Mechanism and Machine Theory 104 (2016) 287–302

Contents lists available at ScienceDirect

Mechanism and Machine Theory

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

Design optimization of an angular contact ball bearing for the

main shaft of a grinder
Seung-Wook Kim a, Kibong Kang a, Kichan Yoon b, Dong-Hoon Choi c,⁎
a
Graduate School of Mechanical Convergence Engineering, Hanyang University, Seoul, South Korea
b
BU Chassis System, Automotive, Schaeffler Greater China, Shanghai, China
c
School of Mechanical Engineering, Hanyang University, Seoul, South Korea

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

Article history: A conventional trial and error approach toward the design of non-standard bearings takes a
Received 13 July 2015 significant amount of time to obtain an adequate design. In this study, a non-standard angular
Received in revised form 19 April 2016 contact ball bearing for the main shaft of a grinder was optimized using design automation and
Accepted 7 June 2016
optimization techniques. To manufacture a product as precisely as possible with a grinder, the
Available online xxxx
radial and axial stiffness values of the grinder bearing must be selected as objective functions.
To treat two objective functions, this study employed a global criterion method as a multi-
Keywords: objective optimization methodology. Eight constraints on the manufacturing, film thickness,
Angular contact ball bearing
friction, and fatigue life were imposed. Six geometric variables and an axial preload were se-
Multi-objective discrete optimization
lected as design variables. All design variables were regarded as discrete because they should
Quasi-static analysis
Grinder have manufacture-possible dimensions. Quasi-static analysis taking dynamic effects into ac-
count was employed to analyze bearing performance. For efficient discrete optimization, this
study proposed a hybrid method in which a micro-genetic algorithm and regression-based se-
quential approximate optimizer were both employed. Optimization results revealed that both
stiffness values were enhanced while satisfying all design constraints.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction

Rolling bearings are important rotating elements in a variety of mechanical systems and are standardized by different types and sizes.
The requirements of rolling bearings include long lifetimes, high stiffness, and low torque; requirements have become more stringent due
to the sophisticated demands of mechanical systems such as weight reductions, miniaturization, and high reliability [1,2]. However, the
existing standardized rolling bearings for general use purposes have failed to meet these requirements. If the boundary dimensions (the
bore diameter, outer diameter, and bearing width) of a rolling bearing were changed to satisfy a set of requirements, the redesign of a
correspondingly-sized grinder would be demanded and would likely be costly. Accordingly, the development of non-standard rolling
bearings that fulfill requirements without any change to their boundary dimensions has increased. With the rising development of
non-standard rolling bearings, engineers have developed CAE tools to predict the performance of rolling bearings in order to improve
their design efficiency, and have conducted analysis-based designs [3]. However, depending on the expertise of the engineer and through
the process of trial and error, adequate designs were made that met target requirements. As a result, the design of rolling bearings is time
consuming. Therefore, increased research with regard to non-standard bearings based on optimization techniques has been performed to
achieve design optimization and shortened design times.

⁎ Corresponding author.
E-mail address: dhchoi@hanyang.ac.kr (D.-H. Choi).

http://dx.doi.org/10.1016/j.mechmachtheory.2016.06.006
0094-114X/© 2016 Elsevier Ltd. All rights reserved.
288 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302

Changsen [4] introduced design problem formulation methods for deep groove ball bearings and tapered roller bearings by
applying design optimization concepts. Choi and Yoon [5] optimized the system life of an automotive wheel bearing by
performing a static analysis of the ball bearings. Chakraborthy et al. [6] performed a design optimization of deep groove ball bear-
ings to maximize their dynamic load rating. Building on Chakraborthy's research, Rao and Tiwari [7] additionally took into account
bearing widths and other constraints related to the thicknesses of the inner and outer raceways in design optimization. Gupta,
Tiwari, and Nair [8] chose the dynamic capacity, static load capacity, and elastohydrodynamic minimum film thickness as objec-
tive functions and used NSGA II (non-dominated sorting based genetic algorithm II) as a multi-objective evolutionary optimizer to
obtain Pareto optimal fronts of the deep groove ball bearings. Kumar, Tiwari and Reddy [9] performed a design optimization of
cylindrical roller bearings to maximize their dynamic capacity. Kumar, Tiwari, and Prasad [10] maximized the fatigue life of
crowned cylindrical roller bearings. Wei and Chengzu [11] selected the rating life and spin frictional power loss as objective func-
tions and used NSGA II to obtain Pareto optimal fronts of the high-speed angular contact ball bearing subjected only to an axial
load. Lin [12] implemented design optimizations for deep groove ball bearings to maximize their dynamic load rating and used
differential evolution combined with real-valued genetic algorithms as an optimizer, comparing its performance with those of
existing binary-code genetic algorithms. Tiwari, Sunil, and Reddy [13] obtained optimum designs to maximize the dynamic
load capacity for tapered roller bearings. Waghole and Tiwari [14] maximized the dynamic capacity of needle roller bearings
and used an artificial bee colony algorithm (ABCA), a differential search algorithm (DSA), a grid search method (GSM), a hybrid
method combining the ABCA and GSM, and a hybrid method combining the DSA and GSM as optimizers. This study performed a
design optimization of an angular contact ball bearing for the main shaft of a grinder.
Compared to the aforementioned studies, contributions of this study toward the advancement of applying optimization tech-
niques to the design of non-standard rolling bearings consisted of: (1) the use of an efficient hybrid optimization method taking
into account a relatively long quasi-static analysis time for the angular contact ball bearing under radial and axial loads, (2) the
use of a comprehensive design problem formulation including stiffness, manufacturability, film thickness, friction, and fatigue
life, and (3) the treatment of all design variables as discrete variables taking manufacturability into account.
Objective functions and constraints of all the aforementioned studies can be expressed through explicit functions of design
variables, and thus no analysis procedures were necessary except for the design of an automotive wheel bearing [5] and angular
contact ball bearing [11]. In the case of the automotive wheel bearing, static analysis was employed, and in case of the angular
contact ball bearing, quasi-static analysis was employed. Although quasi-static analysis procedures were used for the angular con-
tact ball bearing in Wei and Chengzu [11], the bearing was only subjected to an axial load; thus, all balls had equivalent inner and
outer contact angles. However, the high-speed angular contact ball bearing for the main shaft of a grinder in this study was sub-
jected to both radial and axial loads; thus, each ball had different inner and outer contact angles, resulting in longer times to ob-
tain converged analysis solutions compared to the case of the bearing only subjected to an axial load. It took an average of 10 s
(CPU: Intel Core i7 960, RAM: 24GB) to carry out one quasi-static analysis. If one of the genetic algorithm (GA) or meta-heuristic
algorithm frequently used in the aforementioned bearing design optimization studies were applied, it would require considerable
design time. Therefore, this study proposed a hybrid method using both a micro-genetic algorithm (MGA) [15] and progressive
quadratic response surface method (PQRSM) [16] implemented in PIAnO (process integration, automation, and optimization)
[17], a commercial PIDO (process integration and design optimization) software. The MGA that required fewer function evalua-
tions than GA was first used to obtain near global optima in the feasible region; these were then used as multi-starting initial
points of a function-based local optimizer (PQRSM) to obtain the best optimum solution.
All the aforementioned studies commonly accounted for the life and manufacturability of rolling bearings in their design prob-
lem formulations. Additionally, [5,9,10,13,14] included contact stress, [8] included film thickness, and [11] included friction. How-
ever, a comprehensive design problem formulation of this study took into account the radial stiffness and axial stiffness due to
simultaneously imposed radial and axial loads in addition to all of the design requirements considered in prior studies. Note
that the contact stress was implicitly accounted for when maximizing stiffness values.
Other than the number of balls or rollers, all the aforementioned studies treated all design variables as continuous variables
except for the design of an automotive wheel bearing [5] and cylindrical roller bearing [9,10], where only the roller diameter
was treated as a discrete design variable. In practice, however, design variables must be selected from sets of specified discrete
values. The sets of this study are listed in Table 2. The treatment of all design variables as discrete variables of this study resulted
in optimum discrete design variable values that could be directly adopted in practice without further discretization. Post-optimally
discretized solution yielded deteriorating design results compared to those of the discrete optimization in this study.
This paper is comprised of the following: Section 2 for bearing design problem formulations explains a design object, design
variables, objective functions, and constraints in detail; Section 3 briefly introduces quasi-static analysis; Section 4 describes the
proposed design optimization procedure; Section 5 shows the results of multi-objective discrete design optimization; Section 6
presents the conclusions of this study.

2. Bearing design problem formulation

2.1. Design object and components

A grinder is a machine with which to grind surface structures with a rapidly-rotating grindstone. In order for the ground sur-
face to be more precise than that of a surface cut by a regular bite, the stiffness of the grinder is of great significance. Insufficient
stiffness can result in deformations at the main shaft of the grinder during operation, leading to imprecise fabrication. To improve
 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302 289

Fig. 1. Components and boundary dimensions of an angular contact ball bearing.

Table 1
Values of boundary dimensions and a rotational
speed.

d [mm] 30
D [mm] 55
B [mm] 13
n [rev/min] 15,000

the stiffness of the grinder, this study employed an angular contact ball bearing for the grinder main shaft as a design object. The
angular contact ball bearing was composed of balls, a retainer, an inner raceway, and an outer raceway; its boundary dimensions
such as the bore diameter (d), outer diameter (D), and bearing width (B) were specified and were not changed during the design
optimization, illustrated in Fig. 1. Values of the boundary dimensions and rotational speed (n) of the angular contact ball bearing
of this study are listed in Table 1.

2.2. Design variables

To improve the objective functions described in Section 2.3 while satisfying the constraints described in Section 2.4 with fixed
boundary dimensions, it was necessary to change the internal geometric shape of the bearing. In this study, geometric variables as
the ball diameter (Db), pitch circle diameter (Dm), inner and outer raceway curvature ratios (fi and fo), number of balls (N), and
free contact angle (α0) were chosen as design variables, illustrated in Fig. 2. Note that the inner and outer raceway groove cur-
vature radii were denoted as ri and ro, respectively; fi and fo were defined as:

ri
ro

fi ¼ Db
; fo ¼ Db
: ð1Þ

Preloading was a general method to improve the stiffness of the angular contact ball bearing used. There were two methods
for preloading: fixed-position preloading and fixed-pressure preloading. This study employed fixed-pressure preloading which
used a spring to impose a certain level of preload on the bearing in the axial direction. This preloading method was mainly

Fig. 2. Geometric design variables of the angular contact ball bearing.

290 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302

Table 2
Initial values, lower bounds, upper bounds, and discretized design variable sets.

Design variable Initial Lower bound Upper bound Interval Discretized set

Db [mm] 7.144 3.175 9.000 – 3.175, 3.500, 3.572, 3.969, 4.000, 4.500, 4.763, 5.000,
5.500, 5.556, 5.953, 6.000, 6.350, 6.500, 6.747, 7.000,
7.144, 7.500, 7.938, 8.000, 8.500, 8.731, 9.000
Dm [mm] 42.5 40.5 44.5 0.1 40.5, 40.6, ⋯, 44.4, 44.5
fi [−] 0.520 0.510 0.540 0.001 0.510, 0.511, ⋯, 0.539, 0.540
fo [−] 0.520 0.510 0.540 0.001 0.510, 0.511, ⋯, 0.539, 0.540
N [ea] 16 12 20 1 12, 13, ⋯, 19, 20
α0 [deg.] 15 15 40 1 15, 16, ⋯, 39, 40
Fpre [N] 120 30 600 1 30, 31, ⋯, 599, 600

applied to bearings for high-speed rotation [18]. In this study, the magnitude of the axial preload (Fpre) imposed on the bearing
was also selected as a design variable. A total of seven design variables chosen in this study were:
h i
X ¼ Db ; Dm ; f i ; f o ; N; α 0 ; F pre : ð2Þ

To obtain a directly applicable optimum solution for actual bearing manufacture, all seven design variables were discretized in
their lower bound and upper bound ranges. Table 2 shows the initial values, lower bounds, upper bounds, and discretized sets of
the design variables. With regard to ball diameter, taking into account a standardization of ball sizes in units of inches and mm,
this study adopted all available standard balls in units of inches and mm within the upper and lower bound ranges; the
discretized sets of ball diameters in units of mm are listed in Table 2. Other design variables except for the number of balls
were discretized at specific intervals within their upper and lower bound ranges in consideration of actual manufacturing prac-
tices; their discretized sets are listed in Table 2.

2.3. Objective functions

If the angular contact ball bearing mounted on the main shaft of a grinder possessed insufficient stiffness, shaft deformation may occur
in the radial and axial directions during operation; the grinder may thus fail to grind precisely. For precise fabrication, this study chose
both the radial stiffness and axial stiffness of the angular contact ball bearing as objective functions. The radial stiffness (Kr) and axial stiff-
ness (Ka) values could be calculated via Formula (3) and Formula (4), [19], and the methods of calculating Ki, Ko, and δj in Eqs. (5)–(7) are
described in Reference [20]. Refer to the nomenclature in Appendix A for other notations.

XN h i  
∂F z 2 2 2πj
Kr ¼ ¼K U j þ V j cos β j cos ; ð3Þ
∂z j¼1
N

XN h i
∂F x 2
Ka ¼ ¼K U j þ V j sin β j ; ð4Þ
∂x j¼1

where
" #−3=2
1 1
K¼ 2=3
þ 2=3
; ð5Þ
Ki Ko

3=2
δj
Uj ¼ ; ð6Þ
δ j þ ðð f i þ f o −1ÞDb Þ

h i
1=2
δj δ j þ 3ðð f i þ f o −1ÞDb Þ
Vj ¼ h i : ð7Þ
2 δ j þ ðð f i þ f o −1ÞDb Þ

Although the weighted-sum method was popular due to its simplicity in implementation and use to solve multi-objective
problems, it could not reach the non-convex parts of the Pareto set, as shown in Fig. 3(a). Because we do not know whether
the Pareto front of two objective functions (radial and axial stiffness) was convex or not a priori, the global criterion method
was adopted, which could reach the non-convex parts of the Pareto set as shown in Fig. 3(b). This method determined the closest
 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302 291

Fig. 3. (a) Weighted-sum method, (b) global criterion method.

solution in the objective function space from the goal values of the objective functions by minimizing a scalarized objective func-
tion called a global criterion [21]. The particular global criterion employed in this study was:
!2 !2
K r −K r;G K a −K a;G
  þ   : ð8Þ
K  K 
r;S a;S

In Eq. (8), stiffness values with subscripts G and S represent goal values and scale factors, respectively. The procedure of ap-
plying the global criterion method in this study is described below.

Step 1: Prior to applying the global criterion method, the radial stiffness and axial stiffness scaled by their respective initial
values were separately maximized to determine their extreme objective values, which are depicted by a blue circle with a
(KRadial Radial Axial Axial
r,opt , Ka,opt ) label and an orange circle with a (Kr,opt,Ka,opt) label, respectively, in Fig. 4(a). These extreme objective values
comprised two boundaries of the Pareto front, and their larger radial and axial stiffness values were used as a goal value
for the next step, as depicted by a red circle with a (KStep2 Step2
r,G , Ka,G ) label in Fig. 4(b).

Fig. 4. Procedure of applying the global criterion method: (a) Step 1, (b) Step 2, (c) Step 3, and (d) Step 4.
292 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302

Fig. 5. Four geometrical parameters relevant to manufacturability constraints.

Step 2: Using the goal values depicted by a red circle, the third Pareto optimum was obtained by minimizing the global crite-
rion, Eq. (8), and is depicted by a green circle with a (KStep2 Step2
r,opt ,Ka,opt ) label in Fig. 4(b). Scale factors for radial and axial stiffness
were set as the difference between the goal value and the smaller value of the radial and axial stiffness (at the two employed
Pareto optima to specify the current goal), respectively, which are depicted by KStep2 r,S and KStep2
a,S , respectively, in Fig. 4(b).
Step 3: Using the goal specified by the first Pareto optimum (the blue circle in Fig. 4(a)) and the third one, the fourth Pareto
optimum was obtained by minimizing the global criterion, Eq. (8), and is depicted by a green circle with a (KStep3 Step3
r,opt ,Ka,opt ) label
in Fig. 4(c).
Step 4: Using the goal specified by the second Pareto optimum (the orange circle in Fig. 4(a)) and the third one, the fifth Pa-
reto optimum was obtained by minimizing the global criterion, Eq. (8), and is depicted by a green circle with a (KStep4 Step4
r,opt ,Ka,opt )
label in Fig. 4(d).

2.4. Constraints

2.4.1. Constraints on manufacturability

For a bearing to be manufactured, the minimum distance between balls (C), the minimum space to install a seal (S), and the
minimum thickness of the inner and outer raceways (ti and to) must be larger than or equal to their limit values [5]. Four geo-
metrical parameters (C, S, ti and to) are illustrated in Fig. 5, and the four constraints are formulated in Eqs. (9)–(12), where the
limit values are denoted with the subscript LV.

g1 ¼ C ≥C LV ð9Þ

g 2 ¼ S ≥SLV ð10Þ

g 3 ¼ t i ≥t i;LV ð11Þ

g 4 ¼ t o ≥t o;LV ð12Þ

2.4.2. Film thickness constraints

The angular contact ball bearing of this study rotated with a high speed of 15,000 rpm. Thus, lubrication of the bearing was
important to prevent severe friction between the balls and raceways, resulting in a greatly shortened bearing life. To assess the
lubrication appropriateness, the film parameter (Ʌ) defined in Eq. (13) was used. In Eq. (13), h denoted an oil film thickness
at a contact point between a ball and raceway, and values with regard to the surface finish of the raceways (fraces) and surface
 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302 293

finish of the balls (fballs) were set to 0.175[μm] and 0.0625[μm], respectively. If the film parameter value was larger than or equal
to 3, there was a sufficient oil film thickness and thus surface friction would not occur [22].

hξj
Λ ¼ 1=2 ð13Þ
f 2races þ f 2balls

To evaluate the oil film thickness (the numerator of the film parameter in Eq. (13)), elastohydrodynamic lubrication theory
was adopted [4,8], presented in Eqs. (14)–(17). In the equations, the subscript ξ refers to an inner or outer raceway and the sub-
script j denotes the jth ball. The angular contact ball bearing for a grinder subjected to both radial and axial loads possessed a
different contact angle between the (inner or outer) raceways and each ball, and the load imposed on each ball was also different.
Thus, the oil film thickness between a (inner and outer) ring and each ball was different, as denoted by hξ,j in Eq. (14). In Eq. (14),
Qξ,j denotes a load between the (inner or outer) raceway and the jth ball, which was evaluated as a result of the quasi-static anal-
ysis described in Section 3. Also, α1 and ηo denote the pressure coefficient of viscosity and a dynamic viscosity at atmospheric
pressure, respectively, and were set to 2.3 × 10−8[Pa−1] and 0.021[Pa s], respectively, in this study. Refer to the nomenclature
in Appendix A for other notations.
8  90:68 " ( !0:636 )#
<πnDm ηo 1−γξ 2 = Ry;ξ
0:49 0:466 −0:117 −0:073
hξj ¼ 3:63α 1 Rx;ξ E0 Q ξj 1− exp −0:703 ; ð14Þ
: 120 ; Rx;ξ

where

Db ð1−γi Þ f i Db
Rx;i ¼ ; Ry;i ¼ ; ð15Þ
2 ð2f i −1Þ

Db ð1 þ γ o Þ f o Db
Rx;o ¼ ; Ry;o ¼ ; ð16Þ
2 ð2 f o −1Þ

Db cosα ij Db cosα oj
γi ¼ ; γo ¼ : ð17Þ
Dm Dm

To guarantee appropriate lubrication at every contact point between a raceway and ball, the minimum film parameter (Ʌmin)
among all contacts was constrained to be larger than or equal to 3, as shown in Eq. (18).

hξj
g 5 ¼ Λ min ¼ min  1=2 ≥3 ð18Þ
ξ; j f 2races þ f 2balls

2.4.3. Constraints on friction

2.4.3.1. Spin-to-roll ratio. A spin-to-roll ratio is the ratio of the spinning angular speed of a ball with regard to its rolling angular
speed at the contact point between a ball and raceway. A large spin-to-roll ratio results in severe sliding between the ball and
raceway, causing excessive friction and frictional heat. Therefore, the spin-to-roll ratio is an important response related to friction
and proper design that can minimize this is essential.
Like the film thickness in Eq. (14), the spin-to-roll ratio is different at each contact between the inner raceway and each ball.
Assuming that outer raceway control occurs, there is no need to consider the spin-to-roll ratio between the outer raceway and
each ball. The spin-to-roll ratio between the inner raceway and jth ball (StRRj) can be calculated by Eq. (19) [4]. The maximum
spin-to-roll ratio (StRRmax) is constrained to be smaller than or equal to a limit value (StRRLV), as shown in Eq. (21). The limit
value is set to the maximum spin-to-roll ratio for the initial design, 0.214, in this study. Refer to the nomenclature in Appendix
A for other notations.
      D  
Db b
StRR j ¼ 1− cos α ij tan α ij −β þ sin α ij ð19Þ
Dm Dm

where
!
−1 sinα oj
β ¼ tan ; ð20Þ
cosα oj þ ðDb =Dm Þ
294 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302

Table 3
Values of f0 for angular contact ball bearings.

Oil mist Oil bath or grease Oil bath (vertical shaft), oil jet

Angular contact ball bearing

Single row 1 2 4
Double row 2 4 8

g 6 ¼ StRR max ¼ max StRR j ≤ StRRLV : ð21Þ

j

2.4.3.2. Friction moment. The spin-to-roll ratio is a response to judge friction when only accounting for bearing geometry, without
taking into account the load imposed on a bearing. Since axial preload is one of the design variables in this study, the load im-
posed on a bearing continued to change in the design optimization process, and thus we took into account friction moments in-
duced by bearing loads in this study. The friction moment at a contact between a (inner or outer) raceway and jth ball (Mξj) can
be calculated by Eq. (22) [4].

Mξj ¼ M o þ M 1;ξj ð22Þ

Mo in Eq. (22) is calculated by:

−7 2=3 3
if νn ≥ 2000; Mo ¼ 10 f 0 ðνnÞ Dm ; ð23Þ

−7 3
if νn b 2000; M o ¼ 160  10 f 0 Dm ; ð24Þ

where v is the kinematic viscosity of a lubricant at operating temperature, 25 mm2/s, n is the rotational speed in rpm, Dm is the
pitch circle diameter, and f0 is a factor dependent on the bearing type and lubrication method presented in Table 3 [4].
M1,ξj in Eq. (22) changes as the bearing load changes, and is calculated by:

M1;ξj ¼ f 1;ξj P 1 Dm : ð25Þ

In Eq. (25), f1,ξj and P1 are evaluated by the formula shown in Table 4, where P0,ξj and C0,ξj denote an equivalent static load
and a static load rating at a contact between a (inner or outer) raceway and jth ball, respectively. Refer to Reference [4] for de-
tailed formulas for P0,ξj and C0,ξj. If P1 is found lower than the radial load (Fr), P1 is replaced by the radial load.

Table 4
Evaluation of f1,ξj and P1 for angular contact ball bearings.

Type of bearings f1,ξj P1

Angular contact ball bearings

Single row 0.0013(P0 ,ξj/C0, ξj)0.33 Fa − 0.1Fr
Double row 0.001(P0 ,ξj/C0, ξj)0.33 1.4Fa − 0.1Fr

Table 5
Life adjustment factor for reliability.

Reliability [%] a1

90 1
95 0.62
96 0.53
97 0.44
98 0.33
99 0.21
 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302 295

The maximum friction moment (Mmax) among all the contacts was constrained to be smaller than or equal to the limit value
(MLV), as shown in Eq. (26). The limit value was set to the maximum friction moment for the initial design, 51.121 [N mm], in this
study.

g 7 ¼ M max ¼ max M ξj ≤ MLV ð26Þ

ξ; j

2.4.4. Constraints on fatigue life

Fatigue failure is one of the most frequent causes of bearing damage. Since the angular contact ball bearing for a grinder pos-
sesses different contact angles between the (inner or outer) raceway and jth ball, the bearing life formula reflecting the contact
angles changes of the balls was employed, as proposed by Ioannides et al. [23,24]. According to Ioannides et al., the life of the
rotating inner raceway (Li), nonrotating outer raceway (Lo), and fatigue life for the entire bearing (L10) in 106 rotations can be
calculated as:
2 ( 3
X
N   )−10=9 −0:9
1 Q c;i 3
Li ¼ 4 a2 a3 5 ; ð27Þ
N j¼1 Qi j

2 ( )−10=9 3−0:9
X
N  
1 Q c;o 10=3
Lo ¼ 4 a2 a3 5 ; ð28Þ
N j¼1 Qo j

n o
−10=9 −10=9 −0:9
L10 ¼ a1 ðLi Þ þ ðLo Þ ; ð29Þ

where

!0:41  1:39 !0:3

2f ξ 1∓γ ξ γξ 1:8 −1=3
Q c;ξ ¼ 93:2  1=3 Db N : ð30Þ
2 f ξ −1 1γ cosα ξj
ξ

In the above equations, N is the number of balls, a1, a2, and a3 are adjustment factors for reliability, bearing material, bearing
lubrication, and operating conditions, respectively; fξ is the (inner and outer) raceway curvature ratio and γξ is calculated from
Eq. (17). The a1 values are listed in Table 5. In this study, the values of a1, a2, and a3 were set to 1.
In consideration of a replacement cycle of the angular contact ball bearing for a grinder, the L10 fatigue life was constrained to
be larger than or equal to 4000 h as shown in Eq. (31):

L10  106
g8 ¼ ≥ 4000½hours; ð31Þ
60  n

where n is the rotational speed in rpm.

3. Quasi-static analysis

To evaluate the performance indices in the objective function described in Section 2.3 and the constraints in Section 2.4, quasi-
static analysis took into account dynamic effects such as centrifugal forces and gyroscopic moments performed for the high-speed
angular contact ball bearing. It is well known that a ball bearing rotating at high speed is greatly influenced by centrifugal forces;
thus, the outer raceway control hypothesis should be applied [4]. Therefore, the outer raceway control hypothesis for the quasi-
static analysis was adopted. The bearing coordinate system is depicted in Fig. 6, where the x- and z-directions indicate the axial
and radial directions, respectively.
The output responses of the quasi-static analysis are listed in Table 6, along with the associated bearing performance indices.
The input variables and parameters are listed in Tables 7 and 8, respectively. Note that the first seven input variables are the de-
sign variables described in Section 2.2.

4. Design optimization procedures

As mentioned in the fourth paragraph of the introduction, it took 10 s (CPU: Intel Core i7 960, RAM: 24GB) on average to carry
out one quasi-static analysis for the angular contact ball bearing of this study. If one of the genetic algorithms (GAs) or meta-
heuristic algorithms used in the previous bearing design optimization studies was applied, it would require a large number of
function evaluations (quasi-static analyses), and would take a considerable amount of design time to obtain an optimal solution.
296 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302

Fig. 6. Bearing coordinate system.

Table 6
Output responses of the quasi-static analysis.

Output response Associated bearing performance index

Displacement in x-direction (δx) Radial/axial stiffness

Displacement in y-direction (δy) Radial/axial stiffness
Displacement in z-direction (δz) Radial/axial stiffness
Angular displacement in y-direction (θy) Radial/axial stiffness
Angular displacement in z-direction (θz) Radial/axial stiffness
Inner raceway contact angle at j-th ball (αij) Radial/axial stiffness, film parameter, spin to roll ratio, friction moment, fatigue life
Outer raceway contact angle at j-th ball (αoj) Radial/axial stiffness, film parameter, spin to roll ratio, friction moment, fatigue life
Inner raceway normal load of j-th ball (Q ij) Film parameter, fatigue life
Outer raceway normal load of j-th ball (Q oj) Film parameter, fatigue life

Table 7
Input variables of the quasi-static analysis.

Db Ball diameter
Dm Bearing pitch diameter
fi Inner race curvature
fo Outer race curvature
N Number of balls
α0 Free contact angle
Fpre Axial preload
Fx X-direction force
Fy Y-direction force
Fz Z-direction force
MY Y-direction moment
MZ Z-direction moment
ω Angular shaft velocity
Eo Young's modulus of outer-race
Ei Young's modulus of inner-race
ξo Poisson ratio of outer-race
ξi Poisson ratio of inner-race
ρB Density of the ball
 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302 297

Table 8
Input parameters for the quasi-static analysis.

Input parameter Value

Fx [N] 350
Fy [N] 0
Fz [N] 550
MY [N∙mm] 0
MZ [N∙mm] 0
n [rpm] 15,000
Eo [MPa] 200,000
Ei [MPa] 200,000
ξo [−] 0.3
ξi [−] 0.3
ρB [kg/mm3] 7.8 × 10−6

For example, if 100,000 or 1,000,000 function evaluations were usually required and performed to obtain an optimal solution, it
would take 1,000,000 or 10,000,000 s (about 12 or 120 days). Therefore, this study proposed a hybrid method using both a micro-
genetic algorithm (MGA) [15] and progressive quadratic response surface method (PQRSM) [16] implemented in PIAnO (process
integration, automation and optimization) [17], a commercial PIDO (process integration and design optimization) software. The
MGA that required fewer function evaluations than GA was first used to obtain near global optima in the feasible region, and
were then used as multi-starting initial points of a function-based local optimization algorithm (PQRSM) to obtain the best opti-
mum solution.
Compared to GA, MGA usually used a very small population size with a relatively small number of generations. However, this
did not guarantee that the global optimum would be obtained, which could be accepted in this study because only a near global
optimum was necessary to be used as an initial design point for the subsequent PQRSM. The population size was set to 5 with the
number of generations set to 200. Like GA, MGA employed a stochastic process and thus an MGA solution was different every
time. In this study, 10 MGA replications were performed and 10 MGA solutions denoted as [Db,Dm, fi, fo, N, α0, Fpre]MGA
1 ,2, ⋯ ,9,10 were

Fig. 7. Design optimization procedures using the proposed hybrid method.

298 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302

Table 9
Pareto optimum solutions.

Step 1(r) Step 1(a) Step 2 Step 3 Step 4

Design variables Db [mm] 6.350 5.953 6.000 7.000 6.000

Dm [mm] 42.3 41.9 42.2 41.4 42.2
fi [−] 0.510 0.510 0.510 0.510 0.510
fo [−] 0.510 0.510 0.510 0.510 0.510
N [ea] 19 20 20 17 20
α0 [deg.] 15 19 17 15 18
Fpre [N] 369 365 350 509 369
Constraints g1 [mm] −0.013 −0.002 −0.002 −0.008 −0.002
g2 [mm] −2.450 −2.847 −2.800 −1.800 −2.800
g3 [mm] −1.610 −1.694 −1.810 −0.695 −1.810
g4 [mm] −1.810 −2.294 −2.110 −1.795 −2.110
g5 [−] 3.463 3.376 3.404 3.479 3.403
g6 [−] 0.200 0.215 0.204 0.208 0.209
g7 [N mm] 51.256 50.362 50.865 51.256 51.247
g8 [hours] 4385 3989 3994 4506 3992
Objective functions Kr [kN/mm] 299.681 286.285 294.128 298.589 291.549
Ka [kN/mm] 97.839 114.528 106.019 101.794 111.116

obtained in the first step in Fig. 7. The total number of function evaluations in this step was 10,000. Note that MGA could handle
discrete design variables without any difficulty.
The 10 MGA solutions were used as 10 initial values for subsequent local optimizations using PQRSM. Note that, to treat dis-
crete design variables, gradient-based local optimization algorithms could not be employed, and thus an efficient function-based
local optimization algorithm called PQRSM was employed in this study. Using the 10 initial values, 10 PQRSM solutions denoted as
[Db, Dm, fi, fo, N,α0, Fpre]PQRSM
1,2, ⋯ ,9,10 were obtained in the second step in Fig. 7. The number of function evaluations to obtain a PQRSM
solution depended on the initial value, and the average number of function evaluations to obtain a PQRSM solution was found to
be 196. Thus, the total number of function evaluations in this step was 1960.
The last step just selected the best solution among the ten PQRSM solutions, as shown in the last step in Fig. 7. The total num-
ber of function evaluations to obtain the best solution amounted to 11,960, corresponding to 119,600 s (about 1.4 days).

5. Results

The design optimization procedure described in Section 4 was applied to find each Pareto optimum satisfying all the con-
straints described in Section 2.4, and five Pareto optima were obtained using the procedure of applying the global criterion meth-
od described in Section 2.3. Five Pareto optimum solutions are presented in Table 9.
Columns under Step 1(r) and Step 1(a) list optimum solutions obtained by only maximizing the radial stiffness and axial stiff-
ness, respectively, while satisfying all the constraints. The last three columns list the Pareto optimum solutions obtained by Step 2,
Step 3, and Step 4 in the procedure of applying the global criterion method described in Section 2.3 while satisfying all the con-
straints. Fig. 8 graphically compares some design variable values of the initial design with the five Pareto optimum solutions.

Fig. 8. Comparison of design variable values of the initial design and five Pareto optimum solutions: (a) the initial design, (b) Step 1(r) solution, (c) Step
1(a) solution, (d) Step 2 solution, (e) Step 3 solution, and (f) Step 4 solution.
 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302 299

Fig. 9. Five Pareto optimum solutions and the initial design point in the objective function space.

Fig. 9 shows the five Pareto optimum solutions and the initial design point in the objective function space. Compared to the
initial design, the radial stiffness and axial stiffness of the five Pareto optimum solutions were improved by 27–33% and 107–
142%, respectively, as shown in Fig. 10. Improvements in the radial and axial stiffness compared to the initial design could be at-
tributed to the following: (1) as the ball diameter became smaller and the number of balls increased, the bearing load was dis-
tributed well, and thus the contact stress was reduced; (2) as the inner and outer raceway curvature ratios became smaller,
the contact area between the ball and raceway became bigger and thus the contact stress was reduced; (3) as the axial preload
became larger, the rate of deflection increase declined as the bearing load increased [20].
To clearly observe the Pareto front shape, an enlarged view near the Pareto front is displayed in Fig. 11. The Pareto front was
found to be non-convex, confirming that the global criterion method should be adopted to solve the multi-objective optimization
problem of this study, as mentioned in the second paragraph of Section 2.3.

6. Conclusions

In this study, a non-standard angular contact ball bearing for the main shaft of a grinder was optimized using a design auto-
mation and optimization technique. The main contributions of this study can be summarized as follows:

1. A comprehensive design problem was formulated to maximize both the radial stiffness and axial stiffness of the angular contact ball
bearing. Eight constraints on manufacturing, film thickness, friction, and fatigue life were imposed. Six geometric variables and an
axial preload were selected as design variables. All design variables were regarded as discrete because they should have manufactur-
able dimensions.
2. Quasi-static analysis taking into account dynamic effects such as centrifugal forces and gyroscopic moments was performed for
the analysis of a high-speed angular contact ball bearing.

Fig. 10. Percent improvement of (a) the radial stiffness and (b) axial stiffness of five Pareto optimum solutions compared to the initial design.
300 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302

Fig. 11. An enlarged view near the Pareto front.

3. To obtain the Pareto optimum solutions, this study employed a global criterion method as a multi-objective optimization methodol-
ogy. Prior to applying the global criterion method, the radial stiffness and axial stiffness were separately maximized to find the ex-
treme Pareto optima. Using the goal values set using previous Pareto optima, additional Pareto optimum solutions were obtained
by minimizing the global criterion. Five Pareto optima were obtained using the procedure of applying the global criterion method.
4. It took 10 s (CPU: Intel Core i7 960, RAM: 24GB) on average to carry out one quasi-static analysis for the angular contact ball bear-
ing of this study. To considerably reduce the time to obtain a Pareto optimum solution, this study proposed a hybrid method that
used both an MGA and a PQRSM.
5. Compared to the initial design, the radial stiffness and axial stiffness of five Pareto optimum solutions improved by 27–33% and
107–142%, respectively. Also, the Pareto front was found to be non-convex, confirming that the global criterion method should
be adopted to solve the multi-objective optimization problem in this study.

With regard to future work, it is expected that reliability based design optimization reflecting uncertainties due to manufactur-
ing tolerances, material properties, and manufacturing environments would greatly contribute to the advancement of applying
optimization techniques for the design of non-standard rolling bearings.

Acknowledgements

This work was supported by the Korea Institute of Energy Technology Evaluation and Planning(KETEP) and the Ministry of
Trade, Industry & Energy(MOTIE) of the Republic of Korea (No.20164010200860). The authors express gratitude to PIDOTECH
Inc. for providing the PIAnO software as a PIDO tool for this study.

Appendix A. Nomenclature

Symbol Description Units

a1 Adjustment factor for reliability

a2 Adjustment factor for the bearing material
a3 Adjustment factor for bearing lubrication and operating conditions
B Bearing width mm
C Minimum distance between balls mm
C0 Static load rating N
d Bore diameter mm
D Outer diameter mm
Db Ball diameter mm
Dm Pitch circle diameter mm
E0 Equivalent modulus of elasticity MPa
 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302 301

Appendix A (continued)
(continued)

Symbol Description Units

E Young's modulus MPa

f0 Factor dependent on the bearing type and lubrication method
fi Inner raceway curvature ratio
fo Outer raceway curvature ratio
fraces Surface finish of races μm
fballs Surface finish of balls μm
F Load N
h Film thickness μm
K Load-deflection factor
Li Life of the rotating inner raceway 106 rotations
Lo Life of the nonrotating outer raceway 106 rotations
L10 Fatigue life of the entire bearing 106 rotations
M Friction moment N mm
My Y-direction moment N mm
Mz Z-direction moment N mm
n Rotational speed rpm
N Number of balls ea
Po Equivalent static load N
Q Rolling element load N
Qc Dynamic load rating N
ri Inner raceway groove curvature radius mm
ro Outer raceway groove curvature radius mm
S Minimum space to install a seal mm
StRR Spin to roll ratio
t Minimum thicknesses of the raceways mm
α Contact angle degree
α0 Free contact angle degree
α1 Pressure coefficient of viscosity Pa−1
βj Static operating contact angle rad
δ Displacement mm
δj Total deflection between the balls and the inner and outer raceways mm
η0 Dynamic viscosity at atmospheric pressure Pa s
θ Angular displacement rad
Ʌ Film parameter μm
v Kinematic viscosity of the lubricant at operating temperature mm2/s
ξi Poisson ratio of inner-race
ξo Poisson ratio of outer-race
ρB Density of the ball kg/mm3
Subscripts
a Axial direction
G Goal values
i Inner raceway
j jth ball
LV Limit value
o Outer raceway
pre preload
r Radial direction
S Scale factors
x X-direction
y Y-direction
z Z-direction
ξ Inner or outer raceway
0 Initial model

References

[1] A. Bourdon, J.F. Rigal, D. Play, Static rolling bearing models in C.A.D. environment for the study of complex mechanisms: part i - rolling bearing model, Trans.
ASME J. Tribol. 121 (2) (1999) 205–214.
[2] A. Bourdon, J.F. Rigal, D. Play, Static rolling bearing models in C.A.D. environment for the study of complex mechanisms: part II - complete assembly model, Trans.
ASME J. Tribol. 121 (2) (1999) 215–223.
[3] H. Aramaki, Basic Technology Research & Development Center, Motion and Control No. 3, NSK, Rolling Bearing Analysis Program Package “BRAIN”, 1997.
[4] W. Changsen, Analysis of Rolling Element Bearings, Mechanical Engineering Publications Ltd., London, 1991.
[5] D.H. Choi, K.C. Yoon, A design method of an automotive wheel bearing unit with discrete design variables using genetic algorithms, Trans. ASME J. Tribol. 123 (1)
(2001) 181–187.
[6] I. Chakraborthy, K. Vinay, S.B. Nair, R. Tiwari, Rolling element bearing design through genetic algorithms, Eng. Optim. 35 (6) (2003) 649–659.
[7] B.R. Rao, R. Tiwari, Optimum design of rolling element bearings using genetic algorithms, Mech. Mach. Theory 42 (2) (2007) 233–250.
[8] S. Gupta, R. Tiwari, S.B. Nair, Multi-objective design optimisation of rolling bearings using genetic algorithms, Mech. Mach. Theory 42 (10) (2007) 1418–1443.
302 S.-W. Kim et al. / Mechanism and Machine Theory 104 (2016) 287–302

[9] S.K. Kumar, R. Tiwari, R.S. Reddy, Development of an optimum design methodology of cylindrical roller bearing using genetic algorithms, Int. J. Comput. Methods
Eng. Sci. Mech. 9 (6) (2008) 321–341.
[10] S.K. Kumar, R. Tiwari, P.V.V.N. Prasad, An optimum design of crowned cylindrical roller bearings using genetic algorithms, Trans. ASME J. Mech. Des. 131 (5)
(2009) (14 pages).
[11] Y. Wei, R. Chengzu, Optimal Design of High Speed Angular Contact Ball Bearing Using a Multiobjective Evolution Algorithm, 2010 International Conference on
Computing, Control and Industrial Engineering, June 5–6, China, 2010.
[12] W.Y. Lin, Optimum design of rolling element bearings using a genetic algorithm—differential evolution (GA—DE) hybrid algorithm, Proc. Inst. Mech. Eng. C J.
Mech. Eng. Sci. 225 (3) (2011) 714–721.
[13] R. Tiwari, K.K. Sunil, R.S. Reddy, An optimal design methodology of tapered roller bearings using genetic algorithms, Int. J. Comput. Methods Eng. Sci. Mech. 13 (2)
(2012) 108–127.
[14] V. Waghole, R. Tiwari, Optimization of needle roller bearing design using novel hybrid methods, Mech. Mach. Theory 72 (2014) 71–85.
[15] K. Krishnakumar, Micro-Genetic Algorithms for Stationary and Non-Stationary Function Optimization, Proc. SPIE 1196, Intelligent Control and Adaptive Systems,
Vol. 289, 1990.
[16] K.J. Hong, M.S. Kim, D.H. Choi, Efficient approximation method for constructing quadratic response surface model, KSME Int. J. 15 (7) (2001) 876–888.
[17] PIDOTECH, PIAnO User's Manual, Version 4.0, Korea, 2015.
[18] NSK, Super Precision Bearings, CAT.NO. 1254 K, 2006.
[19] A.B. Jones, A general theory for elastically constrained ball and radial roller bearings under arbitrary load and speed conditions, J. Basic Eng. 82 (2) (1960)
309–320.
[20] T.A. Harris, M.N. Kotzalas, Rolling Bearing Analysis, fifth ed. Taylor and Francis group, Florida, 2007.
[21] G. Chiandussi, M. Codegone, S. Ferrero, F.E. Varesio, Comparison of multi-objective optimization methodologies for engineering applications, Comput. Math. Appl.
63 (5) (2012) 912–942.
[22] J. Hamrock, D. Bernard, Dowson, Ball Bearing Lubrication: the Elastohydrodynamics of Elliptical Contacts, Wiley, New York, 1981.
[23] E. Ioannides, T.A. Harris, A new fatigue life model for rolling bearings, Trans. ASME J. Tribol. 107 (3) (1985) 367–377.
[24] E. Ioannides, T.A. Harris, M. Ragen, Endurance of aircraft gas turbine main shaft ball bearings-analysis using improved fatigue life theory: part 1-application to a
long-life bearing, Trans. ASME J. Tribol. 112 (2) (1990) 304–308.