You are on page 1of 18

Design Optimization of Rolling Element

Bearings Using Advanced Optimization


Technique

R. D. Dandagwhal & V. D. Kalyankar

Arabian Journal for Science and


Engineering

ISSN 2193-567X
Volume 44
Number 9

Arab J Sci Eng (2019) 44:7407-7422


DOI 10.1007/s13369-019-03767-0

1 23
Your article is protected by copyright and
all rights are held exclusively by King Fahd
University of Petroleum & Minerals. This e-
offprint is for personal use only and shall not
be self-archived in electronic repositories. If
you wish to self-archive your article, please
use the accepted manuscript version for
posting on your own website. You may
further deposit the accepted manuscript
version in any repository, provided it is only
made publicly available 12 months after
official publication or later and provided
acknowledgement is given to the original
source of publication and a link is inserted
to the published article on Springer's
website. The link must be accompanied by
the following text: "The final publication is
available at link.springer.com”.

1 23
Author's personal copy
Arabian Journal for Science and Engineering (2019) 44:7407–7422
https://doi.org/10.1007/s13369-019-03767-0

RESEARCH ARTICLE - MECHANICAL ENGINEERING

Design Optimization of Rolling Element Bearings Using Advanced


Optimization Technique
R. D. Dandagwhal1 · V. D. Kalyankar2

Received: 10 May 2017 / Accepted: 5 February 2019 / Published online: 22 February 2019
© King Fahd University of Petroleum & Minerals 2019

Abstract
This article concentrates on optimization of fatigue life, i.e., dynamic capacity of rolling element bearings by using a novel
optimization approach known as Teaching–Learning-Based optimization. The optimization technique is applied to two cases
of bearing, i.e., deep groove ball bearing and cylindrical roller bearing, by considering a large number of bearings and their
associated constraints. The chosen problems involve around 9 design variables and the optimized result obtained shows
considerable improvement over the previous results, standard catalogues and handbooks. Efforts are also made to validate
the obtained results using finite element analysis approach and the simulated results of contact stress and deformation at
the contact between inner race and roller are found in close agreement with the obtained optimum results. Thus, this article
proves good applicability of proposed optimization approach in bearing design which will be useful to the on field designers
to improve the performance of bearings.

Keywords Deep groove ball bearing · Cylindrical roller bearing · Dynamic capacity · TLBO algorithm · Optimization

List of symbols φ0 Maximum tolerable assembly angle


D Outer diameter of bearing r1 Outer ring chamfering height
d Bore diameter of bearing r2 Outer ring chamfer width
B Width of bearing r3 Inner ring chamfer height
Dm Pitch diameter r4 Inner ring chamfer width
Db Ball diameter le Effective length of roller
fi The inner and outer raceway curva- Dr Mean diameter of the roller
ture radius coefficient λ Reduction factor,
fo The inner and outer raceway curva- ν Factor to account for edge loading
ture radius coefficient bm Factor to put up improvements in
Z Number of rolling elements geometrical accuracies
Cd Dynamic load rating γ Ratio of diameter of rolling ele-
∝ Contact angle ment and pitch diameter
L Life of the bearing d Number of design variables
a Constant defined by Lundberg–Pal- Nn Population size
mgren [2] m1, m2, . . . , mn Mean of population column wise
K D min , K D max , ε, e, β Constraints to basic design vari- TF Teacher factor
ables [9]

B R. D. Dandagwhal 1 Introduction
dandgwhalpes@gmail.com
1 Department of Mechanical Engineering, MPSTME, NMIMS, Rolling element bearings (REB) are expansively worn in
Shirpur, India high-speed operations, almost in all industrial sectors. Due to
2 Mechanical Engineering Department, S.V. National Institute their nonlinear, statistically indeterminate behavior, design
of Technology, Surat, India of REB has great importance in all global industries. Any

123
Author's personal copy
7408 Arabian Journal for Science and Engineering (2019) 44:7407–7422

critical failure of it should not only affect on the overall shown application potential of NSGA-II to different domain
system efficiency, but also its steadfastness, safety, avail- of research.
ability and overall cost of the system. To withstand the Kumar et al. [13] applied Genetic Algorithm (GA) to a
competitiveness, higher fatigue life is desirable for REB. The constrained nonlinear design problem of cylindrical roller
fatigue life is directly proportional with dynamic capacity of bearings (CRB) and chosen four design variables for opti-
the bearing. For the enhancement of operating life of the mization. In another study, Kumar et al. [14] applied GA
bearing, it is necessary to optimize the internal design vari- for design optimization of CRB with the logarithmic profile
ables of bearing. As the bearing has a complex structure, a crowning. NSGA-II on design optimization of high-speed
large number of design variables and constraints have been angular contact ball bearing by considering two objec-
involved in it. This makes design optimization of the bear- tive functions; spin frictional power loss and rating life on
ing a dedicated job to solve by conventional optimization 7007AC bearing used by Wei and Chengzu [15]. Subse-
method [1]. quently, Kumar et al. [13] attempted for design optimization
The term rolling element includes that bearing in which on a tapered roller bearing. Waghole and Tiwari [16] success-
load is carried out and transferred by rolling element. These fully formulated nonlinear problem of needle roller bearing
are widely popular due to its important properties like low and optimized by using hybrid method which involved arti-
starting torque, low kinetic friction, higher stiffness and the ficial bee colony algorithm (ABC) algorithm, differential
widest available range of load, speed and operating tem- search algorithm and grid search method. In another study,
peratures [2]. A whole unit of the bearings includes inner Tiwari and Waghole [17] applied ABC for the design opti-
ring, outer ring, rolling element and the cage which form a mization of spherical roller and validation was carried out by
statically indeterminate system. The bearing, which gener- comparing with grid algorithm results.
ally has several load-transmitting rolling elements, is itself It is observed from the above literature that relatively
also a high-grade statically indeterminate spring system [3]. less work is done on the design optimization of the REB.
With a minimum friction, a bearing can sense two require- Many deterministic optimization techniques can be applied
ments, i.e., supports the load and allows relative motion for the nonlinear optimization problem of bearings; however,
between two parts, like in the shaft and the housing [4]. the problem arises when numbers of design variables are
The complexity involved in the loading mechanism of the more. With continuous research in the field of optimization
bearings is due to the presence of local defects, and to pre- techniques, nature-inspired heuristic optimization methods
vent long-term breakdowns or catastrophic failures, it is very provide a better solution instead of classical optimization
necessary to detect a defect at its initial stage [5]. Accord- method. There are many nature-inspired well known and
ing to Yang et al. [6], bearing malfunctioning is broadly recently developed optimization algorithms such as Genetic
responsible for many losses in rotational mechanical com- Algorithm (GA), Artificial Bee Colony (ABC), artificial
ponents. In another study, Mehrjou et al. [7] claimed that neural network (ANN), particle swarm optimization (PSO),
the malfunctions for induction motors are about 40% bear- Harmony Search (HS), water cycle algorithm (WCA), mine
ing generated. Hence, optimization of the design variables blast algorithm (MBA), etc. and many of those proved
of bearings plays a very important role to overcome the said their suitability to many engineering optimization problems.
troubles. However, some of these advanced optimization techniques
Design of REB is a nonlinear optimization problem with have some limitations due to algorithm specific parame-
associated constraints [4]. In the literature, few articles are ters. In the present work, an attempt has been made to
reported showing the use of few optimization techniques for implement Teaching–learning-based optimization (TLBO)
different types of bearings and most of the relevant literatures for optimization of internal geometrical parameters of cylin-
among those are summarized here. Choi and Yoon [8] opti- drical roller bearing to determine the dynamic load bearing
mized design variables for double row angular contact ball capacity and deformation. The TLBO technique, developed
bearings of an automobile wheel-bearing unit using genetic by Rao et al. [18], is one of the recent techniques, becoming
algorithm (GA). Chakraborty et al. [9] applied optimized popular in the optimization aspects of various engineering
deep groove ball bearing (DGBB) using GA; however, certain problems. The objective of the present work is to apply TLBO
conditions like assembly angle was kept constant for all pairs technique for design optimization of REB, and working of
of solutions. Rao and Tiwari [10] attempted a parametric the same is explained in next section.
study on the design variables specified by Chakraborty et al.
[9] and applied bounds to the five constant variables involved
in the constraints. Gupta et al. [11] applied Non-dominated 2 Optimization Algorithm
Sorting Genetic Algorithm-II (NSGA-II) to a mathematical
model which was a multi-objective function and paramet- The concept of TLBO algorithm is based on the teaching–
ric study was carried out. Ganesan and Mohankumar [12] learning process happening in a class between the teacher and

123
Author's personal copy
Arabian Journal for Science and Engineering (2019) 44:7407–7422 7409

students, also referred as learners. The teacher is expected to Most of the popular optimization techniques like Genetic
attain best accord based on the output of learners within the Algorithm (GA), Particle swarm optimization (PSO), Artifi-
class. Objective of the teacher is to enhance the knowledge cial Bee Colony (ABC), etc., are also a population-based
level of the class and subsequently, output of the class is technique; however, these techniques require appropriate
evaluated by the teacher based on the grading system. Apart algorithm specific parameters that affect the performance of
from gaining knowledge from teacher, the learners also try to the algorithm [19–21,23,24]. For example, GA requires the
enhance their knowledge by interacting among themselves. fine tuning of crossover probability, mutation rate, and selec-
This basic theory of teaching–learning process is adopted tion method. PSO requires fine tuning of inertia weight, social
by Rao et al. [18] to develop this algorithm and the overall parameter and cognitive parameter. HS requires the harmony
working of this algorithm, as described by them is narrated memory consideration rate, pitch adjusting rate, and number
below. Like most of the nature-inspired algorithms, this tech- of improvisations and similar is the case with some other
nique is a population-based and a cluster of learners in a class advanced optimization techniques. However, the TLBO algo-
is considered as the population, whereas different subjects rithm does not require any algorithm specific parameters to
offered to the learners are analogous with different design be tuned, thus making the implementation of TLBO simpler
variables and the result of the learners is considered as the with better results.
fitness. It is mainly divided into two phases namely teacher TLBO finds a wide range of application in almost every
phase which covers interaction between teacher and learners field of engineering. Rao and Patel [25] used teaching–
and learner phase which covers interaction between learn- learning-based optimization (TLBO) and artificial bee colony
ers. (ABC) algorithms for determining the optimum operat-
Teacher in a class is considered as the most knowledge- ing conditions of combined Brayton and inverse Brayton
able person trying to increase knowledge level of the class cycles. Maximization of thermal efficiency and specific work
to his or her level. In mathematical way, the teacher moves of the system were considered as the objective functions
the mean of a class to higher level. This follows a random and treated simultaneously for multi-objective optimization.
process depending on many factors. In this phase, the best Krishnanand et al. [26] applied a multi-objective TLBO
solution out of initial population acts as teacher. The teacher algorithm with non-domination-based sorting to solve the
will try to move the mean of class to his level which will act environmental/economic dispatch problem containing the
as a new mean for the iteration. The mathematical approach, incommensurable objectives of best economic dispatch and
as used by Rao et al. [18] and subsequently experimented and least emission dispatch. Togan [27] applied TLBO technique
used in [19–24] is considered here. After updating the results for the discrete optimization of planer steel frame. The design
in teacher phase, fitness value for each learner correspond- algorithm aimed to obtain minimum weight frames sub-
ing to upgraded results is obtained and the superior function jected to strength and displacement requirements imposed by
value is considered after comparison with initial solution. the American Institute for Steel Construction (AISC) Load
This upgraded result of teacher phase then becomes the input and Resistance Factor Design (LRFD). Pawar and Rao [28]
for learner phase. presented the application of the TLBO to find the optimal
Learners can also improve their knowledge level through combination of process parameters of the various machining
the interaction with other learners. During their interaction, processes. Based on above implementation, basic steps of the
learner learns something new, if a knowledge level of other TLBO are summarized as the pseudo code, as follows Rao
learner is high. The mathematical approach compares the et al. [18].
results of learners after interaction with the each other, as
explained by Rao [24], and accepts the superior function
value and moves to the next iteration or stops as per the
termination criteria.

123
Author's personal copy
7410 Arabian Journal for Science and Engineering (2019) 44:7407–7422

123
Author's personal copy
Arabian Journal for Science and Engineering (2019) 44:7407–7422 7411

Subsequently, the TLBO algorithm is modified by various


researchers to make it compatible for a particular situation
[24]. However, the basic version is also equally capable to
handle complicated problems. Hence, to make the approach
easy to understand for the end users, the basic version of
TLBO algorithm is considered in the present work. The
TLBO algorithm is now applied for the design optimization
of two different cases of bearings, deep groove ball bear-
ings (DGBB) and cylindrical roller bearings (CRB), and the
description of the problem along with the result discussion
is presented in next sections.

3 Design Optimization Problem of Deep


Groove Ball Bearing

3.1 Internal Geometries and Terminology

A brief history of rolling element bearing (REB) published by


Hamrack and Anderson [2] in which internal geometry and
kinematics as well as material and manufacturing process
involved in the bearing manufacturing had been explained.
In DGBB, two rings, i.e., inner and outer ring, enclose the
rolling elements. Raceway is referred as the rolling area of
rollers on the rings. Though, rolling bearing appears as a
simple geometry, its inner geometry is mainly responsible
for the deflections, load distribution and amount of stress.
Figure 1a, b shows the internal geometry and nomenclature
of distinctive REB.
In general, three boundary dimensions can define the bear-
ing geometry that are outer diameter (D), bore diameter (d)
and width (B). Apart from this, the other variables required
to define internal geometry are pitch diameter (Dm ), ball
diameter (Db ), the inner and outer raceway curvature radius
coefficient ( f i and f o ), and number of rolling elements (Z).

3.2 Problem Inception for Bearing


Fig. 1 a Internal geometry of rolling element bearing [2], b sectional
From the literature, it is observed that various objective func- view of bearing races [2]
tions were proposed by different researchers based on various
reasons of failure, such as increased friction, fatigue, vibra-
where as per the Lundberg–Palmgren, value of ‘a’ is 3 and
tion, corrosions, noise and over-heating, etc. But out of these,
10/3 for point contact and line contact respectively. Life of
at normal operating conditions, the foremost reason of fail-
the bearing (L) is the total number of revolutions or, the num-
ure is contact fatigue. Longer the contact fatigue life means
ber of hours at a given constant speed of bearing operations
longer bearing life. Basic aim of the design optimization is
required for the failure criteria to develop [29].
to increase the fatigue life which is directly proportional to
the dynamic load rating (Cd ). The expression for fatigue life
3.3 Objective Function and Constraints
is given by Hamrack and Anderson [2] in Eq. (1)
As explained above, the fatigue life is a function of the
 a
Cd dynamic load rating. In present optimization problem, max-
L= 106 (1) imization of fatigue life is the main objective, similar to
F

123
Author's personal copy
7412 Arabian Journal for Science and Engineering (2019) 44:7407–7422

Table 1 Boundary and design constraints for DGBB


Design variables Constraint range Constraint condition
 
Db ≤ ϕ ϕ0
Number of balls (Z) 2 (Z − 1) sin−1 D 0 g1 (X ) = 
Db
 − Z +1≥0
m 2 sin−1 Dm

ϕ0 = the maximum tolerable assembly angle is depends on the geometry of the bearing
Ball Diameter (Db ) 2 ≤ Db ≤
K Dmin D−d g2 (X ) = 2Db − g3 (X ) = K Dmin (D − d) − 2Db ≥ 0
K Dmax D−d K Dmin (D − d) ≥ 0
2
KDmin and KDmax are unknown constants which decide the possible minimum and
maximum diameter and strength of the rolling element
Pitch Diameter (Dm ) g4 (X ) = Dm − (0.5 − e) (D + d) ≥ 0 g5 (X ) = (0.5 + e) (D + d) − Dm ≥ 0
e is a constant and obtained on the mobility conditions of balls
g6 (X ) = 0.5 (D − Dm − Db ) − ε Db ≥ 0
ε is constant obtained from strength consideration of the outer ring of bearing
Race conformity g7 (X ) = 0.52 ≥ f i ≥ 0.515 g8 (X ) = 0.53 ≥ f o ≥ 0.515

Chakraborty et al. [9], which is represented by Eqs. (2)–(4) remaining variables provide constraints to the basic design
as: variables.

max [ f (X )] = max [Cd] (2) 3.5 Constraint Conditions




⎪ 2
⎨ max f c Z 3 Db1.8 Db ≤ 25.4 mm
max [ f (X )] =
The process for obtaining bounds of design variables, which

⎩ max 3.647 f c Z 23 D 1.4
b Db > 25.4 mm defines more appropriate feasible search space for these vari-
(3) ables, is similar to Chakraborty et al. [9]. These bounds of
design parameters depend on the boundary dimensions of
With each bearing. Table 1 summarizes eight different problem
constraints associated with the problem.
⎡ ⎤
  1.72  0.41  103 −0.3
1 − γ f i (2 f o − 1) 3.6 Results and Discussion
f c = ⎣1 + 1.04 ⎦
1+γ f o (2 f i − 1)
  0.41 The aim of the present work is to optimize all important inter-
γ 0.3 (1 − γ )1.39 2fi
× (4) nal geometrical parameters which affects on the fatigue life.
1
(1 + γ ) 3 2fi −1
For this purpose, five design variables (Db , Z , Dm , f o , f i )
have been considered similar to Chakraborty et al. [9]. Along
Here, γ = DbDCosα
m
, is not an independent variable. For given with five main variables, other five constant parameters have
type of bearing, contact angle ∝ is zero. been used in the constraint equations that are: K D min =
0.5, K D max = 0.8, ε = 0.1, e = 0.1 and φ0 = 4.7124 radi-
3.4 Design Variables ans, which are obtained from the mobility condition, strength
of ball and rings.
From the observation of the Eqs. (1)–(4), design variables After making various trials for the algorithm settings, pop-
of the bearing and some unknown constraint constants, ulation size is considered as 50 for every individual run, and
explained in subsequent section, can be expressed as shown the convergence is attained for 100 generations to all the com-
by Eq. (5) bination of the design variables. The necessary MATLAB
code is prepared for the problem under consideration and
X = [Db , Z , Dm , f o , f i , K D min , K D max , ε, e, β] (5) checked a large number of times for consistency of results.
The optimum results obtained from the TLBO algorithm are
where Db is diameter of the ball, Z is number of balls, Dm tabulated in Table 2.
is pitch diameter, f o and f i is race conformity ratio ( f o = It is found that, as the bearing size increases, the pitch
ro /Db , f i = ri /Db ), r is referred as raceway groove curva- diameter has a propensity to increase in proportion and simi-
ture radius where ‘i’ and ‘o’ used to signify inner and outer, lar with the ball diameter. The numbers of ball are found to be
respectively. Out of the above variables, first five variables are changed in number with bearing. However, for some bearing
required to define the internal geometry of the bearing. The numbers, the converged numbers of balls are found to be the

123
Author's personal copy
Arabian Journal for Science and Engineering (2019) 44:7407–7422 7413

Table 2 Optimized design variables for 10 different standard DGBB


Type of bearing Boundary dimensions (Input) Design variables (output) Dynamic capacity
D (mm) d (mm) Dm (mm) Db (mm) fi fo Z Cd (N)

1 30 10 20.5592 7.8673 0.515 0.5157 7 8308.2


2 35 15 25.3202 7.9805 0.515 0.5150 8 10,279.9
3 47 20 34.0396 10.7998 0.515 0.5160 8 17,663.0
4 62 30 46.4499 12.8000 0.515 0.5180 9 27,076.7
5 80 40 61.0646 15.7794 0.515 0.5170 10 42,991.8
6 90 50 69.0237 15.9870 0.515 0.5170 11 47,691.7
7 110 60 86.0582 19.8275 0.515 0.5180 11 70,169.2
8 125 70 98.5332 21.9995 0.515 0.5180 11 84,853.9
9 140 80 111.1583 23.9997 0.515 0.5200 11 99,018.2
10 160 90 127.4944 27.0872 0.515 0.5170 11 128,168.8

of TLBO are around 10% superior to GA. Also, after com-


paring the results with Timken bearings for the light 200 W
series, it is observed that obtained results are 15% better in
average than the exiting bearing catalogue values [29].

4 Design Optimization Problem of


Cylindrical Roller Bearing (CRB)

4.1 Bearing Geometry

Roller bearings are usually used in the high-speed applica-


tion and for exceptionally large load-supporting conditions
which is difficult to obtain from ball bearings of same spec-
ifications. Fatigue life of CRB is very much affected by the
Fig. 2 Convergence plot for dynamic capacity of DGBB (bearing spec- internal geometry. At first sight, it appears simple; however,
ifications: D = 30, d = 10) it has varying effect on the stresses, deformation and load
distribution. CRB encompasses an inner ring, outer ring and
a single or multiple rows of the rollers. Standard boundary
same, even though they have different boundary conditions. dimensions include the bore diameter (d), outer diameter of
Hence, larger number of balls is not always advantageous, the bearing (D), and the width of the bearing (B) as shown
even balls with larger diameter with less in number provided in Fig. 3. To reduce edge loading, it is necessary to cham-
better fatigue life. The convergence of the result is shown in fer inner and outer raceways. Figure 3 shows the outer ring
Fig. 2 for the first type of bearing (i.e., D = 30 and d = 10). chamfering height (r1 ), outer ring chamfer width (r2 ), inner
Convergence rate of the solution with respect to number of ring chamfer height (r3 ) and the inner ring chamfer width
generation and search capability for TLBO is sufficiently (r4 ). The values of these heights and widths of chamfer can
higher. be obtained from standard catalogue with boundary dimen-
The obtained results of the TLBO in the present work are sions. In the present problem, standard boundary dimensions
compared with the results obtained by previous researcher are selected from the SKF general catalogue [31].
using GA [9], Timken ball bearing catalogue [30] and the
standard values available in books [29]. Comparison on the 4.2 Problem Inception for Design of CRB
basis of dynamic capacity is listed in the Table 3 which clearly
shows that the results of the TLBO have shown considerable From the literature, it is observed that many failure crite-
improvement over the previous results. ria were shortlisted by previous researches such as contact
Results of the TLBO are approximately two times superior fatigues, corrosions, over-heating, increased friction, vibra-
to standard values of Cd , which may be a sign of the best opti- tion and noise. Recent research in modern manufacturing
mality of the solution. Along with the standard values, results techniques enables production of the bearings with extremely

123
Author's personal copy
7414 Arabian Journal for Science and Engineering (2019) 44:7407–7422

Table 3 Comparison of optimized bearing geometry


Type of bearing Performance comparisons between (input) Dynamic capacities of the bearing (output)
D (mm) d (mm) Cd _std [17] Cd _Timken [26] Cd _GA [12] Cd _TLBO

1 30 10 3580 6800 7306.6 8308.2


2 35 15 5870 8650 9553.7 10,279.9
3 47 20 9430 14,400 16,213.4 17,663.0
4 62 30 14,900 22,200 25,785.0 27,076.7
5 80 40 22,500 36,200 38,979.1 42,991.8
6 90 50 26,900 40,000 45,161.4 47,691.7
7 110 60 40,300 62,100 64,542.3 70,169.2
8 125 70 47,600 69,000 81,701.2 84,853.9
9 140 80 55,600 81,300 95,915.9 99,018.2
10 160 90 73,900 109,000 121,401.9 128,168.8

Fig. 3 A typical cylindrical


roller bearing [12]

accurate internal and external geometries, with smooth where


rolling contact surfaces. Also modern steel-making process
⎡ ⎤ −2
increases outstanding homogeneity with few impurities, and 29    143  2
9 9

lubricants filtration method minimizes the contamination 2 1−γ 27 ⎢ 1−γ 108 ⎥


f c = 207.9λνγ 9 × ⎣1 + 1.04 ⎦ (7)
in rolling contact zones. This considerably reduces many 1+γ 4
1
1+γ
failures as listed above. However, at normal operating con-
ditions, contact fatigue plays important role in defining life and
of the bearings [32,33].
Dr
γ = (8)
Dm
4.3 Objective Functions and Constraints
where i is number of rows of rolling elements, le is effec-
Expressions for dynamic capacity and constraints for the tive length of roller, Z is number of rolling elements, Dr is
design optimization of CRB are similar to Kumar et al. [13] mean diameter of the roller, Dm is pitch diameter of the bear-
and the expression for dynamic capacity is represented by ing. λ is reduction factor, ν is the factor to account for edge
Eqs. (6)–(8), as given by Hamrock and Anderson [2]. loading. The factor bm in the Eq. 6 is to put up improve-
ments in geometrical accuracies. The present problem is of
29
single objective function for maximization of fatigue life of
7 3
Cd = bm f c (ile ) 9 Z 4 Dr27 (6) the bearing.

123
Author's personal copy
Arabian Journal for Science and Engineering (2019) 44:7407–7422 7415

Table 4 Constraint defining bounds for basic design variables of CRB


Design variables Constraint range Constraints
     
Pitch diameter (Dm ) d + 2r3min ≤ Dm ≤ C1 (X ) = Dm − d + 2r3min ≥ 0 C2 (X ) = D + 2r13min − Dm ≥ 0
 
D − 2r1min

Roller diameter (Dr ) Dr L B ≤ Dr ≤ Dr U B where, C3 (X ) = Dr − 268.71 σQcmax ≥ 0  ) =
C4 (X   
max
DrLB = 268.71 σQcmax And
1
2 D − 2r1min − d + 2r3min −
max Dr ≥ 0
Dr UB =   
1
2 D − 2r1min − d + 2r3min
       
π d+2r3min π D−2r1min π d+2r3min π D−2r1min
Number of rollers (Z) DrUB ≤Z≤ DrLB C5 (X ) = Z − DrUB ≥0 C6 (X ) = DrLB −Z ≥0

Table 5 Design constraints for designing of CRB


Constraint no Constraint conditions for design variables

Constraint 7 and 8 K Dmin D−d


2 ≤ Dr ≤ K Dmax D−d
2 0.4 ≤ K Dmin ≤ 0.5
0.6 ≤ K Dmax ≤ 0.7
C7 (X ) = 2Dr − K Dmin (D − d) ≥ 0 C8 (X ) = K Dmax (D − d) − 2Dr ≥ 0
Constraint 9 and 10 C9 = Dm − (0.5 − e) (D + d) ≥ 0 C10 = (0.5 − e) (D + d) − Dm ≥ 0
0.03 ≤ e ≤ 0.08
Constraint 11 C11 (X) = 0.5 (D − Dm − Dr ) − ε Dr ≥ 0 0.3 ≤ ε ≤ 0.4
Constraint 12 C12 (X ) = 1
2 (D − Do ) − 2r1min ≥ 0
Constraint 13 C13 (X ) = 1
2 (Di − d) − 1
2 (D − Do ) ≥ 0
Constraint 14 C14 (X ) = σcsa f e − σclimax ≥0
Constraint 15 C15 (X ) = σcsa f e − σclmax
o
≥0
 
π
Constraint 16 C16 (X ) = 2π − 2Z sin −1 Dr
Dm − Z 180 ≥0
Constraint 17 and 18 C17 = β B − le C18 = (B − le − 2r ) − 2r1 ≥ 0
0.7 ≤ β ≤ 0.85
Constraint 19 C19 = 1
2 (D − Do ) − 3Z static ≥ 0 with Z static = 0.626bo

Table 6 Input parameters used for design of CRB (TIMKEN Ball Bearing Catalog 2011)
Bearing no. D d B Standard boundary dimensions, mm Cd (kN) Limiting speed, rpm
r1min r2min r3min r4min

NU 202 35 15 11 0.6 0.6 0.3 0.3 12.50 26,000


NU 203 40 17 12 0.6 0.6 0.3 0.3 17.20 22,000
NU 303 47 17 14 1.0 1.0 0.6 0.6 24.60 20,000
NU 204 47 20 14 1.0 1.0 0.6 0.6 25.10 19,000
NU 304 52 20 15 1.0 1.0 0.6 0.6 35.10 18,000
NU 2205 52 25 18 1.0 1.0 0.6 0.6 34.10 16,000
NU 305 62 25 17 1.1 1.1 1.1 1.1 46.50 15,000
NU 2206 62 30 16 1.0 1.0 0.6 0.6 44.00 14,000
NU 207 72 35 17 1.1 1.1 0.6 0.6 56.00 12,000

123
Author's personal copy
7416 Arabian Journal for Science and Engineering (2019) 44:7407–7422

Table 7 Material properties of the bearing steel


Sr. no. Description Values

1. Young’s modulus 210 GPa


2. Poisson’s ratio 0.3
3. Safe contact stress 4000 MPa

4.4 Design Variables

Design variables for the design optimization of CRB are


expressed as:

 
Cd = f Dm , Dr , le , Z , K Dmin , K Dmax , ε, e, β (9)

Fig. 4 Convergence plot of dynamic capacity of CRB (Bearing no. NU


Out off above parameters, first four are independent vari- 202)
ables which need to describe internal geometry of the
bearings. Therefore, these four self-determining variables
Constraints C3 and C4 : Dynamic capacity of the CRB is
(Dm , Dr , le , Z ) are selected as design variables. Other
 than directly proportional with mean diameter of the roller. Limits
self-determining  variables, constraints constants K Dmin ,
of the roller mean diameter are obtained by using maximum
K Dmax , ε, e, β are also taken as a design variable [13] which
compressive stress at the line contact. Least value of the mean
provides constraint to basic design variables.
diameter must withstand the Hertz contact stress developed
due to external loads.
4.5 Constraints Constraints C5 and C6 : Number of rolling elements is
directly appear in the objective functions and has great impor-
The constraints involved in the problem reduce the search tance as it is directly proportional to dynamic load rating.
domain to feasible search domain. The problem involves Lower bound of pitch diameter is used to find lower bound
considerable number of constraints which are categorized of the number of rollers and roller mean diameter used to
and described as below. Constraints 1–6 are associated with find the upper limit.
the bounds of the design variables and are summarized in Constraint C7 and C8 : An unknown constant given in
Table 4, whereas 9–19 are associated with the design of CRB equations decides the upper and lower value of the mean
and those are summarizes in Table 5. roller diameter. The bound of constant is given as 0.4 ≤
Constraints C1 and C2 : These constraints ensure the space K Dmin ≤ 0.5, 0.6 ≤ K Dmax ≤ 0.7 which is determined by
for chamfering of corners of rings. For this, pitch diameter the parametric study [11].
must lie between inner and outer diameters of the roller bear- Constraint C9 and C10 : These constraints used to ensure
ing. the operational mobility of the bearings where e is unknown

Table 8 Optimized design parameters for CRB obtained through TLBO


Bearings no. Optimized basic design variables CdTLBO CdGA CdSKF
Dm Dr Z le kDmin kDmax e ε β

NU 202 25.00 6.25 11 8 0.4780 0.6347 0.0609 0.3652 0.8361 22.64 18.48 12.50
NU 203 28.50 6.90 12 9 0.4868 0.6991 0.0694 0.3299 0.7789 27.16 23.58 17.20
NU 303 32.00 7.80 12 11 0.4187 0.6106 0.0429 0.3891 0.8083 36.27 31.87 24.60
NU 204 33.50 7.95 13 11 0.4015 0.6796 0.0333 0.3339 0.8123 39.09 32.23 25.10
NU 304 36.00 8.36 14 12 0.4587 0.6211 0.0319 0.3187 0.8442 46.42 45.48 35.50
NU 2205 38.50 7.62 17 14 0.4786 0.6477 0.0576 0.3096 0.7852 44.22 39.75 34.10
NU 305 44.45 8.45 18 12 0.4374 0.6321 0.0781 0.3230 0.7951 54.44 50.39 46.50
NU 2206 46.00 8.25 18 12 0.4017 0.6226 0.0418 0.3980 0.8483 52.46 51.64 44.00
NU 207 53.50 7.81 22 12 0.4007 0.6783 0.0569 0.3314 0.7842 64.60 60.97 56.00

123
Author's personal copy
Arabian Journal for Science and Engineering (2019) 44:7407–7422 7417

Fig. 5 Average value of fitness


versus iterations

Fig. 6 Standard deviation


versus iterations

constant whose values are taken as 0.03 ≤ e ≤ 0.08 which Constraint C14 : In CRB, large amount of force is applied
is determined by the parametric study [11]. between rolling elements and raceway develops only small
Constraint C11 : Thickness of the bearing, the outer race- area of contact between mating members. This results in very
way of the bearing ring should not be less than ε Db where ε high contact stress which may cause permanent deformation.
is constant obtained from strength deliberation of outer ring However, due to the rigid nature of the rolling members, this
of bearing whose value lies between 0.3 ≤ ε ≤ 0.4. deformation is of very low order, e.g., 0.025 mm or less. The
Constraint C12 : To perform chamfering on the outer race, safe contact stress value is limited to 4000 N/mm2 .
it should be large enough. Therefore, these constraints keep Constraint C15 : As like constraint 14, the contact stress in
restriction on inner diameter of outer ring. the outer ring should be within the safe limit.
Constraint C13 : In CRB, inner ring is subjected to more Constraint C16 : This constraint will put the limitation on
stress due to convex curvature and hence, this constraint is number of rollers in the bearings. For a smooth rolling of
necessary for the inner ring thickness. rollers in the inner and outer race and to avoid collision among
them, a proper spacing is required between rollers. Minimum

123
Author's personal copy
7418 Arabian Journal for Science and Engineering (2019) 44:7407–7422

Fig. 7 Number of iterations


versus fitness value of all 50
iteration for NU202

of one degree angular spacing is to be allowed as in standard All the necessary precautions are taken to develop the
bearing. MATLAB code for this problem as it involves considerable
Constraint C17 and C18 : In this work, the effective length number of constraints and large numbers of trials are taken
of roller is taken to the corresponding standard roller diam- to check the consistency of results and the results obtained
eter. This constraint limits the effective length of the roller from TLBO algorithm are tabulated in Table 8. The results
with respect to width of the bearing. βis unknown constraint are obtained after 100 generations with the population size
constant and its range is between 0.7 ≤ β ≤ 0.85. of 50 and all cases of bearings under consideration are run
Constraint C19 : For the worst case, outer ring of bear- for 50 times each to check for consistency of the results. The
ing should be thick enough to withstand maximum dynamic convergence of fitness value for the first bearing NU 202
shear stress that occurs at the center of the ring. (D = 35, d = 15, B = 11) is shown in Fig. 4.
Efforts are made to take into account all these constraints In the manufacturing scenario of bearing, some percentage
so as to obtain the realistic design of the bearing. of deviation is possible in the values of dependent variables.
For the better evaluation of bearing design, it is necessary to
4.6 Results and Discussion observe the sensitivity of the variables to obtain performance
measures. After many generation and number of runs, it is
The aim of the present work is to optimize the 9 design vari- observed that the percentage variation of dynamic capacity
ables (Dm , Dr , le , Z , K Dmin , K Dmax , ε, e, β) involved in the are very small in magnitude. For the better understanding
design of CRB so as to maximize the dynamic capacity of the of converged result, average value and standard deviation of
bearing similar to Kumar et al. [13]. Out of the 9 variables, fitness function of all 50 iterations is shown in Figs. 5 and 6.
first four are independent variables and remaining five are Deviation between maximum and minimum average values
constraint constants which appears in constraint equations. of all iterations is less than 4%.
Numbers of CRBs are selected from the standard catalogue All 50 iterations results also have shown graphically
which is listed in Table 6. Boundary dimensions such as in Fig. 7. It is observed from the figure, those maximum
outer–inner diameters, width, chamfering size and limiting solutions lies in 22000 to 22700 ranges which shows the
speed of operation of bearings are taken from the SKF general consistency of the TLBO algorithm. Maximum values of
catalogue [31]. Standard dynamic capacity (Cd ) is also listed pitch diameter lie in between 25 and 25.4 mm. Figure 8a
in the Table 6 and for the calculation purpose, equivalent shows the variation of pitch diameter of the best solution
radial load assumes to be 60% of the total standard dynamic with the number of iterations. The pitch diameter fluctuates
capacity. Table 7 shows the necessary material properties like up to 9th generations, then decreases suddenly, and then, it
Young’s modulus, Poisson’s ratio and safe contact stress con- stabilizes to a value of 25 mm. After that, a slight increment
sidered during the design. All the data given in Tables 6 and 7 in the pitch diameter to the value of 25.35 mm takes place
are used as input for the considered algorithm for the design beyond the 20th generations. Sudden step rise in the roller
optimization of CRB.

123
Author's personal copy
Arabian Journal for Science and Engineering (2019) 44:7407–7422 7419

Fig. 8c. It is due to new approach of constraint handling used


in the programming in which a special loop is inserted at
the initial step of algorithm which only allowed constraint
satisfying value for the next step. This has an advantage that
it automatically satisfies the concept of duplicate removal
and penalty functioning. Figure 9a–e shows the variation of
constraint constants with the number of generations.
Similar to NU 202, the same procedure is carried out for all
the selected bearing series and the results obtained for inter-
nal geometries of the bearings are summarized in Table 8.
The obtained results of the TLBO in the present work are
compared with the results obtained by previous researcher
using GA, SKF cylindrical roller bearing catalogue values.
It is observed that results of TLBO have shown considerable
improvement over the previous results.

5 Result Validation

Efforts are also made to validate the optimum results obtained


from the TLBO so as to make the approach more realistic.
For this purpose, the well-developed approach of FEA is con-
sidered by using the ANSYS tool. Similar approach related
to design simulation of bearings using FEA is reported in the
literature and few important among those are summarized
here. In design of CRB, Poplawski et al. [32,33] used four
rolling element theories to predict and compare the dynamic
capacity and life of the bearings, by considering stress and
life analysis in FEA. Cavallaro et al. [34] presented a sys-
tematic method to account for the structural deformation in
REB ring and housing based on Roark’s formulas and defor-
mations were compared with those obtained with the FEA.
Stress and displacement distributions of rings of CRB using
FEA investigated by Demirhan and Kanber [35]. A work
on the crowned CRB by considering three profiles that are
standard, modified partially crowned and logarithmic roller
Fig. 8 a Variation of pitch diameter with number of iterations for the crowing proposed by Fujiwara et al. [36]. Lostado et al. [37]
cylindrical roller bearing NU 202, b variation of roller diameter with applied multiple repose optimization combined with FEA for
number of iterations for cylindrical roller bearing, NU 202, c variation
the design optimization of tapered roller bearings.
of number of roller with number of iterations
Thus, the validation of bearing design using FEA approach
is a well-established practice, and hence, in this article, the
diameter is due to the discrete nature of the standard pitch same approach is used to validate the obtained optimum
diameter. design setting. Out of the various bearings, CRB NU 202
Figure 8b shows the variation of roller diameter with num- is chosen for validation purpose, as it is more complicated
ber of generations. It can be observed that value of roller compared to DGBB. By considering the corresponding bear-
diameter lies in between 6 and 6.25 mm. All the values ing contact and total deformation, a 3D model is developed
in between above-mentioned range satisfy all the constraint using ANSYS and subsequently the contact conditions are
related to roller diameter. Maximum value of dynamic capac- analyzed.
ity is obtained at value which is near to 6.25 mm. Similar to ANSYS supports two surface contact elements like rigid–
pitch diameter, sudden step rise and down the value of roller flexible or flexible–flexible and in the present case, type
diameter is due to discrete nature of given variable. contact element considered is flexible–flexible and type of
It can be observed that the number of rollers in given contact is line contact. By using fine meshing type, the
problem, settled at 11 throughout the simulation shown in whole model has 318,048 nodes and 194755 elements. All

123
Author's personal copy
7420 Arabian Journal for Science and Engineering (2019) 44:7407–7422

Fig. 9 Variation of constraint constants with number of iterations for, ning mobility of bearings, d ε, which maintain the thickness of outer
a K Dmin , which decides the lower bound of pitch diameter, b K Dmax , ring, e β, which apply the constraint on length of roller
which decides upper bound of pitch diameter, c e, which ensure the run-

Table 9 Comparison between theoretical and analytical results


Sr, no. Theoretical value by TLBO Analytical value by ANSYS

1. Total maximum contact stress on inner ring 3399 MPa 3531 MPa
2. Total maximum contact stress on outer ring 2633 MPa 2688 MPa
3. Total deformation on roller and ring contact 0.023 mm 0.02268 mm

the obtained results for Hertz contact stress and total defor- number of bearings involving relevant constraints. A latest
mation is tabulated in Table 9 and Fig. 10a–g illustrates the optimization approach known as Teaching–Learning-based
approach used for FEA starting from development of model optimization approach is considered to obtain the optimum
to end results. design of the chosen bearings. The dynamics capacity of
The optimum output design variables obtained by TLBO DGBB has been found to be increased by approximately two
are checked by means of FEA simulation for contact stress, times than the standard values available in the data book. The
deformation and fatigue life for inner and outer ring of the results are also 10% superior to the values obtained from GA
bearing so as to validate the results. The obtained FEA sim- and 15% better on average than existing bearing catalogue
ulation results are shown in Table 9 which clearly indicates values.
that both results are very near to each other, thereby closely In case of optimal design of CRB, which involves a large
validating the optimum results presented in this work. number of constraints, TLBO shows considerable improve-
ment over the previous results of GA and SKF standard
catalogue. Efforts are also made to validate the obtained
6 Conclusions results of CRB using FEA approach of ANSYS which
shows close agreement with the obtained theoretical solu-
In the present work, optimum methodology for the design tions. Convergence study ensured the comprehensive point
of DGBB and CRB is proposed by considering a large in the design.

123
Author's personal copy
Arabian Journal for Science and Engineering (2019) 44:7407–7422 7421

Fig. 10 a Model development, b geometry meshing, c boundary conditions, d stress distribution on whole body, e total stress distribution on inner
ring, f total maximum contact stress on outer ring, g total deformation on inner and outer race

123
Author's personal copy
7422 Arabian Journal for Science and Engineering (2019) 44:7407–7422

References 20. Rao, R.V.; Patel, V.: An elitist teaching–learning-based opti-


mization algorithm for solving complex constrained optimization
1. He, S.; Prempain, E.; Wu, Q.H.: An improved particle swarm opti- problems. Int. J. Ind. Eng. Comput. 3(4), 535–560 (2012)
mizer for mechanical design optimization problems. Eng. Optim. 21. Rao, R.V.; Patel, V.: Comparative performance of an eli-
36(5), 585–605 (2004) tist teaching–learning-based optimization algorithm for solving
2. Hamrock, B.J.; Anderson, W.J.: Rolling-elements bearing. NASA unconstrained optimization problems. Int. J. Ind. Eng. Comput.
reference publication. 1105 (1983) 4, 29–50 (2013)
3. Harris, T.A.: Rolling Bearing Analysis, 4th edn. Wiley, New York 22. Rao, R.V.; Kalyankar, V.D.: Parameter optimization of modern
(2001) machining processes using teaching–learning-based optimization
4. Changsen, W.: Analysis of Rolling Element Bearings. Mechanical algorithm. Eng. Appl. Artif. Intell. 26, 524–531 (2013)
Engineering Publications Ltd., London (1991) 23. Rao, R.V.: Teaching Learning Based Optimization and its Engi-
5. Gupta, P.K.: Current status of and future innovations in rolling neering Applications. Springer, London (2015)
bearing modeling. Tribol. Trans. 54(3), 394–403 (2011) 24. Rao, R.V.: Review of applications of TLBO algorithm and a tutorial
6. Yang, H.; Mathew, J.; Ma, L.: Fault diagnosis of rolling element for beginners to solve the unconstrained and constrained optimiza-
bearings using basis pursuit. Mech. Syst. Signal Process. 19, 341– tion problems. Decis. Sci. Lett. 5, 1–30 (2016)
356 (2005) 25. Rao, R.V.; Patel, V.: Multi-objective optimization of combined
7. Mehrjou, M.R.; Mariun, N.; Marhaban, M.H.; Misron, N.: Rotor Brayton and inverse Brayton cycles using advanced optimiza-
fault condition monitoring techniques for squirrel-cage induction tion algorithms. Eng. Optim. (2011). https://doi.org/10.1080/
machine-A review. Mech. Syst. Signal Process. 25(8), 2827–2848 0305215X.2011.624183
(2011) 26. Krishnanand, K. R.; Panigrahi, B. K.; Rout, P. K.; Mohapatra, A.;
8. Choi, D.H.; Yoon, K.C.: A design method of an automotive Application of multi-objective teaching–learning-based algorithm
wheel-bearing unit with discrete design variables using genetic to an economic load dispatch problem with in commensurable
algorithms. J. Tribol. 123, 181–187 (2001) objectives. In: Swarm, Evolutionary, and Memetic Computing,
9. Chakraborty, I.; Kumar, V.; Nair, S.B.; Tiwari, R.: Rolling element Lecture Notes in Computer Science, vol. 7076. Springer, Berlin
bearing design through genetic algorithms. Eng. Optim. 35(6), (2011)
649–659 (2003) 27. Togan, V.: Design of planar steel frames using Teaching–Learning
10. Rao, B.R.; Tiwari, R.: Optimum design of rolling element bearings based optimization. Eng. Struct. 34, 225–232 (2012)
using genetic algorithms. Mech. Mach. Theory 42, 233–250 (2007) 28. Pawar, P.J.; Rao, R.V.: Parameter optimization of machining pro-
11. Gupta, S.; Tiwari, R.; Nair, S.B.: Multi-objective design optimiza- cesses using teaching–learning-based optimization algorithm. Int.
tion of rolling bearings using genetic algorithm. Mech. Mach. J. Adv. Manuf. Technol. 67(5–8), 995–1006 (2013)
Theory 42, 1418–1443 (2007) 29. Shigley, J.E.: Mechanical Engineering Design. McGraw-Hill Book
12. Ganesan, H.; Mohankumar, G.: Optimization of machining tech- Company, New York (2011)
niques in CNC turning centre using genetic algorithm. Arab. J. Sci. 30. TIMKEN: Ball Bearing Catalog (2011)
Eng. 38, 1529–1538 (2013) 31. SKF: General Catalogue. Germany (2005)
13. Kumar, K.S.; Tiwari, R.; Reddy, R.S.: “Development of an opti- 32. Poplawski, J.V.; Peters, S.M.; Zaretsky, E.V.: Effect of roller profile
mum design methodology of cylindrical roller bearings using on cylindrical roller bearing life prediction–Part I: comparison of
genetic algorithms. Int. J. Comput. Methods Eng. Sci. Mech. 9(6), bearing life theories. Tribol. Trans. 44(3), 339–350 (2001a)
321–341 (2008) 33. Poplawski, J.V.; Peters, S.M.; Zaretsky, E.V.: Effect of roller profile
14. Kumar, K.S.; Tiwari, R.; Prasad, P.V.V.N.: An Optimum design on cylindrical roller bearing life prediction–Part II: comparison of
of crowned cylindrical roller bearings using genetic algorithms. J. roller profiles. Tribol. Trans. 44(3), 417–427 (2001b)
Mech. Des. 131(5), 051011 (2009) 34. Cavallaro, G.; Nelias, D.; Bon, F.: Analysis of high-speed inter
15. Wei, Y., Chengzu, R.: Optimal design of high speed angular contact shaft cylindrical roller bearing with flexible rings. Tribol. Trans.
ball bearing using a multi-objective evolution algorithm. In: Paper 48(2), 154–164 (2005)
presented at the International Conference on Computing. Control 35. Demirhan, N.; Kanber, B.: Stress and displacement distributions
and Industrial Engineering. IEEE, Wuhan, June 5–6 (2010) on cylindrical roller bearing rings using FEM. Mech. Based Des.
16. Waghole, V.; Tiwari, R.: Optimization of needle roller bearing Struct. Mach. Int. J. 36, 86–102 (2008)
design using novel hybrid methods. Mech. Mach. Theory 72, 71–85 36. Fujiwara, H.; Kobayashi, T.; Kawase, T.; Yamauchi, K.: Optimized
(2014) logarithmic roller crowning design of cylindrical roller bearings
17. Tiwari, R.; Waghole, V.: Optimization of spherical roller bearing and its experimental demonstration. Tribol. Trans. 53(6), 909–916
design using artificial bee colony algorithm and grid search method. (2010)
Int. J. Comput. Methods Eng. Sci. Mech. 16(4), 221–233 (2015) 37. Lostado, R.; Garcia, R.E.; Martinez, R.F.: Optimization of operat-
18. Rao, R.V.; Savsani, V.J.; Vakharia, D.P.: Teaching-learning-based ing conditions for a double-row tapered roller bearing. Int. J. Mech.
optimization: a novel method for constrained mechanical design Mater. Des. (2015). https://doi.org/10.1007/s10999-015-9311-4
optimization problems. Comput. Aided Des. 43, 303–315 (2011)
19. Rao, R.V.; Savsani, V.J.; Balic, J.: Teaching-learning-based
optimization algorithm for unconstrained and constrained real-
parameter optimization problems. Eng. Optim. 44(12), 1447–1462
(2012)

123

You might also like