Sådhanå (2018) 43:9 Ó Indian Academy of Sciences

https://doi.org/10.1007/s12046-017-0775-9 Sadhana(0123456789().,-volV)FT3
](0123456789().,-volV)

Design optimization of deep groove ball bearings using crowding

distance particle swarm optimization
APARNA DUGGIRALA1, R K JANA2,*, R VENKAT SHESU3 and PRASUN BHATTACHARJEE4
1
Department of Industrial Engineering and Management, Maulana Abul Kalam Azad University of Technology,
Kolkata 700064, India
2
Indian Institute of Management Raipur, GEC Campus, Sejbahar, Raipur 492015, India
3
Indian National Centre for Ocean Information Services, Ministry of Earth Sciences, Hyderabad 500090, India
4
Department of Mechanical Engineering, Jadavpur University, Kolkata 700032, India
e-mail: [email protected]

MS received 28 July 2016; revised 31 May 2017; accepted 3 June 2017; published online 7 February 2018

Abstract. This paper presents a crowding distance particle swarm optimization technique to optimize the
design parameters of deep groove ball bearings. The design optimization problem is multi-objective in nature.
The considered objectives are maximizing dynamic and static load bearing capacities and minimizing elasto-
hydrodynamic film thickness. The technique is applied to bearings used in transmission system of a tractor for
the purpose of farming. Pareto optimal solutions are obtained using the proposed technique. The results obtained
from the technique are found to be superior compared with NSGA-II and catalogue values.

Keywords. Mechanical design; design optimization; deep groove ball bearings; multi-objective optimization;
particle swarm optimization; crowding distance.

1. Introduction variants of PSO proposed by the researchers [14–19].

They have wide and varied applications in mechanical
Deep groove ball bearings are dynamic mechanical com- design [20–26]. The variant of PSO that deals with
ponents that minimize the frictional losses during power multiple objectives is known as multi-objective PSO
and motion transfer processes. They belong to the class of (MOPSO) [27]. Crowding distance PSO (PSOCD) is a
rolling element bearings [1] and are used extensively in class of MOPSO. It ensures an even spread of non-
industrial applications occupying a major market share [2]. dominated solutions to cover the entire Pareto front [28].
Bearings can sustain radial and moderate amount of thrust Researchers have implemented the concept of crowding
loads exerted by the rotating shafts on a small area, giving distance to obtain better results for a variety of problems
rise to high magnitude contact stress. If a bearing element is [29, 30]. PSOCD is a little more time consuming than
provided with proper lubrication and alignment, without MOPSO. However, it is competitive for selecting the
any abrasive particles, moisture or corrosion-causing ele- global best [31, 32]. However, in the field of ball bearing
ments with appropriate load, it is unlikely to fail except due research, PSOCD has not been explored much.
to contact fatigue [3]. Bearings look trivial but their failure In the present work, the application of deep groove ball
is costly. This has led to the increasing interest in research bearings in farm equipment like tractors has been consid-
on bearings design and other operating parameters. ered. An attempt has been made to increase the load bearing
The design optimization of bearing literature mentions capacity of deep groove ball bearings using PSOCD. A
application of traditional optimization techniques [4, 5]. modified mutation strategy is used to maintain the diversity
The limitations of these techniques are well documented. in swarm for including the entire range of design variables.
To overcome these drawbacks of traditional techniques, The obtained results were compared with the NSGA-II
researchers have proposed different artificial intelligence solutions and catalogue values. Also, the possible increased
(AI) techniques [6, 7]. In the field of mechanical design, reliability of tractors in harsh soil and environmental con-
different types of AI techniques have been used [8–12]. ditions is discussed.
Particle swarm optimization (PSO) [13] is an AI The paper is organized as follows. The problem state-
technique that explores the best solution by sharing the ment, presented in section 2, gives a detailed description of
information about global or local best. There are many the objective functions, constraints and their conditions.
The PSOCD is presented in section 3. An application of the
*For correspondence deep groove ball bearings is presented in section 4. Results

1
9 Page 2 of 8 Sådhanå (2018) 43:9

and discussion are presented in section 5. Conclusions are resist before reaching permanent deformation of the
presented in section 6. component. If the applied load exceeds Cor then brinel-
ling will take place, which is furthermore influential for
failure. The static load rating is influenced by the highest
2. Problem statement contact stress that occurs between the balls and either of
the inner or outer raceways. It is affected by the material
We focus on deep groove ball bearings, which are used to used for bearing, number of balls used and their size,
carry radial load in the transmission system of a tractor. curvature of inner and outer raceway and their depth
Figure 1 indicates the details of the internal geometry of the along with the contact angles [1].
system. The design variables and objective functions are 2.2b Dynamic capacity (Cr): The dynamic load bearing
explained in the following sub-sections. capacity is defined using the maximum shear stress theory.
The theory states that the total strain energy in the com-
plex state of stress system reaches the shear strain at the
2.1 Design variables sub-surface near the area of contact. Dynamic load ratings
The design parameters of this study include pitch diameter are found by using the variables like geometry of the
of bearing (D1), ball diameter (D2), curvature coefficient of bearing, number of balls, size of the balls used, pitch
the inner raceways (f1), curvature coefficient for outer diameter of bearing and the material used for ring and the
raceway (f2) and the number of rolling elements (Z). These balls. This load rating is used along with the actual
parameters form the internal geometry of the bearings. applied radial load for calculating bearing fatigue life. It
Also, the boundary dimensions are essential to define the has been observed that the dynamic capacity of the
major geometry of the bearings, which are standardized to bearing varies with the number of rollers and the diameter
facilitate interchangeability. These standardized dimen- of the ball [10].
sions include bearing bore diameter, outside diameter, 2.2c Minimum elasto-hydrodynamic film thickness: The
width and chamfer. Thus, to optimize the overall life and elasto-hydrodynamic film thickness is related to the
performance of the bearing, variables that define the maximum wear life of the bearing. Theoretical and
internal geometry are used as these variables directly experimental results have proved that the elasto-hydro-
influence the study objectives. dynamic lubrication is a manipulating factor for the basic
pattern of stresses developed in the contacting surfaces
that intensifies the failure due to fatigue. It has been
2.2 Objective functions proved that the bearing life decreases when the film
The objective functions considered are maximizing the thickness decreases [33–35]. This is due to the increase
static and dynamic load bearing capacities and minimizing in friction and thereby increase in temperature. This
the elasto-hydrodynamic film thickness. They are defined as process enhances the lubrication loss and damages the
follows. raceways. As a consequence, the sliding metal to metal
contact damages the surfaces in contact and the bearing
2.2a Static load bearing capacity (Cor): The static load life decreases rapidly. Hence, an optimum elasto-hydro-
bearing capacity is related to the loads applied to non- dynamic film thickness must be maintained for smooth
rotating bearings. It is expressed as a function of vari- and pure rolling in the contact surfaces.
ables corresponding to the internal geometry of the
system. Cor is the maximum load that any bearing can
2.3 Constraints
The constraints defined in Chakraborty et al [8] and
Gupta et al [10] are abridged and adopted for this study.
Ranges of the design variables are defined as follows
[36]:
i. the mean diameter ranges between 0.5(D?d) and
0.6(D?d);
ii. the ball diameter varies between 0:5ðD  d Þ and
0:6ðD  dÞ;
iii. the number of balls is provided by the condition
ð4  Z  50Þ;
iv. the inner and outer raceways vary according to the
Figure 1. Deep groove ball bearing. inequalities
0:515  f1  0:6; 0:515  f2  0:6.
Sådhanå (2018) 43:9 Page 3 of 8 9

Bearings should be assembled properly to obtain better 3. Crowding distance PSO (PSOCD)
life. To ensure proper assembly with required assembly
angle, the number of balls should satisfy the following The proposed PSOCD is presented here. The algorithm
constraints: selects the global best particle and deletes the external
  archive after selection. The mutation operator is used to
Db
2ðZ  1Þ sin1  u0 ; ð1Þ maintain diversity of the solution. Crowding distance
Dm
will be computed for each non-dominated solution sep-
u0 arately and the non-dominated solution with highest
g1 ð xÞ ¼    Z þ 1  0; ð2Þ
1 Db crowding distance will act as the swarm leader. The
2 sin Dm pseudo-code of the approach is presented below. The
notations used to describe the pseudo-code are presented
where u0 is the maximum tolerable assembly angle. This is
first.
influenced by the bearing geometry.
N: swarm size
The diameter of balls should vary between the following
x[i]: particle
limits:
t: iteration count
Dd Dd v[t]: velocity of the particle
KDmin  Db  KDmax ; ð3Þ AR: external archive where the non-dominated solutions
2 2
will be stored
where KDmin and KDmax are the minimum and maximum m: objective value
values of the ball diameter, respectively. This is associated R1 ; R2  U ð0; 1Þ.
with the diametric series of bearings and the ball strength. The pseudo-code is presented as follows:
The constraints related to these conditions can be con-
structed as follows:

g2 ð xÞ ¼ 2Db  KDmin ðD  dÞ  0; ð4Þ

g3 ð xÞ ¼ KDmax ðD  dÞ  2Db  0: ð5Þ

In order to ensure the running mobility of bearings, the
difference between the pitch diameter and the average
diameter should be less than a given value. Therefore, we
write

g4 ð xÞ ¼ Dm  ð0:5  eÞðD þ dÞ  0; ð6Þ

g5 ð xÞ ¼ ð0:5 þ eÞðD þ dÞ  Dm  0; ð7Þ

where e is a constant and its value is obtained based on the
mobility conditions of the balls. In the design optimization
of bearings, the pitch diameter is typically greater than the
related average diameter of a bearing. Therefore, for the
bearing ring at the outer raceways, the thickness should not
be lower than eDb , where e is a constant that depends on
simple strength consideration of the outer ring. The corre-
sponding constraint is defined as follows:

g6 ð xÞ ¼ 0:5ðD  Dm  Db Þ  eDb  0: ð8Þ

For the inner and outer raceways of a bearing, the groove
curvature radius should be designed in such a way that it
does not exceed 0.515Db. There is no upper limit for the
groove curvature radius, but when it is greater than 0.52Db
and 0.53Db, respectively, for the inner and outer raceways,
the dynamic load rating of the bearing reduces. Therefore

g7 ð xÞ ¼ 0:52  fi  0:515; ð9Þ

g8 ð xÞ ¼ 0:53  fo  0:515: ð10Þ

9 Page 4 of 8 Sådhanå (2018) 43:9

4. Application load capacities have been mentioned in N and the elasto-

hydrodynamic film thickness has been calculated in lm.
Tractors and farm implements should be able to function
smoothly in huge farms under harsh working conditions
that place severe demands on the bearing in the equipment. 5.1 NSGA II results
In such working conditions the customers expect bearings
For finding the NSGA II results, we have used the
to be highly reliable while offering superior cost perfor-
MATLAB MOGA toolbox. In figures 3, 4 and 5, the Pareto
mance. Various components of a tractor, including engine,
fronts between static and dynamic load capacity, static load
transmission system and wheel hub, are embedded with
capacity and elasto-hydrodynamic film thickness, and
bearings. The present work is focussed on the transmission
dynamic load capacity and elasto-hydrodynamic film
system that invloves heavy duty radial load bearings. These
thickness are shown, respectively.
bearings should withstand high amount of meshing loads
The NSGA II results have been found out to be reason-
that are caused by the engagement and disengagement of
ably practical for both static load capacity
gears. For this purpose, the bearings should be designed in
(25000–40200 N) and elasto-hydrodynamic film thickness
such a way that they should be reliable in the highly vari-
(0.02–0.17 lm). However, the dynamic load capacity
able loading conditions. An attempt is made to increase the
obtained by NSGA II is within a range of 36.5–39 kN,
load bearing capacity by applying the proposed PSOCD so
which is quite low. Even the number of Pareto optimal
as to attain the best solution. Figure 2 shows a sliding mesh
solutions obtained by NSGA II algorithm is fewer (within a
gear transmission system. The ball bearings are employed
range of 5–12), resulting in more restricted design options
to work under the radial load that arises out of the meshing
available to the decision makers.
of gears. Generally, the deep groove ball bearing with
bearing number 6204 will be used in this particular
application.
450000
Static load capacity

5. Results and discussion

350000

In this study, the optimum design parameters of deep grove

ball bearing used in transmission systems of mechanical
applications like tractor have been investigated. NSGA II
and PSOCD algorithms have been applied simultaneously
to analyse the performance of both algorithms. A set of
250000

Pareto optimum solutions have been obtained by imple-

menting both the algorithms considering any two of the 36.5 37 37.5 38 38.5
stated objectives concurrently. The boundary values for Dynamic load capacity
each of the variable have been kept identical according to
proper consideration of the application and constraints Figure 3. Pareto front for dynamic load capacity and static load
capacity using NSGA II.
applicable to the case under study. The static and dynamic

36.95
36.9
Dynamic load capacity

36.85
36.8
36.75
36.7
36.65
36.6
0 0.05 0.1 0.15 0.2
Elasto-hydrodynamic film thickness

Figure 4. Pareto front of static load capacity and elasto-hydro-

Figure 2. Transmission system of a tractor. dynamic film thickness using NSGA II.
Sådhanå (2018) 43:9 Page 5 of 8 9

402000 and f1: 0.5159298, f2: 0.5274896. The elasto-hydrodynamic

film thickness obtained is 0.2713662 lm and the values of
401500 the design variables corresponding to the thickness are D1:
Static load capacity

39.99165, D2: 13.57279, Z: 50 and f1: 0.5150475, f2:

401000 0.5284733. The trend in the variation of static load bearing
capacity along with the elasto-hydrodynamic film thickness
400500 can be observed from figure 7.
The graph indicates that with a raise in load the film
400000 thickness decreases. The minimum film thickness that
0.1 0.12 0.14 0.16 0.18 keeps the balls is always advantageous. However, the
Elasto-hydrodynamic film thickness
minimum film thickness occurs at lower static load capacity
than that obtained without considering the film thickness.
Figure 5. Pareto front for dynamic load capacity and elasto- When dynamic load bearing capacity and the elasto-
hydrodynamic film thickness using NSGA II. hydrodynamic film thickness are under consideration the
dynamic capacity obtained is 30586.36 N and the corre-
sponding elasto-hydrodynamic film thickness is
5.2 PSOCD results
0.5233867 lm, whereas the highest value obtained in the
When the optimization is carried out comprehensively for optimization is 0.5431483 lm. Though the value of the
an archive size of 1000 and a maximum of 50 generations, highest dynamic load bearing capacity is not compromised,
the maximum value of static load bearing capacity is the value of the film thickness has increased from its
obtained as 65261.55 N. The values of the design variables minimum obtainable value. However, the trend is accept-
corresponding to this highest static capacity are D1: able as the theory also confirms that with increase in the
33.56480; D2: 16.00000; Z: 50 and f1: 0.5674568, f2: load the film thickness also increases. The variation of
0.5673589 as per PSOCD algorithm. The highest value of dynamic load bearing capacity and the elasto-hydrody-
the dynamic load bearing capacity is 30578.67 N and the namic film thickness is indicated in the graph given in
corresponding design variables are D1: 40.00000; D2: figure 8.
15.99996; Z: 50 and f1: 0.5495156, f2: 0.5677343. The When the optimization is carried out for all three
variation of static load bearing capacity along with the objectives simultaneously, the graph shown in figure 9 is
dynamic load bearing capacity is shown in figure 6. obtained.
When the optimization is carried out between the elasto- The afore-mentioned solutions achieved by PSOCD are
hydrodynamic film thickness and static load bearing more implementable and diverse when compared with
capacity, the maximum static load bearing capacity NSGA II. Thus, PSOCD provided the decision makers with
obtained is 10787.678 N and the corresponding value of the more design choices with more viable opportunities to set
design parameters are D1: 39.99737; D2: 16.00000; Z: 50 parameters for designing the deep grove ball bearing.

Figure 6. Pareto front of dynamic and static load bearing Figure 7. Pareto front of elasto-hydrodynamic film thickness and
capacity using PSOCD. static load bearing capacity using PSOCD.
9 Page 6 of 8 Sådhanå (2018) 43:9

Figure 10. Dynamic load bearing capacity vs D1.

Figure 8. Pareto front of elasto-hydrodynamic film thickness and

dynamic load bearing capacity using PSOCD. functions and a swarm size of S, the complexity is
O(FS9log N) Leung et al [31].
5.3 Time complexity of the algorithm
In the present study, the three objectives and five inequality 5.4 Sensitivity analysis
constraints are considered for calculating the crowding
Pitch diameter: The pitch diameter directly impacts the
distance and the non-dominated solution is obtained by
dynamic load bearing capacity. Figure 10 indicates that
comparing the solutions in the repository to determine the
with increase of diameter, load bearing capacity also
complexity in computing the solutions. This problem needs
increases.
a computational time of 1.6293 s to generate the Pareto
Ball diameter: The ball diameter remained almost con-
front for the three objectives taken simultaneously. Most of
stant over all the optimized results, indicating that the
the computation time is spent on sorting the solutions in the
change in ball diameter has low impact on the variation in
repository for each of the objective with an archive size of
dynamic capacity of the bearing.
500. The variation of the computational time with archive
Number of balls: According to the obtained results, the
size will be logarithmic according to big ‘O’ notation. If
number of balls always remained constant. Also, the value
there are N solutions in the archive for F objective
that is obtained is the maximum value based on the con-
straints taken. Hence, the number of balls should be max-
imum to obtain better load bearing capacity.
Inner and outer curvature radius: The values of these
two variables do not seem to influence the output, but they
are important from the designer point of view.

6. Conclusion

A multi-objective design optimization problem corre-

sponding to deep groove ball bearings with application to
transmission system of tractor has been studied using a
PSOCD. There are three objectives in the problem. The
dynamic load bearing capacity, which is the most impor-
tant objective for increasing the sustainability of ball
bearings under varying load, is increased considerably
from the standard catalogue values. Static load bearing
capacity is also increased substantially, indicating that the
Figure 9. Pareto front considering of all three objectives bearings can withstand higher non-rotating load. This too
simultaneously. has significant impact on the durability of the bearings
Sådhanå (2018) 43:9 Page 7 of 8 9

under static conditions. The elasto-hydrodynamic film [12] Rao R B and Tiwari R 2007 Optimum design of rolling
thickness has an optimum value, which is neither too low element bearings using genetic algorithms. Mech. Mach.
nor too high. Hence, the bearings can maintain the Theory 42: 233–250
appropriate lubricant film thickness under increased con- [13] Eberhart R and Kennedy J 1995 A new optimizer using
ditions, leading to higher wear life for the bearings. The particle swarm theory. In: Proceedings of the Sixth Interna-
tional Symposium on Micro Machine and Human Science,
results obtained from the PSOCD show that the objective
pp. 39–43
values increase four times with respect to their respective [14] Chauhan P, Deep K and Pant M 2013 Novel inertial weight
catalogue values. As a result, the life of the bearing will strategies for particle swarm optimization. Memetic Comput.
increase considerably, making the transmission system 5(3): 229–251
reliable under variable loading conditions. We have [15] Chen M C, Hsiao Y H, Reddy R H and Tiwari M K 2016 The
solved the problem using NSGA II. It has been observed Self-Learning Particle Swarm Optimization approach for
that the PSOCD results are superior and advantageous routing pickup and delivery of multiple products with
from the engineering design perspective. The PSOCD can material handling in multiple cross-docks. Transp. Res. Part
be applied to other design optimization problems with E: Logist. Transp. Rev. 91: 208–226
multiple objectives. [16] De A, Mamanduru V K R, Gunasekaran A, Subramanian N
and Tiwari M K 2016 Composite particle algorithm for
sustainable integrated dynamic ship routing and scheduling
optimization. Comput. Ind. Eng.96: 201–215
Acknowledgements [17] De A, Kumar S K, Gunasekaran A and Tiwari M K 2017
Sustainable maritime inventory routing problem with time
The authors would like to thank the reviewers and the window constraints. Eng. Appl. Artif. Intell. 61: 77–95
Associate Editor for their great help in improving the [18] Eberhart R C, Shi Y and Kennedy J 2001 Swarm intelligence.
quality of the paper. The first author would like to thank the Elsevier
TEQIP Cell of Maulana Abul Kalam Azad University of [19] Kumar R S, Kondapaneni K, Dixit V, Goswami A, Thakur L
Technology, Kolkata, India, for providing financial support S and Tiwari M K 2016 Multi-objective modeling of pro-
for carrying out this research. duction and pollution routing problem with time window: a
self-learning particle swarm optimization approach. Comput.
Ind. Eng. 99: 29–40
[20] Lu H, Du B, Liu J, Xia H and Yeap W K 2016 A kernel
References extreme learning machine algorithm based on improved
particle swam optimization. Memetic Comput. https://doi.
[1] Childs P R N 2014 Mechanical design engineering hand- org/10.1007/s12293-016-0182-5
book. Oxford, UK: Butterworth-Heinemann [21] Nayak S K, Padhy S K and Panigrahi S P 2012 A novel
[2] Makhecha N and Patel R C 2015 Optimization of dynamic algorithm for dynamic task scheduling. Future Gener.
load carrying capacity of deep groove ball bearing using Comput. Syst. 285: 709–717
teaching–learning based optimization technique. Int. J. Sci. [22] Noroozi A, Mokhtari H, Nakhai I and Abadi K 2013
Res. Dev. 3: 2321–2613 Research on computational intelligence algorithms with
[3] Harris T A 2001 Rolling element bearing analysis. London: adaptive learning approach for scheduling problems with
Wiley batch processing machines. Neurocomputing 101: 190–203
[4] Changsen W 1990 Analysis of rolling element bearings. [23] Pant S, Kumar A, Kishore A, Anand D and Singh S B 2015
London: Mechanical Engineering Publications Ltd Application of a multi-objective particle article swarm opti-
[5] Seireg A 1972 A Survey of optimization of mechanical mization technique to solve reliability optimization problem.
design. J. Eng. Ind. 94(2): 495–499 In: Proceedings of the First International Conference on
[6] Chan F T S and Tiwari M K 2007 Swarm intelligence: focus Next Generation Computing Technologies, Dehradun, India,
on ant and particle swarm optimization. Vienna, Austria: September 04–05
I-Tech Education and Publishing [24] Tyagi S and Panigrahi S K 2017 An improved envelope
[7] Jong C, Gen M, Tiwari M K and Chang P C 2013 Meta- detection method using Particle Swarm Optimisation for
heuristics for manufacturing scheduling and logistics prob- rolling element bearing fault diagnosis. J. Comput. Des. Eng.
lems. Int. J. Prod. Econ. 141(1): 1–3 4(4): 305–317, https://doi.org/10.1016/j.jcde.2017.05.002
[8] Chakraborty I, Kumar V, Nair S B and Tiwari R 2003 [25] Yang Y, Xiaoxing L and Chunqin G 2007 Hybrid particle
Rolling element bearing design through Genetic Algorithms. swarm optimization for multi-objective resource allocation.
Eng. Optim. 35: 649–659 J. Syst. Eng. Electron. 18: 959–964
[9] Engelbrecht A P 2007 Computational intelligence – an [26] Yu-zhen Y U, Xin-yi R, Feng-shan D U and Jun-jie S H I
introduction. London: Wiley 2012 Application of improved PSO algorithm in hydraulic
[10] Gupta S, Tiwari R and Nair S B 2007 Multi-objective design pressing system identification. J. Iron Steel Res. Int. 19(9):
optimization of rolling bearings using genetic algorithms. 29–35
Mech. Mach. Theory 42: 1418–1443 [27] Coello Coello C A and Lechuga M S 2002 MOPSO: a pro-
[11] Lin W Y 2011 Optimum design of rolling element bearings posal for multiple objective particle swarm optimization. In:
using genetic algorithm – differential evolution hybrid Proceedings of the 2002 IEEE Congress on Evolutionary
algorithm. J. Mech. Eng. Sci. 225: 714–721 Computation, vol. 2, pp. 1051–1056
9 Page 8 of 8 Sådhanå (2018) 43:9

[28] Raquel C R and Naval P C Jr 2005 An effective use of Computation, pp. 1990–1997, https://doi.org/10.1109/cec.
crowding distance in multiobjective particle swarm opti- 2014.6900449
mization. In: GECCO ‘05, Proceedings of the 7th Annual [32] Zhang J and Li H 2015 A global-crowding-distance based
Conference on Genetic and Evolutionary Computation, multi-objective particle swarm optimization algorithm. In:
pp. 257–264 Proceedings of 2014 Computational Intelligence and Secu-
[29] Li W, Zhou Q, Zhu Y and Pan F 2012 An improved MOPSO rity, IEEE Xplore, https://doi.org/10.1109/cis.2014.160
with a crowding distance based external archive maintenance [33] Dawson D and Higginson G R 2006 Elasto-hydrodynamic
strategy. In: Tan Y, Shi Y and Ji Z (eds) Advances in swarm lubrication. London: Pergamon Press
intelligence. ICSI 2012. Lecture Notes in Computer Science, [34] Sibley L L and Orcutt F K 1961 Elasto-hydrodynamic
vol 7331. Springer, Berlin, Heidelberg lubrication of rolling contact surfaces. ASLE Trans. 4:
[30] Li W X, Zhou Q, Zhu Y and Pan F 2012 An improved 234–249
MOPSO with a crowding distance based external archive [35] Tallian T E 1966 Special research report on rolling contact
maintenance strategy. In: Proceedings of ICSI 2012, Part I, failure control through lubrication. SKF Report AL66Q028
LNCS 7331, pp. 74–82 [36] Rao V R and Svasani V J 2011 Mechanical design opti-
[31] Leung M F, Ng S C, Cheung C C and Lui A K 2014 A new mization using advanced optimization techniques. London:
strategy for finding good local guides in MOPSO. In: Pro- Springer Verlag, pp. 39–44
ceedings of the IEEE Congress on Evolutionary