ENOC 2008, Saint Petersburg, Russia, June, 30–July, 4 2008

OPTIMIZATION METHODS FOR SPUR GEAR DYNAMICS

Marco Barbieri Giorgio Scagliarini

Dep. of Mechanical and Civil Engineering Dep. of Mechanical and Civil Engineering
Univ. of Modena and Reggio Emilia Univ. of Modena and Reggio Emilia
Italy Italy
[email protected] [email protected]

Giorgio Bonori Francesco Pellicano Gabriele Bertacchi

Dep. of Mech. and Civil Eng. Dep. of Mech. and Civil Eng. Dep. of Mech. and Civil Eng.
Univ. of Modena Univ. of Modena Univ. of Modena
and Reggio Emilia - Italy and Reggio Emilia - Italy and Reggio Emilia - Italy
[email protected] [email protected] [email protected]

Abstract uate the static transmission error, assuming that low

In the present paper two different approaches for spur excursions of the STE correspond to low vibrations,
gear noise reduction using micro-geometrical modifi- i.e. low fluctuations of the dynamic transmission
cations are compared. The two approaches are based error (DTE) [2]. Gear tooth deformations were also
respectively on the reduction of Static Transmission taken into account using a simplified cantilever beam
Error (STE) and Dynamic Transmission Error (DTE) model [3]. Beghini et al. [4] proposed an iterative
fluctuations. The dynamic behaviour of the system is method to reduce the peak to peak of the STE based
computed through a simple lumped parameter model. on FEM models. In their paper, a sequential approach,
A genetic algorithm is proposed to find the best solu- spanning couples of profile modification parameters,
tions inside the parameters space because the evalua- is described. Despite the accuracy in stiffness evalu-
tion of the objective functions requires finite elements ation, Finite Element Methods do not allow to bring
calculations and numerical ODE integrations. A relia- out a search in the parameter space using common
bility analysis is afterwards performed to evaluate the optimization techniques, e.g. steepest descent, because
effect of manufacturing errors on the dynamic perfor- the objective function is not known in an explicit
mance of the achieved optimum. formulation. It is to note that all approaches found in
literature are not global in nature, because only one
load value is considered, while, due to nonlinearities,
Key words both STE and DTE depend on the applied torque.
Optimization, gears, NVH, genetic algorithms, manu- The relationship between static and dynamic transmis-
facturing errors. sion error can be investigated using lumped parameter
models, which permits to find the dynamic behaviour
1 Introduction of the gear pair for varying speed starting from a series
The reduction of vibration and noise is one of the of static FEM calculations. On this topic, previous
main issue in spur gear design. Literature offers works from Parker et al. [5] were considered to de-
many approaches to evaluate the dynamic behaviour velop a tool for studying the dynamic effectiveness of
of such systems and their design optimization. Most profile modifications. In this paper STE is interpolated
of the design methods are based on the use of tip an by means of Discrete Fourier Transform (DFT) starting
root relief, which consist in a removal of material from fifteen reciprocal positions within a mesh cycle.
from the tooth flank in order to reduce the fluctuation This work focuses on the comparison between two dif-
of the static transmission error (STE) at a specific ferent objective functions in gear profile optimization:
torque level. Kahraman and Blankenship [1] proved the first one takes into account the harmonic content
the influence of profile modifications on the dynamic of the STE, the second considers mean amplitudes
response with experiments. of vibration of the system. Both approaches perform
In the last century, optimization approaches were simulations at three different torque levels, so that the
mainly based on simplified analytical model. For minima, found by the optimization processes, keep
example, rigid models of gear teeth were used to eval- their effectiveness despite the applied load condition.
 d
Since both objective functions can be estimated Tg2 (t) = Tg1 (t) dg2
g1
and Tg1 (t) = Tg1 are assumed to
only for a finite number of configurations, a genetic be constant.
algorithm is developed and applied to both static and The dynamic transmission error x(t) along the line of
dynamic optimization. The effectiveness of the opti- action is defined as:
mum is afterwards checked by means of the method
proposed in [10]: a random profile is created starting dg1 dg2
x(t) = θg1 (t) − θg2 (t) (4)
from the optimal one and the dynamic behaviour of 2 2
the new gear pair is compared to the unperturbed
case. It is to stress that usual tolerances in the profile kbs (t) is the back side contact stiffness.
manufacturing are larger than the magnitudes of profile Smoothing backlash functions are considered in order
modifications, for this reason such a comparison is to simulate clearances:
quite important to understand whether or not profile
1
modifications are suitable for technological applica- f1 (t) = 2 [(x(t) − b) {1 + tanh[λ(x(t) − b)]}]
tion.
1
f2 (t) = [(x(t) + b) {1 + tanh[−λ(x(t) + b)]}]
2
(5)
where 2 b is the backlash along the line of action and λ
2 Dynamic model is the shape parameter (λ = 108 ). The gear pair mesh
The dynamic model used in this paper is the simple stiffness along the line of action is given by:
lumped mass model showed in Fig. 1. Such model
considers spur gears as rigid disks, coupled along the 2Tg1 4Tg1
k(t) = = 2 (6)
line of action through a time varying mesh stiffness dg1 ST E(t) dg1 δ(t)
k(t) and a constant mesh damping c; θg1 (t) is the
angular position of the driver wheel (pinion), θg2 (t) is where
the angular position of the driven wheel (gear); Tg1 (t)
is the driving torque, Tg2 (t) is the braking torque; Ig1 dg2 dg1 δ(t)
and Ig2 are the rotary inertias; dg1 and dg2 are the base δ(t) = θ1 (t) − θ2 (t) ST E(t) = (7)
dg1 2
diameters.
δ(t) is the difference between the nominal position of
the wheel 1 (pinion) given by the exact kinematics and
Pinion Gear the actual position influenced by the teeth flexibility;
e(t)
dg2 ST E(t) is the static transmission error along the line
k(t)
of action, it depends on time because, during meshing,
both the contact point and the number of teeth in con-
c
tact can change.
dg1 Parker et al. [5] described a methodology to compute
Tg1 Qg1
Ig1 Ig2 Qg2 Tg2 θ1 (t) and θ2 (t), referred as the rotational degrees of
freedom of the pinion and the gear, for small “rigid-
Figure 1. Dynamic model of a spur gear pair body” motions; this approach is followed here for static
analyses.
Since no manufacturing errors are included, the mesh
stiffness is periodic within a mesh cycle; therefore, it is
According to literature [8], the relative dynamics of expanded in terms of Fourier series:
gears along the line of action can be represented by the
N
following equation of motion: X
k(t) = k0 + kj cos(jωm t − ϕj ) (8)
j=1
me ẍ(t)+c(ẋ(t))+k(t)f1 (x(t))+kbs (t)f2 (x(t)) = Tg (t)
(1) where ωm is the mesh circular frequency, amplitudes
˙ = d(·) , me is the equivalent mass:
where (·) kj and phases ϕj are obtained using the Discrete
dt
Fourier Transform (DFT); the number of samples n is
!−1 related to the number of harmonics N = (n − 1)/2; in
d2g1 d2g2 the following, n = 15 is considered to ensure enough
me = + (2)
4Ig1 4Ig2 accuracy in the expansion.
Similarly we have:
Tg (t) is the equivalent applied preload: N
X sts,1
kbs (t) = k0 + kj cos(−(jωm t − ϕj + j ))

dg1 Tg1 (t) dg2 Tg2 (t)

j=1
dg1
Tg (t) = me + (3) (9)
2Ig1 2Ig2
where sts,1 is the thickness of the pinion tooth space at Parameter Start End
αts Roll angle at operating pitch diameter Roll angle at tip diameter
the pitch operating diameter, see equation (3.2.32) of
magt 0 40 µm
Ref. [9] for details. αrs Roll angle at operating pitch diameter Roll angle at initial point of contact diameter
A dimensionless form of equation is obtained by let- magr 0 40 µm

ting:
Table 1. Parameter ranges for the optimization
q
k0 c
ωn = me ; ζ= 2me ωn ; τ = ωn t ;
(10)
Tg x strings, it is possible to use them to find minima of
T̄g = 2
bme ωn ; x̄ = b multivariable functions through a discretization of the
domain. The three transformations resemble the evo-
and: lution of biological systems. At the first stage, selec-
tion, the objective function is evaluated for each mem-
N   ber of the actual population and a new set of points is
kj
k̄j cos j ωωm
P
k̄j = 2
me ωn ; k̄(τ ) = 1 + n
τ − ϕj ; extracted from the actual one by means of a stochastical
j=1
N   sampling in which better elements have higher proba-
sts,1
−j ωωm bility of extraction. Crossover, which acts like chro-
P
k̄bs (τ ) = 1 + k̄j cos n
τ + ϕj − j dg1
j=1 mosome recombination, combines the extracted solu-
(11) tions in order to search different areas of the parameter
Dynamic transmission error is computed using an im- space. Since some regions of the space could not be
plicit Runge-Kutta numerical integrator. reached by crossover, mutation adds a random varia-
tion of points inside the population.
3 Profile modifications An optimization strategy based on genetic algorithms
Profile modifications are micro-geometrical removals is proposed and applied to an actual gear pair. Tab. 2
of material both from the tip and from the root of the shows data of the case study.
tooth. The parameters that define these profile modi-
fications are the roll angle of start and magnitude re-
lief at the tip, the roll angles of start and magnitudes Data Pinion Gear
of relief at the root; this way the parameter space is Number of teeth 28 43
8-dimensional. Fig 2 shows the parameters defining
Module [mm] 3 3
Pressure angle [Deg] 20 20
mag Base radius [mm] 39.467 60.610
TIP RELIEF t
Theoretical pitch radius [mm] 42 64.5
!ts
Thickness on theoretical pitch circle [mm] 6.1151 6.7128
Addendum modification [mm] 1.927 2.748
!rs
Face width [mm] 20 20
!re
Hob tip radius [mm] 0.9 0.9
ROOT
RELIEF Outer diameter [mm] 93.1 139.7
magr
Root diameter [mm] 79.1 126.2
Inner diameter [mm] 40 40
Figure 2. Representation of profile modification parameters
Mass [kg] 0.71681 1.9823
Inertia [kg m2 ] 0.0008076 0.0047762
Young’s modulus [M P a] 206000 206000
tooth profile modifications. The type of the modifica- Poisson’s coefficient 0.3 0.3
tions can be linear or parabolic with respect to the roll Center distance [mm] 111
angle. Since parabolic modifications give worst results Backlash [mm] 0.3461
in optimization [6], linear modifications are consid-
Backlash (2b) on the line of action [mm] 0.312
ered. Ranges spanned by each parameter are reported
Backside stiffness phase [rad] 1.594232
in Tab. 1.
Transmission ratio 0.6511
Contact ratio 1.28565
4 Optimization Torque (Tg1 ) [N mm] 470000
Genetic algorithms are optimization methods that start
Damping coefficient (ζ) 0.01
from an initial population of points and improve it
through the iterative application of three transforma-
Table 2. Geometrical data for the case study (courtesy of CNH Case
tions: selection, crossover and mutation. Although dis-
New Holland)
crete in nature, since genetic algorithms work on bit
4.1 Effectiveness of optimization Number of strings in the population
At a first stage, the proposed algorithm is used to find 50
npop
an optimum set of profile modifications for a specific
torque value, i.e. the nominal torque. Two differ- Crossover probability pc 1
ent objective functions, peak to peak of STE (CASE Mutation rate pm 0.4
B) and mean value of the first seven harmonics of
STE (CASE C) are applied and their dynamic effec- Multiplier for the fitness scaling 1.5
tiveness is compared to that of a pure involute pro- cmult
file (CASE A). Fig 3 shows the dynamic results: the Number of iterations niter 60
amplitude-frequency diagram is obtained starting from
strongly perturbed initial conditions. This approach,
Table 3. Parameters of the genetic optimization
even though produces lower vibration levels, is con-
sidered to be more realistic because the operating en-
vironment is subjected to strong perturbations. Mini-
mizations at the nominal torque value of peak to peak different torque values.
of STE produces the best results in terms of vibration
reduction. In the following, we will consider the har-
monic content of STE, instead of the peak to peak, be-
0.35
cause it is the direct source of excitation and for this 100% T
66% T
reason can be related to the dynamic behaviour. 0.3 33% T

0.25

0.45
RMS Amplitude

Case A
0.2
0.4 Case B
Case C

0.35 0.15

0.3
0.1
RMS Amplitude

0.25

0.05
0.2

0.15 0
0 0.5 1 1.5 2 2.5 3
Normalized frequency !/!nat
0.1

Figure 4. Amplitude frequency diagrams: CASE D - Static opti-

0.05
mization
0
0 0.5 1 1.5 2 2.5 3
Normalized frequency !/!nat

Figure 3. Amplitude-Frequency diagrams: CASE A - no profile

modifications; CASE B - peak to peak of STE; CASE C - mean value
of the first seven harmonics of STE 4.3 Dynamic optimization
In this case for each solution three numerical integra-
tions are performed for three different torque values.
In this way, the Root Mean Square (RMS) of the so-
lution is evaluated in the range between 500 to 25000
4.2 Static optimization rpm. The target of optimization is the mean value of
In the present section, the same algorithm has been the RMS for the three different torque levels. Fig 5
applied to minimize the mean amplitude of the first shows the dynamic scenario for the dynamically op-
seven harmonics of STE at three different torque lev- timized gears. The dynamic optimization approach is
els (CASE D): nominal load, 66% and 33% of nominal applied to the same case study (Case E) and the optimal
load. This way, for each point in the parameter space, solution shows the same behaviour of the static one: vi-
3x15 FEM simulations are performed, so that the first bration levels are higher at the maximum torque, lower
seven harmonics of STE could be evaluated by means at 33% of the nominal torque.
of DFT. With reference to the algorithm described in Tab 4 presents the sets of optimized profile modifi-
[6], the parameters used in the present simulations are cations for CASE D and CASE E: it is to note that
collected in Tab. 3: both solutions present very small modifications at root.
The optimal set of profile modification, which should Since root modifications are more difficult to manufac-
be capable to reduce vibrations at different loads, is ture than tip modifications, this observation suggests
tested to find out its dynamic behaviour. Figure 4 sum- that tip only modifications could be considered in de-
marizes the dynamic scenarios obtained for the three signing gear transmissions.
 0.35
100% T
66% T
0.3 33% T

0.25
RMS Amplitude

0.2

0.15

0.1

0.05

!
0
0 0.5 1 1.5 2
Normalized frequency !/!nat
2.5 3 t,s

Figure 5. Amplitude frequency diagrams: CASE E - Dynamic op-

timization

Case D Case E
!
Pinion Gear Pinion Gear
r,s
Tip Relief
αts [deg] 34.488 30.006 32.715 30.886
magt [mm] 0.039 0.037 0.022 0.020

Root Relief !
αrs [deg] 23.455 25.168 21.947 22.545 r,e
αre [deg] 14.433 20.576 14.433 20.576
magr [mm] 0.010 0.000 0.006 0.002

Table 4. Optimal profile modifications according to optimizations:

cases D and E

4.4 Reliability analysis Figure 6. Example of k-chart used to represent tolerances in gear
Considering the set of optimal profile modifications profile
proposed in Tab. 4 Case D , it can be seen that the mag-
nitudes lie in the range 0 − 40µm, which is usually 0.45
the same size of the allowable manufacturing error. To Case A
0.4 Case D
simulate the profile error, a random profile is consid- Case F

ered. This profile is defined by means of random nor- 0.35

mally distributed points with standard deviation set up

0.3
considering prescribed tolerances. In gear profile, tol-
RMS Amplitude

erances are defined using a ”k-chart”, that is a sketch 0.25

of the domain of allowable profile, see Fig. 6. 0.2

In the present case study, since profile modification
0.15
manufacturing is less accurate than pure involute pro-
file, tolerances are set up as in the following: ∆i = 0.1

8µm, ∆r = 20µm. Starting from these values, a ran- 0.05

dom profile is defined; the corresponding transmission
error is calculated and represented using the first four 0
0 0.5 1 1.5 2 2.5 3
Normalized frequency !/!nat
harmonic components; which are added to the dynamic
model as e(t) in Fig. 1.
Figure 7. CASE A - no profile modifications; CASE D - Static op-
Figure 7 shows the dynamic behaviour of the per-
timum; CASE F - Effect of manufacturing error on D case
turbed profile: vibration levels are higher than in Case
D, but still lower than pure involute profile (Case A).
levels: the first is based on static finite element calcula-
5 Conclusion tions, the second involves a numerical ODE integration
Two optimization strategies are proposed to find op- as well. The main focus is to reduce vibrations in spur
timal sets of profile modifications at different torque gears. The effectiveness in vibration reduction has been
checked for both approaches, showing a good reduc-
tion of noise level. The first method gives good results
with much lower computational time. For this reason,
the optimum obtained on a static basis is assumed as
a standard. Since the magnitudes of the modifications
are of the same size of the manufacturing error, a per-
turbed profile is considered in order to prove that vi-
bration level of the optimized profile is still better than
pure involute even if profile errors are taken into ac-
count.

References
Kahraman A., and Blankenship G. W., 1999, Effect
of Involute Tip Relief on Dynamic Response of Spur
Gear Pairs, Journal of Mechanical Design, 121, pp.
313-315.
Harris, S.L., 1958, Dynamic Loads on the Teeth of Spur
Gears, Proceedings of the Institution of Mechanical
Engineers, 172(2), pp. 87-100.
Tavakoli, M. S., and Houser, D. R., 1986, Optimum
Profile Modifications for the Minimization of Static
Transmission Errors of Spur Gears, Journal of Mech-
anism, Transmissions, and Automation in Design,
108, pp. 86-95.
Beghini, M., Presicce, F., and Santus, C., 2004, A
Method to define Profile Modification of Spur Gear
and Minimize the Transmission Error, AGMA Tech-
nical Paper, 04FTM3.
Parker, R. G., Vijayakar, S. M., and Imajo, T., 2000,
Nonlinear Dynamic Response of a Spur Gear Pair:
Modeling and Experimental Comparisons, Journal of
Sound and Vibration, 237, pp. 435-455.
Bonori, G., Barbieri, M., Pellicano, F., 2008, Opti-
mum Profile Modifications of Spur Gears by Means
of Genetic Algorithms, Journal of Sound and Vibra-
tion, 313, pp. 603-616
Faggioni, M., Pellicano, F., Andrisano, A. O., Bertac-
chi, G., 2007, Dynamic Optimization of Spur Gears,
2007, Proceedings of ASME 2007 IDETC/CIE
Kahraman, A., and Singh, R., 1990, Non-linear Dy-
namics of a Spur Gear Pair, Journal of Sound and
Vibrations, 142 (1), pp. 49-75.
Bonori, G., 2006, Static and Dynamic Mod-
elling of Gear Transmission, Ph.D. thesis, Uni-
versity of Modena and Reggio Emilia, Italy,
http://www.vibrazioni.unimo.it/lab1/index.htm.
Bonori, G., Pellicano, F., 2007, Non-smooth dynamics
of spur gears with manufacturing errors , Journal of
Sound and Vibration, 306, pp. 271–283