Computers ind. Engng Vol. 35, Nos 3--4, pp.

523-526, 1998
© 1998 ElsevierScienceLtd. All rights reserved
Pergamon Printed in Great Britain
PII: S0360-8352(98)00149-1 0360-8352/98 $19.00 + 0.00

A Solution Method for Optimal Weight Design Problem

of the Gear Using Genetic Algorithms

Takao YOKOTA, Takeaki TAGUCHI and Mitsuo GEN

Department of Industrial and Systems Engineering

Ashikaga Institute of Technology, Ashikaga 326, Japan

Abstract
In t h i s paper, we formulate an optimal weight designproblem of a gear f o r a constrained
bending strength of gear, t o r t i o n a l strength of shafts and each gear dimension as a nonlinear
integer programing(NIP) problemand solve i t d i r e c t l y by keeping the nonlinear c o n s t r a i n t by
using an improved genetic algorithm (GA). We discuss the efficency ofthe proposed method.
© 1998 Elsevier Science Ltd. Allrights mserved.

Key words:Gear Design Problem, Nonlinear Integer Programming, Genetic Algorithm

1 Introduction result, the number of decision (design) variables

does not increase and easily get the best
The approach to solve certain design problemshas compromised solution. We discuss the efficency of
so for r e l i e d heavily on the t r a i l- a n d - c u t methods the propose method.
which, because of their methodology, take
considerable time to obtain the optimal solution 2 An O p t i m a l Weight Design
The gear is one of the machine elements. It Problem o f Gear
transmits power with accuracy to p a r a l l e l shafts,
skew shafts and intermittent action gear etc. Let us consider an 0WD problem of a gear
Therfore it has various uses in industrial with a minimum weight as shown in Fig. 1.
production. b
When designing a gear we usuallyuse the
t r a i l - a n d - c u t methods to determine factors such as
input power, rotation frequency, transmission
ratio, bending strength of the gear, tortional
strength of shafts and each gear dimension. However,
~this method does not include the method of
optimal weight design.
The mathematical model of an optimal weight
design problem of gear for minimizing objective
function includes the above mentioned design
factors [1]-[3].
In this paper, we formulate an optimal weight
design(0WD) problem of gear for a constrained
bending strength of gear, tortional strength of
shafts and each gear dimension as a NIP problem and Fig. 1: An OWD problem of the gear
solve it directly by keeping the nonlinear
constraint by using improved GA[4]-[5]. As a We formulate the optimal design problem with a

523
524 Selected papers from the 22nd ICC&IE Conference

s a t i s f i e d bending strength of spur gear, t o r t i o n a l axes(C). (bL, dlL, ~L) and (/~, dlh~, ~u) are lower/upper
strength of shafts and each gear dimension with a limit values of the design variable, respectively.
minimized gear weight F(&d~,d~z~,m). I t is the / ) 1 / / ~ and Ni/Nz are pitch diameter of pinion/gear
following NIP problem: and speed of pinion/gear, z~ is number of teeth in
gear. v, H a n d v are pitch line velocity, input
min F(b, dl,de,zl,m)=W (l) power and allowable shearing s t r e s s of shafts,
s.t. Gt(b,m) = Fs > b/ (2) respectively.
_ Fs
G2(b'm'zt) ---~e > b2 (3)
3 Representation and Evaluation
G~(d,) = d] > b~ (4) Function
6,(d~) = d~ ~ b, (S)
Gs(o~,zt)= (l+a)m~ <b5 (6) 3.1" Representation and I n i t i a l z a t i o n
2 In t h i s paper, each design variable of spur gear
is represented by the gene of chromsome Vk which
b~ < b < bU, integer
is the integer-valued decision variable Vk=
d~ < d t < a~, integer
[b, all, 4, z~, m]. pop_size denotes the size of
d~ _<d~ _<dev, integer thepopulation. The i n i t i a l population is randomly
z~ _<zt _<zv, integer generated by s a t i s f y i n g the constraints (2) to (6).
m~ < m < mv, integer
3.2 Evaluation Function
W=~ P / b n C ~ g ( l + a 2 ) _ ( / ~ _ 4 )
4 1000"
Generally, in GA the f i t n e s s value of a
(l-bw)-nadpbw-(d:+dff)b}
chromosome is obtained by the evaluation function.
Or = m(az1- 2.5)J, = 2.5m,
In t h i s paper, the evaluation function eval (Vk) was
Di =Dr-21w, bw=3.Sm, do=d2+25,
d e = 0.2509~ -do), defined using the objective function (1).
Yokota et al. [4] reported an improved GA to cope
Fs = ~ fvf.o bray,Fe = f~f,~Dtb 2z2
zl+z2 with such a problam. That is, for the constraint
01 = mz. D~ = amzI, N2 = NI /a condition, the following measure is introduced:
0; G~ < bi, i = 1,2,...,5
g2 = - - ~V =--'-"'-'-7,hi =
Dt 60×lff v ~i = (G, - b~)/bj; otherwise

b3 = 4"97x106H,b4 = 4"97×106H, The measure A i adopted here indicates how much

the left-hand side of the expression corresponding
to the constraint condition exceeds the r i g h t -
where F (b, dl, dz zl, m ) is the weight function, b hand side, and the evaluation function i s defined
and d t / ~ are face width and diameter of pinion/gear as follows:
shaft, respectively, z ~ and m are number of teeth 1 s
(8)
in pinion and module, respectively. W i s weight of
k J
gears, o is allovnable s t r e s s of gear. a is gear
This improved GA includes information about
ratio. Gi is bending strength of teeth(Fs:Lewis
unfeasible chromosomes as near as possible to
formula, pinion). G2 is sureface d u r a b i l i t y (k). F~
the f e a s i b l e region in the evaluation function
i s wear load. bi ( i = l , . . . 5) are constraint quantity We can select the best chromosome with the
each other. Dr is dedendum circle. D~, do and I are
following equation:
inside diameter of rim, outdiameter of boss and
length of boss. b,, a~ and n are thickness of web, V* = argmin{ evat(V, )} (9)
d r i l l holl diameter and number of a~. p is density mineral = min{ eval (V, )} (i0)
of gear. f,/f, is velocity/load factor, y is form Where the "arg" of argumin is an abbreviation for
factor. G..~/G4 is cubic diameter of pinion/gear argument and means that we adopt the chromosome
shaft, respectively. Q is distance between the whose evaluation returns the lowest value.
 Selected papers from the 22nd ICC&IE Conference 525
4 Algorithm Table 1: The range of variable and coefficient values

20 <b <_32 lO<d~ <30

The procedures for solving the O C D problem of
30 <d? <40 m=2.75,3t3.5~4
multiple beam by genetic algorithms is proposed in
18_z t <_25 a=4
this section. H=7.5 N 1 =1500
0=30 7=2
[Step 1] Parameter setting b2 =0.193 f, =0.389
The size of population: pop_size, probability fw =0.8 y=O.102
of crossover:Pc, probability of mutation: P., p=8 n=6
and maximum number of generation maxgen are
set. Additionally, the generation counter
genand the initial evaluation function value "(2) Mutation
mineval are both zero. Let the number of chromosomes generated by
mutation be mcnt---- 0. Generate the random
[Step 2] Generation of the initial number rk (k=1,2..... pop_size + n*ccnt) from
chromosome population the range[0, 1]. Select a gene that satisfies
The initial chromosome V k consists of elements rk < P. and set mcnt = mcnt+l. There is no
for variables b, all,~, zl,m. increase in the value of mcnt in the case of
Vk=[b,d,,dz, z,,m], k = l . . . . . pop_size (11) a mutation in the same chromosome. Mutate the
These variables generate V randomly element that is selected in the region of
s a t i s f y i n g the constraint (12). eq. (12). Let the chromosome that is newly
G, <b/,i =1,2...,5 generated be Vcc,t,~,t..
b L < b < b v, integer (3) S e l e c t i o n
Calculate the evaluation functions evaJ
d [ < d I < d~, integer
(Vl'), (1= 1,2 . . . . . ccnt~mcnt) that are newly
d~ < d 2 _<d~2, integer
generated. Select the chromosomes among the
z~ < z t < z ~ , integer parents Vk and V~' s t a r t i n g with the highest
m L < m < m U, integer values and choosing the subsequently highest
I f chromosomes s a t i s f y i n g the constraint variables. The number to be selected is pop_size.
(12) can not be found, then i t is assumed that Let these chromosomes be the Vk for the
there are no f e a s i b l e solutions for the generation where the chromosome, that has the
problem, and the algorithm terminates. same structure, is not selected.

[ S t e p 3] Calculation of fitness value [Step 5] Calculate the best chromosome

The generation counter is updated(gen=gen+l), Update the best chromosome V* and the
and the f i t n e s s chromosome is calculated from corresponding evaluation function from eq. (9).
the evaluation function (8).
[ S t e p 6] Next generation If gen maxgen,
[Step 4] Genetic operators then go to Step 3. Otherwise output V* and stop.
(I) Crossover
Let the number of chromosomes generated be ccnt 5 Numerical Example
= 0. Create the random number rk (k =1, 2.....
pop_size) from the range[0, I]. Select Vk that In this paper, when we are given the design
satisfies rk < Pc. Make a pair of V, and set parameters, we discuss the optimal gear design
ccnt=-ccnt+2. Choose the position for crossover problem for minimizing the weight as shown in Fig. 1
at random and exchange each other' s elements. and Table I.
Let offspring be V,.c,~,and V,~.,,respectively. Wesolve the OWD problem of the gear by using
our proposed algorithm.
526 Selectedpape~from &e22ndlCC&lEConference
We set up the following i n t i a l parameters for Table 2: The r e s u l t of calculation and standard
the numerical example: gear' s specification
result numerical
pop size = 20, Pc = 0.4, Pm= O.1,maxgen = 1000 example
t o t a l g e a r ' s weight: W 3512. 6 6802.5
We apply the proposed method to this problem and )iinon' weight: W l 233.8 723.5
repeat the process for 1,000 generations. At the gear' weight: W 2 2868.0 6079

648th generation of the chromosome and constraint fase width: b 24 32

diameter of pinion s h a f t : dl 30 32
design variables, the following results are diameter of gear s h a f t : d2 30 45
obtained: number o f t e e t h ( p i n i o n ) : zj 18 20
number of t e e t h ( g e a r ) : z z 72 81
V'=[ b, d], de, z], m ] module: m 2.75 3
=[ 24 30 30 18 2 . 7 5 ] )itch circle(pinion): Dl 49.5 60
)itch circle(gear): Ds 198 243
The above values obtained are the best compromised
between the axes: C 123.75 151.5
solutions. The corresponding evaluation function sureface durabulity: k 0. 374 0. 180
value, that is, the minimized weight of the gear, dedendum circle(gear): Dr 191.1 235.5
was F (b, d~, dz, zt, m )=3513.6. inside diameter of rim: di 177.4 212
thickness of web: b, 9.63 I0
outdiameter of boss: do 55 75
6 Evaluation drill holls: dp 30.6 16

By applying the method proposed in this paper gears (W) /space area improved by 48. 8% and 33.6%,
to the numerical example described in the previous and confirmed the efficency of the proposed method.
section, it was possible to obtain results shown
in Table 2. And we also include the specification References
for spur gear in Table 2. [i] A.0syczka:"Pareto Set Distribution Method for
After comparing the gear figures and our Multicriteria Design Optimization Using Genetic
calculation results, the weight of gears (W)/mean Algorithm", Seminar Note, 12pp, Dept. of ISE, Ashikaga
radius(R) improved by 44. 8% and 18. 5%. Here, the Inst. of Tech., 1997.
improved mean radius 18. 5% means a decrease in [2] Yokota, T. ,T. Taguchi and M. Gen;"GA-Based Method for
space area of 33.6~ Optimal Weight Design Problem with a Duble Coil Spring",
Proc. of 5th Europian Congress on Intelligent
7 Conclusion Techniques and Soft Computing, putting Aachen, 1997. 9
[3] Onishi, K. : Elementary Mechanical eesgn, Rikougakusya,
In this paper, we formulate an OWD problem of 1994 (in Japanese). "Fuzzy Multi-Objective
gear for a constrained bending strength of teeth, Combinatorial Optimization by an Improved Genetic
tortional strength of shaft and each gear dimension Algorithm Model", J. of Japanese Society for Fuzzy
as a NIP problem and solve i t directly by keeping Theory and Systems}, 6, 3, pp, 177-186, 1994 (in
the nonlinear constraint by using improved GA. Japanese).
We discuss the efficency of the proposed method. [4] Yokota T., M. Gen, K. Ida and T. Taguchi: "Optimal
As a result, the number of decision (design) Design of System Reliability By Modified Genetic
variables did not increase, and easily got Algorithm", Trans. Inst. Electronics, Information and
thebest compromised solution. Conm Engineers, A, J78-A, 6, pp. 702-709, 1995 (in
When a chromosome is not contained in the Japanese).
feasible region, i t will include information about [5] Gan, M. and R. Chang: Genetic Algorithms and
the infeasible region's chromosome in the Engineering Design, John Wiley & Sons, New York, 1997.
evaluation function in order to improve i t s search
efficency.
As a result, we obtained a minimumweight of
F(b, d~, d~, z~, m )=3513.6, then the weight of