# 1 ELASTIC-PLASTIC CONTACT OF A ROUGH

## Comments

## Transcription

1 ELASTIC-PLASTIC CONTACT OF A ROUGH

ELASTIC-PLASTIC CONTACT OF A ROUGH SURFACE WITH WEIERSTRASS PROFILE Yan-Fei Gao and A. F. Bower Division of Engineering, Brown University, Providence, RI 02912, USA The classical approach to modeling contact between rough surfaces follows a model developed by Greenwood and Williamson (1966), in which surface roughness is idealized as a collection of discrete asperities, with a Gaussian height distribution, and (in the simplest model) identical curvatures. Despite the widespread acceptance of this model, it has proved difficult to apply to realistic surfaces, whose roughness resembles a self-affine fractal over a wide range of length scales. For a perfectly fractal surface, the Greenwood-Williamson model breaks down completely, because it is impossible to detect individual asperities on the surface, and to determine their curvature. Several recent models of rough surface contact have therefore been based on a fractal description of the surface geometry: the Weierstrass profile is one such example. Ciavarella et al (2000) have recently reported a rigorous analysis of elastic contact of the Weierstrass profile. Our goal is to extend their analysis to the more realistic case of an elastic-perfectly plastic solid. To this end, we consider an isotropic, elastic – perfectly plastic solid with a rough surface, which is characterized by the two-dimensional Weierstrass profile. The solid is indented by a smooth frictionless cylinder. We attempt to compute quantities such as the number of contact spots, their size (or more precisely, their size distribution); the true contact pressure acting on the contact spots, and the total true area of contact. In addition, we calculate critical conditions that will initiate plastic flow in the asperities, and determine the extent of this plastic flow as a function of the applied loading, the material properties of the deforming solid, and the surface roughness. An important conclusion of our study is that a perfectly fractal description of surface roughness appears to lead to unphysical predictions of the true contact size and number of contact spots, for both elastic and elastic-plastic solids. Possible approaches to resolving these difficulties are discussed. Keywords: self-affine fractal, contact mechanics, rough surface plasticity 1. Introduction When two rough surfaces are pressed into contact, they meet only at the highest points on the two surfaces. To predict the size of the resulting contact spots, the true contact pressure, and the true area of contact is a venerable problem in contact mechanics, and is clearly of considerable practical importance, since phenomena such as friction, wear, and contact fatigue are controlled by the nature of asperity contacts. Nevertheless, a fully satisfactory solution remains elusive. The essential character of surface roughness, and the implications on the nature of contact between rough surfaces, were recognized in the pioneering work of Archard (1957). His approach was to model a rough surface as a sphere, which has an array of smaller spherical bumps on its surface. These spheres in turn have still smaller spheres on their surfaces. If this geometric process is continued indefinitely, one constructs what would today be regarded as a 1 surface with fractal geometry. Archard analyzed elastic contact between such a surface and a flat indenter, and predicted the number of contact spots, the contact size, and contact pressure for each successive scale of spheres. One important result of this calculation is that if the surface roughness contains a finite number of scales of spheres, the true area of contact is proportional to the load, thereby providing an elegant explanation for Amonton’s laws of friction. For the limit of a perfectly fractal surface (with an infinite number of scales of spheres), one finds that the true contact area consists of an infinite number of contact spots with zero size, which are subjected to infinite contact pressure. Archard’s model contains an excellent qualitative description of a rough surface, but cannot easily be related to actual measurements of real surfaces. Greenwood and Williamson (1966) therefore devised an alternative approach to characterize surface roughness, and to model contact between two rough surfaces. In their original work, Greenwood and Williamson idealized a rough surface as a collection of asperities with spherical tips of equal radius, but with a Gaussian distribution of heights. They neglected interactions between neighboring asperities, and so were able to relate the behavior of the surface to the force-displacement relation for an individual asperity and the asperity height distribution function. This approach predicts a well-defined number of total contact spots, contact size, contact pressure and contact area, which depend on the curvature of the asperity tips, the asperity height distribution, and the elastic and plastic properties of the two contacting surfaces. The Greenwood-Williamson approach has been refined and extended by a number of authors (e.g. Whitehouse & Archard 1970; Nayak 1971; Bush et al 1975; McCool 1985; Greenwood 1984), and is widely used in many applications. Since the early work of Greenwood and Williamson, a great deal of work has also been done to characterize the statistical properties of surface roughness. Various approaches have been developed, including through its autocorrelation function (Whitehouse & Archard 1970), power spectral density (Sayles & Thomas 1978) or as a semi-deterministic fractal (Majumdar & Tien 1990; Majumdar & Bhushan 1990, 1991). There has been some debate as to the most accurate description of realistic surfaces, but all models agree that surfaces have an approximately selfaffine fractal character over a wide range of length scales, and usually continue to exhibit fractal behavior down to the limits of the resolution of standard surface profilometers. Indeed, Majumdar and Bhushan (1990, 1991) show data that suggests that roughness spectra for hard disk drives remain fractal down to wavelengths of a few hundred nanometers. This behavior leads to some serious problems with Greenwood and Williamson’s approach, because it is difficult to detect individual asperity tips on a surface, and even more difficult to measure their curvature. Surface profiles are measured by sampling the surface at discrete intervals. In two dimensions, asperity tips are identified by finding points that are higher than their neighbors. Finding summits in three dimensions is more difficult, but conceptually similar. The curvature of each asperity tip can be determined by fitting a curve (in two dimensions) or surface (in three dimensions) through the points surrounding the peak. For a differentiable surface, this procedure converges to identify a well-defined set of asperities as the sampling interval is reduced. When this procedure is applied to experimentally measured surfaces, however, the number of asperities and their curvature appear to increase without limit even as the sampling interval is reduced to extremely small values. This, of course, is a consequence of the 2 fractal nature of surface roughness. For a perfectly fractal surface, Greenwood and Williamson’s procedure breaks down completely, since such a surface does not contain well-defined summits. Many recent studies of rough surface contact have therefore abandoned the GreenwoodWilliamson description of surface geometry in favor of one that captures in some way the fractal character of surface roughness. Various approaches have been proposed, including ad-hoc procedures for adapting Greenwood and Williamson’s calculations to fractal surfaces (Majumdar & Bhushan 1990, 1991), direct numerical simulations of experimentally measured profiles (e.g. Sayles, 1996; Polonsky & Keer 2000a, 2000b; Borri-Brunetto et al 1999; Hyun et al 2004; Pei et al 2005), or hybrid schemes which combine numerical simulations with Greenwood and Williamson’s approach (see Mihailidis et al 2001 for a review). Attempts to adapt GreenwoodWilliamson to a fractal surface appear to suffer from the same difficulties as the original Greenwood-Williamson analysis: there is no unambiguous way to count, or to characterize, asperities on a fractal surface. Numerical simulations provide considerable insight into the deformation fields associated with rough surface contact, but there is a limit to the spatial resolution of any numerical scheme, which artificially truncates the roughness spectrum. This again introduces problems very similar to those associated with the Greenwood-Williamson model: for a fractal surface, predictions of the contact size, number of contacts, and true contact area will not converge as the resolution of the numerical scheme is refined. Great care must be taken to ensure that this truncation does not lead to spurious predictions (Borri- Brunetto et al 1999). A significant step towards a satisfactory model of the elastic contact between rough fractal surfaces was made recently by Ciavarella et al (2000) and by recent work of Persson (2001, 2005). The approximate approach of Ciavarella et al (2000) will be followed closely in our work, and so will be described in more detail in Sections 2 and 3 of this paper. Briefly, they chose to approximate surface roughness using the two-dimensional Weierstrass function given in Eq. 1 below. The Weierstrass function consists of a sum of sinusoidal profiles, with progressively decreasing wavelengths Λ m = Λ0γ − m where m = 0,1,2K ∞ and Λ 0 and γ are geometrical parameters. The amplitude of each term in the series varies with m as G = G0γ ( D− 2) m , and for 1 < D < 2 the function is a self-affine fractal with dimension D. It is difficult to analyze contact of rough surfaces with realistic values of γ , but Ciavarella et al show that much progress can be made by taking γ >> 1 . Under these conditions, the wavelengths in the Weierstrass profile are widely separated, and each term can be analyzed independently of the rest. The surface then resembles Archard’s early model, and can be treated in a similar way: instead of attempting to identify actual summits on the surface, and to predict the true contact size, pressure and area for these summits, Ciavarella et al compute the apparent contact size, pressure and total area for each successive scale of roughness m. They formally take the limit m → ∞ , and recover Archard’s original prediction that the true contact area consists of an infinite number of contact spots, with zero size, infinite pressure, and vanishing total area (Ciavarella et al 2000; Ciavarella & Demelio 2001). This is a purely formal mathematical result, of course: in practice phenomena such as plastic flow or adhesion might alter these predictions, and in any event no real surface is perfectly fractal. There are some applications, such as modeling friction, where the true contact size and area are of great interest, and predicting these quantities remains an unresolved problem. For modeling the compliance of the surface, or its electrical contact resistance, and possibly 3 physical phenomena such as plastic flow, wear, and the initiation and growth of contact fatigue cracks, the behavior of very short wavelength roughness is less important, since damage events occur at sub- micron length scales rather than nanometer length-scales. One advantage of Archard’s model, as extended by Civarella et al, is that the behavior of long-wavelength roughness scales can be predicted in terms of measurable surface parameters without needing to identify real asperities on the surface. This work has provided an important step towards understanding contact between surfaces with a fractal character. Their predictions leave several issues open, however. Their analysis shows clearly that for elastic contact between perfectly fractal surfaces, the true contact area consists of an infinite number of contact spots with vanishing size, which are subjected to infinite contact pressure. The total true contact area also vanishes. This is clearly an unphysical prediction. Of course, real surfaces differ from the idealized model in several respects, which include: (i) real surfaces are not perfectly elastic; (ii) the surface roughness of a real surface cannot be perfectly fractal – the spectrum must be truncated at some very short length-scale; (iv) adhesive forces may act between real surfaces at very short length-scales; and (v) Ciavarella et al’s model is restricted to surfaces with widely separated scales of roughness. Weierstrass profiles with large values of real γ clearly do not accurately characterize real surface roughness, even if one accepts that the Weierstrass representation with γ → 1 is a realistic model of a surface. The question is, which of these effects plays the most important role in controlling the true contact area in real solids? These are important issues for further research. In this paper, our goal is to extend the calculations of Ciavarella et al to account for plastic deformation. The problem to be solved is illustrated in Fig. 2. We consider an isotropic, elastic – perfectly plastic solid with Young’s modulus E , Poisson’s ratio ν and yield stress σ Y . The solid has a rough surface, whic h is characterized by the two-dimensional Weierstrass profile given in Eq. 1. The solid is indented by a smooth cylinder with radius R, which is subjected to a force P per unit length out of the plane. The goal of our analysis is to estimate quantities such as the number of contact spots, their size (or more precisely, their size distribution); the true contact pressure acting on the contact spots, and the total true area of contact. It is of particular interest to calculate critical conditions that will initiate plastic flow in the asperities, and to determine the extent of this plastic flow as a function of the applied loading, the material properties of the deforming solid, and the surface roughness. The remainder of this paper is organized as follows. In the next section, we give a more careful statement of the problem to be solved, the assumptions underlying our calculations, and the results that will be calculated. Section 3 gives a brief review of the indentation response of an elastic-perfectly plastic solid with sinusoidal roughness, as the first step towards analyzing the Weierstrass profile. In Section 4, we review the procedure proposed by Ciavarella et al to analyze contact of the Weierstrass profile, and devise a simplified, approximate procedure that can be used to derive analytical results for both elastic and plastic contacts. Section 5 presents detailed results, and section 6 contains conclusions. The principal conclusion of our calculation is that for a perfectly fractal elastic-plastic surface, the true contact size consists of an infinite number of contact spots with zero size, which are subjected to a finite contact pressure. This limiting contact pressure is approximately twice the 4 material hardness, due to lateral interactions between neighboring contacts. In addition, the total area of contact between the two surfaces is finite. Thus, plasticity resolves some, but not all, of the singular behavior associated with elastic contact between ideally fractal surfaces. Determining the true contact spot size, and the number of true contacts, remain open questions. 2 100 3 S(ω) (µm ) 10 10-2 10 -4 10-6 10-2 10-1 ω (rad/µm) (a) 100 (b) Fig 1. (a) A typical rough surface topography of single crystal copper, measured by the atomic force microscope (courtesy of B.C.P. Burke). (b) Fractal properties can be deduced from the power spectrum. The straight line is by least squares fitting, giving rise to D ≈ 1.47 . 2. Rough Surface Model Majumdar and Bhushan (1991) suggest that the two dimensional Weierstrass function ∞ h ( x ) = G0 ∑ γ ( D −2 )m m =0 cos(2π xγ / Λ 0 ) m (1) with γ ≈ 1.5 and D ≈ 1.4 provides a good qualitative description of many practical surface profiles. As a representative example, we show in Fig 1(a) a three dimensional profile z ( x , y ) of a surface (in this case a copper single crystal) obtained with an atomic force microscope. To compare the surface profile with the representation in Eq.(1), we take a two-dimensional trace z ( x, y = 20µ m ) of the surface in the x direction. We then compute the Fourier transform zˆ(ω ) of the two-dimensional trace, and plot the power spectral density S (ω ) in Fig. 1(b). The power spectrum for the ideal Weierstrass profile is a series of Dirac delta functions, but Berry and Lewis (1980) argue that the Weierstrass distribution can be used to approximate a continuous power spectrum by ensuring that the integral of the continuous power spectral density matches that of the Weierstrass function. The resulting continuous approximation is given by G2Λ S (ω ) = 0 0 (Λ 0ω )2 D−5 (2) 2 log(γ ) 5 and would correspond to a straight line with slope 2D-5 on Fig 1(b). The Weierstrass function is clearly not a fully accurate description of real surface roughness. The actual power spectrum clearly appears to be better approximated by a continuous function than a series of Dirac delta functions. In addition, roughness wavelengths in the range 50 µ m 1mm are strongly sensitive to processing and in general are not well characterized by even the continuous approximation to the Weierstrass function. Similarly, the Weierstrass profile becomes unphysical at very short wavelengths: as m → ∞ , the surface becomes needle- like, with roughness amplitude much greater than its wavelength. Any departure from ideal behavior at long wavelength (low m) will lead to quantitative, but not qualitative changes in our predictions and could readily be incorporated in the computations. The departure from ideal behavior at short wavelengths (large m) has a critical influence on the true size of the contact spots, and for ideally elastic surfaces, on the total true area of contact and contact pressure. It will not, however, have a significant influence on the behavior of lower scales of roughness. Experiments suggest that γ ≈ 1.5 and D ≈ 1.4 in Eq. (1) give the best qualitative fit to real surfaces. Under these conditions, roughness scales with successive values of m do not clearly decouple, however, and analyzing the contact between such surfaces is difficult. Ciavarella et al (2000) show that calculations are greatly simplified by approximating a fractal surface with a larger value of γ > 5 . The surface then consists of a series of harmonic profiles, with widely separated wavelength, each of which can be analyzed independently. Although the resulting surface does not resemble any actual surface, it does capture the self-affine fractal character of real surfaces, and so provides an invaluable qualitative picture of the behavior of two contacting fractal surfaces. We now consider an elastic perfectly-plastic solid, with Young’s modulus E , Poisson’s ratio ν and yield stress σ Y , which has a surface roughness defined by Eq. (1). The solid is indented by a smooth cylinder with radius R as illustrated in Fig. 2. The cylinder is loaded by a total force (per unit out of plane length) P, to induce a contact pressure distribution p ( x) . Of course, the two surfaces actually meet only at the tips of asperities. We shall find that for an elastic-perfectly plastic solid with an ideal Weierstrass profile, the true contact consists of an infinite number of contact spots of zero size, which are all subjected to the same, uniform contact pressure. This is a purely formal result, however, and does not characterize the loading or deformation of the solid in a useful manner. We obtain a better picture by regarding the loading and deformation to consist of a series of separate scales, n = −1,0,1...∞ , as illustrated in Fig. 2. The lowest scale, n = −1 corresponds to the nominal contact. Successive terms n = 0,1,2...∞ characterize the behavior of the solid at length scales comparable to successive wavelengths in the Weierstrass profile. The fields associated with each term interact weakly through the requirement that the pressure at scale n must be statically equivalent to the pressure at scale n+1. 6 Scale -1 P R p-1 (x) a-1 Scale 0 p0 p0 (x) 2a0 λ0 Scale 1 p1 p 1 (x) 2a1 λ1 Figure 2: Idealized two-dimensional model of contact between a smooth cylinder and a surface with Weierstrass profile with widely separated scales. Successive scales interact through the condition that the nominal pressure acting on scale n is equal to the true pressure acting on scale n-1. The nominal contact area and pressure (scale n=-1) can be computed using results for smooth contacts (see e.g. Johnson 1985). It is helpful to review these results briefly. The severity of the nominal load is conveniently characterized by a dimensionless load parameter PE * / σ Y2 R . For PE* / σ Y2 R < 3.2 , the contact is elastic, and the contact width and nominal contact pressure distribution follow as 4 PR a−1 = π E* (3) 4 mean 2 2 p−1 ( x) = p− 1 1 − x / a−1 π where p−mean = P / 2 a−1 ≤ 1.4σ Y (4) 1 7 is the average nominal contact pressure. For PE * / σ Y2 R > 40 , the contact is fully plastic, and the contact width and stress distribution can be approximated using the rigid plastic slip-line field solution, which gives a−1 = P /(2 H 0σY ) (5) p−1 ( x) = H0σ Y where H 0 ≈ 2.9 is the ratio of the plane strain hardness of the solid to its tensile yield stress. In the intermediate regime 3.2 < PE* / σY2 R < 40 , the contact width and pressure distribution would need to be computed using numerical simulations. This will not be attempted here: we shall focus instead on the limiting cases of perfectly elastic and fully plastic bulk contacts. It is important to note that the nominal contact geometry introduces a natural and well defined long wavelength cut-off for the roughness spectrum. Roughness wavelengths greater than a −1 clearly do not influence the contact pressure. The roughness component with wavelength closest to a −1 should be regarded as part of the nominal contact geometry (its effect is equivalent to modifying the shape of the indenter). The first component of the profile to influence the contact pressure distribution has index m = 1 + log(Λ 0 / a−1 )/log γ , and has wavelength and amplitude λ0 = a−1 / γ g0 = G0 ( γΛ0 / a−1 ) (6) Henceforth, we will characterize the surface roughness within the area of contact with a truncated Weierstrass profile given by ( D −2) ∞ h ( x ) = g0 ∑ γ ( D −2) n n= 0 cos(2π xγ / λ0 ) n (7) where g 0 and λ0 represent the amplitude and wavelength of the first roughness component that has wavelength less than the nominal contact width. It will be important to keep in mind that while γ , D, G0 and Λ 0 are properties of the surface geometry only, g 0 and λ0 depend also on the nominal contact geometry as shown in Eq. (6). We shall return to this issue when presenting our results in Section 5. If we now examine a small region of the nominal contact, with length λ1 << dx << a−1 and centered at some point x within the contact area, we observe a sinusoidal surface with wavelength λ0 in contact with a nominally flat indenter. The indenter is subjected to a uniform nominal contact pressure p0 = p−1 ( x) , which induces a contact width 2a0 and pressure p0 ( x) on the zero-th roughness scale. We can then examine more closely a region with size λ2 << dx << λ0 within one of these contacts. Here, we observe a sinusoidal surface with wavelength λ1 , subjected to nominal contact pressure p1 = p0 ( x) , inducing a contact size a1 , and contact pressure p1( x) . This process can evidently be continued indefinitely. The goal of our analysis is to determine, for each roughness scale n, the total number of contact spots N n , the apparent size of the contact spots a n , and the apparent contact pressure distribution pn ( x ) . Calculations are complicated by the fact that the qua ntities of interest vary 8 with position on the surface. It is possible (but exceedingly cumbersome) to calculate this spatial variation. In practice, it is preferable to quantify the variation by computing distribution functions, which, for example, specify the fraction of asperity contacts with size between a n and an + dan , or the fraction of the nominal contact area that is subjected to pressure between pn and pn + dpn , for each scale n = 0...∞ . The distribution function for the pressure plays a particularly important role in subsequent calculations. For each scale of roughness, we define a dimensionless function qn ( p / σ Y ) such that qn ( p / σ Y ) ( dp /σ Y ) is the fraction of the nominal contact (which has total width 2a−1 ) that is subjected to a dimensionless pressure between pn / σ Y and ( pn + dpn ) / σ Y . As an example, the distribution function q−1 ( p / σ Y ) characterizing the nominal contact can easily be computed. The cumulative distribution Q ( p) (the proportion of the contact over which the pressure exceeds p) can be computed from the contact pressure distribution as Q−1 ( p ) = x ( p ) / a−1 . The distribution function q−1 ( p) then follows as q −1 ( p ) = − ∂Q−1 p . For example, if the nominal contact is elastic, we have Q−1 ( p) = x ( p) / a−1 = 1 − ( π p / 4 p−mean ) 1 2 (8) The distribution function follows as ξ (π / 4ξ −mean ) dQ 1 q−1 (ξ ) = −σ Y = mean dp 1 − (πξ / 4ξ −1 ) 2 2 (9) where ξ = p σ Y , ξ −mean = p−mean / σ Y . Similarly, for fully plastic contact, the distribution function 1 1 is evidently q−1 (ξ ) = δ (ξ − H 0 ) (10) where δ denotes the Dirac delta distribution and H 0 ≈ 2. 9 is the ratio of the hardness to the tensile yield strength of the solid, as before. Similar distributions will characterize the pressure distribution for each higher roughness scale. Although it is not possible to write down a simple expression relating the contact pressure pn ( x ) to qn ( p) for n ≥ 0 , Ciavarella et al (2000) show that this is not necessary. Their procedure for computing q n will be reviewed in detail in Section 4. The distribution function qn ( p) provides a convenient description of the contact pressure acting on the nth roughness scale, but is difficult to interpret physically. To provide a more direct measure of the deformation of each roughness scale, we introduce the contact size distribution function Sn ( a) defined such that dn = Nn Sn ( a) da (11) with N n the total number of contact spots at the nth scale, gives the number of contacts with size between a and a + da . Similarly, as a more direct measure of the loading applied to the contacts in the nth scale, we define the mean contact pressure for an individual contact spot as 9 p mean = a 1 p ( x ) dx 2 a −∫a (12) We then introduce the contact pressure distribution Pn ( p mean ) such that dn = Nn Pn ( p mean )dp mean (13) gives the number of contacts in the nth scale that are subjected to mean pressure between p mean and p mean + dp mean . Naturally, since there is a direct relationship between the contact size and mean contact pressure for individual contact spots, there will be a similar direct relationship between Sn ( a) and Pn ( p mean ) . Although we shall formally compute expressions for Sn ( a) and Pn ( p mean ) for our model surface, we readily concede that in view of the rather crude description of surface topography that forms the basis for our calculations, the exact distribution functions may have little practical significance. In many cases the average values of contact size and pressure for each scale of contact are sufficient. These are formally related to Sn ( a) and Pn ( p mean ) through ∞ ∞ an = ∫ S n ( a )a da pnmean = ∫ Pn ( p) p dp 0 (14) 0 In practice, however, it is simpler to compute an and distribution functions. The total contact area An = Nn 2 an and the nominal contact pressure pnnom = P / An are also of interest. pnmean directly, rather than via the (15) (16) We shall calculate these quantities using the elegant procedure developed by Ciavarella et al (2000) to model contact between elastic solids with a Weierstrass profile. A key assumption in their work is that the behavior of a particular roughness scale n at a point on the surface can be calculated from the solution to a sinusoidal surface indented by a flat, frictionless punch. Successive roughness scales interact only through the requirement that the nominal pressure pn that acts on scale n is equal to the true pressure pn−1 ( x ) for the preceding scale n-1. This assumes that no interaction occurs between the deformation fields associated with successive roughness scales. For an elastic solid, this assumption can easily be justified using Saint Venant’s principle. For elastic-plastic materials, no rigorous proof exists. We speculate, however, that provided the bulk contact (at scale n-1) consists of a single, isolated contact area, with a size much smaller than any characteristic structural dimension, the same decoupling can be applied for elastic-plastic solids. As an extreme example, consider the limiting case of a rigid-perfectly plastic solid. The slip- line field solutions for contact between a semi- infinite solid and a single indenter invariably show a rigid region under the contact area (Hill 1950; also reproduced in Johnson 1985). Consequently, the deformation associated with the first roughness scale in a Weierstrass surface (scale 0) may be embedded within the rigid region associated with the nominal contact. The behavior of the 10 first scale of roughness is therefore equivalent to that of a sinusoidal surface, with rigid boundary conditions at infinity. Our numerical solution to this problem shows that a rigid regio n forms under each contact area. Consequently, the second roughness scale can similarly be embedded within this region, and this process can be continued for all higher roughness scales. Since the deformation field constructed in this manner forms a kinematically admissible collapse mechanism, our solution should strictly be regarded as an upper bound to the collapse load, and a lower bound to the contact size and contact area. We present some further support for our calculations in the Appendix A, where the analytical calculations are compared with the predictions of full- field finite element computations for up to three roughness scales. It is important to recognize, however, that if the deformation field for the bulk contact (scale -1) is not contained, a rigid region may not form below the bulk contact in the fully plastic limit. In this case, the first scale of roughness is not contained, and this behavior will propagate upwards throughout all subsequent scales of roughness. Such unconstrained flow is particularly likely to occur in metal forming operations. 3. Indentation response of a sinusoidal surface Clearly, establishing the response of a simple sinusoidal surface to indentation loading is a key step towards solving the contact problem outlined in the preceding section. The indentation response of an elastic-perfectly plastic solid with a harmonic roughness was analyzed in detail by Gao et al (2005). Only a few of their results are needed here: these are reviewed briefly below. We focus attention on a single roughness scale with wavelength λn = λ0 / γ n (17) and amplitude g n = g 0γ ( D −2) n (18) as illustrated in Fig. 2. The substrate is an elastic-perfectly plastic solid with Young’s modulus E, Poisson’s ratio ν and yield stress σ Y . Gao et al (2005) show that the strength of the surface is conveniently characterized by a material and geometric parameter given by g E* ψn = n (19) λnσ Y with E * = E 1 − ν 2 . In the discussion to follow, the value of ψ n for the zero-th roughness scale (n=0) plays a central role in characterizing the resistanc e of the rough surface to plastic flow. In terms of this parameter, we have g0 E* ψ0 = , ψ n = ψ 0γ (D −1)n (20) λ0 σ Y Note that the resistance to plastic flow varies in inverse proportion to ψ : a rough surface with a low yield stress has a high value of ψ , while a smooth, hard surface would have a low ψ . In addition, note that ψ n increases with n, so we anticipate that roughness scales with low wavelength will be elastic, but all scales beyond a critical value of n will deform plastically. ( ) 11 The solid is indented by a flat, rigid surface, under the action of a nominal pressure pn , as illustrated in Fig. 2. The loading induces a contact width 2 an , pressure distribution pn ( x ) and mean pressure pnmean given by functional relationships p g 2 an / λn = A n ,ψ n , n λn σY x p pn ( x) g = Φ , n ,ψ n , n σY λn λn σ Y (21) p pnmean g = Φ mean n ,ψ n , n σY λn σY Here, we have shown that formally that A, Φ and Φ mean depend on both ψ n and g n / λn . In fact, if the surface remains elastic, exact calculations show that the solution depends only on ψ n and is independent of g n / λn . If the surface deforms plastically, numerical simulations show that the functions depend only weakly on g n / λn . It would not be difficult to include this variation in our computations, but for simplicity we have chosen to neglect it. 0.5 Interaction between neighboring asperities 0.4 0.3 0.2 a/λ Fully-Plastic Elastic-Plastic 0.1 10 -1 Elastic 100 10 1 ψ=E*g/λσY Fig. 3: A map illustrating the regimes of behavior of an elastic-perfectly plastic solid with sinusoidal surface when indented by a rigid flat punch. Finally, we must also compute the pressure distribution function qn ( p / σ Y ) , which is defined so that qn ( p / σY )( dp /σ Y ) is the fraction of the sinusoidal surface that is subjected to normalized 12 pressure between pn / σ Y and ( pn + dpn ) / σ Y . This distribution function can be determined from the contact pressure, following the procedure given for the nominal contact in equations (8) and (9). The result will be specified by the functional relationship p p pn qn (22) = q , ,ψ n σY σY σY We have calculated approximations to the functions A, Φ , Φ mean and q by fitting curves to the numerical simulations reported in Gao et al (2005). To specify the results, it is necessary to distinguish between four general regimes of behavior: (i) Elastic; (ii) Elastic-plastic; (iii) Fully plastic, with no interaction between neighboring asperities and (iv) Fully plastic, with interaction between neighboring asperities. The distinction between these regimes is discussed in detail in Gao et al (2005), and will be summarized only briefly here. Elastic behavior is observed at low values of applied load; while the elastic-plastic regime occurs for loads that slightly exceed the elastic limit. For sufficiently large values of load, the contact pressure distribution becomes approximately uniform. This condition defines fully plastic behavior. In the fully plastic regime, if the contact spots are small compared with the wavelength of the roughness, the asperity contacts are isolated and the contact pressure is equal to the plane strain hardness of the solid (≈ 2.9σ Y ). If the contact width approaches the roughness wavelength, the plastic zones under neighboring contact spots begin to interact. Under these conditions, the contact pressure remains approximately uniform, but its magnitude increases with applied load, and approaches values up to twice the material hardness as 2 an → λn . Gao et al (2005) show that these four regimes of behavior can be conveniently displayed on a map with a / λ and ψ as parameters. The map is shown in Fig. 3: the elastic limit is given approximately by the condition ψ a / λ = 0.1 , while ψ a / λ = 1 defines the transition to fully plastic behavior. In the sections below, we give expressions for the functions defined in Eqs (21) and (22) in each of the four regimes. 3.1 Elastic Regime 0 < ξ < πψ sin 2 (π /10ψ ) In the elastic regime, A, Φ and Φ mean are given by the Westergaard solution (Westergaard 1939) 2 −1 ξ , ξ < πψ sin A(ξ ,ψ ) = π πψ (23) 1, ξ ≥ πψ 1/2 2ξ cos(πη ) sin 2 (π A /2) − sin 2 (πη )} ξ < πψ { 2 Φ (η ,ξ ,ψ ) = sin (π A /2) (24) ξ + πψ cos(2πη ) ξ > πψ Φ mean πξ −1 (ξ ,ψ ) = 2sin ( ξ /πψ ) ξ 13 ξ ≤ πψ ξ ≥ πψ (25) 2 2 q (η ,ξ ,ψ ) = (πψ − ξ ) + η 0 { η /π }{2ξ (πψ − ξ + ( πψ − ξ ) 2 +η 2 ) } −η 1/2 η < 2 πψξ 2 (26) η > 2 πψξ Note that the elastic solution depends only on the normalized nominal pressure ξ = p / σY and surface property ψ , and is independent of g n / λn . The critical load ξ = p / σ Y = πψ corresponds to full contact ( 2 a / λ = 1 ). This suggests a second physical interpretation for the parameter ψ : a surface with a low value for ψ can easily be flattened. This solution is valid below the elastic limit: Gao et al (2005) suggest that a convenient (but approximate) estimate for the yield point is given by the condition ψ a / λ < 0.1 , which requires that ξ = pn / σ Y < πψ sin 2 (π /10ψ ) . 3.2 Elastic-Plastic regime πψ sin 2 (π /10ψ ) < ξ < 2 H 0 / ψ The elastic-plastic regime of behavior is observed for applied loads that slightly exceed yield. Under these conditions, the elastic and plastic strains under the indenter are comparable, and exact expressions for A, Φ and Φ mean cannot be found. Detailed numerical results are given in Gao et al (2005). For the problem at hand, however, we find that at most a single scale of roughness is in the elastic-plastic regime, so it is not necessary to characterize this regime in detail. Consequently, we have constructed simple functions for A, Φ and Φ mean by using a selfconsistent interpolation between the elastic and fully plastic regimes. The interpolation is constructed to predict correctly: (i) the elastic limit, (ii) the critical load to transition to fully plastic behavior; (iii) the contact pressure distribution at the elastic limit; and (iv) the contact pressure at the fully plastic limit. In addition, we interpolate the contact size between the elastic and fully plastic limits as follows (ξ − ξ p ) A + (ξ − ξ e )0.9 A A(ξ ,ψ ) = (27) (ξe − ξ p ) e (ξ p − ξ e )0.9 p where π 2H 0 ξ e = πψ sin 2 ξp = H 0 = 2.8 ψ (28) 10ψ Ae = 1/(5ψ ) Ap = 2 / ψ define dimensionless values of the critical nominal pressure at yield; the critical nominal pressure at full plasticity, and the normalized contact widths at yield and the fully plastic limit, respectively. The contact pressure distribution is approximated as 1/2 2α (ξ ,ψ )cos(πη ) Φ (η ,ξ ,ψ ) = sin 2 (π A /2) − sin 2 (πη )} + β (ξ ,ψ ) (29) { 2 sin (π A /2) where the scaling factors α and β are introduced to ensure that the contact pressure has the correct average and maximum values. This requires 14 ( Aξ α (ξ ,ψ ) = max − ξ ) sin(π A /2) 2ξ − ξ max sin(π A /2) (30) 2 A − sin(π A / 2 ) 2 A − sin(π A /2) where the maximum contact pressure is interpolated between its values at yield and full plasticity as ( A − Ap ) ξ + H 0 ( A − A0 ) ξ max = (31) ( A0 − Ap ) A ( Ap − A0 ) β (ξ ,ψ ) = where A0 = ( Ae ( H 0 − ξe / Ae ) + Ap ξe / Ae − 2 πψξ e ) H0 − 2 πψξ e is a constant chosen to ensure that the peak contact pressure has the correct value ξ max = 2 πψξe at the elastic limit ( A = Ae , ξ = ξ e ). The mean contact pressure follows as Φ mean (ξ ,ψ ) = ξ / A In addition, the contact pressure distribution function may be computed as (η − β H 0 )/ π q (η ,ξ ψ , )= 1/2 {ω 2 + (η − β H ) 2} 2α ω + ω 2 + (η − β H ) 2 − (η − β H ) 2 0 0 0 ω = α cot 2 (π A /2) { ( ) } (32) (33) (34) 3.3 Fully plastic regime – non-interacting asperities 2 H 0 /ψ < ξ < 2H 0 / 3 In the fully plastic regime, the plastic strains under the indenter greatly exceed the elastic strains, and the stress fields approach the rigid-plastic limit (given e.g. by slip line field solutions). Here, it is necessary to distinguish between two distinct types of behavior. As long as the contact spots are small (numerical results suggest 2 a / λ < 0.67 ), there is no interaction between neighboring asperity contacts, and A, Π and Π mean can be estimated from slip- line field solutions (or numerical simulations) for a flat punch indenting a rigid-plastic half-space as A(ξ ) = ξ / H0 Φ(x ) = H 0 ( x / a < 1) (35) Φ mean = H0 where H 0 ≈ 2.9 is the hardness of the material normalized by its tensile yield stress. Numerical simulations and slip- line field solutions show that H 0 varies weakly with g / λ , but this variation is insignificant for our present purposes. Note that in this regime the contact pressure distribution is perfectly uniform, and the solution is independent of ψ . Finally, the pressure distribution function follows as q (η ,ξ ,ψ ) = A(ξ ,ψ )δ (η − H0 ) (36) where δ ( x ) denotes the Dirac delta distribution. 15 3.4 Fully plastic regime – interacting asperities 2 H 0 / 3 < ξ When the size of the contact spots becomes comparable to the wavelength, the plastic strain fields associated with neighboring asperities begin to interact. Numerical simulations show that in this regime, the contact pressure distribution remains approximately uniform, but the magnitude of the pressure increases substantially, and in some cases reaches twice the hardness of the material. This behavior has a dramatic influence on the behavior of fine roughness scales. As in Section 3.2 we have devised approximate analytical approximations to numerical simulations to simplify our calculations. It is simplest to specify the relationships between ξ , ψ , A and Φ mean parametrically as Φ mean = H 0 + ( H 1(ψ ) − H 0 )( 3 A − 2 ) 3 (37) Approximation Numerical 5 Mean contact pressure pmean/σY Normalized contact pressure p/σY ξ = AΦ mean (38) where H1(ψ ) is the limiting value of the normalized contact pressure p ( x ) / σ Y at full contact. The following function gives an approximate fit to the results shown in Fig. 8 of Gao et al (2004) H 1(ψ ) = 5.8 − 60/(16 +ψ 3 / 2 ) (39) It is straightforward to solve these equations to determine the functions defined in Eqs (21). The results involve the solution to a cubic equation, however, and so are too lengthy to be recorded here. Finally, the contact pressure distribution function in the interacting asperity regime has the same form as that given in Eq. (36). ψ=50 ψ=10 4 ψ=3.5 3 ψ=2 2 ψ=1.0 1 0 0.2 0.4 0.6 0.8 1 Approximation Numerical 5 ψ=50 ψ=10 4 ψ=3.5 3 ψ=2 2 ψ=1.0 1 0 0.2 0.4 0.6 0.8 1 Contact fraction 2a/λ Contact fraction 2a/λ Fig. 4: The relationship between nominal pressure and mean pressure and contact fraction for a sinusoidal surface in contact with a rigid, flat surface. The symbols show the results of finite element comp utations, while the solid lines show the approximate analytical expressions used in our rough surface model. 16 Our approximate curves are compared with the results of finite element simulations in Fig. 4, which shows a parametric plot of the nominal and mean contact pressures as a function of the contact size as the load increases, for various values of ψ . 4. Contact mechanics for a general Weierstrass profile We proceed to analyze contact of the Weierstrass profile. Our analysis follows exactly the work of Ciavarella et al (2000) and so will be reviewed only briefly. We will first present a fully general procedure for computing distribution functions for contact size and pressure for each scale. Subsequently, we suggest a simple approximation that can be used to obtain closed- form estimates for the number of contact spots, their size, and mean pressure for each scale of loading. 4.1 Results for general surface contact Although it is difficult to compute the spatial contact pressure distribution for the Weierstrass surface, it is straightforward to determine the probability distribution function qn ( p / σ Y ) defined in Section 3. The calculation begins by computing the distribution function q−1 ( p / σY )( dp /σ Y ) , which specifies the proportion of the nominal contact area that is subjected to a pressure between p σ Y and ( p + dp ) σ Y . The result is given in Eq. (9). The distribution q0 ( p ) for the zero-th roughness scale may therefore be computed as follows. Recall that we introduced the dimensionless function q, such that q ( p / σ Y , ξ ,ψ )( dp / σ Y ) specifies the proportio n of the sinusoidal surface that is subjected to pressure between p σ Y and ( p + dp ) σ Y when loaded by a normalized nominal pressure ξ . The zeroth roughness scale is subjected to a distribution of nominal pressure, as quantified by q−1 ( p / σ Y ) . Summing over all possible values of the contact pressure for the nominal contact, we find that ∞ q0 ( p / σY ) = ∫ q ( p / σY , ξ ,ψ 0 ) q−1 (ξ ) dξ (40) 0 The upper limit on the integral here is purely formal, of course: the nominal contact pressure does not exceed p−1 = 2 P / π a−1 , so q−1 (ξ ) = 0 for ξ > 2P / π a−1σ Y . This result can be applied recursively, to see that ∞ qn ( p / σ Y ) = ∫ q ( p / σ Y , ξ ,ψ n ) qn−1 (ξ ) dξ (41) 0 Once the hierarchy of distribution functions qn ( p / σ Y ) has been computed, various quantities of interest can be deduced. For example, to calculate the total (apparent) area of contact for the nth scale, note that the area of contact for one wavelength of the nth scale subjected to pressure p is 2 an = λn A ( p σ Y ,ψ n ) . In addition, there are 2 a−1 / λn such wavelengths within the nominal area of contact. The total contact area therefore follows as ∞ An = 2a−1 ∫ A (ξ ,ψ n ) qn−1 (ξ )d ξ 0 17 (42) The total number of contact spots at the n -th scale within the nominal contact is A N n = n−1 λn whereupon the average size of the contacts at the nth scale follows as A A 2 an = n = λ n n Nn An−1 The average contact pressure can be found by a similar approach, with the result pnmean = (43) (44) ∞ 2a −1 Φ mean (ξ ,ψ n )qn −1 (ξ ) dξ An−1 ∫ (45) 0 Finally, the nominal contact pressure for the nth scale follows as mean n p ∞ P mean = = p−1 ∫ A ( ξ ,ψ n ) qn−1 (ξ ) d ξ An 0 −1 (46) We can also find expressions for the contact size and mean pressure distribution functions Sn ( a) and Pn ( p mean ) defined in Eqs. (11) and (13). To compute the distribution function for the contact size, denote the pressure required to induce a contact size a is given by the functional relationship p / σ Y = A−1(2a / λ ,ψ ) , where A−1 denotes the inverse of the function A defined in equation (21). Given the distribution function for the contact pressure at (n − 1) -th scale, we can easily compute the probability distribution function of the contact fraction for scale n as follows. For a single contact, 2 an / λn = A ( pn−1 σ Y ,ψ n ) , so that let p n−1 / σ Y = A−1 ( 2an / λn ,ψ n ) . The probability of finding a contact fraction in between 2 an / λn and 2(an + dan ) / λn is S n (2a n λ n ) = dA −1 q n−1 A −1 (2a n λn ,ψ n ) d (2a n λ n ) ( ) (47) 4.2 Simplified analysis – the uniform contact pressure assumption The results listed in the preceding section can be simplified dramatically if the contact pressure distribution for each scale in the contact (including the nominal contact area) is uniform. This is an excellent approximation when each scale of contact is fully plastic. At lighter loads, the nominal contact, and some long-wavelength roughness scales, will remain elastic, and for these scales the distribution of contact pressure is not uniform. Surprisingly, however, we have found that averaged quantities such as the total area of contact, the average contact size, and the average contact pressure can be computed remarkably accurately even for elastic surfaces by replacing this non-uniform contact pressure distribution with a statically equivalent, uniform pressure. Proceeding along these lines, therefore, we suppose that the entire nominal contact area is subjected to the same contact pressure. For this approximation, the distribution function q−1 ( p / σ Y ) , which specifies the fraction of the nominal contact area (scale -1) subjected to pressure between p σ Y and ( p + dp ) σ Y , must therefore be given by 18 q−1 (ξ ) = δ (ξ − p−mean /σY ) 1 (48) where δ ( x) denotes the Dirac delta distribution and p = P / 2 a−1 . The distribution function for the zero-th scale of roughness follows from Eq. (41) as mean −1 ∞ q0 ( p / σ Y ) = ∫ q( p / σ Y , ξ ,ψ 0 )δ (ξ − p−mean / σ Y ) d ξ = q( p / σ Y , p−mean / σ Y ,ψ 0 ) 1 1 (49) 0 Recall now that the function q ( p / σY , p−mean / σ Y , ψ 0 ) specifies the fraction of a sinusoidal surface 1 that is subjected to pressure between p σ Y and ( p + dp ) σ Y , when indented by a nominal contact pressure p−mean . If the pressure is approximated by a statically equivalent uniform 1 distribution, with magnitude p mean , it follows that ( q ( p / σ Y , p−mean / σ Y ,ψ ) = A( p−mean / σY ,ψ )δ p / σ Y − Φ mean ( p−mean / σ Y ,ψ , ) 1 1 1 ) (50) where the functions A and Φ mean are defined in equations (21), and given specific forms in Eqs. (23)--(39). Repeating this procedure for subsequent roughness scales gives the following result for the distribution functions qn ( p) { } qn ( p / σ Y ) = Π A( p j −1 / σY ,ψ j ) δ ( p / σ Y − Φ n mean j =0 mean ( pn−1 / σY ,ψ n ) ) mean Here, the mean contact pressure at each scale can be determined from pnmean / σ Y = Φ mean ( pnmean −1 / σ Y ,ψ n ) (51) (52) where the sequence is started with p−mean given in Eq.(48). 1 Given pnmean for each scale, and the distribution functions qn ( p / σ Y ) , we can determine the remaining quantities of interest using Eqs. (42)-(46). The total area of contact at the nth scale follows as ∞ n An = 2a−1 ∫ A (ξ ,ψ n ) qn −1 (ξ )d ξ = 2a −1 Π A( p mean j −1 / σ Y , ψ j ) j= 0 0 (53) where the function A is defined in Eq. (21). The number of contacts follows as n −1 N n = 2( a−1 / λn ) Π A( p mean j −1 / σ Y ,ψ j ) (54) j= 0 and the average contact size is given by 2 an = λn A( pnmean −1 / σ Y ,ψ n ) (55) Finally, equation (46) provides an additional way to compute p { mean n } as −1 n pnmean = p−mean Π A( p jmean 1 −1 / σ Y ,ψ j ) j =0 (56) from which we also recover the expected relationship between the mean pressure at successive scales mean pnmean = pnmean (57) −1 A( pn −1 / σ Y , ψ n ) 19 4.3 Approximate analytical expressions for average contact size, pressure, number of contacts, and total contact area for elastic contact If the contacts remain elastic, the procedure outlined in the preceding section can be used to derive simple analytical expressions for the average contact size, average contact pressure. Under these conditions we may approximate 2 ξ 2 ξ A(ξ ,ψ ) = sin−1 ≈ π πψ π πψ Φ mean (58) πξ π (ξ ,ψ ) = ≈ πξψ −1 2sin ( ξ /πψ ) 2 whereupon Eq. (52) reduces to pnmean π pnmean π −1 = πψ n = σY 2 σY 2 This recursion relation is satisfied by pnmean −1 πψ 0 γ ( D−1) n σY 2 −2− n mean 2− n −1 pnmean π ( D −1) n p−1 = γ D −1 ) πψ 0γ ( σY 2 σ Y πψ 0 The average contact size may be estimated using Eq. (55) as − 2− n p π 2 an = λn γ D−1 ) ( 2 σ Y πψ 0 Similarly, the total area of contact at the nth scale follows as 2− n −1 mean −1 2 2− ( n+1) (60) −(n +1) (61) 1− 2 −n (59) 2 −2 mean − ( D −1)( n−1+ 2− n ) p π −1 An = 2a−1 γ 2 σ Y πψ 0 The total number of contacts at the nth scale can be estimated as N n = An−1γ (n+1) (2a−1 ) For large n, these results may be approximated by their asymptotes −( n +1) (62) (63) 2 pnmean π = πψ 0 γ ( D−1)n σY 2 2 < an >= λn γ (1− D) (64) (65) −2 mean π −2 P −( D −1)n p−1 π An = 2a−1 γ = 2 σY πψ 0 2 σ Y πψ n (66) −2 mean π D− ( D−2 ) n p−1 Nn = γ 2 σ Y πψ 0 (67) As expected, equation (66) shows that the total contact area at the nth scale (for large n) is proportional to the contact load P and independent of the nominal contact size a −1 or the indenter geometry (other than through its effect of setting the wavelength of the zeroth roughness 20 scale). This result also provides an opportunity to check the predictions obtained using the constant pressure approximation (Sect 4.2) with the exact solution. Ciavarella et al. (2000) find that An = 0.391P /(σ Y πψ n ) , giving a coefficient that is within 4% of the approximate value in Eq. (66). A more detailed comparison of the approximate and exact calculations in section 5 shows similar agreement under all conditions. Note also that Eq. (64) shows that the mean contact pressure acting on each roughness scale becomes progressively more severe as n increases. Consequently, for an elastic-plastic solid, the short wavelength roughness scales will always deform plastically. We also see that the formal limit n → ∞ for a perfectly elastic solid consists of an infinite number of contact spots, with zero size, which are subjected to infinite contact pressure. The total true contact area vanishes. 4.4 Deformation and loading for the most heavily loaded region at each scale (elastic contact) To calculate the conditions necessary to initiate yield at each successive scale of roughness, or to identify conditions that will just begin to bring a roughness scale into complete contact with the indenter, it is of interest to determine the contact size and contact pressure for the most heavily loaded asperities at each successive scale of roughness. We assume that all roughness scales, as well as the nominal contact, remain elastic; and assume also that γ ( D−1) > 2 . Equation (3) shows that the maximum contact pressure exerted by the nominal contact is p−max PE* 1 = σY π Rσ Y2 (68) The most severely loaded asperities on the zeroth roughness scale are therefore subjected to a nominal contact pressure p0 = p−max 1 . Equation (24) gives the maximum contact pressure for a sinusoidal surface that is subjected to a normalized nominal pressure ξ = p / σY as whence max 0 p the maximum p max (ξ ,ψ ) / σ Y = 2 πξψ contact pressure on the zeroth roughness (69) scale is / σY = 2 π p ψ 0 / σY . Similarly, the maximum contact pressure on successive scales of max −1 roughness follows as pnmax / σY = 2 π pnmax −1 ψ n / σ Y . The recursion relation is satisfied by 2− ( n+1) max 2− n −1 pnmax 2− 2− n ( D −1) n p−1 = ( 2) γ D −1 ) πψ 0 γ (70) ( σY σ Y πψ 0 Other than a numerical factor, this result is identical to eq (60), which estimates the average contact pressure on each roughness scale. However, Eq. (70) does not assume isolated Hertztype contacts and so is valid for all values of ψ 0 , provided that no scale reaches full contact. 21 4.5 A graphical approach to calculating contact size, pressure and area for roughness scales in the fully plastic regime If the bulk contact is fully plastic (which requires PE * / σ Y2 R > 40 ), and the surface roughness is such that ψ 0 > 16 , then the procedure outlined in section 4.2 gives exact results, since all roughness scales are in the fully plastic regime, so the contact pressure distribution at each scale is uniform. In addition, under these conditions, the method can be illustrated using a simple graphical procedure, originally suggested by Childs (1973, 1977) in a discussion of the persistence of asperities under contact loading. We begin by simplifying the behavior of the sinusoidal surface for the fully plastic regime. Consider a single scale of roughness with wavelength λ , which is indented by a rigid platen. The surface is subjected to a nominal contact pressure (force per wavelength) p , which induces a contact size a. Each contact spot is subjected to a uniform contact pressure with magnitude p mean . For values of ψ 0 > 16 , the relationship between the mean normalized contact pressure and contact size for the fully plastic regimes given in Eqs. (35)-(39) can be approximated by A< 2/3 H0 p mean / σ Y = Φ mean = (71) 3 A > 2/3 H0 + ( H1 (ψ → ∞ ) − H 0 )( 3 A − 2 ) where A = 2a / λ , H 0 ≈ 2.9 and H1(ψ → ∞ ) = 5.8 . In addition, the normalized nominal contact Contact pressure p/σY pressure follows as ξ = p / σ Y = AΦ mean . Both these functions are plotted in Fig. 5. 5 4 3 mean pmean/σ Y p1 /σ Y = p 2/σY pmean /σ Y = p 1/σY 0 mean p -1 /σY = p0/σ Y 2 p/σY 2a2/λ 2a 1/λ 2a0/λ 1 0 0.2 0.4 0.6 0.8 1 Contact fraction 2a/λ Fig. 5: A graphical technique for calculating the contact size and contact pressure for successive scales of a rigid-perfectly plastic surface with Weierstrass profile. The solid lines show the mean pressure p mean / σ Y and nominal pressure p / σ Y for a rigid plastic sinusoidal surface as a function of contact fraction. Successive scales of roughness are related by the condition that p n / σ Y = pnmean −1 / σ Y . 22 Next, we apply the iterative scheme outlined in Section 4.2 for this simplified situation. Since the nominal contact (scale -1) is in the fully plastic regime, the nominal contact pressure is uniform, and has magnitude p−1 / σY = 2.9 . This implies that the nominal pressure exerted on asperities in the zero-th roughness scale is p0 / σ Y = 2.9 . Fig. 5 can be used to deduce the normalized contact size 2 a0 / λ0 corresponding to this nominal contact pressure; whence the true contact pressure p0mean / σ Y can also be found. The same procedure may be used to progress to higher scales of roughness: the condition pnmean / σ Y = p n+1 / σ Y determines the normalized contact size an +1 / λn+1 at each successive roughness scale, whereupon pnmean +1 / σ Y can be deduced. In addition, the apparent total contact area at each scale follows as n 2a j An = 2a−1∏ (72) λ j =0 j In particular, the total contact area in the limit n → ∞ is H0 P Lim An = 2a−1 = (73) n →∞ H 1 (ψ → ∞ ) 5.8σ Y Thus, we find that the contact area is again proportional to the normal load, but instead of converging to zero as n → ∞ , the total contact area converges to a finite value. A similar conclusion was reached by Willner (2004) using a more sophisticated series of numerical simulations. His analyses did not account for interactions between neighboring contact spots, however, and consequently gave a limiting area of Lim An = P H 0σ Y . More n→ ∞ recently, Pei et al (2005) have conducted an extensive study of the indentation response of an elastic-plastic solid with a nominally fractal surface roughness using three-dimensional finite element computations. They found that the true area of contact is a function of the ratio of yield stress to Young’s modulus for the solid: this behavior is most likely due to an artificial truncation of the roughness spectrum resulting from the resolution of the finite element mesh. For sufficiently low values of σ Y / E , however, they found that the true contact area was related to the load by Atrue = P / 6σY , in excellent agreement with our approximate estimate. The convergence of contact size and contact area to their limiting values are illustrated in Fig. 6. Since the normalized contact size 2 an / λn → 1 as n → ∞ , while Lim An / 2a−1 = H 0 / H1 (ψ → ∞ ) , n →∞ we have plotted 2 an / λn − 1 and An /(2a−1 ) − H0 / H1(ψ → ∞) as functions of n. Evidently, both contact size and total contact area converge to their asymptotic values exponentially: indeed, it is straightforward to show that for sufficiently large n , the contact size and total contact area may be approximated by 2 an / λn ≈ 1 − 0.1e −0.2n (74) An /(2a−1) ≈ H 0 (1 − 0.45e −0.2 n ) / H1 (ψ → ∞) 23 Difference from asymptote Total contact area An /a-1 -H0/H1 (ψ→∞ ) -1 10 10-2 10-3 0 Contact fraction 1-2an/λ n 5 10 15 20 Roughness scale n Fig 6: Illustrating the convergence of total contact area and contact fraction towards the ir rigid plastic asymptotes with roughness scale n. These results can also be used to deduce the asymptotic behavior of fine-scale roughness on an elastic-plastic solid. We saw in Section 4.3 that even if the bulk contact and the first few roughness scales remain below yield, the contact pressure on an elastic surface increases for progressively higher scales of roughness. Consequently, there will always be a critical scale of roughness that reaches yield. By the same argument, there will always be a critical roughness scale which reaches the fully plastic regime. For higher (shorter wavelength) scales of roughness the graphical method illustrated in Fig 5 can again be used to deduce the contact pressure and contact fraction. For sufficiently large n, the true contact area and contact fraction will again approach their asymptotes as shown in Fig 6. It is interesting to note the important role played by interactions between neighboring asperities in determining the asymptotic behavior of fine scale roughness. Without this interaction, the contact pressure on each scale of roughness is bounded by the material hardness (i.e. the limiting pressure under an isolated indenter). For a rigid plastic contact, therefore, all roughness scales (including n=0) must reach full contact an / λn = 1 to support the load. For a more general elastic-plastic contact, the total true contact area and the true contact spot size would be determined by the first roughness scale to reach the fully plastic limit, since all shorter wavelength scales of roughness must be at full contact to support the contact pressure. The true contact spot size could then be computed unambiguously. This, in fact, is the simple picture of plastic contact between two rough surfaces proposed by Tabor (1951), although his focus was on computing the total area of contact, and did not address the question of the contact spot size. 24 5. Results We now discuss in detail the predictions of the calculations outlined in the preceding sections. We suppose that the power spectral density S (ω ) has been measured for some (two dimensional) surface of interest, as illustrated in Fig. 2. To extract the fractal parameters Λ 0 , G0 , D for the surface, we fit a straight line through the high frequency portion of the spectrum, and find D by noting that on a log-log plot the line has slope 2 D − 5 . The value of Λ 0 can be chosen arbitrarily, as long as Λ 0 exceeds the expected width of the nominal contact. The value of G0 should then be chosen so that the point ( Λ 0 , G0 ) is consistent with the discrete approximation to the PSD, which requires S (ω = 2π Λ 0 ) = G 02 Λ 0 2 log (γ ) . In addition, an arbitrary value may be used for γ : Majumdar and Bhushan suggest that γ = 1.4 produces the most realistic looking surfaces, but our calculations are strictly only valid for γ >> 1 , so choosing γ = 5 or greater is more appropriate. We now suppose that the surface is indented by a cylinder of radius R, which is subjected to a load P per unit length out of plane, as illustrated in Fig. 1. It is convenient to characterize the applied load, surface geometry and material properties using three dimensionless groups G0E * γΛ02 PE * Σ= Ψ = Κ = (75) RσY2 Λ 0σ Y G0 R Here, Σ is a dimensionless load parameter, and is proportional to the mean nominal contact pressure, through Eq. (4). The parameters Ψ and Κ together characterize the surface roughness and its resistance to plastic flow. As a preliminary step, we express several parameters of interest in terms of Σ , Ψ and Κ . If the nominal contact is elastic ( Σ < 3.2 ) the contact size is a−1 / Λ0 = (2/ π ) γΣ /(ΨK ) , while for a fully plastic nominal contact ( Σ > 40 ) a−1 / Λ0 = γΣ 2 /(2.8Ψ K ) . The wavelength of the zeroth roughness scale (see Fig. 1) follows as λ0 = a−1 / γ , so we find that the plasticity index ψ 0 (defined in Eq. (19)) for the zeroth roughness scale is given by Ψ D π Κ / 2Σ D−1 (elastic nominal contact Σ < 3.2) ψ0 = (76) D −1 Ψ D ( 2.8Κ / Σ 2 ) (plastic nominal contact Σ > 40) ( ) 5.1 Conditions for inducing yield Controlling plastic flow under contact loading is a central issue in designing failure resistant surfaces for tribological applications. With this motivation, we begin by estimating the conditions necessary to initiate yield in our idealized model of a surface with fractal roughness. The general nature of the contact between a smooth cylinder and the Weierstrass surface is illustrated in Fig. 2. As discussed in detail in section 2, we focus attention on each successive scale of roughness under the contact in turn. Strictly, our calculations show that short 25 wavelength roughness scales will always deform plastically, so the surface does not have an elastic limit. Instead, our goal is to calculate the number of roughness scales that remain elastic, as a function of the load (parameterized by Σ ) and surface roughness and material properties, parameterized by Ψ and Κ . We assume for simplicity that the nominal contact remains elastic, which requires Σ ≤ 3.2 (Johnson, 1985). The most heavily loaded region of the zeroth scale of roughness occurs at the center of the nominal contact, and is subjected to a normalized contact pressure p0 / σ Y = p−max 1 / σ Y = Σ / π . The elastic solution for indentation of a sinusoidal surface (Section 3) shows that the zeroth roughness scale will behave in one of two ways, depending on the value of ψ 0 . For ψ 0 < 0.2 , the zeroth roughness scale will remain elastic up to full contact, which occurs at a critical load p0 / σ Y = Σ / π = πψ 0 . For ψ 0 > 0.2 , the zeroth scale of roughness will reach yield before full contact. The critical nominal contact pressure at yield is given by p0 / σ Y < πψ 0 sin 2 (π /10ψ 0 ) . Nominal contact yields 10 -2 Sc ale e las -3 10 10 Scale 1 elastic -1 tic 2 el a sti c 10 Sca l e3 Load factor Σ 10 All roughness scales plastic 0 -2 Scale 0 elastic 10 -1 0 10 Normalized roughness ΨDΚD-1 10 1 Fig. 7: A map showing the behavior of successive roughness scales during indentation by a smooth cylindrical indenter. Dashed lines show conditions necessary to bring each scale to full contact; solid lines mark transitions from elastic to plastic behavior. Similar arguments can be applied to each successive scale of roughness n = 1,2,3...∞ . The maximum nominal contact pressure acting on each roughness scale can be computed from eq max 2 (70); the criteria pnmax −1 / σ Y = πψ n and pn −1 / σ Y < πψ n sin (π /10ψ n ) give the conditions for full contact and yielding on the nth scale of roughness, respectively. The results of these calculations are summarized in Fig 7, which shows the critical normalized load Σ required to (i) just bring each roughness scale to full contact (shown as dashed lines), and (ii) just cause yield in each roughness scale (shown as solid lines). The results depend (weakly) on D and γ , and are shown for the particular case D = 1.5, γ = 5 . 26 At extremely low values of applied load, only the zeroth scale of roughness deforms elastically; all shorter wavelength scales n = 1,2..∞ are plastic. As the load is increased, however, additional roughness scales can transition to elastic behavior. This apparently contradictory behavior is readily explained by recalling that the cont act size increases with load. While the contact is small, all roughness scales with wavelength smaller than the contact size have large amplitude compared with their wavelength, and therefore readily deform plastically. As the load is increased, the nominal contact spreads, and therefore new roughness scales that come under the influence of the indenter have a lower ratio of roughness amplitude to wavelength, and so may remain elastic, and eventually reach full contact. The parameter Ψ DΚ D −1 evidently plays the role of a plasticity index for the surface: a contact with a low value of Ψ DΚ D −1 has a large number of roughness scales that deform elastically up to the load where the nominal contact yields, while a contact with high Ψ DΚ D −1 will experience extensive plastic flow even at modest loads. 0 Plastic zone depth dΨΚ/Λ0 10 ΨDΚD-1 = 1 -1 10 10-2 Ψ Κ D -3 10 D-1 = 0.2 10-4 ΨD ΚD-1 = 0.06 -5 10 -6 10 10 ΨD ΚD-1 = 0.02 -6 10 -5 10 -4 10 -3 10 -2 Load Factor Σ 10 -1 10 0 Fig 8: The depth of the plastically deformed layer induced by indenting a Weierstrass surface with a rigid cylinder. For practical purposes it is of particular interest to estimate the depth of the plastically deformed region under the contact. As a rough guide, the depth of the plastically deformed region can be taken to be one half the wavelength of the first roughness scale to reach yield, since longer wavelengths remain below yield, while all shorter wavelengths will exceed yield. This approximation gives d yield / Λ 0 ≈ a− 1 /(2Λ 0γ k +1) = Σ /(γ k ΨΚ π ) , where k is the number of roughness scales that remain in the elastic regime, which may be estimated from Fig 7. The normalized depth of the plastically deformed region as a function of the load parameter Σ is shown in Fig 8, for various values of the plasticity index Ψ DΚ D −1 , for D = 1.5, γ = 5 . We observe three general regimes of behavior: for large values of Ψ DΚ D −1 , only the zeroth 27 roughness scale remains elastic at light loads, so the depth of the deformed layer is proportional to Σ . For Ψ DΚ D −1 > 0.25 , the zeroth roughness scale yields at large loads, which results in a sudden increase in the depth of the deformed layer. For 0.08 < Ψ DΚ D−1 < 0.15 , the zeroth scale remains elastic up to full contact, while all remaining scales yield – in this case there depth of the deformed layer is proportional to Σ over the full range. Finally, for Ψ DΚ D −1 < 0.08 , the one or more roughness scales transition to elastic behavior at sufficiently large loads. Under these conditions the depth of the deformed layer is proportional to the load parameter Σ at light loads, but is independent of Σ above a critical load. 0 ela c s ti 10-1 0 10-2 al e -3 Sc 10 ic elas t -2 1 Scale 10 las tic -1 e Scale 2 10 All roughness scales plastic Scale 3 elas tic Load factor p-1/σY 10 100 101 Normalized roughness ψ0 Fig. 9: A map showing the behavior of successive roughness scales under a uniform nominal pressure p− 1 . Dashed lines show conditions necessary to bring each scale to full contact; solid lines mark transitions from elastic to plastic behavior. For comparison, we have also calculated the conditions necessary to induce yield in a rough surface with truncated Weierstrass profile given in Eq. (7), subjected to a uniform nominal pressure p− 1 . In this case, we characterize the surface roughness by ψ 0 , defined in Eq. (20), and characterize the load using the dimensionless ratio p −1 / σY . The critical load required to initiate yield in successive scales of roughness is shown as a function of ψ 0 in Fig. 9, for D = 1.5 and γ = 5 . The results resemble the conditions necessary to initiate yield under a cylindrical indenter, except that in this case the zero-th roughness scale has a fixed wavelength and amplitude, and therefore roughness scales do not transition from plastic to elastic behavior. It is interesting to note that the yield criterion for scales n=4 and above is essentially independent of nominal load, and is determined primarily by the roughness amplitude of the zeroth roughness scale and the elastic and plastic properties of the surface, characterized by ψ 0 . 28 n=0 n=1 Fully plastic transition n=2 -1 10 n=3 n=4 n=5 n=6 n=7 n=8 n=9 n=10 10-2 10-3 10 -3 10-2 n=18 10 -1 10 0 Fully plastic transition n=1 n=2 n=3 10 -4 10 n=16 n=0 10 -2 10 -3 -4 n=11 Distribution function Qn(p/σ Y) Distribution function Qn(p/σY) 10 0 101 n=4 n=6 10 -2 n=8 n=10 10 -1 n=12 100 101 Normalized pressure p/σY Normalized pressure p/σ Y (a) (b) Fig 10 Cumulative contact pressure distributions Qn ( p / σ Y ) which specify the proportion of each scale of roughness over which the pressure exceeds p / σ Y . Results are shown for a cylindrical indenter subjected to nominal load Σ = 0.001 and for surfaces with (a) ψ 0 = 0.0002 , (b) ψ 0 = 0.1 . 5.2 Contact pressure distribution functions We proceed next to examine the contact pressure distributions in more detail. The pressure acting on each scale of contact is completely characterized by the distribution functions qn (ξ ) specifying the fraction of the contact. For an elastic surface, qn (ξ ) are smooth functions (see e.g. Fig. 5 of Ciavarella et al 2000). For plastic contacts, qn (ξ ) contains singularities (this is a consequence of the fact that contact pressure under fully plastic contacts is uniform - see Appendix B for a more detailed discussion). Consequently, it is more convenient to plot the cumulative contact pressure distribution functions Qn ( p / σ Y ) , which specify the proportion of each scale of contact over which the pressure exceeds p / σ Y . The cumulative contact pressure distributions can be computed from the definition q n = −σ Y ∂Qn ∂p . Results are shown in Fig 10 for a dimensionless nominal pressure Σ = 0.001 and two values of surface roughness, with γ = 5 and D = 1.5 . Fig 10(a) shows results for ψ 0 = 0.0002 , in which case a large number of scales under the contact are elastic. Under these conditions, the pressure distributions for roughness scales n = 0...9 , which deform elastically, tend towards a self-similar distribution. (Ciavarella et al 2000). The most heavily lo aded region of the 10th roughness scale begins to deform plastically, while the 11th reaches the fully plastic limit (which occurs when the contact pressure reaches 2.9σ Y ), and begins to transition towards a uniform distribution. The pressure acting on subsequent scales becomes increasingly uniform, and gradually increases towards the limiting value 5.8σ Y . A contrasting result for a rougher surface with ψ 0 = 0.1 is shown in Fig. 10(b). In this case, only scales n = 0...1 are elastic; scale n = 2 exceeds yield, and a small fraction of scale n = 3 reaches the fully plastic limit. Scale n = 4 is dominated by contacts in the 29 Plastic asymptote 10 10 10 0 -1 (a) Contact fraction 2<an >/λn (b) Elastic asymptote (c) (d) -2 Fully Plastic 10 -3 10 -4 10 7 10 5 No. contacts or pressure Normalized contact size or area fully plastic regime. The pressure acting on subsequent scales of roughness again becomes progressively more uniform, and increases towards the limit 5.8σ Y . Pei et al (2005) have noted the tendency of plasticity to drive contact size and pressure distributions to uniform distributions in numerical simulations (they refer to plasticity as a `grand equalizer’). Our model suggests an explanation for this phenomenon. The behavior of a sinusoidal surface is shown in Fig. 4, and is approximated by Eqs (23)-(39). Evidently, nominal pressures in the range 2 H 0 /ψ < p / σ Y < 2 H 0 / 3 acting on the sinusoid all induce the same true contact pressure. The first roughness scale to reach fully plastic behavior therefore exerts a nearly uniform pressure on the next scale of roughness. Elastic Elastic asymptote 10 3 10 1 -1 10 Contact area An/2a-1 Elastic asymptote -3 -2 -1 0 1 2 3 4 5 10 10 10 10 10 10 10 10 10 10 Elastic asymptote Plastic asymptote Contact pressure P/AnσY (a) (b) (c) -3 (b) (a) 6 No. contacts Nn /N0 Plastic asymptote 10 Plastic asymptote (d) (c) (d) Fully Plastic Elastic -3 -2 -1 0 1 2 3 4 5 10 10 10 10 10 10 10 10 10 10 6 Roughness scale ψ n Roughness scale ψ n (a) (b) Fig 11: The variations of (a) normalized average contact size and total contact area, and (b) the number of contact spots and normalized nominal contact pressure, for successive roughness scales under a cylindrical indenter. The chain lines mark the transitions from elastic to elasticplastic behavior, and from elastic-plastic behavior to fully plastic behavior. Results are shown for nominal pressure Σ = 0.001 and roughness values (a) ψ 0 = 0.0002 , (b) ψ 0 = 0.001 , (c) ψ 0 = 0.01 , (d) ψ 0 = 0.1 . 5.3 Average contact size, contact pressure, total contact area, and number of contacts for each scale The overall behavior of the contacts at each scale may be computed using the distribution functions given in the preceding section, and may also be estimated using the approximate procedure outlined in Section 4.2. Figures 11 and 12 show the results of these calculations. Figs. 11(a) and 12(a) show the contact fraction 2 an / λn and the ratio of true to nominal contact area An / 2 a− 1 for successive roughness scales n under a contact; while Figs 11(b) and 12(b) show the total number of contacts on each successive scale, normalized by the number of contacts in the zeroth roughness scale, N n / N0 , and the dimensionless nominal contact pressure P / σ Y An on 30 each scale. All variables have been plotted as a function of ψ n = ψ 0γ ( D−1) n to illustrate their asymptotic behavior clearly. Results are shown in Fig 11 for a nominal contact pressure Σ = 0.001 and for several values of surface roughness (parameterized by ψ 0 = Ψ D ( π Κ / 2Σ ) D−1 ); while Fig 12 shows results for a fixed dimensionless surface roughness Ψ DΚ D−1 = 10− 3 and progressively increasing load Σ . In each figure, the symbols show results obtained using the approximate procedure outlined in Section 4.2, while the solid lines show the exact procedure described in Section 4.1 and Appendix B. The approximate recursion scheme is clearly an excellent approximation to the exact solution. Fig. 11 shows clearly the influence of plastic flow on the behavior of successive scales of roughness. For the conditions chosen, the first roughness scale (n=0) is always elastic. As shown in Fig. 11(a), the contact size and contact area for subsequent elastic scales of roughness approach the elastic asymptotes: 2 an / λn → γ (1−D ) and An / 2 a− 1 → Σ / π 5 / 2ψ n , respectively, while the number of contact spots and nominal pressure approach N n / N0 → Σ γ (2− D)n / π 5 / 2ψ 0 and P / Anσ Y → π 3ψ n / 4 . The transition from elastic to elastic-plastic behavior occurs at ψ n ≈ 0.1γ ( D −1) (this approximation is strictly valid only if the contact size reaches its elastic asymptote); while the transition to fully plastic behavior occurs near ψ n ≈ 2 . Only two or three roughness scales lie in the elastic-plastic regime. For roughness scales that are loaded beyond the fully plastic limit, the contact size and contact area approach the plastic asymptotes: 2 an / λn → 1 ; An / 2 a−1 → π Σ / 4 H1(∞ ) ≈ 0.076 Σ , while the number of contact spots and the Contact fraction 2<an>/λn 0 10 (e) -1 10 (d) (c) -2 10 (b) -3 10 10 7 10 5 10 3 10 1 No. contacts or pressure Normalized contact size or area normalized nominal pressure approach N n / N0 → γ n and P / Anσ Y → H1 (∞ ) = 5.8 . (a) -4 10 No. contacts Nn/N0 -1 10 -5 10 Contact area An /2a- 1 -3 -2 -1 0 1 2 3 4 5 10 10 10 10 10 10 10 10 10 10 -3 10 -6 10 6 (e) (d) (c) (b) (a) Contact pressure P/AnσY (e) (d) (c) -2 (b) (a) -1 0 1 2 3 4 5 10 10 10 10 10 10 10 10 10 6 Roughness scale ψ n Roughness scale ψ n (a) (b) Fig 12 The variations of (a) normalized average contact size and total contact area, and (b) the number of contact spots and normalized nominal contact pressure, for successive roughness scales under a cylindrical indenter. Results are shown for constant normalized roughness Ψ DΚ D−1 = 10− 3 and progressively increasing nominal pressure (a) Σ = 10−5 , (b) Σ = 10−4 , (c) Σ = 10 −3 , (d) Σ = 10−2 (e) Σ = 0.03 . 31 The behavior of roughness scales under a surface with fixed roughness, which is indented by a cylinder under progressively increasing load is illustrated in Fig. 12. Results are shown for roughness Ψ DΚ D−1 = 10− 3 and five different progressively increasing loads. At the lightest load ( Σ = 10−5 ) only the zeroth scale of roughness is in the elastic regime; there are 2-3 scales in the elastic-plastic regime, but the majority of roughness scales lie in the fully plastic regime and the contact size, area, pressure and the number of contact spots quickly approach their fully plastic asymptotes. As load is increased, additional roughness scales with a lower ratio of amplitude to wavelength are brought within the contact, so the number of scales in the elastic regime increases with load. At Σ = 0.03 , there are six roughness scales in the elastic regime: a sufficient number to allow the contact size, area, pressure and the number of contact spots approach their elastic asymptotes. Shorter wavelengths again transition to the fully plastic regime. 6. Discussion We began this paper by pointing out that a fully satisfactory model of contact between rough surfaces has not yet been developed, and must reluctantly conclude with the same statement. The central issue is that measured surface geometries appear to be well characterized as a selfaffine fractal even at length-scales of a few hundred nanometers; calculations show, however, that for ideally elastic, perfectly fractal surfaces one cannot obtain meaningful predictions for any features of the true contact area. The situation is slightly better for elastic-plastic solids, where the total contact area and contact pressure are well defined, but it remains impossible to predict the actual number of contacts or their size. The ambiguous nature of the true contact could be resolved in either of two ways: (i) No real surface can be perfectly fractal, and provided that the surface geometry can be measured, and characterized, at extremely short wavelengths where the fractal description breaks down, standard techniques – including perhaps the Greenwood Williamson method – can be used to model the contact. Consequently, a systematic study of surface geometry at nanometer length scales would be of great interest. Available experimental studies provide strong evidence that surfaces cease to be fractal at short length scales. Majumdar and Bhushan (1990) obtained bilinear power spectra for profiles of hard disk surfaces, suggesting that the fractal descriptio n is accurate for roughness scales with wavelengths between about one hundred microns and a few hundreds nanometers. For short wavelength roughness, the spectral density remains linear, but with slope exceeding 3. Since 1 < D < 2 and 2D − 5 < 3 , thus, the surface retains its multi-scale character (and has no well-defined surface curvature) but it is not fractal. Similar results were obtained more recently by Bora et al (2004) for surfaces of MEMS devices. There is, of course, no difficulty in applying Ciavarella’s approach to surfaces of this type, although the GreenwoodWilliamson model remains ill-posed, since the mean asperity curvature cannot be computed. Preliminary calculations along these lines suggest that the contact spot size and area is well defined for surfaces of this type. (ii) It is also possible that some physical process, which was not included in our contact model, can regularize the ill posed nature of contact even for perfectly fractal surfaces. In fact, the present work was motivated by the hope that plasticity would have this effect. As we have shown, however, this is not the case, because of the stiffening caused by 32 lateral interactions between plastically deforming asperities. Some hope remains that this lateral interaction may play a less significant role under sliding contact, or in situations (such as metal forming) where the plastic region of the nominal contact is not contained. Behavior could also be regularized by phenomena such as short-range adhesive forces acting on the surfaces; the presence of a lubricant film; and time-dependent processes such as diffusion, which would tend to quickly spread small contacts. In our view, the most promising approach is to abandon attempts to characterize surface roughness as a perfect fractal. Nevertheless it is clear that a fractal description of a surface has several advantages over the conventional approach; and that some features of fractal behavior must be retained in a complete analysis. In particular, the fractal description captures the important fact that asperity contacts are spatially correlated. Since small asperities sit on top of larger ones, contact spots tend to cluster together. At a distance, a cluster of small spots resembles a single, larger contact – but these larger contacts also tend to cluster together. This progression continues down to the nominal contact. In the Greenwood-Williamson description, the focus is entirely on the smallest scale of contacts, which are assumed to be uniformly distributed spatially, and to exert a nominally uniform pressure. In contrast, the fractal description, together with the Ciavarella model, allows one to calculate the collective effect of the contact clusters, without needing to describe the contacts themselves in detail. For many applications, this is sufficient: one such example is our fractal based plasticity index Ψ DΚ D −1 , which determines the number of roughness scales that exceed yield, and the resulting depth of the plastic zone as illustrated in Figs 7 and 8. Our analytical calculations are also helpful to elucidate the behavior that one would expect to observe in a numerical simulation of contact between rough, ideally fractal, surfaces. The most sophisticated models of contact between rough surfaces rely on large-scale numerical simulations. These have now reached an advanced stage, particularly for elastic solids, and it is possible to compute contact pressure distributions and contact sizes for measured three dimensional surface profiles in considerable detail. The calculations of Ciavarella et al (2000), as well as those reported here, raise some questions regarding the interpretation of such simulations. It appears to be an inescapable conclusion that the true contact area for elastic contact between two perfectly fractal surfaces consists of an infinite number of contact spots of zero size, which are subjected to infinite contact pressure, and have a vanishing total area of contact. Numerical simulations that predict contact sizes, contact pressures, or the true area of contact between two elastic solids are therefore valid only if (i) the actual surface geometry ceases to be fractal at some short wavelength (e.g. the power-spectral density could be truncated at some short wavelength; or else decay sufficiently rapidly to give a differentiable autocorrelation function at the origin); and (ii) the resolutions of both the surface profilometer used to measure the 3D surface, as well as the numerical simulations, are sufficient to detect and to characterize surface geometry at these short wavelengths. Real surfaces must meet the first criterion, but measurements of actual surface roughness spectra suggest that real surfaces remain approximately fractal down to wave lengths of a few hundred nanometers. It is extremely difficult, therefore, to meet the second criterion, both when the surface is measured, and also in the subsequent numerical simulations. 33 Numerical simulations of elastic-plastic contact are somewhat less sensitive to artificial truncation of the roughness spectrum. In an elastic-plastic simulation, we would expect the predicted contact pressure, and the total area of contact, to converge to well-defined limits as the resolution is reduced, even for perfectly fractal surfaces. The total number of contacts, and their size, however, would not converge for a perfectly fractal surface. Again, such simulations are valid only if the surface roughness is not perfectly fractal, and if the simulations can properly resolve and characterize surface topography at short wavelengths. 4 10 Contact pressure P/A nσYψ0 Plastic asymptote n=11 n=8 103 Elastic asymptote n=5 102 101 100 10 1 102 10 3 104 Normalized surface strength 1/ψ0 Fig. 13: The dimensionless true contact pressure for a truncated Weierstrass surface, as a function of the dimensionless strength parameter 1/ψ 0 . The qua litative effects of introducing an artificial short-wavelength cutoff into numerical simulations of contact between fractal surfaces can be estimated using our simple model. As a representative example, we consider a Weierstrass surface (Eq. (7)) containing only terms n = 0,1,2...5 , which is indented by a perfectly flat, rigid indenter under a uniform nominal pressure p− 1 . We restrict attention to modest nominal pressures, so that the true area of contact is much less than the nominal contact area. Under these conditions, the true contact consists of a well-defined set of contact spots. The average size of these spots, and the true contact pressure, are essentially independent of nominal pressure, while the number of contact spots and the total true contact area increase in proportion to the nominal pressure. The contact size and mean contact pressure, as well as the relationship between true contact area and the number of contact spots, is determined by the elastic and plastic properties of the surface and the surface roughness, which may be characterized together by ψ 0 . The elastic map shown in Fig. 9 shows that for ψ 0 < 0.006 , the surface deforms elastically: under these conditions the behavior of the contacts can be estimated using the expressions given in Section 4.3, with n=5. For ψ 0 > 0.006 the contact spots exceed yield. The effect of plastic flow is to increase the size and number of contacts, to increase the true area of contact, and reduce the mean pressure. As ψ 0 increases, the true contact progressively transitions to the limiting conditions for a rigid plastic surface: the 34 contact fraction for the 5th scale approaches a5 / λ5 → 1 , the true contact pressure approaches p5 = 5.8σ Y and the ratio of true to nominal contact area approaches p −1 /5.8σY . The transition with ψ 0 is gradual, however. To illustrate this transition from elastic to plastic behavior, we have plotted the variation of the true contact pressure P Anσ Yψ 0 = Pλ0 An E * g 0 as a function of 1 ψ 0 = σ Y λ 0 E * g 0 in Fig. 13 for a surface with D = 1.5 and γ = 5 . Results are shown for surfaces truncated at scales n=5, n=8 and n=11. We have chosen to plot the results in this form to compare our predictions with the detailed finite element simulations of indentation of a nominally fractal surface reported by Pei et al (2005). In particular, their Fig 6 shows the predicted true contact area as a function of the ratio of yield stress to elastic modulus σ Y / E . The numerical results are in excellent qualitative agreement with our estimate. They observe a clear transition from elastic behavior at high σ Y / E to plastic behavior at low σ Y / E ; moreover, they predict P / An → 6σY in the limit σ Y / E → 0 , in excellent agreement with our estimate of P / An → 5.8σ Y . In our calculations, the elastic-plastic transition is a direct consequence of truncating the Weierstrass series, however. As the series is truncated at progressively shorter wavelengths, the elastic regime disappears. It is probable, therefore, that the elastic-plastic transition observed in numerical simulations is a consequence of a similar truncation introduced by the finite element interpolation. These results illustrate the challenges associated with numerical simulations of rough surface contact. The most satisfactory approach to avoiding these difficulties would be to purposely truncate the roughness spectrum in a controlled manner, and to ensure that the numerical computations can fully resolve the truncated spectrum. Numerical predictions are likely to be sensitive to the details of the truncation, however. Despite these concerns, numerical simulations of rough surface contacts can still provide useful information. Under purely normal indentation, the behavior of long-wavelength scales of roughness can be computed accurately even if the nature of the true contacts is not known. Numerical simulatio ns that artificially truncate the roughness spectrum can therefore be interpreted as modeling the behavior of roughness scales that exceed some critical cut-off. In models of wear, and contact fatigue, the principal concern is to characterize the state of the solid at microstructurally relevant length-scales. Predictions of the true contact area, the number of contact spots, and the actual size of the contact spots are of less interest. For such applications numerical simulations provide a useful way to compare surface roughness textures, even if the details of the very finest scales of roughness are not captured accurately. The situation for simulations of frictional sliding contact between rough surfaces is less clear. As we have seen, under purely normal loading, the contact pressure acting on the nth scale of roughness is determined fully by lower (longer wavelength) roughness scales, and so is not sensitive to truncation. In contrast, the frictional forces acting on the nth scale of roughness are determined by the behavior of higher (shorter wavelength) scales of roughness. Here, it seems that artificial truncation of the roughness spectrum will lead to spurious results. A particularly simple illustration of these difficulties is provided by an attempt to model frictional sliding between two fractal, elastic, surfaces. A model that follows the usual assumption that shear tractions are proportional to the true contact area arrives at the absurd conclusion that the coefficient of friction vanishes. 35 The simple model we have developed here could be extended in several ways. For example, it would be straightforward to extend the calculations described here to three dimensions: this would require numerical simulation and analytical characterization of the indentation of an elastic-plastic solid with a doubly sinusoidal surface; the remainder of the analysis would still apply. In addition, it would not be difficult to extend our calculations to account for the scale dependent nature of plastic flow: due to indentation size effects (Nix & Gao 1998) we might expect short wavelength roughness to return to elastic behavior. It would be of great interest to determine where this transition occurs, because then it would be possible to design surfaces that remain elastic. Finally, the general procedure employed here is clearly not restricted to fractal surfaces, and could be applied to any discrete approximation to a power spectral density, provided that the discrete scales remain widely separated. A more important issue, however, is the need to adopt a more realistic model of the surface roughness. The Weierstrass profile gives a convenient way to characterize the multi-scale nature of surface roughness, but is clearly not a realistic model of surface roughness. Furthermore, the scale decoupling introduced by Ciavarella et al (2000), which forms the basis of our analysis, strictly holds only in the limit γ → ∞ . A number of questions therefore need to be addressed. Firstly, it is of interest to determine the range of γ for which the scale decoupling can be applied. For elastic surfaces, this is straightforward. The scales will decouple as long as there are sufficient contacts of scale n+1 within each contact of scale n. Recall that (for sufficiently large n) the contact spot size for the nth scale can be estimated as 2 < an >= λn γ (1− D) . The number of contact spots of scale n+1 within each contact of scale n follows as m = 2 < an > / λn+1 ≈ γ (2 −D ) . Consequently, the critical value of γ to ensure m contact spots is γ ≥ m1/(2− D ) ≈ m 2 . Taking m = 2 as an optimistic estimate requires γ ≥ 4 ; in practice, an estimate of γ ≥ 10 is more reasonable. It is difficult to obtain similar analytical estimates for plastic contacts, so instead, we have investigated this issue using finite element computations with a finite number of roughness scales, some of which are described in Appendix A. Simulations with two and three scales of roughness suggest that scales of roughness completely decouple for γ ≥ 10 . For smaller γ , we find that a few asperities in the nth scale that are located near the peaks of scale n-1 may coalesce, in contrast to the large γ approximation. This is because at light loads, the nth scale contains only a few contacts, and insufficient lateral constraint develops to increase the contact pressure for the finest scale. However, asperities on scale n that are located in the valleys of scale n-1 are strongly constrained, and so exhibit behavior that is in good agreement with the behavior of a surface with purely sinusoidal roughness. Consequently, the predictions of the scale decoupling analysis continue to predict qualitatively the average behavior of each roughness scale even for low values of γ . Nevertheless, further work is required to properly characterize the behavior of random surfaces, for both elastic and plastic solids. Recent studies by Persson (2001, 2005) are promising steps in this direction. 36 7. Conclusions We have outlined an approximate model that is intended to estimate the behavior of a rough elastic-perfectly plastic solid with (two-dimensional) self-affine fractal surface roughness subjected to contact loading. The surface was idealized by a Weierstrass profile (Eq. (1)) with widely separated scales of roughness ( γ >> 1 ). Under these conditions, each scale of roughness can be regarded as a sinusoidal surface, indented by a flat surface. Successive scales are coupled through the condition that the nominal pressure acting on scale n is equal to the true pressure acting on scale n-1. On this basis, a recursive expression can be found for the contact pressure distribution, the contact size, and the total area of contact for each scale of roughness, in terms of the response of a sinusoidal surface (Ciavarella et al, 2000). In our calculations, the response of an elastic-plastic solid with sinusoidal roughness was approximated by fitting curves to finite element computations, as outlined in Section 3. We found that simple but highly accurate estimates for the average contact fraction, pressure and area of contact on successive roughness scales can be obtained by replacing the pressure distribution acting on each scale by a statically equivalent uniform distribution. We have used this approach to analyze the behavior of an elastic-perfectly plastic Weierstrass surface when indented by a rigid cylinder. We found that the behavior of the solid is governed by three dimensionless functions, defined in Eqs. (75). The parameter Σ quantifies the magnitude of the applied load; while Ψ and Κ together characterize the surface roughness and its resistance to plastic flow. The conditions necessary to induce plastic flow in successive roughness scales under the cylinder are illustrated in Fig. 7. For light loads, the nominal contact area is small, so roughness scales under the indenter have a large amplitude compared with their wavelength, and consequently a low resistance to plastic flow. As a result, at light loads only the zeroth scale of roughness deforms elastically; all other roughne ss scales are fully plastic. As the applied load is increased, the nominal contact area spreads, and new roughness scales that come into the contact area may remain elastic, and for sufficiently high loads will reach full contact. The parameter Ψ DΚ D −1 was identified as a plasticity index for a surface with fractal roughness. Surfaces with low values of Ψ DΚ D −1 have a large number of scales of roughness in the elastic regime, and consequently plastic flow is restricted to a thin layer at the surface, Fig. 8. As Ψ DΚ D −1 increases, the number of plastic roughness scales, and the corresponding depth of the plastically deformed layer, increase. We also calculated the average contact fraction, total contact area, number of contacts, and average contact pressure for successive scales of roughness. The results are illustrated in Figs. 11 and 12. The behavior of the surface depends primarily on the dimensionless parameter ψ 0 defined in Eq. (20), which characterizes the resistance to plastic flow of the longest wavelength scale of roughness under the nominal contact. This parameter can be computed in terms of Ψ and Κ using Eqs. (76). For low values of ψ 0 , a large number of long-wavelength roughness scales under the contact remain elastic: their behavior can be calculated using the closed- form expressions given in Section 4.3. Short wavelength roughness scales always deform plastically, and for large n approach the asymptotes given in Section 4.5. For an elastic-plastic solid with a perfectly fractal surface, the true area of contact consists of an infinite number of contact spots of 37 zero size, with a true contact pressure lim n →∞ P / An = 5.8σ Y . The true contact area is bounded, and proportional to the load. The implications of these findings were discussed in detail in Section 6. Our calculations suggest that idealizing surface roughness as a perfect fractal leads to unphysical predictions for the contact size and number of contact spots. It is likely that these difficulties can only be resolved by taking into account the departure of actual surfaces from perfectly fractal behavior at very short wavelengths. Acknowledgements This work was performed as part of the General Motors Collaborative Research Laboratory on Computational Materials Research at Brown University. The authors are grateful to M. Ciavarella, R.W. Carpick and J.F. Molinari for helpful discussions and for providing pre-prints of their recent publications. 38 Appendix A: Numerical simulations of plastic contact of a truncated Weierstrass profile Mean contact pressure pm/σ Y A central assumption that underlies the Ciavarella et al model is that successive scales of roughness interact only through the requirement that the true contact pressure acting on the nth scale must equal the nominal contact pressure exerted on the n+1th scale, and the deformation fields otherwise fully decouple. Saint Venant’s principle can be used to justify this assumption for elastic solids. For materials that deform plastically, however, there is no rigorous basis for this approximation – and indeed there is good reason to believe that there are circumstances where the deformation fields between successive roughness scales do not decouple. We have therefore conduc ted finite element simulations of elastic-plastic contact between surfaces with two or more scales of roughness, to investigate the validity of this assumption. A few representative results of these simulations will be outlined here. Approximation Asperity 1 Asperity 2 Asperity 3 Asperity 4 Asperity 5 5 4 3 λ2 2 1 2 3 1 4 5 λ 1/2 0.25 0.5 0.75 Contact fraction 2a2/λ2 1 Fig 14: Results of a finite element simulation of the indentation response of an elastic-perfectly plastic solid with two scales of surface roughness. The figure shows the variation of mean pressure with contact fraction for each asperity contact on the finer scale as the load is progressively increased. Our approximation to the behavior of a single roughness scale is shown for comparison. Fig 14 shows the behavior of an elastic-perfectly plastic solid with two roughness scales, which is indented by a rigid flat indenter. Results are shown for γ =λ 1 / λ0 = 10 E σ Y = 200 , g 0 λ0 = 1100 , g1 λ1 = 1 4 0 and ν = 0.3 . These parameters correspond to ψ 0 = 2 , ψ 1 = 5 , and D = 1.4 . We have computed the mean contact pressure as a function of contact size for each asperity in the finer roughness scale, and compared the results with our approximation to the behavior of a surface with a purely sinusoidal roughness. For small values of 2 a1 / λ1 the two cases are indistinguishable. For values of 2 a1 / λ1 > 0.6 (where asperities on the finer scale begin to interact), we find that the pressure on the two-scale surface rises somewhat more gradually than that for a pure sinusoid, but the limiting behavior as 2a1 / λ1 → 1 is correctly modeled by a single-scale surface. 39 The results of a second set of simulations are illustrated in Figs 15 and 16. Here, we compare directly the behavior of surfaces with one, two and three scales of roughness, when indented by a rigid flat surface. Results are shown for γ = 10 , E σ Y = 200 , ν = 0.3 , g 0 λ0 = 1100 , g1 λ1 = 1 4 0 , g 2 λ2 = 1 1 6 , which correspond to D = 1.4 , ψ 0 = 2.2 , ψ 1 = 5.5 , ψ 2 = 13.7 . All three scales of roughness are therefore in the fully plastic regime for sufficiently high loads. Fig 15 shows the distribution of contact pressure for the three scales of roughness, for two values of nominal load. These results clearly agree qualitatively with the predictions of our approximate model. In particular, the computations demonstrate the progressive increase in contact fraction and contact pressure on successive scales of roughness, and show also that asperity interactions cause the average contact pressure on scales n=1 and 2 to exceed the material hardness. Scale 0 Scale 1 Scale 2 4 3 2 1 0 0 0.1 0.2 0.3 Scale 0 Scale 1 Scale 2 5 Contact pressure p/σY Contact pressure p/σY 5 0.4 4 3 2 1 0 0.5 0 Position x/λ0 0.1 0.2 0.3 0.4 0.5 Position x/λ0 (a) (b) Fig. 15: The contact pressure distribution predicted by finite element computations for truncated Weierstrass surfaces with one, two and three scales of roughness that are indented by a rigid, flat surface. Results are shown for two values of nominal pressure: (a) p −1 / σY = 0.624 (b) p− 1 / σ Y = 2.10 A more quantitative comparison of our numerical and analytical predictions is shown in Fig 16. Here, we have plotted the total true area of contact predicted by finite element computations for each of the three roughness scales, and compared the results with our approximate analytical expression. The analytical results clearly provide a good approximation to the finite element computations. 40 Total contact area An/λ0 0.8 Finite element results Approximation 0.7 0.6 0.5 A0 A1 A2 0.4 0.3 0.2 0.1 0 0 1 2 Nominal contact pressure p-1/σY Fig 16: Comparison of the true area of contact predicted by finite element simulations with 1, 2 and 3 scales of roughness with the approximate analytical expressions predicted by assuming each scale behaves as an isolated sinusoid. 41 Appendix B: Numerical computations of pressure distribution functions. Elementary calculations show that the pressure distribution function for an elastic-plastic solid will have the form n qn (ξ ) = qn (ξ ) + ∑ Q (jn )δ (ξ − ξ (j n) ) (B1) j =0 where q n denotes a continuous function, and the sum consists of a series of direct Delta distributions with magnitude Q(j n) , which occur at normalized pressures ξ (jn ) . Given the nth term in the series, we seek to compute the next term. The general expression is given by ∞ qn (η ) = ∫ q (η ,ξ ,ψ n ) qn−1(ξ ) dξ (B2) 0 This integral must be divided into separate contributions that correspond to the elastic regime, elastic plastic regime, and fully plastic regime qn (η ) = e ξ max ∫ q (η ,ξ ,ψ ) q n −1 n (ξ )d ξ + e ξ min + fp ξ max ∫ q (η ,ξ ,ψ ) q n n −1 ep ξ max ∫ q (η, ξ ,ψ ) q n −1 n (ξ ) dξ ep ξ min (ξ )d ξ + fp (B3) fpi ξ max ∫ q (η, ξ ,ψ ) q n n −1 (ξ ) dξ fpi ξ min ξ min where the limits are e ξ min = η 2 / 4πψ n e e ξ max = min(ξ nmax −1 , ξn −1) ep ξ min = max( p max −1 (η ), ξne−1 ) ep −1 ξ max = min(ξnmax −1 , β (η )) fp ξ min = max(ξ nmax −1 , 2 H 0 / ψ n ) fp ξ min = max(ξ nmax −1 , 2 H 0 /3) fp ξ max = min(ξnmax −1 ,2 H 0 /3) (B4) fpi ξ max = min(ξnmax −1 , H 1(ψ n )) where ξ ne denotes the critical value of dimensionless nominal pressure to induce yield in the nth scale; ξnmax is the maximum dimensionless nominal pressure acting on the nth scale; p max −1 denotes the value of ξ that satisfies eq (31) with ξ max = η ; β − 1(η ) is the value of ξ that satisfies the second of Eq. (30) with β = η . Next, recall that that fp fp q (ξ ,η ) = A(ξ ,ψ )δ (η − H0 ) ξmin < ξ < ξmax (B5) fpi fpi q (ξ ,η ) = A(ξ ,ψ )δ (η − Φ mean (ξ )) ξmin < ξ < ξ max Substitute into (B2) to see that 42 qn (η ) = e ξ max ∫ q (η ,ξ ,ψ ) q n −1 n (ξ )d ξ + e ξ min + fpi ξ max ∫ A ( ξ ,ψ ) δ (η − Φ mean n ep ξmax fp ξ max ∫ q (η, ξ ,ψ ) q n n −1 ep ξ min (ξ ) dξ + δ (η − H0 ) ∫ A (ξ ,ψ n ) qn−1 (ξ ) dξ fp ξ min n −1 (B6) (ξ )) qn−1(ξ ) dξ + ∑ Q(jn−1) A(ξ jn−1, ψ n )δ (η − Φ mean (ξ jn−1,ψ n )) j =0 fpi ξ min Hence Q = n 0 fp ξ max ∫ Afp (ξ ,ψ n )qn−1(ξ ) dξ fp ξ min Qnj = Qj( n−1−1) Afpi (ξ j(−n1−1) ,ψ n ) ξ nj = Φ mean (ξ nj−−11,ψ n ) qn (η ) = (B7) e ξ max ∫ q (η ,ξ ,ψ ) q n n −1 (ξ )d ξ + e ∫ q (η, ξ ,ψ ) q n ep ξ min + ξ ep max ξ min fpi ξ max ∫ A (ξ ,ψ ) δ (η − Φ mean n fpi ξ min 43 (ξ )) qn−1(ξ )d ξ n −1 (ξ ) dξ References Archard, J. F. 1957 Elastic deformation and the laws of friction. Proc. R. Soc. Lond. A 243, 190205. Berry, M. V. & Lewis, Z. V. 1980 On the Weierstrass-Mandelbrot fractal function. Proc. R. Soc. Lond. A 370, 459-484. Bora, C. K., Plesha, M. E., Flater, E. E., Street, M. D., Carpick, R. W. & Redmond J. M. 2004 Multiscale roughness of MEMS surfaces. Proceeding of 2004 ASME/STLE Joint International Tribology Conference, TRIB2004-64370. Borri- Brunetto, M., Carpinteri, A. & Chiaia, B. 1999 Scaling phenomena due to fractal contact in concrete and rock fractures. Int. J. Fract. 95, 221-238. Bush, A. W., Gibson, R. D. & Thomas, T. R. 1975 The elastic contact of a rough surface. Wear 35, 87-111. Childs, T. H. C. 1973 The persistence of asperities in indentation experiments. Wear 25, 3-16. Childs, T. H. C. 1977 The persistence of roughness between surfaces in static contact. Proc. R. Soc. Lond. A 353, 35-53. Ciavarella, M. & Demelio G. 2001 Elastic multiscale contact of rough surfaces: Archard’s model revisited and comparisons with modern fractal models. ASME Jl Appl. Mech. 68, 496498. Ciavarella, M., Demelio, G., Barber, J. R. & Jang, Y. H. 2000 Linear elastic contact of the Weierstrass profile. Proc. R. Soc. Lond. A 456, 387-405. Gao, Y. F., Bower, A. F., Kim, K.-S., Lev, L. & Cheng, Y. T. 2005 The behavior of an elasticperfectly plastic sinusoidal surface under contact loading. (In review). Greenwood, J. A. 1984 A unified theory of surface roughness. Proc. R. Soc. Lond. A 393, 133157. Greenwood, J. A. & Williamson, J. B. P. 1966 Contact of nominally flat surfaces. Proc. R. Soc. Lond. A 295, 300-319. Hill, R. 1950 The Mathematical Theory of Plasticity. Oxford University Press. Hyun, S., Pei., L., Monlinari, J. F. & Robbins, M. O. 2004 Finite-element analysis of contact between elastic self-affine surfaces. Phys. Rev. E 70, 026117. Johnson, K. L. 1985 Contact Mechanics. Cambridge University Press. Majumdar, A. & Tien, C. L. 1990 Fractal characterization and simulation of rough surfaces. Wear 136, 313-327. Majumdar, A. & Bhushan, B. 1990 Role of fractal geometry in roughness characterization and contact mechanics of surfaces. ASME Jl Tribol. 112, 205-216. Majumdar, A. & Bhushan, B. 1991 Fractal model of elastic-plastic contact between rough surfaces. ASME Jl Tribol. 113, 1-11. McCool, J. I. 1986 Comparison of models for the contact of rough surfaces. Wear 107, 37-60. Mihailidis, A., Bakolas, V. & Drivakos, N. 2001 Subsurface stress field of a dry line contact. Wear 249, 546-556. Nayak, P. R. 1971 Random process model of rough surfaces. ASME Jl Lubr. Technol. 93, 398407. Nix, W. D. & Gao, H. 1998 Indentation size effects in crystalline materials: a law for strain gradient plasticity. J. Mech. Phys. Solids 46, 411-425. Pei, L., Hyun, S., Monlinari, J. F. & Robbins, M. O. 2005 Finite element modeling of elastoplastic contact between rough surfaces. (In review). 44 Persson, B. N. J. 2001 Elastoplastic contact between randomly rough surfaces. Phys. Rev. Lett. 87, 116101. Persson, B. N. J., Albohr, O., Tartaglino, U. Volokitin, A. I. & Tosatti, E. 2005 On the nature of surface roughness with application to contact mechanics, sealing, rubber friction and adhesion. J. Phys: Condens. Matter 17, R1-R62. Polonsky, I. A. & Keer, L. M. 2000a A fast and accurate method for numerical analysis of elastic layered contacts. ASME Jl Tribol. 122, 30-35. Polonsky, I. A. & Keer, L. M. 2000b Fast methods for solving rough contact problems: a comparative study. ASME Jl Tribol. 122, 36-41. Sayles, R. S. 1996 Basic principles of rough surface contact analysis using numerical methods. Tribol. Int. 29, 639-650. Sayles, R. S. & Thomas, T. R. 1978 Surface topography as a nonstationary random process. Nature 271, 431-434. Tabor, D. 1951 Hardness of Metals. Oxford University Press. Westergaard, H. M. 1939 Bearing pressures and cracks. ASME Jl Mech. 6, 49-53. Whitehouse, D. J. & Archard, J. F. 1970 The properties of random surface of significance in their contact. Proc. R. Soc. Lond. A 316, 97-121. Willner, K. 2004 Elasto-plastic normal contact of three-dimensional fractal surfaces using halfspace theory. ASME Jl Tribol. 126, 28-33. 45