You are on page 1of 29

ed

_____________________________________________________________________________________

Quasi-static and Dynamic Transfer Functions for the


Analytical Modelling of a Helicopter Planetary Gearbox
Dynamic Behavior

iew
Raphaël Mignot-Pastya,b,*, François Malbureta, Olivier Honnoratb , Lionel Roucoulesa

a Arts et Metiers Institute of Technology, HESAM Universite, LISPEN, F-13617 Aix-en-Provence, France

ev
b Airbus Helicopters, Aéroport Marseille Provence, 13725 Marignane, France
_____________________________________________________________________________________
ABSTRACT

r
Gearbox dynamic modeling is a wide topic that has been extensively treated in the literature. This research
is nevertheless still on going to better understand the vibrations produced by the gearboxes and to predict

er
the defect signatures that should be monitored to prevent catastrophic failures. Most planetary gear train
models presented in the literature are proposing results in the rotating reference of the planet-gear carrier,
while real-life monitoring is usually performed through accelerometers positioned outside the gearboxes in
pe
a fix-reference. This paper is proposing an approach based on quasi-static and dynamic transfer functions
to transform the rotating-reference results of a dynamic model considering a helicopter planetary gear train
towards the fix-reference of the sensor positioned outside the ring gear of the gearbox. 2D and 3D transfer
functions are calculated by using Finite-Elements models of the gearbox housings and by performing
ot

impact and shake-test on a real helicopter Main Gearbox. A methodology is finally proposed to apply these
transfer functions to the contacting efforts of the planet gears predicted in the rotating reference, and
frequency spectra are calculated in the fix-reference of the accelerometer. Theoretical fix-reference mesh-
tn

harmonics are finally checked.

Keywords
rin

Dynamic modelling; Planetary gear train; Transfer function; Finite Element Modeling; Impact testing

_____________________________________________________________________________________

1. Introduction
ep

Helicopters Main Gearboxes (MGB) are complex systems composed of several transmission components
such as gears, bearings or splines. Many gearboxes involve planetary gear as they offer high-speed
reduction ratios within a limited space, with high-efficiencies co-axial power transmission. The higher
Pr

stages of a typical Helicopter MGB are illustrated in Fig. 1: they are composed of two serial planetary gear

____________
* Corresponding author.
E-mail address : raphael.mignot-pasty@ensam.eu
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

trains sharing a common ring gear. Power is transmitted in the system from the lowest sun gear; it is shared

ed
between all the planet gears, which all contribute to the rotation of the planet gear carrier. MGB are highly
critical systems as they represent the sole link between the Main Rotor (MR), which provides the aircraft
thrust and lift, and the aircraft structure. Monitoring the gearbox while operating is therefore essential,

iew
especially to detect the emergence of defects such as cracks, spalls (on the gears or on the bearings),
mounting or geometrical errors which could endanger the system. The vibration measurement through
accelerometers is one of the main monitoring strategies: vibrations coming from all the transmission
components are measured in-flight and can either be treated on-ground after the aircraft mission, or during
the mission to alert the pilots in case of any failure. For accessibility and security purposes, sensors are

ev
positioned on fix supports outside the gearbox and can namely be mounted at the junction of the housings
and the ring gear, see Fig. 1.

r
er
pe
ot
tn
rin

Fig. 1 A helicopter Main Gearbox with two stages of planetary gear train

In a planetary gear train system, vibration is due to the dynamic excitation applied to the ring gear through
the simultaneous contacting teeth of all the planet gears, in the rotating reference carried by the planet gear
ep

carrier. Vibration are then transmitted to the fix-reference sensor outside the ring gear through the housing
components: the ring, the lower and the conical housing. To model the dynamic behavior of the planetary
gear train is a practical way of predicting the dynamic excitation of the planet gears on the ring. Many
Pr

researches have already been conducted to model the dynamic behavior of planetary gear systems but
results of such models are expressed in the carrier rotating reference. An additional translation operation

2
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

towards the fix reference of the accelerometer should be performed to calculate the vibration signal, which

ed
would be measured by the sensor under the given operating condition with or without defect. This paper is
focusing on this particular translation operation though the definition of transfer functions. Quasi-static and
dynamic versions are studied through Finite Element Models (FEM) and bench tests to characterize the

iew
system. The MGB illustrated in Fig. 1 is used as a study-case. Modeling strategies used to link those transfer
functions to the results of the dynamic model of the planetary gear trains are also developed and results are
given from the fix-reference accelerometer point of view in the frequency domain.

2. Review of Literature

ev
2.1. Planetary gear train dynamic modelling review

Planetary gear trains dynamic behavior has been extensively studied during the past six decades through
many modelling strategies, as reviewed by Parker and Cooley [1]. Models have been used to understand

r
and predict several physical aspects of those complex systems, like the load sharing between all the involved
gears, determining the fundamentals frequencies or studying different kinds of influences like geometrical

er
modifications or defects on their static / quasi-static behavior or on the produced vibrations.
pe
ot
tn
rin

LPM model Finite Element Model

Fig. 2 LPM versus FEM Planetary gear train model comparison, by Ambarisha & Parker [2]
ep

Many kinds of models have been considered. Lumped Parameters Models (LPM) have been used since the
first researches: they consist in modelling a given system through concentrated masses and springs (see the
LPM model considered by Ambarisha & Parker [2] in Fig. 2) and are used to study the dynamic behaviors
Pr

of various systems, from a pairs of cylindrical gears to multi-stages reduction systems. They can work either
with few Degrees of Freedoms (DoF) to generate fast results, or with complex sets of rotational and

3
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

translational DoF for a refined precision. Translational DoF are often used to take into account the bearing

ed
stiffness influences, through constant spring values, while rotational DoF are actioned through the mesh
stiffness. This parameter can either be consider constant [3], [4] (excitation is introduced through an
external force as the static transmission error) or time-varying [5], [6], to namely take into account the

iew
evolving number of contacting teeth or the shocks. Widely used by researchers, LPM approach is less
represented in the industry where multi-body industrial codes (for instance LMS Siemens® or MSC
Adams®) are more often used: such solutions enable to consider large systems by assuming rigid bodies
and introducing stiffness coefficient in-between. The great simplifications of the contacts in those
approaches decreases their ability to predict accurately the vibration resulting from the complex meshing

ev
process of the planetary gear trains.

Finite Elements Method (FEM) has been later used to push forward the fidelity of the representation, by
considering deformable bodies or more detailed contact conditions, see the FEM models of planetary gear

r
trains proposed by Kahraman & Vijayakar in [7], or the model illustrated in Fig. 2 by Ambarisha & Parker
[2]. FEM models enable to better understand the teeth pressure patterns or root stresses. Quasi-static analysis

er
can be considered with FEM models but dynamic studies require huge CPU resources, making them less
appropriate than LPM approaches. More recently, hybrid approaches have been introduced by coupling
pe
global LPM and local FEM models. The goal was to enhance the representation fidelity of local influences
(mesh contact or gears bodies deformation) in dynamic models: Bettaieb & al. [8] and latter Guilbert [9]
use hybrid approaches to consider thin-webbed gears flexibility in their dynamic behaviors.
ot
tn
rin
ep

Beam elements Ring model Deformed shape of the train

Fig. 3 Model including the ring flexibility, by Abousleiman & Velex [10]
Pr

A large variety of models has therefore already been proposed across the years. Nevertheless, most results
of the proposed works are expressed within the rotating reference associated to the carrier rotation. As an

4
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

illustration, frequency and temporal results of the study conducted by Chaari & al. [11] (with a LPM dynamic

ed
model) are illustrated in Fig. 4 : those results were obtained in the rotating reference by treating one of the
carrier DoF.

iew
r ev
Fig. 4 Frequency domain and time domain results, by Chaari & al. [11]

er
In Fig. 5, teeth contacting load were derived in [10]: they also correspond to carrier-reference results as
planet-gears axes are rotating with the carrier.
pe
ot
tn

Fig. 5 Tooth load result for different ring geometries, by Abousleiman & Velex [10]

Some fix-reference results are often proposed as planet gear trains mode shapes [2], [12], [13]. But none of
rin

the considered results is given from an outer fix-reference accelerometer point of view, despite the fact that
accelerometers positioned outside the gearbox housing constitute a recurrent way of monitoring in
experimental studies [1].
ep

2.2. Frequency domain theoretical results in fix-reference

Some literature references focused on the theoretical harmonics that should be observed on a frequency
spectrum obtained from the signal measured by a fix-reference accelerometer, outside a planetary gear train
Pr

ring: it is namely the case of Inalpolat & Kahraman or Hong & al. [14], [15]. Their studies simulate

5
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

theoretical signals as Fourier series. Planet-ring contacting efforts are expressed as series periodic of the

ed
mesh frequency 𝑓𝑚 (𝜔𝑚 for the corresponding rotational frequency):

𝜔𝑚 = 𝑍𝑎(𝜔𝑎 + 𝜔𝑐) (1)

with 𝑍𝑎 the number of teeth of the ring gear, 𝜔𝑎 the ring rotational frequency (0 for fixed ring) and 𝜔𝑐 the

iew
carrier frequency. A “window function” is introduced as a Hanning window 𝑤(𝑡) to filter the influence of
each planet gear on the sensor signal: planet gears which are close enough from the accelerometer will
influence the resulting signal (higher transmission is supposed at the level of the sensor) while gears far
from the measurement point will not be taken into account.

r ev
er
pe
Fig. 6 Window function represented by Hong & al. [15]

The window function, illustrated by Hong & al. [15] in Fig. 6 is periodic of the carrier frequency 𝜔𝑐.
Accelerometer signal is obtained by summing each planet gear influence, which is the product of its
contacting effort by the window function. Author’s calculations lead to formula for the predicted
ot

accelerometer signal 𝑥(𝑡), which can be summarized as follow (𝐴𝑛 & 𝐵𝑚 are Fourier series coefficients, 𝑁𝑝
the number of planets, 𝑅𝑐 the reminder of the Euclidian division of 𝑍𝑎 by 𝑁𝑝, and 𝑛 & 𝑚 are positive
tn

integers):
𝑁 𝑀 𝑁𝑝 ± 𝑖2𝜋(𝑘 ‒ 1)(𝑛𝑅𝑐 ± 𝑚) (2)
𝑥(𝑡) = ∑ ∑𝐴 𝐵 𝑛 𝑚𝑒
± 𝑖(𝑛𝑍𝑎 ± 𝑚).𝜔𝑐.𝑡 + Φ
× ∑𝑒 𝑁𝑝

𝑛 = 1𝑚 = 1 𝑘=1
rin

The right hand-side of this product goes to zero for every {𝑛, 𝑚} combination that does not fulfill the
condition:
𝑛𝑅𝑐 ± 𝑚 = 𝑘 × 𝑁𝑝 (3)
ep

Consequently, the only harmonics that will emerge from the mesh process in the frequency spectrum will
be the harmonics that will be multiple of the number of planets. To illustrate this conclusion, a fix-reference
frequency spectrum is given for a gear train composed of 4 equally-spaced planet gears and 99 ring teeth
Pr

in Fig. 7 [15].

6
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

ed
iew
ev
Fig. 7 Theoretical spectrum from a fix-reference accelerometer point-of-view by Hong & al. [15] (𝑁𝑝 = 4, 𝑍𝑎 = 99)

All the emerging mesh-harmonics are multiples of 𝑵𝒑 and the principal mesh-harmonic (𝟗𝟗𝝎𝒄) almost vanishes.

r
2.3. Literature synthesis : Dynamic model overview and Reference results

er
This paper is focusing on the transfer function (that would represent a more advanced model of the
theoretical “window function” introduced by Inalpolat & Kahraman or Hong & al. [14], [15]) to be applied
to rotating-reference results coming from planetary-gear trains models. Models that should be concerned
pe
by such an approach are the ones considering rigid rings, as the transfer function will be built using the
flexibility of the housing components, including the ring.
ot
tn
rin
ep
Pr

Fig. 8 LPM Dynamic model of planetary gear train

7
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

Such a model is illustrated in Fig. 8, following the philosophy of the models introduced in [2], [11], [16]:

ed
it consists in a LPM model using rotational and translational DoF for the planet and sun gears and for the
planet gear carrier. The transfer function must be applied between the moving-reference planet-ring
contacting efforts and the accelerometer outside the ring. Base frequency results of the model are displayed

iew
on Fig. 9.

r ev
er
pe
Fig. 9 Reference results of the gear model: no transfer function
ot
tn
rin
ep

Fig. 10 Reference results: MGB bench tests results

The spectrum obtained by Fast Fourier Transform (FFT) on planet efforts in rotating reference is first
shown: the three first principal mesh harmonics are obtained (𝑓𝑚 = 𝑍𝑎.𝑓𝑐). Then, the spectrum of the fix
Pr

reference signal, generated by summing the influences of all the planets simultaneously and without transfer
function, is shown on the right side of Fig. 9. Antagonist influences of the gears all around the ring cancel

8
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

all the main mesh harmonics: only a residual signal can be observed (with negligible amplitudes with

ed
respect to the first harmonic of the rotating reference spectrum).

Frequency results obtained by measuring a real gearbox signal during bench testing are finally shown in
Fig. 10: its planetary gear stage is representative of the one modelled in Fig. 8 and used to generate Fig. 9

iew
results. As predicted in [14], [15], main mesh harmonics have vanished at the benefits of harmonic pairs
𝑓𝑚𝑖 , 𝑓𝑚𝑠 that are multiples of the number of planet gears. All the other frequencies observed in the spectrum
are coming from dynamic effects that are not representative of the meshing process, or representative of
lower transmission stages. Those results will be used as reference results for the next parts of this paper.

ev
3. Quasi-static and dynamic transfer functions establishment

Three approaches have been used to derive the transfer functions to be applied to the LPM dynamic model
of planetary gear train: a first quasi-static approach that uses FEM model of the ring; a FEM model to get

r
dynamic transfer functions; and finally impact- and shake-testing on a real system.

3.1. Quasi-static approach

3.1.1. Method er
pe
ot
tn
rin

Fig. 11 FEM model of the ring for the quasi-transfer approach of the transfer function
ep

Quasi-static approach concept is based on the observations from [14], [15] where a “window function” is
used to modulate the rotating-reference signal when transposing towards the fix-reference one. It states that
the influence of a single planet gear on the sensor at a given time step will not be the same depending on
Pr

its position on the ring and more especially depending on its distance with respect to the measure point. To

9
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

determine the shape of this quasi-static function, a FEM model of the ring based on a hexahedral mesh is

ed
established, see Fig. 11.

Ring screw holes are used to fix the ring using Rigid Body Elements (RBE): RBE and are either clamped
or linked to concentrated stiffness terms to include the influence of the flexibility of the housings in the

iew
behavior, see Fig. 11. A constant nominal effort is applied on an excitation area which consists in two
consecutives teeth surfaces to simulate the contact of planet gear on the ring. Effort is applied with an angle
that correspond to the real-contact pressure angle. Excitation area is moved all around the ring to test all
the contacting positions, while a single RBE is used as a fix measure point to extract the radial and the
tangential resulting efforts at the position of the sensor.

ev
Quasi-static transfer functions 𝐻𝑛 and 𝐻𝑡 are built in radial and tangential directions by calculating the ratio
between applied effort 𝐹 at the excitation position 𝜃 and the measured radial or tangential resulting efforts

r
(𝑅𝑛 or 𝑅𝑡):

{
𝐻𝑛(𝜃) = 𝑅𝑛/𝐹(𝜃) (4)

3.1.2. Results – 2D transfer functions


er
𝐻𝑡(𝜃) = 𝑅𝑡/𝐹(𝜃)
pe
Transfer functions are expressed in 2D giving normalized transmission coefficients. Those are calculated
in absolute value following Eq. (4) for excitation angular positions all around the ring, see Fig. 12. Measure
point (sensor) is set to be the 0deg reference angle in the figure.
ot
tn
rin
ep

Fig. 12 Quasi-static transfer functions


Pr

Results are given for the two boundary conditions (clamped RBE and concentrated mesh stiffness). The
global tendency of the Hanning window used in [14], [15] can be observed in both case: effort transmission

10
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

is cancelled out of the angular area ± 60° with the clamped RBE, while the transmission greatly reduces

ed
out of the angular area ± 80° with the concentrated stiffness. Two maxima are observed on both sides of
the accelerometer while a local transmission minimum stands at the direct proximity of the sensor (-9deg).
Those two pics can be explained by the pressure angle used for the contacting efforts. Bumps observed with

iew
the concentrated stiffness are the result of the efforts passing the screws of the ring.

3.2. Dynamic approach – FEM modeling

3.2.1. Method

FEM modeling is the first approach considered in the research process for the dynamic Transfer Function,

ev
which should take into account the influence of the frequency level on the modulation. Modelling has been
performed using the SAMCEF® FEM code. The modelled system is illustrated in Fig. 13. It is composed
of the main housing components of the MGB:

r
- The ring, internal common gear of the two stages of planetary gear train also assuming a structural
role into the assembly by joining the upper conical housing and the lower housing through bolt

-
links; er
The lower housing that contains all the lower transmission stages composed of bevel gears and
pe
cylindrical gears. This housing components is linked to the engine entries and to the rear output
with the Tail rotor flange, while a lower flange is connecting the MGB to a suspension known as
the “Barbeque Plate”, not considered in this approached;
- The conical housing, upper part of the MGB carrying the planetary gear trains through the carrier
bearing and supporting the MGB by conning it to the aircraft structure through the MGB struts, not
ot

represented and not taken into account in this study.

The three housing components are connected through RBE connections at the ring screw holes locations.
tn

Ring gear is meshed as for the quasi-static study, while conical and lower housings are composed of
tetrahedral elements with crude mesh on the walls and refined mesh at the connections and sensors levels.
Two different sets of Boundary Conditions are used. The first one (BC1 in Fig. 11) uses rigid connections
rin

at all the opened flanges locations (Engine flanges; Lower suspension flange; Rear flange and Upper cover
flange) with a clamped lower flange. This condition tends to limit the flange modes and it stiffens those
locations, which are connected to additional housing components in the real system. The second conditions
simulates a suspended MGB by only applying a clamped rigid connection at the Upper cover level and by
ep

adding the gravity influence in the model. A concentrated mass can also be fixed at the carrier bearing
location inside the conical housing to take into account the inertial properties of the MGB gears. This
condition is also used to represent the system later involved in the testing.
Pr

11
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

ed
iew
ev
Fig. 13 Housing components and boundary conditions in MEF assembly

r
Two calculation steps are involved in the calculation process of the transfer function. A modal analysis is

er
first performed (the DYNAM module has for instance been used in the SAMCEF code) to obtain an
orthogonal modal base and to determine the modal mass and stiffness matrices 𝑀𝑚𝑜𝑑 and 𝐾𝑚𝑜𝑑 that
correspond to the system and its boundary conditions. Those matrices are composed of diagonal terms that
pe
fulfill the following conditions:

{
𝓂𝑘 = 𝑋𝑘𝑇𝑀 𝑋𝑘 (5)

𝓀𝑘 = 𝑋𝑘𝑇𝐾 𝑋𝑘
ot
tn
rin
ep

Fig. 14 Reciprocity hypothesis illustration

A harmonic analysis is next performed (using the REPDYN module in SAMCEF) based on the previously
Pr

calculated modal base. In this step, a harmonic excitation is applied at the accelerometer level on a ring-
gear tooth (the effort is applied following the direction and the teeth load repartition illustrated in Fig. 11).

12
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

A reciprocity hypothesis is made, as illustrated in Fig. 14. Using this hypothesis, a single excitation location

ed
is used while the measuring point is moving, while in the actual system the excitation is moving and the
sensor is fixed. This hypothesis accelerates the calculation process of the transfer function and should be
validated later on by testing. The frequency of the excitation is swept on a frequency range covering the

iew
studied mesh harmonics that will be considered in the future accelerometer spectra. At each point of
measurement, a 2D transfer function (or Frequency Response Function - FRF) stating on the vibrational
amplitude on the given frequency range is extracted.

At the end of the process, all the FRF are assembled into a 3D transfer function with as first dimension the
frequency, as second dimension the angular position of the measurement on the ring, and as a third

ev
dimension the complex value of the FRF. In this model, a global damping coefficient method based on the
Raleigh hypothesis [17] is applied.

r
3.2.2. Results – 3D transfer functions

3D transfer functions obtained from both the clamped and the free / suspended FEM models are presented
in the Fig. 15.
er
pe
ot
tn
rin
ep
Pr

Fig. 15 3D Transfer Functions obtained using the FEM model (a) Clamped model (b) Free Suspended model

13
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

The amplitude of each results (color bars) has been normalized by the highest amplitude of the clamped

ed
model radial transfer function. The accelerometer is located at the 0deg position on the x-axis of the results.

As a first observation, the radial results presents more energy than the tangential ones and the fully clamped
model more than the suspended one. Beyond those energetic differences, both models provide very close

iew
results with similar patterns (dotted line in Fig. 15) meaning that the dynamic modulation of the gears
signals will more depends on the ring close environment than the further boundary conditions. The
identified pattern, easily observed in the radial results, represents the increasing numbers of lobes in the
mode shape with the increasing frequency while the amplitude concentrates near the sensor in the lower
frequencies. Two main angular positions can next be identified on both sides of the sensor’s location

ev
(identified by the arrows in Fig. 15) with the highest amplitudes, while a low-amplitude position stands
close to the 0deg angle: this result matches the observation from the quasi-static approach (see Fig. 12). 2D
FRF extracted from the two identified positions are presented in Fig. 16.

r
er
pe
ot
tn

Fig. 16 Extracted 2D FRF at the principal angles location


rin

Small differences can still be observed between the two models looking at those more precise curves,
namely the amplitude sharing but also some frequency positioning: the second highest-amplitude mode
from the suspended model is 6% lower in frequency than for the clamped one. Nevertheless, many common
ep

markers can once again be noticed. The normalized amplitude is increasing with the frequency and high-
amplitude modes are observed on both sides of the third mesh harmonic. Some low frequency pics can still
be observed below 1 × 𝑓𝑚 with a uniform repartition around the ring, which may be the result of a modeling
artefact, or of a lower-housing mode. Finally, one mode is present at the second mesh harmonics location:
Pr

14
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

this mode is likely to enhance the frequency content close to the 2nd harmonic in the spectrum that will be

ed
predicted by the gear train model.

The FEM approach enables to get a first approximation of the transfer function but the performed
simplifications and hypothesis are likely to decrease the fidelity of the results. It is nevertheless highly

iew
upgradable and the method could easily be applied to models that are more complex or to other
architectures.

3.3. Dynamic approach – Real system testing

3.3.1. Method

ev
An experimental approach has been used to characterize the real system in its entirety, see Fig. 17. All the
housing components studied during the FEM approach are present, as well as additional ones further from
the ring at the rear output and at the entries levels. An upper cover (dedicated to the storage of the system)

r
through which the entire MGB is suspended by a crane closes conical housing. The MGB comprises all the
transmission components, namely the gears, bearings, pumps and shafts.

er
pe
ot
tn
rin

Fig. 17 Tested System: suspended MGB

Accelerometers are finally added to the system to monitor its vibrations: accelerometers are glued with a
regular spacing all around the ring, see Fig. 18. Two of them have been selected for presenting the transfer
ep

functions results in this paper: they are identified as G01 and G06. An additional sensor is mounted on a
fix support at the ring-conical housing junction: M02. This sensor is used for the system monitoring in real
condition operation.
Pr

15
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

ed
iew
r ev
Fig. 18 Sensors positioning around the ring gear

er
Two methods are considered for the system excitation, see Fig. 19.
pe
ot
tn
rin

Fig. 19 Impact tests versus shake tests

First method is the impact testing using a metallic-tip impact hammer. Impacts are performed near each
sensor’s location on the ring web as shown in Fig. 19. The impact corresponds to a Dirac that will excite
ep

all the frequencies simultaneously. Each testing point is averaged using five impacts at the considered
position. When the impact position is not accessible, impact is performed above the sensor on the conical
housing flange. The second method is the shake-test: it involves a shaker that will apply a harmonic
excitation with a constant effort value. The frequency is swept on the studied frequency range (starting
Pr

frequency is taken with a reasonable margin above 0Hz not to damage the shaker) and effort is check using

16
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

an effort sensor at the tip of the threaded shaft that links the shaker to the housing. The shaker is fixed on

ed
the mounted-accelerometer’s support.

2D FRF are finally extracted and assembled into the 3D transfer functions following two methods:

- Excitations to one measure point : FRF are extracted at a single measure point location and for each

iew
excitation point to assemble the 3D transfer function;
- Measures to one excitation point: For a single excitation point, FRF are extracted at each measure
point’s level to assemble the 3D transfer function.

Following the reciprocity hypothesis (detailed in part 3.2.1) the results produces by both methods should

ev
be the same. This hypothesis will be verified when analyzing the results.

The experimental approach should offer a high fidelity, but some gaps remain with the real-life operating
MGB. The influence of the MGB struts and of the Barbecue Plate that normally support the system cannot

r
be considered. In addition, no power is transmitted through the MGB during the testing: no torque is applied
to the gears and the ring is therefore not constrained. Finally, excitation is applied outside the ring and is

er
perpendicular to its surface, while real-life excitation is applied inside the ring on its teeth following the
pressure angle direction.
pe
3.3.2. Results – 3D transfer functions
ot
tn
rin

Fig. 20 3D Transfer function assembling methods comparisons


ep

Results are presented the same way as for the model results. In Fig. 20 are compared the results from the
two assembling methods and from the G01 location point of view: for the Excitation to measurement
method, measures are done by the G01 accelerometer; for the Measures to excitation method, one excitation
Pr

is applied near the sensor.

17
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

Some small differences can be observed between the two results: greater amplitude are namely present at

ed
the -20deg location for the first method. Nevertheless, both remain highly close with the higher energy
concentrated at the sensor’s level (0deg); very low content far from the sensor below the second mesh
harmonic; one pic at the second mesh harmonic level and at the 0deg location and one pic every 60deg at

iew
3.5 × 𝑓𝑚. The reciprocity hypothesis involved for the modeling and for the shake-tests is therefore
validated. In the followings paragraphs, no distinction will be made between both assembling methods.

Radial and tangential transfer functions from the front and rear sensors (G01 & G06 respectively) obtained
by the impact tests are presented in the Fig. 21.

r ev
er
pe
ot
tn
rin

Fig. 21 Impact test transfer function results (a) G01 results (b) G06 results
ep

Front and rear results testify about the anisotropy of the MGB system. Radial results almost vanish at the
G06 location compared with the ones from the G01 sensor, whereas tangential result are more energetic in
the G06 case. Pics are also different between the two radial results: the energy is well concentrated around
Pr

the sensor on the front side of the MGB and more spread when measure is performed on the rear side. This
observation therefore shows the importance of the monitoring location on the MGB, as some locations

18
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

(front side in this study case) will enhance the vibrations, especially at some frequencies of interest like the

ed
second mesh harmonic 2 × 𝑓𝑚.

We can now compare the G01 results with the one of the FEM approach (see Fig. 15), where measurement
was performed at the front of the MGB. As a first observation, we notice the lower precision on the angular

iew
axis compared with the FEM results. This precision is limited by the number of sensors positioned all
around the ring (approximatively at one accelerometer every 20deg in our case). On the contrary, test results
benefits from a higher precision on the frequency axis (0.3Hz), as matching such a precision with a FEM
models would require a high computation time. Looking at the shape of the transfer functions, we can still
observe the characteristic pattern (dotted line in the figure) narrowing towards the low frequencies. As for

ev
the FEM results, the amplitude is concentrated near the accelerometer at the lower frequencies (2 × 𝑓𝑚 and
below) while it is spread all around the ring above 3 × 𝑓𝑚. Some pics are emerging at the second mesh
harmonic location and one or two ring angles come out with the higher amplitudes, as for the model. The

r
model was hence able to catch the general aspect of the transfer function despite the simplifications and the
hypothesis in presence. Differences are still observed between the two positions. Equally spaced pics (every

er
60deg) above the third mesh harmonics are more visible on the model results. The angular positions of the
high-amplitude also changes: the 0deg angle is the most sensitive one in the radial test results; two angles
pe
around -30deg and +30deg in the tangential test results; whereas two positions are observed at -34deg and
+16deg in the radial results and one at -9deg in the tangential results of the FEM model).
ot
tn
rin
ep

Fig. 22 Results comparison between Impact and Shake tests at the M02 accelerometer location

Results of the impact tests and shake tests at the mounted M02 sensor’s location can finally be compared,
see Fig. 22. This accelerometer is located at the front-left side of the MGB whereas the G01 sensor, used
Pr

to extract the results given in Fig. 20 and Fig. 21, was located at the front-right side. The proximity of the

19
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

two sensors explains a global similarity between the impact test results measured by both sensors: on main

ed
angular location around 0deg, on pic at { 2 × 𝑓𝑚, 0𝑑𝑒𝑔}, and more content above 3 × 𝑓𝑚. Less amplitude
is nevertheless observed on the M02 results. This can either be explained by different mounting / positioning
strategies between the two sensors (G01 was glued on the ring web, while M01 was screwed on a support

iew
fixed at the junction between the ring and the conical housing), or by the small difference of location on the
ring. Shake test enhance the energy content of the transfer function without modifying its shape, and the
results get closer to the one of the impact test at the G01 location.

ev
4. Rotating-reference to Fix-reference results using the transfer functions

Three approaches have been proposed to obtain the 2D quasi-static and the 3D dynamic transfer functions.
The particular shapes of each already give an insight of the structural dynamic behavior of the gearbox, by

r
namely highlighting its anisotropy and showing the preferred relevancy of certain monitoring positions.
These functions must finally be applied to the results of the planetary gear train model to transform them

er
from the carrier rotating reference to the accelerometer fix reference.

4.1. Rotating reference results


pe
The gear model used as an example to produce the rotating reference results was illustrated in Fig. 8 in
section 2.3. The results calculating process of such a model is represented in Fig. 23. The model is based
on the classical dynamic equation, see Eq. (6). The mass, stiffness and damping matrices (𝑀, 𝐶 & 𝐾) are
determined and assembled based on the system geometry and its material properties, and the non-
ot

conservative force vector 𝐹 depends on the external loading. Stiffness and damping matrices are step-
dependent and will evolve with the gears mesh-stiffness.
𝑀𝑋(𝑡) + 𝐶(𝑡)𝑋(𝑡) + 𝐾(𝑡)𝑋(𝑡) = 𝐹(𝑡) (6)
tn

The Ordinary Differential Equation (ODE) system (see the algorithm of Fig. 23) involves the state vector 𝑈
(𝑡) and is solved gradually by using an explicit solver scheme. From this solving loop are produced raw
and non-equally time-spaced results: the state matrix 𝑈(𝑡) and its corresponding time vector 𝑇(𝑡). A re-
rin

sampling step enables to produce equally time-spaced results. The data contained in the state matrix (or in
the resampled state matrix 𝑈𝑟𝑒 ) are composed of all the time-evolving DoF of the model, given as
translational or rotational displacements, and of their corresponding speeds. In Fig. 23 are represented the
ep

time-evolving planet angle and the corresponding angular speed, which are direct results of the model and
which are expressed in the rotating reference. A final step of the modeling process consists in calculating
the contacting efforts 𝐹𝑝 ‒𝑖 𝑟(𝑡) between each planet gears 𝑖 and the ring. To do so, the transmission error
𝛿𝑝 ‒𝑖 𝑟 consisting in the compression following the line of action of the planet – ring mesh (direction of the
Pr

𝑘𝑝 ‒𝑖 𝑟 stiffness term in Fig. 8) must be multiplied by the mesh stiffness 𝑘𝑝 ‒𝑖 𝑟, following the formula:

20
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

𝐹𝑝 ‒𝑖 𝑟 = 𝑘𝑝 ‒𝑖 𝑟 × 𝛿𝑝 ‒𝑖 𝑟

ed
With
(7)
𝛿𝑝 ‒𝑖 𝑟 = cos (𝛽) × (𝑢𝑝𝑖 sin (𝛼) ‒ 𝑣𝑝𝑖 cos (𝛼) + 𝑅𝑝𝑖 𝜃𝑝𝑖 + 𝑒𝑠𝑐(𝑡))

iew
where 𝛽 is the helix angle, 𝑢𝑝𝑖 , 𝑣𝑝𝑖 & 𝜃𝑝𝑖 are the translational and rotational DoF of the 𝑖𝑡ℎ planet gear, 𝑅𝑝 the
base radius of the planet gear, 𝛼 the pressure angle and 𝑒𝑠𝑐(𝑡) the possible eccentricity defect.

r ev
er
pe
ot
tn

Fig. 23 Gear train model algorithm and temporal rotating-reference results

The time-varying contacting force calculated in the rotating reference is illustrated in Fig. 23. This result
represents the input for methodology presented in this section, as it will be coupled with the transfer
rin

functions to calculate the fix-reference results.

4.2. Applying the Quasi-static transfer functions

4.2.1. Method
ep

Quasi-static transfer functions 𝐻𝑛 and 𝐻𝑡 are giving one transmission value per contacting angular position
on the ring. To produce the normal and tangential accelerometer signals 𝑆𝑖𝑔𝑛(𝑡), and 𝑆𝑖𝑔𝑡(𝑡), contacting
efforts 𝐹𝑝 ‒ 𝑟 of each planet gear 𝑖 must be multiplied by the value of the transfer functions that corresponds
Pr

to its current location 𝜃𝑖(𝑡):

21
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

{
𝑁𝑝𝑙𝑎𝑛𝑒𝑡

∑𝐹

ed
𝑆𝑖𝑔𝑛(𝑡) = 𝐹𝑎𝑐𝑐 𝑝 ‒ 𝑟(𝜃 (𝑡)) × 𝐻𝑛(𝜃 )
𝑖 𝑖 𝑖
𝑛 (𝑡) =
𝑖=1
𝑁𝑝𝑙𝑎𝑛𝑒𝑡 (8)
𝑆𝑖𝑔𝑡(𝑡) = 𝐹𝑎𝑐𝑐
𝑡 (𝑡) = ∑𝐹 𝑝 ‒ 𝑟(𝜃 (𝑡)) × 𝐻𝑡(𝜃 )
𝑖 𝑖 𝑖

𝑖=1

iew
Following the approaches of Inalpolat & Kahraman and Hong & al. [14], [15], a linearity hypothesis is
performed and the influence of each planet is summed to produce the complete signal.

4.2.2. Results – Frequency spectra

r ev
er
pe
ot
tn
rin
ep

Fig. 24 Frequency spectrum obtained after applying the quasi-static transfer functions (a) Clamped RBE Transfer
Function (b) Concentrated housing stiffness Transfer Function

The signal obtained by applying the quasi-static transfer functions is homogeneous to an effort. It is post-
Pr

treated using a FFT to produce the frequency domain results shown in the Fig. 24.

22
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

The results are given for the two boundary conditions: clamped screw holes or concentrated housing

ed
stiffness. Spectra are normalized using the value of the first mesh-frequency amplitude that was obtained
in Fig. 9. Each of these results is presenting the theoretical harmonic mesh pairs that were predicted by the
𝑖 𝑠 𝑖 𝑠 𝑖 𝑠
Fourier approach given in part 2.3: 𝑓𝑚1 & 𝑓𝑚1 , 𝑓𝑚2 & 𝑓𝑚2 and 𝑓𝑚3 & 𝑓𝑚3 . Also, one observes that the

iew
second mesh harmonic is of a low amplitude compared to the first and the third ones, which follows the
results of the model without transfer functions (see Fig. 9), as the quasi-static approaches does not depend
on the frequency. In the real-system results that were proposed in Fig. 9, the second mesh harmonic is of a
comparable amplitude to the first mesh harmonic.

Several differences can be observed between the results that were obtained using the two boundary

ev
conditions. First, there is a difference in the relative energy sharing between the inferior and the superior
𝑖 𝑠
mesh harmonic: 𝑓𝑚1 is greater than 𝑓𝑚1 using the clamped RBE, whereas the opposite is observed using
the concentrated mesh stiffness. In general, more energy is contained in the sidebands in the Fig. 24 (a)

r
than in Fig. 24 (b): concentrated stiffness of the second model are filtering the side-frequency in a better
way and more energy is concentrated in the main mesh harmonics. In addition, the high-amplitude low

er
frequencies are of a lower importance in this model. Very little amplitude was contained in the low
frequencies of the real-system spectrum, which makes the second model closer the actual physics. All the
pe
sidebands are multiples of the number of planet gears (when the frequency axis is normalized by the carrier
frequency), which follows the conclusion of the theoretical approach given in part 2.2.

4.3. Applying the dynamic transfer functions

4.3.1. Method
ot

Dynamic transfer functions 𝐻𝑛 and 𝐻𝑡 are 3D results giving one complex value per angular position and
frequency pairs {𝜃, 𝜔}. As the gear model effort results are temporal functions, a particular methodology
tn

must be considered to make them compatible with the frequency-domain transfer functions. This algorithm
is detailed in Fig. 25.

First, efforts of each planet gear calculated by the gear model must be transformed into the frequency
rin

domain by using FFT, giving the vectors 𝐹𝐹𝑇𝑝 ‒𝑖 𝑟. The transfer functions that were calculated over a
frequency range that was covering the three third mesh harmonics must then be extended over the frequency
range that is covered by 𝐹𝐹𝑇𝑝 ‒𝑖 𝑟: values that were not covered in the initial target range are set to zero, to
retain only the influence of the desired and studied frequencies. The actualized transfer functions are called
ep

𝐻𝑁 and 𝐻𝑇. 𝐹𝐹𝑇𝑝 ‒𝑖 𝑟, which is expressed as a function of the frequency Ω is repeated for each angular
position to fit the transfer functions dimensions, and then multiplied term by term with 𝐻, giving the same-
𝑖
dimension angular-frequency matrix 𝐹𝐹𝑇𝑝 ‒ 𝑟. By performing an inverse Fourier Transform (iFFT), the
Pr

𝑖
result is set back to the temporal domain: 𝐹𝑇𝑝 ‒ 𝑟. Values are extracted from this matrix depending on the

23
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

position of the planet 𝜃𝑖 at each time-step 𝑡, and then summed over each planet gear to assemble the fix-

ed
reference accelerometer temporal signals 𝑆𝑖𝑔𝑛 and 𝑆𝑖𝑔𝑡 in the radial and tangential directions.

iew
r ev
er
pe
ot

Fig. 25 Algorithm for applying the dynamic transfer functions to the rotating-reference temporal results of the gear
model
tn

4.3.2. Results – Frequency spectra

The signal obtained by applying the dynamic transfer function is homogeneous to an acceleration. This
section presents the spectra that were calculated after applying a FFT to the signals from coming from the
rin

modulation of the planet gear efforts by the various transfer functions from the FEM models, the impact
𝑖
and the shake tests. All the spectra are normalized by the amplitude of the first lower mesh harmonic 𝑓𝑚1
of the real system reference spectrum given in Fig. 10.
ep

First, spectra obtained by applying the transfer functions of the FEM models with the Clamped boundary
conditions and with the Free Suspended boundary condition are given in Fig. 26. As a first observation,
𝑖
results obtained by applying the transfer functions from both boundary conditions are extremely close: 𝑓𝑚3
Pr

is the only harmonic that has a higher amplitude in the Clamped version than in the Free Suspended version;
and the low amplitude domain of the Free Suspended tangential signal is a bit noisier than for the Clamped

24
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

version. This similarity follows the observations that were performed on the transfer functions from both

ed
models in Fig. 15, where there was no first-order difference in the functions between the two boundary
conditions, but the Clamped model had a slightly higher amplitude around the third mesh harmonic.

iew
r ev
er
pe
ot
tn

Fig. 26 Frequency spectrum obtained after applying the dynamic transfer functions obtained by FEM method (a)
Clamped model (b) Free Suspended model
rin

Next, those results can be compared with the ones that were obtained using the quasi-static transfer
functions (see Fig. 24). In a general fashion, results from Fig. 26 seem “cleaner” than the quasi-static ones,
ep

with far less side-bands. Low-frequency domain is free of harmonics and the general shape of the spectrum
𝑖 𝑠
gets closer to the reference results of the real system (see Fig. 10), with the three mesh pairs: 𝑓𝑚1 & 𝑓𝑚1
𝑖 𝑠 𝑖 𝑠 𝑠 𝑖 𝑖 𝑠 𝑖
with 𝑓𝑚1 of a higher amplitude than 𝑓𝑚1 ; 𝑓𝑚2 & 𝑓𝑚2 with 𝑓𝑚2 higher than 𝑓𝑚2 ; and 𝑓𝑚3 & 𝑓𝑚3 with 𝑓𝑚3
𝑠
higher than 𝑓𝑚3 . Some side-bands that are multiples of the number of planet gears are still present nearby
Pr

the third harmonics as for the reference results. Finally, the second mesh harmonics grows higher in term

25
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

of amplitude and get closer to the first harmonics, as in the real system results. This is the direct consequence

ed
of the high level of the second harmonic in the transfer function, see Fig. 15.

Spectra obtained by applying the transfer functions of the impact tests are presented in Fig. 27 for two

iew
monitoring positions: the G01 sensor on the front of the MGB and the G06 sensor on the back of the MGB.

r ev
er
pe
ot
tn
rin

Fig. 27 Frequency spectrum obtained after applying the dynamic transfer functions obtained by Impact testing (a)
G01 front sensor (b) G06 rear sensor
ep

Those results tend to be pretty close to the ones that were obtained using the transfer functions of the FEM
models: the three harmonics pairs are present with similar amplitudes ratios; spectra are even cleaner than
the FEM transfer functions ones; and the second mesh harmonic grows higher in term of amplitude.
Pr

Therefore, the simplifications that were performed on the models seem to be acceptable and the transfer
function seems to mainly rely on the structural components close to the ring-gear environment. Some low

26
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

amplitude pics are nevertheless still present in the low frequency area that is matching the real-system

ed
results, see Fig. 10. Greater amplitudes are obtained using the sensor at the front of the gearbox that follows
the conclusions performed on their respective transfer functions, see Fig. 21.

Finally, results of the impact test and the shake test are compared at the mounted sensor’s location in Fig.

iew
28.

r ev
er
pe
Fig. 28 Frequency spectrum obtained after applying the dynamic transfer functions at M02 mounted sensor location

The impact test result is very close to the one of the impact test at the G01 location, as both the sensors are
ot

close. The three harmonic pairs are once again observed in the shake-test results, but their relative
amplitudes are reversed compared with the Impact test. Some side bands emerge close to the first and the
second mesh harmonics, which was namely observed in the real system spectrum at the second harmonic
tn

location and some low-amplitude frequencies appear between the harmonics as for the real system. In a
general fashion, results see closer to the real system ones for the impact tests at the mesh-harmonic
locations, while side-bands and low-amplitude pics seem to be better represented by using the shake-tests.
rin

One hypothesis that could explain the better correspondence of the impact test over the shake test at the
mesh harmonics level would be that impacts are better representing the excitation caused by the gears.

5. Conclusions
ep

To transform the results of a planetary gear train dynamic model from the planet-gear carrier rotating-
reference to the fix-reference of an accelerometer mounted outside the ring gear, an approach based on
quasi-static and dynamic transfer functions have been proposed. Finite Element Method models composed
Pr

of the ring gear and of the main housing components with different boundary conditions have been
presented to obtain the 2D quasi-static and first 3D dynamic transfer-functions. Impact and shake-tests on

27
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

the real system have also been performed to obtain and validate the 3D transfer functions. Following some

ed
hypothesis of the literature, it has been shown that the closer the planet-gear is with respect to the sensor,
the greater is its influence on the signal. By considering the 3D dynamic functions, it has been shown that
some mesh harmonics should be amplified in the accelerometer signal due to the dynamic behavior of the

iew
housing. By monitoring vibrations at different locations on the gearbox during the testing, its anisotropy
has been demonstrated and it has been established that the front-side of the gearbox would constitute a
better location for monitoring the vibration during operation. Methodologies have finally been proposed to
apply the quasi-static and dynamic transfer functions to the planet gears efforts that are given in the temporal
domain, and accelerometer signals have been predicted. By analyzing the signals frequency spectrum, the

ev
approaches have been validated by obtaining the theoretical mesh harmonic pairs in all the results.
Nevertheless, dynamic approaches and more especially transfer functions obtained by the impact tests are
producing results that are the closest to the ones obtained on the real system: amplitude ratios between the

r
harmonics are correct, low-amplitude pics are present in the low frequencies and sidebands are obtained
close to the third harmonic.

er
Even though a real MGB have been put into testing to obtain the transfer functions, some differences with
the real system in operation can still be pointed at among which, the lack of torque that could pre-constrain
pe
the housing and therefore influence the transfer function shape and the vibration signal. As the FEM
modeling method has been validated by producing very close results with respect to the testing approach,
a future work should consist in introduction the torque in the model to check its influence on the results.

The proposed approach is a good method to be applied to Lumped-Parameters models with rigid ring
ot

hypothesis to post-process their results and introduce afterward the effect of the housing modulation. An
alternative way to treat the problem would be to extend the flexible ring approaches presented in the
literature to extract the vibrations at one of the ring nodes. Housing influence should then be taken into
tn

account as concentrated mass and stiffness components on the ring nodes.

6. References
rin

[1] C. Cooley and R. Parker, ‘A Review of Planetary and Epicyclic Gear Dynamics and Vibrations
Research’, Appl. Mech. Rev., vol. 66, p. 040804, Jul. 2014.
[2] V. K. Ambarisha and R. G. Parker, ‘Nonlinear dynamics of planetary gears using analytical and finite
ep

element models’, J. Sound Vib., vol. 302, no. 3, pp. 577–595, 2007.
[3] D. L. Seager, ‘Conditions for the Neutralization of Excitation by the Teeth in Epicyclic Gearing’, J.
Mech. Eng. Sci., vol. 17, no. 5, pp. 293–299, Oct. 1975.
Pr

[4] A. Kahraman, ‘Planetary Gear Train Dynamics’, J. Mech. Des., vol. 116, no. 3, pp. 713–720, Sep.
1994.

28
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886
R. Mignot-Pasty et al.

[5] G. Liu and R. G. Parker, ‘Dynamic Modeling and Analysis of Tooth Profile Modification for

ed
Multimesh Gear Vibration’, J. Mech. Des., vol. 130, no. 121402, Oct. 2008.
[6] R. G. Parker and X. Wu, ‘Parametric Instability of Planetary Gears Having Elastic Continuum Ring
Gears’, J. Vib. Acoust., vol. 134, no. 041011, May 2012.

iew
[7] A. Kahraman and S. Vijayakar, ‘Effect of Internal Gear Flexibility on the Quasi-Static Behavior of a
Planetary Gear Set’, J. Mech. Des., vol. 123, no. 3, pp. 408–415, 2000.
[8] M. Bettaïeb, P. Velex, and M. Ajmi, ‘A Static and Dynamic Model of Geared Transmissions by
Combining Substructures and Elastic Foundations—Applications to Thin-Rimmed Gears’, J. Mech.
Des. - J MECH Des., vol. 129, Feb. 2007.

ev
[9] B. Guilbert, P. Velex, D. Dureisseix, and P. Cutuli, ‘Modular hybrid models to simulate the static and
dynamic behaviour of high-speed thin-rimmed gears’, J. Sound Vib., vol. 438, pp. 353–380, 2019.
[10]V. Abousleiman and P. Velex, ‘A hybrid 3D finite element/lumped parameter model for quasi-static

r
and dynamic analyses of planetary/epicyclic gear sets’, Mech. Mach. Theory, vol. 41, no. 6, pp. 725–
748, Jun. 2006.

er
[11]F. Chaari, T. Fakhfakh, R. Hbaieb, J. Louati, and M. Haddar, ‘Influence of manufacturing errors on the
dynamic behavior of planetary gears’, Int. J. Adv. Manuf. Technol., vol. 27, no. 7, pp. 738–746, Jan.
2006.
pe
[12]R. G. Parker and X. Wu, ‘Vibration modes of planetary gears with unequally spaced planets and an
elastic ring gear’, J. Sound Vib., vol. 329, no. 11, pp. 2265–2275, May 2010.
[13]A. Zhang, J. Wei, D. Qin, and S. Hou, ‘Analytical coupling characterization of multi-stage planetary
gear free vibration considering flexible structure’, J. Vibroengineering, vol. 19, Oct. 2017.
ot

[14]M. Inalpolat and A. Kahraman, ‘A theoretical and experimental investigation of modulation sidebands
of planetary gear sets’, J. Sound Vib., vol. 323, no. 3, pp. 677–696, Jun. 2009.
[15]L. Hong, J. Singh Dhupia, and S. Sheng, ‘An explanation of frequency features enabling detection of
tn

faults in equally spaced planetary gearbox’, Mech. Mach. Theory, vol. 73, pp. 169–183, 2014.
[16]A. Saada and P. Velex, ‘An Extended Model for the Analysis of the Dynamic Behavior of Planetary
Trains’, J. Mech. Des., vol. 117, no. 2A, pp. 241–247, Jun. 1995.
rin

[17]R. Crambuer, B. Richard, N. Ile, and F. Ragueneau, ‘Experimental characterization and modeling of
energy dissipation in reinforced concrete beams subjected to cyclic loading’, Eng. Struct., vol. 56, pp.
919–934, Nov. 2013.
ep
Pr

29
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4252886

You might also like