mathematics

Article
Stochastic Chebyshev Goal Programming Mixed Integer Linear
Model for Sustainable Global Production Planning
Chia-Nan Wang 1, * , Nhat-Luong Nhieu 1, * and Trang Thi Thu Tran 2

1 Department of Industrial Engineering and Management, National Kaohsiung University of Science and
Technology, Kaohsiung 80778, Taiwan
2 Department of Industrial Engineering and Engineering Management, National Tsing Hua University,
Hsinchu 300044, Taiwan; [email protected]
* Correspondence: [email protected] (C.-N.W.); [email protected] (N.-L.N.)

Abstract: Production planning is a necessary process that directly affects the efficiency of production
systems in most industries. The complexity of the current production planning problem depends on
increased options in production, uncertainties in demand and production resources. In this study,
a stochastic multi-objective mixed-integer optimization model is developed to ensure production
efficiency in uncertainty conditions and satisfy the requirements of sustainable development. The
efficiency of the production system is ensured through objective functions that optimize backorder
quantity, machine uptime and customer satisfaction. The other three objective functions of the
proposed model are related to optimization of profits, emissions, and employment changing. The



objective functions respectively represent the three elements of sustainable development: economy,

 environment, and sociality. The proposed model also assures the production manager’s discretion
Citation: Wang, C.-N.; Nhieu, N.-L.; over whether or not to adopt production options such as backorder, overtime, and employment of
Tran, T.T.T. Stochastic Chebyshev temporary workers. At the same time, the resource limits of the above options can also be adjusted
Goal Programming Mixed Integer according to the situation of each production facility via the model’s parameters. The solutions that
Linear Model for Sustainable Global compromise the above objective functions are determined with the Chebyshev goal programming
Production Planning. Mathematics approach together with the weights of the goals. The model is applied to the multinational production
2021, 9, 483. https://doi.org/ system of a Southeast Asian supplier in the textile industry. The goal programming solution of the
10.3390/math9050483
model shows an improvement in many aspects compared to this supplier’s manufacturing practices
under the same production conditions. Last but not least, the study develops different scenarios
Academic Editors:
based on different random distributions of uncertainty demand and different weights between the
María-Isabel Ayuda and
objective functions. The analysis and evaluation of these scenarios provide a reference basis for
María Dolores Gadea
managers to adjust the production system in different situations. Analysis of uncertain demand
Received: 5 February 2021 with more complex random distributions as well as making predictions about the effectiveness of
Accepted: 22 February 2021 scenarios through the advantages of machine learning can be considered in future studies.
Published: 26 February 2021
Keywords: global production planning; multiple objective; sustainable development; Chebyshev
Publisher’s Note: MDPI stays neutral goal programming; stochastic linear programming; textile industry
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
1. Introduction
The textile industry plays a vital role in many Asian countries’ economies especially
in emerging nations. It is a diverse sector, which has its own chain that includes converting
Copyright: © 2021 by the authors. chemical or natural fibers into consumer goods such as garments or industrial textiles. It
Licensee MDPI, Basel, Switzerland. will require force to transfer from the existing Industry 3.0 to Industry 4.0 and empower
This article is an open access article smart production. Sustainable fabrication is significant for the effective use of waste
distributed under the terms and reduction, resources allocation efficiency and related costs. Textiles are severely interlinked
conditions of the Creative Commons with environmental, governance and social issues. In the past, efforts of the manufacturer
Attribution (CC BY) license (https://
have been primarily concentrated on enhancing the social attributes of textiles, such as
creativecommons.org/licenses/by/
creating suitable working conditions, setting minimum wages, maintaining work safety,
4.0/).

Mathematics 2021, 9, 483. https://doi.org/10.3390/math9050483 https://www.mdpi.com/journal/mathematics

Mathematics 2021, 9, 483 2 of 22

etc. However, over the years, concern about the textile’s industry’s environmental impacts
have been risen especially regarding greenhouse gas (GHG) emissions which may arise
from washing (using water heating and detergents), processing fossil fuels into synthetic
fibers and dyeing textiles. This industry is one of the major sources of GHG emissions
because of its fabrication technologies and transportation [1].
Presently, it is not only faced with a global aggressive manufacturing environment,
but also incessant rising demand. Therefore, the textile industry needs to optimize its
manufacturing operations to enhance productivity, customization, and product quality,
and minimize lead times [2]. Indeed, these factors have led manufacturers to apply plan-
ning strategies that make manufacturing procedure more effective, and encouraging, and
improve production structures [3]. These have driven textile and clothing supply chains
to optimize their production planning to satisfy the challenges demanded by the market.
Global production planning ascertains the production capacity to meet the de-mand, in
which design configurations and production scenarios are evaluated that require a combina-
tion of factors, which are the maximization of profit, machine operation time, and customer
satisfaction level, and the minimization of costs, emissions, and workforce changing at once.
As a theoretical contribution of this study, the proposed mixed-integer programming
model integrates all typical production options such as regular production, production
using temporary workers, overtime and backorder. Moreover, the production-planning
problem that this model deals with has characteristics of globalization, which is reflected
in a multi-product, multi-period, multi-country production system and an uncertain pro-
duction environment. Another novelty of the model is the set of objective functions that
ensures that the production system satisfies the requirements of the new production era
to be efficient and sustainable. The objective functions towards sustainability include
optimization of profits, emissions and changes in employment. Meanwhile, optimization
of backorders, machine use, and customer satisfaction are the target functions that ensure
efficiency. The compromise solution of the above objective functions is achieved through
the Chebyshev variant of goal programming. This variation is believed to be able to avoid
extreme solutions that can be encountered with lexicographic goal programming. Finally,
as a practical contribution, a set of scenarios is developed regarding two factors: uncertainty
of demand and the weight of objective functions. Evaluation of these scenarios provides
production managers with predictions for the refinement of production systems when the
two factors mentioned above change.
The paper is organized as follows. Section 2 reviews the literature and related studies.
The mathematical multi-objective mixed-integer programming model is presented in
Section 3. Section 4 proposes a result from an empirical study for validation with its
optimization approaches. Section 5 concludes the paper, and further work of this study.

2. Literature Review
The production-planning process is seen as the process of finding a way that economi-
cally or efficiently uses resources required for production such as manpower, materials, and
equipment. In other words, the typical decisions in production planning are affected by
trade-offs between financial and productivity efficiency objectives. In particular, financial
targets are mainly related to increasing profits or reducing costs, including production costs,
labor costs, material costs, inventory costs, and so on. In terms of production efficiency,
production planning should consider the ability to supply products as well as the effects
of other factors during operation such as overtime, inventory level, and backorders. For
long-medium, or short-term periods, the production plan is classified into a strategic plan,
a tactical plan, and an operational plan, respectively. The objective of strategic plans is to
sustain growth and create a competitive advantage. Meanwhile, tactical and operational
plans make decisions about product flow, manpower, and machine capacity with more
detailed information to create economic and productivity efficiency [4].
Quantitative models and methods are supposed to assist operations managers with
grasping the complexity of the production system and the decisions associated with it [5].
Mathematics 2021, 9, 483 3 of 22

Production plans can be developed with forecasted demand using DEA-based approaches,
which were proposed by Du et al. [6]. In that study, the production system performance
is measured by the CCR efficiency score, which was introduced by Charnes, Cooper and
Rhodes [7]. Another heuristic-based approach is recommended to determine the effec-
tive inventory level to save inventory-related costs, such as holding cost and backorder
penalty [8]. The results of the synthesis paper published in 2019 by Bagshaw stated that
different quantitative models are developed for different production conditions and prob-
lem characteristics [9]. In other words, a universal model or method for all operational
production problems does not exist. This study also demonstrates that linear program-
ming planning being an effective quantitative analysis approach for the improvement of
production systems [9,10]. An optimal control formulation, introduced by Devizón et.
al., integrates two dynamical systems with typical inventory process parameters such as
production rate, inventory level, and costs [11]. Another linear planning model developed
to minimize costs in the mining industry is suggested by Ozsan et al. The solution of
this model is under different production conditions due to limitations on resources [12].
In research published in 2018, Campo developed an optimized mathematical model for
the whole production planning problem with the objective function and constraints that
represent the nature of the garment industry. In which, the model includes variables that
affect specific production systems of the garment industry such as physical variations
of fabrics, skills of new workers, and so on [3]. For the furniture industry, Gremani’s
mathematical model solves the problem of the trade-off between inventory and production
of semi-finished or finished products. The results of the study show that while inventory of
semi-finished or chopped materials is more cost-effective, it is advantageous to determine
the quantity of finished product inventory [13]. The special requirements for their highly
skilled workforce and long-time flexibility are characteristic elements of the automotive
industry, according to Sillekens et al. Therefore, the workforce planning has been integrated
into the mathematical model that incorporates a heuristic algorithm to make useful deci-
sions for automotive production planning [14]. Another application of linear programming
model in production planning is in the food industry. The mathematical model of Munhoz
and Morabito is concerned with the acidity parameters of the product that are typical of
the fruit juice production process [15].
The review paper by Ali Cheraghalikhani, Farid Khoshalhan, and Hadi Mokhtari
has categorized production planning models into 6 groups, as showen in Table 1 [16].
Accordingly, the group 1 and group 2 models solve the problem of production planning
under certain conditions. Meanwhile, the studies of the remaining groups have considered
the uncertainties when developing models or optimization algorithms. While group 3
and group 4 approach uncertainty solutions using fuzzy theory, the studies in group 5
and group 6 apply the theories and probability distributions. This study also pointed out
important issues in the production planning problem where the number of related studies
decreased over time. Machine utilization, labor characteristics, and multiple facilities are
the issues on the list. In the conclusion of this review, the authors propose potential research
directions for the production planning problem. Among them, there are multi-objective
stochastic models with more than three important issues considered.

Table 1. Production planning model classification.

Single Objective Multiple Objective

Deterministic models Group 1 Group 2
Fuzzy models Group 3 Group 4
Uncertainty models
Stochastic models Group 5 Group 6

According to research by Mula et al., uncertainty factors in the production system

are listed as demand, environment, system resource, lead time, and yield. The study also
synthesizes common approaches to uncertainty including stochastics model, dynamic
programming, fuzzy theory, and simulation-based approaches [17].
Mathematics 2021, 9, 483 4 of 22

Uncertainty in production plans often appears as demand uncertainty or fluctuations

in the production process, such as production times or material loss. Bakir and Byrne
analyzed the difference of the solutions provided by deterministic model and stochastic
model, in which the uncertainty factor is market demand. The analytical results show that
this difference is dependent on the variance of uncertain demand [18].
The uncertainty of parameters in the linear planning model is also approached in terms
of fuzzy numbers. Accordingly, some of the parameters describing the production system
limit are recognized by the authors as uncertain and described as fuzzy parameters [19–21].
In addition, parameters intended to determine the value of the target functions such
as purchasing materials cost or production cost, can also be uncertain. Therefore, the
linear programming model in Vasant’s study in 2006 also introduced a fuzzy objective
function [22]. In a recently published study, the authors used fuzzy numbers to describe the
uncertainty of unit production time, inventory time, and packing time in the automotive
industry. The results of the linear planning model are assessed to have improved the
production system in terms of shortening inventory times as well as increasing economic
efficiency [23]. Zimerman’s method has been applied in the fuzzy mathematical linear
programming model, which minimizes all kinds of costs by Kalaf et al. [24]. Baykasoglu
and Gocken converted the fuzzy numbers to a non-linear crisp equivalent before using a
Tabu Search algorithm to find an optimized solution [25]. Meanwhile, another choice is a
genetic algorithm for solving the crisp model. Moreover, this study also uses a combination
of fuzzy AHP and Topis to determine the solution that is reconciled [26,27].
In addition to the fuzzy number approaches, production planning uncertainty is
also examined through stochastic models with random distributions and the theory of
probability. Orcun presented an integrated non-linear programming model with uncertain
demand. The author then used a heuristics approach to evaluate the trade-off between
the production system’s time and safety stock level [28]. A hybrid programming model is
developed through a combination of stochastic programming and linear programming for
the refinery’s production plan, in which the weighting factor is used to control the uncer-
tainty of the hybrid model [29]. Kazemi Zanjani used a scenario tree to control demand
uncertainty in a multi-product production planning study. The scenarios in this stochastic
model are defined based on the probability distribution of demand during production
planning [30]. With similar approaches, Zhang created scenarios for demand uncertainty
based on optimistic, pessimistic, and neutral forecast values and their probabilities [31].
For the probabilities, Kazemi Zanjani uses the mean values of the uncertainty parameters
as a method for processing the proposed stochastic model [32].
As mentioned above, production plans not only target financial and economic benefits
but also consider objectives such as employment, production time, etc. Consequently, the
presence of studies that place more than one objective function in a linear programming
model is growing. Komsiyah’s research has proposed a multi-objective model with the fuzzy
goal programming approach in the furniture industry. However, the objective functions
of this model revolve around different cost categories that can be combined into a total
cost objective function [33]. Meanwhile, in addition to costs, Tirkolaee’s study has pointed
out another objective function, which is to maximize customer satisfaction. In that study,
the authors applied weighted goal programming to solve the problem of multi-objective
production planning with assumptions that did not consider overtime and outsourcing [34].
A study applied in Hong Kong has developed an optimization model with three target
functions: Maximizing profit, minimizing costs due to defective products, and maximizing
equipment utilization at a certain production facility [35]. A similar study has been devel-
oped for multinational production systems in which ensuring stability in employment and
optimizing the efficiency of the import/export quota becomes the objective functions of the
mathematical model [36].
In recent years, the impact of climate change on social activities has become more
and more apparent. In particular, industry and manufacturing are often considered the
source that generates more emissions compared to other sectors. Therefore, environmental
Mathematics 2021, 9, 483 5 of 22

factors also appear in the optimization models to help managers make sustainable pro-
duction decisions. These environmental impacts in the manufacturing sector are often
considered by researchers in terms of production energy consumption. In 2018, Hahn
and Brandenburg proposed an optimization model in which the environmental impact
factors are converted into a form of costs [37]. Taking a similar approach, Tsai’s model
describes the environmental impacts such as carbon taxes and the cost savings from reused
energy [38]. The optimization model of Modarres and Izadpanahi published in 2016 has
the relationship between the uncertainty level of the parameters and the value of the cost
of electrical energy. Goal programming is again chosen as the approach to solve the multi-
objective model of production planning [39]. In order to increase the efficiency of planning
processes, optimization models and algorithms are increasingly integrated both quanti-
tatively and extensively with project management tools [40]. Research by Valenko and
Klanšek presented an integrated approach to optimization and Microsoft Project software
that increased the cost-effectiveness of the planning process [41].
In summary, the effectiveness of decisions in the production planning process is
supported by the results of numerous studies with different approaches. The approach
that uses optimization mathematical models has been showing useful applications in the
global production planning problems of many industries. Regarding production options, a
few studies consider overtime, backorder or part-time worker production in addition to
regular production. For the model’s objective functions, most studies focus on economic
factors such as profits, costs, and penalties. Meanwhile, studies in recent years consider
the environmental impact of production systems as an objective function. In addition,
indicators showing the system performance such as maximizing utilization of equipment
and customer satisfaction are also considered in the models. Moreover, the uncertainty of
the parameters in the model has also shown to have a significant effect on the optimization
results by researchers. As the results summarize in Table A1, this study develops a model
that integrates the above factors simultaneously.

3. Methodology
3.1. Sets, Parameters and Decision Variables
In this optimization models, parameters and decision variables are formulated based
on the set of product families, manufacturing facilities, the time period and the objective
function as Table 2.

Table 2. Sets.

Set Indices Description

I i The set of the product families
J j The set of the factories
T t The set of the planning periods
K K The set of objective functions

As shown in Table 3, the model parameters can be classified as five groups. The
first group is the parameters related to the selling price of the product along with the
costs in production, labor, inventory, backorder penalties, recruitment and layoffs. The
second parameter group describes the upper and lower boundaries of inventory, backorder,
workforce, and machine capacity. The third group includes parameters related to demand,
the standard deviation of demand and its probability under uncertainty. The parameters
that provide information about the processing time of product families by the machine and
the operator are classified in the fourth group. The fifth group includes the parameters of
total worker time, overtime limits and allowable rates of workforce change in production
facilities. In addition, such parameters as batch size, production specific electricity use of
product families; emission factor of manufacturing facilities; goal value, and weight of
objective functions fall into this group.
Mathematics 2021, 9, 483 6 of 22

Table 3. Parameters.

Group Parameter Description Unit

Pi The unit sell price of product i K VND/unit
PCijF The unit production cost by full-time workers K VND/unit
PCijT The unit production cost by temporary workers K VND/unit
OTCij The unit production cost by full-time workers at overtime K VND/unit
LCjF The labor cost of fulltime worker K VND/man-period
1
LCjT The labor cost of temporary worker K VND/man-period
LCOT
j
The labor cost of fulltime worker at overtime K VND/man-period
ICij The unit inventory cost to hold a product at the end of each period K VND/unit
BOCij The unit backorder cost for a product at the end of each period K VND/unit
HCj The hiring cost for one full-time worker K VND/man
FCj The firing cost for one full-time worker K VND/man
IL jt Maximum inventory capacity in factory j at period t Units
BOLit Maximum backorder in of product i at period t Units
Wjmin Minimum workforce level of full-time worker available in factory j in each period Man
2
Wjmax Maximum work-force level of full-time worker available in factory j in each period Man
TPL jt The temporary worker time limit in factory j at period t Hour
ML jt The machine time capacity in factory j at period t Machine-hour
Ditmin Minimum known demand of product i at period t Units
Ditmost Most-likely demand of product i at period t Units
Ditmax Maximum forecasted demand of product i at period t Units
σimin Standard deviation of minimum known demand standard deviation of product i Units
3 σimost Standard deviation of most-likely demand of product i Units
σimax Standard deviation of maximum forecasted demand of product i Units
Pr max The probability of forecasted maximum demand %
Pr most The probability of most-likely demand %
Pr min The probability of minimum known demand %
PTiF The processing time for product i by full-time workers Hour/unit
PTiT The processing time for product i by temporary workers Hour/unit
4
MTiF The machine time for product i operated by full-time workers Machine-hour/unit
MTiT The machine time for product i operated by temporary workers Machine-hour/unit
δj The working hour of full-time workers in factory j in each period Hour/man-period
W LCj The fraction of workforce allowable for variation in each period %
αj The fraction of overtime hours in factory j used in each period %
BZi The production batch size of product i Units/batch
EFj CO2 Emission factor of factory j tCO2/mwh
5
SE Specific electricity use for production mwh/unit
The inverse distribution function of a standard normal distribution with
z
cumulative probability
β The proportion of demand met from stock (service level Type II) %
Gk The aspiration level of objective function k
WEk The weight of objective function k

The decision variables of the model assist managers in the optimal quantity of products
that can be sold, Sit , based on information about demand under uncertain conditions. In
Table 4, the above production quantities supplied by sources including regular production,
temporary worker production, overtime production, inventory, and backorder are denoted
xijt , yijt , vijt , Iijt , Bijt respectively. In addition, the model also makes decisions about
the workforce level (Wjt ) as well as the amount of recruitment (Hjt ) and layoffs (Fjt ) in
manufacturing facilities in each period. Furthermore, the total overtime (OTjt ), temporary
labor use time (TPTjt ), and machine uptime (MPTjt ) are also calculated by the model.
Finally, deviation variables (ok and uk ) are used to determine the difference between the
goal value and the achieved value of the objective functions obtained by the solution.
Mathematics 2021, 9, 483 7 of 22

The maximal deviation, ω, is used for Chebyshev Goal Programming to ensure balance
between objective functions.

Table 4. Decision variables.

Decision Variable Type Description Unit

Sit Integer The quantity of product i sold at period t Units
The quantity of product i manufactured from factory j by full-time
xijt Integer Batch
worker at regular time at period t
The quantity of product i manufactured from factory j by temporary
yijt Integer Batch
worker in regular time at period t
The quantity of product i manufactured from factory j by full-time
vijt Integer Batch
worker in overtime at period t
Wjt Integer The number of fulltime workers required in factory j at period t Man
Hjt Integer The number of fulltime workers hired in factory j at period t Man
Fjt Integer The number of fulltime workers laid-off in factory j at period t Man
OTjt Float The overtime of fulltime workers in factory j at period t Hour
TPTjt Float The labor time of temporary workers in factory j at period t Hour
MPTjt Float The machine using time in factory j at period t Hour
Iijt Integer The inventory of product i in factory j at the end of period t Batch
Bijt Integer The backorder of product i in factory j at the end of period t Batch
(
1 i f f actory j hiring worker at period t
θ jt Binary =
0 otherwise
(
1 i f product i produced f rom f actory j at period t
ϕijt Binary =
0 otherwise
ok Integer/float The deviation variable of overachievement of the goal Gk
uk float The deviation variable of underachievement of the goal Gk
ω Float The maximal deviation from amongst the goals

3.2. Model Formulation

3.2.1. Objective Functions
In this paper, the authors aim for six objective functions that reflect the efficiency of the
production system in both economic aspects and factors of sustainable development such
as environment and society. First, this model maximizes the profitability of the production
system (VND). In Equation (1), the first term represents the estimated total revenue while
the rest shows the costs. Those costs are production costs, labor costs, inventory costs,
backorder penalties, and cost of hiring/firing workers are presented in the order. The
second object function of this model is to minimize emissions (tCO2) that are estimated
based on parameters such as emission factor and specific electricity use for production as
Equation (2).
!
Maximize ∑ ∑ Pi .Sit − ∑ ∑ ∑(PCijF .xijt + PCijT .yijt + OTCij .vijt ).BZi
i t i j t
!
− ∑ ∑ LCFj .Wjt + LCTj .TPTjt + LCOT
j .OTjt
j t (1)
− ∑ ∑ ∑ ICij .Iijt .BZi − ∑ ∑ ∑ BOCij .Bijt .BZi
i j t i !j t

− ∑ ∑ HC j .Hjt + FCj .Fjt

j t

∑∑∑


Minimize xijt + yijt + vijt .BZi .EFj .SE) (2)
i j t
Mathematics 2021, 9, 483 8 of 22

Besides environmental impacts, social influences are also a part of ensuring sustainable
development. In Equation (3), this study aims to minimize changes in the workforce level
(man). This objective function both ensures production stability through ensuring the
skill of workers, while also minimizing employment disturbances in the area where the
manufacturing facilities are located.

Minimize ∑ ∑ Hjt + Fjt

(3)
j t

Backorder is one of the production planning options that this model considers. Al-
though this option can bring some advantages to the production system such as saving
inventory space, a large backorder could lead to the risk of customers turning to alternative
suppliers. Therefore, as Equation (4), the next objective function is to minimize the amount
of system production backorder (unit).

Minimize ∑ ∑ ∑ Bijt .BZi (4)

i j t

In order to improve the efficiency of investment by increasing the utilization rate

of production machines, the model maximizes the total operating time of production
machinery (hour) in all manufacturing facilities through Equation (5).

Maximize ∑ ∑ MPTjt (5)

j t

In Equation (6), the model’s sixth objective function is to maximize customer satis-
faction level (%), which is determined by the ratio between the amount of product the
system can supply and the expected value of demand. This higher ratio can improve the
popularity of the product in the market, but it can also increase costs, leading to a decrease
in net profit.  
Sit
∑t Pr min Ditmin + Pr most Ditmost + Pr max Ditmax
Mazimize ∑ T.I
(6)
i

3.2.2. Constraints
The constraints presented in this section describe the resource limitation and policy
of the production system. In Equations (7) and (8), the quantity of products expected to
be sold does not exceed the expected value of demand and is above the minimum known
demand. The expected values of demand are estimated through known minimum, most-
likely, and forecasted maximum demand, combined with their probabilities. This expected
sales quantity is balanced with regular production, overtime production, production by
temporary workers, inventory, and backorder, as shown in Equation (9). The binary
decision variable ϕijt in Equation (10) is used to determine the minimum inventory level
in Equation (11). The right-hand side of Equation (11) describes the safety stock of a
product if it is regularly produced at a given facility over a certain period of time with
service level type II (β). In which, z β is the inverse distribution function of a standard
normal distribution with cumulative probability β [42]. Equations (12) and (13) ensure that
inventory level and backorder do not exceed the maximum allowed limit.

Sit ≤ Pr min Ditmin + Pr most Ditmost + Pr max Ditmax i ∈ I, t ∈ T (7)

Sit ≥ Ditmin i ∈ I, t ∈ T (8)

∑


Iijt−1 − Bijt−1 + xijt + yijt + vijt − Iijt + Bijt .BZi = Sit i ∈ I, t ∈ T (9)
j
Mathematics 2021, 9, 483 9 of 22


xijt ≤ ϕijt .BigM
i ∈ I, j ∈ J, t ∈ T (10)
xijt ≥ ϕijt
 
Iijt .BZi ≥ ϕijt .z β . Pr min σimin + Pr most σimost + Pr max σimax i ∈ I, j ∈ J, t ∈ T (11)

∑ Iijt .BZi ≤ IL jt j ∈ J, t ∈ T (12)

i

∑ Bijt .BZi ≤ BOLit i ∈ I, t ∈ T (13)

j

Equation (14) defines change in the workforce by recruiting new workers and layoffs.
In particular, the number of workers at a time cannot be outside the upper and lower
borders according to the policies of each manufacturing facility as Equation (15). Moreover,
Equations (16) and (17) ensure that the number of newly recruited or fired workers cannot
exceed a permitted percentage compared to the workforce level in the preceding period.
Equation (18) is a non-linear variation of the equation Hjt .Fjt = 0, which ensures that
manufacturing facilities cannot hire and fire workers in the same period.

Wjt = Wjt−1 + Hjt − Fjt j ∈ J, t ∈ T (14)

Wjmin ≤ Wjt ≤ Wjmax j ∈ J, t ∈ T (15)

Hjt ≤ WLCj .Wjt−1 j ∈ J, t ∈ T (16)
Fjt ≤ WLCj .Wjt−1 j ∈ J, t ∈ T (17)

Hjt ≤ θ jt .BigM

j ∈ J, t ∈ T (18)
Fjt ≤ 1 − θ jt .BigM
The constraints (19) ensure that the total time required for regular production does not
exceed the total time of the workforce. Similarly, constraints (20)–(23) represent the limits
of the production time by temporary workers and overtime. In Equations (24) and (25), the
total time required for the aforementioned production types is also limited by the design
capacity of the machinery system at the production facility. Finally, all decision variables in
this model are considered as non-negative variables.

∑ PTFi .xijt .BZi ≤ δj .Wjt j ∈ J, t ∈ T (19)

i

∑ PTTi .yijt .BZi ≤ TPTjt j ∈ J, t ∈ T (20)

i

∑ PTFi .vijt .BZi ≤ OTjt j ∈ J, t ∈ T (21)

i

TPTjt ≤ TPL jt j ∈ J, t ∈ T (22)

OTjt ≤ δj .α j .Wjt j ∈ J, t ∈ T (23)

∑ MTFi xijt + ∑ MTiT yijt + ∑ MTFi zijt ≤ MPTjt j ∈ J, t ∈ T (24)

i i i

MPTjt ≤ ML jt j ∈ J, t ∈ T (25)

3.3. Chebyshev Goal Programming (CGP) Application

In order to find compromise solutions for all six objective functions outlined above,
this study uses the Chebyshev goal programming approach, also known as Minmax
Goal Programming. The difference of this variant compared to other variants of goal
programming such as weighted goal programming or lexicographic goal programming is
to assist in finding balanced solutions between the objective functions instead of extreme
solutions. In the first step of this approach, the goal values of the objective functions
Mathematics 2021, 9, 483 10 of 22

Gk are set either through solving the optimization problem for each objective function
individually or consulting decision makers. Next, the deviation variables in both negative
uk and positive ok directions from the goal value of each objective function are added.
The result of the above step is to form new constraints of the model that are presented in
Equations (26)–(31).
!
∑ ∑ Pi .Sit − ∑ ∑ ∑(PCijF .xijt + PCijT .yijt + OTCij .vijt ).BZi
i t i j t
!
− ∑ ∑ LCFj .Wjt + LCTj .TPTjt + LCOT
j .OTjt
j t (26)
− ∑ ∑ ∑ ICij .Iijt .BZi − ∑ ∑ ∑ BOCij .Bijt .BZi
i j t i !j t

− ∑ ∑ HCj .Hjt + FCj .Fjt + u1 − o1 = G1

j t

∑∑∑


xijt + yijt + vijt .BZi .EFj .SE) + u2 − o2 = G2 (27)
i j t

∑∑


Hjt + Fjt + u3 − o3 = G3 (28)
j t

∑ ∑ ∑ Bijt .BZi + u4 − o4 = G4 (29)

i j t

∑ ∑ MPTjt + u5 − o5 = G5 (30)
j t
 
Sit
∑t Pr min Ditmin + Pr most Ditmost + Pr max Ditmax
∑ T.I
+ u6 − o6 = G6 (31)
i
 
u1
. WE1 ≤ ω (32)
G1
 
o2
. WE2 ≤ ω (33)
G1
 
o3
. WE3 ≤ ω (34)
G1
 
o4
. WE4 ≤ ω (35)
G1
 
u5
. WE5 ≤ ω (36)
G1
 
u6
. WE6 ≤ ω (37)
G1
The goal values are then used as a normalized constant according to the percentage
normalization method to convert the deviation variables to the same unit. Additionally,
each deviation variable of each objective function has a weighting parameter WEk that
represents its importance relative to the rest. These deviation variables are controlled by the
maximal deviation variable ω as constraints (32)–(37). Finally, the only objective function
of the model is to minimize the maximum deviation variable with Equation (38).

Minimize ω (38)
 normalization method to convert the deviation variables to the same unit. Additionally,
each deviation variable of each objective function has a weighting parameter 𝑊𝐸𝑘 that
represents its importance relative to the rest. These deviation variables are controlled by
the maximal deviation variable 𝜔 as constraints (32)–(37). Finally, the only objective func-
Mathematics 2021, 9, 483 tion of the model is to minimize the maximum deviation variable with Equation (38). 11 of 22

𝑴𝒊𝒏𝒊𝒎𝒊𝒛𝒆 𝜔 (38)

4.4.Results
Results
4.1.
4.1.Case
CaseStudy
StudyofofTextile
TextileIndustry
Industry
InInthis
thisstudy,
study,a amultiple
multipleobjective
objectiveoptimization
optimizationmodel
modelwas wasdeveloped,
developed,which
whichapplies
applies
for the multinational textile supplier’s production planning process. Their
for the multinational textile supplier’s production planning process. Their production productionsys-
system consists
tem consists of of
twotwo manufacturing
manufacturing facilities
facilities in Vietnam
in Vietnam (VN-1,
(VN-1, VN-2)
VN-2) and and another
another in
in Cam-
Cambodia
bodia (KH). (KH). These
These facilities
facilities employ
employ an average
an average of 500
of 500 full-time
full-time employees
employees to produce
to produce five
five
product families including large-sized handbags (LH), handbags (HB), backpacks(BP),
product families including large-sized handbags (LH), handbags (HB), backpacks (BP),
wallets (W), and key bags (KB) with a total production in 2019 of approximately one million
wallets (W), and key bags (KB) with a total production in 2019 of approximately one mil-
products as presented in Figure 1. According to production managers, the average required
lion products as presented in Figure 1. According to production managers, the average
production time of the above product families for temporary and full-time workers is
required production time of the above product families for temporary and full-time work-
different as described in Figure 2.
ers is different as described in Figure 2.

KB
W
PB
HB
LHB

- 50,000 100,000 150,000 200,000 250,000 300,000

VN-1 VN-2 KH
Mathematics 2021, 9, x FOR PEER REVIEW 12 of 24
Figure1.1.Total
Figure Totalproduction
productioninin 2019
2019 (unit).
(unit).

4.5 1
4
3.5 0.8
3
0.6
2.5
2
0.4
1.5
1 0.2
0.5
0 0
LHB HB PB W KB LHB HB PB W KB

Full-time worker Temporary worker Full-time worker Temporary worker

(a) (b)

Figure
Figure2.2.Required
Requiredprocessing
processingtime
time(hour/unit).
(hour/unit).(a)(a)
Operator (b)(b)
Operator Machine.
Machine.

InInTable
Table5,5,manufacturing
manufacturing facilities have
facilities the the
have same totaltotal
same number of periodic
number working
of periodic work-
hours, but different
ing hours, labor labor
but different policies such as
policies allowable
such rate ofrate
as allowable the of
workforce changing,
the workforce up-
changing,
per/lower
upper/lower limitslimits
of workforce level, level,
of workforce maximummaximuminventory level, overtime
inventory limit, and
level, overtime ma-
limit, and
chine capacity.
machine capacity.
To estimate the production system emissions, this study is based on the textile indus-
Table
try’s 5. Facility electrical
average productionenergy
policy.consumption [43] and the emissions factor of the country in
which the products are manufactured [44]. Under uncertain conditions, the expected value
Facility VN-1 VN-2 KH
of demand for product families is estimated using the three-point estimation technique
Workforce change rate (%) 20% 20% 30%
with the program evaluation and review technique (PERT) distribution. Accordingly, the
Initial workforce (man) 150 70 100
Minimum workforce level (man) 150 50 100
Maximum workforce (man) 200 100 150
Overtime limit (%) 20% 20% 40%
Maximum inventory level (unit/period) 20,000 15,000 20,000
Mathematics 2021, 9, 483 12 of 22

probabilities for the known minimum, most-likely and forecasted maximum demand are
1/6, 4/6, and 1/6 respectively. The optimization problem is solved through IBM CPLEX
software version 12.8 on hardware CPU i7 intel and 16 GB RAM. Table 6 describes the
dimensions of the optimization problem. By solving the model with each objective function
individually, the extreme solutions and their CPU times form the payoff matrix as Table 7.

Table 5. Facility production policy.

Facility VN-1 VN-2 KH

Workforce change rate (%) 20% 20% 30%
Initial workforce (man) 150 70 100
Minimum workforce level (man) 150 50 100
Maximum workforce (man) 200 100 150
Overtime limit (%) 20% 20% 40%
Maximum inventory level (unit/period) 20,000 15,000 20,000
Machine capacity (hour) 62,400 49,920 62,400

Table 6. Optimization problem size.

Decision Variables Non-Zero

Constraints
Binary Integer Float Coefficients
880 72 358 47 5225

Table 7. Objective function payoff matrix.

Minimize Maximize
Maximize Minimize Minimize Maximize Machine
Objective Function Value Workforce Customer
Profit Emission Backorder Operation Time
Changing Satisfaction Level
Profit (mil. VND) 516,838 88,107 138,952 48,253 245,826 391,505
Emission (tCO2) 65.06 19.59 43.17 49.38 89.54 83.44
Workforce changing (man) 132 94 0 198 130 130
Backorder (Unit) 11,550 14,000 7000 0 10,900 11,950
Machine operation time (hour) 473,753 230,519 277,133 294,848 609,647 572,256
Customer satisfaction level (%) 74.45 36.40 44.14 45.28 85.97 88.42
CPU time (second) 2.97 2.51 2.37 2.36 4.10 28.89
No. of Iterations 13,203 2679 123 113 34,650 2,187,084

4.2. Goal Programming Solution

Based on the extreme solutions found above, this study defines the goal values as
the first step in applying CGP to the model. Meanwhile, the weighting parameters of the
objective functions use the non-negative power of ten to represent the dominant nature
or the priority between the objective functions. Table 8 shows that 1% deviation of the
goal value of profit is ten times more important than the goal value related to emissions.
Similarly, every 1% deviation of the emission-related goal value is more important than
a thousand times the goal value relative to machine operation time. In other words, the
objective function related to machine operation time is dominated by the emission-related
objective function.
As shown in Table 9, the solution obtained from the CGP model has some differences
in the value of the objective functions compared to the supplier’s current production plan in
2019 which is named CP solution in this article. According to these results, CGP’s solution
resulted in improvements in profitability, emissions, workforce changing and backorder of
21.02%, 37.16%, 30.77% and 23.44% respectively compared to the CP solution. Although
the expected value of profit increases, the CGP model also sacrifices the objective functions
of machine operation time and customer satisfaction level.
Mathematics 2021, 9, 483 13 of 22

Table 8. Objective function’s goal value and weight.

Objective Function Goal Value (Gk ) Weight (WEk )

Maximize profit 516,838 105
Minimize emission 19.59 104
Minimize workforce changing 0 103
Minimize backorder 0 102
Maximize machine operation time 609,647 101
Maximize customer satisfaction level 88.42 100

Table 9. Chebyshev goal programming solution’s objective function value.

Mathematics 2021, 9, x FOR PEER REVIEW 14 of 24

Objective Function Value CP CGP Improvement

Profit (mil. VND) 363,093 439,417 21.02%
Emission
Workforce changing (tCO2)
(man) 130 77.8690 48.93
30.77% 37.16%
Workforce changing (man) 130 90 30.77%
Backorder (Unit) (Unit)
Backorder 9250 7100
9250 23.24%
7100 23.24%
Machine operation time (hour) 591,511 394,361 −33.33%
MachineCustomer
operationsatisfaction
time (hour)level (%) 591,511 86.84%
394,361 −33.33%
62.26% −28.30%
Customer satisfaction level (%) 86.84% 62.26% −28.30%
In other words, the CGP solution focuses on effectiveness rather than efficiency. In
Figure In3, other
the production plansolution
words, the CGP of the HB, W, on
focuses and KB product
effectiveness families
rather is determined
than efficiency. In by
theFigure
model3, tothebe
production
lower thanplan ofCP
the the production
HB, W, and KBplanproduct families is demand.
and uncertain determined by the
This decision
model
leads to to
thebeconsequences
lower than the CP
of production plan and
low inventory uncertain
levels demand.
as well This decision
as reduced leads with
production
to the consequences of low inventory levels as well as reduced production with
temporary workers or overtime thereby reducing total costs. Meanwhile, the solution of temporary
theworkers
model or overtime thereby
concentrates reducing
resources total costs.
to maintain theMeanwhile,
production thelevel
solution of theeffectiveness
of more model
concentrates resources to maintain the production level of more effectiveness product
product families such as LHB and PB. The CGP model’s production plan of these two
families such as LHB and PB. The CGP model’s production plan of these two product
product families to chase their uncertain demand.
families to chase their uncertain demand.

LHB HB PB W KB

90,000
80,000
70,000
60,000
50,000
40,000
30,000
20,000
10,000
-
CP Sale CGP Sale Expected CP Sale CGP Sale Expected CP Sale CGP Sale Expected CP Sale CGP Sale Expected
demand demand demand demand
Quarter 1 Quarter 2 Quarter 3 Quarter 4

Figure 3. Expected sale and expected demand (unit).

Figure 3. Expected sale and expected demand (unit).
As one of the factors affecting production levels, the workforce level policy also has
As one of the factors affecting production levels, the workforce level policy also has
the difference between the current operation and the solution of the model as shown in
theFigure
difference between the current operation and the solution of the model as shown in
4. For CP solutions, the workforce is replenished in the first two quarters to ensure
Figure 4. For CP
production before solutions, the workforce
being maintained is replenished facilities.
in all manufacturing in the first two quarters
Meanwhile, to ensure
the CGP
production beforethat
solution shows being maintained
the workforce hasindifferent
all manufacturing facilities. Meanwhile,
changes in manufacturing the CGP
facilities. The
solution shows
labor force that
level the workforce
increased has
slightly at thedifferent changes
VN-1 facility in theinfirst
manufacturing facilities. The
quarter and increased
labor
by force
50% atlevel increased
the KH facilityslightly at the
in the first twoVN-1 facility
quarters. in the first
Especially, quarter
at the VN-2 and increased
facility, the by
50% at the of
number KH facilityworkers
full-time in the first two quarters.
is downwardly Especially,
adjusted at the
in the last twoVN-2 facility,
quarters of thethe number
year
to maintain
of full-time economic
workers effectiveness. adjusted
is downwardly In general,inthe
theCGP
last solution suggests
two quarters a lower
of the year work-
to maintain
force level and more variation over time than the manager’s current policy.
Mathematics 2021, 9, 483 14 of 22

Mathematics 2021, 9, x FOR PEER REVIEW 15 of 24

economic
Mathematics 2021, 9, x FOR PEER REVIEW effectiveness. In general, the CGP solution suggests a lower workforce
15 of 24 level and
more variation over time than the manager’s current policy.

200 200 200 170 170 170 170

180 200 200 200 150 170 170150 170 150 170 150
180 150 130 150 150 150
150 150 150 150 130
150 130 150 150 150 100
130 100
100 100 100 100
100 84 84 100 100 100 70 70 70
70 70 70 70 70 62 62 50
50

InitialInitial
Quarter 1 Quarter
Quarter 2 Quarter
1 Quarter 3 3Quarter
2 Quarter Quarter44 Initial Quarter
Initial Quarter1 1Quarter
Quarter 2 Quarter
2 Quarter 3 Quarter
3 Quarter 4 4

VN-1VN-1 VN-2
VN-2 KHKH VN-1
VN-1 VN-2VN-2 KH KH

(a) (a) (b)

(b)
Figure 4. Workforce level changing (man). (a) CP (b) CGP.
Figure 4. 4.Workforce
Figure Workforcelevel changing(man).
level changing (man).(a)(a)
CPCP
(b) (b)
CGP.CGP.
According to detailed production plan, shown in Figure 5, CP solution’s regular pro-
According
duction plan to
According showsdetailed
to detailed production
production
variability in bothplan, plan,
volume shown
andshown in
in Figure
variety of 5,Figure 5, CP regular
CP solution’s
product families solution’s
at each pro- regular
duction
production plan shows
plan shows
manufacturing variability
facility.variabilityin both
In other words, volume
in each and
bothmanufacturing variety
volume andfacility of
varietyproduct
wasof families
product
in charge at each at each
families
of three
manufacturing
to four product facility.
familiesIn other
with words,
volumes each
that manufacturing
fluctuate significantly
manufacturing facility. In other words, each manufacturing facility was in charge facility
over was
time. in charge
For CGP of
so- three
of three to
to four product
lution, besides families
lower with
total volumes
production, that
the rolefluctuate
of significantly
production
four product families with volumes that fluctuate significantly over time. For CGPfacilities over
is alsotime.
more For CGP
clearly so-solution,
defined
lution, to provide
besides lowerstability in production.
total production, theIn which, the productfacilities
families which
is alsoinvolves
besides lower
handbag astotal
LHB production,
and HB are the role
manufactured ofrole
at
of production
production
facilities in facilities
Vietnam. is also
Meanwhile,
more clearly
more clearly
theinvolves
re- defined
defined to provide stability in production. In which, the product families which
to provide stability
mainingasproduction in production. In which, the product families which involves handbag
handbag LHB andlines HB are
areset up at a manufacturing
manufactured facility
at facilities in inVietnam.
Cambodia. This arrange-
Meanwhile, the re-
as LHBment
maining and HB
helps the are
production manufactured
production
lines are set up at aatmanufacturing
system save time facilities
and the costin of
Vietnam. Meanwhile,
re-establishing
facility in Cambodia. the remaining
the production
This arrange-
lines.
production
ment helpslines are set upsystem
the production at a manufacturing
save time and thefacility in Cambodia.
cost of re-establishing theThis arrangement
production
helps the production system save time and the cost of re-establishing the production lines.
lines.
120,000 120,000
100,000 100,000
120,000 120,000
80,000 80,000
100,000 100,000
60,000 60,000
80,000 80,000
40,000 40,000
60,000 60,000
20,000
20,000
40,000 -
40,000
-
Quarter 1
Quarter 2
Quarter 3
Quarter 4
Quarter 4Quarter 1
Quarter 1Quarter 2
Quarter 2Quarter 3
Quarter 3Quarter 4
Quarter 4Quarter 1
Quarter 1Quarter 2
Quarter 3
Quarter 4
Quarter 2 Quarter 1
Quarter 3 Quarter 2
Quarter 4 Quarter 3
Quarter 1 Quarter 4
Quarter 2 Quarter 1
Quarter 2
Quarter 3
Quarter 4
Quarter 1
Quarter 2
Quarter 3 Quarter 3
Quarter 4 Quarter 4

20,000 20,000

- -
Quarter 1
Quarter 2
Quarter 3

Quarter 2
Quarter 3
Quarter 4
Quarter 1

Quarter 3
Quarter 4
Quarter 1
Quarter 2

VN-1 VN-2 KH VN-1 VN-2 KH

LHB HB PB W KB LHB HB PB W KB

VN-1 VN-2 KH VN-1 VN-2 KH

(a) (b)
LHB HB PB W KB LHB HB PB W KB
Figure 5. Regular production plan (unit). (a) CP (b) CGP.

(a) Figures 6 and 7 shows the CP solution’s production plan(b) by temporary workers and
overtime. It can be observed that these two forms of production occur commonly in all
Figure 5. Regular
manufacturing production
facilities. plan (unit).
In contrast, (a) CP
because (b) selection
of CP
the CGP.
Figure 5. Regular production plan (unit). (a) (b) CGP. of an effectiveness level of
Figures 6 and 7 shows the CP solution’s production plan by temporary workers and
Figures 6 and 7 shows the CP solution’s production plan by temporary workers
overtime. It can be observed that these two forms of production occur commonly in all
andmanufacturing
overtime. Itfacilities.
can be observed
In contrast,that these
because of two forms of
the selection of production occur
an effectiveness levelcommonly
of
in all manufacturing facilities. In contrast, because of the selection of an effectiveness
level of production, CGP profits do not use the temporary workforce in production and
only approximately 25,000 units are produced by overtime. This overtime production is
equivalent to 11.63% compared to CP solution.
 Mathematics 2021, 9, x FOR PEER REVIEW 16 of 24
Mathematics 2021, 9, x FOR PEER REVIEW 16 of 24

Mathematics 2021, 9, 483 production, CGP profits do not use the temporary workforce in production and only ap- 15 of 22
proximatelyCGP
production, 25,000 unitsdo
profits arenot
produced
use the by overtime.
temporary This overtime
workforce production
in production andisonly
equiva-
ap-
lent to 11.63%
proximately compared
25,000 units to
areCP solution.by overtime. This overtime production is equiva-
produced
lent to 11.63% compared to CP solution.
4500
4000
4500
3500
4000
3000
3500
2500
3000
2000
2500
1500
2000
1000
1500
500
1000
5000 1 1

2 2

3 3

4 4

1 1

2 2

3 3

4 4

1 1

2 2

3 3

4 4
0
Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter
Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter
VN-1 VN-2 KH
VN-1 LHB HB VN-2
PB W KB KH

LHB HB PB W KB
Figure 6. CP solution’s temporary worker production plan (unit).
Figure 6. CP solution’s temporary worker production plan (unit).
Figure 6. CP solution’s temporary worker production plan (unit).
60,000
50,000
60,000
40,000
50,000
30,000
40,000
20,000
30,000
10,000
20,000
-
10,000
1 1

2 2

3 3

4 4

1 1

2 2

3 3

4 4

1 1

2 2

3 3

4 4
-
Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter
Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter

Quarter
VN-1 VN-2 KH
VN-1 LHB HB VN-2 W
PB KB KH

LHB HB PB W KB
Figure 7. CP solution’s overtime production (unit).
Figure 7. CP solution’s overtime production (unit).
7. CPproduction
FigureSince solution’s overtime
levels areproduction
determined(unit).
more efficiently, the inventory levels of the
CGPSince
solution are also levels
production lower are
with the same safety
determined stock policy
more efficiently, theasinventory
shown in levels
Figureof8.the
In
Sincethe
addition, production levels are determined more efficiently, the disadvantages
inventory levelsin of the
CGP solution variety
are alsooflower
product
withfamilies of CP
the same solution
safety stock also hasasmore
policy shown in Figure 8. In
CGP solution
warehouse are also lower
management at with the same
manufacturing safety stock policy as shown in Figure 8. In
facilities.
addition, the variety of product families of CP solution also has more disadvantages
Mathematics 2021, 9, x FOR PEER REVIEW 17 ofin24
addition,
warehousethe variety ofatproduct
management families
manufacturing of CP solution also has more disadvantages in
facilities.
warehouse management at manufacturing facilities.

25,000 16,000
14,000
20,000 12,000
15,000 10,000
8,000
10,000 6,000
5,000 4,000
2,000
- -
Quarter 1
Quarter 2
Quarter 3
Quarter 4
Quarter 1
Quarter 2
Quarter 3
Quarter 4
Quarter 1
Quarter 2
Quarter 3
Quarter 4

Quarter 1
Quarter 2
Quarter 3
Quarter 4
Quarter 1
Quarter 2
Quarter 3
Quarter 4
Quarter 1
Quarter 2
Quarter 3
Quarter 4

VN-1 VN-2 KH VN-1 VN-2 KH

LHB HB PB W KB LHB HB PB W KB

(a) (b)

Figure 8. Inventory
Figure 8. level(unit).
Inventory level (unit).
(a)(a)
CPCP
(b) (b)
CGP.CGP.

For backorder, both solutions show the backorder that appeared at facility KH where
there was the largest planned output as shown in Figure 9. Accordingly, the backorder of
the CGP solution is concentrated in the last quarters of the year and is approximately
76.76% equivalent to the CP solution.
 For backorder, both solutions show the backorder that appeared at facility KH where
there was the largest planned output as shown in Figure 9. Accordingly, the backorder of
the CGP solution is concentrated in the last quarters of the year and is approximately
76.76% equivalent to the CP solution.
Mathematics 2021, 9, 483 To summarize, the solution that the optimization model provides with the CGP
16 of 22
multi-objective approach shows advantages in terms of economic efficiency, lower
backorder, and environmental and social sustainability factors. On the other hands, this
solution sacrifices the utilization of production machines as well as the risk of the absence
For backorder, both solutions show the backorder that appeared at facility KH where
of goods when market demand increases, because of a low and more secured production
there was the largest planned output as shown in Figure 9. Accordingly, the backorder
level. However, the optimization model in this study still ensures decision-makers’ right
of the CGP solution is concentrated in the last quarters of the year and is approximately
to make decisions about satisfaction and sacrifice objectives through weighting parame-
76.76% equivalent to the CP solution.
ters.

6000

5000

4000

3000

2000

1000

0
CP CGP CP CGP CP CGP CP CGP
Quarter 1 Quarter 2 Quarter 3 Quarter 4

LHB HB PB W KB

Figure9.9.Backorder
Figure Backorderlevel
levelininKH
KHfacility
facility(unit).
(unit).

4.3. To
Experiments
summarize, the solution that the optimization model provides with the CGP multi-
objective approach shows advantages in terms of economic efficiency, lower backorder, and
environmental and social sustainability factors. On the other hands, this solution sacrifices
the utilization of production machines as well as the risk of the absence of goods when
market demand increases, because of a low and more secured production level. However,
the optimization model in this study still ensures decision-makers’ right to make decisions
about satisfaction and sacrifice objectives through weighting parameters.

4.3. Experiments
In this part, the authors examine the changes of the CGP solution according to two
factors: Weights of the objective function and probability distributions of demand. For
weighting, according to combinatorics, the number of weight combinations increases with
a factorial function of the number of objective functions. Thus, there are seven hundred
and twenty weight combinations for the six objective functions. However, examining
all combinations is ineffective because combinations with similar first, second, and third
priority of objective function do not make significant differences in results. Therefore, this
study selected seven weighting combinations that have significant differences in priority
among the objective functions. For demand’s probability distributions, the expected value
of demand is estimated using the three-point estimation method in which PERT distribution
and triangular distribution are widely used. The probabilities for the known minimum,
most-likely and forecasted maximum demand, which are provide by those distribution,
are presented in Table 10.

Table 10. Demand probability distributions.

Minimum Demand Probability Most-Likely Demand Probability Maximum Demand Probability

Distribution
Prmin Prmost Prmax
Program evaluation and review
1/6 4/6 1/6
technique (PERT) distribution
Triangular distribution 1/3 1/3 1/3
Mathematics 2021, 9, 483 17 of 22

Accordingly, the authors develop fourteen scenarios based on the two mentioned
factors as described and denoted in Table 11. In it, the scenario where its solution has been
discussed in the previous section is named S-Base.

Table 11. Experiment scenarios.

Objective Function Weight

Demand Probability Minimize Maximize
Scenario Maximize Minimize Minimize Maximize Customer
Distribution Workforce Machine
Profit Emission Backorder Satisfaction Level
Changing Operation Time
S-Base 100,000 10,000 1000 100 10 1
S-1 100,000 10,000 1 10 100 1000
S-2 10,000 1000 1 10 100 100,000
S-3 PERT 10,000 100 100,000 1000 1 10
S-4 10,000 100,000 10 1 100 1000
S-5 1000 100 10 1 100,000 10,000
S-6 100 10,000 10 100,000 1000 1
S-7 100,000 10,000 1000 100 10 1
S-8 100,000 10,000 1 10 100 1000
S-9 10,000 1000 1 10 100 100,000
S-10 Triangular 10,000 100 100,000 1000 1 10
S-11 10,000 100,000 10 1 100 1000
S-12 1000 100 10 1 100,000 10,000
S-13 100 10,000 10 100,000 1000 1

Table 12 shows the value of the objective functions for scenarios along with the
supplier’s current operations. It can be seen that there are no significant differences in the
results between pairs of scenarios with the same weight combinations such as (S-1, S-8),
(S-4, S-10), etc. This implies that the difference of the PERT and Triangular distributions
does not clearly affect the optimization results. On the other hand, the optimization results
of the model clearly show the influence of the weight combinations, which represent the
will of the decision-makers on the production system. These scenarios all show their own
strengths and weaknesses compared to the rest. However, the solutions of the scenarios
S-4, S-6, S-11, and S-13 have low practical application because of low economic efficiency
compared to CP solution. The results of scenarios S-5 and S-12 have the same level of
customer satisfaction, change in workforce, and machine operation time as CP solution,
but greater profits and emissions. The tradeoffs for profit and environmental impact
that only creates an improvement in backorder can be difficult to convince production
managers. The S-2 and S-9 scenarios are almost identical to the supplier’s current policies,
but generate more profits and lower emissions. However, these two scenarios have much
higher backorder than the rest. Therefore, when applying them, suppliers need to pay
certain attention to the loyalty of customers. Breakthrough scenarios from current operating
policy are S-Base, S-1, S-3, S-7, S-8, and S-10 with marked improvement in profitability and
sustainable development factors. Among them, the S-3 and S-10 scenarios sacrifice a small
profits and environmental impact to achieve less variability in workforce, less backorder,
and greater customer satisfaction level.
Finally, the authors normalize the units between the objective functions of the scenarios
to scores in the interval [0, 1] in order to analyze the trade-offs between them. The scores
presented in Figure 10 show that the S-3 and S-10 scenarios have the greatest number of
objective functions whose scores are greater than 0.7. Next is the group of S-Base, S-1,
S-7, and S-8 scenarios that have three objective functions with scores between 0.4 and 0.7,
besides one objective function that achieves upper extreme scores. The remaining scenarios
also have maximum scores at a certain objective function, but the rest are below 0.5.
Mathematics 2021, 9, 483 18 of 22

Table 12. Scenario objective function value.

Profit Emission Workforce Changing Backorder Machine Operation Time Customer Satisfaction Level
Scenario
(mil. VND) (tCO2) (man) (Unit) (hour) (%)
CP 363,093 77.8580 130 9250 591,511 86.84%
S-Base 439,417 48.9308 90 7100 394,361 62.26%
S-1 439,424 48.9305 92 9000 394,236 62.32%
S-2 379,722 71.5603 139 14,750 529,425 86.08%
S-3 417,553 58.0736 10 2900 449,700 68.46%
S-4 184,562 20.8473 70 5150 220,108 36.40%
S-5 283,317 79.3844 130 6300 606,894 86.60%
S-6 44,177 20.7368 94 1000 252,366 36.98%
S-7 439,889 48.7600 99 8000 392,809 63.77%
S-8 439,880 48.7576 97 7850 392,883 63.78%
S-9 386,575 69.1318 130 13,400 513,127 86.18%
S-10 417,350 58.2894 10 2900 446,747 70.40%
S-11 185,070 20.8426 70 5050 219,941 37.34%
S-12
Mathematics 247,333
2021, 9, 83.9829
x FOR PEER REVIEW 130 9950 603,766 87.53% 20 of 24
S-13 53,648 20.7344 76 1000 252,522 38.03%

1.0000
0.9000
0.8000
0.7000
0.6000
0.5000
0.4000
0.3000
0.2000
0.1000
0.0000
S-Base S-1 S-2 S-3 S-4 S-5 S-6 S-7 S-8 S-9 S-10 S-11 S-12 S-13

Profit Emission Workforce changing

Backorder Machine operation time Customer satisfaction level

Figure 10.
Figure 10. Scenario’s
Scenario’s normalized
normalized score.
score.

In conclusion,
In conclusion,thetheoptimization
optimizationmodel
modelininthis
this study
study provides
provides managers
managers with
with solu-
solutions
tions that can improve the production planning process. These solutions
that can improve the production planning process. These solutions can be determined can be deter-
minedthe
under under the influence
influence of bothofsubjective
both subjective and objective
and objective factors.
factors. Subjective
Subjective factors
factors rep-
represent
the will of production managers or decision-makers in choosing the priority betweenbe-
resent the will of production managers or decision-makers in choosing the priority the
tween thefunctions.
objective objective Atfunctions.
the sameAttime,
the same time,
market market
surveys tosurveys
estimatetothe
estimate the variability
objective objective
variability
of demand of demand
along with along with its probabilities
its probabilities also increase also
theincrease
model’sthe model’s effectiveness
effectiveness in practice.
in practice.
5. Conclusions
5. Conclusions
In this study, the authors develop a stochastic multi-objective mixed-integer program-
mingIn this study,
model the sustainable
for global authors develop a stochastic
multi-product multi-objective
production mixed-integer
planning. The variables pro-of
gramming
this modeldecision
model assist for global sustainable
making multi-product
regarding the quantityproduction planning.
of production The variables
options including
of this model
regular assist decision
production, temporarymaking
workerregarding the quantity
production, of production
overtime production, options includ-
backorder for
ing regular
each period,production, temporary
and production worker
facilities production,
in many overtime
countries. production,
Production backorder
resource policies
for each
can period, by
be adjusted andthe
production
managersfacilities
throughinmodel
manyparameters
countries. Production resource
such as facilities’ policies
maximum
can be adjusted
inventory level, by the managers
inventory servicethrough model parameters
level, workforce such as facilities’
level limitation, overtimemaximum
limitation,
inventory limitation,
backorder level, inventory service
machine level, and
capacity, workforce levelmost
so on. The limitation, overtime
optimistic, limitation,
pessimistic, and
backorder limitation,
most-likely values of themachine capacity,
demand, along and
withso on. PERT
their The most optimistic,
probabilities arepessimistic,
the basis for and
the
most-likely
model values of
to calculate thethe demand,
expected alongofwith
value their PERT
the demand probabilities
and are theThe
the safety stock. basis for the
objective
model to calculate the expected value of the demand and the safety stock. The objective
functions that the proposed model considers are, respectively, maximizing total profit,
minimizing production emissions, minimizing change in workforce, minimizing
backorder, maximizing machine operation time, and maximizing customer satisfaction.
The multi-objective solution in this study is determined by the Chebyshev variant of the
Mathematics 2021, 9, 483 19 of 22

functions that the proposed model considers are, respectively, maximizing total profit,
minimizing production emissions, minimizing change in workforce, minimizing backo-
rder, maximizing machine operation time, and maximizing customer satisfaction. The
multi-objective solution in this study is determined by the Chebyshev variant of the goal
programming approach. This variation is able to help goal programming solutions to avoid
extreme values of high priority objective functions. Next, scenarios are developed based
on the change in the weights of the objective functions and the probability distribution
of demand. Analysis of these scenarios helps decision-makers to understand the overall
picture of the production system when these two factors fluctuate.
The main contribution of this study was to develop an integrated multiple objectives
model as a theoretical reinforcement for the production planning problem. The model
considers typical current production system characteristics such as multi-product, multi-
facility, multi-period, and uncertainty factors. The proposed model also provides assistance
for managers in deciding whether or not the production options are applicable. The options
mentioned include regular production, temporary worker production, overtime production,
and backorder. The applicability of these production options can be adjusted through the
parameters to which the model relates to resource constraints. In addition, the model’s
objective functions are defined to ensure both efficient and sustainable development of the
production system. Profit maximization, emission minimization, and employment volatility
minimization are the objective functions for which the model ensures a balance between
the three pillars of sustainable development: Economic, environmental, and social. At the
same time, the objective functions that minimize backorder, maximize machine uptime, and
maximize customer satisfaction towards optimizing production system efficiency. In order
to achieve solutions that are weighted and balanced at the same time for these objective
functions, this study proposes the Chebyshev goal programming approach. Another
contribution of this study is the development and evaluation of scenarios that depend on
the random distributions of uncertain demand and the weights of the objective functions
from the point of view of the production managers. This assessment is a useful reference
for managers to adjust their production plans for uncertainties and production changes
according to the global situation.
Future studies may consider the complexity of other random distributions in terms of
the uncertainties of demand as well as production resources. In addition, the advantages
of machine learning are also suitable approaches to predict the effectiveness of scenarios as
their number may increase exponentially.

Author Contributions: Conceptualization, N.-L.N.; methodology, C.-N.W., N.-L.N.; formal analysis,

N.-L.N., and T.T.T.T.; investigation, N.-L.N.; data curation, N.-L.N.; writing—original draft prepara-
tion, N.-L.N., and T.T.T.T.; writing—review and editing, C.-N.W., N.-L.N.; project administration,
C.-N.W.; funding acquisition, C.-N.W. All authors have read and agreed to the published version of
the manuscript.
Funding: The research was partly supported by the National Kaohsiung University of Science and
Technology, and MOST 109-2622-E-992-026 from the Ministry of Sciences and Technology in Taiwan.
Acknowledgments: The authors appreciate the support from the National Kaohsiung University of
Science and Technology and the Ministry of Sciences and Technology in Taiwan.
Conflicts of Interest: The authors declare no conflict of interest.
Mathematics 2021, 9, 483 20 of 22

Appendix A
Table A1. Production planning linear programming models.

Problem Characteristics Multiple Objectives Production Options

Author Year
Uncertainty Factor Multiple Facility Multiple Product Economy Environment Others Part-Time Worker Production Backorder Overtime
Bakir, M.A. et al. 1998 X X
Vasant, P. et al. 2004 X X
Kanyalkar, A.P. et al. 2005 X X X
Vasant, P.M. 2006 X X
Li, C. et al. 2008 X X X
Orcun, S. et al. 2009 X X
Kezemi Zanjani, M. et al. 2009 X X X X
Elamvazuthi, I. et al. 2009 X X X
Leung, S.C.H. and Chan, S.S.W. 2009 X X X X X X
Ozsan, O. et al. 2010 X X
Baykasoglu, A. and Goken, T. 2010 X X X X X
Leung, S.C.H. et al. 2010 X X X X X X X
Gramani, M.C.N. et al. 2011 X X
Sillekens, T. et al. 2011 X X
Zhang, X. et al. 2011 X X X X
Ning, Y. et al. 2012 X X X X X
Kezemi Zanjani, M. et al. 2013 X X X X
Mortezaei, N. et al. 2013 X X X X
Munhoz, J.R. et al. 2014 X X
Madadi, N. and Wong, K.Y. 2014 X X X X X
Davizón, Y. et al. 2015 X
Kalaf, B.A. et al. 2015 X X X X
Modarres, M. and Izadpanahi, E. 2016 X X X X X
Campo, E.A. et al. 2018 X
Komsiyah, S. et al. 2018 X X X
Hahn, G.J. and Brandenburg, M. 2018 X X X X X
Tsai, W.-H. 2018 X X X X
Djordjevic, I. et al. 2019 X X
Tirkolaee, E.B. et al. 2019 X X X X X
Proposed model X X X X X X X X X
Mathematics 2021, 9, 483 21 of 22

References
1. Sustainable Production Methods in Textile Industry. 2019. Available online: https://www.intechopen.com/books/textile-
industry-and-environment/sustainable-production-methods-in-textile-industry (accessed on 20 December 2020).
2. Huynh, N.-T.; Chien, C.-F. A hybrid multi-subpopulation genetic algorithm for textile batch dyeing scheduling and an empirical
study. Comput. Ind. Eng. 2018, 125, 615–627. [CrossRef]
3. Campo, E.A.; Cano, J.A.C.; Gómez-Montoya, R.A. Linear Programming for Aggregate Production Planning in a Textile Company.
Fibres Text. East. Eur. 2018, 26, 13–19. [CrossRef]
4. Pochet, Y. Mathematical Programming Models and Formulations for Deterministic Production Planning Problems. In Computa-
tional Combinatorial Optimization; Springer: Berlin/Heidelberg, Germany, 2001; pp. 57–111.
5. Salomon, M. Deterministic Lotsizing Models for Production Planning; Springer-Verlag: Berlin/Heidelberg, Germany, 1991.
6. Du, J.; Liang, L.; Chen, Y.; Bi, G.-B. DEA-based production planning. Omega-Int. J. Manag. S. 2010, 38, 105–112. [CrossRef]
7. Charne, A.; Cooper, W.W.; Rhodes, E. Measuring the efficiency of decision making units. Eur. J. Oper. Res. 1978, 2, 429–444.
[CrossRef]
8. Wang, C.-N.; Dang, T.-T.; Nguyen, N.-A.-T. A Computational Model for Determining Levels of Factors in Inventory Management
Using Response Surface Methodology. Mathematics 2020, 8, 1210. [CrossRef]
9. Bagshaw, K.B. A Review of Quantitative Analysis (QA) in Production Planning Decisions Using the Linear Programming Model.
Am. J. Oper. Res. 2019, 9, 255–269. [CrossRef]
10. Kanyalkar, A.P.; Adil, G.K. An integrated aggregate and detailed planning in a multi-site production environment using linear
programming. Int. J. Prod. Res. 2005, 43, 4431–4454. [CrossRef]
11. Davizón, Y.; Martínez-Olvera, C.; Soto, R.; Hinojosa, C.; Espino-Román, P. Optimal Control Approaches to the Aggregate
Production Planning Problem. Sustainability 2015, 7, 16324–16339. [CrossRef]
12. Ozsan, O.; Simsir, F.; Pamukcu, C. Application of Linear Programming in Production Planning at Marble Processing Plants. J.
Min. Sci. 2010, 46, 57–65. [CrossRef]
13. Gramani, M.C.N.; França, P.M.; Arenales, M.N. A linear optimization approach to the combined production planning model. J.
Frankl. Inst. 2011, 348, 1523–1536. [CrossRef]
14. Sillekens, T.; Koberstein, A.; Suhl, L. Aggregate production planning in the automotive industry with special consideration of
workforce flexibility. Int. J. Prod. Res. 2011, 49, 5055–5078. [CrossRef]
15. Munhoz, J.R.; Morabito, R. Optimization approaches to support decision making in the production planning of a citrus company:
A Brazilian case study. Comput. Electron. Agric. 2014, 107, 45–57. [CrossRef]
16. Cheraghalikhani, A.; Khoshalhan, F.; Mokhtari, H. Aggregate production planning: A literature review and future research
directions. Int. J. Ind. Eng. Comput. 2019, 309–330. [CrossRef]
17. Mula, J.; Poler, R.; García-Sabater, J.P.; Lario, F.C. Models for production planning under uncertainty: A review. Int. J. Prod. Econ.
2006, 103, 271–285. [CrossRef]
18. Bakir, M.A.; Byrne, M.D. Stochastic linear optimisation of an MPMP production planning model. Int. J. Prod. Econ. 1998, 55,
87–96. [CrossRef]
19. Vasant, P.; Nagarajan, R.; Yaacob, S. Decision making in industrial production planning using fuzzy linear programming. IMA J.
Manag. Math. 2004, 15, 53–65. [CrossRef]
20. Elamvazuthi, I.; Ganesan, T.; Vasant, P.; Webb, J.F. Application of a Fuzzy Programming Technique to Production Planning in the
Textile Industry. Int. J. Comput. Sci. Inf. Secur. 2009, 6, 238–243.
21. Madadi, N.; Wong, K.Y. A Multiobjective Fuzzy Aggregate Production Planning Model Considering Real Capacity and Quality
of Products. Math. Probl. Eng. 2014, 2014, 1–15. [CrossRef]
22. Vasant, P.M. Fuzzy production planning and its application to decision making. J. Intell. Manuf. 2006, 17, 5–12. [CrossRef]
23. Djordjevic, I.; Petrovic, D.; Stojic, G. A fuzzy linear programming model for aggregated production planning (APP) in the
automotive industry. Comput. Ind. 2019, 110, 48–63. [CrossRef]
24. Kalaf, B.A.; Bakar, R.A.; Soon, L.L.; Monsi, M.B.; Bakheet, A.J.K.; Abbas, I.T. A Modified Fuzzy Multi-Objective Linear Program-
ming to Solve Aggregate Production Planning. Int. J. Pure Appl. Math. 2015, 104, 339–352. [CrossRef]
25. Baykasoglu, A.; Gocken, T. Multi-objective aggregate production planning with fuzzy parameters. Adv. Eng. Softw. 2010, 41,
1124–1131. [CrossRef]
26. Mortezaei, N.; Zulkifli, N.; Hong, T.S.; Yusuff, R.M. Multi-objective aggregate production planning model with fuzzy parameters
and its solving methods. Life Sci. J. 2013, 10, 2406–2414.
27. Ning, Y.; Liu, J.; Yan, L. Uncertain aggregate production planning. Soft Comput. 2012, 17, 617–624. [CrossRef]
28. Orcun, S.; Uzsoy, R.; Kempf, K.G. An integrated production planning model with load-dependent lead-times and safety stocks.
Comput. Chem. Eng. 2009, 33, 2159–2163. [CrossRef]
29. Li, C.; He, X.; Chen, B.; Xu, Q.; Liu, C. A Hybrid Programming Model for Optimal Production Planning under Demand
Uncertainty in Refinery. Chin. J. Chem. Eng. 2008, 16, 241–246. [CrossRef]
30. Kazemi Zanjani, M.; Nourelfath, M.; Ait-Kadi, D. A multi-stage stochastic programming approach for production planning with
uncertainty in the quality of raw materials and demand. Int. J. Prod. Res. 2009, 48, 4701–4723. [CrossRef]
31. Zhang, X.; Prajapati, M.; Peden, E. A stochastic production planning model under uncertain seasonal demand and market growth.
Int. J. Prod. Res. 2011, 49, 1957–1975. [CrossRef]
Mathematics 2021, 9, 483 22 of 22

32. Kazemi Zanjani, M.; Ait-Kadi, D.; Nourelfath, M. A stochastic programming approach for sawmill production planning. Int. J.
Math. Oper. Res. 2013, 5, 1–18. [CrossRef]
33. Komsiyah, S.; Meiliana; Centika, H.E. A Fuzzy Goal Programming Model for Production Planning in Furniture Company. Procedia
Comput. Sci. 2018, 135, 544–552. [CrossRef]
34. Tirkolaee, E.B.; Goli, A.; Weber, G.-W. Multi-objective Aggregate Production Planning Model Considering Overtime and
Outsourcing Options Under Fuzzy Seasonal Demand. In Advances in Manufacturing II; Springer International Publishing: Cham,
Switzerland, 2019.
35. Leung, S.C.H.; Chan, S.S.W. A goal programming model for aggregate production planning with resource utilization constraint.
Comput. Ind. Eng. 2009, 56, 1053–1064. [CrossRef]
36. Leung, S.C.H.; Wu, Y.; Lai, K.K. Multi-site aggregate production planning with multiple objectives: A goal programming
approach. Prod. Plan. Control 2010, 14, 425–436. [CrossRef]
37. Hahn, G.J.; Brandenburg, M. A sustainable aggregate production planning model for the chemical process industry. Comput.
Oper. Res. 2018, 94, 154–168. [CrossRef]
38. Tsai, W.-H. Green Production Planning and Control for the Textile Industry by Using Mathematical Programming and Industry
4.0 Techniques. Energies 2018, 11, 2072. [CrossRef]
39. Modarres, M.; Izadpanahi, E. Aggregate production planning by focusing on energy saving: A robust optimization approach. J.
Clean. Prod. 2016, 133, 1074–1085. [CrossRef]
40. Dasović, B.; Galić, M.; Klanšek, U. A Survey on Integration of Optimization and Project Management Tools for Sustainable
Construction Scheduling. Sustainability 2020, 12, 3405. [CrossRef]
41. Valenko, T.; Klanšek, U. An integration of spreadsheet and project management software for cost optimal time scheduling in
construction. Organ. Technol. Manag. Constr. Int. J. 2017, 9, 1627–1637. [CrossRef]
42. Nahmias, S.; Olsen, T.L. Production and Operations Analysis, 7th ed.; Waveland Press, Inc.: Grove, IL, USA, 2015.
43. Çay, A. Energy consumption and energy saving potential in clothing industry. Energy 2018, 159, 74–85. [CrossRef]
44. List of Grid Emission Factors. 2020. Available online: https://pub.iges.or.jp/pub/iges-list-grid-emission-factors (accessed on
21 January 2021).