Friction and Wear Volume Measurement
During the wear tests, the coefficient of friction was recorded and wear volumes were estimated using surface profile measurements as described above. It was found that the wear volume of the cylinder was three orders of magnitude less than the wear of the steel disk at 10 N load. In case of the brass disk, it was found that the brass transfer film formed on the surface of the cylinder and no significant wear of the cylinder was detected. Therefore, it was assumed that most of the analyzed particles originated from the disks.
The regime of lubrication was assessed using \(\lambda = {\raise0.7ex\hbox{${h_{ \hbox{min} } }$} \!\mathord{\left/ {\vphantom {{h_{ \hbox{min} } } {R_{q} }}}\right.\kern-0pt} \!\lower0.7ex\hbox{${R_{q} }$}}\) [37], where \(h_{ \hbox{min} }\) is the minimum film thickness obtained by Dowson–Higginson equation [38, 39]. For the case of steel–steel contact at 2.5 N load \(\lambda = 0.051\), and since \(\lambda < 0.1\) the boundary lubrication regime was encountered [40]. Under higher loads or with brass material, \(\lambda\) is even smaller, and therefore, all the considered tests operated in boundary lubrication regime. It can be seen from Fig. 5 that the coefficient of friction does not change significantly with load for both material combinations. This suggests that the dominating wear mechanism will also be the same for all measurements. The friction with brass disk is approximately twice as high as with the steel disk.
There was no significant variation of the friction with temperature. This was to be expected, since, in the considered range of temperatures, the mechanical properties of the materials are not influenced significantly. A COF-temperature dependence is also not expected from a possible formation of a tribo-layer. In these tests, no extreme pressure and anti-wear additives are used. However, even for additive-rich oils, in the range of temperatures considered, friction will not depend on temperature [41]. In this range of temperatures, the oxide type remains the same [42].
Results of wear volume measurements are shown in Fig. 6. For both materials, there is a linear relation of wear volume with load. The wear of the brass is higher, since it has twice as high COF as the steel. The wear volume of brass as a function of load increases stronger than that of steel disk.
Wear coefficients were obtained from the Archard’s wear equation in the following form [43]:
$$W = k \times F \times s,$$
(3)
where \(F\) is the applied normal load, \(s\) is the sliding distance, \(W\) is the wear volume and \(k\) is specific wear rate. It was found that the mean value of \(k\) for steel disks was around \(350 \times 10^{ - 7} \, {\raise0.7ex\hbox{${{\text{mm}}^{ 3} }$} \!\mathord{\left/ {\vphantom {{{\text{mm}}^{ 3} } {\text{Nm}}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\text{Nm}}$}}\) and for brass disks \(800 \times 10^{ - 7} \, {\raise0.7ex\hbox{${{\text{mm}}^{ 3} }$} \!\mathord{\left/ {\vphantom {{{\text{mm}}^{ 3} } {\text{Nm}}}}\right.\kern-0pt} \!\lower0.7ex\hbox{${\text{Nm}}$}}\).
From Fig. 7, it can be clearly seen that for both materials there is a slight change of the specific wear rate coefficient with load. For brass, there is a slight increase, whereas for steel a decrease.
The evolution of surface roughness during the wear test was studied as a function of load in a following form:
$$\delta R_{q} = \frac{{\left( {R_{{q \,{\text{worn}}}} - R_{{q\, {\text{initial}}}} } \right)}}{{R_{{q\, {\text{initial}}}} }} \times 100\,\% ,$$
(4)
where \(\delta R_{q}\) is the increase in roughness in %, \(R_{{q\, {\text{initial}}}}\) and \(R_{{q \,{\text{worn}}}}\) are roughness of initial and worn surfaces.
Variation of \(\delta R_{q}\) with load is given in Fig. 8. It can be seen that for 2.5 N load, the roughness after the test somewhat decreased for both materials. However, for the loads of 5 and 10 N, the “worn roughness” is significantly higher than the initial for brass. In case of steel, the worn roughness is lower than the initial in all cases, so the surface was smoothened during the wear test.
SEM/EDS Measurements
The major goal of the particle isolation procedure is to collect wear particles from the oil for further analysis. Firstly, it was confirmed by the presence of iron in the collected particles that they originate from the base material (in this case of the steel–steel) sliding pair.
The SEM/EDS results are shown in Fig. 9. A SEM image of a particle is shown in Fig. 9a and the corresponding EDS element mapping in Fig. 9b. It can be clearly seen that iron is present in the particle. There is also C (carbon) present, which originates from the carbon tape used for preparation of the sample. The presence of oxygen shows that the particle is oxidized. The measurement of elemental composition was performed for several particles, and iron was found in all of them, and therefore, it can be concluded that the wear particles can be collected using the developed isolation procedure.
DLS Measurements of the Particles
Significant effort in this study was devoted to the DLS analysis of the size of the wear particles. First, a measurement (including the particle isolation procedure) was taken with fresh oil which confirmed that no particles were observed. Coupled with the SEM/EDS measurements, it proves that the isolation procedure does not introduce artifacts in the measurement.
Figure 10 shows the relation between the characteristic particles radius (the radius obtained from DLS) on load. This figure shows that there is a trend in the particles radius with load in the case of steel. The difference of the mean radius at 10 N and at 2.5 N is approximately 7 %. This does not apply to the case of the brass disks. Apparently, the size does not depend on the load here. The particles have smaller radius roughly by 20–40 nm than formed in a steel–steel pair.
Figure 11 shows the characteristic radius as a function of temperature for the case of steel disks. Rubbing at different temperatures gives wear particles with the similar radii.
By dividing the total wear volume by a volume of a characteristic wear particle, the total number of particles can be found. The results are shown in Fig. 12a.
Dividing Archard’s Eq. (3) by a volume of a characteristic particle results in the following relation:
$$N_{\text{p}} = k_{N} \times F \times s,$$
(5)
where \(N_{\text{p}}\) is the number of particles and \(k_{N} = k/W_{\text{p}}\) is the specific wear particle coefficient with \(W_{\text{p}}\) being the volume of characteristic particle.
The calculated values of \(k_{N}\) as a function of load are shown in Fig. 12b. It should be noted that \(k_{N}\) does not depend on load significantly for brass. This makes it convenient to estimate the total number of particles without knowing the mean size of a particle in similar conditions.
AFM Measurements of Steel Wear Particles
After the DLS measurements, several samples from the tests with steel disks under 2.5 and 10 N load were analyzed using AFM to obtain 3D information regarding particle size. Overall, 2500 particles were analyzed with nearly 500 particles for the samples at 10 N and 2000 particles for the samples at 2.5 N. Based on these measurements, the distributions of length, width and thickness were built.
In order to compare the results of AFM measurements with those obtained by DLS, equivalent radii of the particles were calculated. These are based on the radius of a spherical particle with the same volume given by:
$$R_{i} = \left( {\frac{{V_{i} }}{\pi }\frac{3}{4}} \right)^{1/3} ,$$
(6)
where \(R_{i}\) is the equivalent radius of the particle and \(V_{i}\) is its volume obtained from AFM analysis. The probability distributions of the equivalent radius of particles obtained under 10 N load are given in Fig. 13.
The number-weighted mean value of the equivalent radius was found to be around 175 nm, whereas the volume-weighted mean value was found to be 358 nm. It means that most of the volume is composed by a small amount of relatively large particles, whereas the number of smaller particles is dominant. DLS results for the same sample showed an equivalent radius of approximately 265 nm. The tendency of the DLS approach toward higher values of particle’s size compared to AFM was already reported in the literature [44]. The equivalent hydrodynamic diameter may be biased toward higher radius particles, since it is proportional to the scattered intensity, and for small particles, it is proportional to the sixth power of the radius [35, 44]. It can be noticed that the distribution of the particles size is not Gaussian. However, the distribution is mono-modal. This indicates that there are only limited wear particles originating from the cylinder, otherwise it could be expected to get at least two peaks in the distribution, corresponding to the disk and to the cylinder.
The situation is similar for the particles generated at 2.5 N, see Fig. 14. The total wear volume is dominated by the small amount of large particles, and this effect is even more pronounced compared to the samples tested at 10 N.
The distributions of the length, width and thickness of the particles were also obtained for both loads and both materials. Qualitatively similar distributions were obtained, and for brevity only results at 10 N for steel disks are shown, see Fig. 15. The thickness of the particles was found to be approximately 4–5 times less than the width, which means that the particles are flat. It can be noticed that the distributions are negatively skewed and diverge from the Gaussian distribution, i.e., there are relatively many particles with a small radius.
It is also necessary to compare the surface area-weighted mean radius of the particles. Comparison of the distributions is shown in Fig. 16. In both cases, the number-weighted mean radius is smaller than the surface area weighted. This fact and the fact that the shape of the distributions show that most of the surface area comes from the large particles, although the number of the small particles according to number-weighted distributions gets larger. Therefore, even though the number of small particles gets higher, their impact on the total surface area or wear volume becomes less.
Finally, comparison of the DLS and AFM data is shown in Fig. 17. A slight trend of the increase in the particles size with load can be found from both DLS and AFM results for steel disk. In case of brass disks, DLS results do not show such trend and the obtained radius stays approximately constant with load. The AFM results show smaller particles sizes due to bias of DLS toward larger particles. At the same time, AFM also has a bias, although it is apparently less expressed. Due to the height threshold, a number of small particles are not considered, next to the fact that in AFM only the upper contour is obtained, and thus, volume of undercuts is added to the wear particles. Since the number of these particles is relatively large, a large number of particles are not considered, which leads to the bias of AFM results toward larger sizes. This bias is more pronounced for low loads, when the size of the particles is close to the height threshold.
Discussion
The main challenge of the wear particles analysis lies in the preparation of the samples and related artificial filtering. There is a possibility of the wear particles to aggregate into larger clusters, and sometimes these clusters may not be broken by the sonication [29]. Therefore, the obtained results can be considered as the upper bound estimation for the wear particles size. The particle isolation procedure also introduces a filtering of particles smaller than approximately 25 nm radius due to the limited centrifugation time. It should also be noted that hypothetically there is no minimum wear particles size (the theoretical smallest possible wear particle is an ion).
On the other hand, there is no need in putting efforts to capture smaller particles during the centrifugation, since both DLS and AFM techniques introduce coarser filtering. In case of DLS, the particles with the radius smaller than 100 nm will have significantly less weight in the determination of the mean value of the wear particle. This results in relatively large mean particle radius observed by DLS. In case of AFM, the filtering is introduced explicitly through the height profile threshold. This threshold is necessary to distinguish the particle from the background roughness and possible noise. In this work, the threshold was taken to be 50 nm. Due to filtering, it is also clear that the obtained values of the wear particles size are larger than the actual sizes in case of both DLS and AFM analyses.
On the other hand, the contribution of the smallest particles to the overall wear volume or surface area of worn particles sometimes can be limited. It can be argued that in the presence of large particles which make most of the wear volume, the mechanical damage from wear particles can be ascribed to these large particles. They are capable of making relatively large dents, scratches and other mechanical damage, as in case of large abrasive particles [45]. Smaller particles, within the range of the surface roughness, are going to create relatively minor damage [46, 47], since they may carry less load. The fact that the number of small particles is high but their contribution to the total wear volume is low is found by the comparison of the number- and volume-weighted distributions obtained using AFM, see Figs. 13 and 14. Moreover, with the decrease in the overall mean particles size, this effect becomes even more pronounced. Wear particles also play significant role in chemical processes, such as chemical deterioration of the grease in bearings [11]. The degree of catalytic effect of wear particles on chemical reactions is mostly determined by the total surface area of the particles rather than by their volume [48]. From the AFM results regarding the surface area-weighted particle size, it can be concluded that the contribution of the small particles to the total surface area of the worn particles decreases rapidly with a decrease in the particles size, see Fig. 16. In addition, it must be noted that relatively large particles will always be generated in the system in the running-in stage of the wear process [8].
Based on the results of AFM measurements, a slight dependence of the particles size on load can be observed for both materials. This behavior is confirmed by DLS for steel samples. It should be noted that with the decrease in the applied load, the size of the particles becomes smaller and the effect of the filtering (both by AFM and DLS) becomes more pronounced. Therefore, the actual dependence of the size on load may be more noticeable, especially at lower loads. Based on DLS results for brass disks, it is not possible to draw a conclusion on the relation between load and size. This may be due to resolution of the DLS and only slight changes of the particles size.
AFM measurements also revealed that the shape of particles is characterized by a high aspect ratio (around 4–5) meaning that the particles are flat, as it is frequently reported in the literature [1, 49]. The formation of flat particles suggests that severe plastic deformation takes place on the surface and subsurface in the presence of a tangential load. The wear particle may form in such regions [50, 51].
Specific wear rates obtained in these measurements indicate a running-in stage of the sliding friction [52]. It is suggested that most of the wear volume is generated during this initial stage [22, 53], and therefore, it can be concluded that most of the particles are also formed during the running-in stage. In the present work, the values of \(k_{N} \approx 330 - 750 \,\frac{\text{particles}}{{{\text{mm}}\,{\text{N}}}}\) for steel and \(k_{N} \approx 1600 - 1900 \,\frac{\text{particles}}{{{\text{mm}}\,{\text{N}}}}\) for brass were obtained. These values then can be considered as characteristic to the running-in stage of the discussed system. It should be noted that calculation of the number of particles based on DLS is approximate and underestimates the real number of particles due to bias toward large particles.