Multi-Objective Teaching-Learning-Based Optimization Algorithm For Reducing Carbon Emissions and Operation Time in Turning Operations

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/271853578
Multi-objective teaching–learning-based optimization

algorithm for reducing carbon emissions and operation time in
turning operations
Article in Engineering Optimization · December 2014

DOI: 10.1080/0305215X.2014.928818
CITATIONS READS
32 282
8 authors, including:
Wenwen Lin Shaojun Wang

Ningbo University South East Missouri State University
13 PUBLICATIONS 158 CITATIONS 37 PUBLICATIONS 1,395 CITATIONS
SEE PROFILE SEE PROFILE
Chaoyong Zhang
Huazhong University of Science and Technology
153 PUBLICATIONS 1,873 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Dual-driving system View project
View project
All content following this page was uploaded by Chaoyong Zhang on 28 May 2015.
The user has requested enhancement of the downloaded file.

This article was downloaded by: [Huazhong University of Science & Technology ]
On: 28 May 2015, At: 02:16
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Engineering Optimization
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/geno20
Multi-objective teaching–learning-based
optimization algorithm for reducing
carbon emissions and operation time in
turning operations
a a b a
Wenwen Lin , D.Y. Yu , S. Wang , Chaoyong Zhang , Sanqiang
a c d e
Zhang , Huiyu Tian , Min Luo & Shengqiang Liu
a
School of Mechanical Science and Engineering, Huazhong
University of Science and Technology, Wuhan, PR China
b
Department of Industrial & Engineering Technology, Southeast
Missouri State University, Cape Girardeau, Missouri, USA
Click for updates c
Shan Dong Hoteam Software Co. Ltd, Jinan, PR China
d
School of Electrical & Information Engineering, Hubei University
of Automative Technology, Shiyan, PR China
e
China National Heavy Duty Truck Group Co. Ltd, Jinan, PR China
Published online: 30 Jun 2014.
To cite this article: Wenwen Lin, D.Y. Yu, S. Wang, Chaoyong Zhang, Sanqiang Zhang, Huiyu Tian,
Min Luo & Shengqiang Liu (2015) Multi-objective teaching–learning-based optimization algorithm for
reducing carbon emissions and operation time in turning operations, Engineering Optimization, 47:7,
994-1007, DOI: 10.1080/0305215X.2014.928818
To link to this article: http://dx.doi.org/10.1080/0305215X.2014.928818
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
Downloaded by [Huazhong University of Science & Technology ] at 02:16 28 May 2015
Engineering Optimization, 2015
Vol. 47, No. 7, 994–1007, http://dx.doi.org/10.1080/0305215X.2014.928818
Multi-objective teaching–learning-based optimization algorithm

for reducing carbon emissions and operation time in turning
operations
Wenwen Lina , D.Y. Yua , S. Wangb , Chaoyong Zhanga∗ , Sanqiang Zhanga , Huiyu Tianc ,
Min Luod and Shengqiang Liue
a School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan,
PR China; b Department of Industrial & Engineering Technology, Southeast Missouri State University,
Cape Girardeau, Missouri, USA; c Shan Dong Hoteam Software Co. Ltd, Jinan, PR China; d School of
Electrical & Information Engineering, Hubei University of Automative Technology, Shiyan, PR China;
e China National Heavy Duty Truck Group Co. Ltd, Jinan, PR China
(Received 22 January 2014; accepted 14 May 2014)
In addition to energy consumption, the use of cutting fluids, deposition of worn tools and certain other
manufacturing activities can have environmental impacts.All these activities cause carbon emission directly
or indirectly; therefore, carbon emission can be used as an environmental criterion for machining systems.
In this article, a direct method is proposed to quantify the carbon emissions in turning operations. To
determine the coefficients in the quantitative method, real experimental data were obtained and analysed in
MATLAB. Moreover, a multi-objective teaching–learning-based optimization algorithm is proposed, and
two objectives to minimize carbon emissions and operation time are considered simultaneously. Cutting
parameters were optimized by the proposed algorithm. Finally, the analytic hierarchy process was used
to determine the optimal solution, which was found to be more environmentally friendly than the cutting
parameters determined by the design of experiments method.
Keywords: metal-cutting parameter optimization; sustainable manufacturing; orthogonal arrays; multi-

objective optimization; teaching–learning-based optimization
1. Introduction
According to the Annual Energy Review 2011 (EIA 2012), industrial energy consumption was
responsible for approximately 40% of global energy consumption and has nearly doubled in the
past 60 years. The use of fossil fuels causes the direct emission of greenhouse gases such as
carbon dioxide (CO2 ). Although the consumption of electricity does not generate CO2 directly,
CO2 is released into the atmosphere during power generation. Besides energy consumption, other
manufacturing activities cause carbon emissions, such as the use of cutting fluids, deposition of
worn tools and material consumption (Li et al. 2013). Moreover, unit production cost increases
with increasing energy price.
Benardos and Vosniakos (2003) reported that turning was the most common metal-cutting
operation in machining systems. The material removal rate (MRR), surface roughness, tool wear,
∗ Corresponding author. Email: zcyhust@mail.hust.edu.cn
© 2014 Taylor & Francis

Engineering Optimization 995
production cost and operation time have been the most commonly used optimization criteria in the
published literature (Yusup, Zain, and Hashim 2012). Recently, several studies have emphasized
energy saving in metal-cutting operations. Fratila and Caizar (2011) investigated the effects of
metal-cutting parameters on surface roughness and power consumption using orthogonal arrays,
signal-to-noise ratio and analysis of variance. Yan and Li (2013) aimed to develop a milling
operation model to find a trade-off solution using three criteria: energy consumption, MRR and
surface roughness. However, carbon emissions were not considered by the above-mentioned
studies.
Metal-cutting parameter optimization methods can be classified into three categories: design
of experiments (DOE), artificial neural networks (ANNs) and metaheuristics. The DOE method
analyses the relations between the input and response variables experimentally. A criticism of
this technique is that it is limited by the number of variables and cost of experiments, leading
to local optimal solutions. Moreover, the conflicting criteria have to be combined into a single
objective using grey relational analysis (GRA) to solve the multi-objective optimization prob-
lems in the DOE method (Yan and Li 2013). The ANN is a good modelling technique when
faced with difficulties in building empirical models for machining operations. Then, the model is
determined as the objective function of optimization algorithms. Metaheuristics could efficiently
optimize machining operations with one or more criteria (Pawar and Rao 2013). However, under-
lying relationships between the cutting parameters and objective functions should be determined
experimentally for real applications.
The objective of this research was to investigate the optimization problem of metal-cutting
parameters using a new optimization criterion, carbon emission, which thoroughly considers the
environmental impacts of machining systems. A direct method is proposed to quantify the carbon
emissions during an entire turning operation. Real experimental data were obtained and analysed
in MATLAB® to determine the coefficients in the proposed method. Then, the studied problem was
optimized using a multi-objective teaching–learning-based optimization (MOTLBO) algorithm.
Finally, the analytic hierarchy process (AHP) was used to determine the optimal solution.
The remainder of the article is organized as follows. Section 2 briefly describes the quantitative
method for calculating the carbon emissions during the entire turning operation. In Section 3, the
experiments and regression results are described in detail. In Section 4, the details of the MOTLBO
algorithm are described. The comparison results are presented in Section 5, and conclusions and
directions for future research are described in Section 6.
2. Quantitative measurements of carbon emissions in machining systems
2.1. Carbon emissions of machining systems under dry cutting conditions
The total carbon emissions, CEtotal , of a machining system are mainly caused by the energy
conversion for the machining system, CEelec , material processing system, CEm , removed materials,
CEchip , and cutting tools, CEtool (Li et al. 2013). CEm and CEchip are determined at the stage of
product design, and CEelec and CEtool are determined during the machining operation. Therefore,
the carbon emissions caused during a machining operation, CEopt , can be calculated as follows:
CEopt = CEelec + CEtool (1)
2.1.1. Carbon emissions of energy conversion for machining system, CEelec
Electricity is the primary energy consumed by machine tools. CEelec is equal to the product of the
electricity carbon emission factor, CEFelec , and the energy consumption of the machine system,
996 W. Lin et al.
Figure 1. Power measurement example of a turning operation.
ECmachine (Equation 2):

CEelec = CEFelec × ECmachine (2)
The CEFelec values of the main power grids were provided by the National Development and
Reform Commission in China (NDRC 2011). The average value of these factors is 0.1524 g
CO2 /kJ.
Balogun and Mativenga (2013) classified the machine tool states during machining operations
into three categories: basic, ready and cutting states. In the basic state, the required machine
components are activated to ensure the operational readiness of the machine tools. In the ready
state, more power is required for drives and spindle movement to bring the tool and workpiece to the
correct positions. In the cutting state, additional power is required by the tool tip to remove material,
as well as for heat generation, noise, friction and other energy loss. The energy consumption cycle
of a turning operation can be simplified into six phases (Figure 1). In phase 2, an additional state,
the transition state, is observed, in which more power is needed to turn on drives and spindle
motors and bring the machine tool to the ready state (Figure 1). Thus, ECmachine can be calculated
by adding the energy consumptions during various states (Equation 3):
ECmachine = ECb + ECt + ECr + ECc (3)
where ECb , ECt , ECr and ECc denote the energy consumptions during the basic, transition, ready
and cutting states, respectively.
The machine tool consumes a constant power during the basic state and can be measured by
experiments (Mori et al. 2011). Therefore, ECb can be calculated using Equation (4):
ECb = Pb × (t1 + t6 ) (4)
where Pb is the basic power consumption during the basic state and ti is the cycle time in phase i
(i = 1, 2, …, 6).
The energy profile of the transition state is complex. The mathematical relation between the
energy consumption during the transition state, ECt , and the rotation speed, n, can be expressed
as Equation (5) (Liu et al. 2012):
ECt = x1 n2 + x2 n + x3 (5)
In the ready state, the total input power, Pr , consists of two parts: the basic power, Pb , and the
required operational power, Poperational , because of the essential operations related to the cutting
process, such as the rotation of the spindle and movement of the tool. Li and Kara (2011) proved
that Poperational has a linear relationship with the rotation speed, n, as shown in Equation (6):
Poperational = k1 × n + k2 (6)
where k1 and k2 are the coefficients of the linear model and n is the rotation speed. Therefore,
ECr can be obtained using Equation (7):
ECr = Pr × (t3 + t5 ) = (Pb + Poperational ) × (t3 + t5 ) (7)
Li and Kara (2011) introduced a new concept, specific energy consumption (SEC), which is the
energy required for removing 1 cm3 of material. SEC could be modelled as an inverse function
of MRR (Equation 8). It is simpler to evaluate ECc using the SEC model (Equation 9) (Balogun
and Mativenga 2013):

C1
SEC = C0 + (8)
MRR
ECc = MRV × SEC (9)
where C0 and C1 are the coefficients of the inverse model and can be obtained using some
mathematical regression tools. MRV is the material removal volume, and MRR is the material
removal rate, which can be calculated using Equation (10):
v × f × ap
MRR = (10)
60
where v, f and ap are the cutting speed, feed rate and cutting depth, respectively.
ECc is measured by a power monitoring system; thus, SEC can be calculated using
Equation (11):

ECc Pinput dt Pinput
SEC = = = (11)
MRV MRV MRR
where Pinput is the input power of the machine tool during the cutting state and Pinput is the average
value of Pinput .
Substituting CEelec in Equations (2)–(4), (7) and (9), CEelec can be expressed as follows:

P × (t1 + t6 )+ ECt + (P b + Poperational ) × (t3 + t5 )
CEelec = CEF elec × b (12)
+ MRV × C0 + MRR C1
2.1.2. Carbon emissions caused by cutting tools, CEtool
CEtool is generated by the production of tools. Thus, CEtool is uniformly distributed during the
entire expected tool lifetime and can be calculated using Equation (13):
CEtool = ξ × CEFtool × Wtool (13)
where ξ is the used tool life and is considered as a part of the entire tool life, which is consumed
in metal-cutting operations, CEFtool is the tool carbon emission factor, and Wtool is the tool mass.
998 W. Lin et al.
ξ can be calculated using Equations (14) and (15):

MRV
ξ= × 100% (14)
MRR × T
Aα
T = 60 α β (15)
v f
where T is the tool life, and α and β are the coefficients related to the tool and workpiece materials,
respectively.
Rajemi, Mativenga, and Aramcharoen (2010) investigated the energy footprint for tool inserts.
The weight of the tool insert was 9.5 g and the energy embodied was 5.3 MJ. CEFtool can be
calculated using Equation (16):
CEFelec × Total energy per insert
CEFtool =
Wtool
(16)
0.1524 × 5.3
= = 95.048g CO2 /g
9.5
Substituting CEopt in Equations (1), (12) and (13), the carbon emission objective function, CEopt ,
for the metal-cutting parameter optimization model can be expressed as follows:

P × (t1 + t6 )+ ECt + (P b + Poperational ) × (t3 + t5 )
CEopt = CEFelec × b + (17)
+ MRV × C0 + MRR C1
ζ × CEFtool × Wtool
2.2. Operation time
Operation time, topt , is another objective considered in the proposed model. topt can be measured
as the total time required for the entire turning operation and equals the sum of the cycle times,
ti (i = 1, 2, . . ., 6) (Equation 18):
topt = t1 + t2 + t3 + t4 + t5 + t6 (18)
The cycle time during the cutting state, t4 , depends on the cutting parameters and can be calculated
using Equation (19). Other cycle times are related to the operator’s skill and machine tool’s
characteristics, and are considered as constants (Mori et al. 2011).

MRV tct
t4 = × 1+ (19)
MRR T
where tct is the tool changing time. Therefore, topt can be re-expressed as follows:

MRV tct
topt = t1 + t2 + t3 + × 1+ + t 5 + t6 (20)
MRR T
3. Acquisition of experimental data and modelling
3.1. Experimental details
The experiments were performed on a CNC lathe CK0628, which was equipped with a FANUC-
0i-TD CNC system. The spindle speed of the lathe ranged from 100 to 2000 rpm, and the minimum
Table 1. Energy consumption during the transition state, ECt , in various test cases.
Test case v (m/min) f (mm/r) ap (mm) MRR (cm3 /s) ECt (kJ) Poperational (kW) SEC
1 62.83 0.2 0.5 0.1047 0.7535 0.3723 8.3591

2 62.83 0.25 0.75 0.2618 0.7267 0.3645 4.1662
3 62.83 0.3 1.5 0.3534 0.7398 0.3747 3.4536
4 94.25 0.2 0.75 0.3142 1.1154 0.5409 3.6680
5 94.25 0.25 1.5 0.4418 1.1360 0.5374 3.0032
6 94.25 0.3 0.5 0.2356 1.2089 0.5633 4.6528
7 125.66 0.2 1.5 0.4712 1.7090 0.7016 2.9455
8 125.66 0.25 0.5 0.2618 1.7580 0.7110 4.4144

9 125.66 0.3 0.75 0.6283 1.8460 0.6939 2.4329
feed rate fmin was 0.001 mm/r. The tool insert used in the experiments was a YT15 75 ◦ cylindrical
turning carbide tool. Moreover, a power monitoring platform was developed to acquire real data: a
three-phase power sensor Weibo WB9128, a programmable logic controller OMRON CJ2M-ETN
with an Ethernet interface, and a personal computer with power monitoring software developed
by KingView 6.55. The workpiece was a 45# carbon steel cylindrical bar. The original diameter,
cutting length and maximum cutting depth were 20, 40 and 1.5 mm, respectively. This operation
was conducted under a dry cutting environment.
The central composite design needs five levels, whereas the orthogonal array method needs
only three levels. Therefore, the latter requires less time and cost than the former. Moreover, the
orthogonal array method is more effective, because it adopts an extensive level interval between
the factors (Kim et al. 2005). The orthogonal array method has commonly been used to investigate
the energy consumption of machining systems (Fratila and Caizar 2011; Li and Kara 2011). In
this experiment, the orthogonal array method was used, and three levels were selected for three
processing parameters (i.e. v, f and ap ). Thus, only nine tests were needed for this experiment
(Table 1).
3.2. Regression modelling
Table 1 shows the values of ECt , Poperational and SEC. The mathematical functions of ECt , Poperational
and SEC were developed using MATLAB Curve Fitting Toolbox. The results indicate that the best
empirical values of coefficients in Equations (5), (6) and (11) had 95% confidence level (Table 2).
The results indicate that the regression models could explain more than 98% of data validation.
The values of Pb , tc and ti (i = 1, 2, 3, 5, 6) were obtained by experiments (Table 3). The
coefficients shown in Equation (15) were obtained from the Machinery Processing Technical
Handbook. (Shaonong 1992) (Table 3). The weight of a YT15 tool insert was 13 g and the nose
radius was 0.8 mm. Narang and Fischer (1993) proved that feed rate and nose radius had the most
influential effects on surface roughness, and the relationship between them can be expressed as
follows:
125 f 2
Ra = (21)
rε
Table 2. Summary of the regression models.
Regression model SSE R square Adjusted R square RMSE
ECt = 4.083 × 10−7 × n2 − 1.938 × 10−4

× n + 0.5255 0.003316 0.9869 0.985 0.02177
Poperational = 3.317 × 10−4 × n + 0.04247 0.01482 0.9909 0.9879 0.04971
SEC = 1.226 + 0.785/MRR 0.006329 0.9783 0.9752 0.03007
1000 W. Lin et al.
Table 3. Coefficients in the turning operation.
Pb (kW) tc (s) t1 (s) t2 (s)
0.4186 12 2.5 0.5

t3 (s) t5 (s) t6 (s)
5.0 2.0 7.0
A α β Rmax (μm)
900 2.13 1 20
where Ra and rε are the surface roughness and nose radius, respectively. Let Rmax denote the
maximum allowable surface roughness (Table 3). Thus, the maximum feed rate, fmax , can be
calculated using Equation (22):

Rmax × rε
fmax = = 0.3578 mm/r (22)
125
3.3. Metal-cutting parameter optimization model
Using the parameters and coefficients listed in Subsections 4.1 and 4.2, the multi-objective
optimization model of the turning operation can be expressed as follows:
⎧
⎪
⎪Eopt = 6.222 × 10−8 × n2 + 3.2432 × 10−4 × n
⎨ −3
Minimize +1.8824 + 2.1252×10
v−1.13 a
+ 32.466
⎪
⎪
vfa
⎩ 2.30644×10−5
topt = 17 + v−1.13 a + vfa 226.194
πn
100 ≤ n ≤ 2000 or 62.83 ≤ v ≤ 125.66, v =
50
0.001 ≤ f ≤ 0.3578
0 < ap ≤ 1.5
4. Multi-objective teaching–learning-based optimization algorithm
The basic teaching–learning-based optimization (TLBO) algorithm consists of two phases: teacher
and learner phases. Waghmare (2013) reported the correct understanding of the TLBO algorithm in
an objective manner and commented on the note of Črepinšek, Liu, and Mernik (2012). Črepinšek,
Liu, and Mernik (2014) claimed in another article that the different success rates reported by
Waghmare (2013) were incorrect. Notably, the better performance of the TLBO algorithm com-
pared to the other evolutionary algorithms has already been established by various researchers,
such as Baykasoğlu, Hamzadayi, and Köse (2014) and Chandra Satapathy and Naik (2014). This
study developed a new MOTLBO algorithm.
4.1. Construction of non-dominated set
In general, the non-dominated set of the current population was constructed after the initialization
or at the beginning of one generation. Any individual in the non-dominated set was not dominated
by any individual in the population. Li and Zheng (2004) proposed a method to construct the
non-dominated set by removing those dominated solutions. The steps in the method are described
as follows. (1) The initial size of the non-dominated set was set to zero. (2) Each individual, Xi ,
in the class was compared to each individual, Xj , in the non-dominated set. If any individual, Xj ,
in the non-dominated set was dominated by Xi , then Xj was abandoned; if Xi was not dominated
by any individual in the non-dominated set, then Xi was considered as a new non-dominated
individual and added to the set. Finally, the crowding distance assignment method (Deb et al.
2000) was used to calculate the crowding distances of the individuals in the non-dominated set.
4.2. Teacher phase
The individuals with the largest crowding distance values were selected as the teachers. In this
phase, the learners improved themselves by learning from their teachers. In general, some learners
may have high learning abilities, while others may have a tough time learning. Let Mk denote
the average level of learners, and Tkj denote the jth teacher at iteration k. Mk can be expressed as
follows:
Mk = [x̄1 , x̄2 , . . . , x̄D ]
where x̄d (d = 1, 2, . . . , D) is the mean of the xi,d , i = 1, 2, · · · , pop_size, and pop_size is the
population size. The difference between Mk and Tkj can be calculated using Equation (23):
Xdiff ,k, j = rk (Tkj − TF, j Mk ), j = 1, 2, . . . , tea_num (23)
where rk is a value randomly generated within the range [0, 1], TFj is the teaching factor mimicking
the teaching abilities of teacher j, and tea_num is the teacher number. Because learners may learn
any proportion taught by teachers (Venkata Rao and Patel 2012), this study used an adaptive
teaching factor between 1 and 2. When TF is 2, this indicates that the learners learn everything
taught by the teachers; the learners learn nothing when TF is 1. The teaching factor can be
re-expressed as in Equations (24) and (25):
Mk
TFj = , j = 1, 2, . . . , tea_num (24)
Tkj
⎧
⎪
⎨2, if TFj > 2
TFj = TFj , if 1 ≤ TFj ≤ 2 (25)
⎪
⎩
1, if TFj < 1
Then, each learner learns from the teachers using Equation (26):
Xnew,i = Xold,i + Xdiff ,k,i , i = 1, 2, . . . , pop_size, j = 1, 2, . . . , tea_num (26)
A selection operator was used to accept Xnew,i or Xold,i : If Xnew,i dominates Xold,i , it will be added to
the population, and Xold,i will be deposed; if Xnew,i is Pareto-equal to Xold,i , a coin will be flipped
to determine which one is to be accepted.
4.3. Learner phase
Besides learning from teachers, the learners can also improve themselves by interacting with their
peers. In this phase, a learner acquires more knowledge if his or her peers have more knowledge.
Each learner, Xi , in the class was updated by the following step:
(1) A learner Xj (j = i) was randomly selected from the class and considered as the peer of Xi .
1002 W. Lin et al.
(2) If Xi dominates Xj , then Xi was updated using Equation (27); if Xj dominates Xi , then Xi was
updated using Equation (28); otherwise, a coin was flipped to randomly select one equation
from Equations (27) and (28) to update Xi .
(3) Then, the selection process was used to determine whether the new or old individual would
be accepted at the next iteration.
Xnew,i = Xi + ri × (Xi − Xj ) (27)

Xnew,i = Xi + ri × (Xj − Xi ) (28)
Figure 2 shows the flowchart of the MOTLBO algorithm.
Figure 2. Flowchart of the multi-objective teaching–learning-based optimization (MOTLBO) algorithm.

5. Experimental results and discussion
The MOTLBO algorithm was compiled in the C++ program on a personal computer using a
Windows 7 operating system, a dual-core Pentium CPU, and 2 GB RAM. In this study, the pop_size
and tea_num were 100 and 3, respectively. The maximum evaluation number, max_eva_num, was
20,000.
5.1. Results of multi-objective teaching–learning-based optimization algorithm

The Pareto-optimal frontier obtained by the MOTLBO algorithm contained 30 Pareto-optimal

solutions (Figure 3). Solutions 1 and 30 are the two extreme solutions in the Pareto-optimal
frontier, and their objectives are shown in Figure 3. Solution 1 has a minimum value of CEopt ,
whereas solution 30 has a minimum value of topt . Moreover, a large difference was observed in
the objective values of the two extreme solutions. The value of CEopt of solution 30 was 10.10%
less than that of solution 1, and the value of topt of solution 30 was 10.17% larger than that
of solution 1. Therefore, a traditional best solution with the minimum objective values in the
proposed optimization problem does not exist.
5.2. Analytic hierarchy process to select the optimal solution
Manufacturers are usually confused about how to find the optimal solution from Pareto-optimal
solutions. Using the inputs of manufacturers’ preferences, the AHP can select an appropriate
solution from those solutions (Saaty 1988). Two criteria were used in the AHP method: an energy-
related criterion, C1 , and a time-related criterion, C2 . The energy-related criterion, C1 , consists
of two subcriteria: carbon emission subcriterion, C11 , and direct energy consumption subcrite-
rion, C12 . The time-related criterion, C2 , includes operation time subcriterion, C21 , and tool life
subcriterion, C22 . Saaty’s scale was used to define the intensities of importance between two cri-
teria or two subcriteria (Saaty 1988). Table 4 shows the intensities of importance and priorities.
Thirty Pareto-optimal solutions shown in Figure 3 are considered as the alternative solutions,
and their priorities can be calculated using Equation (29). The researchers find that solution 7
(v = 84.24 m/min, f = 0.357 mm/rev, ap = 1.5 mm) has the largest priority, indicating that this
Figure 3. Pareto-optimal frontier.

1004 W. Lin et al.
Table 4. Intensity of importance between

criteria or subcriteria.
C1 C2 Priority
C1 1 2 0.667
C2 1/2 1 0.333
C11 C12
C11 1 3 0.750
C12 1/3 1 0.250
C21 C22
C21 1 5 0.833
C22 1/5 1 0.167
solution is the optimal solution.
pi = 0.667 × (0.75Ĉ11
i
+ 0.25Ĉ12
i
) + 0.332 × (0.833Ĉ21
i
+ 0.167Ĉ22
i
),
(29)
i = 1, 2, . . . , n
where Ĉjki is a normalized subcriterion Cjk of solution i, and n is the number of alternative solutions.
Because both CEopt and topt are ‘the smaller the better’, Equation (30) was used to normalize them:
max(Cjki ) − Cjki
Ĉjki = , i = 1, 2, . . . , n; j = 1, 2, . . . , m (30)
max(Cjki ) − min(Cjki )
where m is the number of criteria or subcriteria.
5.3. Multi-objective optimization based on design of experiments method
Yan and Li (2013) proposed a new framework for multi-objective optimization in the cutting
parameter optimization field. The method was based on the response surface method (RSM) and
weighted GRA, which determine the weights of the dependent responses according to the degree
of influence of cutting parameters’ variations on the responses. Therefore, there is no need to
provide the preferences for the responses. The steps adapted in Yan and Li’s (2013) method can
be shown as follows:
(1) The experimental results of both dependent responses, i.e. CEopt and topt , were normalized.
(2) The grey relational coefficient (GRC) was calculated using Equation (31):
min( 0j (k)) + ξ max( 0j (k))
GRCk = γ (C0 (k), Ĉij (k)) = (31)
0j (k) + ζ max( 0j (k))

0j (k) = C0 (k) − Cij (k) , C0 (k) = 1, k = 1, 2, . . . , m (32)
where C0 (k) is the reference sequence, m is the number of dependent responses, 0j (k) is the
deviation sequence of C0 (k) and comparability sequence Cij (k), and ξ is the distinguishing
coefficient, with a value of 0.5 normally.
(3) The grey relational grade (GRG) is the weighted sum of the GRCs, as shown in Equation (33):

m
GRG = γ (C0 , Ĉij ) = wk GRCk (33)
k=1

m
wk = 1 (34)
k=1
where wk is the weight of GRCk . The sum of wk is 1, which is supported by Equation (34). If
the response is not sensitive to the change in the cutting parameters, then the weight will be
small. Otherwise, the weight will be large. The weights can be calculated using Equations (35)
and (36):
Rki = max{Kk,i,1 , Kk,i,2 , . . . , Kk,i,l } − min{Kk,i,1 , Kk,i,2 , . . . , Kk,i,l } (35)

D
Rki
wk = m i=1 D , k = 1, 2, . . . , m (36)
k=1 i=1 Rki
where l is the number of experimental levels, R is the GRC range, D is the number of
independent variables, and K is the average GRC for each parameter at each level of each
response.
(4) The regression model was constructed using the RSM. The mathematical model of the GRG
is a second-order polynomial regression expressed by Equation (37):

D
D
GRG = β0 + βi xi + βij xi xj + βii xi2 + ε (37)
i=1 i<j i=1
where x1 = (v − 62.83)/62.83, x2 = (f − 0.2)/0.1, x3 = (ap − 0.5)/1. β is the RSM coeffi-

cient of each term, and ε is the residual error.
(5) Finally, the GRG model was solved using the MATLAB Genetic Algorithm Solver.
The regression model (Equation 38) was constructed using MATLAB based on the experimental
data obtained in Section 4.
GRG = 0.3333 + 0.2963x1 + 0.19x2 + 0.5025x3 − 0.0509x1 x2

(38)
+ 0.07x1 x3 + 0.1697x2 x3 − 0.1559x12 − 0.068x22 − 0.1658x32
Equation (38) was used as the fitness function of the MATLAB Genetic Algorithm Solver. The
population size and maximum number of generations were 100 and 200, respectively. The cutting
parameters of the optimal solution were 94.25 m/min, 0.3 mm/r and 1.5 mm.
5.4. Comparative analysis and verification
Normally, the cutting parameters are determined based on experience or the Machinery Processing
Technical Handbook (Shaonong, 1992). The initial cutting parameters in the turning experiment
were 125.66 m/min, 0.2 mm/rev and 1.5 mm, which is test case 7, shown in Table 1. Let A, B
and C denote the turning operations using the initial parameters, those obtained by Yan and Li’s
(2013) DOE method, and those obtained by the proposed algorithm and AHP method.
The values of CEopt and topt are listed in Figure 4. When compared to the CEopt of operations B
and A, operation C decreased the carbon emissions by 4.52% and 18.36%, respectively. Moreover,
operation C has the lowest value of topt , followed by operations B and A. Therefore, operation C is
much more production efficient and environmentally friendly than operations B and A. The carbon
emissions of the entire turning operation are mainly caused by electricity, as the percentages of
CEelec in CEopt were greater than 90% (Figure 4). It was observed that the carbon emissions were
mainly produced during the ready and cutting states in the lathe. The results of Yan and Li’s
(2013) DOE method are highly related to those selected factor levels and usually trapped in the
local optimal points. However, the experiments and proposed algorithm could always obtain the
global optimal solutions.
1006 W. Lin et al.
Figure 4. Carbon emission constituents of various turning operations.
6. Conclusions
This study proposed a method to quantify the carbon emissions during an entire turning operation
under a dry cutting environment. This carbon emission model was established based on the gen-
eralized boundary graph of a machining system. A MOTLBO algorithm was proposed to achieve
the turning operation optimization. A trade-off was shown to be involved in the optimization when
considering carbon emissions and operation time simultaneously. In this problem, the AHP was
used to find the optimal solution among the Pareto-optimal solutions. Finally, the performances
of the optimal solution were compared to those of the initial cutting parameters and Yan and Li’s
(2013) DOE method. The optimal solution was found to be the most production efficient and
environmentally friendly solution.
There are some interesting directions for future research. First, this study could be extended
to other manufacturing activities such as the cutting fluids, minimum quality liquid and material
consumption. Secondly, more constraints could be considered, such as the cutting force, tool
interface temperature and radial rake angle. Finally, the experiments and proposed algorithm
could be used to optimize other machining operations.
Funding
This work was supported by the State Key Program of National Natural Science of China [grant no. 51035001]; and
National Natural Science Foundation of China [grant no. 51275190].
References
Balogun, Vincent Aizebeoje, and Paul Tarisai Mativenga. 2013. “Modelling of Direct Energy Requirements in Mechan-
ical Machining Processes.” Journal of Cleaner Production 41 (0): 179–186. doi:http://dx.doi.org/10.1016/j.jclepro.
2012.10.015
Baykasoğlu, Adil, Alper Hamzadayi, and Simge Yelkenci Köse. 2014. “Testing the Performance of Teaching–Learning
Based Optimization (TLBO) Algorithm on Combinatorial Problems: Flow Shop and Job Shop Scheduling Cases.”
Information Sciences 276 (0): 204–218. doi:http://dx.doi.org/10.1016/j.ins.2014.02.056
Benardos, P. G., and G.-C. Vosniakos. 2003. “Predicting Surface Roughness in Machining: A Review.” International
Journal of Machine Tools and Manufacture 43 (8): 833–844. doi:http://dx.doi.org/10.1016/S0890-6955(03)00059-2
Chandra Satapathy, Suresh, and Anima Naik. 2014. “Modified Teaching–Learning-based Optimization Algorithm for
Global Numerical Optimization—A Comparative Study.” Swarm and Evolutionary Computation 16 (0): 28–37.
doi:http://dx.doi.org/10.1016/j.swevo.2013.12.005
Črepinšek, Matej, Shih-Hsi Liu, and Luka Mernik. 2012. “A Note on Teaching–Learning-Based Optimization Algorithm.”
Information Sciences 212 (0): 79–93. doi:http://dx.doi.org/10.1016/j.ins.2012.05.009
Črepinšek, Matej, Shih-Hsi Liu, and Marjan Mernik. 2014. “Replication and Comparison of Computational Experiments
in Applied Evolutionary Computing: Common Pitfalls and Guidelines to Avoid Them.” Applied Soft Computing 19
(0): 161–170. doi:http://dx.doi.org/10.1016/j.asoc.2014.02.009
Deb, Kalyanmoy, Samir Agrawal, Amrit Pratap, and Tanaka Meyarivan. 2000. “A Fast Elitist Non-Dominated Sorting
Genetic Algorithm for Multi-Objective Optimization: NSGA-II.” Lecture Notes in Computer Science 1917: 849–858.
doi:10.1007/3-540-45356-3_83
EIA. 2012. “Annual Energy Review 2011.” US Energy Information Administration. http://www.eia.gov/totalenergy/data/
annual/archive/038411.pdf
Fratila, Domnita, and Cristian Caizar. 2011. “Application of Taguchi Method to Selection of Optimal Lubrica-
tion and Cutting Conditions in Face Milling of AlMg3 .” Journal of Cleaner Production 19 (6): 640–645.
doi:http://dx.doi.org/10.1016/j.jclepro.2010.12.007
Kim, Sung-Il, Ji-Young Lee,Young-Kyoun Kim, Jung-Pyo Hong,Yoon Hur, andYeon-Hwan Jung. 2005. “Optimization for
Reduction of Torque Ripple in Interior Permanent Magnet Motor by Using the Taguchi Method.” IEEE Transactions
on Magnetics 41 (5): 1796–1799. doi:10.1109/TMAG.2005.846478
Li, W., and S. Kara. 2011. “An Empirical Model for Predicting Energy Consumption of Manufacturing Processes: A
Case of Turning Process.” Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering
Manufacture 225 (9): 1636–1646. doi:10.1177/2041297511398541
Li, Congbo, Ying Tang, Longguo Cui, and Pengyu Li. 2013. “A Quantitative Approach to Analyze Car-
bon Emissions of CNC-Based Machining Systems.” Journal of Intelligent Manufacturing 23 (0): 1–12.
doi:10.1007/s10845-013-0812-4
Li, Lirong, and Jinhua Zheng. 2004. “Multi Objective Genetic Algorithm Based on Pareto Front.” Natural Science Journal
of Xiangtan University 26 (1): 39–41. doi:10.3969/j.issn.1000-5900.2004.01.012
Liu, Fei, Shuang Liu, Jun Xie, and Qiulian Wang. 2012. “Energy Forecasting Method for CNC Machine Tool During its
Working Time.” China Patent CN102621932 filed May 2.
Mori, M., M. Fujishima, Y. Inamasu, and Y. Oda. 2011. “A Study on Energy Efficiency Improvement for Machine Tools.”
CIRP Annals—Manufacturing Technology 60 (1): 145–148. doi:http://dx.doi.org/10.1016/j.cirp.2011.03.099
Narang, R. V., and G. W. Fischer. 1993. “Development of a Framework to Automate Process Planning Functions
and to Determine Machining Parameters.” International Journal of Production Research 31 (8): 1921–1942.
doi:10.1080/00207549308956832
NDRC [National Development and Reform Commission]. 2011. “Baseline Emission Factors for Regional
Power Grids in China.” National Development and Reform Commission. Accessed December, 2013.
http://cdm.ccchina.gov.cn/WebSite/CDM/UpFile/File2720.pdf
Pawar, P. J., and R. Venkata Rao. 2013. “Parameter Optimization of Machining Processes Using Teaching–Learning-
Based Optimization Algorithm.” International Journal of Advanced Manufacturing Technology 67 (5–8): 995–1006.
doi:10.1007/s00170-012-4524-2
Rajemi, M. F., P. T. Mativenga, and A. Aramcharoen. 2010. “Sustainable Machining: Selection of Optimum Turn-
ing Conditions Based on Minimum Energy Considerations.” Journal of Cleaner Production 18 (10): 1059–1065.
doi:http://dx.doi.org/10.1016/j.jclepro.2010.01.025
Saaty, Thomas L. 1988. “What is the Analytic Hierarchy Process?” Chap. 5. In Mathematical Models for Decision Support,
109–121. Berlin: Springer.
Shaonong, Meng. 1992. Machinery Processing Technical Handbook. Beijing: Machinery Industry Press.
Venkata Rao, R., and Vivek Patel. 2012. “Multi-Objective Optimization of Heat Exchangers Using a Modi-
fied Teaching–Learning-Based Optimization Algorithm.” Applied Mathematical Modelling 37 (3): 1147–1162.
doi:http://dx.doi.org/10.1016/j.apm.2012.03.043
Waghmare, Gajanan. 2013. “Comments on ‘A Note on Teaching–Learning-Based Optimization Algorithm’.” Information
Sciences 229 (0): 159–169. doi:http://dx.doi.org/10.1016/j.ins.2012.11.009
Yan, Jihong, and Lin Li. 2013. “Multi-Objective Optimization of Milling Parameters—The Trade-Offs
Between Energy, Production Rate and Cutting Quality.” Journal of Cleaner Production 52 (0): 462–471.
doi:10.1016/j.jclepro.2013.02.030
Yusup, Norfadzlan, Azlan Mohd Zain, and Siti Zaiton Mohd Hashim. 2012. “Evolutionary Techniques in Optimizing
Machining Parameters: Review and Recent Applications (2007–2011).” Expert Systems with Applications 39 (10):
9909–9927. doi:http://dx.doi.org/10.1016/j.eswa.2012.02.109
View publication stats

Multi-Objective Teaching-Learning-Based Optimization Algorithm For Reducing Carbon Emissions and Operation Time in Turning Operations

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Multi-Objective Teaching-Learning-Based Optimization Algorithm For Reducing Carbon Emissions and Operation Time in Turning Operations

Uploaded by

Copyright:

Available Formats

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

Multi-objective teaching–learning-based optimization

Article in Engineering Optimization · December 2014

Wenwen Lin Shaojun Wang

SEE PROFILE SEE PROFILE

Dual-driving system View project

The user has requested enhancement of the downloaded file.

To link to this article: http://dx.doi.org/10.1080/0305215X.2014.928818

PLEASE SCROLL DOWN FOR ARTICLE

Multi-objective teaching–learning-based optimization algorithm

(Received 22 January 2014; accepted 14 May 2014)

Keywords: metal-cutting parameter optimization; sustainable manufacturing; orthogonal arrays; multi-

∗ Corresponding author. Email: zcyhust@mail.hust.edu.cn

© 2014 Taylor & Francis

2. Quantitative measurements of carbon emissions in machining systems

2.1. Carbon emissions of machining systems under dry cutting conditions

CEopt = CEelec + CEtool (1)

2.1.1. Carbon emissions of energy conversion for machining system, CEelec

Figure 1. Power measurement example of a turning operation.

ECmachine (Equation 2):

ECmachine = ECb + ECt + ECr + ECc (3)

ECb = Pb × (t1 + t6 ) (4)

as Equation (5) (Liu et al. 2012):

ECr = Pr × (t3 + t5 ) = (Pb + Poperational ) × (t3 + t5 ) (7)

2.1.2. Carbon emissions caused by cutting tools, CEtool

CEtool = ξ × CEFtool × Wtool (13)

ξ can be calculated using Equations (14) and (15):

2.2. Operation time

3. Acquisition of experimental data and modelling

3.1. Experimental details

1 62.83 0.2 0.5 0.1047 0.7535 0.3723 8.3591

8 125.66 0.25 0.5 0.2618 1.7580 0.7110 4.4144

3.2. Regression modelling

Table 2. Summary of the regression models.

Regression model SSE R square Adjusted R square RMSE

ECt = 4.083 × 10−7 × n2 − 1.938 × 10−4

Table 3. Coefficients in the turning operation.

Pb (kW) tc (s) t1 (s) t2 (s)

0.4186 12 2.5 0.5

3.3. Metal-cutting parameter optimization model

4. Multi-objective teaching–learning-based optimization algorithm

4.1. Construction of non-dominated set

4.2. Teacher phase

Xdiff ,k, j = rk (Tkj − TF, j Mk ), j = 1, 2, . . . , tea_num (23)

Xnew,i = Xold,i + Xdiff ,k,i , i = 1, 2, . . . , pop_size, j = 1, 2, . . . , tea_num (26)

4.3. Learner phase

Xnew,i = Xi + ri × (Xi − Xj ) (27)

Figure 2 shows the flowchart of the MOTLBO algorithm.

Figure 2. Flowchart of the multi-objective teaching–learning-based optimization (MOTLBO) algorithm.

5. Experimental results and discussion

5.1. Results of multi-objective teaching–learning-based optimization algorithm

The Pareto-optimal frontier obtained by the MOTLBO algorithm contained 30 Pareto-optimal

5.2. Analytic hierarchy process to select the optimal solution

Figure 3. Pareto-optimal frontier.

Table 4. Intensity of importance between

C22 1/5 1 0.167

solution is the optimal solution.

5.3. Multi-objective optimization based on design of experiments method

Rki = max{Kk,i,1 , Kk,i,2 , . . . , Kk,i,l } − min{Kk,i,1 , Kk,i,2 , . . . , Kk,i,l } (35)

where x1 = (v − 62.83)/62.83, x2 = (f − 0.2)/0.1, x3 = (ap − 0.5)/1. β is the RSM coeffi-

GRG = 0.3333 + 0.2963x1 + 0.19x2 + 0.5025x3 − 0.0509x1 x2

5.4. Comparative analysis and verification

Figure 4. Carbon emission constituents of various turning operations.