Accepted Manuscript

Numerical simulations of oil flow inside a gearbox by Smoothed Particle

Hydrodynamics (SPH) method

Zhe Ji, Milos Stanic, Erwin Adi Hartono, Valery Chernoray

PII: S0301-679X(18)30268-8
DOI: 10.1016/j.triboint.2018.05.034
Reference: JTRI 5248

To appear in: Tribology International

Received Date: 6 February 2018

Revised Date: 30 April 2018
Accepted Date: 24 May 2018

Please cite this article as: Ji Z, Stanic M, Hartono EA, Chernoray V, Numerical simulations of oil flow
inside a gearbox by Smoothed Particle Hydrodynamics (SPH) method, Tribology International (2018),
doi: 10.1016/j.triboint.2018.05.034.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Numerical simulations of oil flow inside a gearbox by Smoothed

Particle Hydrodynamics (SPH) method
Zhe Jia , Milos Stanica , Erwin Adi Hartonob , Valery Chernorayb
a
FluiDyna GmbH, Edisonstrasse 3, D-85716 Unterschleissheim, Germany

PT
b
Department of Applied Mechanics, Chalmers University of Technology, Gothenburg, Sweden

RI
Abstract

SC
In this paper, numerical simulations of oil flow inside a gearbox are presented and com-
pared to experimental particle image velocimetry (PIV) results. Instead of a traditional
grid-based finite volume method, a mesh-free Smoothed Particle Hydrodynamics (SPH)
method is employed. A multi-phase SPH formulation is utilized to resolve the complex mul-

U
tiphase fluid flow. A total of nine simulations are carried out for three oil levels and three
Reynolds numbers, to investigate the flow field behavior and to compare with the experi-
AN
mental results. The aeration effect is first considered to qualitatively analyze the quantity
and size of bubbles generated due to the rotation of the gears. Furthermore, the velocity
field and velocity profile beneath the oil surface are comprehensively analyzed. Our simu-
M

lation results exhibit physically consistent behavior of the oil flow and good agreement is
achieved compared to the experimental results. Flow structures, splashing and recirculation
areas are very well captured by the simulation. However, some discrepancies of the velocity
D

field between the numerical and experimental results are also observed, and discussed.
TE

Keywords:
Smoothed Particle Hydrodynamics, Gearbox, Oil Distribution, CFD
EP

1. Introduction
In transmission systems, oil churning losses can represent a significant portion of the
C

overall energy losses. Excessive oil churning loss increases the energy dissipation and conse-
quently, hinders the improvement of the efficiency in transmitting power [1].
AC

As one of the most common lubricating solutions in automotive applications [2], oil-
bath lubrication systems suffer, in particular, from over-dissipation due to the churning
losses. In an oil-bath lubrication system, gears are partially immersed in the lubricant,
in such a way that the rotation of the gears ensures the formation of a lubrication film on
critical gear surfaces in the contact area of gear teeth. The adherent lubrication film on

Email addresses: [email protected] (Zhe Ji), [email protected] (Milos Stanic),

[email protected] (Erwin Adi Hartono), [email protected] (Valery
Chernoray)
Preprint submitted to XXX May 26, 2018
 ACCEPTED MANUSCRIPT

the teeth of the gears prevents direct metal-to-metal contact during the meshing. Moreover,
the oil sump serves as a heat sink and fulfils the purpose of the gear cooling. To design an
oil-bath lubrication system, a precise fill level of the lubricant is vital in order to achieve
optimal performance. Sufficient oil level is needed to achieve good lubrication and
cooling of the assembly, yet excessive oil levels lead to increased churning losses
[3]. Thus, study of the oil distribution and the flow field inside a gearbox is essential for

PT
predicting the oil churning losses, lubrication and cooling effects [4]. The topic of pursuing
an optimal balance between the energy loss and sufficient lubrication of the system remains
open, but will be addressed in future work.

RI
Various approaches have been developed to study the churning losses inside a gearbox.
Moreover, numerous experiments have been conducted to investigate the losses due to a wide
set of parameters, e.g. Reynolds and Froude number [5], tip diameter and gear width [6], oil

SC
level and circumferential speed [7], etc. Empirical equations are derived accordingly from
the experimental data to provide guidelines for the design of gearboxes. However, empirical
equations are often limited by constraints and the application to general design purposes

U
is therefore not straightforward. More recently, because of the rapid growth in the
computational capability of Computational Fluid Dynamics (CFD) codes as well
AN
as the increase in High-Performance Computing (HPC), numerical simulations
are becoming a viable option for the investigation of gearboxes.
Surprisingly few detailed flow studies by experiments and CFD are present in the litera-
ture. The number of such experimental studies is particularly limited. Probably only works
M

[2] and [8] performed at Chalmers University of Technology provide experimental studies of
the flow field in the gearbox. The number of CFD studies on the oil distribution and viscous
D

losses of an oil-bath lubrication system is also very limited [9] [10] [11]. In these works,
the influence of the oil level, temperature and Reynolds number are studied respectively
TE

using different numerical models. The torque predictions from CFD are overall in a good
agreement with the experimental results [11].
The aforementioned researches generally provide us with a comprehensive understanding
of the churning losses due to different design parameters or operating conditions from both
EP

experimental and numerical perspectives. However, there is a general lack of flow field com-
parison studies between the experiments and the simulations[4], where accurate capturing
of fluid structures is essential in predicting the oil distribution. Moreover, the study of oil
C

flow inside the gearbox is essential in finding ways to reduce churning losses and gain a good
heat dissipation.
AC

Topic of multiphase-flow simulation has historically been a challenging one

and there is a large body of literature addressing it. Most of the finite volume
based CFD codes rely either on the volume of fluid (VOF) [12] or level set [13]
methods. The VOF method is particularly popular among the industry-oriented
commercial CFD packages, as it is capable of accurately simulating a range of
multiphase scenarios [14]. Some successful applications of the VOF methods
relevant to this work include [10][11][15][16]. A big advantage of any Eule-
rian code over Lagrangian codes is the capability to easily implement variable
mesh resolution and therefore focus the computational effort where it is needed.
2
 ACCEPTED MANUSCRIPT

Nonetheless, when applied to multiphase flows Eulerian codes inevitably en-

counter numerical diffusivity at fluid interfaces. This issue is well-known and
can be solved by increasing the resolution of the mesh or dynamically tracking
the interface, both which can make the simulation prohibitively computationally
expensive [17][18]. Alternatively it is possible to introduce a set of numerical
improvements in an attempt to mitigate the interface diffusion [17]. Moving

PT
complex geometries found inside of a typical gearbox assembly require creation
of elaborate meshes in order to gain reasonable results. Moreover, special treat-
ments, e.g. scaling of gear size, are required to handle numerical singularities

RI
due to the gap between the tooth flanks. Because of these drawbacks the Eule-
rian CFD methods are yet to establish themselves in this particular domain of
the industry.

SC
SPH multiphase methodology finds its roots in works of Colagrossi [19] and
more recently Monaghan [20]. An exhaustive overview of SPH multiphase ap-
proaches and numerical details can be found in a recent study by Wang et al.

U
[21]. For gear-train type of applications, e.g. complex moving geometries, SPH
[22] possesses several significant advantages over the grid-based methods. In
AN
SPH, the flow field is characterized by particles, which carry physical proper-
ties, and are updated following the movement of the flow in a fully Lagrangian
way. SPH inherently ensures Galilean invariance and conserves mass, momen-
tum and energy. The governing equations are discretized on scattered points,
M

i.e. SPH particles, through a smoothed kernel function, thus no specific geomet-
ric topology, i.e. mesh, is required. In this scenario, mesh distortion due to the
D

boundary motion can be completely avoided, and no geometric modification is

necessary to avoid numerical difficulties. In addition, preprocessing and setting
TE

up the particles also requires less effort than meshing in the grid-based methods.
Prominent downside of the SPH method is that the implementation of a variable
particle resolution is not straightforward and most of SPH codes are limited to
a uniform resolution throughout the domain of the simulation. This introduces
EP

computational inefficiency and limits accuracy in certain regions.

As mentioned, a set of experiments was carried out at Chalmers to study the oil flow in
a FZG back-to-back gearbox [2][4]. A spatial flow-field measurement method Particle Image
C

Velocimetry (PIV), is employed to capture the flow structure and velocity distribution.
Three different oil levels as well as three Reynolds numbers are, accordingly, considered to
AC

study the impact on flow and particularly on the aeration, size of the recirculation area, and
shape of the velocity profiles.
In the current paper, however, our main focus is on numerical simulations of oil flow inside
the gearbox and the comparison with the aforementioned experimental results. Instead of
using a traditional grid-based finite volume method, we develop a mesh-free CFD model
using the SPH method, utilizing a multi-phase formulation as well as some of the state-
of-the-art SPH techniques [23][24][25]. The presented numerical model is able to resolve
the complex fluid evolution and handle the interaction between the interfaces of different
materials. Numerical studies following the exact conditions in [2] are setup accordingly.
3
 ACCEPTED MANUSCRIPT

The fluid flow field is calculated and compared with PIV measurements. The objective is
to validate the proposed numerical model and to have a more comprehensive understanding
of the oil flow inside a gearbox from a numerical perspective. The SPH code used in this
study is nanoFluidX, developed by FluiDyna GmbH. nanoFluidX is a fully GPU (Graphical
Processing Unit) code. One of its main application domains is specifically power- and gear-
train components, where it efficiently harvests the aforementioned advantages of the SPH

PT
method.
The rest of the paper is arranged as follows: (1) The governing equations as well as
our SPH models are first elaborated in section 2; (2) the setups and parameters for our

RI
numerical experiments are introduced in section 3; (3) the simulation results are presented
and discussed in section 4. The aeration effect, velocity field and the discrepancies between
numerical and experimental results are studied respectively; (4) lastly, the conclusions are

SC
drawn in section 5.

2. Governing equations and SPH methods

U
In this section, the governing equations as well as the numerical methods utilized for
AN
multi-phase simulation are introduced.

2.1. Governing equations

The governing equations for isothermal fluid in a Lagrangian framework can be described
M

by the continuity equation

dρ
= −ρ 5 ·v, (1)
D

dt
and the momentum equation
TE

dv
ρ = − 5 p + Fv + Fs + ρg, (2)
dt
where ρ denotes the fluid density, v the velocity, p the pressure, Fv the viscous force, Fs the
EP

interfacial surface force, and g the body force.

The viscous force Fv can be expressed by
C

Fv = η 52 v, (3)
AC

where η is the dynamic viscosity.

The interfacial surface force is modeled by the continuum surface force (CSF)
method proposed in [26]. A color function c is defined to describe different
phases. A finite transitional band is used to characterized the interface, where
the gradient of the color function does not vanish. Within the transitional band,
the surface-tension force is approximated as a continuous force. The employed
CSF method recovers the prescribed surface tension and converges to the exact
value with increasing resolution [23]. In the CSF model, Fs is calculated following
Fs = −ακnδΣ , (4)
4
 ACCEPTED MANUSCRIPT

where δΣ is the surface delta function, α the surface-tension coefficient, κ the curvature, n
the normal vector of the interface.
The position of SPH particles is updated in a Lagrangian way following
dx
= v(t). (5)
dt

PT
In this paper, the fluid is modeled by the weakly-compressible approach. The
weakly compressible approach assumes simulating an incompressible fluid with

RI
lowered speed of sound, such that we allow for maximum of 1% of artificial
compressibility. Given that the code is fully explicit, lower speed of sound
relaxes the numerical CFL time step condition and allows for significantly faster

SC
computational times. The approach mitigates the need for solving of the stiff
Poissons pressure equation. Therefore, in order to calculate the pressure, we
require an artificial, stiff equation of state [27]

p = p0 [(
ρ γ
) − 1] + pb ,
U (6)
AN
ρ0
ρ0 c2s
where p0 = is the reference pressure, pb the background pressure, ρ0 the reference
γ
M

density, and cs the artificial sound speed.

In this paper, thermal effects were assumed negligible as the experiments
were performed at room temperature with a low RPM. Furthermore, simulation
D

of viscous heating or any type of fully coupled fluid-thermal solver requires

prohibitively long simulation times due to discrepancy in time scale of physics
TE

between fluid and heat dynamics.

2.2. Numerical discretization

EP

In the SPH method, the simulation domain is discretized with a set of particles carrying
the fluid properties, e.g. density, mass, velocity, etc. In the current multi-phase SPH
method, particles from different phases are assigned distinct identifiers and are
evolved by solving the same set of governing equations.
C

With a smoothing function W , any field quantity φ at coordinate r can be calculated

AC

by
Z
φ̄(r) = φ(r0 )W (r − r0 , h)dr0 . (7)

where r0 is the coordinate of an integration point, and dr0 is a differential volume

element. The radially symmetric smoothing function (kernel) W should reduce to a delta
function, i.e. lim W (r−r0 , h) = δ(r−r0 ), in the limit of a vanishing smoothing-length h. The
h→0
partition of unity property, i.e. W (r − r0 , h)dr0 = 1, is required to reproduce a constant
R

field.
5
 ACCEPTED MANUSCRIPT

The discrete form of the integration using particles can be expressed as the summation
over all the neighboring particles j within the smoothing-length h of the position of particle
i following
X
φ̄(r) = φ(rj )W (r − rj , h)Vj , (8)
j

PT
where V denotes the particle volume [22]. We employ the quintic-spline kernel with
cut-off radius of three times particle distance throughout the paper. One can
refer to [27] for a detailed description of the kernel.

RI
In order to tackle multi-phase problems, the gradient operator employed in this paper
follows [28] [29]. The governing equations are discretized as

SC
dρ X mj ∂W
= −ρi (vi − vj ) · eij , (9)
dt j
ρ j ∂r ij

U
and
dv 1 X 2 ∂W vi − vj ∂W αi κi ∂W
(Vi + Vj2 )(−p̃ij
AN
=− eij + η̃ij + c̃ij eij ) + g, (10)
dt mi j ∂rij rij ∂rij Vi ∂rij

∂W
where eij denotes the vector connecting particle i and j, the derivative of kernel function,
M

∂rij
ρj pi + ρi pj
and g the body force. p̃ij = is the density-weighted inter-particle averaged
ρi + ρj
D

pressure, which guarantees the conservation of linear and angular momentum exactly. η̃ij =
2ηi ηj
is a combined viscosity to enforce a continuous shear force across the interface.
TE

ηi + ηj
ρj ρi
c̃ij = cii + ci is the inter-particle averaged color function. The density-
ρi + ρj ρi + ρj j
weighted inter-particle color function c̃ij is utilized to reflect the reality of an
EP

asymmetrically distributed surface-tension force, and to obtain a stable surface

behavior for problems with large density ratio. The color function used for
calculating the surface-tension force is defined as
C

(
1 if the ith particle does not belong to the phase of particle j;
cij . (11)
AC

0 if the ith particle belongs to the phase of particle j;

For more details one can refer to [23][29].

Furthermore, in order to avoid the tensile instability and the creation of void regions in
the fluid field, we modify the governing equation by introducing a transport velocity formu-
lation, which is first proposed in [25]. An additional acceleration term is introduced
following
 
dvi 1
= − 5 p∗b , (12)
dt c ρ
6
 ACCEPTED MANUSCRIPT

where p∗b is the background

  pressure, with a value of the same order of p0 . The
dvi
additional term is employed to modify the momentum velocity
dt c
 t
dvi
ṽt+δt
i = vt+δt
i + δt , (13)
dt

PT
c

where ṽ is the transport velocity used to update particle positions.

It is well-known that the traditional SPH method has the difficulty of zeroth-order in-

RI
consistency, which implies the inability of reproducing a constant field, especially when the
particles have non-uniform distribution. The transport velocity facilitates the homog-
enization of the particle configuration, and at the same time reduces artificial

SC
numerical dissipation [25]. It has been recently proved that, with the presence
of transport velocity, higher order of integration error can be achieved [30].

2.3. Boundary conditions

U
To simulate oil flow inside a gearbox, it is vital that the numerical method is capable
AN
of handling interactions between fluid and rigid body with complex geometries
and motions. In the SPH method, various approaches can be applied to model solid wall
boundary conditions. To achieve physically-consistent simulation results near the solid walls,
two constraints should be complied, i.e. the impermeable condition and local load balance.
M

In this paper, the solid wall treatment is as described in [24]. The solid wall particles are
treated as ghost particles and are constructed along the interface in four layers
of particles to ensure the compact support of the kernel area, see Fig 1 (a). A
D

pair-wise force equilibrium relation is solved between fluid and solid particles to achieve local
load balance by
TE

Z Z
5p · dl = ρf (g − aw ) · dl, (14)
EP

where dl is the vector connecting the centerline of a fluid and a solid particle.
The indices f and w denote fluid and wall particle, respectively. Fig 1 (b) shows
the interaction between a solid particle and the surrounding fluid particles. The
C

impermeable condition is satisfied by imposing a pressure condition and including solid

particles in the density evolution of fluid particles. The free-slip and non-slip conditions are
AC

enforced by extrapolating fluid velocity to solid particles and including boundary velocity
in the force calculation. For a detailed description, one can refer to [24].

The boundary condition is general and able to handle complex geometries in three di-
mensions. Moreover, both static and moving walls can be simulated by simply including
boundary velocity in the solid particles.

7
 ACCEPTED MANUSCRIPT

gear particle
pinion particle
air particle

PT
oil particle

RI
(a) (b)

SC
Figure 1: Sketch of (a) the particle packing at the meshing area and (b) the dummy particle boundary condition.

2.4. Time integration

U
The time-marching algorithm employed in this paper is the Leapfrog integration
scheme [31]. This scheme achieves 2nd-order and is reversible in time in the absence of
AN
viscosity. The reader can refer to [32] for a more detailed discussion on the accu-
racy of the scheme.
First, the intermediate momentum velocity and the transport velocity are
M

updated
 n
n+1/2 n δt dvi
vi = vi + , (15)
2 dt
D

 n
δt dvi
TE

n+1/2 n+1/2
ṽi = vi + , (16)
2 dt c
 n
dvi
where is the correction term for the particle acceleration. The density
EP

dt c
and particle position is then updated for the new time step by
 n
n+1 n dρi
ρi = ρi + δt , (17)
C

dt
AC

rn+1
i = rni + δtṽni , (18)
Finally, the momentum velocity at the new time step is updated following
 n+1
n+1 n+1/2 δt dvi
vi = vi + . (19)
2 dt

The time-step size of the simulation is calculated with respect to the physical model
involved. In this paper, constant time-step size is used throughout the simulation.

8
 ACCEPTED MANUSCRIPT

The time-step size is decided by the CFL condition, the condition of viscosity, the condition
of body force and the condition of surface tension respectively [23], see Eq. 20:

h h2 h ρh3 1/2
4t = M IN (0.25 , 0.125 , 0.25( )1/2 , 0.25( ) ). (20)
cmax + |vmax | υ |g| 2πσ

PT
3. Numerical and experimental setup
In this section details regarding to our simulation setups are presented. Firstly, the

RI
geometry of the gearbox as well as the operating conditions are introduced in sub-section
3.1. Then, in sub-section 3.2, setups for numerical models and parameters are elaborated
upon.

SC
3.1. Geometry and operating conditions
The experimental rig utilized in this study is designed at Chalmers University of Tech-
nology (referred to as Chalmers gearbox rig), which is based on the FZG back-to-back gear

U
test rig [2]. The dimensions and layout of the gearbox model were kept the same as the FZG
AN
gear test rig [33]. Fig. 2 illustrates the detailed measurements of the Chalmers gearbox rig.
The FZG type C gear [33] is employed to study the oil behavior inside the gearbox. The
test section of the gearbox model was made from PMMA to accommodate optical access
for the PIV measurement. The method used for the PIV measurement is 2D-2C PIV, also
M

known as planar PIV. The details of the experimental setup and results can be found in [2].
A hydrotreated process oil, Nytex 810 [34], is chosen as the test lubricant due to its good
optical properties. During the experiment, the temperature of the test section is maintained
D

at 20 ◦ C, thus the oil viscosity is kept approximately constant at 64 cSt.

TE

Gearbox outer wall

EP

Gear
C

Pinion
AC

Figure 2: Detailed geometry of the gearbox.

9
 ACCEPTED MANUSCRIPT

Three oil levels, i.e. centerline, 2x module of gear, and 2x module of pinion, are studied
in this paper. Fig. 3 shows the oil levels with respect to the position of gears. Here
we describe the oil level relative to the gear module to be consistent with the
experiment. Module is the unit of size that indicates how big or small a gear is,
and relating the oil level to the gear module can better characterize the mutual
relationship between the gear size and the lubricating level. In addition, three

PT
pitch-line velocities Vt , i.e. 0.55 m/s, 1.1 m/s and 1.62 m/s, are also investigated in order
to study the influence on oil distribution.

RI
U SC
AN
Figure 3: Three oil levels: (a) oil level centerline; (b) oil level 2x modules of pinion; (c) oil level 2x modules of gear.

3.2. Numerical models

M

Three-dimensional models of the test rig are built for this study. The pre-processing of
particle generation is accomplished by using Altair HyperMesh (version v2017.0.0.8) com-
D

mercial software.
In our numerical model, the computational domain is setup including all the structures
TE

and fluid field inside the housing (outer wall of gearbox). To characterize the system,
components of the test rig are treated and built separately. The housing is modeled as a
static wall. The gear and pinion are represented by moving walls with different rotational
speeds. The rest of the domain inside the housing is filled with fluid particles. The volume
EP

ratio between oil and air corresponds to the filled oil level. According to the previous
description in section 2.3, we fill the boundary of all solid walls with four layers of dummy
particles to prevent the penetration of fluid particles.
C

The spatial pitch (4x) of SPH particles for the initial conditions is set at constant
throughout the computational domain. Since, in the SPH framework, particles move in a
AC

full Lagrangian way, no special treatment or scaling of particle pitch is necessary to model the
region between the teeth flanks. In order to find a compromise between the quality
of the results and the computational cost, it is important to carefully choose
the particle size. Through various numerical tests, e.g. see [35], we found that
setting four particles at the tip of gear teeth, i.e. 4x = 1 mm, is an acceptable
particle scale for all the cases in the current paper. Consequently, the total number
of particles created for the model is 3,088,805.
We simulate each case for 5s of physical time. Steady states are achieved for all
the cases by monitoring the total kinetic energy of the system. A damping time
10
 ACCEPTED MANUSCRIPT

of 0.2s is set to boost the rotating velocity of the gear and pinion gradually from 0 to the
operation conditions in order to alleviate the violent change of fluid state and the generation
of instabilities from the initial condition.
For numerical parameters, we use the same set of values for all the models for the sake of
consistency. We set γ = 1 and pb = 0 in the EOS. The background pressure p∗b for transport
velocity is set equal to p0 . α in Eq.4 is equal to 0.03 for both air and oil phase.

PT
4. Results and discussion

RI
In this section, the simulation results are presented and characterized into three cate-
gories, corresponding with the PIV measurements [2].

SC
4.1. Bubble size and quantity
Due to the rotation of gears, voids and low-pressure regions are created between the
teeth, and consequently, air is entrained into the oil periodically. The trapped air inside the

U
liquid forms into bubbles of distinct sizes. Bubble generation is inevitable inside a gearbox
and may cause serious issues according to [36]. This phenomenon is usually characterized
AN
as aeration. The impact of aeration on the performance of gearboxes is twofold. Firstly,
extra heat is generated due to the work of compressing bubbles formed between the teeth,
and the aerated oil reduces the heat conductivity from gear to oil and from oil to housing,
behaving as an insulator. Therefore, the temperature of the oil and gears rises. Meanwhile,
M

the existence of bubbles creates a dampening effect on the flow and increases the energy
required to pump the oil, thus resulting in increased churning losses [8].
In our simulation, since we model both air and oil phases, it is straightforward to quantify
D

the amount of air particles entrained into the oil. Fig. 4 illustrates the comparison of bubble
size and quantity around the gear between experimental laser induced fluorescence images
TE

and simulations. The images are from the highest oil fill level cases and illustrate the
impact of the pitch-line velocity on the aeration. Two visualization methods are used
to characterize air bubbles for numerical results. In Fig. 4 row (b), we present air
EP

by isolated spheres with the same diameter as the SPH particle. In Fig. 4 row
(c), a zoomed-in view near the gear teeth is illustrated. In addition to the plot
of SPH particles with individual spheres, iso-surfaces are created for air phase
C

to characterize individual bubbles entrapped in the oil. It is worth noticing

that the group of air particles enclosed by one iso-surface is counted as a single
AC

bubble instead of a cluster of small bubbles, since their motions are confined by
a continuous surface-tension force at the phase interface.
According to the experimental results [2], three types of bubbles can be observed in
the flow field, and they are classified by size and concentration (see Fig. 4, a). The first
kind are the largest bubbles trapped between the gear teeth. Similarly, large air particle
clusters can be observed in our simulation results. In Fig. 4 (b), the large clusters of air
particles are concentrated in the region between the teeth, and they can be considered as a
single bubble. The size of the largest bubbles grows with increasing of rotational
speed. At the lowest Vt , the bubble size is approximately 6mm. The size grows
11
 ACCEPTED MANUSCRIPT

(a)

PT
RI
SC
(b)

U
AN
0.02m 0.02m 0.02m
M
D

(c)
TE
EP

0.01m 0.01m 0.01m

Vt=0.55 m/s Vt=1.1 m/s Vt=1.62 m/s

C

Figure 4: Comparison of bubble size and quantity around the gear with respect to different flow Reynolds numbers. (a).
Experimental laser induced fluorescence images. (b). Simulation results. Air is represented by white particles. (c). Zoomed-
AC

in simulation results. Air is represented by white particles, and iso-surfaces are created for air phase from
Paraview.

to about 10mm and 12mm for Vt = 1.1 and Vt = 1.62 respectively. The second kind
of bubbles are the medium-sized bubbles, which are primarily concentrated near the gear.
In particular, these bubbles are visible in experimental images in front of the gear. From
our simulation results, air bubbles of size 2mm to 5mm can be observed near the gear,
and we can classify them as the medium-sized bubbles. The quantity and diameter of
the medium-sized bubbles increases with Vt . Lastly, there are small-sized bubbles

12
 ACCEPTED MANUSCRIPT

distributed in the entire oil bath. A large number of isolated SPH particles scattered in the
oil can also be observed. We consider them as smaller-sized bubbles not fully resolved at
current numerical resolution, and the size of these bubbles is 1mm due to the applied
resolution. In the experiment, since there is no ruler installed as reference, the
size of bubbles can only be approximated. It can be observed that the bubble
sizes from simulation are generally larger than the experimental result. The

PT
discrepancy may due to the limitation of current resolution.
To conclude, the present result is able to resolve all types of bubbles observed
from the experiment. The variation of both bubble size and quantity due to the

RI
increase of rotational speed is captured qualitatively.

4.2. Velocity field

SC
In this section, the simulation results of velocity field for all studied operation conditions
are presented and compared with the experimental data. The physical time we simulated is
5s, including a damping interval of 0.2s. All the time-averaged results illustrated here are

U
calculated from 1s to 5s to exclude the influence of the transient period of the system. In
addition, to calculate the time-averaging physical field, we interpolate particle data to a
AN
2D background cell mesh in the middle plane with identical resolution, i.e. 1mm. The kernel
employed for extrapolation is quintic spline. Before plotting the final result, P we renormalize
the extrapolated data with the Shepard coefficient, which is defined as j W (ri − rj , hi ),
M

to ensure the integration of particle weights inside the cut-off radius is always equal to 1.
To obtain a better understanding of the velocity distribution and more comprehensive
comparison with the PIV results, the velocity field is plotted with contours/cloud maps,
D

stream-lines and vectors (Figs. 5, 6 and 7). In each of these figures, the circumferential
velocity is the same, while results from different oil levels are compared. It is worth not-
TE

ing that, in the PIV result, the velocity distribution above the oil level is considered as a
qualitative result due to the non-uniform illumination intensity [2].
First, we have an overview of all the contour maps of the velocity field (2nd row of Figs.
5, 6 and 7), the results show that the oil at the tip of the teeth is always propelled at the
EP

same velocity as the local circumferential velocity of the gear. The calculation manifests a
similar physically consistent result, since, according to the Stokes adhesive condition, the
velocity of a Newtonian fluid particle on a moving wall is identical to the wall velocity.
C

However, at a small radial distance from the gears, the flow velocity experiences a vast
drop in magnitude. The boundary layers formed on the gears do not stay attached on the
AC

entire gear surface and separate at an angular position of 45 degrees (corresponding to 3

teeth) from the point of entrance into the oil bath. The same location of the separation
is observed both in experiments and in CFD. In Figs. 6 and 7, in the c) columns,
the experiments show presence of oil (velocity) around both the gear and the
pinion, whereas the simulation does not. This can be attributed partially to the
high experimental uncertainty above the oil surface, but also the code’s inability
to resolve thin films on the surface of the gears due to lack of variable particle
resolution.

13
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP

(a) Oil level = centerline (b) Oil level = 2 x module of pinion (c) Oil level = 2 x module of gear
C

Figure 5: Comparison of overall mean velocity fields with Vt = 0.55m/s. Row No. 1, 3 and 5 are experimental results, and
are presented with velocity clouds, streamlines and vectors respectively. Row No. 2, 4 and 6 are experimental results, and are
presented with velocity clouds, streamlines and vectors respectively.
AC

The influence of oil level on the velocity field is significant. At the higher oil fill level,
more oil is accelerated and splashed into the air. For oil level centerline and 2x module of
pinion, the splash raised by the gear is more prominent than by the pinion, since the tip
diameter is the dominant factor in spin effect [9]. Whereas, for oil level 2x module of gear,
the pinion does not touch the oil bath, thus no oil is propelled from the pinion side. From
the PIV results, we can make the same conclusions.
Moreover, the flow behavior becomes remarkably different as the pitch-line velocity

14
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP

(a) Oil level = centerline (b) Oil level = 2 x module of pinion (c) Oil level = 2 x module of gear
C

Figure 6: Comparison of overall mean velocity fields with Vt = 1.1m/s. Row No. 1, 3 and 5 are experimental results, and
are presented with velocity clouds, streamlines and vectors respectively. Row No. 2, 4 and 6 are experimental results, and are
presented with velocity clouds, streamlines and vectors respectively.
AC

changes. In general, the splash generated by gears is increased with the increasing speed
of rotation. When the pitch-line velocity increases, more energy is transmitted to the oil
and the momentum of the oil particles increases. As a result, the flow behavior shifts from
viscous-dominated to inertia-dominated. In the case Vt = 1.62m/s, the splash momentum
is sufficiently high enough to hit the top wall and strike back to hit the gear again (Fig 7).
The same behavior is observed both in experiments and CFD.
Regarding to the flow structure, i.e. the recirculation region, beneath the oil surface,

15
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP

(a) Oil level = centerline (b) Oil level = 2 x module of pinion (c) Oil level = 2 x module of gear
C

Figure 7: Comparison of overall mean velocity fields with Vt = 1.62m/s. Row No. 1, 3 and 5 are experimental results, and
are presented with velocity clouds, streamlines and vectors respectively. Row No. 2, 4 and 6 are experimental results, and are
presented with velocity clouds, streamlines and vectors respectively.
AC

the simulation results are generally in a good agreement with the PIV results. The simulation
is able to capture most of the flow structures. The number of the recirculation areas and
the position of the recirculation vortices are predicted well. However, it can be observed
that there are two discrepancies in the results from the gear side at oil level centerline with
Vt = 0.55m/s and Vt = 1.1m/s (see the left side of Fig. 5 (a) and Fig. 6 (a)). Regarding
to Vt = 0.55m/s (Fig. 5 (a)), the PIV results show a counter flow formed on the gear
side close to the crossflow plane at the meshing region, and the counter flow ends on the

16
 ACCEPTED MANUSCRIPT

left of the gear centerline. However, from the simulation results, the recirculation region is
located across the region below the gear. In the case Vt = 1.1m/s (Fig. 6 (a)), the PIV
result reveals one recirculation area at the pinion side, whereas, from the simulation result
one additional recirculation area can be observed at the left corner of the box. Thus, the
highly decelerated flow region below the gear appears different in the CFD and experiments;
however, the location of the flow separation and the decelerated region in the vicinity of the

PT
gear are very similar in the experiments and CFD.

4.3. Velocity profile

RI
U SC
AN
M

(a) Oil level = centerline (b) Oil level = 2 x module of pinion (c) Oil level = 2 x module of gear

Figure 8: Comparison of mean absolute velocity profile at gear centerline. The abscissa shows the local velocity scaled by pitch
line velocity (V /Vt )and the ordinate is distance from the bottom of the housing (d) in meters.
D

In order to better understand the flow feature beneath the oil level, the mean absolute
TE

velocity (V /Vt ) profile is plotted and compared with experimental results. The profiles
are extracted beneath the gear and beneath the pinion at the centerline of their axes. Fig.
8 illustrates the results from the gear centerline and Fig. 9 the results from the pinion
centerline. Since the pinion does not touch the oil bath at oil level 2x module of gear, the
EP

velocity profiles are not presented for this case.

The results presented in these figures support the above conclusions. Good agreement
is observed between experimental and numerical velocity profiles. For the profiles beneath
C

the gear centerline at oil level centerline the experiments show thinner boundary layers
and more uniform velocity distribution, see Fig. 8 (a). The calculated profile exhibits a
AC

kink at Vt = 0.55m/s, whereas the PIV result suggests that no counter flow is created.
For case Vt = 1.1m/s and Vt = 1.62m/s, no counter flow is observed in the calculations
and in the PIV results. A larger discrepancy between the experimental and numerical
velocity profiles is observed in Fig. 8 (a) compared to other discussed figures. Other profiles
depicted in Figs. 8 and 9 show a good agreement between experiments and CFD. CFD very
well captures the main features of the flow and reproducing the velocity profiles. Several
explanations might exist for the discrepancy observed in Fig. 8 (a). From the
experiment conducted in [4], the velocity field in the direction perpendicular
to the PIV measurement plane is not symmetric. However, in the numerical
17
 ACCEPTED MANUSCRIPT

simulation the velocity field is symmetric coinciding with the geometry. The
asymmetry of flow field in the crossflow plane may cause the discrepancy in the
measurement performed in the presented plane. From numerical side, turbulent
effects are a common source of error in the current method. The code is capable
of producing turbulent-like behavior to an extent demonstrated by [37]. For
wall-bounded flow simulations, improved wall modeling is required to better

PT
resolve the turbulent effect. Other possible explanations are more exotic and
would include additional physical models, e.g. aeration effects and heat transfer.

RI
U SC
AN
(b) Oil level = centerline (d) Oil level = 2 x module of pinion
M

Figure 9: Comparison of mean absolute velocity profile at pinion centerline. The abscissa shows the local velocity scaled by
pitch line velocity (V /Vt )and the ordinate is distance from the bottom of the housing (d) in meters.
D

4.4. Uncertainty analysis of the PIV results

TE

In this paper, the PIV uncertainty is evaluated by using Davis 8.3.1 software. The
method used for the uncertainty calculation is the correlation statistics. Details of this
method can be found in paper [38].
EP

Fig. 10 shows contour plot of uncertainty magnitude for different pitch-line velocities
and different immersion depths. The highest uncertainty value is observed in the air regions
where the gear is not immersed in the oil. This is expected due to smaller amount of statistics
C

in the data. Medium uncertainty is observed in the regions of high turbulence, such as the
area close to the gear teeth in the submerged region. In the region far from the gear teeth
AC

the uncertainty value is very close to zero.

Table 1: Relative uncertainty for centerline oil level.

Relative uncertainty scaled by the pitch line velocity in %

Pitch Line Velocity
Air region Submerged near region Submerged far region
0.55 m/s 0.03 − 0.5 0.02 − 0.5 0 − 0.009
1.1 m/s 0.02 − 0.12 0.05 − 0.12 0 − 0.02
1.62 m/s 0.08 − 0.2 0.015 − 0.04 0 − 0.015

Analyzing the magnitude of the uncertainty, it can be seen that it is proportional to

the local velocity magnitude. Table 1 shows the relative uncertainty value scaled by the
18
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M

Figure 10: Uncertainty contour plots for different immersion depths (top to bottom: centerline, 2 module pinion, 2 module
gear) and different pitch line velocities (left to right: 0.55, 1.1, 1.62 m/s).
D

pitch line velocity. The table shows the minimum and maximum uncertainties in three
TE

different regions: air region, submerged near region, and submerged far region. The relative
uncertainty supports the previous statement that, in high noise and high fluctuation regions,
the uncertainty value is higher. Overall, the experimental uncertainties of the PIV algorithm
itself are very low.
EP

5. Conclusions
C

In this paper, a particle-based CFD model using a SPH method is presented to investigate
the oil flow inside a gearbox. Three oil fill-levels, i.e. centerline, 2x module of gear, and
AC

2x module of pinion, and three circumferential velocities, i.e. Vt = 0.55 m/s, 1.1 m/s, and
1.62 m/s, are considered to study the impact on the flow behavior. The calculation results
are compared with PIV measurements. The conclusions from the study are listed below:

1. The aeration effect at oil level centerline is studied. Three types of air bubbles are
observed. Both the quantity and size of the bubbles generated in the oil increase due to
the increased rotational speed of the gears. The simulation results reveal good qualitative
agreement with experimental observations at various operating conditions.

19
 ACCEPTED MANUSCRIPT

2. The calculated flow field is analyzed in detail via various visualization methods. Phys-
ically consistent results are obtained by our CFD model. The calculation results manifest
good agreement with experimental results, and flow structures are captured well.

3. The study of the influence of the oil level reveals that, as the gear submerging depth
is increased, larger oil mass is accelerated and splashed into the air. The gear tip velocity

PT
plays a dominant role in spin effect of generating splashes.

4. The velocity profiles at two locations, gear centerline and pinion centerline, are analyzed
to study the flow beneath the oil free surface. Good agreement between numerical and

RI
experimental results is observed. The profile shape and velocity magnitudes are well pre-
dicted. However, some discrepancies are observed for the velocity profiles beneath the gear

SC
centerline at oil level centerline.

5. The uncertainty analysis reveals that the PIV measurements of the velocity field is reli-
able. To improve our numerical results, additional physical models, (e.g. aeration

U
effects, heat transfer), improved wall modeling, and higher-resolution simula-
tions are to be considered.
AN
In the future, we will carry out the optimization of gearbox in order to study the oil flow
with respect to different geometries using the same numerical model. The objective is to
investigate the viscous effect and decrease the churning losses.
M

Acknowledgements
References
D

[1] S. Chu, A. Majumdar, Opportunities and challenges for a sustainable energy future, Nature 488 (7411)
TE

(2012) 294.
[2] E. Hartono, M. Golubev, V. Chernoray, Piv study of fluid flow inside a gearbox, in: PIV13; 10th
International Symposium on Particle Image Velocimetry, Delft, The Netherlands, July 1-3, 2013, 2013.
[3] B.-R. Höhn, K. Michaelis, H.-P. Otto, Influence of immersion depth of dip lubricated gears on power
EP

loss, bulk temperature and scuffing load carrying capacity, International Journal of Mechanics and
Materials in Design 4 (2) (2008) 145–156.
[4] E. A. Hartono, A. Pavlenko, V. Chernoray, Stereo-PIV study of oil flow inside a model gearbox, in: 17th
International Symposium on Applications of Laser Techniques to Fluid Mechanics, 2014, pp. 07–10.
C

[5] R. Boness, Churning losses of discs and gears running partially submerged in oil, in: Proc. ASME Int.
Power Trans. Gearing Conf, Vol. 1, 1989, pp. 255–359.
AC

[6] C. Changenet, P. Velex, A model for the prediction of churning losses in geared transmissions-
preliminary results, Journal of Mechanical Design 129 (1) (2007) 128–133.
[7] B. Höhn, K. Michaelis, H. Otto, Influence on no-load gear losses, in: Proceedings of the Ecotrib
Conference, Vol. 2, 2011, pp. 639–644.
[8] E. Hartono, Study of fluid flow inside the gearbox, PhD dissertation, Chalmers University of Technology
(2014).
[9] H. Liu, T. Jurkschat, T. Lohner, K. Stahl, Determination of oil distribution and churning power loss
of gearboxes by finite volume CFD method, Tribology International 109 (2017) 346–354.
[10] C. Gorla, F. Concli, K. Stahl, B.-R. Höhn, M. Klaus, H. Schultheiß, J.-P. Stemplinger, CFD simulations
of splash losses of a gearbox, Advances in Tribology 2012.
20
 ACCEPTED MANUSCRIPT

[11] F. Concli, C. Gorla, Numerical modeling of the power losses in geared transmissions: Windage, churning
and cavitation simulations with a new integrated approach that drastically reduces the computational
effort, Tribology International 103 (2016) 58–68.
[12] C. Hirt, B. Nichols, Volume of fluid (VOF) method for the dynamics of free boundaries, Journal of
Computational Physics 39 (1) (1981) 201 – 225. doi:https://doi.org/10.1016/0021-9991(81)90145-5.
URL http://www.sciencedirect.com/science/article/pii/0021999181901455
[13] S. Osher, R. P. Fedkiw, Level Set Methods: An Overview and Some Recent Results, Journal of Com-

PT
putational Physics 169 (2) (2001) 463 – 502. doi:https://doi.org/10.1006/jcph.2000.6636.
URL http://www.sciencedirect.com/science/article/pii/S0021999100966361
[14] J. E. Olsen, S. T. Johansen, Progress in Applied CFD: Selected papers from 10th International Con-

RI
ference on Computational Fluid Dynamics in the Oil & Gas, Metallurgical and Process Industries
(2015).
[15] F. Concli, C. Gorla, Oil squeezing power losses in gears: a CFD analysis, WIT Transactions on Engi-
neering Sciences 74 (2012) 37–48.

SC
[16] F. Concli, A. Della Torre, C. Gorla, G. Montenegro, A new integrated approach for the prediction of
the load independent power losses of gears: development of a mesh-handling algorithm to reduce the
CFD simulation time, Advances in Tribology 2016.

U
[17] V.-T. Nguyen, W.-G. Park, A free surface flow solver for complex three-dimensional water impact
problems based on the VOF method, International Journal for Numerical Methods in Fluids 82 (1)
(2016) 3–34.
AN
[18] K. E. Wardle, H. G. Weller, Hybrid multiphase CFD solver for coupled dispersed/segregated flows in
liquid-liquid extraction, International Journal of Chemical Engineering 2013.
[19] A. Colagrossi, M. Landrini, Numerical simulation of interfacial flows by smoothed particle hydrody-
M

namics, Journal of Computational Physics 191 (2) (2003) 448–475.

[20] J. Monaghan, A. Rafiee, A simple SPH algorithm for multi-fluid flow with high density ratios, Inter-
national Journal for Numerical Methods in Fluids 71 (5) (2013) 537–561.
[21] Z.-B. Wang, R. Chen, H. Wang, Q. Liao, X. Zhu, S.-Z. Li, An overview of smoothed particle hydrody-
D

namics for simulating multiphase flow, Applied Mathematical Modelling 40 (23) (2016) 9625 – 9655.
doi:https://doi.org/10.1016/j.apm.2016.06.030.
TE

URL http://www.sciencedirect.com/science/article/pii/S0307904X16303419
[22] J. J. Monaghan, Smoothed particle hydrodynamics, Annual Review of Astronomy and Astrophysics
30 (1) (1992) 543–574.
[23] S. Adami, X. Hu, N. Adams, A new surface-tension formulation for multi-phase SPH using a reproducing
EP

divergence approximation, Journal of Computational Physics 229 (13) (2010) 5011–5021.

[24] S. Adami, X. Hu, N. Adams, A generalized wall boundary condition for smoothed particle hydrody-
namics, Journal of Computational Physics 231 (21) (2012) 7057–7075.
[25] S. Adami, X. Hu, N. A. Adams, A transport-velocity formulation for smoothed particle hydrodynamics,
C

Journal of Computational Physics 241 (2013) 292–307.

[26] J. Brackbill, D. B. Kothe, C. Zemach, A continuum method for modeling surface tension, Journal of
AC

Computational Physics 100 (2) (1992) 335–354.

[27] J. P. Morris, P. J. Fox, Y. Zhu, Modeling low Reynolds number incompressible flows using SPH, Journal
of Computational Physics 136 (1) (1997) 214–226.
[28] X. Hu, N. A. Adams, An incompressible multi-phase SPH method, Journal of Computational Physics
227 (1) (2007) 264–278.
[29] X. Y. Hu, N. A. Adams, A multi-phase SPH method for macroscopic and mesoscopic flows, Journal of
Computational Physics 213 (2) (2006) 844–861.
[30] S. Litvinov, X. Hu, N. Adams, Towards consistence and convergence of conservative SPH approxima-
tions, Journal of Computational Physics 301 (2015) 394–401.

21
 ACCEPTED MANUSCRIPT

[31] J. J. Monaghan, Smoothed particle hydrodynamics, Reports on Progress in Physics 68 (8) (2005) 1703.
[32] B. J. Leimkuhler, S. Reich, R. D. Skeel, Integration methods for molecular dynamics, in: Mathematical
Approaches to Biomolecular Structure and Dynamics, Springer, 1996, pp. 161–185.
[33] B. R. Hoehn, P. Oster, T. Tobie, K. Michaelis, Test methods for gear lubricants, Goriva i maziva 47 (2)
(2008) 141–152.
[34] Product data sheet - nytex 810, https://nyport.nynas.com/Apps/1112.nsf/%20wnpds/Nytex_810_
ASTM/$File/PDS_Nytex_810_EN.pdf, accessed 2016.

PT
[35] H. Liu, G. Arfaoui, M. Stanic, L. Montigny, T. Jurkschat, T. Lohner, K. Stahl, Numerical modelling
of oil distribution and churning gear power losses of gearboxes by smoothed particle hydrodynamics,
Proceedings of the Institution of Mechanical Engineers, Part J: Journal of Engineering Tribology (2018)

RI
1350650118760626.
[36] J. Fitch, The perils of aerated oil-let your machine burp, Machinery Lubrication.
[37] S. Adami, X. Y. Hu, N. Adams, Simulating 3D turbulence with SPH, Tech. rep., Center for Turbulence

SC
Research, Stanford University (2012).
[38] B. Wieneke, PIV uncertainty quantification from correlation statistics, Measurement Science and Tech-
nology 26 (7) (2015) 074002.

U
AN
M
D
TE
C EP
AC

22
 ACCEPTED MANUSCRIPT
Research highlights

1. The aeration effect is well captured by the simulation

2. The flow structure is consistent with experimental measurements
3. The velocity profile is in a good agreement with PIV results
4. The PIV results of the velocity field are reliable

PT
5. The numerical model is verified by the comparison with experimental results

RI
Yours sincerely,
Zhe Ji

SC
FluiDyna GmbH, Edisonstrasse 3

U
D-85716 Unterschleissheim, Germany
E-mail: [email protected]
AN
M
D
TE
C EP
AC