Modeling of Traction Drive in Elastohydrodynamic Lubrication Contact and Traction Coefficient Enhancement Using Surface Texture Under High-Speed Conditions of up to 50,000 rpm

Due to increasing demand for reductions in carbon dioxide emissions, many conventional vehicles are expected to be replaced by hybrid, electric, or fuel-cell vehicles. Motor miniaturization is important for reducing vehicle weight and maximizing occupant space. Motor output is the product of torque, which depends on the motor diameter and rotational speed, and thus motors can be made smaller while maintaining output by increasing the rotational speed. For example, the Toyota Prius increased the motor rotational speed from 5000 rpm to 17,000 rpm when it was first released, achieving a significant reduction in size while increasing output [1]. The European consortium DRIVEMODE has developed a unit for electric vehicles with a maximum rotational speed of 20,000 rpm [2]. Volk and Leighton have proposed a 30,000 rpm unit [3] and the Technical University of Munich studied the design of a 50,000 rpm unit [4]. Although increased losses, centrifugal force, vibration, cooling, and other factors must be taken into account, higher rotational speeds have become a major trend.

To provide sufficient driving force to a vehicle with a high-speed motor, a combination of reduction gears is required. Traction drives are an alternative to gears for high-speed rotation. They transmit power by shearing an oil film that forms between two rolling elements. Traction drives are suitable for high-speed rotation because they have no meshing vibration and low oil agitation resistance. However, the traction coefficient (ratio of transmission force to contact force on the transmission portion) decreases with increasing rolling speed of a traction drive because an increase in velocity causes the oil film to thicken in accordance with fluid lubrication theory [5], resulting in a decrease in shear strain rate and an increase in oil film temperature. To maintain the required transmission force, the contact force must be increased to compensate for a decreased traction coefficient, resulting in lower transmission efficiency and the requirement of a larger transmission. The author previously designed and built a high-speed traction drive testing machine (Fig. 1(a)) to investigate traction characteristics at high rotational speed, and experimentally confirmed that the traction coefficient drops by up to 20% at 75 m/s compared to that at several meters per second (Figs. 1(b) and 1(c)) [6]. Such a large drop in the traction coefficient has a significant impact on vehicle performance.

Various methods have been developed to increase the traction coefficient, including those that modify the traction oil and those that apply a surface texture. Tsubouchi and Hata derived an equation that expresses the correlation between the molecular structure (e.g., stiffness and shape) and the traction coefficient [7]. One study attempted to calculate traction coefficients from molecular structures using molecular dynamics simulation [8]. However, there is a trade-off between the traction coefficient of oil and low-temperature flowability. It is thus difficult to increase the traction coefficient without reducing the low-temperature flowability of automotive oils, which are operated at extremely low temperatures. Nambu et al. applied microgrooves, a type of surface texture, to the transmission surface and thinned the oil film to increase the strain rate and improve the traction coefficient (an improvement of approximately 5% was observed) [9]. However, direct contact tends to occur at high loads, resulting in peeling, so the use of microgrooves is limited to low loads. The traction coefficient was measured at 15 m/s, so the effect at high speeds is unclear. There is thus a need for a texture that effectively improves the traction coefficient at high speeds and that can withstand high loads.

To design a surface texture that has both a high traction coefficient and high strength under high-speed operation, a traction model that agrees with experiments at high speeds is required. Itagaki et al. built a high-power traction drive testing machine and fit a nonlinear Maxwell model [10] to the obtained traction coefficients [11]. However, the model was fitted at peripheral velocities of 40 m/s or lower. The accuracy of the model at higher velocities has not been confirmed.

This study constructs a model for calculating the traction coefficient at high rotational speeds and uses it to create a texture that suppresses the decrease in the traction coefficient with increasing rotational speed. The model considers the non-Newtonian viscosity of the oil, changes in oil properties due to an increase in temperature, and the oil film shape and pressure distribution due to the surface microstructure. A test roller with a texture designed using the model is fabricated and its effectiveness is experimentally confirmed using a high-speed traction drive testing machine.

Traction drives generate heat due to micro slip, which is referred to as creep. Changes in rheological properties due to an increase in temperature affect the oil film thickness and the traction force. In order to account for the increase in temperature, traction calculations can be substituted into thermal elastohydrodynamic lubrication (EHL) equations to simultaneously calculate the oil film shape, pressure distribution, and traction force [12]. However, this method is highly nonlinear and requires a long calculation time. The objective of this research is to optimize texture design. Because many calculations are performed, a method with a small computational load is necessary. Therefore, the proposed method is based on an iterative calculation of the oil film thickness and pressure distribution, a traction calculation, and a temperature calculation. These calculations are repeated until the results become converge. For application to texture design, the oil film thickness and pressure are distributed based on EHL numerical calculations instead of using average values, and the traction force and increase in temperature at each location on the contact surface are calculated.

Under the operating conditions used in this study, the calculated minimum oil film thickness is about 0.2–1 µm and the surface roughness of the roller is less than Ra = 0.04, so the film thickness ratio is more than five. An analysis by Johnson et al. indicated that two rollers are sufficiently separated by an oil film if the film thickness ratio is greater than three [13]. Therefore, the friction due to direct contact is not considered here and the oil film is assumed to transmit all power.

It is difficult to theoretically determine the high-pressure properties of oil given in Eqs. (1) and (2). Therefore, experimental equations were derived based on the experimental values for the operating conditions. The oil used in this study was Idemitsu Kosan’s traction fluid KTF-1, which was developed for automotive traction drive transmissions [14].

The temperatures were calculated as described below.

The roller bulk temperature is difficult to calculate because it depends on heat generation and cooling, heat capacity of the testing machine, heat conduction, and other factors. Therefore, the roller surface temperature at 90 deg before the contact point was measured using a thermocouple for moving surfaces, as shown in Fig. 3. This temperature was taken as the bulk temperature.

Therefore, the results of steady-state calculations were compared with those of transient calculations. The upper limit of time integration was defined as the time required to pass through one mesh, i.e., the peripheral velocity divided by the length of the mesh. Figure 4 shows the temperature increase rate distribution in the rolling direction for a heat source at a single point on the surface at a peripheral velocity of 75 m/s and a maximum Hertzian pressure of 3 GPa. In the unsteady-state calculations, the range over which the temperature increases is significantly narrower. When the temperature rise is integrated over the entire contact surface using Eq. (10) under the condition, the unsteady-state temperature rise is 5 °C compared with a steady-state rise of 26 °C, which is a difference of approximately five-fold, indicating that unsteady-state calculations are required at high speeds.

Tomita et al. considered the shear position of the oil film in the thickness direction to be uniform in the viscous region and near the center in the plastic region, which is consistent with the temperature measurement results [24]. This consideration is used in the present study.

For simplicity, let the temperatures of both rollers be equal and ΔT_f = 0 at z = 0 and z = h. Then

The thermal conductivity of oil under high pressure is calculated using an experimental equation [25] derived from Larsson’s equation [26].

The average oil film temperature is used in the EHL calculation and the maximum oil film temperature is used in the traction calculation.

The overall flow is shown in Fig. 5, where the oil film shape and pressure distribution were obtained from the EHL calculation, the traction force was calculated by substituting the obtained oil film shape and pressure distribution into the Maxwell model, the oil film temperature was calculated from the obtained traction force, and the EHL calculation was performed by modifying the temperature. The calculations were repeated until the traction force and temperature converged. The calculation methods and modifications made to stabilize and speed up the calculations are described below.

Ichimaru’s method [25] was adopted in the present study. In the region where an oil film forms, the elastic deformation equation, high-pressure viscosity equation, and high-pressure density equation are substituted into the difference Reynolds equation to obtain the pressure distribution by the relaxation method. Because the Reynolds equation is not valid in a region where the oil film thickness is zero or negative, this is the direct contact region, and the pressure is obtained by setting the left-hand side of the elastic deformation equation to zero. In cavitation regions, where the pressure is negative, the pressure is set to zero. Methods for speeding up the calculations are described in the  Appendix.

This equation is integrated using the fourth-order Runge–Kutta method to obtain the shear stress. However, if τ > τ_c, Eq. (2) is used. The shear modulus G, Eyring stress τ₀, and critical shear stress τ_c are calculated by considering the pressure and temperature at each node (Eqs. (6)–(8)).

The heat generation rate is determined from the obtained shear stress and slip rate, and the temperature of the oil film is calculated. This temperature is substituted into the EHL calculation to obtain the pressure distribution and oil film shape, which are then substituted into the traction calculation. This process is repeated until convergence is reached. However, the Maxwell model contains a hyperbolic function in the viscosity term, which is highly nonlinear, and the temperature increase equation switches among elasticity, viscosity, and plasticity, causing the calculation to easily diverge, especially under high-speed conditions, which is the case considered in this study. Therefore, a Gaussian filter is applied to the calculated increase in temperature to smooth it out and improve convergence. The standard deviation is set at 0.5.

The experimental results measured using the high-speed traction testing machine shown in Fig. 1(a) are compared with the calculation results obtained using the model developed in this study. The experimental and calculated values under typical conditions are shown in Figs. 6(a)–6(c) and the calculated coefficients of the experimental equations for the Eyring stress and limiting shear stress are shown in Table 1. The calculations generally agree with the experiments regardless of the rotational speed and pressure. Figure 6(d) plots the maximum traction coefficient versus the rotational speed. The difference between the calculations and experiments is about 6% at maximum. The results indicate that the calculation model developed in this study can predict the traction coefficient with high accuracy.

To reduce the decrease in the traction coefficient with increasing speed, a texture was designed using the developed computational model. In the design of the texture, the traction model and the high-pressure rheological properties of the oil were taken into consideration to determine a design policy to increase the traction coefficient.

Ohno et al. showed that the state transition and maximum traction coefficient for oil are determined by the temperature and pressure of the oil based on measurements of oil properties under high pressure and the results of a two-cylinder traction test [19]. A higher pressure and a lower temperature lead to a higher traction coefficient. In particular, when the oil transitions from viscoelastic to elastoplastic, it becomes solid and the traction coefficient significantly increases. Therefore, from Eq. (25), the temperature decreases as the oil film thickness h decreases. In addition, from Eq. (24), the shear force increases as the oil film thickness h decreases, resulting in a higher traction coefficient.

To increase the pressure, the radius of curvature of the roller can be reduced and the Hertzian contact pressure can be increased; however, this reduces the strength and life of the roller. Therefore, the aim is to equalize the pressure and increase the pressure in the low-pressure region within the contact surface without increasing the maximum pressure (Fig. 7(a)), and to increase the pressure at low loads without increasing the pressure at high loads (Fig. 7(b)). Table 2 summarizes the aforementioned studies. It shows that thinning the oil film and increasing the pressure in the low-pressure area effectively increase the traction coefficient.

To make the oil film thinner, fine grooves parallel to the rolling direction are formed on the roller surface to increase the flow coefficient [27]. However, in calculations performed at 43,750 rpm with various groove depths, widths, pitches, and angles, no combination could increase the traction coefficient. This is likely due to the oil film becoming thinner in flat areas but thicker in grooves, which reduces the traction coefficient by decreasing the pressure.

Next, consider the use of high-pressure in low-pressure regions. By changing the point contact to line contact, the pressure can be averaged in the direction perpendicular to the axis. The logarithmic function geometry used for cylindrical roller bearings can make the pressure distribution nearly constant in the perpendicular direction without generating edge loading [28].

In Hertzian point contact, when the radius of curvature of the roller is constant, the pressure is proportional to the cube root of the load. In contrast, as shown in Fig. 8(a), if the radius of curvature increases moving away from the center of contact, the degree of increase in the contact width increases as the load increases, and the slope of pressure versus load can be smaller than the 1/3 power. Therefore, if the radius of curvature at the center of contact is kept small, the characteristics are expected to be those shown by the orange line in Fig. 7(b). A shape with multiple curved surfaces in multiple steps in the direction of load is also possible. At low loads, a tip with a small radius of curvature makes contact to increase the pressure. As the load increases, the tip elastically deforms to contact surfaces with a large radius of curvature to suppress the pressure (Fig. 8(b)). Combined with the logarithmic shape described above, an increase in the traction coefficient in both the low- and high-load ranges is expected.

The proposed shape of the texture for increasing pressure in the low-pressure area, as described in the previous section, was examined in terms of the test roller specifications for evaluation in a high-speed traction testing machine.

The pressure at high and low loads is used as the evaluation index. All combinations are calculated for the parameters shown in Fig. 9. A smooth roller with a constant crowning radius of 23.5 mm is used for comparison. The Hertzian maximum pressure is about 1.5–4.0 GPa at pressing forces of 291–5510 N. With the texture, the pressure at a maximum load of 5510 N should be less than 4 GPa and the pressure at a minimum load of 291 N should be as high as possible.

Figure 10(a) shows the results of all calculations, where the horizontal axis represents the pressure at high load and the vertical axis represents the pressure at low load. The data points for the textured roller are located to the upper left of those for the smooth roller, which is the desired effect. Table 3 shows the results of the pressure and traction coefficient calculations for various parameters, where the pressure at high loads is near 4 GPa. The highest traction coefficient (an increase of 19%) is obtained for A = 600 and K = 16. A = 500 and K = 10 also increase the traction coefficient by 19%, with the advantage that the small K (i.e., low peak) makes processing easier. The latter parameters were thus selected.

Figure 10(b) shows the calculated pressure versus load and traction coefficient. There was concern that the pressure would drop when the projections on the two sides begin to make contact, but the pressure increases over the entire load range, and the traction coefficient also increases. At high loads, the traction coefficient increases even though the pressure remains the same. This is due to the logarithmic shape averaging the pressures and the increase in the pressure in the low-pressure region on the contact surface.

The specifications for the texture and the calculated pressure and traction coefficient are shown in Table 4. For a roller with a constant crowning radius (23.5 mm), the pressure at low loads is increased and the pressure at high loads is maintained, resulting in a 5–8% increase in the traction coefficient.

The effect of a multi-stage logarithmic function shape, which is effective at increasing the traction coefficient, was experimentally confirmed using a high-speed traction drive testing machine. The texture was machined onto the low-speed rollers. The high-speed rollers were smooth. A photograph of the prototype roller and the designed and measured geometry of the texture are shown in Fig. 11. The measured convexities are somewhat flat, but the peak widths and height offsets are approximately equal to the designed values.

The traction curves obtained at high rotational speeds and low load (maximum pressure of 1.5 GPa without texture) are shown in Fig. 12(a). The traction coefficient for the textured rollers is greatly increased compared with that for the smooth rollers. Figure 12(b) plots the maximum traction coefficient versus rotational speed. At a low load (291 N), the texture increases the traction coefficient over the entire speed range, with an increase of up to 19%. At a load of 689 N (2.0 GPa without texture), there is no difference between the cases with and without texture (i.e., there is no improvement effect). This may be because almost the entire area of the oil film had transitioned from viscoelastic to elastoplastic, as predicted by Fig. 10(b).

Texture shape measurements made before and after the experiment are shown in Fig. 13. There is almost no change in shape. The amounts of plastic deformation and wear are considered to be extremely small because the experiments were conducted for a short period of time (approximately 2–3 min per condition, or several tens of minutes in total). Longer operation time would change the shape of the wheel, resulting in a decrease in the traction coefficient and a decrease in the life of the wheel. Confirmation of the change in the traction coefficient over time, for example during 100,000–200,000 km of driving (the life of an automobile), is a major issue for future research.

To suppress the reduction of the traction coefficient at high rotational speeds, a model for predicting the traction coefficient was developed and used to design a texture, which was then evaluated experimentally. In summary:

1. A traction calculation model was created by coupling the nonlinear Maxwell model, EHL calculation, and oil film temperature calculation.

2. The model takes into account the pressure distribution, oil film shape, and temperature distribution, and can be used for texture design.

3. The calculated traction coefficients were in good agreement with the experimental results.

4. The nonlinear Maxwell model and the high-pressure rheological properties of oil were used to formulate design guidelines for textures.

5. The designed prototype texture greatly increased the traction coefficient.

This research was supported by the “Project for Building Simulation Platforms to Accelerate Development of Next-Generation Vehicles” (TRAMI-Transmission Research Association for Mobility Innovation) grant of 2020.

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.

a =

Hertzian contact radius, m

h =

thickness of oil film, m

p =

pressure of oil film, m

q =

heat per unit area, W/mm²

t =

time, s

u =

rolling speed in x-direction, m/s

v =

rolling speed in y-direction, m/s

G =

elastic shear modulus, Pa

T =

temperature of oil, °C

c_s =

specific heat capacity of roller, J/kg · K

q_p =

heat per unit area in plastic region, W/mm²

q_v =

heat per unit area in viscous region, W/mm²

G_f =

elastic shear modulus of oil film, Pa

K_f =

thermal conductivity of oil, W/mK

K_s =

thermal conductivity of roller, W/mK

T_b =

bulk temperature of roller, °C

T_f =

temperature of oil film, °C

E′ =

elastic modulus, Pa

x, y, z =

coordinates, m

α_p =

pressure viscosity index under high pressure, Pa⁻¹

α₀ =

pressure viscosity index under atmospheric pressure, Pa⁻¹

$γ˙$ =

shear rate of oil film, m

Δu =

slip speed in x-direction, m/s

Δv =

slip speed in y-direction, m/s

ΔT_f =

temperature rise of oil film, °C

ΔT_s =

temperature rise of roller surface, °C

η_p =

viscosity under high pressure, Pa · s

η₀ =

viscosity under atmospheric pressure, Pa · s

ν₀ =

kinematic viscosity under atmospheric pressure, m²/s

ρ_s =

density of roller, kg/m³

ρ₀ =

density of oil, kg/m³

τ =

shear stress, Pa

τ_c =

limiting shear stress, Pa

τ₀ =

Eyring stress, Pa

The Reynolds equation is computed using the relaxation method, in which the pressure at a node is calculated from the values of the four surrounding nodes in the previous calculation cycle. Therefore, propagation of the effect of pressure modification is only transmitted to one neighboring node in one calculation, which requires an extremely large number of calculations when the mesh is fine. The multigrid method [29], which alternates between coarse and fine mesh calculations, is considered effective for this problem. However, if the texture cannot be represented by a coarse mesh, errors may occur when switching mesh size, increasing the number of calculations, and diverging the viscosity term in the highly nonlinear Maxwell model. Therefore, for a smooth shape with no texture, a coarse mesh is used to obtain a rough pressure distribution and oil film shape, and then a fine mesh is used to obtain the pressure distribution and oil film shape, taking texture into account. The traction, temperature, and EHL calculations are repeated using only the fine mesh.

The center of the calculated result is the original convolution sum.