933 55, 569. (28) RHINES,R. N. Origins of quantitative metalography, The Sorby Cent. Symp. on the History of &etallurgy 1965, 417 (Gordon and Breach Scientific Publications, New York, American Institute for Metallurgical Engineering). (29) GRADSHTEYN,S. and RHYZHIK, M. Table of integrals I. 1. series and products 1965, 284 (Academic Press Inc., New York).

B. B. Mikic Cambridge, Massachusetts The authors attempted to answer the question whether the contact problem can be generally treated by assuming only one of the contacting surfaces to be rough. In the paper, the surfaces were described by a peak height distribution and all asperities were assumed to have identical shape. Since the answer to the question might depend on a model one uses to describe the surfaces in contact, the consideration presented in the paper is more specific than general. Moreover, we are inclined to accept the papers conclusions more as an indication of adequacy of the method used to describe the surfaces rather than as a proof that the contact problem can be treated generally by assuming only one surface rough. The latter, we believe, is a priori possible provided one uses an adequate surface description capable of forming a representative equivalent rough surface. In respect to the authors model, it might be pertinent to mention that the use of some alternative, and probably less restrictive, surface descriptions shows that the expected peak curvature is a function of the peak height (10) (30). This does not agree with the description used in the paper. It is interesting to note that in spite of these discrepancies, when we applied a model which represented a rough surface by Gaussian distribution of surface height (not peaks) and arbitrary (but independent of heights) distribution of profile slope to the problem of elastic contact, we reproduced some of the conclusions obtained by the authors model. In particular, it was found (31) that the actual contact area was exactly directly proportional to the pressure
A, _A,
1 P = - erfc (Etan8)/42 4

T. I?.Thomas Middlesbrough It is interesting to compare the results of the authors with recent theories (32) (33) of non-genuine two-roughsurface models. These recent theories treat only the case of plastic contact and assume initially that the real area of contact is equal to the ratio of nominal contact pressure to the flow pressure of the softer material. Their result for the variation of separation with load may be written in the authors terms as

PlH =

+ww

This may be compared with equation (14) in Fig. 11, which also shows an earlier result (34) which took some account of the mutual interference of rough surfaces. The authors have also tackled the long-neglected question of the distribution of the highest points on a finite surface. This is an important problem with obvious engineering applications. Many surface profiles appear Gaussian in the middle regions of their height distributions, but the shape of the central tiles of a distribution is a notoriously unreliable guide to the properties of its tails. As the largest sample that has so far been taken is only about 20 000 ordinates (35) the hypothesis that surfaces
3,
I

(5)

--- Equation (14). - - - - Theory of (32) and (33).


Fig. 17 from (34).
Fig. 11. Variation of separation with dimensionless load
Vol185 48/71

where tan 0 is c.1.a. of the slope.


Proc lnstn Mech Engrs 1970-71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

D248

DISCUSSION O N J. A. GREENWOOD AND J. H. TRIPP

are Gaussian at their upper extremities must be regarded as still unproved. Again, the extreme properties in question are not those of a discrete distribution, but rather those of a three-dimensional continuous random process, with a consequent difficulty in the definition of the sample size.
REFERENCES

J. A. Greenwood Cambridge and J. H. Tripp Hartford, Connecticut (Authors) We can only agree with T. R. Thomas and with R. M.

(32) TSUKIZOE, T. and HISAKADO, On the mechanism of T. contact between metal surfaces: Part 2-The real area and the number of contact points, Trans. A m . SUC. mech. Engrs 1968 9OF,81. (33) KIMURA, Estimation of the number and the mean area Y. of real contact points on the basis of surface profiles, Wear 1970 15,47. (34) IWAKI, and MORI,M. On the distribution of surface A. roughness when two surfaces are pressed together, Bull. J.S.M.E. 1958 1, 329. (35) WILLIAMSON,B. P. Microtopography of surfaces, Proc. J. Imtn mech. Engrs 1967-68 182 (Pt 3K), 21.

T. Tsukizoe Osaka, Japan The analysis presented by the authors is lacking in experimental information for contact area, separation and number of contacts. The main attraction of the paper lies in the loadcompliance curve with a slope such that a movement of one standard deviation of the joint roughness gives a load increase by a factor of about 50 (Fig. 5). Tsukizoe and Hisakado have shown (32)that the load-compliance curve (Fig. 12) has a similar slope to the authors results. With regard to the value of separation, however, it would appear that the authors values are considerably below ours which were compared with the experimental results. The agreement between experiment and theory was fairly good (32).

- 10-6

Io - ~ NOMINAL PRESSURE,

Io- w=Wlp,

10-2 L,Ly

10-1

Fig. 12. Relation between load and separation


Proc lnstn Mech Engrs 1970-71

Baul and W. Scott that more experimental evidence is needed on the tails of height distributions to settle the question of whether heights are truly Gaussian or not. Unfortunately, if the counting is carried out on an extended distance along the surface, doubt arises about the stability of the datum. The collection of data from a number of traverses within a small area (which is how the 20000 ordinates mentioned were obtained) is far from straightforward. It is perhaps worth repeating that the Gaussian has no particular properties making it distinctive as far as surface contact is concerned, although it does 1) play a central role in the analysis of surface roughness ( 0 . We must apologize for not referring to T. Tsukizoes work in our paper. This was because our first draft was written before his paper appeared, and the difference in approach made it difficult to incorporate. A particular success of T. Tsukizoes method is that it predicts the decrease in the number of contact spots at very high loads, to which we referred in Appendix 1. However, or loads greater than +Ha!, his model predicts that the number of contacts becomes negative, so that, just like our model, his must not be taken too far. The difference in the value of the separation which he mentions, and which T. R. Thomas illustrates, is, we think, purely a question of definition; we use the distance between the mean levels of the peaks while he uses the (more useful) distance between both the mean surfaces. Measurements (9) and theory (10) suggest that the difference will be about 0.8(u1 u) 2. We rather agree with B. B. Mikic and P. K. Gupta that the equivalence between two rough surfaces and a single rough surface is axiomatic. However, in the past it has been asserted that it was neither axiomatic nor true. We are somewhat confused over the correlation between peak curvature and peak height. A noticeable correlation is predicted in (10)as well as in B. B. Mikics paper; but P. K. Gupta claims the correlation to be small, and a report by Nayak (36) in which a profile is regarded as a section of a random isotropic surface and not merely as a onedimensional random walk (as in (10)) also shows a small correlation. As the discussers state, a variation in curvature which is independent of height can easily be incorporated. Archard (37) has recently shown that for contact between a single rough surface and a plane, the correlation between curvature and height found in (10) can be included, and leads to an even closer proportionality between area of contact and load. It seems probable that the results will not be much affected whatever correlation is assumed. We are grateful to J. J. Kauzlarich for putting numbers to our somewhat casual assumption that friction is unimportant. This is now seen to be justified except when high friction and rough surfaces occur together and, even then, merely adjusts the values of the proportionality constants. We are pleased that our model passes H. A. Francis test, for we think the test is a good one. Indeed, we con-

Vol 185 48/71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

THE CONTACT OF TWO NOMINALLY FLAT ROUGH SURFACES

D249

sider that the bearing area approach (the profilometric approach of (17))is the superior way of studying contact, and that once the idea of an effectively constant contact pressure has been firmly established by the use of a range of asperity models, then asperity models should be abandoned. We have thought this for some time; but since the appearance of Whitehouse and Archards theory (10) suggesting that asperities, slopes, and peak curvatures are all instrumental properties which would be found to be infinite with a sufficiently refined instrument, we have felt it much more strongly ! We look forward to H. A. Francis paper using this approach. It is interesting to learn from R. M. Baul and W. Scott that they find grinding wheel surfaces to be non-Gaussian. This agrees with the findings of Bhateja (38). The wheel can have an almost linear height distribution, but produces a more or less Gaussian height distribution on the workpiece. Presumably the explanation involves the central limit theorem, as described in (14). We are perfectly happy about our neglect of the displaced material in calculating contact area. The plasticrigid incompressible material of metal-working theory plays little part in contact theory. It is helpful to remember that there are two separate, competing effects; a tendency for incompressible material to be displaced sideways and increase the area above the geometrical value, and a tendency for material displaced vertically to drag down adjacent material and decrease the area below the geometrical value. For purely elastic contact of spherical asperities, the second effect is dominant and the contact area is exactly half the geometrical value. For a plasticrigid material, the area is increased. Recently, Johnson (39) has shown that for an elastic-plastic material the material displaced by a rigid indenter can be accommodated by a radial elastic expansion of the surrounding material. We believe this to be true in asperity contact. In reply to V. E. Gough, dissimilar properties of the two surfaces should not affect the theory. There is a certain amount of surplus material to be accommodated, determined by the combined asperity distributions. This is absorbed in a way determined by the combined material properties. For purely elastic contact the material is shared in the proportion
1-v,2
1--Y,2

a contact between two elastic-plastic materials, but it seems unlikely that the overall predictions are affected. However, although this will be true for contact between nominally flat surfaces of roadstone and a long chain highly elastic polymer, it may not be true about contact between roadstone and a highly curved tyre, for which the variation in the apparent area of contact with load may well be dominant. M. M. El Refaie is unhappy about our use of an unlimited height distribution. As long as we keep to generalities, it is quite possible to work in terms of a distribution $(z) which we subsequently set equal to zero outside a range -L < Z < +L, or equally to work with a finite distribution with a range fL and subsequently let L tend to infinity. We know of no experimental evidence for the existence of a finite cut-off. We do not agree that an infinite height distribution implies an infinite surface area, or that the load and real area become infinite. The deficiencies noted in Appendix 1 apply equally to a finite distribution, except that infinite is replaced by very large. It is true that the upper limits of the integrals in (2), (3),(4), (5) should be related to the surface area, but it is reasonably clear that the approximation is close except for the asperities along the perimeter of the surfaces, and that there will be too few of these to matter. We are not clear why the validity of the assumption of a uniform asperity density should now be questionable; clearly it always was. As far as we can see, (22) does not discuss this point, while (10)seems to us to suggest a uniform (random) distribution, which is what we assume. Cramb (40)shows that very high peaks occur randomly. No theoretical answer is known for others. Finally, while we are fully prepared to agree that our theory has deficiencies, we cannot agree that these deficiencies will alter the conclusions.
REFERENCES

For purely plastic contact, the softer material presumably absorbs it all. Unfortunately, we know rather little about

(36) RANGANATH NAYAK, Random process model of rough P. surfaces, Report TIR-75 1970 ASMEIASLE Joint lubric. Conf. (Bolt Beranek and Newman Inc., Cambridge, Massachusetts). (37) ARCHARD, F. Personal communication, 1970. J. (38) BHATEJA, P. Influence o dressing and performance of C. f grinding wheels Ph.D. Dissertation, University of Salford 1970. (39) JOHNSON, K. L. The correlation of indentation experiments,J. Mech. Phys. Solids 1970 18, 115. (40) CRAMBR, Extreme value problems in stochastic processes. H. Unpublished proceedings of Oceanography Conference, Portugal, 1968.

Proc lnstn Mech Engrs 1970-71

Downloaded from pme.sagepub.com at NORTHEASTERN UNIV LIBRARY on May 29, 2012

You might also like