ME 558

Discrete Design Optimization

Final Report

Design optimization of a two-stage

compound gear train
Team #3
Team Members
Nikhil Kotasthane
Priyank Gajiwala
Pratik Baldota

Abstract
Gear train is pertinent part of the majority of mechanical power transmission system. It has to be
robust enough to sustain the transmitted power over the prolong period of time and also light
enough to reduce the overall weight of the system and wastage of material. But since to increase
the power to be transmitted by the gear train also causes increase in weight and vice versa, these
two objectives rather generates contradicting solutions. Thus, the optimization of gear becomes
very significant in order to have a good trade-off between these two entities.
The paper involves the optimization of a two-stage compound gear-train wherein minimization
of weight and maximization of power are considered as the two objective functions. The decision
variables considered in this report are number of teeth, modules and material. The gear-train
system is subjected to constraints such as tooth strength, dynamic loading, minimum module and
wear strength. Since it is multi-objective function with contradicting objective function, it
becomes difficult to optimize using the conventional optimization techniques. Hence, we will use
non-traditional Non-dominated Sorting Genetic Algorithm-II (NSGA-II) which is much more
proficient to solve this kind of problem. MATLAB code will be used to implement this
algorithm.

Problem Statement
Design and optimize a two-stage compound gear train considering the following multi-objective
functions:
Minimization of the overall weight (f1)
Maximization of Power transmitted (f2)
For Illustration purpose, we have considered a two-stage compound gear-train having the input
power as 15 KW, reduction ratio as 2.5 and speed as 1440 rpm.
Certain assumptions for gear parameters : Type of gear tooth profile= Involute
The Pressure angle of the gear= 20 degrees
Minimum Number of teeth on the spur gear = 17 (to
avoid interference)
Young's Modulus of the materials = 2.1*10^5 N/mm2
Carefully cut gears are considered
Value of c in Buckingham's Dynamic Load = 11860*e
where e-error is expected error in tooth profile
(Ratio of Face Width by module) = 0.3
Fig. 1

Mathematical Model

Objection function:
1)
Minimize - Overall weight of the gears (f1)
f1 = [(/4)*(m*Z1) ^2*b*] + [(/4)*(m*Z2) ^2*b*]
2)
Minimize - Power Transmitted (f2)
f2 = (15/100)*((100-50*f*(Hs2 +Ht2)/(cos()*(Hs+Ht))
Definition of design variables:
Decision variable
Notation Variable type Range
Module
m
Discrete
{1, 1.25,1.5, 2, 2.5, 3, 4, 5, 6}
Number of teeth on first
gear
Material selection

Z1

Discrete

{ 18,20,22,24,26,28,30,32}

Discrete

{0,1}
0 for Material 1 - 40Ni2Cr1Mo28
1 for Material 2 - 40Cr1

Table 1

Available Materials:
1) 40Ni2Cr1Mo28
Crushing strength[ ]- 11000kgf/cm2
Tensile strength[ ]- 4000kgf/cm2
Density = 7800 kg/m3
2) 40Cr1
Crushing strength [ ] - 9750kgf/cm2
Tensile strength [ ] - 2375kgf/cm2
Density = 7100 kg/m3

Definition of equations and constraints:

Tooth strength Fs = [ b ]*b*Y*m

where Y = (0.154 - )
= b/a =0.3 (Taking recommended value of as 0.3)
Number of teeth on a gear is given by
Z2 = i * Z1
The minimum module for gear is given by

0.912

[ ]

mmin= (1.26)*

Y[ ]

[mm]

where = b / m = 10 (Taking recommended value of as 0.3)

Twisting torque [Mt ] = 60*1000*P / (2*pi*N) [Nmm]
Pitch line velocity Vm= * d1 * N / 60
[mm/s]
The module m obtained from optimization process should be greater than mmin.
This is given by the following constraint mmin< m.......................... (1)
Diameter of pinion d1 = m*t1 [mm]
Diameter of Gear d2 = m*t2
[mm]
[N]
Tangential force on gear tooth Ft = P/Vm
The constraint for this equation is given by Ft<Fs......... (2)
Buckinghams dynamic load is given by
Fd = Ft +
0.164 * Vm *(C*b + Ft) [N]
0.164*Vm + 1.485 * C b + Ft
Where
C = 11860 * e
Where Expected error in tooth profile (e) varies according to module
Module (mm)
Error (mm)
Upto 4
0.025
Upto 5
0.025
Upto 6
0.030
Upto 7
0.035
Table 2
The constraint for Buckinghams dynamic load is given by Fd<Fs............ (3)
Wear strength of the gear is given by

Fw = d1 * Q * b *K,
[N]
where Q = 2 * i / (1 + i)
K = [ ]2 * sin () * ((1/E1) + (1/E2))/1.4
The constraint for this equation is given by Fd<Fw................ (4)
Available choices of the design variables:
m, Z1, Z2, B {as specified in Table 1}
Power Loss
Friction Factor :- 0.08
= 20 degrees involute gear
Ro = R + one addendum
Addendum of 20 degrees involutes system = one module (m)
Ro= R+m
[mm]
ro= r+m
[mm]
2

Ht= ((i + 1)/i) *( ()2 2))^(1/2))

2

Hs= (i + 1) *( ()2 2))^(1/2))

Power Loss percentage (PL) = 50*f*(Hs2 +Ht2)/(cos()*(Hs+Ht))

Efficiency() = 100 - PL
Power Transmitted= * 15 KW

Optimization method
The task of multi-objective optimization is different from that of single-objective optimization
because in multi-objective optimization, there is usually no single solution which is optimum
with respect to all objectives. Since our two objective functions are conflicting it is expected to
generate an pareto-front. It is only known that none of the generated solutions dominates the
others. We are going to use the Non-dominated Sorting Genetic Algorithm-II (NSGA-II).
Choice of algorithm
The standard methods for solving multi-objective problem by using the GA is through weighted
sum method. The weighted sum method cannot find certain Pareto-optimal solutions in the case
of a non-convex objective space. Single-objective optimization algorithm with equal-spaced
weight vectors does not always guarantee a uniformly distributed Pareto-optimal front. This
implies the limitation of weighted sum method for real world problem.
On the other hand, NSGA-II produces the Pareto-optimal solutions in a single run only. It is
capable of handling discrete and real valued decision variables for complex versions of gearbox
problem. It is an improvement over the original NSGA algorithm with less computational
complexity, consideration of elitism and does not require any parameter calibration.
Hence, we have chosen NSGA-II for solving our problem.

Final Implementation:
We implemented NSGA-II using the code available on the Mathworks website. It includes
separate Matlab files for objective function, initializing variables, tournament selection,
replacing chromosomes and non-domination sort. It solves for minimization of objective
functions. The algorithm requires input optimization parameters, constraints and the objective
function. In our case it will minimize the weight and power loss (i.e. increases output power
transmitted).
Steps for implementation :
The lower bound and upper bounds of decision variables are given input as vectors. We set
the algorithm to run for integer type of decision variables. However since module used are
standard values they are discrete choices. Changes were made in the algorithm for integer to
discrete transformation for module selection.
Mutationfraction and Cross-overfraction are tuned till we get a proper pareto-front
We run the algorithm for two runs, for stage one and stage two of compound gear sets, with
different population size and generations to get pareto fronts. The first run will give paretooptimal solution values of decision variables for stage one. The input conditions for stage two
are then calculated. We assume the gear ratio of each stage remains the same. The second run
then gives a pareto-optimal solution values for second stage.

Results
Incumbent solution
Considering the problem as optimization of spur gear train with decision variables and objective
function as discussed above, we solve the problem by traditional method using hand calculations.
We obtain the following results for the two stages:

Module - m (mm)
Teeth on pinion -z1(mm)
B
Tooth strength - Fs (N)
Dynamic load - Fd (N)
Wear strength - Fw (N)
Weight -W (kg)
Power transmitted - P (kW)

Stage 1
Material 1 Material 240Ni2Cr1Mo28
40Cr1
4
5
18
18
0
1
20776
19275
9851.2
11093
11583
14219
9.2097
16.373
14.43
14.43

Table 2

Stage 2
Material 1 Material 240Ni2Cr1Mo28
40Cr1
4
5
18
18
0
1
20776
19275
11284
11079
11583
14219
9.2097
16.373
13.882
13.882

Optimization results
The implementation of NSGA-II algorithm was performed in Matlab. The stopping condition is
the number of generations. Starting with an initial population of 50 and number of generations as
100, the code was executed. However, the solution gave optimal points with many pareto fronts.
On increasing the initial population to 150 and number of generations to 400, the following
pareto optimal points were obtained.
For stage 1, the plots are as follows:
Material 2

Material 1
14.475

14.467

14.47

Power (kW)

Power (kW)

14.465
14.463
14.461
14.459

14.465
14.46
14.455
14.45
14.445
14.44

14.457

14.435
14.43

14.455
7.2

8.2

9.2

10.2

14

19

24

29

Weight (kg)

Weight (kg)

Fig 3

Fig 2

As expected from the graphs, our goal of reducing weight is conflicting with the objective of
maximizing the transmitted power. The optimal points satisfy all the constraints and bounds.
For stage 2, the plots are as follows:
Material 2
13.955

13.975
13.97
13.965
13.96
13.955
13.95
13.945
13.94
13.935
13.93
13.925

13.95

Power (kW)

Power (kW)

Material 1

13.945
13.94
13.935
13.93
13.925

13

18

Weight (kg)
Fig 4

23

28

12

17

22

Weight (kg)
Fig 5

27

Example of an optimal solution obtained from NSGA-II

Module - m (mm)
Teeth on pinion -z1(mm)
B
Tooth strength - Fs (N)
Dynamic load - Fd (N)
Wear strength - Fw (N)
Weight -W (kg)
Power transmitted - P (kW)

Stage 1
Material 1 Material 240Ni2Cr1Mo28
40Cr1
3
4
25
20
0
1
13291
12941
8708.9
9652.9
9049.1
10111
7.49491
10.35
14.4565
14.44

Stage 2
Material 1 Material 240Ni2Cr1Mo28
40Cr1
4
4
18
22
0
1
20776
13436
11298
10582
11583
11122
9.20974
12.5231
13.908
13.815

Table 3

Discussions
Comparison of initial (incumbent) solution and optimized solution
In the incumbent solution it is difficult to consider all combinations of module and teeth, so the
general procedure is to increase the module to satisfy strength criteria because increasing teeth
will improve strength by lower margin. But this often leads to over-structure design leading to a
higher weight (Table 2 ).
From Table 3, it is seen that the module and teeth of pinion are optimized resulting in adequate
amount of strength to withstand the dynamic loads and wear
Table 3 represents one of the optimal solution from the Pareto front obtained. The designer will
have the option of choosing the best solution out of these based on the priority. For example if
weight is the most important criterion, then material 1 can be chosen for both stages.
The optimized solution for stage 1 and material 1 gives 22.8% weight reduction, for stage 1 and
material 2 gives 36%, for stage 2 and material 2 gives 23.5%.
However if cost is relatively more important, material 2 can be chosen (material 1 is stronger so
in general it would be more expensive).
Also, if power transmitted is important (higher efficiency required) then there has to be a tradeoff between the power transmission required and the weight.

Conclusion
The paper has successfully implemented NSGA- II on a conflicting multi-objective function of
maximizing power transmitted and minimizing the overall weight of the gear train. The results
obtained a pareto-front which gave an optimal set of solution for the multi-objective function.
So, a particular value from the available set of solution is selected based on the trade- off
between power transmitted and weight. Also, complex mechanical power transmission problems
can be effectively modeled using this algorithm. However, some of the weakness of the
algorithm are not considering the center distance between the input and output shaft. Also, gear
tooth deflection and vibrations were neglected. In future, this algorithm can be easily extended
to include cost as another objective function which also plays a crucial role in design and
fabrication of gear train system. The gear train casing dimensional constraint can also be
introduced to design the gear train for a fixed value of center distance. Also, it can be further
extended to a multi-stage gear train system where the algorithm can even give the optimized
number of stages that should be selected for a particular constrained and center distance.
Moreover, other gears trains consisting of helical, bevel and worm gears can also be optimized
by including the static, dynamic, wear and other constraint.

References
1. Padmanabhan S., Ganesan S., Chandrasekaran M., Srinivasa Raman V., "Gear pair design
optimization by Genetic Algorithm and FEA," Frontiers in Automobile and Mechanical
Engineering (FAME), 2010 , vol., no., pp.301,307, 25-27 Nov. 2010
2. DebK., Jain S.,Multi-Speed Gearbox Design Using Multi-Objective Evolutionary Algorithms,
Kanpur Genetic Algorithms Laboratory (KanGAL)
3. Deb K., Pratap A., Agarwal, S., Meyarivan T., "A fast and elitist multiobjective genetic
algorithm: NSGA-II," Evolutionary Computation, IEEE Transactions on , vol.6, no.2,
pp.182,197, Apr 2002
4. Pomrehn LP, Papalambros PY.,Discrete Optimal Design Formulations With Application to
Gear Train Design, J. Mech. Des., 1995
5. H Huang, Z Tian, MZuo, Multiobjective Optimization of Three-Stage Spur Gear Reduction
Units Using Interactive Physical Programming, Journal of Mechanical Science and Technology
( KSME Int. J.), Vol. 19, No. 5, pp. 1080~1086, 2005
6. Y.K.Mogal, D.D.Palande, V.D.Wakchaure, Optimization of Gear Pairs Using Genetic
Algorithm, IOSR Journal of Engineering (IOSRJEN),ISSN: 2250-3021
7. YallamtiMurali Mohan, T.Seshaiah, Spur Gear Optimization By Using Genetic
Algorithm,International Journal of Engineering Research and Applications (IJERA) ISSN:
2248-9622 Vol. 2, Issue 1, pp. 311-318, Jan-Feb. 2012
8. Stephen P. Radzevich, DudleysHandbook ofPractical GearDesign andManufacture