Lubrication simulation in

automotive gearbox with SPH-flow

A. Bannier*, F. Sarret & M. Farih, P. Chessé, N. Perrot, A. Mucherie & A. Frecinaux,
Nextflow Software, LHEEA UMR 6598,
Nantes, France, École Centrale de Nantes,
[email protected] Nantes, France,

F. Ravet, B. Pelourdeau & L. Dutfoy, N. Dabert,

Renault SAS, Digital Product Simulation,
France France

Abstract— Transmission efficiency is one of the major focus for

car consumption reduction. Automotive manufacturers look The latter aspect is considered in §III, where the previously
deeper into gearbox lubricant management to reduce oil-induced calibrated and validated numerical methodology is applied on a
losses while maintaining a reliable lubrication. Facing the complete automotive gearbox. Comparison with experimental
complexity of the involved physical phenomena, advanced results assesses the ability to predict the resistive torque for
numerical tools are required to supplement the understanding and several engine speeds and thermal conditions.
the design assistance brought by experimental tests. This paper
aims at assessing the capability of the SPH method in simulating Finally, the success story that led Renault Sport to improve the
the lubricant flow involved in splash-lubricated gearboxes.
gearbox of its Formula-E thanks to the SPH approach is
Experimental-versus-numerical comparisons have been
conducted on two validating studies, respectively designed presented in §IV.
towards oil flow topology in a simplified gearbox and oil-induced
resistive torque in an automotive gearbox. Finally, the numerical Simulations in this article have been performed with the SPH-
methodology led Renault to successfully solve a lubrication flow solver, co-developed by Nextflow Software, Ecole
problem on its Formula-E racing car gearbox. Centrale Nantes and CNR-INM. High fidelity results are made
possible thanks to its Arbitrary Lagrangian Eulerian (ALE)
method with Riemann solver flux computation [3]. Boundaries
I. INTRODUCTION
are accurately considered by its fast and consistent surface
Gearbox lubrication is a complex mechanism involving many integration process, even with complex tridimensional
physical phenomena (viscosity, multifluid, surface tension, geometries [4]. Among the physical contributions that can be
thermal effects, etc.). Oil lubrication has two major roles: simulated, viscosity is based on the Morris et al. model [5]
limiting friction and ensuring a sufficient cooling of mechanical while surface tension is inspired from Akinci et al. work [6].
parts. An effective lubrication can significantly increase the life Using the MPI protocol, SPH-flow is efficiently parallelized
of gearboxes while limiting energy loss [1] and therefore CO2 and demonstrates scalability performances from 85% to
emission. Numerical simulations, based on Smoothed Particles 100% [7].
Hydrodynamics (SPH) in particular, can help designing
appropriate gearbox lubrication system and assessing the II. STUDY OF GEARBOX FLOW TOPOLOGY
energy losses due to oil churning phenomenon [2].
This paragraph aims at assessing the capability of the SPH-flow
The present article aims at assessing the capability of the SPH solver in accurately simulating the flow topology within a
method in simulating the lubricant flow involved in splash- splash lubricated gearbox. An experimental-against-numerical
lubricated gearboxes. The focus is especially made on two comparison approach has been adopted.
central aspects: the oil flow topology and the oil-induced
resistive torque. Among the key phenomena to capture is the adhesion of the oil
on the dipping gear. The amount of oil that clings on the
The former aspect is addressed in the following paragraph, §II, emerging part of the gear depends on a balance between
where qualitative and quantitative comparison is made with the viscosity, surface tension, gravity and centrifugation. The
experimental test rig of a simplified gearbox that has been former two tend to favor adhesion, while the latter two cause oil
especially developed. separation.
 2019 international SPHERIC workshop Exeter, United Kingdom, June 25-27, 2019

Another phenomenon to appropriately capture is the jet and C. Numerical setup

droplets creation. The detachment of the oil from the rotating
gear, along with the jet atomization and the droplets Besides physical parameters, whose values are explicitly
coalescence, may directly affect the amount and the distribution imposed by the test condition, some input numerical parameters
of lubricant within the gearbox. may have an influence on the accuracy and the speed of the
computation. To ascertain appropriate setup and
A. Study plan design simplifications, various numerical parametric studies have been
conducted.
Conducting the present study on a real industrial gearbox would
have brought unnecessary complexity (complex 3D design, As a result, tests reveal that the air phase does not play a
high number of rotating parts, etc.). Instead, a simplify design significant role in the studied phenomena. Thus, only the oil
has been preferred. Its picture is represented below Figure 1. A phase will be considered.
single dipper gear is considered. No gears meshing is required
to examine the desired phenomena. On the gear emerging side, For such free-surface flow simulation, the wall boundary
the carter is circular and conforms the gear. This allows the condition may be adapted to let the flow eventually detach from
channeling of the oil conveyed by the gear, with a teeth-to- walls (that is, enforcing the wall boundary normal velocity to
carter gap similar to industrial gearboxes. On the other side, the be positive or zero, rather than necessary null). The application
carter is made vaster, such that oil splashing is not obstructed. of such process did not evidence any notable improvement in
the present gearbox simulation.
Even though parametric studies would have been possible with
numerous design parameters (carter dimensions, gear thickness, Within the weakly-compressible framework, the numerical
diameter or modulus, oil properties or amount, etc.), the present sound speed may be chosen smaller than the actual physical
work focuses on the influence of the gear rotational speed. All sound speed to reduce the computational cost. However, its
other parameters are kept unchanged. The gear is 10 mm wide value must remain sufficient to satisfy the almost
and has a 74 mm reference diameter (37 teeth, module 2). A incompressibility of the flow (in other terms, the flow and
10W40 oil has been retained, whose density is 845 kg/m3 and acoustic energy spectra must remain disjointed from each
dynamic viscosity 74 mPa·s. other). For numerical sound speed set to 50 m/s, a posteriori
check reveals that the Mach number remains below 0.1 and
B. Experimental setup density variation below 3%.

The experimental test rig is pictured in Figure 1. It consists in a The last but not least numerical parameter is the fluid spatial
wide cuboid box, whose planar front and back sides are made discretization. A convergence study has been led with
in transparent polycarbonate to ease visualization. 3D-printed discretization values 0.5, 0.75 and 1 mm for a gear rotational
parts are used to obtain on-demand gears and carter lateral speed set to 600 rpm. Whatever the discretization, the simulated
shapes. The gear is set in motion by a three-phase motor of time evolution of the jet flow rate fluctuates around the same
nominal speed 1500 rpm at 50 Hz and controlled by a constant value of around 200 L/h. Thus, such discretization is
Motrivac LTP-B frequency converter. considered enough for reaching flow rate convergence. A
0.5 mm discretization is chosen for conducting the study.
The gear rotational speed is measured by Hall effect with a
precision of 1 rpm. Additionally, high-speed cameras are used D. Experimental versus numerical comparison
for visual analysis of the flow topology.
Experimental and numerical results are compared for gear
rotational velocity ranging from 100 to 900 rpm.

A qualitative analysis of the flow can first be performed by

visual observation of the oil flow topology. Such observations,
Figure 2, reveals the presence of different lubrication flow
regimes:

• At low speed (100 rpm), Figure 2 (a, b), inertial effects

are weak. The oil flow is mainly driven by viscosity
and remains attached to the dipper gear.

• For higher speeds (200 rpm and 500 rpm), Figure 2 (c-
f), inertia comes into play. Under the centrifugal force,
the oil film that coats the gear teeth detaches from
Figure 1: Experimental test rig of the simplified gearbox them. A lubricant film is first formed and, when
 2019 international SPHERIC workshop Exeter, United Kingdom, June 25-27, 2019

rotational speed increases, it gradually turns into a

main jet along with filaments.

• Finally, at high rotational speed regime (900 rpm),

Figure 2 (g, h), the oil jet becomes chaotic. This last
regime is caused by a blockage effect: the faster the
rotation, the greater the amount of oil carried away by
the gear, the longer the gear perimeter section along
which the carter is totally filled by oil and in which the
flow rate saturates. The bound of this filled carter (a) (b)
section is marked by the position of a free-surface
meniscus on the left side of the gear. The chaotic jet
regime appears when this meniscus reaches the
uppermost part of the carter. With such blockage
effect, part of the oil cannot be ejected forwardly by
the gear and leaks along the carter lateral sides.

A clear agreement in the experimental-versus-numerical

comparison of the flow topologies is evidenced for all rotational
speeds. The three different regimes – viscous, inertial and
chaotic – are clearly observed at same speeds experimentally (c) (d)
and numerically. The free-surface meniscus location agrees.
Only notable discrepancies are the absence of wetting on the
gear at 100 rpm, due to the limitation of wettability
consideration within our surface tension numerical model, and
the under-refinement of the oil film and filaments at 200 and
500 rpm respectively.

In light of the above, the numerical simulation has fulfilled the

objectives of simulating the flow topology within a splash
lubricated gearbox. Further quantitative comparisons have been
planned but required additional arrangement on the (e) (f)
experimental side: oil jet flowrate measurement must not be
intrusive, and torque measurement turns out to be extremely
sensitive on this reduced scaled gearbox. The torque
measurement comparison is addressed on realistic full-scaled
gearbox within following paragraph.

III. STUDY OF GEARBOX RESISTIVE TORQUE

Hereafter, a comparison between experimental and numerical

oil-induced resistive torque measured on a complete automotive (g) (h)
gearbox is presented.
Figure 2: Experimental and numerical results: 100 rpm (a, b), 200 rpm (c, d),
A. Study plan design 500 rpm (e, f) and 900 rpm (g, h).

The oil flow in a gearbox depends, among others, on the

rotational regime and the oil properties. In the present study, Temperature Density Dynamic viscosity
three rotating speeds are chosen to cover the application range: (°C) (kg/m3) (mPa·s)
idle engine (1 000 rpm), high speed (6 000 rpm) and limit speed 20 863 56.3
(10 000 rpm). Moreover, since the oil properties is directly 50 847 21.1
influenced by the temperature, three thermal conditions have 80 829 9.07
been considered: cold engine (20°C), warm engine (80°C) and
a transient state (50°C). Density and viscosity of oil for these Table 1: Oil density and dynamic viscosity
temperatures are given in Table 1.
 2019 international SPHERIC workshop Exeter, United Kingdom, June 25-27, 2019

C. Numerical setup

Numerical simulations have been conducted with the SPH-flow

solver to numerically estimate this resistive torque. The full
gearbox carter and rotating parts are considered. The setup for
the numerical parameters is similar than the one used for the
study described in the previous paragraph §II. The spatial
discretization, that is the size of the SPH particles, is set to
1 mm. Although this discretization is too rough to model the
thin oil films that can be experimentally observed (along the
carter on the right part of Figure 4 (a) for instance), it proves to
be fine enough to model the bulk of the lubricant flow and thus,
a good estimation of the resistive torque it induced.
Figure 3: Set-up of the automotive gearbox test bench
D. Experimental versus numerical comparison

B. Experimental setup Experimental and numerical torque measurements are plotted

in Figure 5. While there are some slight discrepancies for very
The test rig set-up is presented in the Figure 3. It is composed high rotating speeds (10 000 rpm), these results show that the
of a gearbox, driven by an electrical engine (rotating machine), SPH approach matches experimental measurements for
a clutch in-between, a robot to change gear and calibrating classical usage conditions (1 000 rpm and 6 000 rpm at 50°C
tools. The torque is measured on the drive shaft. This test rig and 80°C). Numerical results are within the experimental
can measure the torque of whole system up to 50 N·m when measurement tolerance of 0.1-0.2 N·m. At cold temperature,
engine speed ranges from -6 000 rpm to +13 000 rpm. The the simulation under-estimates the torque by an absolute
precision of the torque measurement is estimated to about difference of about 0.25 N·m which remains constant from idle
0.1 N·m. engine to very high rotating speed. The cause for this disparity
has not been proved yet.
Preliminary results show that the highest amount of oil is
observed around the differential gear (the left most gear on
Figure 4). Thus, the main part of the resistive torque induced by
the lubricant is expected to be imposed through this gear.

(a)

Figure 5: Experimental (dashed line) and numerical (solid line) torque

measurements

(b) Among prospects of improvements are, on the experimental

side, the mastery of the external environment effects and, on the
numerical side, the consideration of the thin lubricant film that
spreads on the carter surface.

Figure 4: Experimental (a) and numerical (b) oil lubrication in an automotive

gearbox
 2019 international SPHERIC workshop Exeter, United Kingdom, June 25-27, 2019

IV. APPLICATION: RENAULT FORMULA-E SUCCESS STORY would not be appropriately lubricated after strong breakings
that would last more than 1 second (as shown on Figure 7) or
The volume of lubricant in a gearbox is optimized as a balance after right turns that would last even less (cf. Figure 8). Thus,
between the reliability of the system and the onboard mass of simulation evidences that lengthy breakings and right-turns are
oil. This is especially the case for Formula-E cars. A lack of oil the critical events that may cause lubricant heating.
may induce a poor lubrication, leading to dramatic damages of
the gearbox and may eventually put a stop to the race. On the
contrary, an excess of oil may reduce the gearbox efficiency,
which limits battery life and consequently the car speed and
endurance.

However, during races, Formula-E cars experience strong (a)

accelerations, whose intensities and orientations depend on
track and race configurations: high speedups, short breakings
and sharp right or left turns. Under such accelerations, the fluid
lubricant within the gearbox of the electric-powered cars
undergoes substantial inertial forces. Thus, some parts of the
gearbox system may temporarily run out of lubricant. A too
long non-lube delay may cause a significant warm up of the
mechanical parts within the friction area. When lubrication is
finally restored, after the speedup, breaking or turn event, the
oil cools down the heated parts, but substantially increases its
(b)
own temperature.

In this way, the temperature of the lubricant may increase lap

after lap. Thus, its viscosity coefficient is affected and both the
efficiency and the reliability of the gearbox may be significantly
disturbed.
Figure 7: Lubrication before (a) and after (b) 1 second-long braking

(a)

Figure 6: Monza race circuit [8]

On the Monza race circuit (cf. Figure 6), a temperature rises of

about 2°C per lap of the gearbox lubricant has been measured
by the Renault Sport team. To avoid possible failure or
damages, the racing driver had to stop his car before the end of
the race. (b)
Engineers at Renault performed simulations using the SPH-
flow solver to investigate the cause of the temperature rising.
Based on the G-forces recorded along the circuit during the
race, they identify the strongest events and applied realistic
external forcing on the fluid lubricant during the SPH
computations. Figure 8: Lubrication before (a) and after (b) 1 second
of “Curva Parabolica”-like right turn
Simulation reveals that gears remain well lubricated during
speedups and left turns. However, it demonstrates that gears
 2019 international SPHERIC workshop Exeter, United Kingdom, June 25-27, 2019

On the Monza circuit, all braking events last less than one REFERENCES
second. However, the time spent by the racing car in the “Curva [1] Changenet, C., Oviedo-Marlot, X., & Velex, P. (2006). Power loss
Parabolica” corner (cf. Figure 6), which is the strongest right predictions in geared transmissions using thermal networks-applications
turn event of the race, turns out to last approximately 3 seconds. to a six-speed manual gearbox. Journal of Mechanical Design, 128(3),
618-625.
This is the only critical event for the lubrication of the system.
[2] Mettichi, M., Gargouri, Y., Groenenboom, P., & el Khaldi, F. (2017).
Simulating Oil Flow for Gearbox Lubrication using SPH. Proc. 10th
Thanks to these simulation results, modifications in the SPHERIC International workshop, Parma.
lubrication system have been made possible to solve the issue. [3] Oger, G., Marrone, S., Le Touzé, D., & De Leffe, M. (2016). SPH
Since the re-design phase, the lubrication of the differential gear accuracy improvement through the combination of a quasi-Lagrangian
shifting transport velocity and consistent ALE formalisms. Journal of
is not interrupted any more. In particular, the Renault Formula- Computational Physics, 313, 76-98.
E can undergo never-ending right turns without jeopardizing its [4] Chiron, L., de Leffe, M., Oger, G., & Le Touzé, D. (2019). Fast and
transmission system. The following year, the racing car accurate SPH modelling of 3D complex wall boundaries in viscous and
succeeds to end its race. non viscous flows. Computer Physics Communications, 234, 93-111.
[5] Morris, J. P., Fox, P. J., & Zhu, Y. (1997). Modeling low Reynolds
number incompressible flows using SPH. Journal of computational
CONCLUSIONS physics, 136(1), 214-226.
[6] Akinci, N., Akinci, G., & Teschner, M. (2013). Versatile surface tension
Good lubrication and low energy loss are the two main and adhesion for SPH fluids. ACM Transactions on Graphics (TOG),
challenges driving an automotive gearbox design. This study 32(6), 182.
demonstrated that the SPH approach can enhance the design of [7] Oger, G., Le Touzé, D., Guibert, D., De Leffe, M., Biddiscombe, J.,
gearbox regarding this these two purposes. First, an Soumagne, J., & Piccinali, J. G. (2016). On distributed memory MPI-
based parallelization of SPH codes in massive HPC context. Computer
experimental study on a simplified gearbox has been carried Physics Communications, 200, 1-14.
out. It has resulted in a reliable SPH-flow numerical setup that [8] "Autodromo Nazionale Monza – Company profile". Autodromo
has been applied to a complete automotive gearbox. The oil- Nazionale Monza. MonzaNet.it. 2007. Archived from the original on 25
induced resistive torque has been numerically estimated on the July 2009. Retrieved 17 September 2009.
whole system. Finally, when applied to a Formula-E racing
incident, this numerical tool has proved to be a first-class tool
in the understanding and the solving of gearbox splash
lubrication, even under high G-force constraints.