Vous êtes sur la page 1sur 14

Mechanism and Machine Theory 46 (2011) 544557

Contents lists available at ScienceDirect

Mechanism and Machine Theory


j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / m e c h m t

Dynamic optimization of spur gears


Marcello Faggioni a, Farhad S. Samani a,b, Gabriele Bertacchi a, Francesco Pellicano a,
a
b

Department of Mechanical and Civil Engineering, University of Modena and Reggio Emilia, Modena, Italy
Department of Mechanical Engineering, Bahonar University of Kerman, Kerman, Iran

a r t i c l e

i n f o

Article history:
Received 21 January 2010
Received in revised form 2 November 2010
Accepted 16 November 2010
Available online 31 December 2010
Keywords:
Gears
Optimization
Vibration

a b s t r a c t
This paper presents a global optimization method focused on gear vibration reduction by
means of profile modifications. A nonlinear dynamic model is used to study the vibration
behavior; such model is validated using data available in literature. The optimization method
considers different regimes and torque levels; the objective function can be the static
transmission error or the maximum amplitude of the gear vibration in terms of dynamic
transmission error. The procedure finds the optimal profile modification that reduces the
vibrations over a wide range of operating conditions. In order to reduce the computational cost,
a RandomSimplex optimization algorithm is developed; the optimum reliability is estimated
using a Monte Carlo simulation. The approach shows good performances for the computational
efficiency as well as the reliability of results. Finally, an application to High Contact Ratio (HCR)
gears is presented and an extremely good performance is obtained by combining optimization
procedures and HCR properties.
2010 Elsevier Ltd. All rights reserved.

1. Introduction
The noise generated by spur gears is mainly due to the transmission error uctuation, which is the difference between the
theoretical relative position of two unloaded gears (without manufacturing errors) and the relative position of a gear pair under
actual operating conditions [1].
Introducing suitable prole modications allows to change the phase and the ratio of the teeth load exchange and to reduce the
transmission error during a single meshing period [2]; this approach is valid when only the teeth elastic deections are considered.
In 1983 Sato et al. [3] studied analytically and experimentally the inuence of prole modications on gear vibrations. This
paper was one of the rst works that obtained experimental conrmation regarding gear optimization.
One of the most interesting studies on prole modications, toward noise gear reduction, is due to Tavakoli and Houser [4],
1986, which developed an original optimization algorithm based on the modied Complex method, the steepest descent and the
sequential one-variable search. The optimization method was based on a local approach. An objective function based on the mean
value of the transmission-error-harmonics was used; out of design torques were analyzed on the optimal gears. The dynamic
behavior was not studied and the static transmission error was evaluated by means of a cantilever beam model.
In 1989, Simon [5] used a regressive analysis, based on numerical computations, in order to minimize the uctuation of the
teeth load distribution. In 1990 Munro et al. [6] developed a simple optimization approach, based on the Harris maps; the method
was qualitative, because it did not consider the load distribution due to the real teeth deformations.
In 1992, Cai and Hayashi [7] used a nonlinear dynamic model to evaluate the effects of the static optimization; the transmission
error was evaluated using a simple model based on elementary formulation and the InceStrutt diagram was determined from the
Corresponding author. Department of Mechanical and Civil Engineering, University of Modena and Reggio Emilia, V. Vignolese, 905, 41100 Modena, Italy.
Tel.: +39 0592056154; fax: +39 0592056126.
E-mail address: francesco.pellicano@unimore.it (F. Pellicano).
0094-114X/$ see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mechmachtheory.2010.11.005

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

545

Mathieu equation. Moreover, the inuence of the main parameters (torque, contact ratio) on the dynamic scenario was
analyzed. Applying the optimization they found about 50% vibration amplitude reduction, over a reasonable range of design
parameters.
In 1996, Velex and Maatar [8] developed a new model for the dynamic analysis of 3D gear pairs, such model allowed to carry
out a deep analysis of the effect of misalignments on the gear dynamics. In 1999, Lin et al. [9] presented a computer simulation
program, which was able to predict the dynamic behavior of a gear pair; the inuence of the torque and the angular velocity was
analyzed. They found that parabolic modications give smaller amplitude of vibration for out of design conditions; in particular,
they claimed that parabolic modications generally reduce the sensitivity to load variations.
Litvin et al. [10] attacked the problem of gear noise using three dimensional models (helical, hypoid and worm gears); such
work concluded that two dimensional parabolic modications are optimal for noise reduction. Note that in Ref. [10] the effect of
the geometry on the vibration is estimated in a heuristic way, without solving the governing equations; this is conrmed by the
authors, who stated that their model is a rst guess in terms of dynamic analysis.
In 2005, Fonseca et al. [11] optimized the harmonics of the static transmission error, using the same static model of Tavakoli
and Houser [4], by means of a genetic optimization algorithm. Ref. [11] is one of the rst global optimization approaches applied to
gearing.
Recently Bonori et al. [19] applied genetic algorithms for optimizing spur gears in terms of static transmission ratio, the
procedure considered a nonlinear nite element analysis for the evaluation of the transmission error within the optimization
algorithm; moreover, in such paper some dynamic analyses were carried out to conrm the quality of the optimum.
A further brief discussion must be focused on High Contact Ratio (HCR) gears. It is well known that increasing the average
number of teeth in contact leads to a reduction of the vibration amplitude; moreover, in order to get a further reduction of the
vibration, HCR gear proles can be optimized. Sato et al. [3] found that HCR gears are less sensitive with respect to manufacturing
errors. In particular, such kind of gears allows larger tolerance in the tip relief length. Moreover, they found that, in the absence of
pressure angle error, the best contact ratio should be about 2; otherwise, it is better to have a contact ratio about 1.7 or higher than
2.3. Kahraman and Blankenship [12] published an experimental work on HCR gear vibration; they found that the best behavior is
obtained with an integer contact ratio, even though other specic non integer (rational) contact ratios can minimize the amplitude
of some specic harmonics of the static transmission error. It is important to note that in Ref. [12] HCR gears were obtained by
modifying the outside diameter; the other macro-geometric parameters, e.g. the number of teeth, were left unchanged.
The analysis of the literature shows that the optimization approaches based on the minimization of the static transmission
error uctuation are effective for the torque level used during the optimization. Moreover, optimizations based on actual vibration
levels, including the effects of out of design conditions, are not present in the literature.
A discussion presented at the end of Ref. [4] is of great interest and strictly related to the present work; in such discussion R. W.
Cornell commented in detail the work of Tavakoli and Houser [4], he wrote: It appears that the next step in applying
optimization to tooth prole design would be to include the dynamic effects. If this is done, the objective function would have to
cover the operating load and speed spectrum....
The present work is a contribution toward the suggestions expressed by Cornell. An optimization algorithm, based on the
Random sampling and the Simplex method [13], is used to reduce the gear vibrations with an acceptable computational cost. The
optimization is carried out considering the evaluation of the teeth deection using a nonlinear Finite Element Method (FEM), time
varying stiffness, different torque levels, backlash and the actual vibration level over a wide speed range; the latter two items are
considered in the case of the dynamic optimization. The objective functions are the peak to peak of the static transmission error or
the vibration level. The optimization approach is able to nd the global optimum due to the sequential use of the Random sampling
(global) and the Simplex method (local). The rst one selects a point in the parameter space that is close to the optimum, and the
latter one renes the solution and nds the optimum; the combined use of these strategies allows to handle a large parameter
space and to carry out an effective global optimization. A reliability analysis is performed to check the robustness of the optimum.
It is worthwhile to underline that the reduction of the vibration leads to a reduction of stresses. For example, in Ref. [14] a
formula was developed for estimating the dynamic factor in terms of bending stresses, using the dynamic transmission error; such
formula implicitly proves that there is a strict correlation between the dynamic transmission error and the stresses on gears.
After applying the optimization on an actual gear pair, a further improvement of the dynamic behavior has been attempted. The
initial gear pair has been modied in terms of macro geometry: the numbers of teeth as well as all geometric parameters have been
changed and a new equivalent gear pair having a higher contact ratio (HCR) is found; such new gear pair has the same strength
and transmission ratio as the initial gear pair. The HCR gear pair is then optimized by correcting tooth proles, using the proposed
algorithm. The purpose is to nd a new gear pair, having similar mechanical characteristics of the original one, which exhibits a
strong reduction of vibration as torque and speed vary.
2. Physical model
In order to model the dynamic behavior of a spur gear pair, some physical characteristics must be considered. First of all, the
parametric excitation due to the meshing stiffness uctuation and the transmission error variations, caused by the tooth elastic
deections, must be included in the model. Moreover, there is the possibility of loss of contact, due to the circumferential clearance
between teeth. The present model has been widely used in the literature [1517], and it allows to take into account the
aforementioned phenomena; conversely, the compliance of bearings and shafts is neglected as well as the effect of manufacturing
errors and misalignments.

546

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

Fig. 1. Torsional model for a gear pair.

The meshing of a spur gear pair is represented by a two-DOF model (Fig. 1); the dynamics can be studied by means of a oneDOF model in terms of transmission error, without loss of generality (see Ref. [16] for details):
::
x + 2 x + gx e = M

where t is the time [s]; is the nondimensional time; x = r11 r22 is the transmission error; x = x / b; 2b is the backlash
(Fig. 2); k(t) is the time varying meshing stiffness [N/m] (Fig. 1); () is the nondimensional meshing stiffness; g(d ) is the
backlash function; e() is the rigid body displacement due to intentional deviations from the pure involute (tip and root relief);
e = e/b; (x) = d(x) / d; and 1 and 2 are gear angular positions. Moreover:
8
T1
I I
>
>
M=
; m= 2 12 2 ;
>
2
>
r
bm
r
I
>
1
n
1 2 + r 2 I1
>
>
>
s
>
<
k
c
> n = 2m; = 2m ; = n t;
>
>
n
>
>
>
>
k = n
>
>
: =
;
m2n

where I1 and I2 are moments of inertia; r1 and r2 are base radii; k is the average of k(t); T1 is the driving torque; n is the
natural frequency of the linearized and averaged system; and is the damping ratio.
The dynamical system of Fig. 1, governed by Eq. (1), is a piecewise linear time varying system, which can be represented by the
one-dof model of Fig. 2.
The mesh stiffness is periodic; therefore, it is expanded in Fourier series:

= A0 + Ai cosip + Bi sinip

i=1

1 z1
is the dimensionless excitation frequency, 1z1 is the fundamental excitation frequency (meshing frequency), 1
n
is the rotational speed of the gear 1 and A0 is the mean value of the stiffness, (). The backlash function g(x()) is given by:
where p =

8
x N 1;
< x1;
g x =
x + 1;
xb1;
:
0;
otherwise:

Differential Eq. (1) is solved numerically using a standard adaptive step size integration method. The mesh stiffness ()
depends on the relative position of gears, i.e. the number of tooth pairs in contact, and the gear geometry. Note that () is

Fig. 2. SDOF model.

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

547

Fig. 3. a) Finite element model, b) intentional deviations from involute, tip and root relief.

evaluated considering micro geometric prole modications (tip and root reliefs). A static nonlinear FEM analysis is carried out to
evaluate the mesh stiffness: the inner ring of the driven gear is supposed xed; the inner ring of the driving gear is loaded with a
torque (obtained through suitable tangential forces); and the inner ring of the driving gear is supposed radially rigid and is free to
rotate about the gear axis. Once the gear deformation eld is evaluated, the rotation of the inner ring of the driving gear is
measured and the stiffness is trivially evaluated as the ratio of the torque to the rotation.
The aforementioned procedure takes into account possible rigid body rotations due to the initial gaps between teeth resulting
from prole modications; the presence of rigid body rotations is represented by e(t).
During the optimization process, the geometry is continually changed, as tip and root reliefs are the parameters of the
optimization; this means that the whole FEM model must be continually rebuilt (geometry, nodes location, elements denition
and so on, see Fig. 3a). The strain eld, and consequently (), is computed for each micro-geometrical modication and each
angular position during the meshing period. The previously mentioned procedure cannot be performed manually, as a huge
amount of geometries must be analyzed during the optimization process. Therefore, the software GearDesign has been
developed to carry out the following tasks: i) gear geometry generation; ii) automatic meshing of gears (FEM model and
mounting: see Fig. 3a); iii) generation of an output, compatible with a FEM software (MSC/Marc); and iv) launching calculations
for a desired load case and extracting results in terms of meshing stiffness.
Prole modications, i.e. tip and root reliefs (Fig. 3b), are considered in the FEM model; therefore, the effect of such micro
geometric modications is modeled with extreme accuracy. This means that () includes all geometrical and physical aspects of
the gears.
When designers develop gear boxes, generally they apply tip and root relief in order to improve gear performances. The problem in
determining the best set of prole modications is readily understood by computing the number of parameters. In order to dene tip
and root reliefs for a tooth, one needs four parameters; therefore, for a gear pair the number of parameters is eight. Having an eightdimensional parameter space makes it impossible to nd out the best set of parameters without developing automatic procedures.

Table 1
Data of the benchmark test case.
Case study

Pinion

Gear

Number of teeth
Normal module [mm]
Normal pressure angle []
Pitch diameter [mm]
Outside diameter [mm]
Root diameter [mm]
Face width [mm]
Center distance [mm]
Nominal torque [Nm]
Young Modulus [MPa]
Poisson's ratio
Mass density [kg/m3]
Damping ratio

50
3
20
150
156
140.68
20
150
340
206000
0.3
7850
0.01

50
3
20
150
156
140.68
20

548

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

Fig. 4. Comparison with the experiments. Amplitude frequency diagram: experimental data [15]; present approach.

The present model is validated through comparisons with experimental data


found in literature [15]; in particular, the gear set
q
described in Table 1 is considered; Fig. 4 shows the RMS (root mean square T0 x2 t dt = T ) of the dynamic transmission error of a
spur gear pair, as the meshing frequency spans a domain from 0.2 up to 1.2 resonance frequency (the dimensionless
frequency is the ratio of the actual excitation frequency to the natural frequency of the system). There is a good agreement
among the present model and the experiments. The response is clearly nonlinear (softening) and saddle-node bifurcations are
present. For some excitation frequencies, e.g. before the resonance, two different amplitudes can be reached depending on the
initial conditions. The nonlinearity takes place when the inertia forces cause contact loosing, i.e. when the oscillator of Fig. 2 passes
into the dead zone (restoring force equal to zero). Note that in Fig. 4 peaks at low frequency are present; such peaks depend both
on the presence of higher harmonics in the excitation source and the appearance of super harmonic resonances due to the
nonlinearity.
The behavior of the present model is slightly less softening than the experiments, Fig. 4; this is in full agreement with the
literature, see e.g. Ref. [14], the reason could be addressed to the strong simplication of the one-dof model that does not consider
the dependence of the time varying stiffness on the amplitude of oscillation. In order to carry out such comparison, the procedure
assumes that all solutions have the same basin of attractivity [18], and tries to follow, as well as possible, stable solutions.
3. Case study and dynamic analysis
The numerical analysis is carried out on the spur gear pair described in Table 2, which is part of an agricultural vehicle gearbox
(courtesy of Case New Holland Italy S.p.A.); prole modications reported in Table 2 follow the Case New Holland standard; the
gear pair with such modications will be considered the reference system (standard modications). It is important to note that the
Table 2
Data of the test case.
Case study

Pinion

Gear

n of teeth
Normal module [mm]
Pitch diameter [mm]
Outside radius [mm]
Root radius [mm]
Prole shift [mm]
Hob tip radius [mm]
Shaft diameter [mm]
Normal pressure angle []
Start tip radius [mm]
Linear tip amplitude [mm]
Face width [mm]
Center distance [mm]
Backlash [mm]
Nominal torque (pinion) [Nm]
Young Modulus [MPa]
Poisson's ratio
Mass density [kg/dm3]
Damping ratio
Contact ratio

28
3
84
46.55
39.55
1.927
0.9
40
20
44.54
0.016
27
111
0.312
470
206000
0.3
7.85
0.01
1.54

43
3
129
69.55
63.1
2.748
0.9
40
20
67.64
0.018
22.5

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

549

Dimensionless mesh stiffness

1.1

1.05

0.95

0.9
0.85

0.8

0.2

0.4

0.6

0.8

Dimensionless mesh period


Fig. 5. Mesh stiffness () and Fourier series approximation (-); test case of Table 2. = 1 k = 305572 N/m; n = 4865 2 rad/s.

standard prole modications are used for comparison only; indeed, the optimization algorithm presented here does not need any
seed set.
The meshing period has been sampled considering 14 angular positions; three different torques are considered: 33%, 66% and
100% of the nominal torque (load cases 1, 2 and 3); see Table 2. In the dynamic simulation, 1000 regimes (angular speed) are
considered; this procedure assumes that all solutions have xed initial conditions.
Fig. 5 shows the normalized mesh stiffness during a mesh period and the corresponding Fourier approximation using 6
harmonics; the peak to peak of this function is about 10% of the mean value, rather small for spur gears, because the standard test
case of Table 2 presents prole modications designed for reducing such uctuation.
Due to the non-smoothness of Eq. (1), the response presents nonlinear behaviors (see Fig. 6a,b). Note that the highest Root
Mean Square (RMS) is not directly related to the torque, for example in the case of Fig. 6a (standard case) we have: 33% torque
leads to the largest RMS and gives the highest nonlinearity; 100% torque gives stronger super harmonics (p 0.5). For standard
prole modications and 33% torque one can see a high amplitude branch for p N 1.2 (p is the nondimensional frequency of
excitation, Eq. (3)), such branch represents a period two subharmonic response, due to a parametric instability; this response is
not found for other cases analyzed in this paper; therefore, this phenomenon is not investigated further.
4. The objective function
It is well known from the literature [7,9] that reducing the peak to peak of the transmission error leads to a reduction of
vibration; unfortunately, this kind of optimization is strongly dependent on the torque level [3,4]: e.g. an optimal gear set for the
maximum load can behave worse than a non-optimal one for reduced loads.

Dimensionless RMS(x)

0.25

(b) 0.4

load case 1
load case 2
load case 3

0.2

0.15

0.1

0.05

0.2

load case 1
load case 2
load case 3

0.35

Dimensionless RMS(x)

(a)

0.3
0.25
0.2
0.15
0.1
0.05

0.4

0.6

0.8

Dimensionless frequency

1.2

1.4

0.2

0.4

0.6

0.8

1.2

1.4

Dimensionless frequency

Fig. 6. Amplitude frequency diagram; test case of Table 2. Load cases 1, 2 and 3 (33%, 66% and 100% of the nominal torque); a) standard prole modication, b) pure
involute.

550

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

In order to circumvent this problem, in the present work, the vibration amplitude is evaluated over a wide range of operating
conditions and the maximum is considered the objective function; i.e. the vibration level at different torques and rotation speeds is
considered. A static objective function is also considered for comparison purposes.
4.1. Static objective function FPPS
The Peak to Peak of the static transmission error within a mesh period (PPS) is evaluated; the static objective function is
FPPS = MAXtorque(PPS), i.e. the maximum of the PPS is evaluated as the torque varies. For the case of Table 2, with modications we
have FPPS = 8.00 m at T = 0.33 Tmax; without modication (pure involute tooth) we have FPPS = 9.75 m at T = Tmax. Note that the
FPPS returns the value of the PPS at 33% of the nominal torque (with standard modications) and 100% for the pure involute case;
this is easily explained: prole modications of Table 2 work well for high torque; conversely, for low torques the prole
deviations exceed the teeth deection and cause a rigid body motion, which increases the PPS.
4.2. Dynamic objective function FDTE
The system response is determined in terms of Dynamic Transmission Error (DTE = x xmean); the dynamic objective
function is FDTE = MAXtorque(MAXfreq.(RMS(DTE))), i.e. the maximum of the RMS is determined as torque and frequency are varied.
It is worthwhile to stress that such objective function is obtained after analyzing the whole dynamic scenario; here the choice is to
consider the worst condition in terms of vibration amplitude, but other strategies could also be considered, for example the
average of the RMS instead of the MAX. For the case of the Table 2 (standard case), including modications, FDTE = 0.296 at
T = 0.33Tmax (Fig. 6a) and without modication (pure involute tooth), FDTE = 0.402 at T = Tmax (Fig. 6b).
Comparing the FPPS with the FDTE one can argue that these functions are strictly related; however, from Fig. 6 the complexity of
the dynamic scenario is evident and the direct relationship between the forcing (STE, static transmission error) and the response
(DTE, dynamic transmission error), in some cases, is not straightforward.
5. Brute force optimization
In this section the optimization is carried out using two parameters and linear prole modications. The parameter space is
uniformly spanned in order to get directly the optimal parameter set. This also allows checking the smoothness of the objective
functions.
5.1. Variation of the start tip radius
Start tip radius is varied for the pinion and the gear between 42 mm (64 mm for the gear) and the outside radius using 15
samples; tip amplitudes are left unchanged with respect to the standard case, 0.016 mm for the pinion and 0.018 mm for the gear
(Table 2).
The static optimization function FPPS is smooth (Fig. 7); however, it presents two local minima, and the minimum is indicated
by the yellow circle (the circle is present both in Figs. 7 and 8 even if it indicates the minimum of Fig. 7). The dynamic optimization
function FDTE is also regular and presents several minima, lying in a at region; the topology is not very complex and the

Fig. 7. Variation of start tip radius, FPPS; tip amplitude from Table 2. The circle indicates the minimum; the square indicates the minimum of Fig. 8 (FDTE).

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

551

Fig. 8. Variation of start tip radius, FDTE; tip amplitude from Table 2. The square indicates the minimum; the circle indicates the minimum of Fig.7 (FPPS).

minima are uniformly distributed in a horseshoe shaped valley; the dynamic optimum is indicated by an orange square both in
Figs. 7 and 8.
Both for static and dynamic optimizations, local methods are not reliable because they can lead to a local minimum; therefore,
the use of global optimization approaches is welcome and, a fortiori, it is almost mandatory for the dynamic optimization.
It is remarkable that static and dynamic optima do not coincide in the parameter plane (see Table 3 and Figs. 7 and 8); this
means that the classical static optimizations can lead to a system having a non-optimal dynamic behavior. Moreover, the
comparison between Figs. 7 and 8 shows that the static minimum corresponds to a point next to non negligible gradients in terms
of FPPS (Fig. 7); however, the situation is less critical in terms FDTE, i.e. perturbing the static optimum does not produce big
variations of FDTE (see Fig. 8, the yellow circle) that remains reasonably small.
Such analysis clearly proves that a global dynamic optimization is mandatory when the goal is the reduction of vibration.
5.2. Variation of the tip amplitude
The tip amplitude is varied both for the pinion and the gear between 0.003 mm and 0.030 mm, using 15 samples; tip radii are
unchanged with respect to the standard modications (Table 2).
After optimizing, one observes that, using the PPS optimization, the tip amplitude variation does not allow to reduce
the vibration with respect to the standard modications (see Table 4 and Fig. 9); however, it is effective with respect to the
pure involute prole. FPPS is very smooth (Fig. 9); the FDTE is less regular but it presents a very large at region in correspondence of
the minimum (Fig. 10); moreover, in this case a single minimum is present. The dynamic optimization leads to the best results.

Table 3
Variation of start tip radius, static and dynamic optimizations.

Pinion start tip radius [mm]


Gear start tip radius [mm]
FPPS [m]
FDTE [nondimensional]

Static optimization

Dynamic optimization

45.48
68.64
4.63
0.269

44.90
66.85
9.63
0.232

Table 4
Variation of tip amplitude, static and dynamic optimizations.
Optimal set

Static optimization

Dynamic optimization

Pinion start tip amplitude [mm]


Gear start tip amplitude [mm]
FPPS [m]
FDTE [nondimensional]

0.012
0.010
4.75
0.315

0.008
0.017
6.48
0.237

552

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

Fig. 9. Variation of tip amplitude, FPPS; tip radii from Table 2.

It is to be stressed that this result is valid for the specic limits of the parameter domain considered here.
6. RandomSimplex optimization
The second optimization strategy proposed in this paper is based on the combination of a Random search of the optimum,
followed by a renement carried out using the Simplex method [13]; the latter one is very robust and does not require derivative
evaluation of the objective function; however, it is not suitable for global optimization.
Let us consider the minimization of a nonlinear function of n variables F[x1 xj,, xn]. The rst step is to perform a Random
sampling using a uniform distribution, and the optimum is the seed for the Simplex method.
The Simplex method is an iterative algorithm based on the concept of a polyhedron of n + 1 vertices that is moved, iteration by
iteration, toward the optimal point. The algorithm starts with an initial polyhedron (P1,, Pj,, Pn) of points in the parameter
space, and each vertex is provided by the user or randomly chosen; then the centroid of the Simplex is evaluated as
1
c
j
xi = nj=1 xi ; the objective function is computed for every vertex; then, such vertices are ordered. The vertex with the worst
n
value of F is moved. Three types of motions are possible: reection (the vertex is reected toward the opposite face of the simplex),
expansion (the vertex is expanded toward the minimum), and contraction (the vertex is moved toward the simplex face; for
example, in Fig. 11 the point R would move toward the segment AD). The single iteration starts with a reection move; then, if the

Fig. 10. Variation of tip amplitude, FDTE; tip radii from Table 2.

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

553

A
R

C
B
D
Fig. 11. An iteration of the Simplex algorithm: reection.

new vertex is better than the old one, the new vertex is moved further with an expansion in the same direction; otherwise, it is
contracted. In Fig. 11 a simple reection move of the worst point B is represented; after reection, the new vertex R is accepted.
The algorithm stops when it cannot nd an improved solution or when the maximum number of iterations is reached.
In order to test the reliability of the optimum, the probability of failure has been estimated by a Monte Carlo simulation using
100 trial functions with a Gaussian distribution. The feasibility of a sample, in the neighborhood of the optimum, is evaluated
considering the objective function limit equal to 0.5; it means that the probability of failure is estimated by evaluating how many
cases lead to a vibration amplitude larger that 25% backlash (FDTE = 0.50). The ranges of manufacturing errors have been supposed
equal to 0.2 mm for the radius and 0.012 mm for the tip amplitude; in such ranges the Gaussian distribution simulates errors. Note
that the reliability analysis has been carried out perturbing the parameters used for the optimization; other disturbances, such as
misalignment, run out, spacing errors, prole errors and roughness are not considered as they are beyond the purposes of the
present work; see Ref. [8] for a deep analysis of misalignment in gear dynamics.

6.1. Four parameter optimization


Static and dynamic optimizations are now carried out using a random sampling followed by the Simplex method, in order to
nd the optimal parameters regarding tip relief of pinion and gear. The limits for the pinion tip radius are 42 mm and the outer
pinion radius; the limits for the gear tip radius are 64 mm and the outer gear radius. The limits for pinion and gear tip amplitudes
are 0.003 and 0.035 mm. The domain is sampled by means of a uniform pseudo Random generation using 180 sets of parameters.
Then the optimal set, obtained from the random sampling, is used as starting point for the Simplex method (80 iterations). Results
are shown in Table 5.
Fig. 12 shows the dynamic scenario for dynamically and statically optimized gears: in the case of dynamic optimization, Fig. 12a,
the vibration amplitude is reduced for all frequencies and loads, see Fig. 6 for comparison; in the case of the static optimization,
Fig. 12b, excellent results are obtained for load case 1 (33% nominal torque); conversely, for load case 3 (nominal torque) results are
not satisfactory, there is no improvement with respect to the standard prole modications and the improvement with respect to
the pure involute is small. In the case of dynamic optimization, Fig. 12a, the vibration reduction is about 65%, with respect to the
standard prole modications, the reduction is effective over a wide range of operating conditions (torque and rotational speed).
However, nonlinearities due to the contact loosing are still present. This is a direct consequence of the optimization strategy
proposed here: we are searching for a gear pair that has good performances over a wide range of operating conditions. Fig. 12a,b
clearly shows that there is not a direct correlation between the static optimum and the optimal vibration reduction.

Table 5
Four parameter optimizationRandomSimplex algorithm.

Pinion start tip radius [mm]


Gear start tip radius [mm]
Pinion tip amplitude [mm]
Gear tip amplitude [mm]
FPPS [m]
FDTE [nondimensional]
Probability of failure

Static optimization

Dynamic optimization

44.653
68.529
0.020
0.008
4.25
0.345
0.21

45.001
67.81
0.010
0.020
6.46
0.229
0.26

554

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

(a)

(b)
load case 1

load case 1

0.3

load case 2
load case 3

Dimensionless RMS(x)

Dimensionless RMS(x)

0.3
0.25
0.2
0.15
0.1
0.05
0.2

load case 2
load case 3

0.25
0.2
0.15
0.1
0.05

0.4

0.6

0.8

1.2

1.4

0.2

Dimensionless frequency

0.4

0.6

0.8

1.2

1.4

Dimensionless frequency

Fig. 12. Amplitude frequency diagrams: four parameter optimization. a) dynamic optimization, b) static optimization. Load cases 1, 2 and 3 (33%, 66% and 100% of
the nominal torque).

7. Macro-geometric optimization
It is well known in literature that High Contact Ratio (HCR) gears allow to obtain good results in terms of vibration behavior; in
particular, if the contact ratio (average number of teeth in contact) is an integer value (for HCR it means at least 2), the vibration is
reduced (see Ref. [12]). Our intent is to apply the RandomSimplex technique to a suitable HCR gear pair equivalent to the
original one, having a higher contact ratio with respect to the original gears. The geometry of the original gears is altered with the
following constraints: i) the same operating center distance; ii) the same face width; iii) variation of transmission ratio less than
1%; and iv) the same or smaller stresses. The selection consists of changing all other gear parameters (10 parameters), for example
the number of teeth. Each parameter is varied within reasonable limits; such a new parameter space is uniformly sampled (109
cases), for each sample the contact ratio, the contact pressure and the bending stresses are evaluated using standard formulae, in
order to select candidate gears having high contact ratio and reasonable stresses. However, standard formulations for evaluating
stresses are not accurate; therefore, a certain number of candidate gears are tested by means of a nite element analysis, in order
to select the best HCR gear pair from the strength point of view. The characteristics of the selected HCR gears are reported in
Table 6. The contact ratio increases from 1.54, Table 2 (standard gears), to 2.04, Table 6 HCR gears.
7.1. Four parameter optimization (HCR)
The new gear pair is now optimized with the RandomSimplex method; both linear and parabolic prole modications are
considered (see Ref. [9] for denition of parabolic modications).
The limits for the pinion tip radius are 43 mm and the outer pinion radius; the limits for the gear tip radius are 66 mm and the
outer gear radius. Limits for the pinion and the gear tip amplitudes are 0.003 and 0.030 mm, Table 7. 500 rotational speeds, 200
Random steps and 80 Simplex iterations are considered.
Fig. 13 shows that the HCR gear pair, without prole modications (pure involute), behaves quite well; after optimizing such
gear pair (tip relief) using linear prole modications, the vibration level is sensibly reduced.
The results are very good for all ranges of torque and velocity (see Table 8); moreover, resonances are linear, i.e. the teeth
contact loosing almost disappears (Fig. 13). It is important to note that the probability failure is zero, because the non-optimized

Table 6
Data of the HCR gears.
Case study

Pinion

Gear

n of teeth
Pitch diameter [mm]
Outside radius [mm]
Root radius [mm]
Prole shift [mm]
Hob tip radius [mm]
Shaft diameter [mm]
Normal module [mm]
Normal pressure angle []
Contact ratio

40
84
46.8
41
0.159
0.6
40
2.2
18
2.04

61
129
69
63
0.766
0.6
40

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

555

Table 7
Input data for four parameter optimization, RandomSimplex algorithm. Domain limits (Tip relief).
Parameter

Pinion

Gear

Minimum amplitude [mm]


Maximum amplitude [mm]
Minimum radius [mm]
Maximum radius [mm]

0.003
0.03
43
Addendum radius

0.003
0.03
66
Addendum radius

(a)

load case 1
load case 2
load case 3

0.1

Dimensionless RMS(x)

0.16

Dimensionless RMS(x)

(b) 0.12

load case 1
load case 2
load case 3

0.18

0.14
0.12
0.1
0.08
0.06

0.08

0.06

0.04

0.04
0.02

0.02
0.2

0.4

0.6

0.8

1.2

1.4

0.2

0.4

Dimensionless frequency

0.6

0.8

1.2

1.4

Dimensionless frequency

Fig. 13. Amplitude frequency diagram, HCR gears. Load cases 1, 2 and 3 (33%, 66% and 100% of the nominal torque); a) no prole modication, FDTE = 0.193,
FPPS = 4.79 b) Four parameter dynamic optimization, FDTE = 0.121, FPPS = 2.68.

HCR gear pair does not undergo high vibration amplitude; therefore, none of the samples, distributed around the optimum,
reaches a maximum RMS larger than 0.5.
The parabolic tip relief behaves better than the linear one when the static optimization is carried out, see Table 8. No data is
available yet for parabolic modications and dynamic optimization; this analysis is beyond the purposes of the present work.
By checking Table 8 it is clear that the dynamic optimization is the best strategy.
A static tooth contact analysis of the optimized HCR gears has been carried out at the nominal torque; the maximum contact
pressure is about 1400 MPa, which is sufciently low to guarantee no pitting problems.
7.2. Eight parameter dynamic optimization
Tip and root amplitudes of prole modications have been varied from 0.003 mm to 0.03 mm, the start radius of the root relief
is varied from the active involute start radius up to 42.9 mm for pinion (65.9 mm for the gear), the tip start radii are the same of
Section 7.1. 300 random steps and 80 Simplex iterations have been considered.
In Table 9 a summary of the present results is reported: using the RandomSimplex approach allows to reduce the
computational cost and to attack higher dimensional parameter spaces. The use of eight parameters requires more steps than the
case of four parameters; however, the computational cost is still reasonable.
Results obtained using eight parameter optimization (tip and root reliefs) are similar to those obtained considering a four
parameter space (tip relief only).
Table 8
Results for four parameter optimizationRandomSimplex algorithm.
Case A

Pinion start tip radius [mm]


Gear start tip radius [mm]
Pinion tip amplitude [mm]
Gear tip amplitude [mm]
FPPS [m]
FDTE [nondimensional]

Case A

Case B

Dynamic opt.

Static opt.

Static opt.

Linear prole

Linear prole

Parabolic prole

43.96
68.42
0.005
0.018
2.68
0.121

44.00
68.39
0.010
0.01
2.08
0.202

43.32
66.45
0.019
0.015
1.82
0.197

556

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

Table 9
Dynamic optimization using eight parameters.

Pinion start tip radius [mm]


Gear start tip radius [mm]
Pinion tip amplitude [mm]
Gear tip amplitude [mm]
Pinion root amplitude [mm]
Gear root amplitude [mm]
Pinion end root radius [mm]
Gear end root radius [mm]
FPPS [m]
FDTE [nondimensional]
Probability of failure

Minimum

Maximum

Optimal

43
66
0.003
0.003
0.003
0.003
41.1
63.1

Addendum radius
Addendum radius
0.030
0.030
0.030
0.030
42.9
65.9

46.36
68.57
0.004
0.01
0.022
0.019
41.74
64.16
2.49
0.129
0.59

load case 1
load case 2
load case 3

Dimensionless RMS(x)

0.12

0.1

0.08

0.06

0.04

0.02

0.2

0.4

0.6

0.8

1.2

1.4

Dimensionless frequency
Fig. 14. Amplitude frequency diagramoptimized HCR gears: eight parameter RandomSimplex optimization. Load cases 1, 2 and 3 (33%, 66% and 100% of the
nominal torque).

Fig. 14 shows the dynamic scenario for the dynamically optimized gears. The dynamic behavior is similar to the gear pair
optimized using four parameters (Fig. 13b). The vibration amplitude is smaller than the standard case, see Fig. 6a.
8. Conclusions
A global optimization method, based on the Random plus Simplex approach, has been developed in order to suggest the best
geometry of spur gears versus vibration reduction; tip and root relief parameters are considered for optimizing the gears. The
optimization strategy considers the whole dynamic scenario, i.e. the vibration level at different rotational speeds and different
torques. This allows to nd an optimized gear pair having good dynamic behavior over a wide range of operating conditions.
The effectiveness of the dynamic optimization strategy is proved by means of comparisons with the classical methods based on
the peak to peak of the transmission error. The dynamic optimization produces better results than the static optimization; it
increases the computational cost of 15% only. The approach includes a reliability analysis, which is limited to the parameter space;
it allows to evaluate the robustness of the optimum when design parameters cannot be controlled with extreme precision.
The global optimization approach has been nally applied to an HCR gear pair, which is equivalent to the initial gear pair.
Coupling macro and micro geometric optimizations allows to reduce dramatically the vibration level for all operating conditions.
An application to an actual industrial test case shows that using both root and tip reliefs does not produce better results with
respect to the use of tip relief only; such comment applies to the present test case, but it should not be considered as a general rule.
However, the optimization approach presented here works well up to eight parameters, obtaining in any case reasonable results.
Acknowledgment
This paper is dedicated to the memory of Marcello Faggioni, who was the team leader of the present research and died
prematurely before the publication of this work. The authors would like to thank the Lab SIMECH/INTERMECH (HIMECH District,
Emilia Romagna Region) for supporting the research.

M. Faggioni et al. / Mechanism and Machine Theory 46 (2011) 544557

557

References
[1] H.N. zgven, D.R. Houser, Mathematical models used in gear dynamicsa review, Journal of Sound and Vibration 121 (1988) 383411.
[2] S.L. Harris, Dynamic loads on the teeth of spur gears, Proceedings of the Institution of Mechanical Engineers 172 (2) (1958) 87100.
[3] T. Sato, K. Umezawa, J. Ishikawa, Effects of contact ratio and prole correction of spur gears on the rotational vibrations, Bulletin of the JSME 26 (221) (1983)
20102016.
[4] M.S. Tavakoli, D.R. Houser, Optimum prole modications for the minimization of static transmission errors of spur gears, Journal of Mechanism
Transmissions and Automation in Design 108 (1986) 8695.
[5] V. Simon, Optimal tooth modications for spur and helical gears, Journal of Mechanism, Transmissions, and Automation in Design 111 (1989) 611615.
[6] R.G. Munro, N. Yildirim, D.M. Hall, Optimum prole relief and transmission error in spur gears, Proceedings of Institution of Mechanical Engineers C 404/013
(1990) 3542.
[7] Y. Cai, T. Hayashi, The optimum modication of tooth prole for a pair of spur gears to make its rotational vibration equal to zero, ASME Proceedings of Power
Transmission and Gearing Conference DE-vol. 432 (1992) 453460.
[8] P. Velex, M. Maatar, A mathematical model for analyzing the inuence of shape deviations and mounting errors on gear dynamic behaviour, Journal of Sound
and Vibration 191 (5) (1996) 629660.
[9] H.H. Lin, F.B. Oswald, D.P. Townsend, Dynamic loading of spur gears with linear or parabolic tooth prole modications, Mechanism and Machine Theory 29
(8) (1989) 11151129.
[10] F.L. Litvin, D. Vecchiato, K. Yukishima, A. Fuentes, I. Gonzalez-Perez, K. Hayasaka, Reduction of noise of loaded and unloaded misaligned gear drives, Computer
Methods in Applied Mechanics and Engineering 195 (2006) 55235536.
[11] D.J. Fonseca, S. Shishoo, T.C. Lim, D.S. Chen, A genetic algorithm approach to minimize transmission error of automotive spur gears sets, Applied Articial
Intelligence 19 (2) (2005) 153179.
[12] A. Kahraman, G.W. Blankenship, Effect of involute contact ratio on spur gear dynamics, ASME Journal of Mechanical Design 121 (1999) 112118.
[13] J.A. Nelder, R. Mead, A Simplex method for function minimization, Computer Journal 7 (1965) 308313.
[14] V.K. Tamminana, A. Kahraman, S. Vijayakar, A study of the relationship between the dynamic factors and the dynamic transmission error of spur gear pairs,
Journal of Mechanical Design 129 (2007) 7584.
[15] A. Kahraman, G.W. Blankenship, Experiments on nonlinear dynamic behaviour of an oscillator with clearance and periodically time-varying parameters,
Journal of Applied Mechanics 64 (1997) 217226.
[16] G.W. Blankenship, A. Kahraman, Steady state forced response of a mechanical oscillator with combined parametric excitation and clearance type nonlinearity, Journal of Sound and Vibration 185 (5) (1995) 743765.
[17] R.G. Parker, S.M. Vijayakar, T. Imajo, Nonlinear dynamic response of a spur gear pair: modeling and experimental comparisons, Journal of Sound and
Vibration 237 (2000) 435455.
[18] A.H. Nayfeh, B. Balachandran, Applied Nonlinear Dynamics Analytical, Computational, and Experimental Methods, John Wiley and Sons Inc., New York, 1995.
[19] G. Bonori, M. Barbieri, F. Pellicano, Optimum prole modications of spur gears by means of genetic algorithms, Journal of Sound and Vibration 313 (2008)
603616.

Vous aimerez peut-être aussi