Mechanical Systems and Signal Processing 92 (2017) 213–225

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Two methods for modeling vibrations of planetary gearboxes

including faults: Comparison and validation
J. Parra, Cristián Molina Vicuña ⇑
Laboratorio de Vibraciones Mecánicas, Universidad de Concepción, Edmundo Larenas 219, Concepción, Chile

a r t i c l e i n f o a b s t r a c t

Article history: Planetary gearboxes are important components of many industrial applications. Vibration
Received 27 October 2016 analysis can increase their lifetime and prevent expensive repair and safety concerns.
Received in revised form 6 January 2017 However, an effective analysis is only possible if the vibration features of planetary gear-
Accepted 22 January 2017
boxes are properly understood. In this paper, models are used to study the frequency con-
Available online 1 February 2017
tent of planetary gearbox vibrations under non-fault and different fault conditions. Two
different models are considered: phenomenological model, which is an analytical-
Keywords:
mathematical formulation based on observation, and lumped-parameter model, which is
Planetary gearbox
Vibration model
based on the solution of the equations of motion of the system. Results of both models
Local tooth fault are not directly comparable, because the phenomenological model provides the vibration
on a fixed radial direction, such as the measurements of the vibration sensor mounted
on the outer part of the ring gear. On the other hand, the lumped-parameter model pro-
vides the vibrations on the basis of a rotating reference frame fixed to the carrier. To over-
come this situation, a function to decompose the lumped-parameter model solutions to a
fixed reference frame is presented. Finally, comparisons of results from both model per-
spectives and experimental measurements are presented.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction

Planetary Gear sets (PG) are widely used because they are compact and have large torque-to-weight ratio. They are found
in many applications such as helicopter transmissions, lifting cranes, wind turbines, and mining excavators. It is well known
that vibration analysis has the potential to prevent expensive repair, downtimes and safety concerns of machines. However
its application in PG is not straightforward due to the multiple movable meshing pairs that simultaneously take place inside
the unit. These situations lead to a variable transmission path between the gear pairs and the transducer, which difficult the
understanding of the vibration behavior of the PG. The vibrations of PG have been studied using two different models: First,
there is the phenomenological model, where the vibration contributions resulting from the different meshing processes are
directly modeled as periodic functions considering the system kinematics. That is, it is an analytical-mathematical model.
Then, there is the lumped-parameter model, or dynamic model, which is based in the numeric resolution of the equations
of motion of the system.
The study of phenomenological model started three decades ago with the work of McFadden and Smith [1], who con-
tributed to the understanding of the difference between the dominant frequency and gear meshing frequency. They also
used the model to explain the asymmetry of the spectrum measured by a transducer mounted on the outside of the ring gear

⇑ Corresponding author.
E-mail address: crimolin@udec.cl (C.M. Vicuña).

http://dx.doi.org/10.1016/j.ymssp.2017.01.032
0888-3270/Ó 2017 Elsevier Ltd. All rights reserved.
214 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225

(Fig. 1). Years after, McNames [2] used the Fourier series to complement the work of McFadden and Smith. Parker and Lin [3]
studied the vibration phases of each gear pair inside a PG. Inalpolat [4] and Vicuña [5] generalized the previous work by
expanding the analysis to different geometries of PG. Both works took the amplitude modulation effect (AM) caused by
the carrier rotation into account, but did not consider any frequency modulation effect caused by meshing stiffness variation
or gear damage. The latter was included by Feng in the analysis of his work [6]. More recently, Hong [7] used Fourier series
analysis to explain the distinct sideband patterns in healthy and faulty PG. All studies mentioned propose models that bring
solutions referred to the measurements that the sensor would take placed as in Fig. 1.
The first studies of the lumped-parameter model started as a method to determine the natural frequencies and vibration
modes of PG. Kahraman [8] presented a torsional model for this purpose. Then Lin [9] worked in modal analysis on an equa-
tion of motion with multiple degrees of freedom. Chaari et al. [10] expanded the previous work by studying the influence of
manufacturing errors. Additionally, there are works on other research topics like load sharing among planet gears [11,12],
nonlinear dynamics and tooth faults [13–17]. Among these we highlight the work of Chaari et al. [18], where changes in
gearmesh stiffness due to tooth spalling and breakage were studied, although not in PG.
Unlike the fixed transducer scheme of Fig. 1, generally all lumped-parameter model solutions are referred to a non-
inertial reference frame, fixed to and rotating with the carrier. In the case of comparing the solutions with experimental mea-
surements, it is desirable to know the vibration in the inertial basis as seen by a ground-based observer [19]. In brief, solu-
tions of lumped-parameter model are not directly comparable with results from phenomenological model and real
vibrations measured on the ring gear. Other works include the lumped-parameter model and present their solutions as rep-
resenting the fixed transducer response, but no explanation of the method used to obtain the solutions are provided [7,20].
Other authors propose different methods to decompose the rotatory reference frame solutions into a fixed reference frame
like the transducer measurements, but they make incorrect assumptions: Inalpolat and Kahraman [21] expressed the sensor
measurements in function of the dynamic forces of the planets-sun and planets-ring without considering the difference
between the direction of their lines of action. Torregrosa and Vicuña [22] proposed a different function, but missed the fact
that the direction of the lines of action of the dynamic forces change with the rotation of the shafts. Liang et al. [16] propose
that the sensor measurements are obtained by the summation of weighted vibration of each planet gear, which doesn’t con-
sider the vibrations of other PG components like ring, sun, and carrier. Finally, Karray et al. [23] include a transmission error
function in the line of action, to explain the modulation sidebands pattern of the vibrations, thus incorrectly including an
artificial modulation of the gearmesh forces.
This study examines the frequency content of an equally spaced planetary gear set with the use of the lumped-parameter
model and phenomenological model, in both healthy and faulty condition. Faults considered are local faults on sun gear, pla-
net gear and ring gear. To compare the lumped-parameter model results with the phenomenological model results and
experimental measurements (to account for additional frequency components related to the local faults), we propose a
method that provides a solution to the decomposition of rotatory frame to fixed reference frame vibrations.

2. Planetary gearbox vibration models and frame decomposition

In this section, vibration models of the PG used in this study are presented. The configuration and measurement arrange-
ment considered is shown in Fig. 1. According to the classification suggested in [4], the PG pertains to group A, i.e., with N ¼ 3

Fig. 1. Measurement arrangement.

 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225 215

equally spaced planet gears at the angular positions wi (where i ¼ 1; 2; . . . ; N and w1 ¼ 0), and in-phase gear meshes. Tooth
number of gears are Z s ¼ 18 (sun), Z p ¼ 26 (planet), and Z r ¼ 72 (ring).

2.1. Phenomenological model

This model is presented in detail in [5], and describes directly the measurements as a sensor according to the arrange-
ment of Fig. 1 would take. It models the vibrations of all gear pairs as periodic functions with fundamental frequency equal
to the gear mesh frequency f g ¼ Z r f c (where f c is the carrier frequency). Fig. 2 presents a simulation example of a PG con-
sidering planet-ring gear meshing only. Individual planet-ring vibrations v ri ðtÞ with the corresponding phase difference
wi =ð2pf c Þ can be seen in Fig. 2a. Vibration contributions of each planet-ring gear meshing to the transducer measurements,
including the amplitude modulation effect ari ðtÞ produced by the variable transmission path between the fixed sensor and
the planets are presented in Fig. 2b. The total sensor measurement yr ðtÞ is shown in Fig. 2c, and its spectrum, in Fig. 2d. Note
that the model qualitative predicts the vibration spectrum of the PG. For instance, in this example it shows that vibration
measured in the PG would produce a spectrum with a component at the gear mesh frequency (PG with other planet config-
urations and tooth number combinations can produce no component at the gear mesh frequency [1,2,4,5]). It is obtained that
the spectrum only has components at frequencies f =f c ¼ pN (pN0 ), as reported in [2,4,5,7]. Other planetary gear sets can be
simulated using this model by changing the geometry parameters (number of planets, planet positions, and number of
teeth).

2.2. Lumped-parameter model

Lumped parameter model used in this work is shown in Fig. 3. It models a spur planetary gear for the in-plane case. It
consists of a ring gear, a sun gear, a carrier and N planet gears. Each of these elements has three degrees of freedom (DOFs):
two translational motions and one rotational motion. For all elements, the rotational DOFs are wk (k ¼ r; s; c; 1; 2; . . . ; N),
while the translational DOFs are xj and yj for the ring, sun and carrier (j ¼ r; s; c), and gi and ni for the planet gears
(i ¼ 1; 2; . . . N). The translational DOFs are fixed to and rotating with the carrier reference frame (X; Y) at an angular frequency
Xc . Bearings and gear meshes are modeled by linear springs. The gear mesh springs act on the lines of action of the corre-
sponding tooth pairs. A spring with stiffness kri ðtÞ is considered for the meshing between the ring gear and i-th planet gear.
Likewise, a spring with stiffness ksi ðtÞ is considered for the meshing between the sun gear and i-th planet gear. Proportional
damping C is considered to guarantee convergence of solution.
The matrix equation of motion is:

€ þ ½C þ Xc Gq_ þ ½K b þ K e ðtÞ  X2c K X q ¼ T þ FðtÞ

Mq ð1aÞ
where

q ¼ fxc ; yc ; wc; xr ; yr ; wr; xs ; ys ; ws; g1 ; f1 ; w1 ; . . . ; gN ; fN ; wN g0 ð1bÞ

Is the vector of DOFs, T is the torque applied to the transmission and FðtÞ is the force that can be considered for simulations
associated to transmission error and gear eccentricity. The matrices in Eq. (1a) and the model specifications are detailed in
[9,10].

2.3. Lumped-parameter model: Frame decomposition

Direct comparison of results from lumped-parameter model and phenomenological model simulations is not possible,
because of the different reference frame that both models consider. Inertial reference frame (as in phenomenological model)
is more useful, because of how vibrations are typically measured in practice. In order to be able to compare results between
the models and to experimental measurements, we propose the following formulation for lumped-parameter model simu-
lation results:
X
N X
N
yr ðtÞ ¼ sri ðtÞF ri ðtÞ sinðXc t þ ar þ wi Þ ¼ sri ðtÞF riy ðtÞ ð2Þ
i¼1 i¼1

where yr ðtÞ is the transducer vibration, sri ðtÞ accounts for the variable transmission path due to the movement of the planet
gears with respect to the fixed transducer, and includes the amplitude modulation effect (similar to function ari ðtÞ used in
phenomenological model [5]), Xc ¼ 2pf c , ar is the pressure angle of the planet-ring gear mesh, and F ri ðtÞ is the gear meshing
force that acts in the line of action of each planet-ring tooth pair in mesh. The latter is calculated by:
F ri ðtÞ ¼ kri ðtÞdri ðtÞ ð3Þ
where dri ðtÞ ¼ yr cos wri  xr sin wri þ gi sin ar  fi cos ar þ wr  wi is the displacement in the line of action. It considers the ring
gear and planet gears vibrations, as opposed to the function proposed by Liang et al. [16], that only considers the planet
vibrations. Notice that even though sun-planet interactions are not explicitly included in the proposed formulation, it is con-
216 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225

Fig. 2. Phenomenological model representation. (a) Ring-planet vibrations. (b) Contribution of each planet to transducer measurement. (c) Total transducer
vibration measurement. (d) Vibration spectrum of total transducer vibration.

Fig. 3. Lumped-parameter model.

sidered since the response of planet movements are determined in part by these interactions. This is in agreement with the
work of Inalpolat and Kahraman [21]. The last term of Eq. (2), sinðXc t þ ar þ wi Þ, performs the decomposition of the gear force
in the sensor direction (Fig. 4), and solves the issues of the other solutions found in literature [21–23].
In order to find the spectral composition of the theoretical sensor measurements, an analytical procedure is carried out.
For this we assume that F ri ðtÞ is a function with fundamental frequency equal to f g . The spectral structure of sensor measure-
ment yr ðtÞ is thus obtained by taking the Fourier transform (F ) of yr ðtÞ as follows:

FT X
N X
N
yr ðtÞ$Y r ðf Þ ¼ Y ri ðf Þ ¼ F fsri ðtÞg  F fF ri ðtÞg  F fsinðXc t þ ar þ wi Þg ð4Þ
i¼1 i¼1

where  is the convolution product. Considering wi =2pf c as the initial phase of the planet i, it is possible to rewrite each term
of the function in Eq. (4), as time-shifted versions of the corresponding terms for the planet 1, as follows:
 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225 217

Fig. 4. Decomposition of the gear force F ri ðtÞ into the sensor direction.

XN         
w w w
Y r ðf Þ ¼ F sr1 t þ i  F F r1 t þ i  F D1 t þ T 1 þ i ð5Þ
2pf c
i¼1 |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
2pf c 2pf c
ffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
ðIÞ ðIIÞ ðIIIÞ

where T 1 ¼ ar =2pf c , D1 ðtÞ ¼ sinðXc t þ w1 Þ, with w1 ¼ 0 and Xc ¼ 2pf c . Note that at this point we consider that the gear mesh
forces F ri , i ¼ 1; . . . ; N have the same amplitude and shape. They differ only in their phases. This consideration is analyzed at
the end of this section. Now, recognizing that term ðIÞ in Eq. (5) is a function with fundamental frequency f c , then it can be
expressed as a Fourier series, as:
X
sri ðtÞ ¼ sq expðj2pqfc ðt þ wi =2pf c ÞÞ ð6Þ
q2Z

where q 2 Z is an index and sq are the Fourier coefficients of sr1 ðtÞ. From this, we take the Fourier transform of sri ðtÞ:
X
FTfsri ðtÞg ¼ sq expðj2pqfc ðwi =2pf c ÞÞdðf  qfc Þ ð7Þ
q2Z

where dðxÞ ¼ 1, if x ¼ 0; and dðxÞ ¼ 0, if x–0. The same treatment is given to term ðIIÞ and term ðIIIÞ in Eq. (5), to obtain the
expression of the Fourier transform of yr ðtÞ:

XXXX
N
Y r ðf Þ ¼ sq F g Dh expðj2pf c hT 1 Þexpðjðq þ h þ gZ r Þwi Þdðf  ðq þ hÞf c  gfg Þ ð8Þ
q2Z g2Z h2Z i¼1

where g, h 2 Z are indexes, F g and Dh are the Fourier coefficients of F r1 ðtÞ and D1 ðtÞ, respectively.
Observing Eq. (8) we see that the spectrum only has non-zero values for f ¼ ðq þ hÞf c þ gfg . In addition, the term
expðjðq þ h þ gZ r Þwi Þ ¼ expðj2pði  1Þðq þ h þ gZ r Þ=NÞ yields non-zero values only for values of q þ h þ gZ r that are integer
multiples of N. Now, coming back to the consideration made before about the gear mesh forces having the same amplitude
and shape and being different in their respective phases only. If this condition is not fulfilled for the gear mesh forces F ri ,
i ¼ 1; . . . ; N, then it can be shown, using the presented formulation, that a spectrum with a change in the amplitude of
the existing components and the appearance of new sidebands spaced at f c around f g is obtained. These results are in accor-
dance with [2,4,5,7]. Hence, this analysis demonstrate that, with the use of the proposed decomposition, the frequency con-
tent of yr ðtÞ obtained from lumped-parameter simulation results is theoretically equivalent to the frequency content of the
same variable obtained from the phenomenological model.

3. Fault modeling

In this section faults are simulated using the two models presented in the previous section. Results are presented for three
different faults: local fault on the planet, sun and ring gears.

3.1. Procedure

Considering their differences, the methodology for fault modeling is presented separately for the phenomenological and
lumped-parameter model.
218 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225

3.1.1. Phenomenological model

For the simulation we employ periodic perturbations that modulate the healthy vibration waveform resulting from the
gear meshing between a faulty tooth and another in a healthy condition. An example of a vibration in the presence of fault is
presented in Fig. 5. The faulty vibration waveform (Fig. 5c) is simulated by multiplication of the function presented in Fig. 5a
and the healthy vibration presented in Fig. 5b. The mentioned function has a value of 1, except for intervals that are period-
ically repeated each 1=f fault . In each of these intervals the function is defined as a Hanning window with amplitude AP . As an
example, we consider the local fault is constrained to one gear mesh period T g ¼ 1=f g ; thus, the duration of each Hanning
window is T g . The fault frequency f fault associated with the element including the local fault, and the phase difference of
the perturbation function acting in the different planet vibrations are listed in Table 1, where f p , f s and f c are the rotational
frequencies of the planet gears, sun gear and carrier, respectively.

3.1.2. Lumped-parameter model

Faults on lumped parameter model are simulated by modifying the gear mesh stiffness functions. In this study we sim-
ulate different cases of spalling: a sun gear spalled tooth, ring gear spalled tooth and a planet gear spalled tooth. Specifically,
with this type of fault, a stiffness reduction is observed as shown in Fig. 6 [18]. The magnitude of reduction Ad and the time
duration depend on the width and length of the spall respectively. Depending on the location of the simulated fault, we
introduce the change of amplitude with a determined frequency on the planet-ring gear mesh stiffness, for planet and ring
gear fault, and planet-sun gear mesh stiffness for sun gear fault. Frequencies and phase differences of the amplitude mod-
ification between the different planet vibrations are presented in Table 1.

3.2. Results

For the lumped-parameter simulations, numerical solution of the matrix equation of motion, Eq. (1), is obtained by New-
mark implicit method. A proportional damping matrix C ¼ 0:07M þ 105 K  is taken, where K is the mean stiffness matrix. All
modeling parameters are listed in Table 2. For the implementation of the fault, a 50% of the maximum gear mesh stiffness is
chosen for spall simulation parameter Ad , and a duration of 0:5T g is chosen. The parameters related to the features of the
planetary gearbox simulated are in agreement with the real planetary gearbox used for experimental measurements (Sec-
tion 4). These are presented in Table 2. The same gearbox is modeled according to the phenomenological approach, in which
a Hanning window amplitude AP ¼ 10 is considered for fault modeling. It is necessary to comment that the magnitudes cho-
sen for defining the simulated faults are not relevant in this study, since the analysis is qualitative.

Fig. 5. Fault modeling in phenomenological model. (a) Perturbation function. (b) Vibration waveform without fault. (c) Vibration waveform with fault.
 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225 219

Table 1
Frequency and phase difference of perturbation for local fault cases

Fault location Fault frequency f fault Phase difference between planets

Ring gear fc ði  1Þ=Nfc

Sun gear fs  fc ði  1Þ=Nðf s  f c Þ
Planet gear fp þ fc –

Fig. 6. Gear mesh stiffness modification for tooth spalling simulation.

Table 2
PG parameters for lumped-parameter model.

Sun Ring Planet Carrier

Number of teeth 18 72 26 –
Mass (kg) 0.178 1.711 0.2 0.81
Inercia (kg m2) 5.5e6 7.18e3 4.6e5 5.7e3
Module (mm) 1.5 1.5 1.5 –
Bearing stiffness (N/m) ksx ¼ 150e6 krx ¼ 2e10 kpn ¼ 150e6 kcx ¼ 150e6
ksy ¼ 150e6 kry ¼ 2e10 kcy ¼ 150e6
Torsional stiffness (N/m) ksw ¼ 0 krw ¼ 1e15 kpw ¼ 0 kcw ¼ 0
Number of teeth 18 72 26 –
Number of planets N¼3
wi 0; 2p=3; 4p=3
Transmission ratio 1:5
Pressure angle as ¼ ar ¼ 20
Contact ratio rcs ¼ 1:57, rcr ¼ 1:94
Max. gear mesh stiffness (N/m) max max max
kri ¼ 2  108 ; ksi ¼ 0:8  kri
Min. gear mesh stiffness (N/m) min
kri ¼
max
kri =2;
min max
ksi ¼ ksi =2
Input torque (sun) (N m) 10
Input speed (sun) (RPM) 1800

Vibration waveforms and frequency spectra of simulated vibration of PG with local fault in a planet gear, sun gear and
ring gear are presented in Figs. 7–9, respectively. In the vibration waveform of the faulty planet, sun and ring, impacts at
frequencies related to fault frequencies are observed in the results from both models (see Table 1).
In the case of the local fault on a planet gear, the frequency spectra obtained with the two models, Fig. 7c and d, reveal
multiple sidebands spaced at nðf p þ f c Þ measured from other sidebands spaced at nfc measured from f g , n 2 Z.
In the case of the local fault on the sun, Fig. 8c and d show the spectra around f g for phenomenological and lumped-
parameter model respectively. In Fig. 8c, multiple sidebands spaced at nNfc and spaced at nNðf s  f c Þ, both around f g , are
observed. On the other hand, the lumped-parameter model results coincide with the phenomenological model results on
both sideband families (the first are not marked in the Fig. 8d). Additionally, multiple sidebands spaced at nfNðf s  f c Þg mea-
sured from other sidebands spaced at nfS measured from f g are observed, n 2 Z.
For the local fault on the ring gear, we can see in the results from both models multiple sidebands spaced at nNfc , but only
in lumped-parameter model results, multiple sidebands spaced at nfc are observed.
220 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225

Fig. 7. Simulated vibration of PG with a local fault on a planet gear. (a) Waveform of phenomenological model. (b) Waveform of dynamic model. (c) Freq.
spectrum of phenomenological model. (d) Freq. spectrum of dynamic model.

Fig. 8. Simulated vibration of PG with a local fault on the sun gear. (a) Waveform of phenomenological model. (b) Waveform of dynamic model. (c) Freq.
spectrum of phenomenological model. (d) Freq. spectrum of dynamic model.

4. Experimental measurements and result discussions

The experimental measurements are taken on two test rigs, TR 1 and TR 2. Both test rigs include the same single-stage
spur-gear planetary gearbox, an asynchronous motor and data acquisition system. The only difference is the loading unit: TR
 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225 221

Fig. 9. Simulated vibration of PG with a local fault on the ring gear. (a) Waveform of phenomenological model. (b) Waveform of dynamic model. (c) Freq.
spectrum of phenomenological model. (d) Freq. spectrum of dynamic model.

1 uses a DC generator, whereas TR 2, a magnetic particle brake controlled by a current signal. The TR1 is showed in Fig. 10.
The parameters of the planetary gearbox were presented in Table 2. Fig. 11a–c show the faulty planet, faulty sun and faulty
ring, respectively (artificially seeded). The case of faulty planet is carried out on TR 1 and the faulty sun and faulty ring are
carried out on TR 2. Waveform and frequency spectra of measured vibration of measured vibrations of planetary gear with a
local fault on planet gear, sun gear and ring gear are presented in Figs. 12–14, respectively.

4.1. Planet gear fault

In Fig. 12a, we can see impacts related to the fault, as both models predicted (Fig. 7). In Fig. 12b, which shows the spec-
trum of the measured vibration, the higher amplitude sidebands are spaced at nðf p þ f c Þ from other sidebands at nfc mea-
sured from f g , n 2 Z. This is in accordance with the phenomenological and lumped-parameter model results. Moreover,
multiple sidebands spaced at nfc are observed in experimental measurements and models too. Thus, both models correctly
predict the composition of the vibration spectrum.

4.2. Sun gear fault

In Fig. 13a, we can see impacts related to the fault, as both models predicted (Fig. 8). In Fig. 13b, multiple sidebands are
observed: (i) spaced at nfs around f g , (ii) spaced at nfNðf s  f c Þg measured from the first sidebands, and (iii) spaced at

Fig. 10. Test rig 1.

222 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225

Fig. 11. Local faults (artificially seeded). On (a) planet gear, (b) sun gear, (c) ring gear.

Fig. 12. Experimental measurements of PG with a local fault on a planet gear. (a) Vibration waveform. (b) Frequency spectrum.

Fig. 13. Experimental measurements of PG with a local fault on the sun gear. (a) Vibration waveform. (b) Frequency spectrum.

nfNðf s  f c Þg around f g , n 2 Z. These results are in agreement with predictions from phenomenological and lumped-
parameter model (Fig. 8). The appearance of two sideband families are due to the existence of two modulations of gear mesh
vibrations: one modulation due to the defect; and other due to the variable transmission path. In addition to this, there is the
suppression effect of spectral components, as reported in non-faulty planetary gearboxes [1,2,4,5]. The first and second
groups of sidebands suggest the existence of an additional modulation process at f s , which is not considered in the phe-
nomenological model, and for which we don’t have a convincing explanation till date. Thus, lumped-parameter model cor-
 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225 223

Fig. 14. Experimental measurements of PG with a local fault on the ring gear. (a) Vibration waveform. (b) Frequency spectrum around f g . (c) Frequency
spectrum around 6f g .

rectly predict the composition of the vibration spectrum, while the phenomenological model also does it, if it considers the
additional modulation.

4.3. Ring gear fault

In Fig. 14a no impacts are observed, as it was the case for the simulation results from both models (Fig. 9). Fig 14b, shows
the spectrum around f g , where the difference between the case with and without fault is not evidenced. Fig. 14c shows the
spectrum around 6f g , where multiple sidebands spaced at nfc , n 2 Z, are observed in the faulty case. This is in accordance
with the lumped-parameter model spectrum presented in Fig. 9d. Differently, the phenomenological model spectrum,
Fig. 9c, only presents sidebands spaced at nNfc . The reason why the phenomenological model does not produce a spectrum
with sidebands at nfc is because the vibrations v ri ðtÞ originated in each gear meshing pair were considered equal in shape and
magnitude. If variations between these vibrations are considered, which is a more general modeling approach, then these
sidebands appear in the spectrum.

Fig. 15. Distributed fault on ring gear (artificially seeded).

224 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225

Fig. 16. Experimental measurements of PG with distributed fault on the ring gear. (a) Vibration waveform. (b) Frequency spectrum around f g . (c) Frequency
spectrum around 7f g .

Taking into account that evidence of the fault is observed around the 6th multiple of f g , and that this evidence is in accor-
dance with the predictions from both models, we conclude that both models do represent the real situation in an acceptable
manner. However, the evidence is not as clear as in the simulation results. The difference is more noticeable in the vibration
waveform (Fig. 9a and b vs. Fig. 14a). The hypothesis we consider for this difference is that the contact ratio of the planet-
ring gear meshing (value 1.94), is higher compared to planet-sun gear meshing (value 1.57). This leads to a shorter time in
which the load is supported by one tooth pair; most of time two ring teeth carry the load of each planet-ring gear mesh con-
tact. That is to say, no impacts are observed in the single-tooth fault of the experimental case because, due to the geometry of
the internal gear meshing, the transition over the fault is smoother. To test this hypothesis, we took measurements with a
distributed fault on the ring gear, considering two adjacent teeth as is presented in Fig. 15. The waveform and spectrum of
this experimental measurement are presented in Fig. 16, where now the impacts and the expected sidebands are observable.
These results suggest the contact ratio hypothesis is appropriate. They also highlight the increased difficulty in diagnosing
single-tooth faults in the ring gear of planetary gearboxes.

5. Conclusions

The frame decomposition methodology presented in this paper permits a unification between two modeling approaches
commonly used for planetary gearbox vibrations (i.e. phenomenological and lumped-parameter modeling). Moreover, since
results from both models describe vibrations on the basis of a fixed reference frame, they are directly comparable to real
measurements on the gearbox’ case.
Planetary gearbox vibrations under different faulty conditions were simulated using the two models. Results from both
approaches were equivalent in all cases studied. In addition, model results were also found to be in agreement with exper-
imental measurements in all cases.
In the case of damaged ring gear, experimental results showed no evidence of the fault in the vibration waveform and
spectrum, when the fault is localized in one tooth. This is not the case when one-tooth localized defects occur in planet gear
or sun gear. The hypothesis of higher contact ratio of internal gearing with respect to external gearing, presented in the
paper, is considered as plausible.
Condition monitoring of planetary gearboxes under variable operating conditions (VOC) is currently and important and
active field of research. The methodologies presented in this paper can be directly applied in the study and modeling of plan-
etary gearbox vibrations under variable speed, by properly considering the system’s kinematics in relation to the rotational
speed of one reference shaft. Load dependency is another crucial factor to consider in planetary gearbox condition monitor-
 J. Parra, C.M. Vicuña / Mechanical Systems and Signal Processing 92 (2017) 213–225 225

ing under VOC. From both presented modeling methodologies, only lumped-parameter approach is load dependent. There-
fore, the use of this model including the proposed reference frame decomposition offers a good platform for the study of
planetary gearbox vibrations under VOC.
Finally, localized faults, as considered in this paper, occur in real gearboxes and can lead to catastrophic damage. How-
ever, as reported by [24], this defect type is not as usual as internal misalignment of components due to normal wear. Propo-
sitions such as the Load Susceptibility Characteristics (LSCh) [25] appear as interesting in this context.

Acknowledgments

This work was supported by CONICYT/PCHA/MagísterNacional/2015 – 22150475.

References

[1] P. McFadden, J. Smith, An explanation for the asymmetry of the modulation sidebands about the tooth meshing frequency in epicyclic gear vibration,
Proc. Inst. Mech. Eng. 199 (C1) (1985) 65–70.
[2] J. McNames, Fourier series analysis of epicyclic gearbox vibration, J. Vib. Acoust. 124 (2002) 150–152.
[3] R. Parker, J. Lin, Mesh phasing relationships in planetary and epicyclic gears, J. Mech. Des. 126 (2004) 365–370.
[4] M. Inalpolat, A. Kahraman, A theoretical and experimental investigation of modulation sidebands of planetary gear sets, J. Sound Vib. 323 (2009) 677–
696.
[5] C.M. Vicuña, Theoretical frequency analysis of vibration from planetary gearboxes, Forsch. Ingenieurwes. 76 (1) (2012) 15–31.
[6] Z. Feng, M.J. Zuo, Vibration signal models for fault diagnosis of planetary gearboxes, J. Sound Vib. 331 (2012) 4919–4939.
[7] L. Hong, J.S. Dhupia, S. Sheng, An explanation of frequency features enabling detection of faults in equally spaced planetary gearbox, Mech. Mach.
Theory 73 (2014) 169–183.
[8] A. Kahraman, Natural modes of planetary gear trains, J. Sound Vib. 173 (1) (1994) 125–130.
[9] J. Lin, R. Parker, Analytical characterization of the unique properties of planetary gear free vibration, J. Vib. Acoust. 121 (1999) 316–321.
[10] F. Chaari, T. Fakhfakh, R. Hbaieb, J. Louati, M. Haddar, Influence of manufacturing errors on the dynamic behavior of planetary gears, J. Mech. A/Solids
27 (2006) 738–746.
[11] A. Singh, Load sharing behavior in epicyclic gears: physical explanation and generalized formulation, Mech. Mach. Theory 45 (2010) 511–530.
[12] X. Qiu, Q. Han, F. Chu, Load-sharing characteristics of planetary gear transmission in horizontal axis wind turbines, Mech. Mach. Theory 92 (2015) 391–
406.
[13] Z. Chen, Y. Shao, Dynamic simulation of planetary gear with tooth root crack in ring gear, Eng. Fail. Anal. 31 (2013) 8–18.
[14] Z. Chen, Y. Shao, Dynamic features of a planetary gear system with tooth crack under different sizes and inclination angles, J. Vib. Acoust. 135 (2013) 1–
12.
[15] Lei F., Shaoping W., Xingjian W., Feng H., Huawei L. (2016). Nonlinear dynamic modeling of a helicopter planetary gear train for carrier plate crack fault
diagnosis. Chinese Journal of Aeronautics. Article in press.
[16] X. Liang, M.J. Zuo, M.R. Hoseini, Vibration signal modeling of a planetary gear set for tooth crack detection, Eng. Fail. Anal. 48 (2015) 185–200.
[17] F. Chaari, T. Fakhfakh, M. Haddar, Dynamic analysis of a planetary gear failure caused by tooth pitting and cracking, J. Fail. Anal. Prev. 6 (2) (2006) 73–
78.
[18] F. Chaari, W. Baccar, M.S. Abbes, M. Haddar, Effect of spalling or tooth breakage on gearmesh stiffness and dynamic response of a one stage spur gear
transmission, J. Mech. A/Sol 27 (2008) 691–705.
[19] C.G. Cooley, R.G. Parker, The geometry and frequency content of planetary gear single-mode vibration, Mech. Syst. Signal Process. 40 (2013) 91–104.
[20] Z. Chen, Z. Zhu, Y. Shao, Fault feature analysis of planetary gear system with tooth root crack and flexible ring gear rim, Eng. Fail. Anal. 49 (2015) 92–
103.
[21] M. Inalpolat, A. Kahraman, A dynamic model to predict modulation sidebands of a planetary gear set having manufacturing errors, J. Sound Vib. 329
(2010) 371–393.
[22] J.P. Torregrosa, C.M. Vicuña, Dynamic and phenomenological vibration models for failure prediction on planet gears of planetary gearboxes, J. Braz. Soc.
Mech. Sci. Eng. 36 (2014) 533–545.
[23] M. Karray, F. Chaari, A. Fernandez, F. Viadero, M. Haddar, Modulation sidebands of planetary gear set, Adv. Condition Monit. Mach. Non-stationary
Operations, Lecture Notes Mech. Eng. (2014), http://dx.doi.org/10.1007/978-3-642-39348-8_18, Springer.
[24] R. Zimroz, W. Bartelmus, T. Barszcz, J. Urbanek, Diagnostics of bearings in presence of strong operating conditions non-stationarity—a procedure of
load-dependent features processing with application to wind turbine bearings, Mech. Syst. Signal Process. 46 (2014) 16–27.
[25] R. Zimroz, W. Bartelmus, A new feature for monitoring the condition of gearboxes un in-stationary operating conditions, Mech. Syst. Signal Process. 23
(2009) 1528–1534.