You are on page 1of 22

Mechanism and Machine Theory 70 (2013) 298–319

Contents lists available at ScienceDirect

Mechanism and Machine Theory


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

Analytical investigation of tooth profile modification effects on


planetary gear dynamics
Cheon-Jae Bahk a, Robert G. Parker b,⁎
a
Caterpillar, Inc., East Peoria, IL 61630, United States
b
Department of Mechanical Engineering, Virginia Tech, Blacksburg, VA 24061, United States

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

Article history: This study investigates the impact of tooth profile modification on spur planetary gear
Received 18 June 2013 vibration. An analytical model is proposed to capture the excitation from tooth profile
Accepted 24 July 2013 modifications at the sun–planet and ring–planet meshes. The accuracy of the proposed model
Available online 28 August 2013
for dynamic analysis is correlated against a benchmark finite element analysis. Perturbation
analysis yields a closed-form approximation of the vibration response with tooth profile
Keywords: modifications. The perturbation solution is used to investigate the effects of tooth profile
Tooth profile modification modification. The tooth profile modification parameters that minimize response are readily
Planetary gear
obtained. Static transmission error and dynamic response are minimized at different amounts
Vibration
of profile modification, which contradicts common practical thinking regarding the correlation
Nonlinear
Perturbation method between static transmission error and dynamic response. Contrary to expectations, the
optimal sun–planet and ring–planet tooth profile modifications that minimize response when
applied individually may increase dynamic response when applied simultaneously. System
parameters such as mesh stiffness and mesh phase significantly affect the influence of tooth
profile modification.
© 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Excessive gear tooth deflection due to applied torque causes undesirable tooth contact patterns and increases gear vibration,
noise, and contact stress. In more severe cases, tooth surface damage like pitting and other gear failures can occur. Tooth profile
modification (TPM) is extensively used to compensate for the elastic gear and tooth deflection from the applied torque. The
amount of modification is small (μm scale) and depends on the gear tooth deflection. Its significant impact on the tooth
deformation and contact pattern can be utilized to increase the gear durability and reduce gear dynamic response, noise, and
contact stress.
Since Harris [1] developed maps showing the relationship between static transmission error and tip relief, numerous works
(e.g., [2–5]) can be found regarding TPM. Kahraman and Blankenship [6] conducted gear dynamics experiments for a single spur
gear pair and showed that there exists a particular tip relief to minimize dynamic transmission error for a given load.
Mathematical models of single mesh gear pairs including tooth profile errors were developed to study the impact of TPM on gear
dynamics [7–12]. Liu and Parker [13] introduced a dynamic model of a multi-mesh idler gear system with tooth profile
modification. Meanwhile, the effects of TPM on planetary gear dynamics have received much less research attention. Litvin et al.
[14] proposed tooth surface modifications to improve load distribution in planetary gears. Kahraman [15] and Lin and Parker [16]
included tooth profile modifications in their analytical models of planetary gears, but these were neglected in the dynamic
analysis. Abousleiman and Velex [17] showed that tip relief greatly reduces dynamic tooth loads and displacement amplitudes by
numerical simulations of their three-dimensional planetary gear model.

⁎ Corresponding author. Tel.: +1 540 231 3382.


E-mail address: r.parker@vt.edu (R.G. Parker).

0094-114X/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.mechmachtheory.2013.07.018
C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319 299

A nonlinear dynamic model of a planetary gear with tooth profile modification is proposed in this paper. The accuracy of the
TPM model is evaluated by comparisons with finite element software specially formulated to analyze contact of high-precision
surfaces like gear teeth. Perturbation analysis is employed to obtain a closed-form approximation of the dynamic response. This
solution gives insight on the parameter dependencies. The mesh stiffness only needs to be calculated once for each planetary gear,
and this is done for the gears without profile modification. The mesh stiffness is not re-calculated when the perturbation solution
is changed by different tooth profile modifications because micro-scale removal of material from the gear tooth flank should not
meaningfully alter the mesh stiffness. The perturbation solution quickly yields the peak vibration amplitude at resonance. The
analytical solution allows fast dynamic analysis to investigate the large design space of parameters and can be used as a tool for
comprehensive parameter studies. The influence of TPM on planetary gear dynamic response is studied in terms of the amount
and the length of modification. The examples of different modifications are chosen to illustrate certain features that emerge from
the perturbation solution and that differ from conventional understanding. For example, the minimum static transmission error
and dynamic response are achieved by different amounts of modifications. This discrepancy between static and dynamic response
emphasizes the importance of dynamic analysis when seeking an optimum TPM to reduce planetary gear vibration. Contrary to
the expectation of further vibration reduction, increased dynamic response is observed when the sun–planet and ring–planet
TPMs that give minimum response when applied individually are combined. The effect of TPM on planetary gear dynamic
response is greatly influenced by system parameters such as mesh stiffness fluctuation and relative mesh phase between the
sun–planet and ring–planet meshes.

2. Tooth profile modification model

The two-dimensional lumped-parameter model of a spur planetary gear developed in [16] and [18] is adopted as the basic
model. This is augmented with a tooth profile modification (TPM) model. Fig. 1, from [18], shows a schematic of the planetary
gear model. The matrix equation of motion with N planets is

::
Mx þ Kðx; t Þx ¼ F
 T ð1Þ
x ¼ xc ; yc ; uc ; xr ; yr ; ur ; xs ; ys ; us ; ζ 1 ; η1 ; u1 ; ⋯; ζ N ; ηN ; uN :

Fig. 1. A schematic of two-dimensional lumped parameter model of a planetary gear system.


300 C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319

K = Km(x,t) + Kb is the stiffness matrix including the nonlinear, time-varying mesh stiffness Km(x,t) and the bearing stiffness Kb
(refer to [16] for the detailed system matrices). ksn(t) and krn(t) in [16] are replaced with the nonlinear, time-varying stiffnesses


ksn ðt Þ δsn ≥ 0;
ksn ðx; t Þ ¼ ksn ðt ÞΘðδsn Þ ¼ n ¼ 1; 2; ⋯; N
 0 δsn b 0;
krn ðt Þ δrn ≥ 0;
krn ðx; t Þ ¼ krn ðt ÞΘðδrn Þ ¼ ð2Þ
0 δrn b 0;
δsn ¼ ys cosψsn −xs sinψsn −ζ n sinα s −ηn cosα s þ us þ un
δrn ¼ yr cosψsn −xr sinψrn þ ζ n sinα r −ηn cosα r þ ur −un ;

where ψsn = ψn − αs and ψrn = ψn + αr, ψn is the circumferential position angle of the nth planet, and αs and αr are the pressure
angles of the sun–planet and ring–planet meshes.
Fig. 2 illustrates the sun–planet mesh stiffness for the planetary gear system 1 in Table 1 calculated by quasi-static finite
element analysis at multiple points in a mesh cycle. The transition between single tooth contact and double teeth contact occurs
at the highest point of single tooth contact (HPSTC) and the lowest point of single tooth contact (LPSTC), which are 0.185 and
0.765 in Fig. 2, respectively.
When a geared system is loaded, the elastic tooth deflection causes undesirable contact behavior also known as corner contact.
In practice, TPM is employed to eliminate this corner contact. Because TPM changes the gear tooth contact pattern, the mesh
stiffness function adjusts accordingly as seen in Fig. 2. In order to take account of the profile modification in the dynamic model,
an additional excitation E(x,t) will be derived. E(x,t) acts in addition to the externally applied torque/force vector F(t). Two
different TPM models are introduced, and the accuracy of each model is evaluated by comparisons with finite element solutions.

2.1. Tooth profile error function model (model 1)

In the first model, the designed geometric deviations from the involute tooth profile are directly included in the mesh
deflection. These deviations or TPM functions are available from a tooth profile chart measured by a gear inspection or coordinate
measuring machine. Fig. 3 illustrates a typical linear TPM function. The excitation force generated by the TPM is mathematically
expressed as the product of the mesh stiffness and the TPM function. The basic concept of this TPM modeling was introduced in
prior works [7,13,15,16,19,11,12,20]. This model agreed well with dynamic analysis of a benchmark finite element solution for a
multi-mesh idler gear system [13], and this encourages its application to planetary gears.
One or more tooth pairs may be in contact at each mesh as the gears rotate. Each tooth pair has its own stiffness and TPM
functions. At a given instant, the resultant TPM excitation force at a mesh is the sum of the product of the individual time-varying
tooth pair stiffness and TPM functions. Considering the individual tooth pair loads allows one to model the partial tooth
separation that occurs when only some of the tooth pairs lose contact [13].
With tooth profile modification, the nth sun–planet and ring–planet mesh deflections are

b
δsn ¼ δsn −hsn
b ð3Þ
δrn ¼ δrn −hrn ;

x 108
9
Total mesh stiffness without TPM
8

7
Total mesh stiffness with TPM
Mesh stiffness, N/m

4 2nd pair stiffness without TPM 1st pair stiffness without TPM

2nd pair stiffness with TPM


2
1st pair stiffness with TPM
1

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Mesh cycle

Fig. 2. Sun–planet mesh stiffness for example system 1 described in Table 1.


C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319 301

Table 1
System parameters of the example planetary gear models. There is no relationship between these two different planetary gears.

Parameter System 1 System 2

Number of planets, N 3 6
ksp (N/m) 698e6 683e6
krp (N/m) 691e6 842e6
2π ðn−1Þ ½0;24;49;73;97;122π
ψn (rad) N 73
αs and αr (deg) 25 25
Sun number of teeth 27 40
Planet number of teeth 34 33
Ring number of teeth 96 106
Sun tip diameter (mm) 82.8 225
Planet tip diameter (mm) 103 187
Ring tip diameter (mm) 267 557
Sun inertia, Is/rs2 (kg) 1.76 0.478
Planet inertia, Ip/rp2 (kg) 2.90 0.330
Sun mass (kg) 2.27 8.30
Planet mass (kg) 1.39 3.65
Carrier mass (kg) 43.7 90.5
Sun bearing stiffness (N/m) 2e9 23.3e6
Planet bearing stiffness (N/m) 2e9 2.49e9
Carrier bearing stiffness (N/m) 2e9 11.1e9
Input torque to sun (N m) 1130 11,300

where hbsn and hbrn are the sun–planet and ring–planet TPM functions of the bth tooth pair. They are functions of the roll angle and
are positive when material is removed (Fig. 3). The link between the TPM functions and tip and root modifications described in
Fig. 3 is determined by which gears the modification is applied on and the driving direction. For example, when a sun gear has
root and tip modifications and drives the planet gears, the root modification enters the gear tooth mesh first followed by the tip
modification and engages with the first tooth pair. Therefore, h1s1 and h2s1 are TPM functions of the sun root and tip modification for
the sun–planet 1 mesh, respectively. These δsn and δrn replace the δsn and δrn given in Eq. (2) in the equations of motion for each
gear. For example, after inclusion of TPM, the equations of motion for the sun from [16] are

::  XN X
Bs
2 b b 
ms x s −2Ωc ẏs −Ωc xs − ksn Θ δsn δsn sinψsn þ ks xs ¼ 0
n¼1 b¼1
 ::  XN X
Bs
2 b b 
ms y s þ 2Ωc ẋ s −Ωc ys þ ksn Θ δsn δsn cosψsn þ ks ys ¼ 0 ð4Þ
n¼1 b¼1
  N X
X Bs
2 :: b b 
Is =r s us þ ksn Θ δsn δsn þ ksu us ¼ T s =r s ;
n¼1 b¼1
Amount of tooth profile modification

Root modification Tip modification

Start of root modification

Start of tip modification

SAP Tip
Roll angle

Fig. 3. Description of tip and root modification. SAP stands for start of active profile.
302 C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319

where Ωc is carrier rotation speed, Is is the sun moment of inertia, rs is the sun base radius, us = rsθs, Ts is the external torque
on the sun, and Bs is the total number of tooth contact pairs in each sun–planet mesh, which is the nearest integer greater than
the contact ratio. kbsn is the sun–planet mesh stiffness of the bth tooth pair. The mesh stiffnesses of the two contacting tooth
pairs for the example system in Table 1 calculated by finite element analysis are illustrated in Fig. 2. The first pair mesh
stiffness is active from 0 to 0.765 in a mesh cycle under no load conditions, but the engagement of the tooth pair extends to
0.87 due to the elastic tooth deflection. Assuming that TPM eliminates such extended contact, the first pair mesh stiffness is
adjusted to end at 0.765 when TPM is applied, as shown in Fig. 2. Similar adjustment is applied to the second pair mesh
stiffness.
By moving hbsn to the right side of the equation, Eq. (4) is re-written as

::  XN X
Bs XN X
Bs
2 b b  b b b 
ms x s −2Ωc ẏs −Ωc xs − ksn Θ δsn δsn sinψsn þ ks xs ¼− ksn hsn Θ δsn sinψsn
n¼1 b¼1 n¼1 b¼1

::  XN X
Bs XN X
Bs
2 b b  b b b 
ms ys þ 2Ωc ẋ s −Ωc ys þ ksn Θ δsn δsn cosψsn þ ks ys ¼ ksn hsn Θ δsn cosψsn ð5Þ
n¼1 b¼1 n¼1 b¼1

  N X
X Bs N X
X Bs
2 :: b b  b b b 
I s =r s us þ ksn Θ δsn δsn þ ksu us ¼T s =r s þ ksn hsn Θ δsn :
n¼1 b¼1 n¼1 b¼1

The left sides of Eq. (5) are identical to the equations of motion for the sun in [16] except for the tooth separation function
Θb δsn , and this is true for the equations of motion for the other components. Additional excitation terms generated by the
tooth profile modification hbsn are evident on the right sides of the equations. The excitation for the sun can be expressed
as

N X
X Bs
b b b  T
Es ðx; t Þ ¼ ksn hsn Θ δsn ½−sinψsn ; cosψsn ; 1 : ð6Þ
n¼1 b¼1

Applying similar steps for the other gear components, the TPM excitation corresponding to the vector x in Eq. (1) is defined
as
T
Eðx; t Þ ¼ ½0; Er ðx; t Þ; Es ðx; t Þ; E1 ðx; t Þ; ⋯; EN ðx; t Þ
N X
X Br
b b b  T
Er ðx; t Þ ¼ krn hrn Θ δrn ½−sinψrn ; cosψrn ; 1
n¼1 b¼1
ð7Þ
X
Bs
b b b  T X
Br
b b b  T
En ðx; t Þ ¼ ksn hsn Θ δsn ½−sinα s ; −cosα s ; 1 þ krn hrn Θ δrn ½sinα r ; −cosα r ; −1 ; n ¼ 1; 2; ⋯; N;
b¼1 b¼1

where kbrn is the ring–planet mesh stiffness of the bth tooth pair and Br is the total number of tooth pairs in contact at each ring–
planet mesh.
Examination of the additional TPM excitation E(x,t) in Eq. (7) shows that the excitation is formed by the additional mesh force
Bs
b b
from TPM that acts along the gear mesh lines of action. For the example of the sun excitation Es(x,t) in Eq. (6), ∑ ksn hsn Θb is the
b¼1
additional mesh excitation along the line of action of the nth mesh. This additional mesh force generates an additional torque on
the sun (and the planet), as is evident from the unit value of the third element in the vector in Eq. (6) and noting that the
rotational motion coordinate is us = rsθs. The translational force excitation components of Es(x,t) are calculated by projecting this
mesh force to the corresponding xs and ys coordinate directions. The TPM excitation of the other gears can be interpreted similarly
as shown in Eq. (7). This examination of E(x,t) suggests that one can have another TPM excitation model by replacing the
additional mesh force excitation terms, which are obtained in model 1 by using the mesh stiffness and the TPM function, with
those calculated in different ways, as done in model 2.

2.2. Loaded static transmission error model (model 2)

Tooth profile modification affects the gear meshing action, causing different tooth contact patterns and loaded static
transmission error. The change of loaded static transmission error indicates that the mesh force equilibrium without TPM is
altered. Additional excitation terms are needed to correct the unbalanced forcing terms when TPM is applied, and these additional
excitations are another candidate of the mesh excitation term for E(x,t).
Geared system models using loaded static transmission error as an internal excitation to the system have been
developed [9,21–24]. When the time-invariant average mesh stiffness is used in the model, poor comparison to a
C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319 303

benchmark FE model is shown when TPM is considered [13]. Thus, model 2 retains the nonlinear, time-varying mesh
stiffnesses on the left-hand side of Eq. (1) and uses loaded static transmission errors only to calculate the TPM mesh
excitations.
The planetary gear is decoupled into the sun–planet and ring–planet pairs to calculate loaded static transmission errors. The
applied torque is chosen such that the nominal mesh force is applied to the individual sun–planet and ring–planet gear pairs. For
the example of the sun gear paired with the nth planet gear, the loaded static transmission error b
δsn of the gears without TPM is
determined by the equilibrium condition

0 ¼ bf n −ksn b
δsn ; n ¼ 1; 2; ⋯; N; ð8Þ

where bf n is the nominal mesh force. Because TPM changes the loaded static transmission error b δsn in Eq. (8), the loaded static
transmission error with TPM becomes e δsn . An additional mesh force term e
esn is required for force equilibrium in the gear pair
system according to

0 ¼ bf n −ksn e
δsn þ e
esn ; n ¼ 1; 2; ⋯; N: ð9Þ

Incorporating the tooth separation nonlinearity of mesh stiffness that is needed in the dynamic model, the mesh excitation e
esn
is calculated from Eq. (9) as
 
e δsn −bf n ;
esn ¼ ksn Θ δsn e n ¼ 1; 2; ⋯; N: ð10Þ

Note that individual tooth pairs at a given tooth mesh are not modeled separately, so the contact loss of only one tooth pair out
of multiple contacting pairs is not considered in this model. For the ring–planet pair,
 
e δrn −bf n ;
ern ¼ krn Θ δrn e n ¼ 1; 2; ⋯; N: ð11Þ

esn and e
The additional mesh excitation forces e ern are generated to satisfy the mesh force equilibrium when TPM is applied. They
can be calculated from Eqs. (10) and (11) once the loaded static transmission errors e δsn and e
δrn are measured or calculated by, for
example, finite element analysis of the individual sun–planet and ring–planet gear pairs. In contrast to explicit inclusion of the
TPM functions hbsn and hbrn as in model 1, the mesh excitation forces in this model are expressed with the loaded static
transmission error as seen in Eqs. (10) and (11). Substitution of these additional mesh excitation forces into Eq. (7) yields the
alternative formulation of TPM excitation

T
Eðx; t Þ ¼ ½0; Er ðx; t Þ; Es ðx; t Þ; E1 ðx; t Þ; ⋯; EN ðx; t Þ
X
N
T
Er ðx; t Þ ¼ e
ern ½−sinψrn ; cosψrn ; 1
n¼1
ð12Þ
X
N
T
Es ðx; t Þ ¼ e
esn ½−sinψsn ; cosψsn ; 1
n¼1
T T
En ðx; t Þ ¼ e
esn ½−sinα s ; −cosα s ; 1 þ e
ern ½sinα r ; −cosα r ; −1 :

This E(x,t) is added to the right-hand side of Eq. (1) exactly as done in model 1.

3. Comparison of analytical tooth profile modification models to finite element model

The two analytical TPM models are compared to a benchmark finite element (FE) model for cases with and without TPM.
A unique commercial finite element/contact analysis tool named Calyx [25,26] is used for the FE analysis. The tool precisely
calculates the gear tooth deformation and the contact load with a relatively coarse FE mesh, which reduces simulation time
and makes FE dynamic simulation of geared systems viable. The accuracy and efficiency of this FE code have been shown in
past studies [27,28,13,18]. Two different example systems 1 and 2 described in Table 1 are used. These are distinct
examples, not two variations on a similar planetary gear. The ring gear is fixed as is common in practice to achieve the
maximum gear ratio. Power flows from the sun input to the carrier output. While the carrier rotates, its rotational deviation
from nominal kinematics is constrained to be zero, as occurs with large output inertia. Other power flows are also possible.
There are no floating central members in the examples of this work, and any effects that might arise from floating members
are not explored. Nevertheless, the methods developed in this paper are appropriate to investigate planetary gears with one
or more floating members.
Profile modifications are applied simultaneously to the sun gear tip and root areas. Unless otherwise stated, sun tip
modification starts from the highest point of single tooth contact (HPSTC), and sun root modification starts from the lowest point
of single tooth contact (LPSTC). Throughout the paper, two different modifications are used for each example system. For example
system 1, the modification is 25.4 μm for TPM A and 50.8 μm for TPM B. For example system 2, the modification is 27.94 μm for
TPM C and 38.1 μm for TPM D. All the modifications vary linearly with roll angle (Fig. 3).
304 C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319

For a rotational mode, the sun, carrier, and ring rotate but have no translational deflection, and all planets have identical
motion [16,29]. A decreasing speed sweep covering the first rotational mode natural frequency at 1815 Hz is conducted for
system 1 with and without TPM. Both the analytical and FE models use time domain numerical integration. The root mean square
(RMS) of sun rotational deflection is plotted in Fig. 4 (throughout the paper, rotational deflection is us = rsθs for the sun and us =
rpθi for the planets). Without TPM, the analytical model, which is identical for models 1 and 2 in this case, agrees with the FE
solution. Numerical solutions of both TPM models 1 and 2 match the FE solution reasonably well for TPM A and B. The difference
between models 1 and 2 in their deviation from the FE solution is small, although model 1 is slightly more accurate. Fig. 5
compares the analytical and FE models for example system 2 in Table 1. The analytical model predicts that the peak amplitude
occurs near 1700 Hz while the peak amplitude occurs at 1740 Hz with the FE model. This 2% difference is minor. Except for this
small difference of the peak resonant response frequency, both models 1 and 2 agree with the FE solution for TPM C and D.
Good agreement with the benchmark FE model for two different planetary gears and two different TPMs for each planetary
gear builds confidence in the analytical TPM models. While the accuracy of both models 1 and 2 is acceptable, model 1 is simpler.
It only requires the definition of TPM parameters such as the amount and starting point of the modification. In contrast, one needs
the extra effort to calculate static transmission error with each TPM for model 2. Model 1 will be used throughout the rest of this
paper. Subsequent results in the paper show additional comparisons with finite element analysis to further build confidence in
the model.
The question arises whether the presence of TPM at the multiple system meshes, and the associated no-load kinematic
deviations from nominal motions these TPM create, might sufficiently change the contact conditions at a given mesh so that a
model like the present one must be more detailed in considering all meshes when calculating the effective TPM, contact
conditions, and mesh force at a specific mesh. While such models might improve accuracy, we find in the above and subsequent
comparisons with the finite element benchmark solution that the analytical model gives accuracy suitable for practical and
research use. Furthermore, these full-system TPM effects between meshes would be most evident at low torque conditions where
kinematic movements of the gears from TPM are large compared to motions from elastic tooth deformations. Fig. 6 compares
analytical model and finite element results for example system 2 at 10% of the nominal torque. Example system 2 has
unequally-spaced planets and out-of-phase planet meshes where the influences between the multiple system meshes would be
most pronounced. Nevertheless, the agreement is good.
The presence of TPM combined with multiple tooth meshes that are out of phase may affect planet load sharing. To consider
this, Fig. 7 illustrates the planet load sharing of example system 2 where the tooth meshes are out of phase and the planets are
unequally spaced. Fig. 7(a) plots the variations of planet bearing loads without TPM. The bearing loads fluctuate because the tooth
meshes are out of phase. Unequal spacing between planets results in different fluctuations of the bearing load. For example,
planets 1 and 4 show larger amplitude of variation than the other planets, while planets 2 and 5 and planets 3 and 6 have mirror
symmetry expected from the planet spacing. As a result of the variations of planet bearing load, unequal load sharing is observed
at every instant, and the inequality changes over time. Fig. 7(b) compares the variations of planet 1 and 4 bearing loads with and
without TPM. When TPM is applied, the bearing loads still fluctuate, but the fluctuation amplitude is smaller. The same is true for

40

35 No TPM

30
RMS of sun rotation, µm

TPM A
25

20
TPM B

15

10

0
1500 1600 1700 1800 1900 2000 2100 2200
Mesh frequency, Hz

Fig. 4. Root mean square (mean removed) sun rotational deflection without TPM, with TPM A, and with TPM B. Comparison of FE model and analytical models 1
and 2. Gear parameters and input torque are from example system 1 in Table 1. (- - -) Model 1, (-.-) model 2, and (···) FE solution. The solid line is the common
solution from models 1 and 2.
C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319 305

35

30 No TPM

RMS of sun rotation, µm


25

20 TPM C

TPM D
15

10

0
1600 1650 1700 1750 1800 1850 1900
Mesh frequency, Hz

Fig. 5. Root mean square (mean removed) sun rotational deflection without TPM, with TPM C, and with TPM D. Comparison of FE model and analytical models 1
and 2. Gear parameters and input torque are from example system 2 in Table 1. (- - -) Model 1, (-.-) model 2, and (···) FE solution. The solid line is the common
solution from models 1 and 2.

other planets. Consequently, the planet load sharing at each instant is changed by TPM. Although it is not shown in this paper, the
planet load sharing calculated by averaging the bearing loads over a mesh cycle is equal, and this averaged load sharing with TPM
changes negligibly compared to the system without TPM. While TPM affects the static load sharing, Figs. 5 and 6 show that the
dynamic response from the analytical model compares well with the FE solutions for example system 2. This good agreement
demonstrates that the analytical model captures the impact on dynamic response of static loading sharing changes caused by
TPM.
Fig. 4 shows the positive effect of TPM on reducing dynamic response and tooth contact loss. From the FE solution without
TPM, tooth contact loss begins at 1950 Hz for decreasing speed and softening nonlinear response occurs. Nonlinear jump down
occurs near 1650 Hz. The frequency range with tooth contact loss is 300 Hz. This range is shortened to 100 Hz with TPM A.
Contact loss is eliminated completely with TPM B. Compared to the peak amplitude with no modification, the peak amplitude of
the response from the FE solution is reduced by 30% and 58% for TPM A and B, respectively.
Fig. 8 compares the static transmission error (STE) of the example system 1 sun–planet pair calculated from the FE model for a
torque equal to the nominal sun torque divided by the number of planets, 376.7 N m = 1130/3 N m. TPM A reduces the peak–

20

15
SP mesh deflection, µm

10

0
0 1 2
Mesh cycle

Fig. 6. Sun–planet mesh deflection with TPM C. Comparison of FE model (···) and analytical model 1 (- - -). Gear parameters and 10% input torque (1130 N m)
are from example system 2 in Table 1 at 1000 Hz.
306 C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319

a 45

40

Planet bearing load, kN


35

30

25
0 1
Mesh cycle

b 45

40
Planet bearing load, kN

35

30

25
0 1
Mesh cycle

Fig. 7. Variations of planet bearing load (a) without and (b) with TPM. For (a), the linestyles denote bearing load of planets 1 and 4 (—), planets 2 and 5 (- - -), and
planets 3 and 6 (···). For (b), the linestyles denote bearing load of planets 1 and 4 without TPM (—), with TPM C (- - -), and with TPM D (···). The gear
parameters and input torque are from example system 2 in Table 1.

peak STE compared to the case without modification, while TPM B increases the peak–peak STE. The reduction of dynamic
response for TPM B as shown in Fig. 4 is opposite to the increased STE fluctuation for TPM B in Fig. 8. There exists common
thinking [9,21,22] that STE can be viewed as an excitation source and that there is therefore a strong correlation between the
amplitude of STE fluctuation and dynamic response. Consequently, tooth modifications are often designed to minimize the
fluctuation of STE. This thinking is based on single pair gear systems. With such thinking, one would expect, based on Fig. 8, the
largest dynamic response with TPM B. The response with TPM A is larger, however, indicating a need for better insight and
modeling of TPM in planetary gears. The discrepancy between static and dynamic response in terms of TPM will be discussed
later.

4. Perturbation analysis with tooth profile modification

Although numerical solutions of the analytical TPM model are accurate, the numerical solution is valid only for a specific case.
Numerical simulation is not efficient to investigate the effect of TPM on planetary gear dynamic response. For instance, when one
needs to examine the variation of dynamic response with many different combinations of TPM parameters to find an optimal TPM
that minimizes gear vibration, numerical simulation requires running many cases and a great deal of computational effort. A
closed-form approximation of dynamic response is desirable. Perturbation analysis has proven effective for investigation of TPM
effects on idler gearsets [13] and for planetary gear dynamics [30]. This paper uses perturbation analysis to obtain an analytical
solution of the TPM model and utilizes the solution to study planetary gear dynamics with TPM.
C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319 307

40

35

Static transmission error, µm


30

25

20

15

10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Mesh cycle

Fig. 8. Static transmission errors of the sun–planet gear pair with different tooth profile modifications. Gear parameters are from example system 1 in Table 1 with
an input torque on the sun gear of 376.7 N m. (—) No TPM, (- - -) TPMA, and (-.-) TPM B.

4.1. Perturbation solution using the method of multiple scales

The total mesh stiffnesses as illustrated in Fig. 2 are used for perturbation analysis such that each of the total mesh stiffnesses
Bs Br
b b
ksn ðt Þ ¼ ∑ ksn and krn ðt Þ ¼ ∑ krn is a sum of the individual tooth pair stiffnesses for the nth planet meshes. The total mesh
b¼1 b¼1
stiffnesses with TPM are truncated at the left and right sides as shown in Fig. 2 and discussed earlier. When these total mesh
stiffnesses are used, the partial tooth separation is not considered. Fig. 9 compares the numerical solutions with and without
consideration of the partial tooth separation for the two TPM cases of example system 1 in Table 1. Ignoring the partial tooth
separation only slightly changes the dynamic response. Although an excessive amount of modification may increase the
difference, such a tooth profile modification could result in the increase of dynamic response and is of no interest from a practical
point of view.
The time-varying mesh stiffnesses, which are presumed known from finite element analyses of the individual meshing gear
pairs or approximated as trapezoidal functions over a mesh period, are Fourier expanded as
" #
X ∞
ðlÞ jlωt
ksn ðt Þ ¼ ksp þ b
csn e þ cc
" l¼1 # ð13Þ
X ∞
ðlÞ jlωt
krn ðt Þ ¼ krp þ b
crn e þ cc ; n ¼ 1; 2; ⋯; N;
l¼1

ðlÞ ðlÞ
where ω is the mesh frequency, bcsn and b
crn are the harmonics of the mesh stiffness variations of the sun–planet and ring–planet
meshes, ksp and krp are the average sun–planet and ring–planet mesh stiffnesses, and cc denotes the complex conjugate of the
previous term. The averages and Fourier coefficients in Eq. (13) of example systems 1 and 2 are given in Tables 1 and 2. With
these Fourier expansions, the governing equations of model 1 including the TPM excitations in Eq. (7) are re-written as

" #
:: XN   XN  
Mx þ b K Θ ðδ Þ þ
ksp 1 þ Q k 1 þ μ b K Θ ðδ Þ x
Q
sn sn sn sn rp rn rn rn rn
 n¼1 n¼1
2
þ O  ¼ FþE ð14Þ
b ¼ cð1Þ e jωt þ cc Q
Q b ¼ cð1Þ e jωt þ cc
sn sn rn rn
1 δjn N 0
Θjn ¼ ; j ¼ s; r;
0 δjn ≤ 0

. .
ð1Þ ð1Þ ð1Þ
cr1 =krp ≪1 are small parameters, cðsnlÞ ¼ bcðsnlÞ
cs1 =ksp ≪1 and μ ¼ b
where ¼ b cs1 ¼ OðÞ for l ≥ 2, cðrnlÞ ¼ b
b cðrnlÞ bcðr11Þ ¼ OðÞ for l ≥ 2,
and Ksn and Krn are matrices consisting of the coefficients of ksn and krn in Km [30].
A simplification related to the tooth separation functions is used in Eq. (14). The possible influence of TPM on the onset of
tooth separation is not considered; the nominal mesh deflections δsn and δm determine tooth separation in Θsn and Θrn. Fig. 9
compares the numerical solutions with and without the simplification for the two TPM cases of example system 1 in Table 1. The
308 C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319

25

TPM A
20

RMS of planet rotation, µm


15

TPM B
10

0
1500 1600 1700 1800 1900 2000 2100 2200
Mesh frequency, Hz

Fig. 9. Root mean square (mean removed) planet rotational deflection of model 1 for example system 1 in Table 1. (—) no simplification is made. (···) without
consideration of the partial tooth separation and (- - -) without consideration of the influence of TPM on the tooth separation.

difference of dynamic response due to the simplification is small, demonstrating the validity of the simplification. Unrealistically
large modifications are needed to cause meaningful differences between dynamic response with and without the simplification.
Both numerical and FE simulations reveal that only one tooth separation is observed during the mesh cycle, and the separation
interval is small, that is O(), compared to the mesh period. As a result, the tooth separation functions are

Θsn ¼ 1 þ θsn Θrn ¼ 1 þ μθrn ; ð15Þ

where both θsn and θrn are O(1). These tooth separation functions are not known a priori.
Each element of the excitation vector E in Eq. (7) is periodic at the mesh frequency, and their Fourier expansion gives

XN
E¼ ½ F sn Esn Θsn þ F rn Ern Θrn 
n¼1 " #
XBs
b b
X∞
ðlÞ jlωt
F sn ¼ ksn hsn ¼ F sp þ  e
ηsn e þ cc ð16Þ
b¼1 " l¼1 #
XBr
b b
X∞
ðlÞ jlωt
F rn ¼ krn hrn ¼ F rp þ  e
ηrn e þ cc ;
b¼1 l¼1

!
Bs Br
b b b b
where Esn(Em) is the constant coefficient vector of ∑ ksn hsn Θb ∑ krn hrn Θb in Eq. (7). The mean and Fourier coefficients of Fsn
b¼1 b¼1

and Fm are known from the specified mesh stiffnesses and TPM functions.

Table 2
Fourier coefficients of sun–planet and ring–planet mesh stiffnesses.

Harmonics Example system 1 Example system 2

Without TPM With TPM Without TPM With TPM


ðlÞ
b
csn , 1e6 1 −68.91 − 12.06i −84.40 − 13.74i −81.76 − 1.876i −92.65 + 2.998i
2 −28.74 − 9.660i −10.31 − 2.927i −26.07 − 1.612i 4.178 + 0.365i
3 −4.494 − 1.988i 22.81 + 11.86i 6.001 + 0.123i 31.66 − 3.006i
4 7.627 + 5.854i 14.47 + 10.03i 14.04 + 1.426i 1.500 − 0.665i
5 7.092 + 7.330i −5.803 − 6.242i 6.885 + 1.137i −19.18 + 2.883i
ðlÞ
b
csn , 1e6 1 −76.03 − 22.21i −30.14 − 3.379i −70.38 − 7.587i −108.8 − 13.89i
2 3.789 + 8.179i 12.31 + 43.86i −37.21 − 1.363i −9.660 + 0.74i
3 2.856 + 9.019i 19.87 − 36.34i −14.71 + 0.500i 35.19 + 12.54i
4 −0.796 + 2.751i −27.70 + 7.617i −2.650 + 0.883i 16.26 + 5.680i
5 1.546 − 0.572i 14.73 + 8.313i −1.187 + 0.210i −14.55 − 9.790i
C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319 309

The vibration modes are calculated from the eigenvalue problem of Eq. (1) as
2
K0 vi ¼ ωi Mvi ; ð17Þ

where KO is the mean stiffness matrix of K(x,t) in Eq. (2). The modal matrix V = [v1, …, v3N + 9] is normalized as VTMV = I.
With Eq. (15), Eq. (16), and introduction of modal coordinates as x = Vz, the qth modal equation of Eq. (14) is

XX
3Nþ9 N  
:: 2
z q þ λq ż q þ ωq zq þ ksp Q sn Gsnqw þ μkrp Q rn Grnqw zw
w¼1 n¼1
XX
3Nþ9 N
− vwq ðP sn Esnw þ P rn Ernw Þ ¼ f q þ eq ; q ¼ 1; 2; ⋯; 3N þ 9
w¼1hn¼1 i h i ð18Þ
ð1Þ jωt ð1Þ jωt
Q sn ¼ csn e þ θsn þ cc Q rn ¼ crn e þ θrn þ cc
" # " #
X

ðlÞ jlωt
X∞
ðlÞ jlωt
P sn ¼ F sp θsn þ e
ηsn e þ cc P rn ¼ F rp θrn þ e
ηrn e þ cc ;
l¼1 l¼1

where Gsnqw and Gmqw are the (q,w) elements of Gsn = VTKsnV and Grn = VTKrnV, vwq is the (w,q) element of V, Esnw and Ernw are
3Nþ9 N  
the wth elements of Esn and Ern, fq is the qth element of VTF, and eq ¼ ∑ ∑ vwq F sp Esnw þ F rp Ernw , and these are constants. Small
w¼1 n¼1
modal damping is introduced as λq = 2ζqωq.
With the multiple time scales tn = nt, the asymptotically approximated qth modal coordinate is

 
2
zq ðt 0 ; t 1 Þ ¼ zq0 ðt 0 ; t 1 Þ þ zq1 ðt 0 ; t 1 Þ þ O  ; q ¼ 1; 2; ⋯; 3N þ 9: ð19Þ

Insertion of Eq. (19) into Eq. (18) gives the perturbation equations for order 0 and 1 as

∂2 zq0 2
þ ωq zq0 ¼ f q þ eq ð20Þ
∂t 20

XX N h i XX
∂2 zq1 2 ∂2 zq1 ∂zq0 3Nþ9 3Nþ9 N
þ ωq zq1 ¼ −2 −λq − ksp Q sn Gsnqw þ gkrp Q rn Grnqw zw0 þ vwq ðP sn Esnw þ P rn Esnw Þ; ð21Þ
∂t 20 ∂t 0 ∂t 1 ∂t 0 w¼1 n¼1 w¼1 n¼1

 . 
ð1Þ ð1Þ
where g ¼ b
cr1 =krp b
cs1 =ksp ¼ μ= ¼ Oð1Þ.
For the case where the mesh frequency is close to a natural frequency, the mesh frequency is expressed as ω = ωi + σ,
where σ = O(1). The leading order solution of (20) for q = i is

jðω−σ Þt 0 f i þ ei
zi0 ¼ Ai ðt 1 Þe þ cc þ : ð22Þ
ω2i

After substitution of Ai ðt 1 Þ ¼ 12 ai ðt 1 Þe jβi ðt 1 Þ ; Eq. (22) becomes

1 j½ωt −γ ðt Þ f þe
zi0 ¼ a ðt Þe 0 i 1 þ cc þ i 2 i ; ð23Þ
2 i 1 ωi

where γi(t1) = σt1 − βi(t1). The tooth separation function is periodic at the mesh frequency and its Fourier expansion gives

( " #)
ð0Þ
X

ðmÞ jm½ωt 0 −ϕs ðt 1 Þ
Θsn ¼ 1 þ  θsn þ θsn ðt 1 Þe þ cc ð24Þ
m¼1
310 C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319

and s → r for the ring–planet mesh. The onset of tooth separation is determined by the mesh deflection, so the phases ϕs and ϕr are
chosen such that the tooth separation is in-phase with the mesh deflection. Details are discussed in [30]. Substitution of Eqs. (22) and
(24) into Eq. (21) yields the solvability condition for bounded zq1 as

∂Ai XN h i
ð0Þ ð2Þ j2ðσ t 1 −γi Þ
j2ωi þ jωi λi Ai þ ksp θsn Ai þ θsn e Ai Gsnii
∂t 1 n¼1
XX
3Nþ9 N h i
ð1Þ jðσ t 1 −γ i Þ ð1Þ jσt
− vwi F sp θsn e þeηsn e 1 Esnw þ ðs→r Þ
w¼1 n¼1
ð25Þ
XX
3Nþ9 N
ð1Þ jðσ t 1 −γ i Þ f w þ ew
þ ksp θsn e Gsniw þ ðs→r Þ
w¼1 n¼1 ω2w
XX
3Nþ9 N
ð1Þ jσt 1 f w þ ew
þ ksp csn e Gsniw þ ðs→r Þ ¼ 0;
w¼1 n¼1 ω2w

jβi ðt 1 Þ
where all occurrences of ksp become gkrp for the terms abbreviated by (s → r). With substitution of Ai ðt 1 Þ ¼ 12 ai ðt 1 Þe , the real and
imaginary parts of Eq. (25) are

∂ai 1
ωi ¼ − ωi ai λi −jχ 2 jsinðγ i þ ψÞ
∂t 1 2
ð26Þ
∂γi
ωi ai ¼ ωi ai σ −χ 1 −jχ 2 jcosðγ i þ ψÞ;
∂t 1

X
N
ksp h ð0Þ ð2Þ
i XX
3Nþ9 N
ð1 Þ f þ e
XX
3Nþ9 N
ð1Þ
χ1 ¼ θsn ai þ θsn ai Gsnii þ ksp θsn w 2 w Gsniw þ vwi F sp θsn Esnw þ ðs→r Þ
n¼1
2 w¼1 n¼1 ω w w¼1 n¼1
ð27Þ
XX
3Nþ9 N
f w þ ew h ð1Þ ð1Þ
i 3Nþ9
XX N h
ð1Þ ð1Þ
i
χ2 ¼ ksp csn Gsniw þ gkrp crn Grniw − vwi e
ηsn Esnw þ e
ηrn Ernw ;
w¼1 n¼1 ω2w w¼1 n¼1

where ψ is the phase angle of χ2. For steady state periodic response, the conditions ∂ ai/∂ ti = ∂ γi/∂ ti = 0 in Eq. (26) give the
amplitude–frequency approximation as

" rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi#
1  2
2
ω ¼ ωi þ Ξ1 ai þ 2Ξ2  2 jΞ3 j2 − ωi ai ζ i
2ωi ai
N h
X     i
ð1Þ ð0Þ ð2Þ ð1Þ ð0Þ ð2Þ
Ξ1 ¼ cs1 θsn þ θsn Gsnii þ b
b cr1 θrn þ θrn Grnii
n¼1
ð1Þ ð28Þ
N b
XX
3Nþ9 N
f w þ ew  ð1Þ ð1Þ ð1Þ ð1Þ
 3Nþ9
XX cs1 h
ð1Þ ð1Þ
i
Ξ2 ¼ b
c s1 θsn Gsniw þ b
c r1 θrn Grniw − vwi F sp θsn Esnw þ F rp θrn Ernw
w¼1 n¼1 ω2w w¼1 n¼1
k sp
ð1Þ
N b
XX
3Nþ9 N
f w þ ew  ð1Þ ð1Þ
 3Nþ9
XX cs1 h
ð1Þ ð1Þ
i
Ξ3 ¼ bcsn Gsniw þ b
c rn Grniw − vwi eηsn Esnw þ e
ηrn Ernw :
w¼1 n¼1 ω2w w¼1 n¼1
k sp

4.2. Validation of perturbation solution

Fig. 10 compares the perturbation and numerical solutions from model 1 for example system 1 in Table 1 with TPM A and TPM
AA (12.7 μm modification). The resonant response of the first (wi = 1815 Hz) and second (wi = 4863 Hz) rotational modes are
considered. Fig. 10(a) plots sun rotational deflection. Both perturbation and numerical solutions capture the softening nonlinear
response. A nonlinear jump phenomenon occurs at the first rotational mode while linear response occurs at the second rotational
mode. For both TPM A and TPM AA, the perturbation solution compares well with the numerical solution for both vibration
modes. Good agreement between the perturbation and numerical solutions with TPM A and TPM AA is also evident for the planet
translation in Fig. 10(b).
Note that the peak amplitude with TPM AA is higher than the peak amplitude with TPM A at the first rotational mode for both
sun rotation and planet translation, while the peak amplitude with TPM A is higher at the second rotational mode. Both the
perturbation and numerical solutions confirm this opposite TPM effect. This indicates that the TPM effect can change for the
different vibration modes. The sensitivity of the TPM effect to the vibration mode is discussed later.
C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319 311

a 45 First rotational mode

40

35 Second rotational mode

RMS of sun rotation, µm


30

25 TPM A

20

TPM AA
15

10

0
1500 2000 2500 3000 3500 4000 4500 5000 5500
Mesh frequency, Hz

b 10 First rotational mode Second rotational mode

8
RMS of planet translation, µm

2.5
7
2
TPM A
6
1.5
5
1
4 1600 1700 1800 1900 2000

3
TPM AA
2

0
1500 2000 2500 3000 3500 4000 4500 5000 5500
Mesh frequency, Hz

Fig. 10. Comparison of (- - -) numerical and (—) perturbation solutions with TPM A and TPM AA for the first and second rotational modes of example system 1 in
Table 1.

5. Results and discussion

With the condition of zero deviation from the backbone curve in Eq. (28), i.e., vanishing of the square root, the peak amplitude
of the resonant response is

peak jΞ 3 j
ai ¼ : ð29Þ
ω2i ζ i

All parameters in Ξ3 in Eqs. (29) and (28) are known, so the variation of the peak amplitude can be readily examined. The
ð1Þ ð1Þ
appearance of the TPM excitation terms eηsn and eηrn in Ξ3 illustrates that the peak amplitude is affected by TPM, so there is an
opportunity to reduce the vibration by applying proper TPM.

5.1. Optimal TPM for minimum dynamic response

The variation of the peak sun rotation vibration amplitude with the amount of modification is plotted in Fig. 11. The first
rotational mode with natural frequency 1815 Hz for example system 1 in Table 1 is considered here and for the rest of the
discussions unless otherwise stated. The perturbation solution is compared to the numerical and finite element (FE) solutions. All
312 C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319

45

40

Peak amplitude of sun rotation, µm


35

30

25

20

15

10

0
0 20 40 60 80 100 120 140
Amount of modification, µm

Fig. 11. Variation of sun rotation peak amplitude with the amount of sun TPM for the first rotational mode of example system 1 in Table 1. Tip modification starts
from HPSTC and root modification starts from LPSTC. (—) Perturbation, (- - -) numerical, and (···) finite element solutions.

three solutions agree. The perturbation and numerical solutions of TPM model 1 give the minimum peak amplitude near 75 μm
modification, while the FE solution has a broader range of minimized response.
Fig. 12 illustrates the effect of TPM length. The amount of modification is fixed at 76.2 μm. The tip roll angle is 37.9 deg, and the
starting roll angle of tip modification varies from 28 deg to 35 deg. The starting roll angle of root modification is adjusted so that
the root modification has the same length as the tip modification. The perturbation solution shows that the peak amplitude is
minimized near 30 deg, which is at the HPSTC. The minimum peak amplitude is achieved between 29 deg and 31 deg for the
numerical and FE solutions. The perturbation solution shows that the slope of the peak amplitude variation is different before and
after the roll angle. The peak amplitude gradually decreases as the roll angle approaches where the response is minimized
(30 deg) and increases sharply for roll angles above 30 deg. Such results give design guidance to set tolerances. While it is
desirable in the present case to apply TPM starting at 30 deg roll angle, it would be more acceptable to allow longer (i.e., lower
starting roll angle) modification rather than shorter (i.e., higher starting roll angle) modification. For example, TPM with starting
roll angle of 28 deg (error of − 2 deg) gives 10 μm peak amplitude while 32 deg (error of + 2 deg) gives greater than 20 μm peak
amplitude. Both the numerical and FE solutions confirm this pattern.
The combined effect of the amount and the length of TPM can be examined quickly using the perturbation solution. Fig. 13 shows
the contour plot of the sun rotational dynamic response varying with these two parameters for example system 1 in Table 1. The
minimum peak amplitude is achieved when the amount of TPM is near 80 μm and the starting roll angle is near 30 deg. This contour
plot shows where the response is minimized, and it shows how sensitive the response is to the TPM parameters. The variation of the

50

45
Peak amplitude of sun rotation, µm

40

35

30

25

20

15

10

0
28 29 30 31 32 33 34 35
Starting roll angle of TPM, deg

Fig. 12. Variation of sun rotation peak amplitude with the length of sun TPM for the first rotational mode of example system 1 in Table 1. Amount of modification
is 76.2 μm. (—) Perturbation, (- - -) numerical, and (···) finite element solutions.
C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319 313

120 50

20
15 25 40

100 10 30

Amount of TPM, µm
5
80
5

25
15
10
60
15

20
40
25

30
20

40 40
0
28 29 30 31 32 33
Starting roll angle of TPM, deg

Fig. 13. Contour plot of sun rotation peak amplitude varying with the amount and the starting roll angle (or length) of sun TPM for the first rotational mode of
example system 1 in Table 1.

peak amplitude with the starting roll angle (i.e., the length of modification) becomes more noticeable for large modification. For small
TPM such as 20 μm, the peak amplitude remains almost constant until the starting roll angle of 32 deg. When larger modification such
as 80 μm is applied, the dynamic response varies more significantly as the length of modification changes.

5.2. Correlation between static and dynamic response

It is widely believed that there is strong correlation between loaded static transmission error (STE) and dynamic transmission
error (DTE). From the gear vibration modeling point of view, STE is often used as an excitation source and as a relative measure of
expected gear vibration. This leads to the conclusion that DTE can be controlled by STE, or, in other words, that vibration
reduction can be achieved by decreasing STE fluctuation. Therefore, when TPM is used to reduce gear dynamic response, gear
designers seek the TPM that minimizes STE fluctuation. In addition, it is common to analyze sun–planet and ring–planet pairs
separately for planetary gear designs, partially due to the limited availability of complete planetary gear analysis tools. The
desired TPM is determined from these two individual gear pair analyses, and the two calculated TPM are then applied to the sun–
planet and ring–planet meshes of the full planetary gear. This process assumes that the beneficial effects of TPM for the individual
sun–planet and ring–planet gear pairs remain valid for the full planetary gear. These ideas are examined below.

10 45

9 40
Peak amplitude of sun rotation, µm

8
RMS of static deflection, µm

35
7
30
6
25
5
20
4
15
3
10
2

1 5

0 0
0 20 40 60 80 100 120
Amount of TPM, µm

Fig. 14. Variation of static and dynamic response with the amount of sun TPM. Tip modification starts from HPSTC and root modification starts from LPSTC. Gear
parameters are from example system 1 in Table 1. (- - -) Static transmission error of FE model of sun–planet gear pair, (-.-) sun rotational static deflection in
complete planetary gear FE model, (···) FE solution of sun rotational dynamic response from complete planetary gear system, and (—) perturbation solution of
sun rotational dynamic response from complete planetary gear system.
314 C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319

60

Peak amplitude of planet rotation, µm


50

40

30

20

10

0
0 2 4 6 8 10
Amplitude of modification, mm

Fig. 15. Variation of planet rotation peak amplitude with the amount of ring TPM for the first rotational mode of example system 1 in Table 1. Tip modification
starts from HPSTC and root modification starts from LPSTC. (—) Perturbation and (- - -) numerical solutions.

In order to examine the effect of TPM on individual gear pairs, the FE model of the sun–planet pair separated from example system
1 in Table 1 is analyzed. The torque yielding the same sun–planet mesh load as in the full planetary gear is applied to the sun–planet
pair. The variation in RMS of the sun–planet pair STE of the FE model is plotted in Fig. 14. The RMS of STE is minimized near 15 μm,
while the peak amplitude of sun rotational dynamic response is minimized near 75 μm for perturbation and between 80 and 100 μm
for FE analysis. This shows that the STE may not be a good indicator predicting the reduction of dynamic response. The variation of sun
rotational static deflection in the FE model of the full planetary gear is also plotted in Fig. 14. The 30 μm modification giving the
minimum RMS of static sun rotational deflection is far from the 75–100 μm range where sun dynamic response is minimized. One can
still reduce the peak amplitude of the resonant response with either 15 or 30 μm modification (compared to no modification), but this
misses the opportunity of large additional vibration reduction that can be achieved by larger TPM. This demonstrates that one cannot
rely on STE of an individual gear pair nor static deflection in the complete planetary gear to find an optimal TPM that minimizes DTE.
Instead, it is necessary to conduct dynamic analysis of the full system to determine TPM for reduction of vibration.

5.3. Correlation between sun–planet and ring–planet TPM

One might want to apply TPM at both the sun–planet and ring–planet meshes to maximize the vibration reduction. Fig. 15
shows that 5 μm modification on the ring–planet mesh yields minimum response. This corresponds to point A in Fig. 16. From the

120 80
Amount of sun-planet mesh tip relief, µm

50 60
70
100 30
35
15

80
40 50 60
20
7

30
10

60 12
15
35
20
10

40 7
10
40
12
30
20 3 7
5 15
5
15

7
25

3
0
0 5 10 15
Amount of ring-planet mesh tip relief, µm

Fig. 16. Contour plot of planet rotation amplitude varying with sun–planet and ring–planet mesh TPM for the first rotational mode of example system 1 in Table 1.
Tip modification starts from HPSTC and root modification starts from LPSTC.
C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319 315

18

16

Peak amplitude of sun rotation, µm


14

12

10

0
0 5 10 15 20 25 30
Amount of modification, µm

Fig. 17. Comparison of the variation of sun rotation peak amplitude with the amount of sun TPM for the second rotational mode of example system 1 in Table 1.
Tip modification starts from HPSTC and root modification starts from LPSTC. (—) Perturbation, (- - -) numerical, and (···) finite element solutions.

fact that the response is minimized at 75 μm sun–planet TPM when there is no ring–planet TPM, as discussed in relation to Fig. 11,
one may think that simultaneously applying both 75 μm sun–planet TPM and 5 μm ring–planet TPM will optimally reduce the
dynamic response (point C in Fig. 16). Fig. 16 illustrates contrary results, however. The response at point C is larger than the
response when either of these two TPM is applied individually. Instead, the response is minimized at point D where 3 μm of ring–
planet TPM and 25 μm of sun–planet TPM are applied. This illustrates that the combination of the optimal (when applied
individually) sun–planet and ring–planet modifications does not necessarily yield the optimal (or even an effective) solution for
vibration reduction.
Point D is located roughly along the diagonal line connecting 75 μm sun–planet TPM and 5 μm ring–planet TPM. While the
response amplitude varies along this line, the response at point D is similar to the response at either 75 μm sun–planet TPM or
5 μm ring–planet TPM. In this case, no significant benefit is obtained from the combination of sun–planet and ring–planet
modifications. In practical situations, such a conclusion offers significant manufacturing cost savings.

5.4. Sensitivity of TPM effects to system parameters

Fig. 17 shows the variation of sun rotation peak amplitude with the amount of TPM for the second rotational mode at natural
frequency 4863 Hz. Perturbation analysis predicts the minimum peak amplitude at around 9 μm. Numerical and finite element
solutions show reasonably good agreement. Compared to the case of the first rotational mode as shown in Fig. 11, the difference

Fig. 18. Rectangular wave mesh stiffnesses and mesh phase between the sun–planet and ring–planet mesh stiffnesses.
316 C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319

between the modifications minimizing the peak amplitude is obvious. If one chooses 75 μm modification to minimize the
vibration at the first rotational mode, the resonant response at the second rotational mode will be larger than the response
without TPM. This opposite effect of TPM at different vibration modes was seen in Fig. 10: smaller dynamic response with the
larger modification (TPM A) at the first rotational mode in contrast to smaller response with the smaller modification (TPM AA) at
the second rotational mode. Fig. 17 explains why the response with TPM A is larger than the response with TPM AA at the second
rotational mode. Both TPM A and AA have larger modification than the optimal amount of 9 μm. TPM A, the larger of the two,
yields higher response than TPM AA because the response increases continuously for modifications larger than 9 μm.
Eq. (29) shows that the peak amplitude is proportional to |Ξ3|. Ξ3 includes TPM excitation terms as well as system parameters
and modal deflections. Ξ3 is divided into two parts Ξ3a and Ξ3b as

ð1Þ
N b
XX
3Nþ9 N
f w þ e w  ð1 Þ ð1Þ
 XX
3Nþ9 cs1 h
ð1 Þ ð1Þ
i
Ξ3 ¼ b b
csn Gsniw þ crn Grniw − vwi e ηsn Esnw þ e ηrn Ernw
ωw 2 ksp : ð30Þ
w¼1 n¼1 w¼1 n¼1
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Ξ3a Ξ3b

a 30
Peak amplitude of sun rotation, µm

25

20

15

10

0
0 10 20 30 40 50 60 70 80
Amount of modification, µm

b 20

18
Peak amplitude of sun rotation, µm

16

14

12

10

0
0 5 10 15 20 25
Amount of modification, µm

Fig. 19. Variation of sun rotation peak amplitude with the amount of sun TPM for the (a) first and (b) second rotational modes of example system 1 in Table 1. The
minimum sun–planet mesh stiffnesses are (━) 600e6 N/m and (- - -) 700e6 N/m. Lines with symbols, ○ for 600e6 N/m and □ for 700e6 N/m, are from numerical
solutions.
C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319 317

Depending on the values of Ξ3a and Ξ3b, |Ξ3| can be either decreased or increased by the TPM contribution in Ξ3b. Both Ξ3a and
Ξ3b are complex and can be expressed as Ξ3a = a + bi and Ξ3b = c + di. If the signs of a, b, c, and d are all the same, |Ξ3| will be
minimized by an optimal certain TPM. Otherwise, depending on the values of a, b, c, and d, |Ξ3| could be increased for any TPM.
In order to study the relation between the system parameters and TPM, example system 1 is used for the following analysis
after replacing the actual mesh stiffnesses (Fig. 2 and Table 2) with rectangular wave mesh stiffness functions. The rectangular
mesh stiffness functions allow one to manipulate the stiffness fluctuation and relative mesh phase between the sun–planet and
ring–planet meshes, so their impact on TPM effect can be examined. Fig. 18 illustrates the conceptual rectangular mesh stiffness
functions and their relative mesh phase. In this study, the ring–planet mesh stiffness fluctuates between 600e6 N/m and 800e6 N/m
with contact ratio 1.7. The maximum value of the sun–planet mesh stiffness is fixed at 800e6 N/m, and two different minimum
stiffnesses of 600e6 N/m and 700e6 N/m are considered. The sun–planet contact ratio is 1.5. In order to define the relative mesh
phase γsr [31] between the sun–planet and ring–planet meshes as shown in Fig. 18, a reference point is selected to be at the center of
the maximum mesh stiffness zone. The mesh phase is determined by the difference between the reference points of each mesh
stiffness function. In this way, the mesh phase can be changed for simulation purposes by adjusting the shape of the mesh stiffness
function. The mesh phase can also be calculated for given gear parameters and geometry [31].
Fig. 19 shows how the effects of TPM change with the mesh stiffness and vibration mode. Fig. 19(a) compares the variation of
the sun peak amplitude with the amount of TPM for the first rotational mode. For the minimum sun–planet mesh stiffness of
600e6 N/m, the sun response is minimized at 40 μm modification. A minimum sun–planet mesh stiffness of 700e6 N/m yields the
minimum response at around 30 μm. Smaller modification is needed to minimize the dynamic response for the reduced sun–
planet mesh stiffness fluctuation. Fig. 19(b) shows the similar comparison for the second rotational mode. The peak amplitude
with the minimum stiffness of 600e6 N/m is minimized at 5 μm (which is significantly smaller than the 40 μm for the first
rotational mode in Fig. 19(a)). For a minimum stiffness of 700e6 N/m, the peak amplitude grows continuously for any
modification greater than zero.
Fig. 20 compares the variation of |Ξ3a| and |Ξ3b| for the first and second vibration modes. The minimum sun–planet stiffness of
600e6 N/m is considered. Both |Ξ3a| and |Ξ3b| increase linearly. |Ξ3a| and |Ξ3b| cross at 40 μm modification for the first rotational
mode, where the response is minimized in Fig. 19(a). For the second rotational mode, the slope of |Ξ3b| is much greater than the
slope for the first rotational mode while the slope of |Ξ3a| changes less. The crossing point of |Ξ3a| and |Ξ3b| that minimizes
response occurs at the smaller modification because of the steeper slope of |Ξ3b|.
When one designs profile modifications for planetary gears, it is common to analyze single gear pair such as sun–planet pair
and ring–planet pair. In this approach there is no means to consider the mesh phase. The impact of relative mesh phase γsr
between the sun–planet and ring–planet meshes on the TPM effect is shown in Fig. 21. Fig. 21(a) compares the variation of the
peak sun rotation vibration amplitude with the amount of modification for different relative mesh phase γsr. The first rotational
mode of example system 1 in Table 1 is considered. As the phase increases, the minimum peak amplitude increases (TPM is less
effective) and smaller modification yields this minimized peak amplitude. For γsr = π, the peak amplitude is minimized by having
no TPM; the response continuously increases with increasing modification.
Such a pattern of the response with the relative mesh phase γsr does not occur for all vibration modes. Fig. 20 shows the
impact of the relative mesh phase on the TPM effect for the second vibration mode. The minimum peak amplitude increases as the
relative mesh phase increases until γsr = π / 2. Unlike the case for the first rotational mode, the peak amplitude is minimized by
larger modification for the increased phase. For γsr = π, the response is minimized but at the largest amount of modification,

3000

2500

2000
Absolute value

1500

1000

500

0
0 10 20 30 40 50 60 70 80
Amount of modification, µm

Fig. 20. Variation of (━) |Ξ3a| and (- - -) |Ξ3b| with the amount of sun TPM for the first rotational mode of example system 1 in Table 1. Lines with symbols, ○ for
|Ξ3a| and □ for |Ξ3b|, are for the second rotational mode.
318 C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319

a 70

60

Peak amplitude of sun rotation, µm


50

40

30

20

10

0
0 10 20 30 40 50 60 70 80
Amount of modification, µm

b 16

14
Peak amplitude of sun rotation, µm

12

10

0
0 5 10 15 20 25
Amount of modification, µm

Fig. 21. Variation of sun rotation peak amplitude with the amount of sun TPM for the (a) first and (b) second rotational modes of example system 1 in Table 1. The
minimum sun–planet mesh stiffness is 600e6 N/m. Phase between sun–planet and ring–planet mesh stiffness is (━) 0 π and (- - -) π / 4, (-.-) π / 2, and (···) π.
Numerical solutions for (○) 0 π, (●) π / 4, (□) π / 2, and (■) π.

which contrasts sharply with the γsr = π case for the first rotational mode that has continuously growing response by TPM as
seen in Fig. 21(a).

6. Conclusions

An analytical tooth profile modification (TPM) model for planetary gears has been developed, and its accuracy for dynamic
analysis is evaluated by comparisons with a benchmark finite element analysis. Perturbation analysis yields a closed-form
approximation of the planetary gear dynamic response with TPM. The calculations of mesh stiffnesses are done for the gears
without profile modification. They are not repeated when profile modification is introduced. This makes the perturbation solution
computationally efficient with negligible computational expense. Thus, the perturbation solution can be used for comprehensive
parameter studies.
Using the analytical response solution, an optimal TPM that minimizes dynamic response is easily and quickly identified in
terms of the amount and length of modification. Sensitivity analysis of TPM parameters can help gear designers set an appropriate
design tolerance.
It is common to use the STE of individual gear pairs or the static deflection of the full planetary gear as response metrics when
seeking an optimal TPM to minimize the planetary gear vibration. The peak amplitude of dynamic response, however, is not
minimized at the amount of modification that minimizes either of these STE or static deflection measures. This discrepancy
C.-J. Bahk, R.G. Parker / Mechanism and Machine Theory 70 (2013) 298–319 319

contradicts the common thinking about the strong correlation between STE and DTE and emphasizes the importance of dynamic
analysis for finding an optimum TPM that minimizes gear vibration.
Combined use of the optimum sun–planet and ring–planet mesh TPM, which are determined by minimizing response when
applied individually, is expected to further reduce the dynamic response. Contrary to this expectation, however, dynamic
simulations show that increased response can occur. The minimal dynamic response is achieved at a much different combination
of sun–planet and ring–planet mesh TPM.
System parameters such as the mesh stiffness fluctuation amplitude and the relative mesh phase between the sun–planet and
ring–planet meshes change how TPM affects the dynamic response. Different TPMs are required to minimize the gear vibration
depending on the amount of mesh stiffness fluctuation and the mesh phase. Dynamic response is not guaranteed to be minimized
by TPM. Instead it may continuously grow for larger TPM for certain mesh phase choices. Thus, it is critical to take into account the
mesh phase when designing profile modifications for planetary gears.
Different TPM minimizes the vibration at different vibration modes. In other words, the TPM that reduces dynamic response
for a given vibration mode can increase dynamic response at other modes. Therefore, one needs to consider the operating speed
and which modes are most active when designing the optimal TPM for minimal gear vibration.

Acknowledgments

The authors would like to thank Dr. Sandeep M. Vijayakar of Advanced Numerical Solutions, Inc. for providing the Calyx finite
element software package used in this work.

References

[1] S. Harris, Dynamic loads on the teeth of spur gears, Proceedings of the Institution of Mechanical Engineers 172 (1958) 87–112.
[2] R.G. Munro, The Dynamic Behaviour of Spur Gears, (PhD Thesis) University of Cambridge, England, 1962.
[3] R.W. Gregory, S.L. Harris, R.G. Munro, Dynamic behaviour of spur gears, Proceedings of the Institution of Mechanical Engineers 178 (1963) 207–226.
[4] G. Niemann, J. Baethge, Transmission error, tooth stiffness, and noise of parallel axis gears, VDI-Z 112 (4) (1970), (No 4–No 8).
[5] M.S. Tavakoli, D.R. Houser, Optimum profile modifications for the minimization of static transmission errors of spur gears, Journal of Mechanisms,
Transmissions, and Automation in Design 108 (1986) 86–95.
[6] A. Kahraman, G.W. Blankenship, Gear dynamics Experiments, Part-III: Effect of Involute Tip Relief, 1996. San Diego, USA.
[7] P. Velex, M. Maatar, A mathematical model for analyzing the influence of shape deviations and mounting errors on gear dynamic behaviour, Journal of Sound
and Vibration 191 (5) (1996) 629–660.
[8] M. Amabili, A. Rivola, Dynamic analysis of spur gear pairs: steady-state response and stability of the sdof model with time-varying meshing damping,
Mechanical Systems and Signal Processing 11 (3) (1997) 375–390.
[9] H.N. Özgüven, D.R. Houser, Dynamic analysis of high-speed gears by using loaded static transmission error, Journal of Sound and Vibration 125 (1) (1988)
71–83.
[10] F. Cunliffe, J.D. Smith, D.B. Welbourn, Dynamic tooth loads in epicyclic gears, Journal of Engineering for Industry 95 (2) (1974) 578–584.
[11] C. Lee, H.H. Lin, F.B. Oswald, D.P. Townsend, Influence of linear profile modification and loading conditions on the dynamic tooth load and stress of
high-contact-ratio spur gears, Journal of Mechanical Design 113 (4) (1991) 473–480.
[12] Y. Cai, T. Hayashi, The optimum modification of tooth profile for a pair of spur gears to make its rotational vibration equal zero, 1992. 453–460. Scottsdale,
USA.
[13] G. Liu, R.G. Parker, Dynamic modeling and analysis of tooth profile modification for multimesh gear vibration, Journal of Mechanical Design 130 (12) (2008)
121402.
[14] F.L. Litvin, D. Vecchiato, A. Demenego, E. Karedes, B. Hansen, R. Handschuh, Design of one stage planetary gear train with improved conditions of load
distribution and reduced transmission errors, Journal of Mechanical Design 124 (4) (2002) 745–752.
[15] A. Kahraman, Load sharing characteristics of planetary transmissions, Mechanism and Machine Theory 29 (1994) 1151–1165.
[16] J. Lin, R.G. Parker, Analytical characterization of the unique properties of planetary gear free vibration, Journal of Vibration and Acoustics 121 (1999)
316–321.
[17] V. Abousleiman, P. Velex, A hybrid 3d finite element/lumped parameter model for quasi-static and dynamic analyses of planetary/epicyclic gear sets,
Mechanism and Machine Theory 41 (6) (2006) 725–748.
[18] V. Ambarisha, R.G. Parker, Nonlinear dynamics of planetary gears using analytical and finite element models, Journal of Sound and Vibration 302 (3) (2007)
577–595.
[19] J.P. Raclot, P. Velex, Simulation of the dynamic behaviour of single and multi-stage geared systems with shape deviations and mounting errors by using a
spectral method, Journal of Sound and Vibration 220 (5) (1999) 861–903.
[20] A. Kahraman, Dynamic analysis of a multi-mesh helical gear train, Journal of Mechanical Design 116 (3) (1994) 706–712.
[21] H. Vinayak, R. Singh, C. Padmanabhan, Linear dynamic analysis of multi-mesh transmissions containing external, rigid gears, Journal of Sound and Vibration
185 (1) (1995) 1–32.
[22] M. Maatar, P. Velex, Quasi-static and dynamic analysis of narrow-faced helical gears with profile and lead modifications, Journal of Mechanical Design 119
(4) (1997) 474–480.
[23] A. Kahraman, R. Singh, Non-linear dynamics of a spur gear pair, Journal of Sound and Vibration 142 (1) (1990) 49–75.
[24] J.R. Ottewill, S.A. Neild, R.E. Wilson, An investigation into the effect of tooth profile errors on gear rattle, Journal of Sound and Vibration 329 (17) (2010)
3495–3506.
[25] S.M. Vijayakar, A combined surface integral and finite-element solution for a three-dimensional contact problem, International Journal for Numerical
Methods in Engineering 31 (3) (1991) 525–545.
[26] S.M. Vijayakar, Calyx Users Manual, Hilliard, OH, 43026 USA, http://ansol.us/Manuals/CalyxManual.pdf2004.
[27] R.G. Parker, V. Agashe, S.M. Vijayakar, Dynamic response of a planetary gear system using a finite element/contact mechanics model, Journal of Mechanical
Design 122 (3) (2000) 304–310.
[28] R.G. Parker, S.M. Vijayakar, T. Imajo, Non-linear dynamic response of a spur gear pair: modelling and experimental comparisons, Journal of Sound and
Vibration 237 (3) (2000) 435–455.
[29] J. Lin, R.G. Parker, Structured vibration characteristics of planetary gears with unequally spaced planets, Journal of Sound and Vibration 233 (5) (2000)
921–928.
[30] C. Bahk, R.G. Parker, Analytical solution for the nonlinear dynamics of planetary gears, Journal of Computational and Nonlinear Dynamics 6 (2) (2011)
021007.
[31] R.G. Parker, J. Lin, Mesh phasing relationships in planetary and epicyclic gears, Journal of Mechanical Design 126 (2) (2004) 365–370.

You might also like