Sie sind auf Seite 1von 10

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

net/publication/222795239

A genetic approach to automate preliminary design of gear drives

Article  in  Computers & Industrial Engineering · October 2009


DOI: 10.1016/j.cie.2009.04.006 · Source: DBLP

CITATIONS READS

39 507

2 authors, including:

Metin Zeyveli
Karabuk University
13 PUBLICATIONS   45 CITATIONS   

SEE PROFILE

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

Joining of different material by Clinching rivet View project

All content following this page was uploaded by Metin Zeyveli on 16 October 2017.

The user has requested enhancement of the downloaded file.


Computers & Industrial Engineering 57 (2009) 1043–1051

Contents lists available at ScienceDirect

Computers & Industrial Engineering


journal homepage: www.elsevier.com/locate/caie

A genetic approach to automate preliminary design of gear drives q


Cevdet Gologlu a,*, Metin Zeyveli b
a
Faculty of Engineering, Karabuk University, 78050 Karabuk, Turkey
b
Faculty of Technical Education, Karabuk University, 78050 Karabuk, Turkey

a r t i c l e i n f o a b s t r a c t

Article history: Determination of volume or weight of a gearbox is an important issue in preliminary design of power
Received 15 October 2006 transmission applications. Trial and error procedure or some gear standards information sheets are com-
Received in revised form 17 February 2009 monly used in traditional design. The purpose of this paper is to automate preliminary design of gear
Accepted 13 April 2009
drives by minimizing volume of gear trains. A stochastic approach Genetic Algorithm (GA) was applied
Available online 21 April 2009
to a parallel axis two stage helical gear trains problem. Static and dynamic penalty functions were intro-
duced to the objective function for handling the design constraints. The results were compared with a
Keywords:
deterministic design procedure developed. GA based approach produced quite well results promptly sup-
Preliminary design
Gear pairs
plying preliminary design parameters of gear drives for different gear ratios to the designer.
Minimum volume Ó 2009 Elsevier Ltd. All rights reserved.
Genetic Algorithm

1. Introduction liminary design of multi stage gear drives was proposed (Chong,
Bae, & Park, 2002). A four-step algorithm was run iteratively until
Solving engineering problems, especially design optimization, a desirable solution is found. The steps in the algorithm were
involve multiple and conflicting objectives. Moreover, design vari- mainly conducted manually, by random search and generate and
ables could be in the forms of continuous, 0–1, integer and discrete test methods. In the last step, simulated annealing algorithm was
variable. These considerations make solving mechanical design used for minimizing geometrical volume of a gearbox by means
problems very complex. As power transmitting machine elements of integrating dimensional and configuration design process. An
gear drives are used in various industries. A compound group of optimal weight design problem of a gear pair system was studied
gears is employed to achieve a specific gear ratio between the dri- using GA (Yokota, Taguchi, & Gen, 1998). The system was able to
ver and driven shafts (Fig. 1). In conventional methods, designing a find the number of design variables considering specified con-
gear drive as reduction unit requires very large number of calcula- straints. A generalized optimal design formulation to gear trains
tions based on recommended gear standards, trial and error meth- was presented (Thompson, Gupta, & Shukla, 2000). In this research,
ods, etc. This time consuming process may often finish up with the tradeoff between surface fatigue life and minimum volume
inadequate design outcomes. using a basic multiobjective optimization procedure was studied.
The research on design of gear drives has mainly focused on The results were presented as a Pareto optimal set representing a
dimensional design on single stage gear drives. Researchers have collection of optimal designs. A computer aided design of gears ap-
developed several applications using different design and calcula- proach was proposed (Marcelin, 2001) to optimize one stage gear
tion methods (Deb, Pratap, & Moitra, 2000; Su & Wakelam, pair. GA was employed for minimizing gear volume by reducing
1998). A gearbox producing the required output speed was de- the distance between the centers of gear pairs, and other parame-
signed by using GA (Pham & Yang, 1993). The objective function ters such as transmitting power, reduction ratio. In another study,
of the study was to determine the number of shafts and number an expert system involving a GA module was developed (Abersek &
of teeth for each gear by GA. The objective function was con- Popov, 2004). This GA module was used for optimizing volumes of
strained by number of teeth of gear, maximum transmission ratio, pitch cylinders of gears for a single reduction gearing system.
and maximum number of shafts. An algorithm for automating pre- The volume or weight of a gearbox is an essential parameter in
many applications which require small size. Thus, the purpose of
this research is to automate and present a GA based design optimi-
zation for a parallel axis two stage gearbox. The objective in the
q
This manuscript was processed by Area Editor Gursel A. Suer. optimization is to minimize volume of gear trains subject to bend-
* Corresponding author. Tel.: +90 370 433 82 10 (pbx); fax: +90 370 433 82 04.
E-mail addresses: cgologlu@karabuk.edu.tr (C. Gologlu), mzeyveli@karabuk.
ing strength, contact stress, face width, and number of pinion and
edu.tr (M. Zeyveli). gear teeth based on DIN 3990. The problem definition and GA for-
URL: http://muh.karabuk.edu.tr (C. Gologlu). mulation will be given with a design example. Static and dynamic

0360-8352/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.cie.2009.04.006
1044 C. Gologlu, M. Zeyveli / Computers & Industrial Engineering 57 (2009) 1043–1051

pairs at the stages are presented by b1 and b2. The material volume
of a helical pinion F1 is defined as given in Eq. (4).
hp i
F1 ¼ ðmn1 z1 Þ2 b1 ð4Þ
4
In a two stage gearbox configuration, total volume of two pairs
of helical gears FT can be written as a function of module, number
of teeth and width of gear (Eq. (5)).

FT ¼ F1 þ F2 þ F3 þ F4
hp p i
¼ ððmn1 z1 Þ2 þ ðmn1 z2 Þ2 Þb1 þ ððmn2 z3 Þ2 þ ðmn2 z4 Þ2 Þb2 ð5Þ
4 4

where F1, F2, F3 and F4 are material volumes of pinion at first stage,
gear at first stage, pinion at second stage, and gear at second stage,
respectively.
A number of constraints is introduced into the objective func-
Fig. 1. A parallel axis gear drive.
tion to be minimized. The constraints are used to provide suitable
design choices or as subfunctions to restrict the objective function
in a way that suitable contents are incorporated. The failure types
penalty functions are introduced to the objective function for han- that are mostly encountered are tooth fracture and surface fatigue
dling design constraints. The results obtained by GA are compared in a transmission gear (Akinci, Yilmaz, & Canakci, 2005). Thus,
with a deterministic design procedure developed. bending strength and contact stress restrictions are main con-
straints in designing gears. The others are related to standards
2. Problem definition and premises of sizing gears. These are defined as given in Eq. (6).
8
The effective design variables used in optimizing volume of gear > Bending strength constraint;
>
>
trains are module, number of teeth and face width. In the parallel >
> Contact stress constraint;
>
>
>
< Face width constraint;
axis gearbox, helical teeth are intentionally preferred in order to
increase mechanical performance. The vector of design variables gðj Þ ¼ ð6Þ
> Module constraint;
>
is shown in Eq. (1). >
>
>
> Pinion teeth constraint;
8 >
>
:
< moduleðmn Þ
> Gear teeth constraint;
xðiÞ ¼ number of teethðzÞ ð1Þ
>
: where j = 1 , . . . , number of constraints.
face widthðbÞ
The module variable mn is normal module of helical gear and is 2.2. Forming penalty functions
defined at the interval of 1–5.5 with total of 16 modules. In this
study, larger module values are not taken into consideration as In GA based optimization, the main concern is the constraint
they do not contribute to the objective function. In the design of treatment. The most common way of incorporating constraints
gear pairs, the materials of pinion and gear are accepted as the into GA is done by penalty functions. Two main kinds of penalty
same and design process is conducted based on pinion. Thus num- functions in classical optimization are interior and exterior penalty
ber of teeth of gear is defined subject to number of teeth of pinion functions. In interior penalty functions, the penalty term value is
and gear ratio. The interval for the number of teeth of pinions is chosen as small at points far from the constraint boundaries and
17 6 z1;3 6 24. z1 and z3 are the numbers of teeth of pinions at first tends to infinity as the boundaries are approached. It requires ini-
stage and second stage, respectively. The face width b is deter- tial feasible solutions. In exterior penalty functions, initial feasible
mined at the beginning of design process with respect to face solution is not required as the penalty term value is chosen zero for
width factor um (Eq. (2)) and it is selected as 20 6 um 6 40 based feasible solutions and it increases as the constraint violations in-
on recommendation (Bozaci, Ilknur, & Colak, 2001). crease (Puzzi & Carpinteri, 2008). In the exterior penalty function,
either static or dynamic penalty functions is employed. Static pen-
b ¼ um mn ð2Þ
alty function uses the same value of penalty term throughout the
whole selection process. The later utilizes an increasing penalty
2.1. Forming objective function term in order to boost the selective pressure during the evaluation
process.
An objective function is a quantity to be minimized or maxi- In case of static penalty function method constraint violations
mized by exploring a search space under the imposed constraints are converted into a penalty function PS (Eq. (7)), where PS is only
(Saruhan, Rouch, & Roso, 2004). In this study, minimization of a function of x and not dependent on the current generation num-
material volume is the objective function Fobj and it is defined as ber q in GA.
shown in Eq. (3).
X
Nconst
8 PS ðxÞ ¼ rj ðmax½0; g j ðxÞÞ ð7Þ
> Min: material v olume; mn1 ; z1 ; b1
>
> j¼1
< ðfor 1st stage pinion and gearÞ
F obj ¼ ð3Þ where Nconst is the number of inequality constraints applied to the
> Min: material v olume; mn2 ; z3 ; b2
>
>
: problem; rj penalty coefficients for the jth constraint; g is the con-
ðfor 2nd stage pinion and gearÞ
straint considered. In GA with static penalty, a number of con-
where mn1 is the module of pinion and gear at first stage and mn2 is straints is penalized with the user defined coefficients in such a
the module of pinion and gear at second stage. The width of gear way that the violations are prevented. On the other hand, the
C. Gologlu, M. Zeyveli / Computers & Industrial Engineering 57 (2009) 1043–1051 1045

dynamic penalty PD can be defined as shown in Eqs. (8)–(10) (Joines where rb is bending stress and the allowable stress can be found as
& Houck, 1994): ralw = rD/Kc (N/mm2) while dynamic stress is rD ffi 0:55rK , ultimate
tensile strength rK, stress concentration factor Kc, tangential force Ft
X
Nconst
PD ðx; qÞ ¼ r aq
b
dj ðxÞ ð8Þ at pitch diameter, overlap ratio e, dynamic velocity factor Kd, and
j¼1 form factor Kfe (Bozaci et al., 2001). In helical gears, allowable sur-
 face pressure is presented by Palw and can be defined as
0 if x is feasible
dj ðxÞ ¼ ð9Þ palw ¼ 0:25HB where HB is Brinell hardness. The maximum surface
g i ðxÞ 1 6 j 6 Nconst
pressure is Pmax (Eq. (16)) and the others are material factor Km,
r q ¼ Cq ð10Þ flank transverse coefficient Ka, tooth overlap factor K e , tooth slope
where q is the generation number; C, a, and b are user defined factor Kb, and i is gear ratio. d is pitch diameter (Eq. (17)) and is re-
parameters for adjusting the penalty scales. Setting appropriate val- lated to transverse module mt and number of teeth z. b is the helix
ues to the parameters of C, a, and b is a problem dependent task. angle. Then, the constraints can be formed as in Eqs. (18)–(29).
Then the penalty function is added to the global objective function g 1 ðxÞ ¼ K d K fe K c F t1  eb1 mn1 ð0:55rK Þ 6 0 ð18Þ
Fglobalobj. In case of static penalty function, the global objective func-
g 2 ðxÞ ¼ K d K fe K c F t2  eb2 mn2 ð0:55rK Þ 6 0 ð19Þ
tion with constraints can be presented in Eq. (11). In this study, the  
user defined parameter values C, a, and b are taken as 0.5, 0.4–1, z2 þ z1
g 3 ðxÞ ¼ K d ðK m K a K e K b Þ2 F t1  b1 mn1 z2 ðPalw Þ2 6 0 ð20Þ
and 0.4–1, respectively. z1
 
z4 þ z3
X
Nconst g 4 ðxÞ ¼ K d ðK m K a K e K b Þ2 F t2  b2 mn2 z4 ðPalw Þ2 6 0 ð21Þ
F globalobj ¼ F T þ r j ðmax½0; g j ðxÞÞ ð11Þ z3
j¼1 g 5 ðxÞ ¼ ðð20mn1 Þ  b1 Þ 6 0 ð22Þ
Similarly, in case of dynamic penalty function the global objec- g 6 ðxÞ ¼ ðb1  ð40mn1 ÞÞ 6 0 ð23Þ
tive function becomes as in Eq. (12). g 7 ðxÞ ¼ ðð20mn2 Þ  b2 Þ 6 0 ð24Þ
X
Nconst g 8 ðxÞ ¼ ðb2  ð40mn2 ÞÞ 6 0 ð25Þ
b
F globalobj ¼ F T þ r aq dj ðxÞ ð12Þ g 9 ðxÞ ¼ ð17  z1;3 Þ 6 0 ð26Þ
j¼1
g 10 ðxÞ ¼ ðz1;3  24Þ 6 0 ð27Þ
Constraint optimization problem becomes unconstrained opti- g 11 ðxÞ ¼ ð35  z2;4 Þ 6 0 ð28Þ
mization problem by introducing penalty functions to the objective
g 12 ðxÞ ¼ ðz2;4  145Þ 6 0 ð29Þ
function. With aid of subtracting the function from a large enough
positive number F, the minimization objective function with static The constraints above are classified into the bending stress
penalty and dynamic penalty can be defined in Eqs. (13) and (14), constraints, the contact stress constraints and the face width con-
respectively. straints (see Table 1). A feasible design must satisfy all constraints
! that all gj(x) function values must be less than or equal to zero. The
X
Nconst
F obj;i ¼ F  F i þ r j ðmax½0; g j ðxÞÞ ð13Þ constraints defined in Eqs. (18)–(29) are inequality constraints and
j¼1 there is no equality constraint in the problem considered.
!
X
Nconst
b
F obj;i ¼ F  F i þ r aq dj ðxÞ ð14Þ 3. A design practice
j¼1

where Fi represents F1, F2, F3, and F4. In addition to the constraints some other coefficients and input
As stated earlier, forming of a robust penalty function is not variables such as transmitted power, input speed, total gear ratios,
easy task and is problem dependent (Puzzi & Carpinteri, 2008; etc. are employed for a design practice (see Table 2). Suitable rep-
Saruhan et al., 2004). The solutions depend on proper values of resentation of design vector in coding is key to success in GA. The
penalty coefficients. The weighting of constraints by penalties is formula of mapping design variables to binary equivalences is
difficult to adjust and setting right values are essential to solutions shown in Eq. (30) (Lin & Hajela, 1992). In the equation, x(i)upper
compromised (Gen & Cheng, 2000). Ideally, the penalty should be and x(i)lower are upper and lower bound for design variable vectors,
kept as low as possible, just above the limit, below which infeasible respectively. The l is the number of digits in binary string. The con-
solutions are optimal (Davis, 1987). Making the penalty too high tinuous design variables are represented as a binary string discre-
may direct GA to feasible region very quickly by discouraging the tised to a precision of E = 0.01. As mentioned before, the modules
exploration of the infeasible region and leads to premature conver- larger than 5.5 are not considered in minimization process as they
gence to local optima. On the contrary, making the penalty too low
results in a great deal of search time exploring the infeasible re-
gions (Smith & Coit, 1997). Table 1
Constraints.
2.3. Forming constraints Constraints Related design parameters
g1(x) Bending stress of first pair gears
The formula of bending strength (Eq. (15)), contact stress (Eqs. g2(x) Bending stress of second gear pairs
(16) and (17)) and face width (Eq. (2)) are used for forming con- g3(x) Contact stress of first pair gears
straints for preliminary design of gearbox with helical gears. g4(x) Contact stress of second pair gears
g5(x) First stage pinion width
Ft rD g6(x) First stage gear width
rb ¼ K d K fe 6 ð15Þ
ebmn Kc g7(x) Second stage pinion width
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi g8(x) Second stage gear width
KdFt i þ 1 g9(x) First stage pinion teeth number
pmax ¼ K m K a K e K b 6 palw ð16Þ
bd i g10(x) Second stage pinion teeth number
mn z g11(x) First stage gear teeth number
d ¼ mt z ¼ ð17Þ g12(x) Second stage gear teeth number
cos b
1046 C. Gologlu, M. Zeyveli / Computers & Industrial Engineering 57 (2009) 1043–1051

Table 2 Table 4
Working conditions and coefficients for a preliminary design example. Crossover and mutation probabilities tried with static penalty.

Transferred power (kW) 7.5 Gear ratio Crossover prob. Mutation prob.
Input speed (rpm) 1800
0.1 0.01 0.001
Total gear ratios, i 6, 11, 16, 21
Material (cementite steel) 16MnCr5 Minimum volumes (cm3)
Manufacturing process Fine work
21 0.5 14527.59 13165.08 15380.41
Usage Electricity motor
0.7 12310.39 13156.73 12745.23
Brinell hardness (N/mm2) 1460
0.9 15276.06 14112.91 12052.54
Ultimate tensile strength, rK (N/mm2) 1100
16 0.5 9913.42 9207.90 9664.21
Working factor, Ko 1.25
0.7 9004.25 9417.73 9107.96
Overlap ration, e 1.6
0.9 9763.76 9374.52 8826.93
Stress concentration factor, Kc 1.5
11 0.5 6264.37 6550.50 5886.52
Material factor, Km (N/mm2) 271.11
0.7 5788.43 5616.07 5675.02
Flank transverse coefficient, Ka 1.76
0.9 6212.67 5623.16 5395.69
Tooth overlap factor, K e 0.79
6 0.5 2898.99 3106.19 2990.93
Helical angle, b 18°
0.7 2937.81 3003.56 4139.04
Tooth slope factor Kb 1
0.9 2794.21 2905.71 2849.76
Available modules (mm) 1, 1.125, 1.25, 1.375, 1.5, 1.75,
2, 2.25, 2.5, 2.75, 3, 3.5, 4, 4.5, 5, 5.5

Crossover and mutation operators in GA have direct effect on


do not make sense in reduction of volumes. The modules are de- obtaining optimal solutions. The probability of mutation should
fined discrete with length of four bits. The variable of number of be selected carefully in order to avoid loosing good features of
teeth of pinion is presented with three bits at the interval of 17– strings (Goldberg, 1989). The crossover and mutation probabilities
24 as continuous integer design variable. The interval of 35–145 tried are given in Tables 4 and 5 and the minimum volumes found
is taken for the number of teeth of gear presenting seven bit are depicted as bold. In the GA with static penalty function, the
information. crossover and mutation probabilities were accepted as 0.9 and
0.001, respectively. In the tries for dynamic penalty (Table 5), the
2l P ½xðiÞupper  xðiÞlower =E þ 1 ð30Þ most suitable crossover and mutation probabilities were found to
be the crossover and mutation probabilities of 0.9 and 0.01 (for
In Table 3 the variables used are given as binary strings. A single
gear ratio of 21), 0.9 and 0.001 (for gear ratio of 16), 0.9 and
31-bit string represents an example of solutions from the search
0.001 (for gear ratio of 11), and 0.7 and 0.01 (for gear ratio of 6).
space. The initial GA parameters in this study were chosen as pop-
For the tests, a P4 computer with 2 GHz processor with a 1 Gb
ulation 50; number of generations 100, 200, and 500; length of
RAM was used.
chromosome 31; probabilities of crossover 0.5, 07, and 0.9, and
probabilities of mutation 0.1, 0.01, and 0.001.
4. Deterministic design procedure
Selection for reproduction in GA is done in several ways. The
simplest scheme is construction of a roulette wheel with slots
As it is known that GA is a stochastic method maintaining a
sized according to objective function values. However, there are
population of encoded solutions, and it guides the population to-
some disadvantages as the selection process can lead to premature
wards the optimum solution (Goldberg, 1989). In order to verify
convergence on local optimum. So that the tournament selection
the solutions obtained by GA, a deterministic design procedure
mechanism was adopted with the tournament size of 6. The repro-
was developed considering the same constraints of bending
duction process only discards the poorer strings. Changing of the
strength, contact stress, face width, and pinion and gear teeth
strings is performed by the crossover operator. Since one point
numbers (Zeyveli, 2005). It is actually a full scan algorithm to ob-
crossover has restriction when it cannot combine certain features
tain minimum material volumes of the gear trains based on gear
encoded on chromosomes, use of two-point crossover operator be-
ratios for the design problem under consideration. The flowchart
comes a solution. Thus, a two-point crossover operator was em-
of the procedure developed is shown in Fig. 2. With this procedure,
ployed and the GA was run with the specified number of
the experiments are carried out within the same initial coefficients
generations. It was seen that when static penalty function is em-
and input variables. It is well known that in a gear reduction unit,
ployed, the objective function curve do not change too much after
generation 100. Thus, the generation number 200 was admitted as
a compromised one for the objective function to which static pen-
alty function employed. Similarly, the number of generations in the Table 5
GA with dynamic penalty was accepted as 200. Crosssover and mutation probabilities tried with dynamic penalty.

Gear ratio Crossover prob. Mutation prob.


0.1 0.01 0.001

Table 3 Minimum volumes (cm3)


GA coding of design variables. 21 0.5 12714.76 11714.80 12567.36
0.7 12059.11 12027.10 12445.43
Design variables vectors Vectors Randomised String
0.9 12706.53 11557.16 11666.59
binary digits length l
16 0.5 9329.36 9543.72 9723.97
Module m1 (1st stage) x(1) 1010 4 0.7 10320.78 8602.06 9207.13
Number of teeth of pinion z1 (1st stage) x(2) 101 3 0.9 9168.19 8727.29 8494.60
Tooth (face) width b1 (1st stage) x(3) 10010 5 11 0.5 6198.22 5523.47 5818.25
Number of teeth of gear z2 (1st stage) x(4) 1011100 7 0.7 6134.45 5717.21 6082.97
Module m2 (2nd stage) x(5) 0111 4 0.9 6151.32 5903.44 5405.48
Number of teeth of pinion z3 (2nd stage) x(6) 011 3 6 0.5 2930.59 2884.12 2893.95
Tooth (face) width b2 (2nd stage) x(7) 11001 5 0.7 2949.93 2806.45 2945.01
A single 31-bit individual (chromosome) 1010101100101011100011101111001 0.9 3009.19 3136.26 3217.19
C. Gologlu, M. Zeyveli / Computers & Industrial Engineering 57 (2009) 1043–1051 1047

Start

Initial design variables


i1=1.1, ms =0, zj =0

Determination of gear ratio


-I-
ik =ik +0.1, i(k+1)=i T/ik

Determination of module at
stage k -II-
mks :=mk(s+1)

Determination of pinion teeth


number at stage k -III- k:=1
zkj :=zk(j+1)

Coefficient selection, Sizing,


-IV-
Strength Control k:=k+1
ms :=0

-V- N
gk1(x)≤0
gk2(x)≤0 Y
k>2
...
gk12(x)≤0
Y
N
Calculate total
Y -VI-
volume
17≤zj ≤24

N Print variables and


Y minimum volume
1≤ms≤5.5

N
N
-VII- ik<1.2

i = Gear ratio, i T = Total gear ratio


Y
k = {1, 2}, Stages
ms = {1, 1.2, 1.5, …,5.5}, Number of modules
End
z j = {17, 18,…, 24}, Number of teeth of pinion

Fig. 2. Flowchart of design procedure.

Table 6
Results of design procedure for two stage helical gear pairs.

Transmitted power Input speed Output speed Gear Stage Module Teeth Tooth width Minimum material volume Execution time
(kW) (rpm) (rpm) ratio (mm) number (mm) (cm3) (s)
7.5 1800 85.7 21 1st 2 z1 19 79.8 11488.3 49
z2 91
2nd 3 z3 25 99.9
z4 109
112.5 16 1st 2 z1 19 79.8 8211.89 47
z2 91
2nd 3 z3 25 105.6
z4 83
163.6 11 1st 1.75 z1 24 69.83 5297.81 46
z2 68
2nd 3 z3 25 80.7
z4 76
300 6 1st 1.75 z1 25 68.83 2780.97 48
z2 65
2nd 3 z3 24 68.7
z4 55
1048 C. Gologlu, M. Zeyveli / Computers & Industrial Engineering 57 (2009) 1043–1051

45000 and second stages is recorded. Therefore, the design procedure


First generation conducts a scan with designated coefficients and constraints using
40000
Fitness function (global) cm^3

Last generation
every single values of module, number of teeth, and gear ratio. It
35000
reveals the lowest material volume which meets constraint
30000 requirements.
25000 The disadvantage of the design procedure is that it needs longer
20000 computing time than GAs in order to end the procedure. Because of
the nature of the design process, it is an iterative procedure and
15000
there could potentially be more than one solution. The minimiza-
10000 tion objective of the procedure needs all solutions fitting the con-
5000 straints should be determined and then be refined for the
0
minimum volume. The results of the design procedure for a paral-
0 10 20 30 40 50 lel axis two stage helical gear pairs are shown in Table 6. The exe-
Population cution time needed is roughly 50 s for each gear ratio.

Fig. 3. Distribution of fitness function values for first and last generations of a
sample case. 5. Results and discussion

The preliminary design of two stage helical gearbox problem


initial input speed and output speed determine the total gear ratio was mapped into GA. In Fig. 3, the distribution of fitness values
iT. Thus, iteration starts with the gear ratio of the first stage i1. It is for first and last generations for a sample case is given for popula-
scanned from i1 = 1.1 (with an increment of 0.1) to i2 < 1.2 based on tion size 50 and the gear ratio 6. It is seen in the first generation
i2 = iT/i1 and the gear ratios of first and second stages are deter- that fitness values are different and scattered to one another while
mined (-I-). Then, for the first stage, a module starting from the the same minimum (best) fitness values are converged in the last
smallest among the available modules is chosen (-II-). Later, the generation.
pinion teeth number beginning from the smallest teeth number For the gear ratios of 6, 11, 16, and 21, the average and mini-
is taken and the gear is consequently computed based on mum fitness values with static penalty based on the generations
z2 = i1  z1 (-III-). Then, the strength controls are conducted (-IV-). are pictured in Fig. 4. The results show that the material volumes
If the values of module, pinion and face width prove the con- rapidly approach to a uniform convergence. The optimal values
straints, then the search is directed to the second stage (-V-). The for 6, 11, 16, and 21 gear ratios are obtained at the generations
selection of the module and pinion number for the second stage of 56, 29, 10, and 133, respectively. The design parameters ob-
is done as it is at the first stage. If the constrained are proved, tained by GA with static penalty for a parallel axis two stage helical
the design variable values and volumes for the stages are deter- gear pairs are shown in Table 7. The execution time is 4 s for each
mined (-VI-). The process is repeated until i2 < 1.2 (-VII-). Finally, gear ratio. Similarly, for the same gear ratios, the average and best
the minimum material volume obtained from the volumes of first fitness values with dynamic penalty based on generations are

37000 37500

32000 Average fitness 32500 Average fitness


Fitness function cm3
Fitness function cm3

Min. fitness Min. fitness


27000 27500

22000 Gear ratio = 6 22500 Gear ratio = 11


Crossover prob. =0.9 Crossover prob. =0.9
17000 Mutation prob. =0.001 17500 Mutation prob. =0.001

12000 12500

7000 7500

2000 2500
1 51 101 151 1 51 101 151
Generations Generations

37500 37500

32500 32500 Average fitness


Average fitness
Min. fitness
Fitness function cm3

Fitness function cm3

Min. fitness
27500 27500
Gear ratio = 21
22500 Gear ratio = 16 22500
Crossover prob. =0.9
Crossover prob. =0.9
17500 Mutation prob. =0.001
Mutation prob. =0.001 17500

12500 12500

7500 7500

2500 2500
0 50 100 150 200 0 50 100 150 200
Generations Generations

Fig. 4. Average and minimum fitness values for gear ratios with static penalty.
C. Gologlu, M. Zeyveli / Computers & Industrial Engineering 57 (2009) 1043–1051 1049

Table 7
Results of GA with static penalty for two stage helical gear pairs.

Transmitted power Input speed Output speed Gear ratio Stage Module Teeth number Tooth width (mm) Minimum material Execution
(kW) (rpm) (rpm) (mm) volume (cm3) time (s)
7.5 1800 85.7 21 1st 3 z1 17 60 12052.54 4
z2 119
2nd 4 z3 23 80
z4 69
112.5 16 1st 2.25 z1 22 58 8826.93 4
z2 99
2nd 3 z3 24 80
z4 85
163.6 11 1st 2 z1 22 64 5395.69 4
z2 75
2nd 3.5 z3 22 72
z4 71
300 6 1st 2 z1 22 69.03 2849.76 4
z2 57
2nd 3 z3 24 69.03
z4 56

given in Fig. 5. While the generations progress, a uniform conver- their objective function values and related constraints. The proved
gence occurs as seen at the average and best fitness plots. The opti- constraints are signed with ‘‘+” and the others were signed with
mal values for 6, 11, 16, and 21 gear ratios are obtained at the ‘‘”. It is seen that some of the constraints are not satisfied by
generations of 68, 24, 96, and 25, respectively. The design param- the solutions found in the first generation. However, in the last
eters obtained by GA with dynamic penalty are depicted in Table generations, the overall constrains are proved by the solutions
8. In this case, the execution time is 11 s. Calculation of the penalty found. The results obtained by GA with static and dynamic func-
values at the objective function almost triples the execution time tions (Tables 7 and 8) satisfy all the constraints considered.
on static penalty. The both tables also point that the lower gear ra- The comparison of minimum volumes for GA with static and
tios result in the lower material volumes at the same transferred dynamic penalties and the design procedure developed is given
power and input speed values. in Fig. 6. Comparing with the design procedure, the minimum vol-
The solutions obtained from GA with static and dynamic penal- umes obtained by GA with static penalty at the gear ratios of 6 and
ties were also checked against the constraints whether they are 11 reveal around 98% of successes (Table 10). As it is known, the
fulfilled or not. In Table 9, parts of first and last chromosomes for design procedure scans all design parameter variables throughout
a sample case (gear ratio of 11 with static penalty) are given with the ranges specified. However, at higher gear ratios of 16 and 21,

14500
14000
Average fitness Average fitness
12000
Fitness function cm3

Min. fitness
Fitness function cm3

11500 Min. fitness

10000
Gear ratio=6 Gear ratio=11
Crossover prob.=0.7 Crossover prob.=0.9
8000 8500
Mutation prob.=0.01 Mutation prob.=0.001

6000
5500
4000

2000 2500
0 50 100 150 200 0 50 100 150 200
Generations Generations

29500
30000
Average fitness 26500 Average fitness
Min. fitness 23500 Min. fitness
Fitness function cm3
Fitness function cm3

25000
Gear ratio=16 20500
Crossover prob.=0.5 Gear ratio=21
20000 Mutation prob.=0.001 17500 Crossover prob.=0.9
Mutation prob.=0.01
14500
15000
11500
8500
10000
5500
5000 2500
0 50 100 150 200 0 50 100 150 200
Generations Generations

Fig. 5. Average and minimum fitness values for gear ratios with dynamic penalty.
1050 C. Gologlu, M. Zeyveli / Computers & Industrial Engineering 57 (2009) 1043–1051

Table 8
Results of GA with dynamic penalty for two stage helical gear pairs.

Transmitted power Input speed Output speed Gear Stage Module Teeth Tooth width Minimum material volume Execution time
(kW) (rpm) (rpm) ratio (mm) number (mm) (cm3) (s)
7.5 1800 85.7 21 1st 3 z1 19 42 11557.16 11
z2 138
2nd 5 z3 20 70
z4 58
112.5 16 1st 4 z1 17 34 8494.60 11
z2 87
2nd 4 z3 19 74
z4 59
163.6 11 1st 2.25 z1 24 44 5405.48 11
z2 81
2nd 3.5 z3 24 60
z4 78
300 6 1st 2.5 z1 21 49.68 2806.45 11
z2 51
2nd 3.5 z3 20 67.10
z4 49

Table 9
A sample case for the generations regarding satisfaction of constraints.

Material volume m1 z1 z2 b1 g1(x) g3(x) g5,6(x) m2 z3 z4 b2 g2(x) g4(x) g7,8(x)

First generation
11874147.53 1011110000111111010001111010000 1.00 22 141 32.00 +  + 1.38 21 36 64   
2171621.95 0110111001110000101100111100001 1.25 23 39 56.00 +   2.25 23 150.00 44 +  
7876063.26 0100100101011010010011101110010 1.25 24 106 46.00 +  + 1.75 18 45.00 36   +
3647078.29 1100111100110000110100110101010 3.00 19 40 56.00 + +  1.38 24 120.00 68 +  
9489554.12 1011010000111110000111111110000 1.00 24 132 80.00 +   1.38 21 42.00 62   
6025496.94 0011010101010011111000101110000 1.00 24 62 22.00 +  + 1.75 22 94.00 30 +  
1650904.7 0101001111000000001000111010000 1.00 22 36 24.00 +  + 4.00 20 130.00 38 + + 
1783404.32 1101100111001111011000001111011 3.50 24 142 18.00 + +  4.00 18 33.00 72 +  
2756745.67 1101101110010111011111111110010 1.25 24 86 80.00 +   2.75 20 61.00 72 +  +
3313354.72 1100110010101010101110100100101 1.75 19 109 70.00 +  + 3.00 21 40.00 68 +  +
... ... ...
... ... ...
Last generation
5395.69 1101110110110101110101111010110 2 22 75 64 + + + 3.5 22 71 72 + + +
5395.69 1101110110110101110101111010110 2 22 75 64 + + + 3.5 22 71 72 + + +
5395.69 1101110110110101110101111010110 2 22 75 64 + + + 3.5 22 71 72 + + +
5395.69 1101110110110101110101111010110 2 22 75 64 + + + 3.5 22 71 72 + + +
5395.69 1101110110110101110101111010110 2 22 75 64 + + + 3.5 22 71 72 + + +
5395.69 1101110110110101110101111010110 2 22 75 64 + + + 3.5 22 71 72 + + +
5395.69 1101110110110101110101111010110 2 22 75 64 + + + 3.5 22 71 72 + + +
5395.69 1101110110110101110101111010110 2 22 75 64 + + + 3.5 22 71 72 + + +
5395.69 1101110110110101110101111010110 2 22 75 64 + + + 3.5 22 71 72 + + +
5395.69 1101110110110101110101111010110 2 22 75 64 + + + 3.5 22 71 72 + + +
... ... ...
... ... ...

the differences are broadened with 93% and 95% successes, respec-
tively. These were mainly improved by introducing dynamic pen- 12000 Design procedure
alty to the objective function of GA. The performances of GA with GA with static penalty
Material volumes, cm3

dynamic penalty at the gear ratios of 6, 11, 16, and 21 are 99%, 10000 GA with dynamic penalty
98%, 97%, and 99%, respectively. It should be noticed that all the
constraints introduced are satisfied by GA. In addition, GA ap- 8000
proach is stochastically able to find quite well results within a
short time based on the objective function specified. 6000

4000
6. Conclusions

Designing gear pairs are complex and it requires the use of non- 2000
1 6 11 16 21
linear and discrete design variables. In this study, a parallel axis Gear ratios
gearbox with two stage helical gear pairs was optimized by using
GA with the objective of minimizing volume. Static and dynamic Fig. 6. Comparison of GAs and design procedure.
C. Gologlu, M. Zeyveli / Computers & Industrial Engineering 57 (2009) 1043–1051 1051

Table 10 Bozaci, A., Ilknur, K., & Colak, O. U. (2001). Machine design projects. Istanbul:
Comparison of design procedure and GA with static and dynamic penalties. Caglayan Kitabevi.
Chong, T. H., Bae, I., & Park, G. J. (2002). A new and generalized methodology to
Gear Minimum volumes (cm3) DP
GAS  100 DP
GAD  100 design multi stage gear drives by integrating the dimensional and configuration
ratios design process. Mechanism and Machine Theory, 37, 295–310.
GA with GA with Design
Davis, L. (1987). Genetic algorithms and simulated annealing. London: Pitman.
static dynamic procedure
Deb, K., Pratap, A., & Moitra, S. (2000). In Proceedings of the parallel problem solving
penalty (GAS) penalty (GAD) (DP) from nature VI conference. Mechanical component design for multiple objectives
6 2849.76 2806.45 2780.97 97.59 99.09 using elitist non-dominated sorting GA (pp. 859–868), Paris.
11 5395.69 5405.48 5297.81 98.19 98.01 Gen, M., & Cheng, R. (2000). Genetic algorithms and engineering optimization. New
16 8826.93 8494.6 8211.89 93.03 96.67 York: John Wiley & Sons Inc.
Goldberg, D. E. (1989). Genetic algorithms in search, optimization and machine
21 12052.54 11557.16 11488.3 95.32 99.40
learning. New York: Addison Wesley.
Joines, J., & Houck, C. (1994). On the use of non-stationary penalty functions to solve
nonlinear constrained optimization problems with GAs. In D. Fogel (Ed.),
penalty functions were imposed on the objective function. The aim Proceedings of the first IEEE conference on evolutionary computation, Orlando,
Florida (pp. 579–584). New York: IEEE.
of automating preliminary design of gear drives by minimizing vol-
Lin, C. Y., & Hajela, P. (1992). Genetic algorithms in optimization problems
ume of gear trains was achieved. The optimized design parameters with discrete and integer design variables. Engineering Optimization, 19,
of module, number of teeth, and width of teeth for pinion and gear 309–327.
Marcelin, J. L. (2001). Genetic optimisation of gears. International Journal of
pairs of the stages for desired output speeds were revealed. The re-
Advanced Manufacturing Technology, 179, 910–915.
sults were compared with a deterministic design procedure devel- Pham, D. T., & Yang, Y. (1993). Optimization of multi-model discrete functions using
oped. As it is well known, design is an iterative process that is genetic algorithms. Proceedings of the Institution of Mechanical Engineers, 207,
refined and arrived at a solution. By the GA based approach, the 53–59.
Puzzi, S., & Carpinteri, A. (2008). A double-multiplicative dynamic penalty approach
designers can substantially speed up finding solutions in design for constrained evolutionary optimization. Structural and Multidisciplinary
process for different gear ratios. Optimization, 35(5), 431–445.
Bounding parameter values are very important in GA; therefore Saruhan, H., Rouch, K. E., & Roso, C. A. (2004). Design optimization of tilting-pad
journal bearing using a genetic algorithm. International Journal of Rotating
determination of crossover and mutation parameter values directly Machinery, 10(4), 301–307.
affects solutions. The solutions depend on wisely selected penalty Smith, A. E., & Coit, D. W. (1997). Constraint handling techniques-penalty functions.
coefficient values. The weighting of constraints by penalties is dif- In Back et al. (Eds.), Handbook of evolutionary computation. Oxford: Oxford
University Press.
ficult to adjust and assigning right values are essential to solutions Su, D., & Wakelam, M. (1998). Intelligent hybrid system for integration in
compromised. The use of dynamic penalties has already amelio- design and manufacture. Journal of Materials Processing Technology, 76(1–3),
rated the results of GA. 23–28.
Thompson, D. F., Gupta, S., & Shukla, A. (2000). Tradeoff analysis in minimum
volume design of multi-stage spur gear reduction units. Mechanism and Machine
References Theory, 35, 609–627.
Yokota, T., Taguchi, T., & Gen, M. (1998). A solution method for optimal weight
Abersek, B., & Popov, V. (2004). Intelligent tutoring system for training in design and design problem of the gear using genetic algorithms. Computers & Industrial
manufacturing. Advances in Engineering Software, 35, 461–471. Engineering, 35(3–4), 523–526.
Akinci, I., Yilmaz, D., & Canakci, M. (2005). Failure of a rotary tiller spur gear. Zeyveli, M. (2005). Gearbox wheel design with genetic algorithms. PhD Dissertation,
Engineering Failure Analysis, 12, 400–404. Ankara, Gazi University.

View publication stats

Das könnte Ihnen auch gefallen