Effects of Water Contamination on Micropitting and Rolling Contact Fatigue of Bearing Steels

Water in oil can have detrimental effects on the performance of rolling bearings [1–4] and can be present as either dissolved or free water [5]. Water completely dissolves in oil if the amount of water is less than the saturation point. On the other hand, free water occurs if the amount of water in oil exceeds the saturation level. Water contaminates oil through different means, including absorption and condensation from a humid environment [6], heat exchangers, chemical reactions, free water entry, etc. [7]. Water in lubricants can accelerate hydrogen generation that diffuses into the steel during rolling contact which promotes pitting, cracking, and fatigue failure [6,8], and because of its extremely low viscosity, can significantly reduce the lubricant film thickness in the contact.

Schatzbe and Felsen [1,2] reported that rolling contact fatigue life in precision ball bearings can be significantly reduced by as little as 0.01% water in oil. They proposed that microcracks in the raceway surfaces act as capillaries where water condenses, and aqueous corrosion in the cracks combined with applied stresses reduces fatigue life. Results of full-scale bearing testing performed by Cantley [3] supported a view that the rolling contact fatigue life of bearings was intimately tied to a lubricant’s capacity for water absorption. Specifically, Cantley reported that water concentrations beyond 100 ppm in SAE 20 oil exerted a detrimental effect on bearing life. Soltanahmadi et al. [4] studied the effects of water and humidity on micropitting tests performed in fully formulated oils. They found that micropitting and abrasive wear increased in proportion to the dissolved-water level in the lubricant and that whereas dissolved water increased the amount of micropitting, free-water suppressed micropitting by increasing the amount of mild wear. That is, the progression of micropitting propagation becomes suppressed since mild wear removes the nucleated micropits. In their paper and in this one, the description of mild wear refers to material removal and not a specific wear mode.

Water ingress in oil is a significant concern to wind turbine gearboxes [9]. The damage modes of micropitting and infantile rolling fatigue have been causes for a reduced reliability of wind turbine gearbox bearings [9]. Although the loading conditions on wind turbine gearbox bearings are consistent with those that can cause micropitting and premature rolling contact fatigue, it is of interest and the intent of this study to investigate the effects that water contamination may have on these problematic bearing damage modes.

Micropitting is a special type of surface-initiated fatigue that is sometimes experienced by gears and rolling bearings. Micropitting is the appearance of many micrometer-size pits on the raceway surfaces of rolling bearings. Micropitting progressively removes material from a raceway, which alters the stress profile of the surface. The compromised stress profile creates stress concentrations that can exceed the local yield strength of the material. Micropitting is a root cause for infantile raceway spalling of mainshaft spherical roller bearings in wind turbines [9]. Visually, micropitting is a dull, etched, or stained region on the raceway. The typical size of a micropit has a characteristic depth between 5 and 10 µm and a diameter between 10 and 30 µm. The pits are typically oriented at 10–30 deg to the surface [10].

Based upon tribological and microstructural experiments, Oila and Bull [11] have proposed a mechanism for micropitting. Plastic deformation of contacting asperities creates a region that extends beneath the asperities where the increase in dislocation density causes work hardening of a plastically deformed region (PDR). At the same time, frictional heating (temperature) and mechanical forces (pressure) cause carbon to diffuse creating a region that appears dark when a nital-etched cross-section is viewed with a light optical microscope. At the boundary of the PDR, recrystallization occurs, and dislocation pile up against the PDR boundary creating a slip band. White etching bands also tend to form in the martensite and below the dark region. With continuing stress cycles, dislocations accumulate in the slip band and a crack will initiate. Once initiated, the crack propagates along the PDR boundary until it reaches the surface, where it generates a pit.

Since plastic deformation of asperities is required to initiate the micropitting mechanism, conditions favorable for micropitting exist when the lubricant film thickness is insufficient to separate the asperities in the contact area. Or in other words, when the rolling bearing is operating in boundary layer lubrication, the λ parameter (the ratio of the lubricant film thickness (h) to the composite surface roughness of the mating surfaces) characterizes the lubrication regime as boundary, mixed, elastohydrodynamic, or hydrodynamic. It is in boundary layer lubrication (λ ≤ 1) where asperities intimately interact and generate localized stresses that can lead to PDRs [12–14].

Plastic deformation of asperities and the diffusion of carbon from the martensite are also exacerbated by occurrences of high slide-to-roll ratios between the rolling elements and raceways in rolling bearings operating in low λ conditions. Whereas rolling elements exhibit pure rolling while in the load zone of a radially loaded bearing, the loss of traction outside the load zone leads to an increase of sliding contact between the rolling elements and the raceways.

Under boundary lubrication, mild wear competes with the micropitting process [15]. That is, mild wear can remove the surface layers where microcracks can initiate.

Strategies to mitigate micropitting in rolling bearings include improvements in lubrication conditions, improvements in surface finish, and tribological coatings that work to reduce the asperity/asperity interactions that cause micropitting [16–20].

Rolling contact fatigue involves the initiation and propagation of a crack that eventually leads to a gross loss of material on a raceway or on a rolling element. This loss of material is referred to as a spall. Cyclic stresses and strains can cause cracks to initiate at defects residing on or beneath the surfaces of raceways and rolling elements.

Surface defects that can generate cracks are asperities that protrude through the lubricant film and suffer from plastic deformation in the contact area. Even if the lubricant film is thick enough to fully separate surface asperities on the raceways and rolling elements during normal operation, the presence of hard particles (debris) in the contact area can result in plastically deformed indentations or craters. Surrounding the craters are protrusions or raised areas that can protrude through the lubricant film. The enormous contact stresses encountered by the protrusions create situations where cracks are initiated.

Subsurface defects are inclusions and microstructural alterations of the material. Microstructural alterations (e.g., martensitic decay) of bearing steel are caused by accumulated subsurface damage from cyclic stress and strain. Subsurface defects such as carbides, nonmetallic inclusions, and voids are often the sites of crack initiation.

Inspections of components that have experienced rolling contact fatigue can in many cases indicate whether a spall originated from a subsurface or surface defect. Whereas spalls with an oval shape typically indicate a subsurface-initiated failure, spalls with a V-shape are usually surface-initiated.

As a result of high cycle Hertzian stresses during bearing operation, the near-surface regions of the microstructures of steel alloys experience significant changes. These alterations are concentrated at a depth corresponding to the maximum shear stress (τ_max) encountered by the rolling element-raceway contact. A light optical microscopy (LOM) image of the nital-etched subsurface of an AISI 52100 ball specimen subsequent to rolling contact fatigue testing from Šmeļova et al. [21] is shown in Fig. 1.

Microstructural alterations have occurred as indicated between the two white lines in Fig. 1(a). The dark area, or dark etching region (DER), extends to a depth of about 0.76b, where b is the length of the semi-minor axis of the contact ellipse. The figure indicates that the DER resides approximately 80–600 μm below the raceway surface. Figure 1(b) shows an enlarged area of the DER. The DER is softer (∼53 HRC) than the original martensite (∼61 HRC) and consists of a ferritic disk (thickness ∼ 0.1 μm, ∼1 μm width) mixed with residual parent martensite. The ferritic phase contains a random distribution of excess carbon content that is equivalent to that of the initial martensite. The fine carbides in the original microstructure are believed to reprecipitate into ɛ-carbides (thickness ∼ 1 μm by ∼15 μm) that are sandwiched between disk-shaped regions of ferrite [22]. After prolonged stress cycles, white etching matter (WEM) may appear in the DER. Evidence indicates that WEM is formed as carbon migration out of the martensite precipitates as cementite. The hardness of the region containing the WEM is as much as 9 HRC less than the surrounding regions.

Cracks can initiate at highly concentrated surface stresses and propagate at shallow angles of 15–30 deg to the surface [23]. At a critical depth, the cracks branch toward the surface, and a pit is formed. The elevated surface stresses associated with a pit generate a proliferation of cracks with continued stress cycles, leading to a spall.

White etching matter can be found on crack interfaces within and external to the DER. Voids created at nonmetallic inclusions in the subsurface can initiate cracks in the matrix. The cracks tend to propagate at ∼45 deg to the raceway tangent, or at the angle of the maximum shear stress (τ_max), and in the direction of the rolling motion. As the faces of the crack rub together, intensive mechanical deformation occurs at the fractured surfaces. The mechanical deformation converts the tempered martensite into a region containing ultrafine grains, tiny voids, and fragmented cementite particles that dissolve readily. When etched and examined with LOM, these altered regions appear as wings emanating from the inclusion. Since the general appearance of these phenomena resembles a butterfly, these microstructural alterations are commonly referred to as butterflies.

White etching matter, also called white etching cracks (WECs) and white structure flaking (WSF) have been found in bearings that have experienced rolling contact fatigue in many different applications including various locations in the gearboxes of modular wind turbine designs [24]. The frequency of premature rolling fatigue failures has been unacceptably high in large-scale wind turbine gearboxes, and many examinations of the failed bearings have revealed WEM at crack interfaces in these bearings. For example, Fig. 2 from Ref. [9] shows the inner ring of a cylindrical roller bearing taken from the gearbox of a 1.5 MW wind turbine and an extensive crack network containing WEM residing subsurface.

Proposed root causes of premature rolling fatigue failures of wind turbine gearbox bearings include corrosion, tribochemistry, sliding, impact loading, vibration, bending stress, electrical currents, and hydrogen diffusion [25,26]. Evans [24] summarized the different proposed mechanisms associated with subsurface- or surface-initiated cracks. The subsurface initiation theories suggest that cracks initiate at defects and inclusions (predominately small or short-sized containing oxide) in the maximum shear stress region and that the small cracks link up to form crack networks. The surface initiation hypotheses suppose that cracks initiate from damage or microcracks on raceway surfaces in the vicinities of highly concentrated shear stresses. The presence of diffusible hydrogen in steel enables accelerated crack propagation rates [27–30] and may promote the formation of WEM at crack interfaces [25,31–33].

In this research, the effects of water in oil on two specific kinds of failures in bearing steel, micropitting, and rolling fatigue are studied. It is hypothesized that free or diffusible water in the lubricant can cause capillarity where water becomes attracted to surface-initiated microcracks, or through hydrogen generation and diffusion into the steel, can accelerate the crack propagation rate. Both processes should promote the occurrence of micropitting and surface-initiated fatigue damage in bearing steel.

Test conditions are listed in Table 1. Previous studies [15,34,35] have shown that the λ condition of 0.2 and the slide/roll ratio (SRR) of 10% are effective at generating reproducible micropitting wear on AISI 52100 specimens with martensitic structures. Water was ultrasonically dispersed in the oil, and a series of tests with different water concentrations was performed (Table 2). The areal analysis of micropitting damage, wear depth, surface profile, and surface roughness of the rollers were studied using LOM and scanning electron microscopy (SEM). Water concentration in the oil was quantified by a Karl Fischer titrator. Note that the baseline oil contained ∼11 ppm water.

Rolling contact fatigue testing was carried out using thrust needle roller bearings (AXK 1226) and a thrust bearing tester is shown in Fig. 4. The bearing washer raceways and needle rollers were made of AISI 52100 bearing steel with martensitic structures. The lubricant was a PAO ISO VG 68 base oil with rust and oxidation inhibitors and with a water contamination of less than 10 ppm (the lower limit of the Karl Fisher Titration test). Tests were conducted in the base oil with and without 600 ppm water added to the oil using the conditions and categorized tests identities listed in Tables 3 and 4, respectively. The termination criterion of a test was initially expected to be an exceedance of a torque level that surpassed the set threshold or 14 million revolutions, normally caused by pitting, spalling on the raceway, or the breakage of rollers.

Under boundary lubricated conditions (λ ≤ 1), mild wear competes with micropitting. In rolling/sliding contact conditions, micropitting risk is reduced as the wear-rate increases. It was found that mild wear can reduce the roughness of the surface and remove the layer where microcracks initiate, thereby suppressing the mechanisms that create micropitting [20]. Some researchers have found that water can affect the formation or decrease the thickness of tribo-films typically generated by anti-wear additives (not present in this study) thereby affecting the rate of wear [4,37]. The effect of water on the wear was analyzed under testing conditions shown in Table 1.

Wear of rollers subsequent to testing was obtained by measuring the surface profile of the rollers before and after the tests with a Zygo NewView 7300 3D Optical Profilometer. Figure 5 displays the friction measured in situ during the MPR test (Fig. 5(a)) and a measurement of the material removed from the rollers from the three tests after 0.55 million cycles (Fig. 5(b)). We claim that there is no statistically relevant difference in the three friction traces since the data fall within the scatter in the individual traces. In all three tests, about 14 μm was removed from the roller radii during the 0.55 million cycles. Notably, although the water concentrations in the oil were different, all three tests had similar friction coefficients and similar wear. These results suggest that the various amounts of water added to the oil in these experiments neither did not significantly alter the bulk properties of the lubricant nor did the water significantly affect the wear-rate of the steel.

Micropitting formation on rollers tested in oils with different water concentrations is compared in Fig. 6. Each column has three LOM images collected from the same roller at different locations after 0.55 million cycles. Any minor differences in the specular appearances of the images are likely due to the exposure time required on the 3D optical profilometer to adequately represent the micropits on the specimens. Whereas meager micropitting is evident in MPR1 (11 ppm dissolved water), mild micropitting is present in MPR2 (55 ppm dissolved water), and severe micropitting has occurred in MPR3 (229 ppm free water). The image of the roller from test MPR2 shows several scoring marks, which probably were caused by some of the larger wear debris particles that happened to pass through the contact area. To quantify the surface damage, the roughness of rollers was measured before and after testing (Fig. 7). Roller roughness was found to increase with increasing water concentration in the oil. The roughness is generated mainly by the density of micropits in the three regions evaluated for each roller specimen. This result indicates that water in the lubricant, either dissolved or free water, significantly affects the generation of micropitting. That is, the mechanism of micropitting increases with the concentration of water in the oil.

Micropitting damage was further examined with SEM (Fig. 8). From the SEM images, it is observed that the microcracks initiated along the boundaries of plastic deformation regions (PDR), in agreement with the mechanism proposed by Oila and Bull [38]. Propagation of the cracks along the PDR boundaries eventually liberated the PDR regions from the surface and formed micropits.

Mechanisms of the effect of water contamination in oil on the formation of micropitting are proposed in Fig. 9. Under the positive (for MPR) SRR (slide–roll ratio), cracks initiate at the inlet region (under tensile stress) and are kept open when entering the contact region, allowing the high-pressure oil to enter the cracks. Capillary condensation of water may occur in the microcracks (Fig. 9(a)), or water accelerates hydrogen generation by catalytic reactions with the fresh metal surface (Fig. 9(b)). Both mechanisms can reduce the threshold for crack propagation and accelerate micropitting.

No evidence of WEM was observed in the MPR test specimens neither used in this study nor was any expected. Gould and Greco [39] were able to produce fatigue failures on their MPR specimens with WEM located on crack interfaces, but those experiments where WEM was produced were performed under conditions that generate fatigue failures and not micropitting wear. Although WEM was observed in their tests performed at −30% SRR, no WEM was found on specimens at lower levels of negative SRR and for positive SRR up to 30%.

The effect of water contamination on rolling contact fatigue and concomitant WEM generation was investigated by testing needle roller bearings in a thrust bearing test rig using PAO ISO VG 68 base oil. Table 5 summarizes the key testing conditions and results of this experimental portion. The contact pressures utilized in the rolling contact fatigue tests differ from those used in the micropitting tests due to the differences in the contact areas of the specimens used in the two test types and more importantly, the goal of the rolling contact fatigue tests was to produce fatigue spalls as opposed to micropitting wear. All tests except for test RCF5 were terminated after bearing failure that was associated with the torque exceeding a critical level. Test RCF5 was suspended after completing 14.1 million cycles with no failure. The initial test plan was to terminate tests after achieving 14 million test cycles if the torque did not exceed a critical level. Since no fatigue failure occurred in RCF5, the termination of test RCF6 was determined to occur when a fatigue spall originated. In the case of RCF6, a fatigue spall became evident through torque increases after 57.3 million test cycles. In every case of bearing failure, the source of the failure was determined to be spalling in one or more of the needle roller assemblies, with most of the spalls occurring near the outside end of the rollers where the SRR is the largest.

Figure 10 displays LOM images of a circumferential cross-section and an axial cross-section of a nital-etched roller from RCF4. Both images display near-surface DER and isolated regions of WEM.

Inspection of the data (i.e., RCF1 through RCF4) displayed in Table 5 indicates that as expected, bearing life decreased with increasing contact pressure. Figure 11 displays LOM images of needle rollers from RCF1, RCF2, and RCF3. In tests performed at higher contact stress (RCF1 and RCF2), WEM was found near the roller surfaces that had spalling or cracks, while WEM was found in or around the DER in a roller that was tested at a lower contact pressure (RCF3). That is, at higher contact stresses, WEM appears to be formed after cracks have developed, while at lower contact stresses, WEM emerges from microstructural alterations occurring in the martensitic matrix.

Figure 12 displays LOM and SEM images of a failed roller from RCF4 that experienced 14.1 million cycles at 1.39 GPa in oil with 600 ppm water added. The DER is clearly observable in the circumferential cross-section, and WEM resides adjacent to and within the DER. The SEM image shows that the WEM was formed on the crack interface.

Light optical microscopy and SEM images in Fig. 12 can be compared to those of Fig. 13 taken from a roller removed from RCF5 that achieved 14.1 million cycles at 1.39 GPa in oil without the added water. A DER band clearly exists between 50 and 150 μm below the surface, which is in the vicinity of the maximum shear stress. No cracks or WEM were observed in RCF5 rollers.

All the bearings failed due to spalls appearing on rollers in the tests performed in oil with 600 ppm of water. Cross-sectional images of failed rollers displayed DER, regions where the martensite microstructure was transformed. Within the DER, WEM was found on cracks that appeared in most cases to propagate from the DER.

Finally, RCF6 was performed at 1.39 GPa in oil without added water and was run until failure occurred after 57.3 million cycles. Figure 14 displays LOM and SEM images of a failed roller from the test. Near-surface DER is present, and WEM appears at various locations along cracks.

To ascertain the effect that 600 ppm of free water in the ISO VG 68 oil had on rolling contact fatigue life, a comparison of estimated L₁₀ values calculated according to the ISO 281 standard (in oil with no water added) to the results listed in Table 5 for the tests performed with 600 ppm water added to the oil is made and shown in Table 6.

It is important to point out that while the ISO 281 calculations produce L₁₀ values, there are no Weibull statistics associated with the experimental data presented here. Since it was determined that the RCF3 test was terminated as a result of a broken roller rather than a spall, the failure mechanism for this test appears to be different from the remaining tests and may justify the exclusion of the 1.41 GPa value from the fitting process used to develop Eq. (4).

Several observations can be made of the results of the rolling contact fatigue testing. The thrust bearings that failed suffered from spalls or a fracture on the rollers that were located in the region of high sliding contact; when WEM was present, it was found on cracks that appeared to lie within or adjacent to the DER; corrosive reactions were not found to have taken place between the water and the bearing steel; and to a very good approximation, the fatigue life of the AXK 1226 thrust bearings was reduced by free water contamination according to Eq. (5).

Results of these experiments demonstrated that the presence of water (even a small amount) in lubricant can be considered as an important factor to accelerate micropitting and rolling fatigue. The results presented herein pertain to AISI 52100 bearing steel with a martensitic structure. As pointed out by Bhadeshia [22], AISI 52100 bearing steel with a bainitic structure has ductility larger than the minimal plasticity exhibited by the tempered martensite structure in this steel. As a consequence, the fatigue life of lower bainitic bearings operating in water-containing oils exceeds that of quenched and tempered martensite. High chromium-containing bearing steels such as AISI 440C suffer much less deterioration of fatigue life in water-containing environments than low-alloy ferritic steels. It has been proposed that the high chromium concentration in these materials retards the diffusion of hydrogen into the steel [22]. Two mechanisms were proposed to describe the effect that water has on micropitting: capillarity and/or hydrogen embrittlement. However, the results of experiments performed here are insufficient to elucidate definitively how the free water in the ISO VG 68 oil reduced the fatigue life of these particular thrust needle bearings.

Some of the mechanisms on how water affects rolling contact fatigue are:

Hydrogen-induced fracture: Water enters microscopic cracks by capillary forces and liberates hydrogen that diffuses into and weakens the steel.

Corrosion: Pitted raceways and rolling elements disrupt the elastohydrodynamic oil film.

Restricted oil flow: The polarity of water collects impurities within the oil and forms a sludge that impedes the flow of oil into the contact.

Reduced oil viscosity: The low viscosity of water (∼1 cSt) cannot support high-pressure loads. When water enters the contact between the rolling elements and the raceways, the lubricant film strength collapses. Water can also turn into superheated steam in the load zones further disrupting oil films and potentially fracturing surfaces.

No corrosion of the rolling elements or washer raceway surfaces was observed on the bearings in this study, and no evidence of sludge was observed subsequent to any test. It is therefore concluded that both hydrogen-induced fracture and reduced oil viscosity could account for the fatigue life reduction of the bearings tested in this study.

The authors wish to acknowledge Dr. Aaron Greco and Dr. Benjamin Gould of Argonne National Laboratory for useful discussions and the Department of Energy for the financial support of this project (DE-AC02-06CH11357/4F-31082). Contributions and useful discussions were provided by the students and faculty of The Timken Engineered Surfaces Laboratories (TESL).

There are no conflicts of interest.

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.