PHYSICAL REVIEW LETTERS

PRL 111, 034301 (2013)

Friction Between a Viscoelastic Body and a Rigid Surface with Random Self-Affine Roughness Q. Li,1 M. Popov,1 A. Dimaki,2 A. E. Filippov,3 S. Ku¨rschner,1 and V. L. Popov1,* 1

2

Berlin University of Technology, 10623 Berlin, Germany Institute of Strength Physics and Materials Science, Russian Academy of Sciences, 634021 Tomsk, Russia 3 Donetsk Institute for Physics and Engineering of NASU, 83114 Donetsk, Ukraine (Received 13 May 2013; revised manuscript received 13 June 2013; published 17 July 2013) In this Letter, we study the friction between a one-dimensional elastomer and a one-dimensional rigid body having a randomly rough surface. The elastomer is modeled as a simple Kelvin body and the surface as self-affine fractal having a Hurst exponent H in the range from 0 to 1. The resulting frictional force as a function of velocity always shows a typical structure: it first increases linearly, achieves a plateau and finally drops to another constant level. The coefficient of friction on the plateau depends only weakly on the normal force. At lower velocities, the coefficient of friction depends on two dimensionless combinations of normal force, sliding velocity, shear modulus, viscosity, rms roughness, rms surface gradient, the linear size of the system, and the Hurst exponent. We discuss the physical nature of different regions of the law of friction and suggest an analytical relation describing the coefficient of friction in a wide range of loading conditions. An important implication of the analytical result is the extension of the well-known ‘‘master curve procedure’’ to the dependencies on the normal force and the size of the system. DOI: 10.1103/PhysRevLett.111.034301

PACS numbers: 46.55.+d, 62.20.Qp, 81.40.Pq

Since classical works by Bowden and Tabor [1], it is widely accepted that the roughness plays a central role in friction processes. Greenwood and Tabor [2] have shown that the friction of elastomers can be attributed to deformation losses in the elastomer. In 1963, Grosch supported this idea by a series of experiments of friction between rubber and hard surfaces with controlled roughness [3]. In the following years, the basic understanding of the role of rheology [4] and of surface roughness [5,6] in elastomer friction has been achieved. Most works on elastomer friction discuss the coefficient of friction, thus implicitly implying the validity of Amontons’ law: the force of friction is proportional to the normal load; the coefficient of friction is considered to be a quantity which does not depend on the normal load [7,8]. However, it is well known that this law is only a very rough first approximation and that both the static and the sliding coefficient of friction, even between the same material pairing, can change by a factor of about 4 depending on the geometry of a tribological system as a whole and loading conditions. The load dependence of the elastomer friction was studied experimentally by Schallamach [9]. In a more general context, the strong violations of Amontons’ law were studied experimentally and theoretically in recent papers [10,11]. Deviations from Amontons’ law can be due to macroscopic interfacial dynamics [12–14] or they can be connected with the contact mechanics of rough surfaces. This Letter is devoted to a study of elastomer friction beyond the regions of validity of Amontons’ law due to purely contact mechanical reasons. To achieve the basic understanding of this nonlinear frictional behavior, we consider the following simple model: (i) the elastomer is modeled as a simple Kelvin body, which is completely characterized by its 0031-9007=13=111(3)=034301(5)

static shear modulus and viscosity, (ii) the nondisturbed surface of the elastomer is plane and frictionless, (iii) the rigid counter body is assumed to have a randomly rough, self-affine fractal surface without long wave cutoff, (iv) no adhesion or capillarity effects are taken into account, and (v) we consider a one-dimensional model. These simple assumptions still result in nontrivial and complicated frictional behavior. We do not claim that the reported results can be directly applied for the friction of a true three-dimensional elastomer. However, we would like to note that there is evidence coming from recent studies of contact mechanics of both rotationally symmetric profiles [15,16] and self-affine fractal surfaces [17,18] that suggest that the results obtained with one-dimensional foundations may have a broad area of applicability if the rules of the method of dimensionality reduction (MDR) [19–21] are applied. Following this method, the elastomer was modeled as a row of independent elements with a small spacing x, each element consisting of a spring with normal stiffness k ¼ 4Gx and a dashpot having the damping constant d ¼ 4x, where G is the shear modulus and the viscosity of the elastomer. The counter body was a rough line having the power spectral density C1D / q2H1 , where q is the wave vector and H, the Hurst exponent. The spectral density was defined in the interval from qmin ¼ 2=L, where L is the system size, to the upper cutoff wave vector qmax ¼ =x. The spacing x determines the upper cutoff wave vector and is an essential physical parameter of the model. Surface topography was characterized by the rms roughness h ¼ R ½2 qqmax C1D ðqÞdq1=2 , which is dominated by the long min wavelength components of the power the Rq spectrum 2and 1=2 max rms gradient of the surface rz ¼ ½2 qmin C1D ðqÞq dq ,

034301-1

Ó 2013 American Physical Society

week ending 19 JULY 2013

PHYSICAL REVIEW LETTERS

dominated by the short wavelength part of the spectrum. The rigid surface was generated according to the rules described in [21], and periodic boundary conditions were used. The rigid surface was pressed against the elastomer with a normal force FN and moved tangentially with a constant velocity v. If the rigid profile is given by z ¼ zðx vtÞ, and the profile of the elastomer by u ¼ uðx; tÞ, then the normal force in each particular element of the viscoelastic foundation is given by

“Rigid body”

-1 -1.5 -2 -2.5

“Elastomer”

0

0.004

0.008

0.012

0.016

0.02

x, m

FIG. 1 (color online). One-dimensional contact between a rough surface and a viscoelastic elastomer. Note the difference in vertical and horizontal units.

(1)

For the elements in contact with the rigid surface, this means that f ¼ 4xfG½d zðx; tÞ þ vz0 ðx; tÞg;

v

FN

0 -0.5

(2)

where d is the indentation depth. For these elements, the condition of remaining in contact, f > 0, was checked in each time step. Elements out of contact were relaxed _ tÞ ¼ 0, and according to equation f ¼ 0: GuðxÞ þ uðx; the noncontact condition u < z was checked. The indentation depth d was determined to satisfy the condition of the constant normal force Z FN ¼ 4 ½Gðd zðxÞÞ þ vz0 ðxÞdx; (3) ðreal contÞ

where the integration is only over points in contact. A typical configuration of the contact is shown in Fig. 1. The tangential force was calculated by multiplying the local normal force in each single element with the local surface gradient and subsequently summing over all elements in contact Z Fx ¼ 4 z0 ðxÞ½Gðd zðxÞÞ þ vz0 ðxÞdx: (4) ðreal contÞ

Because of the independence of the degrees of freedom, the algorithm is not iterative and there are no convergence problems. The one-dimensional model is computationally efficient and allows carrying out extensive parameter studies. The following ranges of parameters have been covered in the present study. The length of the system was L ¼ 0:02 m and the number of elements N ¼ L=x was typically 5000 with exception of cases where the dependence on x was studied. Instead of viscosity, the relaxation time ¼ =G ¼ 103 s was used. 11 values of Hurst exponent ranging from 0 to 1 were studied. All values shown below were obtained by averaging over 200 realizations of the rough surface for each set of parameters. Parameter studies have been carried out for 20 different normal forces FN , ranging from 103 to 102 N, 20 values of the G modulus from 103 to 109 Pa, 20 values of rms roughness h from 109 to 105 m, and 20 values of the spacing x from 105 to 107 m, while in each simulation series only one parameter was varied. The presented results are based on approximately 3:5 106 single simulations with the total net computation time of about 50 h. It is well known that the maximum value of the

coefficient of friction in the medium range of velocities is proportional to the rms gradient of the surface profile [19]. We, therefore, present the normalized friction coefficient =rz instead of in this Letter. A typical dependence of the coefficient of friction on the sliding velocity is shown in Fig. 2. At first, it increases linearly with velocity (region I), it then achieves a plateau (region III) and decreases again to a new constant value (region IV). We also marked an intermediate region (II) where transition from the linear velocity dependence to the plateau takes place. This region covers one decade of velocities, and the coefficient of friction increases here by a factor of two. Fig. 3 shows the velocity dependence in double logarithmic scale for 6 different Hurst exponents. It is obvious that at small velocities, the coefficient of friction increases linearly with velocity. The absence of the decreasing region IV in Fig. 3 (and Fig 4 at high loads) is only due to the fact that for high forces this region is outside the scope of practical velocities and is therefore not shown in these figures. Fig. 4 presents velocity dependencies of the coefficient of friction for 20 different normal forces. One can see that the form of the dependence for different forces is approximately the same, only shifted along the axis of the logarithm of velocity. There are two distinctly different regions: in zone 1 there is a partial contact of the rigid surface and the elastomer, while in zone 2 they are in 1.5

1

z

_ tÞg: f ¼ 4xfGuðxÞ þ uðx;

0.5

z, m

PRL 111, 034301 (2013)

0.5

I 0

10 -2

II 10 -1

III 10 0

IV 10 1

10 2

Velocity, m/s

FIG. 2. A typical dependence of the normalized coefficient of friction on the velocity. In this particular case, the results were obtained for the following set of parameters: FN ¼ 0:0034 N, G ¼ 107 Pa, h ¼ 5 107 m, and H ¼ 0:7.

034301-2

¼

100

z

10-1

H=0 H=0.2 H=0.4 H=0.6 H=0.8 H=1

10-2

10-3

10 -2

10 0

10 2

Velocity, m/s

FIG. 3 (color online). Dependence of friction coefficient on velocity for different Hurst exponents and F ¼10 N,G ¼ 107 Pa, h ¼ 5 107 m. Solid lines correspond to the analytical approximation (11).

complete contact. In both of the zones, the shift factor increases linearly with the logarithm of force, the coefficient of friction, thus, being a power function of the normal force. Simulations with different rms gradients of the surface (which were varied by changing the spacing x) show that the coefficient of friction in this region is very accurately proportional to rz2 and depends on the force and shear modulus only over the ratio FN =G. The only form of the dependence which fits these empirical observations and meets the dimensional demands is vrz2 GhL ¼ ; (5) h FN where and are dimensionless constants. Empirical values of these constants extracted from numerical data are shown in Fig. 5. Let us support this result with an analytical estimation. At low velocities, the values of z in the border points of each partial contact region in R the Eq. (4) are the same (z ¼ d); thus, the integral real cont z0 ðxÞ½Gðd zðxÞÞdx vanishes identically. For the coefficient of friction we get

4Lcont rz2cont v: FN

(6)

Here, Lcont is the total contact length and rzcont the rms slope in the region of real contact. The rms slope is dominated by the short wavelength part of the spectrum. It can be approximately replaced by the average rms slope of the entire surface rzcont rz. At the end of the Letter, we discuss the weak dependence of rzcont on loading parameters in more detail. For small forces, in zone 1, the contact length is a power function of the normal force [17]: Lcont / F1=ð1þHÞ , and the coefficient of friction will be given by / FH=ð1þHÞ . Comparing this with Eq. (5) provides an analytical estimation for the exponent : ¼

H : 1þH

(7)

For large normal forces, in zone 2, the contact length achieves a saturation value of Lcont ¼ L. The coefficient of friction becomes ¼

4Lrz2 v; FN

(8)

which is exactly confirmed by numerical simulations. Finally, in the plateau region, the coefficient of friction shows only a weak dependence on the Hurst exponent (Fig. 6). In the range of 0:2 < H < 0:8 and for not too small forces, it is almost constant and can be approximated as pﬃﬃﬃ 2rzcont : (9) This result has a simple physical meaning. In the plateau region, the elastomer behaves practically as a viscous fluid: the elasticity does not play any role and all contacts are ‘‘one-sided.’’ The normal and tangential R R forces reduce to Fx ¼4 ðrealcontÞ v½z0 ðxÞ2 dx, FN ¼4 ðrealcontÞ vjz0 ðxÞjdx. For the normalized coefficient of friction we get 1

-3

10 0

week ending 19 JULY 2013

PHYSICAL REVIEW LETTERS

PRL 111, 034301 (2013)

FNmin =10 N

0.9 0.8

z

0.7 0.6 0.5 2

FNmax =10 N

10 -1

0.4 0.3

Zone 1 Zone 2

10-2

100

α H/(1+H) β

0.2

10 2

0.1

Velocity, m/s

0

FIG. 4 (color online). Double logarithmic presentation of the dependence of the normalized friction coefficient on velocity for 20 exponentially increasing normal forces ranging from 103 to 102 N, as indicated by the arrow (G ¼ 107 Pa, h ¼ 5 107 m, and H ¼ 0:7). The third line from the left corresponds to the data shown in Fig. 2.

0

0.2

0.4

0.6

0. 8

1

Hurst exponent

FIG. 5. Dependence of and [see Eq. (5) on the Hurst exponent in zone 1 (see Fig. 4)]. Analytical estimation of the exponent according to (7) is shown with the bold line. For 0:2 < H < 0:8, it fits the numerical data very well.

034301-3

week ending

Maximum o f

z

PHYSICAL REVIEW LETTERS 19 JULY 2013 PRL 111, 034301 (2013) R 0 2 velocity at which the elastomer is detached from the rigid ðreal contÞ ðz ðxÞÞ dx ¼ R surface on the trailing side of any asperity and all the 0 ðreal contÞ jz ðxÞjdx contacts become ‘‘one-sided.’’ Indeed, according to (2), 1=2 R the condition of detachment f ¼ 0 means d zðxÞ þ 0 2 ðreal contÞ ðz ðxÞÞ dx vz0 ðxÞ ¼ 0. Taking into account that d z has the order : (10) ¼ rzcont R 0 of magnitude of h and z0 has the order of magnitude of ðreal contÞ jz ðxÞjdx rzcont , we come to the conclusion that the one-sided For an exponential probability distribution function of the detachment of the elastomer will occur if ð=GÞvrzcont > gradient of pﬃﬃﬃthe surface, the ratio of the integrals in (10) is h or v > 1. Note that the same conditions are valid in the equal to 2, in accordance with (9), and it depends only corresponding three-dimensional problem: for achieving weakly on the form of the distribution function. the plateau value of contact stiffness (F N 1, [17]) and The results (5) and (7)–(9) can be combined in the for the one-sided detachment of the elastomer (v 1). following equation providing an interpolation between Let us discuss the decrease of the coefficient of friction the three regions I, II, and III: beyond the region of validity of approximation (12), at large 2 1 FN velocities (region IV in Fig. 2). Such a decrease at large ¼ þ velocities is typical for elastomer friction and is usually 2rz2cont 4Lrz2cont v H 2 1=2 associated with a decrease in the ‘‘rheological factor’’ h FN 1þH þ : (11) ImGð!Þ=jGð!Þj at high frequencies [6], where Gð!Þ is rz2cont v GhL the complex modulus of the elastomer and ImGð!Þ its The quality of this interpolation can be seen in Fig. 3 where imaginary part. For the case of the Kelvin body, however, pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ the numerical results for six Hurst exponents are plotted the rheological factor is equal to != G2 þ ð!Þ2 ; it together with analytical dependencies (11). This equation increases monotonically and tends towards 1 at high frecan be rewritten in the dimensionless form quencies. In this case, the decrease of the coefficient of 2H friction is not related to the rheology but rather to the 2 1=2 1þH ððFN =4Þ þ ðFN Þ Þ ¼ 1þ ; (12) dependence of the rms slope on the size of the real contact. 2 v Indeed, for randomly rough surfaces, the rms slope in the with ¼ = contact region can be estimated as pﬃﬃﬃ a normalized coefficient of friction Zq 1=2 ð 2rzcont Þ, dimensionless velocity max 2 rz C1D ðqÞq dq cont ¼ 2 vrz qcont v ¼ pﬃﬃﬃ cont ; (13) 2h 2ð1HÞ ½qmax q2ð1HÞ 1=2 cont and dimensionless force / ; (15) 2ð1 HÞ F N : (14) F N ¼ where the lower integration limit qcont 2=Lcont decreases GhL with increasing size of the real contact. For 0 < H < 1, the Let us discuss the physical meaning of the quantities v and integral (15) depends only weakly on the lower integration F N . The condition F N 1 gives the order of magnitude of limit unless the contact length becomes extremely small so the force at which complete contact is achieved, while the that qcont approaches qmax . Thus, the coefficient of friction in condition v 1 determines the order of magnitude of the region of plateau will decrease with decreasing indentation depth. This happens either at extremely high sliding velocities (Fig. 2, region IV) or at extremely low normal 1.6 forces as illustrated in Fig. 6. The dependence of rzcont on the 1.5 contact size and, thus, on velocity and force is less pro1.4 nounced for small Hurst exponents, H 0, and gets stronger 1.3 H=0 H=0.1 for H 1. Note that the increase of rms slope with increas1.2 H=0.2 H=0.3 ing indentation is closely associated with the assumption of 1.1 H=0.4 H=0.5 the "randomness" of roughness, as the estimation (15) is only 1 H=0.6 H=0.7 valid if the Fourier components of roughness with different 0.9 H=0.8 wave vectors have uncorrelated phases. One can say that H=0.9 0.8 H=1 randomly rough surfaces are always rougher on the slopes 0.7 10 -3 10 -2 10 -1 10 0 10 1 10 2 of waviness than on the summits. Real surfaces, on the F ,N contrary, may have different kinds of correlated roughness. One can easily imagine a surface, which is rougher on the FIG. 6 (color online). Dependence of the normalized coeffisummits than on the slopes; for such surfaces, the rms slope cient of friction in region III (plateau). The coefficient of friction of roughness would decrease with indentation. The general decreases at very small forces. This effect is closely related to the and robust statement, which is independent of the kind of the decrease of the coefficient of friction at high sliding velocities, roughness correlation, is only that the rms slope in the contact Fig. 2, in region IV. N

034301-4

PRL 111, 034301 (2013)

PHYSICAL REVIEW LETTERS

region is a function of indentation depth and, thus, a function of the nondimensional force (14). This statement even remains valid if the linear viscoelastic behavior of the material breaks down at the microscale. Indeed, the statement that the frictional force will depend on the indentation depth is correct for any kind of processes at the microscale. The indentation depth, however, is governed by the contact stiffness which is dominated by the largest wavelength in the power spectrum of the roughness. The general conclusions that the nondimensional force (14) is a governing parameter of the friction process will, therefore, remain valid independently of the particular character of the microscopic processes. We can summarize our results to the following general scaling relation: F N Þ; ¼ rzcont ðF N Þg½v=fð or, in explicit form, FN vrz FN ¼ rzcont g pﬃﬃﬃ cont f : GhL GhL 2h

(16)

(17)

This scaling relation means that the dependence of the coefficient of friction on velocity in the double logarithmic presentation has the same form for different values of all parameters appearing in this equation: force FN , size of the system L, and relaxation time . Changing of any of these parameters will only shift the curves horizontally by the FN ﬃﬃ =fðGhL factor of log½prz Þ and vertically by the factor of 2h logrzcont ðFN =GhLÞ. In particular, the curves will be shifted by changes of temperature (which influences the relaxation time). The shifting procedure with regard to temperature is well known and widely used in the physics of friction of elastomers for constructing ‘‘master curves’’ describing the friction coefficient at any velocity and temperature (see, e.g., [22]). Eq. (17) means that the master curve procedure can be generalized to dependencies on other loading and system parameters. While the particular form (11) of the law of friction is limited by the assumptions of simple viscoelastic rheology, the general scaling relation (17) should have a wider range of application and it should be possible to validate it experimentally. In conclusion, we have shown that the law of friction between a linear viscoelastic body and a rigid fractal surface can be formulated in terms of two dimensionless variables (13) and (14) which are proportional to the sliding velocity and the normal force, correspondingly. Over these variables, the force of friction generally depends on all material, loading, and roughness parameters: sliding velocity, normal force, shear modulus, viscosity, rms roughness, rms slope, and even the size of the system. Generally, the force of friction is not proportional to the normal force; thus, Amonton’s law is violated. However, in the plateau region, where the coefficient of friction achieves its maximum, it is proportional to the rms slope of the roughness in the contact region and depends only weakly on the normal force or any other system parameter. We provided physical interpretation of the dimensionless variables and a simple

week ending 19 JULY 2013

interpolation equation summarizing all numerical and analytical data for a surface with self-affine roughness having Hurst exponents in the rage from 0 to 1. One of the implications of the obtained analytical results is the generalization of the master curve procedure to further variables such as the normal force and the size of the system. We argued that the main physics of the frictional process are dimension invariant. In particular, the general scaling relations should retain their validity for three-dimensional systems. We would like to thank R. Pohrt and J. Benad for valuable discussions. This work was supported by the Deutsche Forschungsgemeinschaft (DFG), the German Academic Exchange Service (DAAD), and the Federal Ministry of Economics and Technology (Germany) under the Contract No. 03EFT9BE55, Q. Li was supported by a scholarship from the China Scholarship Council (CSC).

*Corresponding author. v. [email protected] [1] F. P. Bowden and D. Tabor, The Friction and Lubrication of Solids (Clarendon Press, Oxford, 1986). [2] J. A. Greenwood and D. Tabor, Proc. R. Soc. A 71, 989 (1958). [3] K. A. Grosch, Proc. R. Soc. A 274, 21 (1963). [4] M. Barquins and R. Courtel, Wear 32, 133 (1975). [5] M. Klu¨ppel and G. Heinrich, Rubber Chem. Technol. 73, 578 (2000). [6] B. N. J. Persson, J. Chem. Phys. 115, 3840 (2001). [7] B. Lorenz, B. N. J. Persson, G. Fortunato, M. Giustiniano, and F. Baldoni, J. Phys. Condens. Matter, 25, 095007 (2013). [8] V. L. Popov and A. Dimaki, Tech. Phys. Lett. 37, 8 (2011). [9] A. Schallamach, Proc. R. Soc. Edinburgh, Sect. B Biol. 65, 657 (1952). [10] O. Ben-David and J. Fineberg, Phys. Rev. Lett. 106, 254301 (2011). [11] M. Otsuki and H. Matsukawa, Sci. Rep. 3, 1586 (2013). [12] S. M. Rubinstein, G. Cohen, and J. Fineberg, Nature (London) 430, 1005 (2004). [13] O. Ben-David, G. Cohen, and J. Fineberg, Science 330, 211 (2010). [14] D. S. Amundsen, J. Scheibert, K. Thøgersen, J. Trømborg, and A. Malthe-Sørenssen, Tribol. Lett. 45, 357 (2012). [15] M. Heß, About Mapping of Some Three-Dimensional Contact Problems to Systems with a Lower Spatial Dimensionality (Cuvillier-Verlag, Go¨ttingen, 2011). [16] M. Heß, Phys. Mesomech. 15, 264 (2012). [17] R. Pohrt, V. L. Popov, and A. E. Filippov, Phys. Rev. E 86, 026710 (2012). [18] S. Ku¨rschner and V. L. Popov, Phys. Rev. E 87, 042802 (2013). [19] V. L. Popov, Contact Mechanics and Friction. Physical Principles and Applications (Springer-Verlag, Berlin, 2010), p. 362. [20] V. L. Popov and M. Heß, Method of Dimensionality Reduction in Contact and Friction: A Calculation Methodology for Micro- and Macro Systems (Springer, Berlin, 2013). [21] V. L. Popov, Friction and wear in machinery 1, 41 (2013). [22] A. Le Gal, X. Yang, and M. Klu¨ppel, J. Chem. Phys. 123, 014704 (2005).

034301-5