Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

ARTICLE
Gear-tolerance optimization based on a response
surface method
Rui Xu, Kang Huang, Jun Guo, Lei Yang, Mingming Qiu, and Yan Ru
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

Abstract: To address the low efficiency of gear-tolerance analysis and optimization, a gear-tolerance optimization
method based on a response surface method (RSM) and optimization algorithm is presented. A gear-tolerance
mathematical model, including profile deviation, pitch deviation, and geometric deviation, was developed by com-
bining traditional profile modeling with a small displacement torsor (SDT) method. Based on this mathematical
model, a tooth-contact analysis method, which takes a variety of deviations into account, and a program to com-
pute transmission error were developed. Using the RSM and a genetic algorithm, a gear-tolerance optimization
Transactions of the Canadian Society for Mechanical Engineering

model was created to consider a variety of gear tolerances as design variables and process cost as an optimization
objective. An example of gear-tolerance optimization was analyzed, and the result indicates that the method pre-
sented in this paper may help improve the efficiency of gear-tolerance optimization and is practicable for preci-
sion gear design.
Key words: gears, tolerance modeling, small displacement torsor (SDT), response surface method, tolerance
optimization.
Résumé : Pour répondre à la faible efficacité de l’analyse et de l’optimisation de la tolérance des engrenages, une
méthode d’optimisation de la tolérance des engrenages basée sur la méthode de surface de réponse (MSR) et un
algorithme d’optimisation est présentée. Le modèle mathématique de tolérance d’engrenage incluant la
déviation de profil, la déviation de pas et l’écart géométrique a été développé en combinant la méthode de
modélisation traditionnelle avec un petit Traitement spécial et différencié (méthode TSD). Basé sur ce modèle
mathématique, la méthode d’analyse de contact de dent qui tient compte de divers écarts est discutée et le pro-
gramme informatique pour calculer l’erreur de transmission a été développé. En utilisant le MSR et un algorithme
génétique, un modèle d’optimisation de tolérance d’engrenage a été construit en considérant une variété de
tolérances d’engrenage comme variables de conception et le coût du processus comme objectif d’une optimisa-
tion. Un exemple d’optimisation de la tolérance d’engrenage a été analysé et le résultat indique que la méthode
présentée dans cet article peut aider à améliorer l’efficacité de l’optimisation de la tolérance d’engrenage et qu’elle
est réalisable pour la précision de conception d’engrenages. [Traduit par la Rédaction]
Mots-clés : engrenages, modélisation de la tolérance, traitement spécial et différencié (TSD), méthode de surface de
réponse, optimisation de la tolérance.

1. Introduction Tolerancing research has focused on such aspects as

Tolerances are an issue in almost all aspects of product tolerance modeling, tolerance analysis, and tolerance
design and manufacturing — affecting not only manu- optimization. Requicha (1983) studied the tolerancing of
facturing and assembling processes but also the quality, solid models and proposed a mathematical theory of tol-
functionality, production efficiency, and manufacturing erancing that allows tolerances to be incorporated into
cost of the products. With requirements for product per- geometric modeling systems. Turner and Wozny (1987)
formance and manufacturing cost becoming increas- presented a solid-modeling technology to automate solv-
ingly stringent, the problem of how to manufacture ing tolerance analysis problems. Desrochers and
high-quality products at lower costs has been receiving Clément (1994) developed a representation model called
increased attention. “technologically and topologically related surfaces”

Received 14 January 2018. Accepted 13 April 2018.

R. Xu, K. Huang, J. Guo, L. Yang, and M. Qiu. School of Mechanical Engineering, Hefei University of Technology, Hefei 230009, China.
Y. Ru. P.L.A. Army Academy of Artillery and Air Defense, Hefei 230009, China.
Corresponding author: Kang Huang (email: [email protected]).
Copyright remains with the author(s) or their institution(s). Permission for reuse (free in most cases) can be obtained from RightsLink.

Trans. Can. Soc. Mech. Eng. 00: 1–14 (0000) dx.doi.org/10.1139/tcsme-2018-0006 Published at www.nrcresearchpress.com/tcsme on 30 July 2018.
 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

2 Trans. Can. Soc. Mech. Eng. Vol. 00, 0000

(TTRS) based on an offset tolerance zone model and introducing a response surface method (RSM) to estab-
applied it to dimensioning and tolerancing. Whitney lish the relationship between deviation parameters and
et al. (1994) studied a matrix-based tolerance representa- response goals. First, a tolerance model for a gear is
tion model using an odd matrix transformation to established by combining the traditional modeling
describe the spatial geometric relations, and Desrochers method with the SDT method. Second, a response
and Rivière (1997) applied this approach to three- surface model to describe the relationship between
dimensional tolerance analysis. Gao et al. (1999) studied deviation parameters and transmission performance is
three-dimensional tolerance analysis based on a vector- built using the RSM and Latin hypercube sampling.
loop model, considering the source of variation in Finally, based on the RSM and a genetic algorithm, a
assembly. Bourdet et al. (1996) introduced the small dis- gear-tolerance optimization model is developed by con-
placement torsor (SDT) to the field of tolerance research, sidering a variety of gear tolerances as design variables
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

based on a rigid-body hypothesis and a small-displacement and machining cost as the optimization objective, and
hypothesis, and scholars applied this approach to an example is discussed.
engineering practice (Asante 2009; Guo et al. 2013).
Mujezinović et al. (2004) presented a new mathematical 2. Modeling method for gear tolerance
model called a tolerance map to represent geometric toler- In this section, based on the nominal mathematical
ances, which is compatible with the ASME/ISO standards
model of a tooth flank, a tolerance model of a gear that
Transactions of the Canadian Society for Mechanical Engineering

for geometric tolerances.

mainly refers to three types of deviations — profile
Up to now, these models have mainly been used for deviation, pitch deviation, and geometric deviation —
assembly-oriented tolerance analysis — i.e., analyzing
is developed.
the effects of geometric tolerances on assemblability
(Schleich and Wartzack 2014). However, with an increas- 2.1. Modeling method for nominal tooth flank
ing demand for product performance, the effect of For a pair of involute helical gears in mesh with z(1) and
(2)
tolerances on the functional behavior of products is z teeth, the deviation models vary from one tooth flank
attracting more attention. Stuppy and Meerkamm to another. For convenience, this paper identifies each
(2009) proposed a comprehensive tolerance-analysis tooth flank as follows: (i) the superscript k describes pinion
method for kinematic mechanisms, which took into 1 or wheel 2, and (ii) the subscript i or j describes the ith
account the influence of deviations caused by machining tooth flank of pinion 1 or the jth tooth flank of wheel 2.
and motion on the functional characteristics of the Take pinion 1, for example, to model a nominal
ð1Þ
mechanism. Walter and Wartzack (2013) and Walter tooth flank. In the Cartesian coordinate system S1
et al. (2013) illustrated a tolerance-analysis method for (o1 – x1, y1, z1) in Fig. 1, the nominal tooth flank model of
kinematic mechanisms considering different geometric pinion 1 can be represented as follows:
deviations and their interactions, and the results can be !


ð1Þ ð1Þ ð1Þ
used to optimize kinematic mechanisms. These methods r0ðiÞ φðiÞ ,λðiÞ
can be applied to the general mechanisms with both 8 .
>
> ð1Þ ð1Þ ð1Þ ð1Þ
lower and higher kinematic pairs, which do not require > x0ðiÞ = −r λðiÞ tanβ
>
>
> 
the determination of contact positions between parts, >
>

> ð1Þ
> ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ
but are not suitable for gear mechanisms, in which tooth >
> y = r cosλ sinφ − φ cosφ
>
> 0ðiÞ b ðiÞ ðiÞ ðiÞ ðiÞ
>
>
contact must be considered during meshing. Little analy- ð1Þ >
<

ð1Þ ð1Þ ð1Þ ð1Þ
sis and optimization research of gear tolerance exists in = + sinλ ðiÞ cosφ ðiÞ + φ ðiÞ sinφ ðiÞ
the literature. Watrin et al. (2013) presented a methodol- >
> 
>
>

ogy to specify the tolerance of a bevel gear, but they gave >
> ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ
>
> z = r cosφ cosφ + φ sinφ
>
> 0ðiÞ b ðiÞ ðiÞ ðiÞ ðiÞ
no details about tolerance analysis and tolerance speci- >
>
>
>

fication. Schleich and Wartzack (2014) proposed an >
> ð1Þ ð1Þ ð1Þ ð1Þ
>
: − sinλ ðiÞ sinφ ðiÞ − φ ðiÞ cosφ ðiÞ
approach for analyzing mechanisms respecting various
sources of geometric deviations based on discrete geom-
ð1Þ
etry, which can be applied to gears. where r(1) and rb are the radii of the reference circle and
To analyze the effect of tolerance on the functional base circle of pinion 1, respectively; β (1) is the helix
ð1Þ
behavior of gears, tooth contact analysis (TCA), which angle; λðiÞ is the rotation angle of pinion 1 along the
depends on the solution of a nonlinear system of equa- reference helix, the range of which is [–B tan β (1) /2r(1),
ð1Þ
tions, must be performed. As the number of deviation B tan β(1)/2r(1)]; B is the tooth width; and φðiÞ is the roll
parameters introduced into the nonlinear system of angle of the ith tooth flank of pinion 1.
equations increases, finding solutions becomes more To improve the meshing performance, the tooth
complex and computationally time-consuming, leading form is usually modified by crowning. In this paper,
to low efficiency in gear-tolerance analysis and optimiza- only longitudinal crowning is considered. Figure 2
tion. To solve this problem, this paper proposes a high- shows that for a spur gear, longitudinal crowning can
efficiency approach to optimize gear tolerance by be defined by the radius of longitudinal crowning (Rg),
Published by NRC Research Press
 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

Xu et al. 3

Fig. 1. Nominal tooth flank model of pinion 1. [Colour online.]

(1)
p'

p y1
(1)
B
z1
(1)
x1 r (1) (i)
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

n p p'

n
Transactions of the Canadian Society for Mechanical Engineering

(1)
(1) (i)
r

(1)
(i) rb(1)
z1

o1 y
1

z1
x1

y1

x1

(
Fig. 2. Parameters defining longitudinal crowning. [Colour Rg = Δl=2 þ b2 =ð8ΔlÞ
ð2Þ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
online.] 2 2
lp = Rg − Rg − ðxp Þ

So the mathematical model of a tooth flank of a

helical gear, considering longitudinal crowning, is
described as
!


0 ð1Þ ð1Þ ð1Þ
b/2 r0ðiÞ φðiÞ ,λðiÞ
2 3
1 2 3
6  0  0 0
7 xð1Þ
6 0 cos lp lp
07 6 7
0ðiÞ
ð3Þ 6 ð1Þ ð1Þ
sin ð1Þ ð1Þ 7 6 ð1Þ 7
6 r cos β r cos β 7 6y 7
=6 6   7 · 6 0ðiÞ 7
7 6 ð1Þ 7
6 0 −sin lp lp
6 cos 07 6
7 4 z0ðiÞ 5
7
4 r cos βð1Þ
ð1Þ
rð1Þ cos βð1Þ 5
Rg 1
xp 0 0 0 1

Then the unit normal vector of the tooth flank is

p ! !
0 ð1Þ 0 ð1Þ
∂r0ðiÞ ∂r0ðiÞ
lp
ð1Þ
× ð1Þ
! ∂λðiÞ ∂φðiÞ
0 ð1Þ
ð4Þ n0ðiÞ = ! !
0 ð1Þ 0 ð1Þ
Δl ∂r0ðiÞ ∂r0ðiÞ
ð1Þ
× ð1Þ
∂λðiÞ ∂φðiÞ
the magnitude of crowning (lp), and longitudinal coor-
dinate value (xp) at point p on the tooth profile, which Figure 3 shows the tooth flank with longitudinal
is generally expressed as crowning.

Published by NRC Research Press

 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

4 Trans. Can. Soc. Mech. Eng. Vol. 00, 0000

Fig. 3. Tooth flank with longitudinal crowning. Thus, the rotational angle θ i from S 1(i) to S 2 around
[Colour online.] axis o2 is

2π F pi
(7) θi = ði − 1Þ · þ i ∈ f2, · · · , z1 g
z1 dM

where dM is the diameter of the measurement circle.

The transformation matrix from S1(i) to S2 is
2 3
1 0 0 0
6 0 cos θi − sin θi 0 7
ð8Þ R 12ðiÞ = 6
4 0 sin θi cos θi 0 5
7
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

0 0 0 1

2.4. Modeling method for geometric deviation based

on SDT
2.4.1. Tolerance modeling method based on SDT
Transactions of the Canadian Society for Mechanical Engineering

The SDT, developed by Bourdet et al. (1996) based on a

rigid-body hypothesis and a small-displacement hypoth-
2.2. Modeling method for profile deviation
esis, is used to represent geometric variations. The SDT
According to gear precision standard ISO 1328.1 2013, can be expressed as
the total profile deviation consists of profile form
2 3
deviation and profile slope deviation, as shown in Fig. 4. θ x dx
According to Mucchi et al. (2010), the profile variation (9) ½θ d = 4 θy dy 5
model can be described as θ z dz

eα ðsÞ = ef α ðsÞ þ eHα ðsÞ where θx, θy, and θz are three components of the rotational
 vector θ around the x, y, and z axes in the local reference
(5) s − s0 f f α s − s0
= f Hα þ sin 2πf r ðs0 ≤ s ≤ sf Þ system, respectively; dx, dy, and dz are three components
sf − s0 2 sf − s0
of the translational vector d along the x, y, and z axes,
respectively. All of them are called “screw parameters”.
where ffα is the profile form deviation and fHα the profile To facilitate the matrix calculation, the SDT is gener-
slope deviation; fr represents the number of sine periods ally written as
occurring in the evaluation band; s is the involute rolling 2 3
path length over the profile evaluation band, which is a 1 −θz θy dx
function of the roll angle φ, i.e., s(r) = rb tan φ, as shown 6 θz 1 θ x dy 7
(10) T=6 4 −θy θx
7
in Fig. 1; and s0 and sf are coordinate s evaluated at the 1 dz 5
band’s lower and higher limit, respectively. 0 0 0 1

2.3. Modeling method for pitch deviation

According to gear precision standard ISO 1328.1 2.4.2. Modeling method for geometric deviations
2013, pitch deviation includes individual single-pitch Geometric deviations caused by machining and instal-
deviation fpi, single-pitch deviation fp, individual cumula- lation lead to misalignment between the axis of a gear
tive pitch deviation F pi , and total cumulative pitch ring and the shaft, which creates an eccentric error in
deviation Fp. Among these deviations, individual single- the gear-transmission system, thus producing noise. So
pitch deviation fpi is the most basic. That is, the other mathematical models of these geometric deviations
pitch deviations can be easily obtained when individual must be developed to analyze their effects on transmis-
single-pitch deviation is given. sion performance. The geometric deviations in this
In Fig. 5, tooth flank 1 is the initial tooth flank used to paper mainly include deviation between the gear ring
measure pitch deviation, and the coordinate systems S2 and the gear hole, deviation between the gear hole and
and S 1(i) are fixed on tooth flank 1 and tooth flank i, the shaft, and installation deviations of the shaft.
respectively. If the ith individual single-pitch deviation
f pi has been measured, the ith individual cumulative 1. Modeling deviation between gear ring and gear
pitch deviation can be calculated as hole
In Fig. 6, a deviation exists between the gear ring
X
i axis and the gear hole axis, owing to manufacturing
ð6Þ F pi = f pk error of that gear ring. To ensure that the deviation
k=1 remains within an allowable range, a coaxial
Published by NRC Research Press
 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

Xu et al. 5

Fig. 4. Total profile deviation: (a) profile form deviation and (b) profile slope deviation.
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

(a) (b)

Fig. 5. Coordinate system transforming between the ith Fig. 7. Deviation between the gear hole and shaft.
tooth flank and the initial tooth flank. z3 z4
ptM
Transactions of the Canadian Society for Mechanical Engineering

fpi

Gear hole
kptM Shaft
dM

y3 ( y4)
o3 o4
Fpi

z2
z1
y1

y2
o1 (o2 )

2 3
1 −θz, 23 θy, 23 0
Fig. 6. Deviation between gear ring and gear hole. 6 θz, 23 1 0 dy, 23 7
ð11Þ T23 = 6
4 −θy, 23
7
z3 z2 θx, 23 1 dz, 23 5
0 0 0 1
Reference circle
where θ y, 23, θ z, 23, d y, 23, and d z, 23 are the screw
Gear hole parameters of the axis of the gear ring relative to
the gear-hole axis.
2. Modeling deviation between gear hole and shaft
y2 ( y3)
In Fig. 7, when the gears are mounted on the
o3 o2 shaft, there is a deviation between the gear hole
axis and the shaft axis, owing to manufacturing
error of the shaft. The deviation is restrained within
a coaxial tolerance, and the SDT of the axis of the
gear hole can be expressed as
2 3
1 −θz, 34 θy, 34 0
6 θz, 34 1 0 dy, 34 7
ð12Þ T34 = 6
4 −θy, 34 θx, 34
7
1 dz, 34 5
0 0 0 1
tolerance is used to restrain the variation of the
axis of the gear ring. According to the previous sec- where θ y, 34 , θ z, 34 , d y, 34 , and d z, 34 are the screw
tion, the SDT of the axis of the gear ring can be parameters of the axis of the gear hole relative to
expressed as the shaft axis.

Published by NRC Research Press

 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

6 Trans. Can. Soc. Mech. Eng. Vol. 00, 0000


Fig. 8. Installation deviation of shaft. ð1Þ ð1Þ
R ϕ
2 3
Center distance tolerance 1
z 4 z5 h 0 i h 0 i 0
6 ð1Þ ð1Þ ð1Þ 7
ð16Þ 6 0 cos ϕ − ϕ0 − sin ϕð1Þ − ϕ0 07
6 h i h i 7
=6 7
6 0 sin ϕð1Þ − ϕð1Þ cos ϕ
ð1Þ
− ϕ
ð1Þ
0 7
Vertical plane 4 0 0 5
o4( o5) y5 0 0 0 1
y4


ð2Þ ð2Þ
f

R ϕ
2 3
x5 x4 1 h 0 i h 0 i 0
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

6 ð2Þ ð2Þ ð2Þ ð2Þ 7

f ð17Þ 6 0 cos ϕ − ϕ0 − sin ϕ − ϕ0 07
Plane of axes 6 h i h i 7
=6 7
6 0 sin ϕð2Þ − ϕð2Þ ð2Þ
cos ϕ − ϕ0
ð2Þ
07
4 0 5
0 0 0 1
3. Modeling installation deviations of the shaft
where ϕ(1) and ϕ(2) are the rotation angles of the pinion
In accord with ISO/TR 10064.3 1996, the center ð1Þ ð2Þ
Transactions of the Canadian Society for Mechanical Engineering

and wheel, respectively; and ϕ0 and ϕ0 are the initial

distance deviation and parallelism deviation of the
angle of the pinion and wheel, respectively.
shaft are shown in Fig. 8. Axes 4 and 5 are the actual
To analyze the effect of gear tolerance on transmission
axis and ideal axis of the pinion shaft, respectively;
performance, the variation range of the deviation
axis 5′ is the ideal axis of the wheel shaft. To ensure
that the deviation falls within an allowable range, parameters in eqs. (14) and (15) must be determined.
both the center distance deviation and parallelism Getting the range of parameters in profile and pitch
deviation are used to restrain the variation of the deviation according to established standards of gear
axis of the pinion shaft, the SDT of which can be accuracy is easy; in contrast, the variation range of the
described as geometric deviation parameters, variation inequality,
and constraint inequality must be defined.
2 3
1 −θz, 45 θy, 45 0
6 θz, 45 1 0 dy, 45 7
ð13Þ T45 = 6
4 −θy, 45
7 3. TCA considering deviations
0 1 dz, 45 5
0 0 0 1 The aim of TCA is to obtain the contact path, orienta-
tion, transmission error, etc. This paper mainly evaluates
where θy,45, θz,45, dy,45, and dz,45 are the screw param- the effect of deviations discussed earlier on transmission
eters of the axis of the pinion shaft relative to the error by TCA to pave the way for a response surface
axis of the wheel shaft. model to determine deviation parameters and transmis-
sion error.
2.5. Mathematical model for gear tooth tolerance According to gear geometry, and applying the theory
Based on the deviation models discussed earlier, the of Litvin and Fuentes (2004), tooth flank 1 and tooth
mathematical model for gear tooth tolerance can be flank 2 are continuously tangential, and the contact
obtained by adding the rotation angle. point (as shown in Fig. 9) can be determined using the
The mathematical tolerance model for pinion 1 is following equations:


!  8
ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ > 

!  !


rf ðiÞ φðiÞ , λðiÞ , ϕ = T45 · Rðϕ Þ · T34 >
< rð1Þ φð1Þ , λð1Þ , ϕð1Þ − rð2Þ φð2Þ , λð2Þ , ϕð2Þ = 0
ð14Þ
! ð18Þ
f ðiÞ ðiÞ ðiÞ f ðjÞ ðjÞ ðjÞ
ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ ð1Þ !0  > 

!  !


· T23 · T12ðiÞ · r0ðiÞ ðφðiÞ , λðiÞ Þ + eαðiÞ ðφðiÞ Þ · n0 >
: nfð1Þ φ
ð1Þ ð1Þ
, λ , ϕ
ð1Þ
− n
ð2Þ
φ
ð2Þ ð2Þ
, λ , ϕ
ð2Þ
=0
ðiÞ ðiÞ ðiÞ f ðjÞ ðjÞ ðjÞ
Similarly, the mathematical tolerance model for
wheel 2 is 
! 
! 
! !
ð1Þ ð1Þ ð2Þ ð2Þ
∂rf ðiÞ ∂rf ðiÞ ∂rf ðjÞ ∂rf ðjÞ
!


 ð1Þ
× ð1Þ ð2Þ ð2Þ
×
ð2Þ ð2Þ ð2Þ
rf ðjÞ φðjÞ , λðjÞ , ϕ
ð2Þ
=R ϕ
ð2Þ ð2Þ
· T34 ! ∂λðiÞ ∂φðiÞ ! ∂λðjÞ ∂φðjÞ
ð1Þ ð2Þ
ð15Þ with nf ðiÞ = 
! 
! and n f ðiÞ = ! ! , the


! 
 ! ð1Þ
∂rf ðiÞ
ð1Þ
∂rf ðiÞ
ð2Þ ð2Þ
∂rf ðjÞ ∂rf ðjÞ
ð2Þ ð2Þ ð2Þ ð2Þ ð2Þ ð2Þ ð2Þ 0
· T23 · T12ðjÞ · r0ðjÞ ðφðjÞ , λðjÞ + eαðjÞ φðjÞ · n0 × ×
ð1Þ ð1Þ ð2Þ ð2Þ
∂λðiÞ ∂φðiÞ ∂λðjÞ ∂φðjÞ
In eqs. (14) and (15), R(ϕ(1)) and R(ϕ(2)) are the rotation surfaces’ unit normal vector.
matrices of the pinion and the wheel, respectively, Then the transmission error can be obtained using the
which can be described as follows: following equation:

Published by NRC Research Press

 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

Xu et al. 7

Fig. 9. Meshing relationship of tooth flanks 1 and 2. Fig. 10. Flow chart of TCA program. [Colour online.]

n1((1)i ) ( (1)
, (1)
, (1)
) n1((2)j ) ( (2)
, (2)
, (2)
) (k )
(i ) (i ) ( j) ( j) Input basic parameters of gear and 0
Let i ( j ) 1
T

( j)
M Assign each deviation parameter.
(1)
(2) Let the starting rotation angle value of pinion 1 ( i )start =0
(1)
and solve
(1)
the minimum and maximum value of rotation angle ( i ) min and ( i ) max.
r1((1)i ) ( (1)
(i ) , (1)
(i) , (1)
) r1((2)j ) ( (2)
( j) , (2)
( j) , (2)
) Assign number of contact points N1 and N 2 and Calculate the
interval angle: 1(1) = (1)
– (1) N1
(i ) ( i ) max ( i ) start
z1
2(1)
(i ) =
(1)
– (1)
/ N2
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

( i ) start ( i ) min
(i )
(1) (1)
(1) Let ( i ) = ( i )start
y1
o1
x1
Let (1)
= (1)
(i ) (i 1)2 π / z (1)

Input (1) and solve eq. (18) in Matlab.

Transactions of the Canadian Society for Mechanical Engineering

Calculate the TE with eq. (19) and store the result.


 ð1Þ
ð2Þ ð1Þ z
ð19Þ ε = ϕð2Þ − ϕ0 − ϕð1Þ − ϕ0 ð2Þ
z (1)
(i ) = (1)
( i )start + 1((1)i ) (1)
(i ) = (1)
( i )start 2((1)i )
ð1Þ ð2Þ
where ϕ0 and ϕ0 are the initial angles of pinion 1 and
wheel 2, respectively; and z(1) and z(2) are the tooth num- Yes
Yes No 1(1) (1)
bers of pinion 1 and wheel 2, respectively. 1(1) (1)
( i ) max ( i ) max

With deviations, the nonlinear eq. (18) becomes very No

i =i 1
complex. To avoid the complexity and improve the effi-
ciency of the solving process, a solution is programmed
in Matlab. A flow chart of this program is shown in
Yes i tooth numeber
Fig. 10.
To verify the accuracy of the mathematical tolerance (or set value )
model and the TCA program, two comparative cases are
No
illustrated here. Drawn from Yoshikawa et al. (1997) and
Sanchez-Marin et al. (2016), the basic parameters of the Draw the TE curves by calling the
relevant function in Matlab.
pinion and wheel in two cases are shown in Table 1.
The first case compares the simulation result and the
experimental result of Yoshikawa et al. (1997), who
Table 1. Basic parameters for cases 1 and 2.
reported alignment errors with rotation angle γ around
the y axis and rotation angle δ around the z axis, which Case 1 Case 2
correspond to screw parameters θy, 45 and θz, 45, respec-
Basic parameter Pinion Wheel Pinion Wheel
tively, between input shaft and output shaft. The profile
deviation can be also obtained from the measured data Module (mm) 2.03 2.03 5 5
of the tooth flank. We simulate the transmission error Number of teeth 52 52 20 34
based on these parameters using the TCA program and Press angle (°) 18 18 25 25
Helix angle (°) 28.5 28.5 15 15
compare the simulation result with the experimental
Tooth width (mm) 20 20 60 50
result in Yoshikawa et al. (1997). As Fig. 11 shows, the sim- Center distance (mm) 120 138.96
ulation result is very similar to the experimental result
despite the difference between them due to deviations
not considered in the simulation. The comparison, as shown in Fig. 12, shows that the
The second case compares our simulation results and TCA simulation result is almost the same as the data in
data from Sanchez-Marin et al. (2016). In this case, the Sanchez-Marin et al. (2016). The difference between them
center distance has been increased 0.1 mm, and the is mainly caused by a different choice of starting angle of
wheel has been misaligned with respect to the pinion the double-tooth meshing area, which has little effect on
with a rotation angle of +0.02° around the z axis. the transmission error.
(In Sanchez-Marin et al. (2016), the y axis is the rotation These analyses show that the established mathemati-
axis of the wheel). For comparison, we convert both the cal model of gear tolerance is reasonable. Therefore,
horizontal axis unit (degree) and the vertical axis using this model, we can simulate the variation of the
unit (arcsec) in Sanchez-Marin et al. (2016) to radians. transmission error within required tolerance constraints
Published by NRC Research Press
 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

8 Trans. Can. Soc. Mech. Eng. Vol. 00, 0000

Fig. 11. Comparison between simulation and experiment. based on RSM is proposed and can be divided into the
[Colour online.] following steps:

Simulation result Step 1: Determine design variables and response variables

0.00008
Experimental result
The transmission error is the main source of vibratory
0.00006
and acoustic nuisances that designers and builders of
gears wish to reduce. To ensure the meshing perfor-
TE (rad)

0.00004 mance of a gear pair can meet requirements, design-

ers usually choose suitable parameters to control the
0.00002 amplitude of the transmission error. Therefore, we
choose the maximum and minimum values of the
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

0
transmission error as the response variables and
–0.00002
deviation parameters as the design variables.
0.0 0.2 0.4 0.6 0.8 1.0
Rotation angle 1 (rad) Step 2: Solve the constraint range of design variables
Based on step 1, our aim is to solve the constraint
range of deviation parameters according to the
required gear accuracy. For the deviation parameters
Transactions of the Canadian Society for Mechanical Engineering

Fig. 12. Comparison between simulation results and

Sanchez-Marin et al. (2016). [Colour online.] in profile and pitch deviation, which accord with stan-
dards in gear accuracy, it is easy to get the range of
0.000015
these parameters. In contrast, to determine the varia-
The paper
0.00001 tion range of the parameters in geometric deviation,
Sanchez – Marin
variation inequality and constraint inequality must
5. 10–6
be defined.
TE (rad)

0 Step 3: Design the experiment

For a large design space with many variables, Latin
–5. 10–6
hypercube sampling is a statistical method for gener-
–0.00001 ating a near-random sample of parameter values
from a multidimensional distribution. With this
–0.000015
–0.2 –0.1 0.0 0.1 0.2
method, we need to use just a few experimental
Rotation angle 1(rad) points to achieve a high-accuracy response surface
model. We can obtain n sets of sample values of
deviation parameters by setting the sampling size of
and thus carry out gear-tolerance optimization. With
the Latin hypercube experimental design to be n.
known tolerances, the sample of deviation parameters
Each set of maximum and minimum of transmission
within the tolerance constraints can be obtained by
error corresponding to each set of sample values can
Monte Carlo simulation, which can be used in the TCA
be obtained using the TCA program proposed in
program to calculate the transmission error. However,
section 3. n sets of maximum and minimum trans-
the TCA program involves so many parameters and
mission error values can finally be obtained after
samples that it requires a lot of time and effort, thus
experiment is completed.
resulting in low efficiency of tolerance optimization.
To reduce the computational cost and improve the Step 4: Build and verify the response surface models
efficiency of gear-tolerance optimization, this paper The response surface models between maximum and
presents a response surface model that describes the minimum of transmission error and deviation varia-
relationship between deviation parameters and trans- bles can be built by the RSM based on a Latin hyper-
mission error using the RSM. cube experimental design. To verify the accuracy of
the response surface models, the R2 coefficient is usu-
ally used. It gives the proportion of the total variation
4. Gear-tolerance optimization based on RSM
in the response predicted by the model, indicating
RSM, introduced by Box and Wilson (1951), explores the ratio of the sum of squares due to regression to
the relationships between several explanatory variables the total sum of squares. A high R2 coefficient ensures
and one or more response variables. Many researchers a satisfactory adjustment of the quadratic model to
(e.g., Queipo and Pintos 2005; Steenackers et al. 2009; the experimental data.
Tosi et al. 2015) have studied RSM in recent years.
The main idea of RSM is to use a sequence of Step 5: Make the tolerance optimization of gear
designed experiments to obtain an optimal response. In Based on the model introduced earlier, we can opti-
this paper, a gear-tolerance optimization method mize the tolerance of a gear using a genetic algorithm.
Published by NRC Research Press
 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

Xu et al. 9

Fig. 13. Flow chart of gear-tolerance optimization. [Colour online.]

Determine the design variables and response variable

Solve the constraint range of design variables

Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

Obtain the transmission error curves via TCA program

based on Latin hypercube experimental design

Build and verify the response surface models

Transactions of the Canadian Society for Mechanical Engineering

Do the models meets No

accuracy requirements?

Yes

Optimization gear tolerance based on

response surface models

The flow chart of gear-tolerance optimization is Table 2. Basic parameters of pinion and wheel.
shown in Fig. 13.
Value
5. Analysis of an example Basic parameter Pinion Wheel
Take the profile form tolerance, single pitch toler- Module (mm) 3.175 3.175
ance, and installation tolerance (center distance toler- Number of teeth 28 28
ance and parallelism tolerance) as an example to Press angle (°) 20 20
illustrate the tolerance optimization of a gear. Helix angle (°) 15 15
The basic parameters of pinion 1 and wheel 2 are Addendum diameter (mm) 98.61 98.61
as shown in Table 2. The profile form tolerance f fαT Dedendum diameter (mm) 83.82 83.82
∈[0.006, 0.012], single pitch tolerance f p T ∈[0.006, Tooth width (mm) 25 25
0.012], center distance tolerance Ta ∈[0.08, 0.1], parallel-
ism in-plane tolerance Ty ∈[0.06, 0.08], and out-of-plane
tolerance T z ∈[0.04, 0.06]. The variation of each of transmission error as the response variables and
deviation variable in the corresponding constraint deviation parameters as the design variables.
obeys a normal distribution. The optimization objec-
tive is to minimize the total cost, while maintain- 5.1.1. Solution of constraint range of profile deviation
ing an amplitude of transmission error of less than parameters
1.5 × 10−3 rad. According to the relation between deviation and toler-
ance, the constraint range of profile deviation parame-
5.1. Solution of constraint range of deviation parameters ters can be described as
According to the optimization requirement and sec-
tion 4, we choose the maximum and minimum values (20) f f α ≤ f f αT jf Hα j ≤ f HαT

Published by NRC Research Press

 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

10 Trans. Can. Soc. Mech. Eng. Vol. 00, 0000

Fig. 14. Axis variation along the y axis: (a) case 1, (b) case 2, and (c) case 3.
Ta Ta Ta

Ty Ty Ty
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

(a) (b) (c)

Transactions of the Canadian Society for Mechanical Engineering

5.1.2. Solution of constraint range of pitch deviation and dz along the z axis from the origin, then rotates θy
parameters around the y axis and θz around the z axis. In this proc-
According to the relation between deviation and toler- ess, both ends of each axis should not exceed the boun-
ance, the constraint range of pitch deviation parameters dary of the variation zone.
can be described as The variations of axis 4 along the z axis are con-
ð21Þ j f pi j ≤ f pT strained by the center distance tolerance, and the con-
straint inequality of axis 4 along the z axis can be
The cumulative pitch-deviation curve depends on described as
actual measurement data, but, in general, its trend is
Tz T
similar to a sine curve. Therefore, in this paper, we ð23Þ − ≤ dz + x · θ y ≤ z
adjust the plus or minus of the individual single pitch 2 2
deviation f pi to ensure that the cumulative pitch The variations of axis 4 along the y axis are constrained
deviation curve accords with this relationship. The value by both parallelism tolerance and center distance toler-
of the pitch deviation parameter θi can be obtained by ance, which can be divided into the following three types:
substituting fpi into eqs. (6) and (7).
T Ty − Ta
1. − a ≤ dy ≤ . In this case, as shown in Fig. 14a,
5.1.3. Solution of variation inequality and constraint inequality 2 2
of installation deviation parameters the upper boundary of the variation zone coincides
As is shown in Fig. 9, axis 4 is constrained by both with that of the parallelism tolerance zone, and the
center distance tolerance and parallelism tolerance. The lower boundary of the variation zone coincides
variation of axis 4 along the y axis is constrained by the with that of the center distance tolerance zone. So
center distance tolerance, and the variation range is the constraint inequality of axis 4 along the y axis
 
T T can be described as
− a , a ; the variations of axis 4 along the y and z axes
2 2 Ta Ty
(24) − − dy ≤ x · θ z ≤
are both constrained by parallelism tolerance, and the 2 2
variation zone consists of two pairs of parallel planes,
Ty − Ta Ta − Ty
the distance between which are Ty and Tz, respectively. 2. < dy < . In this case, as shown in
So the variation inequality can be described as 2 2
Fig. 14a, both the upper and lower boundaries of
8
> Ta Ta the variation zone coincide with that of the paral-
>
> − ≤ dy ≤
>
> 2 2 lelism tolerance zone. So the constraint inequality
>
>
>
> T T of axis 4 along the y axis can be described as
> z
< − ≤ dz ≤
z

ð22Þ 2 2 Ty Ty
>
> T T ð25Þ − ≤ x · θz ≤
>
> − z ≤ θy ≤ z 2 2
>
> l l
>
> Ta − Ty
>
> T T T
: − y ≤ θz ≤ y 3. ≤ dy ≤ a . In this case, as shown in Fig. 14a,
l l 2 2
the upper boundary of the variation zone coincides
Set the order of variation as dy → dz → θy → θz — in with that of the center distance tolerance zone, and
other words, axis 4 first translates d y along the y axis the lower boundary of the variation zone coincides
Published by NRC Research Press
 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

Xu et al. 11

Fig. 15. Profile deviation. [Colour online.] Fig. 16. Pitch deviation. [Colour online.]
0.006
0.0010

Cumulative pitch deviation (mm)

0.004
Profile deviation (mm)

0.0005
0.002
0.0000
0.000
–0.0005
0.002
–0.0010
0.004
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

–0.0015
0.006 5 10 15 20 25
0.0 0.1 0.2 0.3 0.4 0.5 Sequence number of tooth
Roll angle (rad)

parameters d y and d z contribute to the constraint of

with that of the parallelism tolerance zone. So the
screw parameters θy and θz, though they have no effect
constraint inequality of axis 4 along the y axis can
Transactions of the Canadian Society for Mechanical Engineering

on transmission error.)
be described as
To reduce the number of experiments and improve
Ty T the efficiency of model development, we adopt the
(26) ≤ x · θ z ≤ a − dy
−
2 2 Latin hypercube experimental design to obtain 200 sets
l of sample values of deviation parameters. These sample
where x equals ± . values are fed into the TCA program to yield 200 sets of
2
maximum and minimum values for transmission error.
5.2. Design of experiment and building response
surface models The deviation parameters of every tooth flank are differ-
Screw parameters dy and dz have no effect on transmis- ent, so a response surface model of the transmission
sion error. Thus, we can take the screw parameters θy and error of a single tooth flank, which is related to the pre-
θz, profile form deviation ffα, single pitch deviation fpi as vious tooth and the following tooth, will be built.
the design variables, and the maximum and minimum Owing to space limitations, we present only ten calcula-
values of the transmission error as the response variables. tion results of the sample points of the third tooth flank,
Set f fαT = 0.012, f pT = 0.012, T a = 0.1, T y = 0.08, and which is shown in Table 3 in accord with Tosi et al.
Tz = 0.06, and choose the value of fr in eq. (5) according (2015). Figure 18 shows the transmission error curves in
to Mucchi et al. (2010). Based on these data, the curves a meshing period.
of profile deviation and pitch deviation can be simu- Using the least squares method to calculate the vector
lated, as shown in Figs. 15 and 16, which show only ten of undetermined coefficients based on experimental
curves for the sake of clarity. Similarly, the varying sam- results, the response surface model can be described as
ple of screw parameters θ y, θ z can be obtained using follows (we show only the response model of ε̃ð3Þ max
eqs. (24)–(26), as shown in Fig. 17. (Notice that screw because of the length of the formula):

ð2Þ ð3Þ ð4Þ

ε̃ð3Þ −4 −3
max =−3.28919×10 þ0.031069f f α þ0.26941f p2 þ9.38359×10 f f α −1.05734f p3 þ0.025955f f α −0.053003f p4
−3 −3 ð2Þ ð2Þ ð3Þ ð2Þ ð2Þ ð4Þ
−3.93064×10 θy −6.19949×10 θz þ22.74066f f α f p2 −0.17392f f α f f α −54.13984f f α f p3 −1.43998f f α f f α
ð2Þ ð2Þ ð2Þ ð3Þ ð4Þ
þ39.53844f f α f p4 þ0.80047f f α θy −0.20074f f α θz −26.53204f p2 f f α −1065.05006f p2 f p3 þ7.41176f p2 f f α
ð3Þ ð3Þ ð4Þ ð3Þ
ð27Þ −524.99490f p2 f p4 þ19.40320f p2 θy þ37.38939f p2 θz þ21.63044f f α f p3 −0.26387f f α f f α −1.06217f f α f p4
ð3Þ ð3Þ ð4Þ ð4Þ
−0.38266f f α θy þ0.51029f f α θz þ23.06571f p3 f f α þ418.64872f p3 f p4 −7.61788f p3 θy −24.63237f p3 θz −27.54479f f α f p4
ð4Þ ð4Þ ð2Þ 2 2
−0.14947f f α θy −0.23574f f α θz −0.90632f p4 θy þ6.57018f p4 θz þ1.70667θy θz −1.13514ð f f α Þ þ1003.95257ðf p2 Þ
ð3Þ 2 2 ð4Þ 2 2 2 2
þ0.027470ðf f α Þ þ98.39405ðf p3 Þ −0.45683ðf f α Þ −63.61701ðf p4 Þ þ0.62841ðθy Þ þ1.94621ðθz Þ

To verify the accuracy of response surface models, an both high, which suggests that the model approximates
R2 coefficient is usually used. As shown in Table 4, the the underlying relationship between the deviation
R 2 coefficients of the response surface models of the parameters and the transmission error quite well. The
maximum and minimum transmission error values are comparison in Fig. 19 also shows that there is good

Published by NRC Research Press

 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

12 Trans. Can. Soc. Mech. Eng. Vol. 00, 0000

Fig. 17. Distribution of variable factors: (a) dz – θy and (b) dy – θz. [Colour online.]
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

(a) (b)
Transactions of the Canadian Society for Mechanical Engineering

Table 3. Design of experiment result.

Design variable Symbol Values generated by Latin hypercube sampling
ð2Þ
Profile form deviation of the f fα 9.5242 8.5121 9.7370 8.4088 10.5626 9.3164 10.1809 11.3295 8.0681 8.5791
2nd tooth flank (×10−3)
Single pitch deviation of the fp2 2.3662 1.8366 0.2121 1.8073 0.8263 0.3741 1.3321 0.5171 1.5007 0.3088
2nd tooth flank (×10−4)
ð3Þ
Profile form deviation of the f fα 6.2155 8.6388 11.4955 10.6728 9.0797 9.7175 9.5393 7.5717 11.0821 11.2231
3rd tooth flank (×10−3)
Single pitch deviation of the fp3 4.3257 2.1972 2.4369 2.0835 3.3727 1.5431 3.5008 1.6724 2.8386 0.9321
3rd tooth flank (×10−4)
ð4Þ
Profile form deviation of the f fα 9.5797 11.8824 11.6796 8.5096 9.4311 11.7158 11.3062 7.5956 8.2683 7.7417
4th tooth flank (×10−3)
Single pitch deviation of the fp4 6.0404 2.3576 4.3590 3.3936 5.0233 2.5060 6.0514 2.3409 5.3425 2.6995
4th tooth flank (×10−4)
Rotational vector around θy −4.1359 3.2701 0.6928 2.8673 −4.5871 1.3347 −1.6245 4.3428 2.2271 −6.5555
the y axis (×10−3)
Rotational vector around θz 3.6108 6.7889 −1.5117 −4.7115 −3.3469 −3.7488 5.3085 −4.1386 3.75922 −6.0205
the z axis (×10−3)

Response variable Symbol Values calculated by TCA program

Maximum transmission error εð3Þ

max −4.6286 −0.4314 −1.6444 −1.3845 −2.6009 −1.2040 −3.2855 −1.7494 −2.6301 −0.3754
value of third tooth flank (×10−4)
ð3Þ
Minimum transmission error εmin −6.7491 −3.1440 −3.8648 −2.6718 −3.6305 −2.3932 −6.5938 −3.5296 −5.9241 −0.2662
−4
value of third tooth flank (×10 )

Fig. 18. Transmission error curves. [Colour online.] Table 4. Accuracy of the RSM.
Transmission error R2
ð3Þ
0.0010 ε̃max 0.9571
ð3Þ 0.9533
ε̃min
0.0005
TE (rad)

0.0000 agreement between the simulation results and the pre-

dicted results of the response surface models for 200
–0.0005 simulation runs.
Therefore, the amplitude of transmission error ε̃ in
–0.0010
meshing can be described as
0 1 2 3 4 5 6 ðiÞ
Rotation angle 1 (rad) (28) ε̃ = ε̃max − ε̃min = maxðε̃ðiÞ
max Þ − minðε̃min Þ i = 1, : : : , z1

Published by NRC Research Press

 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

Xu et al. 13

ð3Þ ð3Þ
Fig. 19. Comparison between simulation and prediction of response surface models for (a) ε̃max and (b) ε̃min . [Colour online.]
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

(a) (b)

Table 5. Parameters of the genetic algorithm. Fig. 21. Amplitude distribution of transmission error.
[Colour online.]
Parameter Value
Transactions of the Canadian Society for Mechanical Engineering

Population size 200

Generation number 100
Precision of variables 10
Generation gap 0.9
Crossover rate 0.7
Mutation rate 0.05

Fig. 20. Optimization process of the objective function.

[Colour online.]

tolerance-optimization model, which seeks the mini-

mum cost where the amplitude of the transmission
error does not exceed 1.5 × 10−3 rad:

min f ð f f aT , f pT , T a , Ty , Tz Þ
such that f f aT ∈ ½0.006, 0.012
(30)
f pT ∈ ½0.006, 0.012 T a ∈ ½0.08, 0.1
−3
T y ∈ ½0.06, 0.08 T z = ½0.04, 0.06 jε̃j ≤ 1.5 × 10

Here, a genetic algorithm is used to solve the optimi-

where ε̃max and ε̃min are the maximum and minimum zation problem. The parameters used in the genetic algo-
values of the transmission error, respectively. rithm are shown in Table 5. Figure 20 shows the
optimization process for the objective function.
5.3. Tolerance optimization of gear
The optimization result of the genetic algorithm is as
To optimize cost, we must determine the relation-
follows:
ship between tolerance and cost. Other researchers
have developed cost–tolerance models; for example, Ta = 0.0851 T y = 0.0698 T z = 0.0595
Chase et al. (1990) developed a model now commonly
f f aT = 0.0105 f pT = 0.0083
used for optimizing gear tolerance:
X
n
Bi To verify the validity of this result, 1000 samples of
ð29Þ f ðTÞ = Ki deviation parameters was generated by Monte Carlo sim-
i=1 T i
ulation and fed into the TCA program to obtain the
Combining the response surface model we developed amplitude distribution of the transmission error, as
with this cost–tolerance model yields the following shown in Fig. 21.
Published by NRC Research Press
 Pagination not final (cite DOI) / Pagination provisoire (citer le DOI)

14 Trans. Can. Soc. Mech. Eng. Vol. 00, 0000

Figure 21 shows that only a few transmission error shaft centre distance and parallelism of axes. Published in
amplitudes exceed the optimization requirement, which Switzerland.
still is acceptable. Thus, the approach proposed in this Litvin, F.L., and Fuentes, A. 2004. Gear geometry and applied
theory. Cambridge University Press, Cambridge, UK.
paper is feasible for improving the efficiency of gear- Mucchi, E., Dalpiaz, G., and Rivola, A. 2010. Elastodynamic analy-
tolerance optimization. sis of a gear pump. Part II: meshing phenomena and simula-
tion results. Mech. Syst. Signal Process. 24(7): 2180–2197.
6. Conclusion doi:10.1016/j.ymssp.2010.02.004.
Tolerance analysis and optimization can effectively Mujezinovi ć, A., Davidson, J.K., and Shah, J.J. 2004. A new
mathematical model for geometric tolerances as applied to
improve processing quality and reduce processing costs.
polygonal faces. J. Mech. Des. 126(3): 504–518. doi:10.1115/1.
But for gear products, the complexity of gear contact 1701881.
calculations restricts the efficiency of gear-tolerance
Downloaded from www.nrcresearchpress.com by St. Francis X Univ on 08/17/18. For personal use only.

Queipo, V., and Pintos, S. 2005. The integration of design of

analysis and optimization. Moreover, as the number of experiments, surrogate modeling and optimization for ther-
deviation parameters considered in the calculations moscience research. Eng. Comput. 20(4): 309–315. doi:10.1007/
s00366-004-0299-x.
increases, finding a solution becomes more complex
Requicha, A.A.G. 1983. Toward a theory of geometric toleranc-
and the computation time becomes longer, thus result- ing. Int. J. Rob. Res. 2(4): 45–60. doi:10.1177/027836498
ing in low computational efficiency. This paper pre- 300200403.
sented a more efficient gear-tolerance optimization Sanchez-Marin, F., Iserte, J.L., and Roda-Casanova, V. 2016.
Transactions of the Canadian Society for Mechanical Engineering

model based on the RSM. The analysis of an example Numerical tooth contact analysis of gear transmissions
through the discretization and adaptive refinement of the
shows that the method has good practicability and can
contact surfaces. Mech. Mach. Theory, 101: 75–94. doi:10.1016/
effectively optimize gear tolerance and provide effective j.mechmachtheory.2016.03.009.
guidance for accurate gear design. Schleich, B., and Wartzack, S. 2014. A discrete geometry
approach for tolerance analysis of mechanism. Mech.
Acknowledgements Mach. Theory, 77(7): 148–163. doi:10.1016/j.mechmachtheory.
The progress of the research is supported by National 2014.02.013.
Steenackers, G., Presezniak, F., and Guillaume, P. 2009.
Natural Science Foundation of China (No. 51775156). Development of an adaptive response surface method for
optimization of computation-intensive models. Comput.
References Ind. Eng. 57(3): 847–855. doi:10.1016/j.cie.2009.02.016.
Asante, J.N. 2009. A small displacement torsor model for toler- Stuppy, J., and Meerkamm, H. 2009. Tolerance analysis of a
ance analysis in a workpiece-fixture assembly. Proc. Inst. crank mechanism by taking into account different kinds
Mech. Eng., Part B J. Eng. Manuf. 223(8): 1005–1020. doi:10.1243/ of deviations. Proc. 11th CIRP International Seminar on
09544054JEM1337. Computer Aided Tolerancing, Annecy, France, 26–27 Mar.
Bourdet, P., Mathieu, L., Lartigue, C., and Ballu, A. 1996. The 2009. pp. 35–40.
concept of the small displacement torsor in metrology. In Tosi, G., Mucchi, E., D’Ippolito, R., and Dalpiaz, G. 2015.
Advanced mathematical tools in metrology II. Edited by Dynamic behavior of pumps: an efficient approach for fast
P. Ciarlin, M.G. Cox, F. Pavese, and D. Richter. Singapore; robust design optimization. Meccanica, 50(8): 2179–2199.
River Edge, NJ: World Scientific. pp. 110–122. doi:10.1007/s11012-015-0142-z.
Box, G.E.P., and Wilson, K.B. 1951. On the experimental attain- Turner, J.U., and Wozny, M. 1987. Tolerances in computer-aided
ment of optimum conditions. J. R. Stat. Soc. Series B Stat. geometric design. Vis. Comput. 3(4): 214–226. doi:10.1007/BF01
Methodol. 13(1): 1–45. 952828.
Chase, K.W., Greenwood, W.H., Loosli, B.G., and Hauglund, L.F. Walter, M., and Wartzack, S. 2013. Statistical tolerance-
1990. Least cost tolerance allocation for mechanical assem- cost-optimization of systems in motion taking into account
blies with automated process selection. Manuf. Rev. 3: 49–59. different kinds of deviations. Proc. 23rd CIRP Design
Desrochers, A., and Clément, A. 1994. A dimensioning and toler- Conference, Bochum, Germany, 11–13 Mar. 2013. Smart prod-
ancing assistance model for CAD/CAM systems. Int. J. Adv. uct engineering. Springer Berlin Heidelberg. pp. 705–714.
Manuf. Tech. 9(6): 352–361. doi:10.1007/BF01748479. Walter, M., Sprugel, T., and Wartzack, S. 2013. Tolerance analy-
Desrochers, A., and Rivière, A. 1997. A matrix approach to the sis of systems in motion taking into account interactions
representation of tolerance zones and clearances. Int. J. Adv. between deviations. Proc. Inst. Mech. Eng., Part B J. Eng.
Manuf. Tech. 13(9): 630–636. doi:10.1007/BF01350821. Manuf. 227(5): 709–719. doi:10.1177/0954405412473719.
Gao, J.S., Chase, K.W., and Magleby, S.P. 1999. Generalized 3-D Watrin, S., Binz, H., and Lindeloof, B.V.D. 2013. Methodo-
tolerance analysis of mechanical assemblies with small logy for requirement-driven tolerance specification of
kinematic adjustments. IIE Trans. 30(4): 367–377. doi:10.1023/ bevel gears. Procedia CIRP, 10: 299–305. doi:10.1016/j.
A:1007451225222. procir.2013.08.046.
Guo, J., Hong, J., Yang, Z., and Wang, Y. 2013. A tolerance analy- Whitney, D.E., Gilbert, O.L., and Jastrzebski, M. 1994.
sis method for rotating machinery. Procedia CIRP, 10: 77–83. Representation of geometric variations using matrix
doi:10.1016/j.procir.2013.08.015. transforms for statistical tolerance analysis in assemblies.
ISO 1328.1-2013. 2nd ed. Cylindrical gears — ISO system of flank Proc. IEEE Conference on Robotics and Automation,
tolerance classification — Part 1: definitions and allow- Atlanta, GA, USA, 8–13 May 1994. pp. 191–210.
able values of deviations relevant to flanks of gear teeth. Yoshikawa, K., Tani, H., Tarutani, I., Suzuki, A., and Maki, H.
Published in Switzerland. 1997. Measurement of helical gear transmission error and
ISO/TR 10064.3-1996. Cylindrical gears — code of inspection improvement of analytical method. Trans. Jpn. Soc. Mech.
practice — Part 3: recommendations relative to gear blanks, Eng. 63(609): 1775–1782. doi:10.1299/kikaic.63.1775.

Published by NRC Research Press