See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/287012523

Optimisation of planetary gear train using

multiobjective genetic algorithm

Article in Journal of the Balkan Tribological Association · January 2011

CITATIONS READS

3 51

4 authors, including:

Stojan Radenović Marija Milojevic

University of Belgrade Imperial College London
477 PUBLICATIONS 3,578 CITATIONS 11 PUBLICATIONS 161 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

common fixed point results for single and multivalued mappings. View project

Best Proximity Points of Proximal Contractions View project

All content following this page was uploaded by Stojan Radenović on 22 December 2015.

The user has requested enhancement of the downloaded file.

Journal of the Balkan Tribological Association Vol. 17, No 3, 462–475 (2011)

Computer simulations

OPTIMIsATION OF PLANETARY GEAR TRAIN

USING MULTIOBJECTIVE GENETIC ALGORITHM
­
B. ROSIca*, S. RADENOVIca, L. J. JANKOVIcb, M. MILOJEVIca
a
Faculty of Mechanical Engineering, University of Belgrade,
16 Kraljice Marije Street, 11 120 Belgrade, Serbia
E-mail: boza.r@eunet.rs
b
Group ‘Zastava Vozila’, Kragujevac, Serbia

ABSTRACT

The purpose of this paper is to present a method for solving multiobjective non-
linear optimisation of planetary gear trains. Optimisation of gear train was suc-
cessfully accomplished using the genetic algorithm (GA). The weighting method
is used to approximate the Pareto set. This method transforms the multiobjective
optimisation problem into single-objective optimisation problem by associating
each objective function a weighting coefficient and then minimising the weighted
sum of the objectives. The GA-based approach produced quite satisfactory re-
sults promptly supplying the designer with the preliminary design parameters of
planetary gear train for different gear ratios. The results showed that the genetic
algorithm is useful and applicable for optimisation of planetary gears desgin.
Keywords: planetary gear train, multiobjective optimisation, pareto-optimal, ge-
netic algorithm.

AIMS AND BACKGROUND

Planetary gear trains take a very significant place among the gear transmissions
which are used in many branches of industry such as automotive transmissions,
aircrafts, marine vessels, machine tool gear boxes, gas turbine gear box, robot
manipulatos, etc.
Planetary gear trains have a number of advantages over the transmission with
fixed shafts1. The advantages of such a transmission are as follows:
(i) Under similar operating conditions the planetary transmissions serve lon-
ger and produce less noise compared to the fixed shaft transmissions.
(ii) This power transmission unit can handle larger torque loads relative to its
compact size than any other gear combination in standard transmission.
* For correspondence.
462
 (iii) Improved efficiency of a gearing system can reduce the requirements on
the capacity of the lubrication system and the gearbox lubricant, thereby reducing
the operation costs of the system.
(iv) Efficiency prediction can assist in estimating the power requirements
during the design stage of a machine and thus ensuring that the system operates
reliably. It can also assist in estimating the power output for a given power input.
(v) The input and output shafts are concentric so no bending moments or
torques are created from radial forces.
(vi) High overall transmission ratio speed.
The effect of instantaneous efficiency of an involute gear drive was studied
by Radzimovsky2 andAnderson3. The problem of efficiency of the planetary gear
train was studied experimentally in Ref. 4. Gearbox efficiency optimisation was
not normally included in the gearbox design methodology in the past due to the
lack of a technique that could assess many design variables.
Today, optimisation is a common method to improve the properties of me-
chanical devices and has received a significant amount of attention. Many engi-
neering problems have multiple objectives, including engineering system design
and nonlinear optimisation. Solving engineering problems, especially design op-
timisation, involve multiple and conflicting objectives.
Extensive work has been done in the field of nonlinear optimisation, and its
application in mechanical engineering design has been dealt with by Arora6 , Pa-
palambros and Wilde7, Rao8, etc. The introduction of a larger number of criteria
considering the desirable performances, even the conflicting ones (axial distance
– efficiency), represents a significant step towards the reality of a planetary gear
train model solved by the multiobjective optimisation methods.
When several criteria are considered simultaneously there is no unique op-
timum solution but a set of mathematically equivalent Pareto-optimal solutions.
There are many papers that present various approaches to find the Pareto optimal
front almost based on the evolutionary algorithms. Articles9–11 are related to nu-
merical analyses based upon the GA, where the Pareto GA method is defined as
based on the characteristic of the GA to search for non-dominated solutions. GA
starts by generating a random initial population. Using an iterative procedure,
the current population is updata and the next population is created by using some
operators, namely: selection, mutation and crossover.
The design of planetary gear trains requires a range of geometrical and kine-
matics conditions in order to perform the mounting and appropriate meshing of
the gears during their work. It is necessary to express the above requirements in
terms of the corresponding functional constraints, whereby all the relevant val-
ues of the gears and planetary gear trains as a system are defined. In developing
the optimisation model, one must start with very strict engineering requirements,
which a planetary gear train should fulfil in respect of efficiency, volume, factor
463
of safety, etc. Based upon the requirements defined, it follows that it is practically
impossible to describe a planetary gear train regarding the desirable performances
with one criterion only. Recently, multiobjective optimisation techniques are in-
corporated into the mechanical design process instead of a single criterion optimi-
sation concept which has been widely used for a long time.

PLANETARY GEAR TRAIN EFFICIENCY

According to their kinematics structure, planetary gear trains are complex toothed
mechanisms which can be decomposed into external and internal toothed gears
with the corresponding interaction. Each planet gear has a supporting link, called
the carrier or arm, which keeps the center distance between the two meshing gears
constant. Planets are free to rotate with respect to the carrier.
The gear trains in operation are characterised by losses in the mechanical
energy arising as a consequence of friction between the contact surfaces of the
meshing teeth and the friction in the bearings. The analysis considers sliding
losses, which are the result of friction forces developed as the teeth slide across
each other, rolling losses resulting
from the formation of an elastohy-
F

F drodynamic film. Contact starts at

the intersection of the tip diameter
of the internal gear with the path
F

F
of contact at A2 (Fig. 1).
F

F
The path of contact is tangent
F

F to the base circles of two gears.

F

Contact ends at the intersection

of the tip diameter of the exter-
F
F

nal gear with the path of contact

F

at E2. In this boundary case, the

a addendum circles of radii ra1 and
ra2 cross the motionless points A2
and E2, respectively. In order to
evaluate the efficiency of an in-
ternal gear pair, we must consider
the equilibrium of the gears. Fig-
ure 1 shows the normal forces Fn,
the rolling friction forces FR, and
the sliding friction forces Fm, with
suffices 1 for teeth in the path of
approach and 2 for teeth in the
Fig. 1. Forces between gear teeth path of recess.
464
 On the basis of the models developed for a gear pair with external and internal
gearing, the efficiency of a planet gear train can be determined. The power losses
within the gears are expressed by means of the efficiency. The instantaneous effi-
ciency for internal gear at any particular instant, from the relevant T1 input torque,
is determined according to the expression:
T2 1
hi = H
(1)
T1 ugb

where T2 is the output torque acting on the driven gear; T1 – input torque acting on
the driving gear; uHgb – relative gear ratio.
The overall efficiency for gearing under consideration may be written as fol-
lows: E
1 2
l A∫2
hgb
H
= hi d x (2)

where l = A 2E 2 is the length of path of contact; x – path of contact distances.

The instantaneous frictional force due to sliding of two gear teeth against
each other is given as:
Fm (x) = m(x) Fn (x). (3)

The friction coefficient is calculated by the Benedict and Kelley method for
mineral oil3:
 29.66 
 Fn 
m(x) = 0.0127 lg  b 2  (4)
 hvR vS 
 
where v R is the rolling velocity; vS – sliding velocity; h – fluid dynamic viscosity;
b – face width of gear.
The instantaneous force due to build of the minimum elastohydrodynamic
lubrication film is given by the following equation3:
FR = C1h(x)b (5)

where C1 = 9  ×  107 is a constant of proportionality.

The gear contact minimum film thickness is calculated by the method of
Dowson and Higginson:
R 0.43
h(x) = 1.6a 0.6 (hn R )0.7 E 0.003 (6)
Fn0.13

where a is viscosity–pressure coefficient of lubricant; R – effective radius of cur-

vature; E – the Young modulus of gear material.
For convenience, the output torque of the train is assumed to be constant.
465
From the equilibrium of gears, we have:
T1 − p1FR2 − xFR1
Fn1 = (7)
d b1 + m 2 p1 − m1x

where
p1 = x + pb ; (8)
p2 = a sin a w + x + pb ; pb = mπ cos a – base pitch; (9)
p3 = a sin a w + x. (10)
The output torque acting on the driven gear at any instant can be expressed
in the form:
T2 = Fn1d b2 + p2 (m 2 Fn1 − FR2 ) − (m1Fn1 + FR1 ) p3 . (11)

On the basis of the models developed for a gear pair with external and inter-
nal gearing, the efficiency of a planetary gear train may be expressed as follows:
1 − hab
H H
uab
haH
b
= (12)
1 − uabH

where
hab
H
= hag
H H
hgb (13)

hHab is the relative efficiency for gear pair a-b; hHgb – relative efficiency for gear pair
g-b; uHab – relative gear ratio.
Based upon the models developed, computer programs for instantaneous
efficiency determination were devised. The computer numerical results for the
determination of the instantaneous efficiency of a gear pair with internal gearing
are shown in Fig. 2.

Fig. 2. Instantaneous efficien-

cies during the contact period

466
PARETO Optimisation

In general, a multiobjective optimisation problem for which a solution technique

is proposed in this paper, can be defined as determining a vector of design vari-
ables within a feasible region to minimise a vector of objective functions that usu-
ally conflict with each other12. Formally, the problem is stated as follows:
Find the values of n design variables (x1,..., xn ) which satisfy n upper and
lower boundaries xia , xie, i = 1,..., n and optimise (minimise or maximise) k objec-
tive functions. Since the problems of minimising and maximising are equivalent
max  f (x) = –min(–f(x)), the general problem can be written as:
minimise { f1 ( x),..., f k ( x)}
subject to g ( x) ≤ 0, (14)
xia ≤ x ≤ xie

where x is a vector of design variables, f i (x) – the i-th objective function, and
g(x) – a constraint vector. The value represents the lower boundary and the upper
boundary of the design variable .
The optimum solution in this case is not unique because the objectives can
contradict each other. Therefore, a set of solutions that is called the Pareto opti-
mal set is considered according to the following definition:
Definition 1. Pareto optimal: Consider a point x* in the feasible solution space,
X, x* ∈ X. The point (a set of decision variables) is Pareto optimal if and only if
there does not exist another point, x ∈ X, that satisfies f(x) ≤  f (x*) and f i (x) ≤  f i( x*),
for at least one function. In other words, this definition states that, for a minimisa-
tion problem, there is no other point which can cause a decrease in one objective
function value without causing a simultaneous increase in at least one of the other
objective function values.
Definition 2. Dominated and non-dominated points: A vector of objective func-
tions, f(x*), is non-dominated if and only if there does not exist another vector,
f(x), that satisfies f(x) ≤  f(x*) with at least one f i (x) ≤  f i (x*). Otherwise, f(x*) is dom-
inated.
Definition 3. Pareto front: The set X* = {x1*, x2*,..., xn*}, which is composed of all
the non-dominated Pareto optimal solutions that compromise the Pareto front of
non-dominated solutions12.
The ‘feasible’ domain is defined by the following relation:
D = {x ∈ R n | g ( x) < 0 ∧ h( x) = 0}. (15)

Figure 3 shows a mapping of the two-dimensional design space D into the

criterion space X where the Pareto-optimal solutions lie on the curved section AB.
467
Fig. 3. Mapping of a feasible set into the criterion space

WEIGHTED SUM APPROACHES

In this paper, the weighting sum method is used to approximate the Pareto set.
This method transforms the multiobjective optimisation problem into a single-
objective optimisation problem by associating each objective function with a
weighting coefficient and then minimising the weighted sum of the objectives,
as follows:
n
min f ( x) = ∑ wi f i ( x), (16)
i =1

where weights wi are non-negative, such that:

n
∑ w (l ) = 1
i =1
i (17)

for n objective functions.

A subset of the Pareto optimal set can thus be generated through systematic
variation of the weights w = {w1, w2,..., wn} and repeated solution of the scalarised
problem (16).
The weights are modified after every certain number of iterations during the
optimisation. The weights are defined by the following equation:
n

∑ random (l ) i
wi (l ) = i =1
n
, (18)
∑ random k (l )
k =1

where l – the index of iteration; randomi (l) – the function to create a uniformly

distributed random value in the range [0,1].

468
FORMULATION OF Optimisation MODEL

The criteria regarding the desired performances are expressed by the criteria func-
tions, which, for the best planetary gear train design, should reach the extreme:
extr f ( x). (19)
x∈D

The function criteria for a one-stage planetary gear train can be written in the
form of the following relations for:
mn za cos a t
– centre distance f1 = (1 + ua-g
H
) (20)
cos b cos a wt

1 − hab
H H
uab
– efficiency f2 = (21)
1 − uab
H

– contact ratio f 3 = e a (a–g) ( x) (22)

– pressure angle f 4 = a w(a-g) ( x) (23)

[s F ]M(a)
– safety factor for bending stress f 5 = S F(a) ( x) = (24)
s F(a)
[s H ]M(a)
– safety factor for contact stress f 6 = S H(a) ( x) = (25)
s H(a)

– volume of material used for gears f 7 = V ( x) (26)

[s F ]M(b)
– safety factor for bending stress for ring gear f8 = S F(b) ( x) = (27)
s F(b)

– outer diameter of planetary gear train f 9 = Dout ( x) (28)

where
2
0.5( d a(a) − d a(b)
2
+ d a(g)
2
− d b(g)
2
) − a sin a wt
– transverse contact ratio ea = (29)
πmt cos a t
xa + xg
– function of pressure angle inva wt = 2 tan a n + inva t (30)
za + zg
Ft
– tooth root stress for the sun gear s F(a) = YFaYSaYeYb K A K V K a Kb (31)
bmn

– critical root stress [s F ]M = s F limYSTYNTYdrelTYRrelTYX (32)

[s F ](a)
– factor of safety from breakage S F(a) = ≥ S Fmi n (33)
s F(a )
Ft u + 1
– effective contact stress: sH = Z H Z E Z e Zb K A K V K Ha K Hb (34)
bd (a) u

469
– for the sun gear – planet gear s Ha = s Hg < [s H ]M = min{[s H ](a) , [s H ](g) } (35)

– critical contact stress [s H ]M = s H lim Z NT ( Z L Z V Z R ) Z V Z X (36)

[s H ](a,g)
– factor of safety from pitting (sun gear – planet gear) S Ha,g = ≥ SH min (37)
sH
In addition, it is also necessary to include the functional constraints:
[s F ]M(a,g,b)
– factor of safety from bending g (1,2,3) = − SF > 0 (38)
s F(a,g,b)
[s H ]M(a,g)
– factor of safety from pitting g4 = − SH > 0 (39)
sH
[s H ]M(g,b)
g5 = − SH > 0 (40)
sH
– radial interference g 6 = ∆x > 0 (41)
π
– space requirements g 7 = 2a sin( ) − f − d a-g ≥ 0 (42)
nw
za z b
– condition for assembly h1 = − INT = 0. (43)
nw D( zg , zb )
Based on the objective functions given and on the functional constraints, all
the relevant values of the planetary gear train have also been identified.

Optimisation PROCEDURE

Every genetic algorihm has its ba-

sic components shown as the flow-
chart in Fig.  4 (Ref. 11). Simple GA
has three basic operators:

3.1. Selection;
3.2. Crossover;
3.3. Mutation.
Each member in this population is
evaluated and assigned a fitness value.
In the selection procedure, some selec-
tion criterion is applied to select a cer-
tain number of strings, namely parents,
from this population according to their
fitness values. Next, in crossover pro-
cedure, selected strings from old popu-
Fig. 4. Genetic algorithm flowchart lation are randomly paired to mate. For
470
Fig. 5. Single-point crossover

binary coding, a cross-site is determined according to some law, and the paired
strings exchange all characters following the cross-site. Crossover usually results
in two new strings, namely, two children that are expected to combine the best
characters of their parents. Parent pairs are randomly chosen from the selected
population and the kind of merging depends on the crossover operator used. Fig-
ure 5 contains a scheme of the single-point crossover with binary codification.
Mutation. Mutation is a random process where one genotype is replaced by anoth-
er to generate a new chromosome. Each genotype has the probability of mutation,
to change from 0 to 1, and vice versa. Mutation is an operator to change elements
in a string which is generated by crossover operator. Mutation may improve an
existing gene string and lead the search to the better level of optimisation. Muta-
tion is an important part of genetic algorithm as it helps to prevent the population
from stagnating at any local optima.
In genetic algorithms variables are represented as coded strings. The coding
discretises the search space of the optimisation problem. Thus, GAs are able to
work with discrete or discontinuous functions. GA is useful method optimisation
when other techniques such as gradient descent or direct analytical discovery are
not possible. This gives great flexibility to a wide range of applications.

NUMERICAL RESULTS AND DISCUSSION

Pareto optimality can be illustrated graphically by considering the set of all feasible
objective values, i.e. the set of all points in objective space corresponding to at
least one setting of the design variable. The results of the computer program can be
shown in the form of a diagram – the criterion space, based upon which all cause –
effect interrelations between certain objective functions can be discovered.
Figure 6 presents the criterion space for the axial distance – efficiency and
Fig. 7 – the criterion space for the axial distance – the outer diameter of the plan-
etary gear train.
Based upon a geometrical interpretation of the results in the criterion space,
the following conclusions can be drawn:
– the criteria f1 and f2 (axial distance – efficiency) are mutually conflicting;
there exists a very strong correlation between the criteria f1 and f9 (axial distance
– outer diameter of the planetary gear train).
The way of coding the variables is shown in the Table 1.
471
 V (cm3)
a (mm) a (mm)

Fig. 6. The criterion space for the axial distance Fig. 7. The criterion space for the axial dis-
– efficiency tance – volume of material used for gears

Table 1. GA coding of design variables

Design variables vectors Vectors Random binary digits String length l

Module mn x1 10101 5
Number of teeth za x2 1101 4
Number of teeth zg x3 11010 5
xa x4 11100 5
xg x5 11100 3
xb x6 111 3
a1 x7 11101 5
nw x8 11111 5
H x9 101 3
rct x10 1111 4
A single 42-bit individual 101011101110101110011100
(chromosome) 11111101111111011111

CONCLUSIONs

This paper describes the optimisation algorithm applied to a multiobjective opti-

misation of a planetary gear train using an approach based on genetic algorithm.
Multiobjective modelling reflects very well the design process in which usually
several conflicting objectives have to be satisfied such as the efficiency of plan-
etary gear trains and the distance between centres of sun gear and planetary gear.
The effect of changes of the design parameters gives useful information regarding
the sensitivity of various features in the model. A Pareto set, presented as a plot of
the efficiency and axial distance of the planetary gear train, gives a quantitative
description of the compromise between efficiency and size. The results of this
paper illustrate the importance of formulating the problem as a multiobjective
optimisation.

472
 The procedure developed in this paper is advantageous because the results
are practical and no further analysis of the planetary gear train is required. In the
present study, an optimisation approach based on Genetic Algorithms is proposed
to improve planetary gear train performances.

ACKNOWLEDGEMENTS

The work presented in this paper was supported by the Ministry of Science and
Technological Development of Serbia.

NOMENCLATURE

za, z b, zg – numbers of teeth of sun gear, ring gear and planet gear, respectively
nw – number of planet gears
xa, x b, xg – addendum modification of sun gear, ring gear and planet gear, respectively
mn – normal module
mt – transverse module
a – centre distance
b – face width
d(a) – reference diameter of sun gear
b – helix angle
u – gear ratio
pb – pitch on the base circle
db – base diameter
rct – hob tip radius
at – transverse pressure angle
awt – operating (working) pressure angle
ea – transverse contact ratio
Dout – outer diameter of planetary gear train
E – modulus of elasticity, the Young modulus
R – effective radius of curvature
SF – factor of safety from breakage
SH – factor of safety from pitting
Ft – transverse tangential force at pinion reference circle
sF – effective tooth root stress
sH – effective contact stress
sF  lim – nominal stress number (bending)
sH  lim – allowable stress number (contact)
KA – application factor
KV – dynamic factor
KFa – transverse load factor (root stress)
KHa – transverse load factor (contact stress)
KHb – face load factor
YF – tooth form factor
YS – stress correction factor
YX – size factor
Yb – helix angle factor

473
Ye – contact ratio factor
YST – stress correction factor
YRrelT – relative surface factor
YdrelT – relative notch sensitivity factor
Zh – zone factor
ZE – elasticity factor
Z e – contact ratio factor
Zb – helix angle factor
Zn – speed factor
ZL – lubricant factor
ZR – roughness factor
ZW – work hardening factor
ZX – size factor

Subscripts
a – sun gear
g – planet gear
b – ring gear

REFERENCES

  1.  H. XU, A. KAHRAMAN: Prediction of Mechanical Efficiency of Parallel – Axis Gear Pairs.
Transactions of ASME, 129, 58 (2007).
  2.  E. I. RADZIMOVSKY, A. MIRAREFI, W. E. BROOM: Instantaneous Efficiency and coef-
ficient of friction of an involute gear drive. J. of Engineering for Industry, 1131 (1973).
  3.  N. E. ANDERSON, S. H. LOEWENTHAL: Efficiency of Nonstandard and High Contact
Ratio Involute Spur Gears. J. of Mechanisms, Transmissions and Automation in Design, 108,
424 (1986).
  4.  R. KASUBA, E. I. RADZIMOVSKY: Bending Stresses in the Bolts of a Bolted Assembly.
Experimental Mechanics, 2, 264 (1962).
  5.  T. LAZOVIC, R. MITROVIC, M. RISTIVOJEVIC: Influence of Internal Radial Clearance
on the Ball Bearing Service Life. J. of the Balkan Tribological Association, 16 (1), 1 (2010).
  6.  J. S. ARORA: Introduction to Optimal Design. McGraw-Hill, New York, 1989.
  7.  P. Y. PAPALAMBROS, D. J. WILDE: Principles of Optimal Design, Modeling and Computa-
tion. Cambridge University Press, Cambridge, UK, 1998.
  8.  S. S. RAO: Engineering Optimisation, Theory and Practice. Wiley, New York, 1996.
9.  R. L. HAUPT, S. E. HAUPT: Practical Genetic Algorithms. John Wiley and Sons, Inc. Publication,
2004.
10.  M. MARTINEZ, M. GARCIA-NIETO, J. SANCHIS, X. BLASCO: Genetic Algorithms Opti-
misation for Normalized Normal Constraint Method under Pareto Construction. Advances in
Engineering Software, 40, 260 (2009).
11.  H. J. Holland: Adaption in Natural and Artificial Systems, an Introductory Analysis with
Aplication to Biology, Control and Artificial Inteligence. The Univeristy of Michigan Press,
Ann Arbor, USA, 1975.
12.  H. ESCHENAUER, J. KOSKI, A. OSYCKA: Multicriteria Design Optimisation. Springer-Verlag,
1990.
474
13.  M. KANDEVA: The Interdisciplinary Paradigm of Tribology. J. of the Balkan Tribological
Association, 14 (2), 241 (2008).
14.  M. GEN, R. CHENG: Genetic Algorithms and Engineering Optimisation. John Wiley and
Sons, Inc., New York, 2000.
15.  A. N. NEASCA, N. ANTONESKU, D. B STOICA: Modern Solutions for Selecting the Cor-
responding Machinery Dedicated to Technological Applications. J. of the Balkan Tribological
Association, 15 (1), 45 (2009).

Received 3 March 2011

Revised 16 April 2011

475

View publication stats