Experimental and numerical investigation of asperities and indents with respect to rolling contact fatigue

Rolling contact experiments with slip were performed of artificial asperities and indents with pile-up. Micro-pits arose on the leading edge of the asperities and classic rolling contact fatigue (RCF) cracks initiated behind the trailing edge of the indents. The elastic-plastic run-in process and the thermal elastohydrodynamic lubrication (TEHL) load cycles were studied numerically. The run-in process caused high tensile residual stresses on the leading edges of the asperities while the TEHL load cycle caused high tensile stresses on the trailing edges of both the asperities and the indents. The conclusion was thus drawn that the classic RCF cracks behind the indents were caused by the THEL load cycle while the micro-pits on the asperities were caused by the residual stresses. This work continues the investigation on the effects of surface roughness in TEHL contacts with focus on well-defined micrometre high asperities. The goal was to attain deeper insights into the initiation mechanisms of RCF through experiments and analyses of individual well defined asperities. The numerical analyses include both the elastic-plastic run-in cycle and the TEHL over-rolling load cycles.


Introduction
Many different technical components utilize hard rolling contacts. This work focused on the highly loaded rolling contacts found in gears and roller bearings. Since these contacts often are subjected to contact pressures in the gigapascal range, they will eventually fail due to rolling contact fatigue (RCF). This work focuses on the surface-initiated form of RCF causing pitting, also known as spalling. The characteristic surface crater is formed by a crack which initiates at one point on the surface. The crack then grows down into the material and simultaneously widens. The crack propagation eventually causes material to fall off, thus forming a sea-shell shaped pit. The damage characteristics are well described by for instance Tallian [1].
Numerous researchers have investigated the circumstances for RCF. The investigations started by Way [2], who among else found that RCF requires lubrication. Another prerequisite for fatigue crack initiation and propagation is the tensile stress. Pitting develops in both point and line shaped rolling contacts. This is intriguing since normal line contacts do not create any tensile stress in the surface [3]. They only produce compressive stresses at the contact and leave the surface outside the contact stress free. Therefore, they can not initiate or propagate fatigue cracks. Olsson [4] proposed the asperity point load mechanism for RCF. The idea is that small local asperities inside the surface roughness will create point loads surrounded by tensile stresses, which may create cracks. Alfredsson and Olsson [5] showed experimentally that point loads can create cracks that are very similar in shape to those creating the pits. Following those experiments Hannes and Alfredsson [6], and Everitt and Alfredsson [7], showed numerically that asperities in the size of surface roughness may create and propagate cracks into the shapes of RCF pits. Numerical studies incorporating the lubricant has shown that the asperity point load mechanism also explains why pitting is more common on surfaces subjected to negative than positive slip [8]. Such conditions exist in the dedendum of pinions. The asperity point load mechanism can thus explain why RCF only develops just below the pitch-line on the pinion, where slip is negative and moderate, and only grows in the forward rolling direction.
Coulon et al. [9] did an experimental investigation on artificially induced indents which caused micropits. The time it took for a micro-pit to develop was related to the stressed volume around the pits. Johansson [10] used artificial indents to investigate how different oils affects the number of cycles until pitting is formed. Nelias et al. [11] investigated both numerically and experimentally how the residual stresses around artificial indents affects the fatigue properties. Cheng et al. [12] found that the pile-up around indents and furrows had a much larger effect than the depth of the indents for the initiation of surface cracks. Gao et al. [13] extended their investigation to include diamond shaped Vickers indents as well as Rockwell indents and furrows. All these investigations suggest that it is the pile-up, the asperity, surrounding the indent that is the damage source, not the indent itself.
Morales-Espejel [14] presented a review in 2014 on the progress of understanding how the surface roughness relates to micro-pitting. It was concluded in the review that the mixed lubrication problem had not yet been properly solved. The goal of the current paper is to highlight some differences between asperities and indents in thermal elastohydrodynamic lubrication (TEHL) contacts. Hannes and Alfredsson [15] performed a parametric study on contacts with single point asperities using linear elastic fracture mechanics. Their investigation showed increasing crack growth life in rolling contacts for reduced surface roughness and increased compressive residual stresses. Another investigation of the effect or the residual stresses was peformed by Al-Mayali et al. [16]. Golmohammade and Sadeghi [17] coupled an elastic-plastic line model with damage mechanics to account for material degradation of pitting around a furrow. Sosa et al. [18] and Clark et al. [19] performed experimental investigations on the evolution of the surface roughness in rolling contacts. Clark et al. also found that the plastic deformation during the run-in process contributed to the formation of micro-pits. EHL experiments on the lubrication regimes by Hanssen et al. [20] show that continued running lead to almost complete separation of the surfaces. Vrček et al. [21] investigated micro-pitting and wear experimentally and found that a sufficiently large difference in hardness between rollers could eliminate micro-pitting.
This work continues the investigation on the effects of surface roughness in TEHL contacts with focus on well-defined micrometre high asperities. The goal was to attain deeper insights into the initiation mechanisms of RCF through experiments and analyses of individual well defined asperities. The numerical analyses include both the elastic-plastic run-in cycle and the TEHL over-rolling load cycles.

Experiments
The experimental procedure contained two stages: imprint of asperities and the RCF experiments in a twin-disc machine. Asperities were created on the test surfaces before the discs were hardened. In the soft state the test discs were rolled against a hardened disc with Rockwell type indents. Due to the high contact force 4-15 kN the indents imprinted 2-15 µm high asperities in the test surface. The subsequent RCF experiments were performed in a twin-disc machine.

Asperity manufacturing
The process of creating the asperities can be divided into three steps according to Fig. 1. During step 1) Rockwell indents were made on a hardened disc. Any pile-ups around the indents were removed by carefully grinding.
2) The hard disc with the indents was pressed and rolled against a soft unhardened test disc to imprint the asperities in this disc surface. Thereafter, the soft disc was case-carburized at 930°C with 1.1% carbon and then equalized at 850°C with 0.75% carbon before it was quenched in an oil bath. After rinsing off the oil, the specimen was annealed at 170°C. At this stage the test disc had hard asperities with an integral material structure with the rest of the disc surface which would be used in the RCF experiments, step 4) in Fig. 1. The manufacturing process is detailed by Everitt et al. [22].
Rockwell indents were also made direct into one asperity-disc and one counter-disc after heat treatment. The surfaces were left as they were with pile-up surrounding the indents. The idea was that the pile-up would act as large asperities and initiate RCF in accordance with the literature [10].

RCF experiments
A Wazau UTM 2000 twin-disc machine was employed. Fig 2a shows the test configuration with two discs in lubricated contact. The machine is described in the reference [10]. Both discs had a diameter of 40 mm, where the asperity/indent disc was flat and the counter-disc was crowned with 46 mm radius [22]. The result was an elliptic contact described in Table 1. Prior to the experiments, all discs were cleaned with heptane and ethanol. After that, the discs were mounted on drive spindles. Before the actual test, the bulk oil was heated to 90°C and circulated into the contact through an oil pump. Each test was ran with 1,5 litre of fresh oil. Two filters were employed after each other where the first had the grid size of 125 μm and the following the grid size of 25 μm in order to remove any fine particles. After the temperature of the bulk oil had stabilized, the test was started by ramping up the rotational speed of both discs to 3 000 rpm without any applied load. When both discs had reached 3 000 rpm, a radial contact load of 2 kN was applied. Then, the speed of both discs were adjusted to achieve the desired entrainment speed and SRR. The test continued until reaching the predetermined number of revolutions, when it was suspended.
Experiments in set-up C in Table 1 were aborted at intermediate stages to observe the development of the asperities with cycle number. In this case, the disc with the asperities was dismounted and cleaned with heptane and ethanol. It was then analysed with the 3D optical interferometer, Zygo 7300. The test discs were then mounted back on the same position and the test was restarted and continued in the same procedure as before. After the experiments, all discs were cleaned with heptane and ethanol before analysed with the optical interferometer and an optical microscope. The vibrations were monitored during the experiments since large pits cause a drastic increase in the noise and vibration levels. The vibration level was around 0.2 g and the experiments were set to stop if the vibrations got higher than 0.22 g. This never happened. Fig. 2a shows the set-up with the rolling direction (RD) pointing outwards, from the contact.
One important feature was the asperity position in the transverse direction (TD). This position was chosen so that the asperities and the counter-disc would be equally loaded over the whole contact width. Otherwise, the asperities may cause some tracks on the counter-disc [22]. The asperities were placed in a structured pattern following Fig. 2b, where the indicated distance D = 200 μm was selected to approximately agree with the asperity diameter ω. The diameter of the indents was the same as the diameter of the asperities they formed. Note that distance in the RD between indents was reduced in the figure in order to visualize the transverse displacement more clearly. Interaction between asperities was avoided with a 4 mm distance between the indents in the rolling direction.  Three different experimental set-ups were used, see Table 1. Set-up A included both asperities and indents. Set-ups B and C comprised asperities at different conditions. Key lubricant parameters are presented in Table 2. The lubricant data were interpolated from the experimental data in Appendix A and supplemented by values recommended by Larsson et al. [23] for the lubricant PAO B.

Numerical set-up
The experimental results were numerically analysed by sequentially combining a finite element (FE) model with the finite difference method (FDM). The FE model was used to analyse the run-in procedure in order to estimate the profile changes and the residual stresses around the defects. It used one load step with an elastic-plastic material description of the asperity. The second model utilized FDM to estimate the TEHL load cycles after run-in. Finally, a non-proportional and multi-axial fatigue analysis was performed based on the residual stresses from run-in and the TEHL stress cycle.

Determination of residual stresses
The commercial FE-programme Comsol 5.4 was used to simulate the run-in of an asperity in order to estimate the chap changes and the residual stresses. The initial asperity shapes were based upon the measurements. They were smoothed and made axi-symmetric to highlight the effect of the overall asperity shape and the rolling direction. The profile of the indent was simulated in order to capture all residual stresses. The indent simulation was performed with a rigid indenter with the shape described in the literature [26]. The indentation depth was specified so that the shape of the indent would match the measurements. In the experiments and in applications the defects meet a new spot on the counterdisc surface at each load cycle. The counter-disc was therefore modelled as a linear elastic cylinder to represent elastic steady-state behaviour after shake-down. Fig. 3 presents the FE-geometry and mesh, which consisted of 27 000 quadratic elements. The front side in Fig. 3 is the symmetry plane in the TD and on the rear side the transverse deformation was set to zero. The displacement of the remaining sides was prescribed to simulate the asperity over-rolling sequence.
The contact between the counter-and asperity-surface was modelled with the augmented Lagrangian method. Since the experiments started with a run-in phase at pure rolling, friction was omitted in the FE-contact. The lubricant was also omitted in this model. The nominal contact was changed from the elliptical shape in the experiments to a cylindrical line contact in the FE-model in Fig. 3. The change in contact geometry at equal pH resulted in an extension of the contact half-width to 0.4 mm for the running-in FE simulation. The mesh size transitions zones were modelled based on the work by Schneiders et al [27].

Asperity-disc
Counter-disc 6 mm 1/3 mm 3 mm An initial compressive stress of 200 MPa was added in both the RD and the TD to account for the casehardening [28]. A cylindrical coordinate system was used to incorporate these in the counter-disc, the top curved part in Fig. 3. The plastic behaviour of the material was described with the Swift material model where σy0 is the initial yield stress, n is the hardening exponent, ε0 is the reference strain and εeff.pl is the effective plastic strain. The material parameters are presented in Table 3.

TEHL load cycle
The numerical model utilized the very high lubricant viscosity inside the rolling contact and that the contact width to lubricant height ratio was more than 1000. The lubricant flow was modelled as laminar in the contact plane and with constant lubrication variables in the height direction. Thus, all variables were assumed constant in the height direction inside the lubricant. The numerical set-up of the lubricant was based on Reynolds equation [30] ( ) which is the foundation for most EHL simulations. The simulations were set-up as fully flooded and cavitation of the lubricant in the outlet was incorporated by the pressure limit p ≥ 0. In Eq. (2), h is the lubricant film thickness, η the viscosity, ρ the density and um = (ua+uc)/2 is the entrainment speed.
The local temperature field in, and around, the contact was computed based on the energy flow from hot to cool locations. Heat sources developed when mechanical energy was converted into thermal energy. The conversion mainly occurred through dry friction and shearing of the lubricant. This thermal energy was then dissipated through and transported by the lubricant and the contacting bodies. The heat flow was described by In Eq. (3) ua is the through thickness average speed of the lubricant, us the sliding speed of the contact, Γ the temperature, Cp the heat capacity and κΓ the thermal conductivity. In addition to Eqs (2) and (3), the lubricant was modelled as Newtonian with Roelands equation [31] together with the non-linear pressure-density relation formulated by Dowson et al. [32] and the linear pressure dependent shear limit formulated by Bair and Winer [33]. The solids were assumed to show a linear elastic material behaviour after run-in.
Eq. (1) was discretized with FDM as set-up by Huang [34]. The lubricant domain was resolved with 257 nodes in the RD and 97 nodes in the TD. A symmetry plane was used in the TD at the centre of the asperity. The global contact coordinates system was placed with the origin on the symmetry plane and at the centre of the contacting flat surface, see Fig. 4. The x-axis was oriented along the RD, the y-axis in the TD and the z-axis in the upward radial direction from the asperity surface. The origin of the z-axis was positioned at the nominal surface height. The flow profile in the vertical direction was described by the Poiseuille and Couette flow terms. The temperature fields in the solids were resolved with twice the spatial distance of the pressure in the horizontal planes and with 39 nodes each in the vertical direction. The vertical distance between metal nodes started at 0.5 µm, at the surface and increased with 0.5 µm for each node plane. A detailed description of the TEHL numerical set-up can be found in [8]. The model parameters are summarised in Appendix B.
The shear stresses at the surface of the asperity-disc were evaluated through (4) where lubricant was present. Metal contact was defined when the dimensionless film thickness H ≤ 80 • 10 -6 . The dimensionless film thickness was defined as H = hr/a 2 , where r is the equivalent contact cylinder radius and a is the Hertzian contact half width in the RD of the cylinder contact. The set-up was based on the formulation by Zhu and Hu [35]. Coulomb friction with µdry = 0.3 was used instead of Eq. (4) for the shear stress at metal contact. The width of the TEHL model was limited to 0.44 mm (-0.67 < Y < 0.67 in Fig. 4) in the TD in order to save computation time and still resolve the results at the asperity with details. The equivalent radius r was changed from 10 to 8 mm to keep the contact half-width a = 0.33 mm and pH = 2.3 GPa. The shapes of the defects were imported from the FE-program in order to get the shapes after run-in.

Fatigue evaluation
The stresses from the TEHL cycle, including the unloaded instance after over-rolling, were added to the residual stresses from the FE run-in simulations. The Findley criterion [36] was used for the fatigue evaluations of the multi-axial and non-proportional stress cycles in the contact surfaces. Earlier studies of hard contacts [37], [38] show that the Findley criterion is well suited for contact fatigue. The criterion was formulated as the Findley fatigue index where the fatigue limit σeF and the normal stress weight factor κF are the fatigue properties of the material. The critical plane was found by scanning all planes at 5° intervals for the one that maximizes the criterion in Eq. (5). If the index Fi > 1, then fatigue was predicted at that position. Since a higher index indicated a more detrimental stress cycle, the index also predicted where damage is most likely to initiate. The values of the material parameters σeF = 625 MPa and κF = 0.627 were based on previous work [7].

Results
The experiments in Fig. 2 were performed with parameters from Table 1 and Table 2. The results were analysed numerically. The local results at the asperities and indents, i.e. the surface defects, are presented in a local coordinate system with the origin at the centre of the investigated defect. The x-axis was in the RD, y-axis in the TD and z-axis in the outward radial direction from the defect surface. The origin of the z-axis was positioned at the nominal height of the tested surface.

Asperities
The asperities were produced by the imprint method described in Section 2.1, which allowed for different asperity heights. Since the imprint method gave quite rotationally symmetric asperities the numerical investigations were based on axi-symmetric asperities with shapes from the measurements. Fig. 5a shows the surface of a worn and representative asperity after 10·10 6 cycles of experiment A. The initial and worn cross-section profiles in the RD were laser measured and are presented in Fig. 5b. The initial asperity height was δinitial = 4 µm and the width ω = 100 µm. After 10·10 6 cycles δ = 1-1.5 µm. No surface cracks were found on the surface of the worn asperity-disc or the counter-disc when investigating them through an optical microscope after the experiment.

Experiment B
Experiment B was identical to experiment A except for slightly higher entrainment speed um · ex = 6.1 m/s and enlarged asperities with δinitial = 5 μm and ω = 200 μm. At some of the asperities a micro-crack appeared at the leading edge. Fig. 6a shows the surface view of an asperity with a micro-crack. Crosssections of 2 initial and 4 worn asperities are presented in Fig. 6b. The asperity height decreased to about 2 μm after 9·10 6 cycles. The figure also shows that some asperities, represented by the yellow and magenta lines, developed micro-pits at their leading edges.

Experiments C -Asperities
Three different C experiments were performed. These experiments utilized a different oil, with higher viscosity, in combination with a higher SRR than experiments A and B. Due to a shorter case carburisation time, the discs in this experiment had a slightly lower hardness and case depth than the discs in experiment A and B. Details of the experiments C1, C2 and C3 are presented in Table 4. Experiments C1 and C3 were interrupted for measurements of asperity cross-section profiles, with the purpose to follow the profile development with cycle number. These asperities were randomly placed in the TD without using the ordered pattern in Fig. 2b, which altered the loading conditions on them since some tracks developed on the counter-discs [22].

Experiment C1 -5 µm high asperities
The final shape of a representative asperity is presented in Fig. 7a. Rolling was from the left to the right with a SRR of -10% and causing the traction to be directed towards the left in the figure. The total profile change, from initial to the shape after experiment termination, of five different asperities are presented in Fig. 7b. The lines which has the same colour are from the same asperity. The solid lines presents the initial shapes while the dashed lines show the asperity shapes after 35·10 6 cycles. The initial height δinit = 4-6 μm was reduced to δfinal = 2-3 μm. The development of the red asperity profile in Fig. 7b is presented in more detail in Fig. 7c. Almost all shape change occurred before 12·10 6 cycles. Thereafter, the height remained constant throughout the experiment. Experiment C1 was analysed numerically with results presented in Fig. 8 and Fig. 9. The right-hand side of the red initial asperity profile in Fig. 7b was smoothed from small undulations, rotated for axisymmetry and used the as FE-asperity. Fig. 8a shows the simulated shape change due to one overrolling, which represented run-in, and the development of the asperity shape in the experiment. The FEsimulations showed a height reduction of about 1 µm while in the experiment, it was about 2 µm after 12·10 6 cycles. Fig. 8b shows the major principal stress σ1 along the symmetry-line of the asperity. The solid blue curve presents σ1 after case-hardening and the run-in over-rolling cycle on the asperity. Plastic deformation during the initial over-rolling caused a significant tensile σ1 at the asperity edges. Rolling was from left to right and the left side of the asperity, which entered the contact first, got a slightly higher residual tensile stress than the right side. Fig. 8c shows the residual σ1 distribution in the cross-section below the imprint and Fig. 8d shows the magnitude and direction of the residual σ1 in the surface. Fig. 8c and Fig.  8d show that tensile residual stresses were only present at the edge of the asperity and that the highest tensile stresses at the symmetry line were directed in the RD. Slightly outside those regions the tensile stresses were instead directed in the TD. The vertical direction in the remaining part of the material in Fig. 8c comes from the initial residual stresses due to the case hardening.
The elastic TEHL stress cycle was the result of contact pressure and traction from slip. The maximum of σ1 at each position over the cycle is included in Fig. 8b  The residual stresses and the cyclic stresses from the TEHL load cycle were combined and inserted into the Findley criterion in Eq. (5) for the fatigue evaluations presented in Fig. 9. Fig. 9a is divided in two halves were the top half shows a surface view of Fi for half of the asperity and the bottom half shows the asperity profile as a reference. The Fi profile displays a circular profile with very high values circumventing the asperity peak. Fig. 9b shows that the highest Fi value occurred at the trailing edge, which agrees with earlier predictions [8]. Since this value is above unity it means that the cracks were predicted to initiate here.
The figure contains Fi profiles computed for the total load, only for the residual stresses and only for the TEHL stress cycle. The Fi evaluation of the residual stresses are proportional to the σ1 since the residual stresses were constant giving τamp = 0 on all planes. The solid lines are along the symmetry-line and the dotted are the transverse maximum. Both the residual-and the cyclic TEHL stresses contributed to the total Fi. However, since the large stress components act in different directions and the Findley criterion is a nonlinear evaluation of the stress cycle, the full evaluation is not equal to the sum of the other two. The divided evaluation of residual and cyclic stresses in Fig. 9b shows which stress-state was the main contributor at different positions. The residual stresses gave a higher contribution near the leading edge of the asperity while the TEHL cycle gave a higher contribution near the trailing edge.

Experiment C2 -8 µm high asperities
Experiment C2 tested slightly higher asperities, δinitial = 7 -9 µm, compared to C1. Fig. 10a, Fig. 10b and Fig. 10c display the surface view of three asperities from C2 after the experiment consisting of of 8·10 6 cycles. Rolling was towards the right with negative −10% SRR and traction to the left in the figure. Fig.  10a shows an asperity that was intact after the experiment. The asperities were however large enough, and were subjected to sufficiently many cycles, to cause some cracks and micro-pitting on them, see Fig.  10b and Fig. 10c. The development of the cross-section profiles of three asperities are illustrated in Fig.  10d. Each colour represents a specific asperity. The initial profiles are illustrated with solid lines while the dashed lines show the final profiles at experiment termination. The asperity illustrated with blue lines developed a micro-pit at the summit. Results from the numerical analysis are presented in Fig. 11 and Fig. 12. The initial and over-rolled profiles in Fig. 11a show that plastic deformations reduced the height of the asperities with about 2 µm, but the height reduction in the experiment was a further 2 µm. Wear was thought to be a major contributor to the remaining height reduction. The solid blue line in Fig. 11b presents the residual σ1 after initial over-rolling and the dashed red line the cyclic σ1 from the TEHL over-rolling simulation. The residual stress was tensile at both ends of the asperity and the highest near the leading edge. The cyclic TEHL stress was the highest near the trailing edge. This is similar to the smaller asperities in experiment C1 presented in Fig. 8. The cross-section in Fig. 11c and the surface view in Fig. 11d of the residual σ1 also show similar behaviour as experiment C1 in Fig. 8. The fatigue evaluation for experiment C2 in Fig. 12 shows a clear distinction of the location with max(Fi) for the residual stress state and the TEHL load cycle but the total Fi is still symmetric.
The stresses in Fig. 11b and max(Fi) in Fig. 12 were compared with the location of the crack in Fig. 10c. The residual stresses seems to have been important for initiating the cracks near the leading edge, not the TEHL load cycle. Such observation suggested crack initiation was early. Furthermore, Fig. 11c indicated some tensile residual stress just underneath the asperity which could promote crack growth into and below the asperity. Fig. 11d show that the tensile stress was aligned with the RD at the leading edge. The state at the trailing edge was less distinct, but the highest tensile stresses were also here directed in the RD.

Experiment C3 -15 µm high asperities
These asperities were originally 14-16 µm high and 180 µm wide. After run-in, 0.1·10 6 load cycles, the height had be reduced to 6-8 µm, see Fig. 13a. Their heights were 5-7 µm after 0.8·10 6 cycles. After 3.4·10 6 cycles micro-pits had started to form near the leading edge on some asperities, as the one presented in Fig. 13b. After 6.7·10 6 cycles a new smaller pit had started to form. The test was terminated after 11·10 6 cycles and at this stage only 11 of the original 22 asperities were still intact. The rest were either flattened or had formed micro-pits similar to those in Fig. 13b. Fig. 13c shows a surface picture of the asperity presented in Fig. 13a after 11·10 6 cycles. This asperity was still intact, but reduced in height, after the experiment. The development of the asperity presented in Fig. 13a correlated well with experimental findings by Sosa et al. [18] and Clark et al. [19] who show that all shape changes occurs during run-in. Fig. 13d displays the surface of the asperity in Fig. 13b after 11·10 6 cycles. The pit had developed further and had removed most of the asperity. Rolling was to the right and friction was to the left. Fig. 13. Asperities after experiment C3: a) development of asperity from initial profile, b) development of asperity profile into pit, c) and d) surface view of asperities in a) and b) after 11·10 6 cycles.
The numerical results for experiments C3 are presented in Fig. 14 and Fig. 15. Almost all height reduction in the first 0.1·10 6 cycles was captured by the plastic deformations during one numerical over-rolling, see Fig. 14a. Figs 14b, 14c and 14d show that the residual stresses comprised a tensile component at the top of, as well as around, the asperity. The tensile residuals stress was believed to be a major contributing factor as to why the pits started at the leading edge of these asperities. The TEHL cycle in Fig. 14b did again provide a large tensile cyclic stress near the trailing edge of the asperity. The fatigue results for experiment C3 in Fig. 15a and Fig. 15b show that the Fi was the highest at the leading edge of the asperities, which agrees with the location of pit initiation in the experiments, see Fig. 13b. Fig. 15b confirms that the σ1 residual stress was the main contributor to the damage. It also shows that the Fi was high at the asperity summit where some small pits developed in the experiment. The residual stresses even reduced the fatigue risk caused by the TEHL load cycle at the trailing edge. The reason was that the different load cases caused tensile stresses in different directions. Fig. 14d shows that the residual σ1 was in the TD, which is perpendicular to the cyclic σ1 from the TEHL [8].

Indents
Artificial indents were placed along the symmetry line of both the flat asperity-disc and the crowned counter-disc. The indents were made with Rockwell cone indents directly on the surfaces. The surface profile of the indents were almost rotationally symmetric with radius ω = 200 µm and indent depth δ = 30 µm. The Rockwell cone indenters are standardized with 120° cone angle and a rounded tip with radius 200 µm [39]. The discs with indents were tested in accordance with experiment A in Table 1. Rolling was from left to right in Fig. 16a contact with 4% negative slip on the indent surface, i.e. contact shear traction to the left, against the rolling direction.
The indents on the asperity-disc caused RCF cracks when tested in the twin-disc machine. The cracks initiated after the trailing edge of the indents, the edge that entered the contact last, see Fig. 16a. The indent is the large almost round shape in the left central part of the figure. To the right of the indent, bright vertical lines illustrate cracks in the surface. Both the initial and worn cross-section profiles were measured in the RD across the indent. Fig. 16b presents the original and long term tested profiles of three indents in the asperity surface of experiment A. Initially there were 5-7 µm high pile-up around the indents. The pile-up was completely removed after 10·10 6 cycles. The crack position in Fig. 16a was compared to the pile-up profiles in Fig. 16b. The crack is located just to the right of the pile-up peak of the original profile. Out of 15 indents on the asperity-disc, 3 initiated RCF cracks such as the one in Fig. 16a. Fig. 17 shows half of the surface and the cross-section in the RD of another indent. Rolling was from the left to the right with traction towards the left. The cross-section in Fig. 17 shows that the crack turned away from the indent and propagated in the forward rolling direction, in a similar manner as surface initiated RCF cracks found in the literature [1], [2]. The crack developed behind the trailing edge of the artificial indent with pile-up. The RCF crack was inclined with a shallow angle β = 35° to the surface, see Fig. 17, which agrees with angles described in the literature [1], [2]. Similar results for indents with pile-up are reported in the literature, see for example the experiments by Gau et al. [13], Morales-Espejel and Gabelli [40] and Vieillard et al. [41]. Only 1 of 15 indents on the counter-disc initiated a crack. The crack in Fig. 18 initiated slightly on the lower side and to the left of the trailing edge of the indent. If the trailing edge of the indent with its pileup is regarded as an asperity, then the crack developed on the leading edge of this asperity. It did not grow as far as the crack in the asperity-disc in Fig. 16 and Fig. 17. The SRR was positive on the counterdisc since the traction was directed in the rolling direction, opposite to the asperity-disc in Fig. 16 and Fig. 17. The SRR of 4% resulted in 0.5·10 6 more revolutions for the counter-disc than the asperity-disc. The indents on the counter-disc were thus subjected to 10.5·10 6 cycles, compared to 10·10 6 cycles for the indents on the asperity-disc. Fig. 18. RCF at indent in counter-disc.

RD Friction
Previous simulations of indents show that it is the trailing edge of the indents which is the crucial part [7]. The locations of the cracks agrees thus with the asperity point load mechanism if the pile-up is viewed as an asperity. For asperities subjected to positive slip the cracks are predicted to occur on their leading edges, while the cracks are predicted to initiate at the trailing edges of asperities subjected to negative slip [8].
The numerical results of the indenting procedure coupled with the run-in is shown in Fig. 19. Fig. 19a show that the indent profiles were not affected by plastic deformations during run-in. The authors' conclusion was thus that wear was the major mechanism removing the pile-up around the indents. However, since the indents seems to be a bit more narrow after run-in, the material might have suffered from creep moving material from the pile-up down into the indents. Fig. 19b show that the residual stresses were large and tensile both before the leading edge and after the trailing edge. Only one set of FE-simulations were performed since the indent procedure was the same on both the asperity-and the counter-disc and the experiments started with pure rolling. The cyclic loading in the twin-disc machine gave different load cycles on the asperity-and the counter-disc since the sliding was negative on the asperity-disc and positive on the crowned counter-disc. The TEHL load cycle with negative slip provided large tensile stresses where the RCF cracks initiated, see Fig. 19b.
The data presented in Fig. 19c show that tensile residual stresses developed outside the high compressive stress state which formed directly underneath the indent. The surface view in Fig. 19d show that the stresses around the indent were oriented in the circumferential direction. The fatigue results presented in Fig. 20 show that fatigue damage were predicted at the edges with normal parallel to the TD instead of around the trailing edge. They show also that significant residual stresses developed where the cracks initiated, outside the trailing edge of the pile-up. The numerical investigation did not capture the location of crack initiation but they did capture that negative sliding was more detrimental than positive in the region of crack initiation. The TEHL load cycles were performed on indents with pile-up since the numerical run-in procedure did not cause any shape changes. Fatigue evaluation of numerical results on indents without pile-up show that the trailing edge was subjected to a more severe load cycle than the leading edge [7].

Discussion
Implications of artificially created asperities and indents were investigated with respect to RCF by both experiments and numerical simulations. The experiments were performed in a twin-disc machine and the numerical simulations included elastic-plastic FE-simulations of the run-in process combined with TEHL simulations of the continued testing.
The experiments were set-up to replicate the conditions of typical heavily loaded spur gears. The height of the asperities on the gear tooth surfaces were found to be around 2-3 µm [8]. The heights of the asperities were enlarged in the experiments in order to decrease the fatigue life of the surfaces. The enlargement was thought necessary since the discs in twin-disc machines has been found to survive up to six-times as many load cycles as gears subjected to similar conditions [42], [43], [44]. The life extension was confirmed by the experiments since the asperities with heights of 4-6 µm survived up to 35 million load cycles. The gear surfaces in the case study were investigated after only 13 million load cycles. They were already then subjected to severe pitting.
The experimental results with micro-pits initiating at the leading edge of the highest asperities, δ = 15 µm, correlated well with the predicted locations from the simulations. The results of the lower asperities, δ = 5 µm, showed however a disagreement between the experiments and the simulations. No cracks were found in the experiments while fatigue damage was predicted at the trailing edge. The prediction of fatigue damage at the trailing edge agreed with results in the literature [8]. The FE simulations showed that there were substantial positive residual stresses at the leading edge of the high asperities. The conclusion was therefore drawn that the tensile residual stresses were the major cause of this crack initiation. The residual stresses in front of the lower asperities seems to have been too low to initiate cracks, they were only about 70 % of the values caused by the higher asperities. The tensile stresses in front of the low asperities were similar to those behind the larger asperities.
Detailed simulations in the literature show that the elastic load cycle caused by surface roughness sized asperities should be large enough to initiate fatigue damage at the trailing edge of asperities subjected to negative slip [8]. This was confirmed by the present numerical TEHL results for the present loading conditions but not by the experiments. It was found that the cracks at the present high asperities initiated due to another damage mechanism. The presented numerical simulations showed that as the height of the asperities was increased, the failure mode changed from being driven by the TEHL load cycle, which was shown to cause fatigue at the trailing edge of the asperity, to being assisted by the residual stresses and thus causing fatigue damage at the leading edge. The experimental results agreed well with the simulations for the 15 µm high asperities but no cracks were found around the trailing edge for the lower asperities.
The elastic-plastic FE simulations showed that parts of the shape changing process could be explained by plastic deformation during run-in. The shape of the asperities after the run-in FE-simulations were however about 2 µm too high for the experiments C1 to C3. The remaining part of the height reduction was thought to be due to wear during the early cycles. Hansen et al. [20] showed that the surface profiles in TEHL contacts quickly changes to a profile where no metal contact is obtained, which suggests that the parts breaking through the lubricant rapidly wears down. Wear of the surfaces could be one reason for why the numerical simulations predicted fatigue damage for the small asperities where no damage was found experimentally.
The damage in the experiments on the indents with pile-up agrees with results in the literature on surface initiated RCF [10]. The cracks initiated after the pile-up of the indents subjected to negative slip and before the last pile-up for indents subjected to positive slip. The cut-through view of the crack presented in Fig. 17 showed that it turned away from the indent in the forward RD with a typical entry angle of β = 35° [6]. The numerical simulations showed that the indentation procedure caused high tensile stresses in the region where the cracks initiated. The same was true for the TEHL load case with negative slip. The positive slip load case on the counter-disc was shown to generate less than half as high tensile stresses in this region. This experiment also showed that negative slip was more detrimental than positive slip which was also captured by the numerical simulations. The current simulations predicted fatigue damage on the transverse edges of the indents. On reason for this could be that the simulations were based on a geometry with pile-up while the pile-up in the experiments got removed. Numerical simulations in the literature show that the trailing edge is the detrimental part of indents without pileup [7].

Conclusions
The asperities with initial heights of 4 -6 µm survived 35 million load cycles while the ones which initially were 15 µm high developed micro-pitting before 3.4 million load cycles. The micro-pits on the 15 µm high asperities initiated at the leading edge of the asperities. Detailed investigation of the stressstate showed that these cracks were primarily caused by the high tensile residual stresses from run-in in combination with some cyclic stresses from the TEHL cycle. The TEHL load cycle caused substantial tensile stresses at the trailing edge of the asperities, where the residual stresses were low. The predicted location of crack initiation correlated well with the experiments for the asperities which were 15 µm high. The numerical investigation predicted a shift in the location to the trailing edge for the lower asperities, where however no damage occurred in the experiments.
The experiment on indents with pile-up subjected to negative slip caused RCF crack initiation at the trailing edge of the pile-up, i.e. at the trailing edge of the asperity around the indent. These cracks displayed the characteristic RCF inclination angle β = 35° to the surface. The numerical investigation showed that both the indentation procedure and especially the TEHL cycle caused high tensile stresses in this region. The indents subjected to positive slip developed less damage. The only crack found initiated before the leading edge of the pile-up, at the trailing edge of the indent. Since the residual stresses were the same the change in the TEHL load cycle was the crucial part. Both the locations and the occurrences of the cracks caused by the indent pile-up agreed with predictions based on the asperity point load mechanism. The damage mechanism for cracks and micro-pits at high asperities seems thus to have been different from that at indents with pile-up.