Sie sind auf Seite 1von 24

Int J Adv Manuf Technol

DOI 10.1007/s00170-017-0802-3

ORIGINAL ARTICLE

A decentralized supply chain planning model: a case study


of hardboard industry
Peiman Ghasemi 1 & Kaveh Khalili-Damghani 1 & Ashkan Hafezolkotob 1 & Sadigh Raissi 1

Received: 27 January 2017 / Accepted: 12 July 2017


# Springer-Verlag London Ltd. 2017

Abstract This paper presents a new decentralized/ natural gas and oil, water resource management, and crisis
distributed decision-making (DDM) supply chain plan- and disaster planning.
ning model. A supply chain is called decentralized when
various decisions are made in different companies trying Keywords Decentralized supply chain . Distributed
to optimize their own objectives. The proposed approach decision-making . Mathematical modeling . Hardboard
is applied to plan a wood supply chain network called industry
Puya Wood Company. The supply chain network design
problem discussed in this paper is an integrated multi-
echelon, multi-product problem considering total costs of 1 Introduction
purchasing raw materials/returned product, transportation
of raw materials, and inventory holding of raw materials. The fierce competition in today’s markets and the rapid chang-
A coordination mechanism is designed to share informa- ing of customers’ priorities, along with the quick expansion of
tion between the retailer model and the manufacturer technology, have enforced organizations to manage as mem-
model called centralized planning. The main contributions bers of a supply chain (SC) instead of acting as individual
of this paper are summarized as follows: (1) presenting a investors. The success of an SC depends on the integration
decentralized (distributed) approach in order to address and coordination of its all entities to form an efficient supply
the production and distribution problem in supply chain chain network structure; an efficient network leads to cost-
including manufacturer and retailers and (2) proposing a effective operations throughout the chain and attempts to react
coordination mechanism in order to handle problems re- quickly in response to customers’ demands [1].
lated to production and distribution in a two-echelon sup- The supply chain management (SCM) is the integration of
ply chain network through a decentralized approach. The main business processes from end users to original suppliers
proposed model is tested using real case study. Sensitivity that prepare products, services, and information which add
analysis is accomplished to find the effects of changing value for clients [2]. One of the most pertaining SCM business
the parameter on the results. The proposed coordination processes is SC collaborative planning (CP) which is connect-
mechanism provides high-quality solutions in reasonable ed to the coordination of planning and control operations
number of iterations for the case study. This coordination among the SC, i.e., production and distribution processes
mechanism can be applied in other supply chains such as [3]. In the SC CP context, there is a requirement to develop
decision support tools to help explore and analyze alternatives
and to predict actions for SC in order to yield entire optimized
economic proficiency as well as high levels of customer
* Kaveh Khalili-Damghani satisfaction.
kaveh.khalili@gmail.com The main objective of supply chain management is how to
coordinate the autonomous players, so that they work together
1
Department of Industrial Engineering, South Tehran Branch, Islamic as a unit, in the following of the common purpose in variable
Azad University, Tehran, Iran market conditions. Thus, investments are often implementing
Int J Adv Manuf Technol

strategies to coordinate between the members of their supply 2 Literature of past works
chain, in order to make more effective use of restricted re-
sources and to capture the knowledge of suppliers and cus- During the recent years, a lot of researches about collaboration
tomers, in an effort to incorporate and coordinate production mechanism and information sharing issues in supply chain
and information flows through the entire supply chain [4–6]. management have been published. This section presents liter-
One of the most important approaches to coordinate and ature reviews focused on the impact of collaboration and in-
integrate supply chains is the use of centralized decision-mak- formation sharing between the members of the supply chain.
ing. In a centralized decision-making discipline, a single plan- Huang et al. [10] reviewed 69 research papers published
ner knows all the system data. The planner not only has the between 1991 and 2002 inclusive, significantly focusing on
authority to plan the associated operations or processes in all Operational Research (OR) and Production Management ori-
the entities within the network but also determines the source ented journal (81% of works they reviewed) according to the
and seeks the optimal solution (e.g., the minimum cost) for the classifications described in the literature. Those works mainly
entire system [7]. presented mathematical models to assess the effectiveness of
The force of information technology can help supply chain collaboration and information sharing on the supply chain
members to establish partnerships for better supply chain sys- management.
tem efficiency. However, the complete information sharing Kumar and Pugazhendhi [11] proposed a review on infor-
between the manufacturer and the other parts of supply chain mation sharing issues in supply chains from the technological
is not possible due to information privacy, and thus, the point of view. Their paper analyzed the impact of the benefits
existing research results cannot be directly used for real supply and barriers of information sharing. Also, it attempted to note
chain planning problem [8]. that technical issues of technology were out of the scope of the
One of the most important approaches to face and preserve present review.
information security is the use of distributed decision-making. The concept of supply chain management (SCM) has
Distributed decision-making (DDM) is a process of decision developed greatly over time [12]. Recently, this concept
theory in which decision-making power is distributed among includes the opinion of collaboration between the members
several decision-making units (DMUs). These decisions are of a chain, in order to achieve individual goals, along with
related to one another because one decision affects the out- considering the environmental impact of joint decisions
come of another. The central question in distributed decision- [13]. Previous researches described the concept of collabo-
making is how to structure these distributed decision problems ration in supply chain management. Phillips et al. [14] de-
into a coordinated problem. “DDM can therefore be character- scribed collaboration mechanism as an organizational rela-
ized as the plan and coordination of connected decisions” [9]. tionship in which the participants are members of a chain
The objective of this paper is therefore to convert a cen- attempting to invest on resources and to share information
tralized (SC) planning model into a distributed decision- as well as joint decision-making to efficiently solve prob-
making model and to compare the performance of both lems. From Seifert [15], collaboration involves shared ef-
models. A review of the literature of past works will be forts for the transformation of different goals, into a com-
performed to define existing DDM systems. The central- mon global goal, through shared efforts. Soosay et al. [16]
ized model will have to be converted into distributed put specific emphasis on creating the strategic approach
models so that a coordination mechanism can be planned. cooperation agreements, virtual integration, and horizontal,
Both model’s performances will then be compared using a vertical, and lateral integration when searching a collabora-
commercial modeling program and solver, MILP and tive structure within a supply chain. The results demonstrat-
GAMS. ed how differing relationships can impact on the operation
This paper is organized as follows: Sect. 2 provides an of firms and their capacities to innovate. From Cannella and
overview of literature encompassing the centralized and Ciancimino [17], collaboration mechanism consists of the
decentralized systems in supply chain management. In transformation of individual sub-optimal solutions, into a
Sect. 3, the mathematical programming model as a central- centralized one, by sharing information about both the de-
ized forward supply chain model is defined for our case mand and the operations. The findings showed that the im-
study (Puya Hardboard Industry), and then, this model is pact of supply chain collaboration on overall supply chain
separated into retailer model and manufacturer model. A performance is greater than that of order smoothing.
coordination mechanism is prepared in the next section. The simplest coordination mechanism was discussed by
Also, this model is tested numerically in Sect. 6. [18] and was introduced as “upstream planning,” which first
Computational results are shown in Sect. 6, and then, sen- solved the models belonging to the nearest optimum parts to
sitivity analysis and the relationship between percentile gap the customers. Then, a model of one SC part has been solved;
and the number of iterations are presented. Furthermore, we the resulting requirements are passed as forecasts to the stage
conclude our manuscript in the final section. which is immediately upstream.
Int J Adv Manuf Technol

Shin [19] presented a model that is more appropriate for to ecological issues. The results suggest that the careful selec-
industrial application. He presented a collaborative model for tion of tolerances in joinery operation and careful selection of
operational coordination among manufacturing facilities in appropriate tools for the operation can prevent the production
manufacturing logistics network. The objective of this paper stops and can increase tool life up to 116% and cut length up to
was to propose a distributed algorithm that collaboratively 30 km.
works between collaborative facility planning agents and the López De Lacalle et al. [26] investigated a new way of
main coordinator. The results showed that the algorithms preparing CNC programs for high-speed milling of molds.
coupled with collaborative coordination framework give solu- They proposed the error-free programs that made use of opti-
tions that are much better than the optimal approach in terms mum cutting strategies in the interest of surface quality, pro-
of minimizing the dissatisfaction of whole participants in dis- ductivity, and more reliable manufacturing process. They pre-
tributed production. sented two risks in the use of high-speed milling on hardened
Pibernik and Sucky [20] presented a hybrid form of inter- steels. One of these was tool breakage, which may be very
domain coordination mechanism based on partial centraliza- costly and may furthermore entail marks on the workpiece.
tion. They pointed out that while centralized decision-making The other was collisions between the tool and the workpiece
achieves better results than decentralized decision-making, or fixtures, the result of which could be damage to the ceramic
there is an important drawback with implementing the central- bearings in the spindles. In order to minimize these risks, they
ized model: the necessary alignment of individual decisions to proposed a novel control and optimization steps that can be
SC-wide information sharing. This mechanism may overcome considered in computer-aided manufacturing methodology.
the impediments associated with centralized planning and can Song and Zhuang [27] proposed a manufacturer-
develop both overall and company specific results of SCMP government-farmer supply chain model with three players,
comparing to the upstream coordination. They described how which presented the risky behaviors in the process of supply
coordination on the basis of centralization can evaluated and chain risk management. Once the contaminated food was con-
derived feasible ranges for benefit sharing among the partners sumed, customers could get sick. The government, thus, was
involved. partially responsible for society’s health risks from the chem-
On the other hand, decentralized systems tend to be more ical additives. The punishment policies were set up by the
robust to failure than centralized systems [21]. Centralized government to regulate and deter farmers’ and manufacturers’
decision-making is usually favored when industry faces a risky behavior, punishment income, balancing tax income,
complex problem [22], for instance, determining the best lo- and society’s health risks. One decentralized model and two
cations for facilities. Therefore, using the centralized or dis- centralized models were presented. They applied the real data
tributed decision-making at a specific temporal level will de- from the 2008 Sanlu milk powder contamination case. The
pend on the supply chain and the problem under study. results showed that the optimal chemical level in the central-
Spitter et al. [23] developed an MILP model for the oper- ized model was higher than that in the decentralized model,
ation planning of multi-echelon SCs with arbitrary product especially when the food price and the slope for sales amount
structures. The result was a deterministic model that included were high and production cost and chemical cost were low.
planned lead times (LTs) with multi-period capacity consump- Toptal and Çetinkaya [28] analyzed the impact of supply
tion. The authors suggested that the orders received at the start chain coordination on environmental measures under two
of a period must be processed before the given planning lead emission-regulation policies: tax and cap and trade. They an-
time is finished and before its due date. To solve the problem, alyzed the impact of centralized and decentralized replenish-
they proposed two alternative methods that were evaluated ment decisions on carbon emissions. The buyer in this model
using a set of experiments. faces a deterministic demand rate for a single product. The
Lin and Chen [24] proposed a centralized model of the vendor produces at a finite rate and makes deliveries to the
multi-echelon and multi-site production planning problem. buyer on a lot-for-lot manner. Both the vendor and buyer aim
This attempted model achieves temporal integration by com- to minimize their annual costs. They proposed centralized and
bining two different time scales in the same model, i.e., daily decentralized models for the buyer and the vendor to deter-
and monthly time buckets. Also, other practical planning char- mine their ordering and production lot sizes under each policy.
acteristics and constraints, which were defined by mathemat- The results showed that even though such coordination mech-
ical expressions from a real thin film transistor-liquid crystal anisms help the vendor and the buyer decrease their costs
display (TFT-LCD) industry, were incorporated into the without violating emission regulations, the cost minimizing
model. solution may result in increased carbon emission under certain
López De Lacalle et al. [25] considered an economical and circumstances.
technical wood machining method for joinery applications. Baboli et al. [29] proposed a two-echelon supply chain
Joinery business has increased in size and investments in the model consisting of one retailer and one warehouse. They
last 10 years due to the good adaptability of wooden structures determined the optimal ordering policy according to
Int J Adv Manuf Technol

transportation and inventory costs and studied this model in Geunes et al. [33] presented two-echelon problems as a
two cases, centralized and decentralized. In the centralized class of Stackelberg games where the supplier announces a
case, the warehouse and the retailer were considered as a set of time-phased ordering costs to the retailer over a discrete
whole firm, while in the decentralized case, the retailer and time period. They provided a set of easily understood and
the warehouse independently minimized their own costs. The implemented mechanisms that can increase efficiency and re-
application of this model into the pharmaceutical downstream duce the negative impacts of uncoordinated decisions. The
supply chain of a public hospital allows obtaining significant results showed that relatively easily implemented mechanisms
savings. The results showed that the centralized case leads to involving one or two parameters can indeed lead to non-trivial
savings for the warehouse and the whole system as compared efficiency improvements without the need for explicit
to the decentralized case. However, centralized model induces coordination.
losses for the retailer.
Bilir et al. [30] proposed a multi-objective clothing supply 2.2 Centralized decision-making
chain centralized model to analyze the findings of ignoring the
impacts of supply chain network decisions on customer de- Claudia and Taraneh [34] proposed a centralized multi-
mand. The company had approximately 150 retail stores objective model for wood supply chain. The applicability of
throughout Turkey, including three multi-story mega stores the presented model is tested through a case study in the region
and over 500 sales points. The firm is one of Turkey’s first of British Columbia where different utilization routs for avail-
500 big industrial organizations in terms of sales volume, able wood residues were investigated. The aim of the case study
number of employees, and other factors. The application of was to evaluate the potential benefits of installing a forest-based
the proposed model was illustrated through a real-world prob- bio-refinery network for investors, the environmental groups,
lem in clothing sector and was solved as single- and multi- and the communities. The multi-objective optimization model
objective models. The results showed that including a utility was solved using a Pareto-generating approach.
function (based on the number and the location of DCs) in Cambero et al. [35] proposed a multi-period mixed integer
demand substantially changes the value of all three perfor- programming (MIP) model for planning the bio-refinery sup-
mance objectives of the model. ply chains for the production of bioenergy. The contribution of
We summarized the detailed specifications of some of the this model lied on the explicit modeling of energy flows
recent researches in the field of forest supply chain network among bioenergy that was required in order to accurately es-
with centralized/decentralized design in Table 1. timate the material flows. The resulting multi-objective opti-
mization model was applied to a case study in a forest-
2.1 Decentralized decision-making dependent area located in the interior region of British
Columbia, Canada. This model was solved using the
Nematollahi et al. [31] proposed a mathematical model for AUGMECON technique that generated a set of Pareto-
collaborative decision-making on order quantity and corporate optimal solutions. The findings of the case study indicated
social responsibility in a two-stage supply chain. They con- that the replacement of current fossil energy sources with
sidered a supplier-retailer supply chain in a newsvendor set- bioenergy could have the potential to generate a NPV of up
ting. The presented supply chain in this research provided a to 560 M $, and up to 235 highly impacting jobs in the region
seasonal product with stochastic demand. The results of com- during the next 20 years.
petitive and decentralized models were compared and some Peter and Niquidet [36] presented a forest-based transpor-
managerial analyses were provided. Results showed that, un- tation model of the three Canadian prairie provinces. They
der some circumstances, alongside with the increment of sup- used the model to estimate residual fiber utilization and pro-
ply chain profits, more corporate social responsibility invest- duction. The result showed that 6.9 million m3 of processing
ment and better corporate social responsibility performance residuals would be traded over the course of a year. Also, it
level will be achieved through the competitive model than showed that increased demand for fiber can develop the prof-
other decision-making methods. itability of facilities that produce residuals, and can help them
Proch et al. [32] presented the problem of coordinating to keep operating.
supplier development in a decentralized supply chain. They Zhang et al. [37] proposed integrated multi-stage, mixed
studied supplier development in a decentralized supply chain MIP model using multimodal transport for a forest biomass
with a single supplier and a single manufacturer. The results of biofuel supply chain logistics. The model coordinated tactical
their numerical analysis indicated that direct supplier develop- and strategic decisions. Strategic decisions included the capac-
ment leads to a significant improvement of supply chain profit ity, and location of storage yards. Tactical decisions consisted
in comparison with indirect supplier development, and that of of the inventories during a time period and amount of biomass
subsidizing a constant share of supplier costs is not econom- shipped. The objective was to minimize the total cost for
ically reasonable from the manufacturer’s model. transport procurement harvest, storage, and process.
Int J Adv Manuf Technol

Table 1 Literature of centralized/decentralized model to the forest sector

Research Model Objective function Solution approach

Centralized Decentralized Other Service level Income Cost

Inventory Shortage Transport Ordering Locating

Nematollahi et al. (2017) * * * * LINGO 11


Proch et al. (2017) * * Exact using Cplex
Geunes et al. (2016) * * * Exact using GAMS
Claudia and Taraneh (2016) * * * Exact using AIMMS 3.11
Cambero et al. (2016) * * * AUGMECON method
Peter and Niquidet (2016) * * * * Exact using GAMS
Zhang et al. (2016) * * * * * * * Exact using Cplex
Sanei et al. (2016) * * * * * Lagrangian relaxation algorithm
Shabani and Sowlati (2015) * * * * Exact using Cplex
Alam et al. (2014) * * * * Nlp method
Ghaffariyan et al. (2013) * * * Exact using GAMS
Gaudreault et al. (2011) * * * * Exact using C++ and the ILOG
Gaudreault et al. (2010) * * * * ILOG Cplex 9.1
Forget et al. (2008) * * * –
Weintraub and Epstein (2002) * * Column generation
This study * * * * * * Exact using GAMS

The asterisks show the attribute of the research.


Int J Adv Manuf Technol

Alam et al. [38] used optimization modeling techniques to than the CP approach after 2 h, but the non-initialized MIP
analyze the forest inventory information system, containing found a better solution after 8 h.
three quality attributes, along with merchandizing yards to Gaudreault et al. [44] proposed a softwood lumber supply
add value in wood industry supply chains. The results showed chain made and defined specific mathematical models to plan
that by having full forest inventory knowledge of three qual- and schedule operations. In order to coordinate these plans
ities, it is possible to increase the gross profit in comparison to between the three agents, they proposed different coordination
a scenario with little or no certainty of tree quality in the forest mechanisms. They simulated the coordination mechanism
inventory. using the agent-based planning. A supply chain planning
Ghaffariyan et al. [39] proposed a centralized model for the model was built, based on the industrial case. The results
wood supply chain of Western Australia. Using a linear pro- showed how an agent-based simulation tool can be used to
gramming, this study investigated the impact of five opera- integrate different coordination mechanisms.
tional factors: transport distance, moisture mass fraction, en- Forget et al. [45] proposed a mathematical agent model
ergy demand, interest rate, and truck payload on total forest using different planning strategies when decisions are support-
residues supply chain cost in Western Australia. The sensitiv- ed by a distributed planning system. In order to develop tools
ity analysis of centralized model showed no strong correlation to increase the agility of the supply chain and to promote the
between supply chain cost and energy demand. decentralized management of such disturbances, agent-based
Weintraub and Epstein [40] described the physical supply technology takes advantage of the ability of agents to make
chain in the case of Chilean forestry firms and discussed using decisions in a decentralized network by using advanced col-
models and computer systems at different stages, as well as laboration mechanisms. The implementation of this method
the weak coordination between some stages. A series of math- was realized through the FOR@C experimental agent-based
ematical models have been successfully implemented along program, dedicated to supply chain planning for the wood
the chain to support decisions. They discussed the flow of industry.
material along the supply chain, as well as links between
decision-making processes, pointing out where developments 2.4 Research gap
should be made to improve coordination mechanism.
Having reviewed the literature, we concluded that a large
2.3 Centralized and decentralized decision-making number of centralized mathematical programming models
have been described to simultaneously optimize the integra-
Sanei et al. [41] presented a model for the harvesting schedule tion of the entire SC. These researches pointed out a wide
over the planning horizon. The proposed model was applied variety of SC configurations ranging from the single-echelon
on a real case study, leading to a large-scale MIP model that SC to the multi-echelon SC. However, in most of these works,
cannot be solved by commercial solvers in a reasonable time. the conditions and justification for using the centralized ap-
They compared the integrated model with the decoupled plan- proach were not explicitly described. Although there are in-
ning models that implemented in the hardboard industry. Also, creasing numbers of contributions combining the mathemati-
they proposed a heuristic algorithm which iteratively updated cal programming approaches with the most realistic
the Lagrangian relaxation method. The results showed that decentralized decision-making, the SCs considered in most
substantial revenue improvement can be reached by using an of these researches were relatively simple if compared with
integrated tactical planning model rather than a decoupled the business reality. Furthermore, only a few centralized SC
planning approach. hierarchical mathematical programming models had simulta-
Shabani and Sowlati [42] proposed a hybrid, multi-stage, neously addressed temporal integrations [46]. Finally, re-
“stochastic programming-robust optimization” model in order searches that address simultaneous temporal and spatial inte-
to include uncertainty in biomass availability and biomass gration were rare and were required for decision-making pro-
quality. Uncertain parameters impacting the net profit of the cesses in CP [47].
power plant include “biomass quality,” namely, higher heating To describe a research gap of this paper, we found that
value and moisture content. Finally, the robust optimization researchers believed that several types of information can be
was used to solve model’s uncertainty in the quality of shared about the production process, the resources, the pro-
biomass. duction order, or about the planning process. In our review, we
Gaudreault et al. [43] proposed a centralized model for observed that numerous references considered studying infor-
drying and finishing operations in a hardboard lumber facility. mation exchange about the process. A considerable amount of
They used a single model that simultaneously performed pro- researches considered the study of sharing information about
cess scheduling and planning. They evaluated two alternative inventory levels and planning process. The exchange of infor-
formulations. The first one was based on CP and the second mation about the order has been considered in 70% of refer-
on MIP. The results showed that mixed approach was better ences analyzed, while a few papers studied the issue of sharing
Int J Adv Manuf Technol

information among manufacturer and the retailer considering product, transportation of raw materials/returned products/fin-
requested demand. ished products, manufacturing products, inventory holding of
The coordination mechanism proposed in this paper attempts raw materials/finished products, shortage, and environmental
to point out that the both manufacturer and the retailer commu- degradation in a logistics network.
nicate with each other by minimal-information sharing, other The proposed model considers the following assumptions
than their private information such as the requested demand. and constraints:

& The storage capacity for each type of raw materials and
3 Problem description products of the manufacturer is known.
& The demand of each product is assumed to be certain.
The motivation for this paper, model, and solution technique Also, each product is made from different raw materials
was a real-life case in the wood industry. Wood is one of the based on its BOM with a known production cost.
most important goods in Asia and particularly within the & Initial inventory of each raw material and product is equal
Middle East. Consequently, the wood market is highly con- to zero.
trolled by the governments of these countries. The model of & The number of suppliers, distribution centers, and cus-
this paper was defined to design the wood SC network for the tomers are known.
Puya Wood Company. & There are some vehicles in each layer responsible for
Raw materials (wood pieces) used to produce hardboard transporting items to the next layer. For each vehicle, a
are initially cut into little pieces by cutting devices. fixed and variable cost of launching and a limited capacity
Classified chunks of wood are impregnated with a special are considered. Also, the traveling time between two
adhesive that is a combination of urea-formaldehyde resins, layers is known.
paraffin, ammonia, and other additives. The mixture is poured & The considered strategies for production planning in a
onto a special tape and goes to be pressed. During pressing, by multi-period planning system are internal production, in-
applying heat and pressure over the specific amount of time, ventory holding, and shortage. Shortage in satisfying the
hardboard is formed with the desired thickness of the manu- customer demands is considered as a lost sale with a
facturer. After reaching on the ambient temperature, the known cost. Also, the inventory holding cost for each item
pressed hardboards will be delivered to the sanding unit in is known.
order for both surfaces to be sanded, and thus, in addition to & Storage capacity for suppliers and customers are consid-
removing the surface roughness, the thickness of the final ered unlimited.
product will also be uniformed throughout the product. & Depletion of natural resources incurs environmental deg-
This factory has three work stations which in each station, radation cost.
there are two devices for cutting, pressing (continuous type
and able to produce hardboard of thickness 4 to 38 mm), and The general structure of the proposed general logistics net-
sanding. The production line of the hardboard can produce in work is illustrated in Fig. 1. In the general model, the suppliers
sizes of 2.50 × 1.25, 2.50 × 1.83, 3.1 × 1.83, and 3.66 × 1.83 are responsible for providing the raw material to the
(according to the order). The size of storage for keeping safety manufacturing company. The products are conveyed from
stock (buffer storage) is 200 m2. the factory to the customers via the distribution centers to meet
This company is facing the significant competition on cost the customer demands.
and might be forced to minimize the information sharing be- These proposed models are general, manufacturer, and re-
tween the members of the supply chain. Therefore, the com- tailer models. These models can encompass various industries
pany had to decentralize its operation for security and minimal such as plastic, paper, and other industries. To describe the
information sharing. mentioned logistics networks, the following notations are ap-
The key objective of this study is to design a coordi- plied in the mathematical model.
nation mechanism for the hardboard industry supply chain
in the Puya Company in Mazandaran Province of Iran. 3.1 Sets
The mixed-integer linear programming model with the
objective of minimizing the total logistic costs including i = {1, 2, …, I} index set of suppliers
purchase of wood raw and waste materials, transportation, j = {1, 2, …, J} index set of distribution centers
inventory holding, manufacturing, and shortage were for- k = {1, 2, …, K} index set of customers
mulated in the next section. Kj index set of customers covered by
The supply chain network design problem discussed in this distribution center j
paper is an integrated multi-echelon, multi-product problem l = {1, 2, …, L} index set of supplier’s vehicles
considering total costs of purchasing raw materials/returned Li index set of all vehicles in supplier i
Int J Adv Manuf Technol

The distributor is responsible


The factory is responsible for
Raw materials are provided by for distributing manufactured Customers use the
processing the raw material
suppliers and delivered to the products by the factory and manufactured products in
and converting wood to
factory. delivering them to the the home, workplace, etc.
hardboard.
customers.
Fig. 1 The general structure of the proposed general logistic network

m = {1, 2, …, M} index set of the factory’s vehicles 3.2.2 Variable costs


n = {1, 2, …, N} index set of distribution center’s vehicles
Nj index set of all vehicles in the distribution j These costs mainly change based on distance, number of
p = {1, 2, …, P} index set of products products, or time. These parameters are defined as follows:
r = {1, 2, …, R} index set of raw materials
t = {1, 2, …, T} index set of time periods

pcðp1Þ unit production cost of product p ($)


3.2 Model parameters pucir unit purchasing cost of raw material r from supplier i
($)
Model parameters are presented in the following sub-sec- ð1Þ
puckp unit purchasing cost of returned product p from
tions. The first sub-section includes fixed costs and fixed customer k ($)
time parameters. The second sub-section includes variable vcir unit cost of transportation of raw material r from
cost parameters. The third sub-section includes capacity of supplier i to the factory ($)
facility parameters. The fourth sub-section includes param- ð2Þ
vcjp unit transportation cost of product p from the factory
eters of the capacity of the storages. The fifth sub-section to the distributor j ($)
includes ratios. The sixth sub-section describes scalar ð3Þ
vcjkp unit transportation cost of product p from distributor
parameters. j to customer k ($)
sckpt unit shortage cost of product p for customer k in
3.2.1 Fixed costs and fixed times period t ($)
dkpt demand of customer k for product p in period t
Fixed costs and fixed times include cases in which costs or hrt(1) unit inventory cost of raw material r in the factory
time will not change based on the number of products. These during period t ($)
parameters are defined as follows: hpt(1)′ unit inventory cost of product p in the factory during
period t ($)
fcil fixed cost of launching vehicle l from supplier i to hjpt(2) unit inventory cost of product p in distributor j
the factory ($) during period t ($)
ð2Þ
fcjm fixed cost of launching vehicle m from the factory
to distribution center j ($)
ð3Þ
fcjkn fixed cost of launching vehicle n from distribution 3.2.3 Capacity of facilities
center j to customer k ($)
timei traveling time from supplier i to the factory (hour) Capacity of facilities includes the upper limit on the capacity
ð2Þ
time j traveling time from the factory to distribution center of transportation vehicles of products or raw materials. These
j (hour) parameters are defined as follows:
ð3Þ
timejk traveling time of distribution center j to customer k
(hour)
SEj the lower limit of time window for distributor j
SLj the upper limit of time window distributor j Vl capacity of supplier’s vehicle l (m3)
Volr unit volume of raw material r (m3) V ðm2Þ capacity of factory’s vehicle m (m3)
Vol′p unit volume of product p (m3). V ðn3Þ capacity of distribution center’s vehicle n (m3)
Int J Adv Manuf Technol

ð2Þ
3.2.4 Capacity of storages yjmt the number of trips by vehicle m from the factory to
the distributor j during period t
ð3Þ
Capacity of storages includes the upper limit on the storage of yjknt the number of trips by vehicle n from distributor j to
raw materials or products in the factory or distribution centers. customer k during period t
These parameters are defined as follows: Zilt 1 if vehicle l is launched to move from supplier i to
the factory at period t; 0 otherwise
ð2Þ
Z jmt 1 if vehicle m is launched to move from the factory to
capr raw material storage capacity of the factory (m3)
the distributor j at period t; 0 otherwise
capp product storage capacity of the factory (m3) ð3Þ
ð2Þ Z jknt 1 if vehicle n is launched to move from distributor j
cap j storage capacity of distribution center j (m3)
to customer k at period t; 0 otherwise

3.2.5 Ratios
3.4 General model formulation
Ratios considered in the model are defined as follows:
3.4.1 Objective function of the general model
CED unit environmental degradation cost due to depletion of
natural resources ($) The objective of the presented general model is to minimize
Brp usage coefficient of raw material r in product p the total costs of the logistics network including purchase of
raw materials, items transportation by vehicles between
layers, manufacturing, inventory, shortage, and environmental
3.2.6 Others degradation which are formulated through components
(1.1)–(1.6).
This section includes scalar parameters. These parameters are
defined as follows: Min Z !
  ð1:1Þ
¼ ∑ ∑ ∑ ∑ pucir * Qilrt
i∈I l∈Li r∈R t∈T
Hor working time in a period (hour)
!
BigM a very large number
þ ∑ ∑ ∑ fcil Z ilt þ ∑ ∑ ∑ ∑ vcir Qilrt
i∈I l∈Li t∈T i∈I l∈Li r∈R t∈T

!
3.3 Decision variables
ð2Þ ð2Þ ð2Þ ð2Þ
þ ∑ ∑ ∑ fcjm Z jmt þ ∑ ∑ ∑ ∑ vcjp Qjmpt ð1:2Þ
j∈ J m∈M t∈T j∈ J m∈M p∈P t∈T
Qilrt quantity of raw material r transferred from supplier i !
to the factory in period t (unit) ð3Þ ð3Þ ð3Þ ð3Þ
ð1Þ0 þ ∑ ∑ ∑ ∑ fcjkn Z jknt þ∑ ∑ ∑ ∑ ∑ vcjkp Qjknpt
Qpt quantity of product p manufactured by the factory in j∈ J k∈K j n∈N j t∈T j∈ J k∈K j n∈N j p∈P t∈T

period t (unit) !
ð2Þ  * ð1Þ 0

Qjmpt quantity of product p transferred from the factory to þ ∑ ∑ pcðp1Þ Qpt ð1:3Þ
distribution center j (unit) p∈P t∈T
ð3Þ
Qjknpt quantity of product p transferred from distributor j to !
0 0
ð1Þ ð1Þ ð1Þ ð1Þ ð2Þ ð2Þ ð3Þ ð3Þ
customer k by vehicle n in period t þ ∑ ∑ hrt srt þ ∑ ∑ hpt spt þ ∑ ∑ ∑ hjpt sjpt þ ∑ ∑ ∑ hkpt skpt
r∈R t∈T p∈P t∈T j∈ J p∈P t∈T k∈K p∈P t∈T
SHkpt shortage quantity of product p in satisfying the
demand of customer k in period t (unit) ð1:4Þ
ð1Þ
srt final inventory of raw material r at the factory at the þ ∑ ∑ ∑ sckpt SH kpt ð1:5Þ
end of period t (unit) k∈K p∈P t∈T
ð1Þ0  
spt final inventory of product p at the factory at the end
of period t (unit) þ CED ∑ ∑ ∑ ∑ Qilrt ð1:6Þ
ð2Þ i∈I l∈L r∈R t∈T
sjpt final inventory of product p of distributor j during the
period t (unit)
ð3Þ
skpt final inventory of product p of customer k at the end 3.4.2 Constraints of the general model
of period t (unit)
STjm start time of the service for distributor j with This sub-section is a representation of the constraints of the
vehicle m presented general model. Inequalities (2)–(4) are space capac-
yilt the number of trips by vehicle l from supplier i to the ity constraints for vehicles in the suppliers, manufacturer, and
factory during period t distribution centers, respectively.
Int J Adv Manuf Technol
0 0 0
ð1Þ ð1Þ ð2Þ ð1Þ
∑ Vol r Qilrt ≤ Vl yilt ∀i∈I; l∈Li ; r∈R; t∈T ð2Þ Qpt þ spðt−1Þ ¼ ∑ ∑ Qjmpt þ spt ∀p∈P; t∈T ð14Þ
r∈R j∈ J m∈M

0 ð2Þ ð2Þ
∑ Volp Qjmpt ≤ V ðm2Þ yjmt ∀ j∈ J; m∈M; t∈T ð3Þ
ð2 Þ ð2Þ
∑ Qjmpt þ sjpðt−1Þ ¼ ∑ ∑ Qjknpt þ sjpt ∀ j∈ J; p∈P; t∈T
ð3Þ ð2Þ
ð15Þ
p∈P m∈M k∈K j n∈N j

0 ð3Þ ð3Þ Equation (16) is demand satisfaction constraint.


∑ Volp Qjknpt ≤ V ðn3Þ yjknt ∀ j∈ J; k∈K j ; n∈N j ; t∈T ð4Þ
p∈P
ð3Þ ð3Þ ð3 Þ
∑ ∑ Qjknpt þ skpðt−1Þ þ SHkpt ¼ dkpt þ skpt ∀k∈K; p∈P; t∈T ð16Þ
j∈ J n∈N j
Constraints (5)–(9) are time capacity limitations for vehicles.
Constraints (17) and (18) are the logical binary and non-
yilt *timei ≤ Hor*Zilt ∀i∈I; l∈Li ; t∈T ð5Þ negativity integer necessities on the decision variables.
0
ð2Þ ð2Þ ð1Þ ð1Þ ð2Þ ð3Þ ð1Þ ð1Þ0 ð2Þ ð3Þ
∑ yjmt *time j ≤ Hor ∀m∈M; t∈T Qilrt ; Qrt ; Qpt ; Qjmpt ; Qjknpt ; srt ; spt ; sjpt ; skpt ; SH kpt ≥0; and is integer
ð6Þ
j∈ J
ð17Þ
ð3Þ ð3Þ
∑ yjknt *timejk ≤ Hor ∀ j∈ J; n∈N j ; t∈T ð7Þ ð2Þ ð3Þ
k∈K Z jmt ; Z ilt ; Z jknt is binary ð18Þ
ð2Þ ð2Þ
yjmt ≤ BigM *Zjmt ∀ j∈J; m∈M; t∈T ð8Þ
3.5 Manufacturer model formulation
ð3Þ ð3Þ
yjknt ≤ BigM*Zjknt ∀ j∈ J; k∈K j ; n∈N j ; t∈T ð9Þ
In this section, the manufacturer model formulation is
presented.
Constraints (10)–(12) are capacity constraints for storages.
Quantity of all types of wood raw materials received by the 3.5.1 Objective function of the manufacturer model
factory from all the suppliers should be less than or equal to
the storage capacity of the factory. The objective function of manufacturer model contains six
terms and is proposed using Eqs. (19.1)–(19.6).
ð1Þ
∑ Volr srt ≤ capr ∀t∈T ð10Þ Min Z !
r∈R
  ð19:1Þ
¼ ∑ ∑ ∑ ∑ pucir Qilrt *

Quantity of all processed products at the factory should be i∈I l∈Li r∈R t∈T

less than or equal to the finished goods storage capacity of the !


factory.
þ ∑ ∑ ∑ fcil Z ilt þ ∑ ∑ ∑ ∑ vcir Qilrt ð19:2Þ
0 i∈I l∈Li t∈T i∈I l∈Li r∈R t∈T
ð1Þ
0
!
∑ Volp spt ≤ capp ∀t∈T ð11Þ
p∈P ð2Þ ð2Þ ð2Þ ð2Þ
þ ∑ ∑ ∑ fcjm Z jmt þ ∑ ∑ ∑ ∑ vcjp Qjmpt
j∈ J m∈M t∈T j∈ J m∈M p∈P t∈T
Quantity of all products transported from the factory to
!
distributor j should be less than or equal to the storage capacity  0

ð1Þ * ð1Þ
of distributor j. þ ∑ ∑ pcp Qpt ð19:3Þ
p∈P t∈T
0 ð2Þ ð2Þ !
∑ Volp sjpt ≤ cap j ∀ j∈ J; t∈T ð12Þ 0 0
ð1Þ ð1Þ ð1Þ ð1Þ
p∈P
þ ∑ ∑ hrt srt þ ∑ ∑ hpt spt ð19:4Þ
r∈R t∈T p∈P t∈T
Equations (13)–(15) are inventory balance constraints for  
raw materials and products at the factory and products at dis-
þ CED ∑ ∑ ∑ ∑ Qilrt ð19:5Þ
tributors. i∈I l∈L r∈R t∈T

0
ð1Þ ð1Þ ð1Þ
∑ ∑ Qilrt þ srðt−1Þ ¼ ∑ Brp Qpt þ srt ∀t∈T; r∈R ð13Þ þ ∑ ∑ ∑ sckpt SHkpt ð19:6Þ
i∈I l∈Li p∈P k∈K p∈P t∈T
Int J Adv Manuf Technol

Counting
iteration

End
Yes

No Can Generate
No Compute
demand be distribution
satisfied? plan
Yes
Compute If Generate
≥0 production
Start
plan
Generate
Compute initial initial
demand distribution
plan
Manufacturer’s model
Retailer’s model

Fig. 2 Coordination mechanism

ð1Þ
3.5.2 Constraints of the manufacturer model ∑ Volr srt ≤ capr ∀t∈T ð25Þ
r∈R

This sub-section is a representation of the constraints of the Quantity of all processed products at the factory should be
proposed manufacturer model. Inequalities (20) and (21) are less than or equal to the finished goods storage capacity of the
space capacity constraints for vehicles of the suppliers and factory.
manufacturer, respectively.
0
ð1Þ
∑ Volr Qilrt ≤ Vl yilt ∀i∈I; l∈Li ; r∈R; t∈T ð20Þ ∑ Vol0 p spt ≤ capp ∀t∈T ð26Þ
r∈R p∈P
0 ð2Þ ð2Þ
∑ Volp Qjmpt ≤ V ðm2Þ yjmt ∀ j∈ J; m∈M; t∈T ð21Þ
p∈P Equations (27) and (28) are inventory balance constraints
for raw materials and products at the factory and products of
Constraints (22)–(24) are time capacity limitations for ve- distributors.
hicles.
ð1Þ ð1Þ0 ð1Þ
yilt *timei ≤ Hor*Zilt ∀i∈I; l∈Li ; t∈T ð22Þ ∑ ∑ Qilrt þ srðt−1Þ ¼ ∑ Brp Qpt þ srt ∀t∈T; r∈R ð27Þ
i∈I l∈Li p∈P
ð2Þ ð2Þ
∑ yjmt *time j ≤ Hor ∀m∈M; t∈T ð23Þ
j∈ J 0 0 0
ð1Þ ð1Þ ð2Þ ð1Þ
ð2Þ ð2Þ
Qpt þ spðt−1Þ ¼ ∑ ∑ Qjmpt þ spt ∀p∈P; t∈T ð28Þ
yjmt ≤ BigM *Zjmt ∀ j∈J; m∈M; t∈T ð24Þ j∈ J m∈M

Constraints (25) and (26) are capacity constraints for


storages. Constraint (29) replaces the constraints that ensured filling
Quantity of all types of wood raw materials received by the demand. It makes sure that the amount of an item shipped
factory from all the suppliers should be less than or equal to from the plant to a retailer plus any shortage equal the request-
the storage capacity of the factory. ed supply quantity by that retailer for that item. If the shipped

Table 2 Unit transportation cost of raw materials


Table 3 Purchasing cost
r1 r2 r3 r4 r5 of raw materials Supplier r1, r2, r3, r4, r5

I1 3540 3545 3550 3500 3450 I1 18,500


I2 3545 3540 3550 3500 3450 I2 18,500
I3 3540 3540 3550 3550 3500 I3 18,500
I4 3560 3545 3540 3550 3500 I4 18,500
I5 3560 3540 3540 3550 3500 I5 18,500
Int J Adv Manuf Technol

Table 4 Demand Table 6 Capacity of vehicles

t1 t2 t3 t4 Capacity

K1. p1 1000 1400 1000 1400 Suppliers l1 = 5,000, l2 = 12,000, l3 = 50,000,


K1. p2 2000 1500 1000 1500 l4 = 11,000, l5 = 10,000, l6 = 5,000,
l7 = 12,000, l8 = 50,000, l9 = 11,000,
K1. p3 1000 2000 1000 2000
l10 = 10,000
K1. p4 1000 0 1000 0
Factory m1 = 7,000, m2 = 11,000, m3 = 10,000,
K2. p1 800 1000 2000 2500 m4 = 12,000, m5 = 8,000
K2. p2 1000 1500 1000 1500 Distribution centers n1 = 4,000, n2 = 3,000, n3 = 2,500,
K2. p3 1000 1000 1500 1500 n4 = 1,500, n5 = 2,500
K2. p4 600 1000 900 500
K3. p1 1000 1200 1300 1500
K3. p2 1000 1000 1000 2000
amounts do not suffice, shortage is positive and the penalty
K3. p3 1000 0 1200 0
will be incurred.
K3. p4 1000 800 1000 1500
K4. p1 1000 1500 1000 1500 ð2Þ
∑ ∑ Qjmpt þ SH kpt ¼ d kpt ∀k∈K; p∈P; t∈T ð29Þ
K4. p2 1000 1000 1000 1000 j∈ J m∈M
K4. p3 500 800 600 1000
K4. p4 2000 2000 1500 1500 Based on the policies and rules conducted between the
K5. p1 1000 0 0 2000 manufacturer and distributor, manufacturer must deliver the
K5. p2 1000 0 2000 0 products to the distributor in a certain time period (for exam-
K5. p3 800 800 1000 1000 ple, between 2 and 4 pm). So, the constraint 30 ensures the
K5. p4 1000 1500 1000 1500 lower limit of time window of the distributor. The constraint
K6. p1 1000 1000 1000 1000 31 ensures the upper limit of time window of the distributor.
K6. p2 2000 2000 2000 2000  
ð2Þ ð2Þ
K6. p3 1000 1500 0 800 ST jm þ time j :Z jmt ≥ SE j ∀ j∈ J ; m∈M ; ∀t∈T ð30Þ
K6. p4 1000 1200 1200 1000
K7. p1 800 800 1000 800  
ð2Þ ð2Þ
K7. p2 1000 2000 1500 1000 ST jm þ time j :Z jmt ≤ SL j ∀ j∈ J ; m∈M ; ∀t∈T ð31Þ
K7. p3 2000 800 1000 0
K7. p4 1000 0 0 1000
K8. p1 2000 1500 1500 2000
Constraints (32) and (33) are the logical binary and non-
K8. p2 1000 1000 1000 1000
negativity integer necessities on the decision variables.
K8. p3 2000 2000 1000 1000 0 0
ð1Þ ð1Þ ð2Þ ð1Þ ð1Þ ð2Þ
K8. p4 800 1200 1300 800 Qilrt ; Qrt ; Qpt ; Qjmpt ; srt ; spt ; sjpt ; SH kpt ≥ 0; and is integer ð32Þ
K9. p1 1500 1000 0 1000
K9. p2 1000 1000 1000 1000
ð2Þ
K9. p3 1000 1000 800 1000 Z jmt ; Z ilt is binary ð33Þ
K4. p4 2000 1500 800 1200
K10. p1 1500 1500 1000 1000
3.6 Retailer model formulation
K10. p2 600 2000 600 1000
K10. p3 800 600 800 800
In this section, the retailer model formulation is presented.
K10. p4 1000 1000 1000 1000

Table 7 Storage capacity

Table 5 Return time Raw materials Products

Factory Factory capr=100,000 capp=150,000


Distribution centers j1 = 30,000, j2 = 32,000,
Suppliers i1 = 3, i2 = 2, i3 = 3, i4 = 3, i5 = 3 j3 = 32,000, j4 = 28,000,
Distribution centers j1 = 3, j2 = 4, j3 = 3, j4 = 2, j5 = 4 j5 = 35,000
Int J Adv Manuf Technol

Table 8 Traveling time of distribution center j to customer k (hour) Table 9 Quantity of product p manufactured by the factory in period t
in centralized model
K1 K2 K3 K4 K5 K6 K7 K8 K9 K10
ð1Þ0
Qpt t1 t2 t3 t4
J1 2 2.5 4 4 1.5 3 2 1 3 3
J2 3 3.5 2 2.5 1.5 2 2 4 3 2.5
J3 3 1 3 4.5 3 3.5 1.5 2 2.5 4 P1 11,711 10,750 9,950 10,336
J4 2.5 2 3.5 3.5 3.5 2 3 2 2.5 4.5 P2 11,617 13,000 12,100 14,892
J5 3.5 2.5 2.5 3 2 2 3 3 3.5 3 P3 11,100 10,782 8,990 10,521
P4 11,300 10,300 9,511 10,410

3.6.1 Objective function of the retailer model Constraint (38) is capacity constraint for storage.
Quantity of all products transported from the factory to
The objective function of retailer model contains two terms distributor j should be less than or equal to the storage capacity
and is proposed using Eqs. (34.1) and (34.2). of distributor j.
Min Z !
0 ð2Þ ð2Þ
¼ ∑ ∑ ∑ ∑
ð3Þ ð3Þ
fcjkn Z jknt þ∑ ∑ ∑ ∑ ∑
ð3Þ ð3Þ
vcjkp Qjknpt ∑ Volp sjpt ≤ cap j ∀ j∈ J; t∈T ð38Þ
p∈P
j∈ J k∈K j n∈N j t∈T j∈ J k∈K j n∈N j p∈P t∈T

ð34:1Þ
! Equation (39) is inventory balance constraints for products
ð3Þ ð3Þ of distributors.
þ ∑ ∑ ∑ hkpt skpt ð34:2Þ
k∈K p∈P t∈T
ð2Þ ð2Þ ð3Þ ð2Þ
∑ Qjmpt þ sjpðt−1Þ ¼ ∑ ∑ Qjknpt þ sjpt ∀ j∈J; p∈P; t∈T ð39Þ
m∈M k∈K j n∈N j

3.6.2 Constraints of the retailer model


Equation (40) is demand satisfaction constraint.
This sub-section is a representation of the constraints of the
presented retailer model. Inequality (35) is space capacity con-
ð3Þ ð3Þ ð3Þ
straint for vehicles in the distribution centers. ∑ ∑ Qjknpt þ skpðt−1Þ ¼ dkpt þ skpt ∀k∈K; p∈P; t∈T ð40Þ
j∈ J n∈N j

0 ð3Þ ð3Þ
∑ Volp Qjknpt ≤ V ðn3Þ yjknt ∀ j∈ J; k∈K j ; n∈N j ; t∈T ð35Þ
p∈P Constraint (41) enforces that the requested amounts of
items from the manufacturer are at most of what the manufac-
Constraints (36) and (37) are time capacity limitations for turer has indicated he can provide. Constraint (42) is newly
vehicles. added to calculate the input variable for the manufacturer’s
ð3Þ
model, dkpt. It is simply equal to Qjknpt , meaning that it could
ð3Þ ð3Þ
∑ yjknt *timejk ≤ Hor ∀ j∈ J; n∈N j ; t∈T ð36Þ also be sent to the manufacturer’s model. However, because of
k∈K
uniformity, this is changed into dkpt.
ð3Þ ð3Þ
yjknt ≤ BigM*Zjknt ∀ j∈ J; k∈K j ; n∈N j ; t∈T ð37Þ
ð3Þ ð2Þ
∑ ∑ Qjknpt ≤ ∑ Qjmpt ∀ j∈ J; p∈P; t∈T ð41Þ
k∈K n∈N j m∈M

1E+09
ð3Þ
800000000 ∑ ∑ Qjknpt ¼ dkpt ∀k∈K; p∈P; t∈T ð42Þ
Retailer cost j∈ J n∈N j
600000000
Cost

400000000 Manufacturer Constraints (43) and (44) are the logical binary and non-
Cost
200000000 negativity integer necessities of the decision variables.
Total Cost
0
ð2Þ ð3Þ ð2Þ ð3Þ
1 2 3 4 5 6 7 8 9 Qjmpt ; Qjknpt ; sjpt ; skpt ; SH kpt ≥ 0; and is integer ð43Þ
Iteration
ð3Þ
Fig. 3 Relationship between cost and the number of iterations
Z jknt is binary ð44Þ
Int J Adv Manuf Technol

Table 10 Quantity of product p transferred from the factory to nor has knowledge of the other’s model (decision-maker)
distribution center in centralized model
implemented in their own model. The quantity of product
ð2Þ
Qjmpt T1 T2 T3 T4 and demand of customers is sharing between the two
decision-making models directly. The coordination mech-
anism starts with the retailer solving his own problem of
J1.M1.P2 4356 0 3163 0
determining how much to request the product from the
J1.M1.P4 0 0 3900 4230
manufacturer based on customer demand. That informa-
J1.M2.P3 0 4580 0 3207
tion is then sent to the manufacturer who returns his re-
J1.M3.P1 4097 0 0 2615
sponse. If the available supply quantity is enough to at
J1.M4.P3 0 2955 0 0
least satisfy demand, the retailers’ model has a feasible
J1.M5.P1 0 0 3700 0
solution, and the iterative exchange starts. The manufac-
J1.M5.P2 0 0 0 3718
turer and retailers’ model exchange updated solutions un-
J2.M1.P3 3210 0 0 0
til they reach a feasible solution where all products are
J2.M2.P1 2600 0 2830 0
delivered without any shortages. The initial solution used
J2.M2.P4 0 4195 0 0
is one that minimizes retailer’s cost, because it is gener-
J2.M2.P3 0 1700 3889 0
ated by the retailers’ model. After starting the coordina-
J2.M3.P2 0 3650 0 0
tion mechanism, the number of iterations will not be fixed
J2.M4.P1 0 0 3752 0
in the model. The entire manufacturer’s model is minimiz-
J2.M5.P2 4752 0 0 4621
ing costs. He obviously does not know how his decisions
J3.M1.P2 0 4782 0 0
affect the retailer’s model, and will only find out after
J3.M1.P1 0 0 0 3500
making the final decision.
J3.M2.P1 0 2400 0 0
The coordination mechanism is represented graphically in
J3.M3.P2 0 0 0 3000
Fig. 2. The distributed model starts with the generation of the
J3.M3.P3 4616 0 3015 0
initial distribution plan by the retailer, in which he computes
J3.M4.P1 2910 0 0 0
J3.M5.P2 0 0 0 2100 the initially requested demand d initial
kpt . These quantities are sent
J3.M5.P4 3950 0 2900 0 to the manufacturer who generates a production to satisfy the
J4.M1.P1 2030 0 0 2200 product quantities in the best way and at minimal cost. If there
J4.M2.P2 0 1980 4000 0 will be no production shortage, then all requested products can
J4.M3.P1 0 3300 0 0 be delivered, which will end the procedure. If the manufactur-
J4.M4.P3 3000 0 0 0 er will not be able to satisfy all that is requested, the product
J4.M5.P3 0 1581 0 3440 quantity, retailer, and time period will be calculated. This in-
J4.M5.P1 0 0 2550 0 formation will then be sent back to the retailer. He will gener-
J4.M5.P4 4000 2400 3400 ate a new plan, checking whether he can satisfy his customers’
J5.M1.P2 0 2900 2900 0 demand. If not, the problem is infeasible. If he can, then he
J5.M2.P2 3600 0 0 2200 will generate a new quantity and will send it to the manufac-
J5.M3.P4 3600 0 0 3000
turer. The procedure continues until there will be no produc-
J5.M4.P1 0 5000 0 0
tion shortages.
J5.M4.P2 0 0 1900 0
The coordination mechanism is proposed in nine steps as
follows:
J5.M5.P3 3000 0 1900 2600
J5.M2.P4 0 0 2500 0
Step1. Generate initial distribution plan.
J5.M5.P4 0 3900 0 0
Step2. Compute initial demand d initial
kpt .
Step3. Generate production and distribution plan from the
manufacturer’s model.
4 Coordination mechanism Step4. If there is production shortage in manufacturer’s
model, go to step 5; otherwise, terminate the optimi-
Information sharing in the decentralized model is mini- zation procedure.
ð2Þ
mal, with only quantity of product and demand of cus- Step5. Compute Qjmpt .
tomers shared between the manufacturer and retailer. The Step6. Counting iteration: iteration number is the step of the
decision-makers do not display opportunistic behavior. manufacturer and retailer solving the model when we
Neither decision-maker takes the other’s interest into ac- find out that a possible solution satisfies the demand
count, and tries to optimize his own objective functions, and capacity constraint.
Int J Adv Manuf Technol

Table 11 Quantity of product p ð3Þ ð3Þ ð3Þ ð3Þ


transferred from distributor j to Qjknpt T1 Qjknpt T2 Qjknpt T3 Qjknpt T4
customer k by vehicle n in period t
in centralized model
J1.K3.N3.P1 1150 J1.K8.N3.P3 2219 J1.K1.N7.P1 1300 J1.K4.N3.P1 1600
J1.K6.N4.P1 1000 J1.K1.N4.P3 2190 J1.K2.N6.P1 2200 J1.K9.N7.P1 1100
J1.K8.N8.P1 2000 J1.K6.N9.P3 1500 J1.K4.N2.P1 900 J1.K6.N5.P2 2100
J1.K1.N5.P2 2230 J1.K9.N7.P3 1000 J1.K8.N3.P2 1090 J1.K10.N2.P2 1140
J1.K4.N10.P2 1430 J1.K4.N2.P3 790 J1.K7.N1.P2 1600 J1.K2.N4.P2 750
J1.K7.N9.P2 1000 J2.K10.N6.P2 2000 J1.K10.N5.P2 600 J1.K1.N1.P3 2320
J2.K2.N1.P1 1000 J2.K9.N1.P2 1000 J1.K8.N2.P4 1300 J1.K6.N8.P3 860
J2.K5.N5.P1 950 J2.K1.N5.P2 700 J1.K6.N8.P4 1400 J1.K10.N6.P3 370
J2.K7.N6.P1 832 J2.K2.N4.P3 1000 J1.K1.N4.P4 1100 J1.K9.N3.P4 1250
J2.K8.N4.P2 1020 J2.K7.N8.P3 700 J1.K2.N9.P4 500 J1.K5.N9.P4 1550
J2.K6.N3.P2 2150 J2.K4.N9.P4 2000 J2.K10.N3.P1 1000 J1.K4.N10.P4 1400
J2.K9.N2.P2 900 J2.K8.N10.P4 1200 J2.K6.N4.P1 1110 J2.K9.N2.P2 1100
J2.K1.N7.P3 1000 J2.K3.N2.P4 800 J2.K7.N10.P1 1000 J2.K8.N5.P2 1240
J2.K2.N5.P3 982 J3.K1.N8.P1 1500 J2.K3.N2.P1 400 J2.K7.N7.P2 1320
J2.K10.N10.P3 590 J3.K6.N2.P1 1000 J2.K4.N8.P1 400 J2.K4.N8.P2 1125
J2.K5.N4.P3 501 J4.K9.N7.P1 1000 J2.K1.N1.P3 1100 J3.K5.N1.P1 2101
J3.K9.N1.P1 1500 J4.K4.N1.P1 1500 J2.K5.N5.P3 1000 J3.K3.N4.P1 1600
J3.K10.N6.P1 1500 J4.K5.N5.P3 800 J2.K3.N4.P3 1700 J3.K2.N8.P1 1900
J3.K8.N2.P3 2000 J4.K6.N7.P3 2000 J2.K10.N3.P2 500 J3.K1.N3.P2 1000
J3.K7.N3.P3 2000 J4.K10.N3.P3 700 J3.K8.N9.P3 1300 J4.K8.N6.P1 2300
J3.K5.N8.P3 455 J4.K7.N8.P3 100 J3.K7.N4.P3 1200 J4.K9.N7.P3 1200
J3.K4.N9.P4 2000 J4.K9.N2.P4 1608 J3.K3.N1.P3 1100 J4.K8.N2.P3 1300
J3.K6.N5.P4 1000 J4.K10.N4.P4 1100 J3.K10.N9.P4 1100 J4.K4.N9.P3 1000
J3.K9.N3.P4 1000 J5.K10.N1.P1 1600 J3.K4.N4.P4 1590 J4.K2.N10.P3 400
J4.K3.N4.P3 900 J5.K2.N7.P1 1220 J3.K9.N5.P4 461 J4.K6.N5.P4 1000
J4.K9.N4.P3 900 J5.K6.N10.P1 1000 J3.K8.N3.P1 1600 J4.K3.N3.P4 1500
J4.K6.N10.P3 1150 J5.K3.N6.P1 1200 J3.K3.N4.P1 1200 J4.K7.N1.P4 800
J4.K4.N1.P3 650 J5.K7.N4.P1 500 J4.K6.N7.P2 2100 J5.K1.N6.P2 400
J4.K1.N3.P4 1300 J5.K5.N3.P4 1600 J4.K5.N5.P2 1190 J5.K2.N7.P2 900
J4.K3.N2.P4 1200 J5.K6.N7.P4 1350 J5.K4.N4.P2 1209 J5.K2.N5.P2 1100
J4.K5.N7.P4 1000 J5.K2.N8.P4 1180 J5.K5.N3.P2 1000 J5.K10.N4.P3 600
J4.K10.N5.P4 1100 J5.K2.N2.P2 950 J5.K5.N10.P3 1000
J5.K2.N10.3 1100 J5.K1.N1. P2 1000 J5.K10.N2.P4 1000
J5.K3.N7.P3 1100 J5.K3.N9.P2 950 J5.K8.N8.P4 900
J5.K5.N8.P2 1000 J5.K10.N1.P3 500 J5.K7.N9.P4 400
J5.K9.N3.P2 600 J5.K9.N2.P3 900 J5.K2.N1.P4 600
J5.K2.N1.P3 2100 J5.K3.N5.P3 200 J5.K2.N1.P3
J5.K3.N3.P2 1000 J5.K4.N10.P3 800 J5.K3.N3.P2
J5.K9.N5.P4 1300 J5.K5.N4.P4 1100 J5.K9.N5.P4
J5.K7.N3.P4 1000 J5.K3.N8.P4 1100 J5.K7.N3.P4
J5.K8.N7.P4 1000 J5.K2.N9.P4 500 J5.K8.N7.P4
J5.K2.N10.P4 755 J5.K9.N1.P4 132 J5.K2.N10.P4

Step7. Generate sales plan from retailer’s model. The distributor receives the order quantity from the cus-
Step8. If the core demands can be met go to step 8, otherwise tomer and, based on the order quantity, declares the amount
terminate the optimization procedure. of demand to the manufacturer (step 2 in the first iteration and
Step9. Calculate request quantity dkpt and go to step 3. step 9 in the subsequent iterations). The manufacturer, after
Int J Adv Manuf Technol

Table 12 Quantity of product p manufactured by the factory in period t Table 13 Quantity of product p transferred from the factory to
in decentralized model distribution center j in decentralized model
ð1Þ0 ð2Þ
Qpt t1 t2 t3 t4 Qjmpt T1 T2 T3 T4

P1 11,600 10,750 9,950 10,000 J1.M1.P2 4000 0 3100 0


P2 11,500 13,100 12,100 14,700 J1.M1.P4 0 0 3900 4200
P3 11,100 10,500 8,900 9,100 J1.M2.P3 0 4400 0 3100
P4 11,400 10,200 9,400 10,300 J1.M3.P1 4000 0 0 2500
J1.M4.P3 0 2900 0 0
J1.M5.P1 0 0 3700 0
J1.M5.P2 0 0 0 3600
receiving the amount of demand (dkpt), according to the con- J2.M1.P3 3100 0 0 0
straint (29) in the producer model, which is the relationship J2.M2.P1 2600 0 2600 0
ð2Þ J2.M2.P4 0 4000 0 0
between SHkpt, dkpt, and Qjmpt , informs the amount of the
J2.M2.P3 0 1700 3900 0
product of type p that can deliver to the distributor, and also
J2.M3.P2 0 3600 0 0
calculates the amount of shortage. If, according to the con-
ð2Þ J2.M4.P1 0 0 3700 0
straint (29), the amount of dkpt is equal to Qjmpt ,
then there is
J2.M5.P2 4000 0 0 4000
no shortage, and the producer exactly delivers the amount of J3.M1.P2 0 4700 0 0
demand to the distributor and the algorithm ends. If the J3.M1.P1 0 0 0 3500
amount of the shortage is positive, the manufacturer will con- J3.M2.P1 0 2400 0 0
tinue the algorithm in order to fulfill the amount of shortage
J3.M3.P2 0 0 0 3000
(step 4). The main reason for this is the information security so
J3.M3.P3 4600 0 3000 0
that the manufacturer does not want the other components of
J3.M4.P1 3000 0 0 0
the supply chain, including the distributor to know his value of
J3.M5.P2 0 0 0 2100
the objective function, his constraints, and variables of his
J3.M5.P4 3900 0 2900 0
basic decision.
J4.M1.P1 2000 0 0 2000
J4.M2.P2 0 2000 4100 0
J4.M3.P1 0 3350 0 0
5 Computational results
J4.M4.P3 3500 0 0 0
J4.M5.P3 0 1500 0 3400
In the general logistic network, there is a set of five suppliers
J4.M5.P1 0 0 2550 0
that provide three types of raw materials with diverse sizes for
J4.M5.P4 4000 2500 3200
PUYA hardboard factory. Four types of products are
J5.M1.P2 0 2800 2900 0
manufactured in the factory and transferred to the five distri-
bution centers in order to satisfy the demand of 10 major J5.M2.P2 3500 0 0 2000
customers. J5.M3.P4 3500 0 0 2900
In the hardboard production industry, processing time, J5.M4.P1 0 5000 0 0
setup time, inspection time, load time, and other produc- J5.M4.P2 0 0 2000 0
tion times can be considered. But, in this study, because J5.M5.P3 3000 0 2000 2600
the production times are shorter compared to the distribu- J5.M2.P4 0 0 2600 0
tion times, the production times are also considered within J5.M5.P4 0 3700 0 0
ð2Þ
the distribution time of time j . Furthermore, the case
study was considered in four periods. Some parameters
are reported in Tables 2, 3, 4, 5, 6, 7, and 8, and others
are mentioned afterwards. percentile gap is 6.3%). This indicates that high production
Average percentile gap in the objective value between the capacity of the manufacturer constrains the feasible area for
CM and the DM and the average number of iterations are production planning and makes retailer that has enough dis-
1.2% and eight iterations, respectively. tribution capacity able to reach an optimized distribution plan
The worst solution is obtained when a demand pattern is that might be distant from the optimal distribution plan.
increasing, the production capacity of the manufacturer is The convergence graphs using three different costs from
high, and the storage capacity of the retailer is low (the Cretailer, Cmanufacturer, and CTotal are shown in Fig. 3.
Int J Adv Manuf Technol

Table 14 Quantity of product p ð3Þ ð3Þ ð3Þ ð3Þ


transferred from distributor j to Qjknpt T1 Qjknpt T2 Qjknpt T3 Qjknpt T4
customer k by vehicle n in period
tin decentralized model
J1.K3.N3.P1 1000 J1.K8.N3.P3 2000 J1.K1.N7.P1 1000 J1.K4.N3.P1 1500
J1.K6.N4.P1 1000 J1.K1.N4.P3 2000 J1.K2.N6.P1 2000 J1.K9.N7.P1 1000
J1.K8.N8.P1 2000 J1.K6.N9.P3 1500 J1.K4.N2.P1 700 J1.K6.N5.P2 2000
J1.K1.N5.P2 2000 J1.K9.N7.P3 1000 J1.K8.N3.P2 1000 J1.K10.N2.P2 1000
J1.K4.N10.P2 1000 J1.K4.N2.P3 800 J1.K7.N1.P2 1500 J1.K2.N4.P2 600
J1.K7.N9.P2 1000 J2.K10.N6.P2 2000 J1.K10.N5.P2 600 J1.K1.N1.P3 2000
J2.K2.N1.P1 800 J2.K9.N1.P2 1000 J1.K8.N2.P4 1300 J1.K6.N8.P3 800
J2.K5.N5.P1 1000 J2.K1.N5.P2 600 J1.K6.N8.P4 1200 J1.K10.N6.P3 300
J2.K7.N6.P1 800 J2.K2.N4.P3 1000 J1.K1.N4.P4 1000 J1.K9.N3.P4 1200
J2.K8.N4.P2 1000 J2.K7.N8.P3 700 J1.K2.N9.P4 400 J1.K5.N9.P4 1500
J2.K6.N3.P2 2000 J2.K4.N9.P4 2000 J2.K10.N3.P1 1000 J1.K4.N10.P4 1500
J2.K9.N2.P2 1000 J2.K8.N10.P4 1200 J2.K6.N4.P1 1000 J2.K9.N2.P2 1000
J2.K1.N7.P3 1000 J2.K3.N2.P4 800 J2.K7.N10.P1 1000 J2.K8.N5.P2 1000
J2.K2.N5.P3 1000 J3.K1.N8.P1 1400 J2.K3.N2.P1 300 J2.K7.N7.P2 1000
J2.K10.N10.P3 800 J3.K6.N2.P1 1000 J2.K4.N8.P1 300 J2.K4.N8.P2 1000
J2.K5.N4.P3 300 J4.K9.N7.P1 1000 J2.K1.N1.P3 1000 J3.K5.N1.P1 2000
J3.K9.N1.P1 1500 J4.K4.N1.P1 1500 J2.K5.N5.P3 1000 J3.K3.N4.P1 1500
J3.K10.N6.P1 1500 J4.K5.N5.P3 800 J2.K3.N4.P3 1500 J3.K2.N8.P1 2000
J3.K8.N2.P3 2000 J4.K6.N7.P3 2000 J2.K10.N3.P2 400 J3.K1.N3.P2 1100
J3.K7.N3.P3 2000 J4.K10.N3.P3 600 J3.K8.N9.P3 1000 J4.K8.N6.P1 2000
J3.K5.N8.P3 300 J4.K7.N8.P3 100 J3.K7.N4.P3 1000 J4.K9.N7.P3 1000
J3.K4.N9.P4 2000 J4.K9.N2.P4 1500 J3.K3.N1.P3 1000 J4.K8.N2.P3 1000
J3.K6.N5.P4 1000 J4.K10.N4.P4 1000 J3.K10.N9.P4 1000 J4.K4.N9.P3 1000
J3.K9.N3.P4 900 J5.K10.N1.P1 1500 J3.K4.N4.P4 1500 J4.K2.N10.P3 400
J4.K3.N4.P3 1000 J5.K2.N7.P1 1000 J3.K9.N5.P4 400 J4.K6.N5.P4 1000
J4.K9.N4.P3 1000 J5.K6.N10.P1 1000 J3.K8.N3.P1 1500 J4.K3.N3.P4 1500
J4.K6.N10.P3 1000 J5.K3.N6.P1 1200 J3.K3.N4.P1 1000 J4.K7.N1.P4 700
J4.K4.N1.P3 500 J5.K7.N4.P1 300 J4.K6.N7.P2 2000 J5.K1.N6.P2 400
J4.K1.N3.P4 1000 J5.K5.N3.P4 1500 J4.K5.N5.P2 1100 J5.K2.N7.P2 900
J4.K3.N2.P4 1000 J5.K6.N7.P4 1200 J5.K4.N4.P2 1000 J5.K2.N5.P2 1100
J4.K5.N7.P4 1000 J5.K2.N8.P4 1000 J5.K5.N3.P2 900 J5.K10.N4.P3 500
J4.K10.N5.P4 1000 J5.K2.N2.P2 1000 J5.K5.N10.P3 1000
J5.K2.N10.3 1000 J5.K1.N1. P2 1000 J5.K10.N2.P4 1000
J5.K3.N7.P3 1000 J5.K3.N9.P2 1000 J5.K8.N8.P4 800
J5.K5.N8.P2 1000 J5.K10.N1.P3 400 J5.K7.N9.P4 300
J5.K9.N3.P2 500 J5.K9.N2.P3 800 J5.K2.N1.P4 500
J5.K2.N1.P3 2000 J5.K3.N5.P3 200 J5.K2.N1.P3
J5.K3.N3.P2 1000 J5.K4.N10.P3 600 J5.K3.N3.P2
J5.K9.N5.P4 1100 J5.K5.N4.P4 1000 J5.K9.N5.P4
J5.K7.N3.P4 1000 J5.K3.N8.P4 1000 J5.K7.N3.P4
J5.K8.N7.P4 800 J5.K2.N9.P4 500 J5.K8.N7.P4
J5.K2.N10.P4 600 J5.K9.N1.P4 100 J5.K2.N10.P4

The most important observations come from the iterations. As shown in Fig. 3, Cretailer has decreased since the initial
In some cases, one iteration is only required because it is dkpt value of retailer’s distribution planning model is adjusted
optimal for the manufacturer to deliver everything requested. by the available supply quantity of manufacturer as planning
This coincides with an increasing pattern of demand behavior iterations. In other words, the initial distribution plan of retail-
and high production capacity. er assuming infinite production input has changed into a more
Int J Adv Manuf Technol

Table 15 Quantity of raw material r transferred from supplier i to the Table 15 (continued)
factory in period t n decentralized model
t1 t2 t3 t4
t1 t2 t3 t4
i5.l10.r4 1800 2000 2500 1400
i1.l1.r1 1800 2000 2500 1400 i5.l10.r5 2000 3500 1000 1500
i1.l1.r2 2000 3500 1000 1500
i1.l1.r3 1000 2000 1000 2000
i1.l1.r4 2000 2000 2000 2000
i1.l1.r5 1000 1500 1500 1500 realistic one overall planning iteration. Conversely,
i1.l2.r1 1000 1000 0 0 Cmanufacturer has been increased since retailer sends a more
i1.l2.r2 3000 1600 2200 2500 realistic request considering the available supply quantity of
i1.l2.r3 1000 1500 2000 1500 manufacturer as planned iterations proceed.
i1.l2.r4 2500 2500 2500 2500 Table 9 shows the amount of product produced in the fac-
i1.l2.r5 1500 1500 1500 2000 tory using a centralized model. As shown in this table, for
i2.l3.r1 800 1000 1500 1500 example, product 2 will be produced in period 1 at 11,617 units
i2.l3.r2 1000 1000 900 800 and product 3 will be produced in period 4 at 10,521 units.
i2.l3.r3 1000 1200 1300 1500 Table 10 shows the amount of product shipped from the
i2.l3.r4 2500 2500 2500 2500 factory to the distributors using a centralized model. As shown
i2.l3.r3 2500 2500 2500 2500 in this table, for example, distributor 1 carries 3900 units of
i2.l4.r1 1000 1000 1000 2000 product 4 by car 1 in the third period.
i2.l4.r2 1800 2000 1800 2000 Table 11 also shows the amount of product shipped from
i2.l4.r3 2500 2500 2000 2500 distributors to customers using a centralized model. As shown
i2.l4.r4 1000 1500 1000 1200 in this table, for example, distributor 1 delivers 2000 units of
i2.l4.r5 1200 1200 1200 1200 product 1 by car 8 to customer 8 in the third period.
i3.l5.r1 1000 1000 1000 1000 Table 12 shows the amount of product produced at the
i3.l5.r2 2000 2000 2000 2000 factory using the decentralized model. As shown in this table,
i3.l5.r3 1500 1500 1500 800 for example, product 2 will be produced in period 1 at
i3.l5.r4 1300 1300 1200 1200 11,500 units and product 3 will be produced in period 4 at
i3.l5.r5 1500 2000 1500 2000 9100 units.
i3.l6.r1 1000 1200 1200 1000 Table 13 shows the amount of product shipped from the
i3.l6.r2 800 800 800 800 factory to the distributors using a decentralized model. As
i3.l6.r3 3000 3000 2500 2500 shown in this table, for example, distributor 1 carries
i3.l6.r4 2000 2000 2000 2000 4200 units of product 4 by car 1 in the fourth period.
i3.l6.r5 1000 1000 1000 1000
Table 14 also shows the amount of product shipped from
i4.l7.r1 3000 3000 3000 3000
distributors to customers using a decentralized model. As
i4.l7.r2 3500 3500 3500 3500
shown in this table, for example, distributor 1 delivers
2000 units of product 1 by car 8 to customer 8 in the third
i4.l7.r3 3000 2800 2800 2500
period.
i4.l7.r4 3000 1600 2200 2500
Table 15 shows the quantity of raw material r transferred
i4.l7.r5 1000 1500 2000 1500
from supplier i to the factory in period (Qilrt).
i4.l8.r1 1000 1000 1000 1000
According to the comparison of Tables 9, 10, 11, 12, 13,
i4.l8.r2 2000 2000 1000 1000
and 14, almost all the results of the variables in the centralized
i4.l8.r3 1200 1200 1300 1200
model, as expected, were equal or better than the decentralized
i4.l8.r4 2500 2500 2500 2500
model.
i4.l8.r5 1000 1000 1000 2000
Table 16 shows the movement of vehicles from factory to
i5.l9.r1 2000 1500 1500 1500
distributers. According to the data of the problem, there are
i5.l9.r2 2500 2500 2500 2500
four distribution centers. Also, five vehicles carry products
i5.l9.r3 3000 2500 3000 2500
from factory to distributers in each period. For example, ac-
i5.l9.r4 1000 1000 1000 1000
cording to Table 16, in the first period, vehicle 3 moves toward
i5.l9.r5 2000 2000 2000 2000
the first distributor, and in the third period, vehicle 1 moves
i5.l10.r1 3000 3000 3000 3000 toward the fifth distributor.
i5.l10.r2 1500 1500 1500 1500 Table 17 shows the movement of vehicles from the supplier
i5.l10.r3 1200 1500 1500 1500 to the factory. According to the data of the problem, there are
Int J Adv Manuf Technol

Table 16 Launched vehicle ð2Þ Table 17 Launched vehicle moved from supplier to the factory
moved from the factory to the Z jmt T1 T2 T3 T4
distributors Zilt L1 L2 L3 L4 L5 L6 L7 L8 L9 L10

J1.M1 1 0 1 0 I1.t1 1 1 – – – – – – – –
J1.M2 0 1 0 0 I1.t2 1 0 – – – – – – – –
J1.M3 1 0 0 1 I1.t3 1 1 – – – – – – – –
J1.M4 0 1 0 0 I1.t4 1 1 – – – – – – – –
J1.M5 0 0 1 1 I2.t1 – – 1 0 – – – – – –
J2.M1 1 0 0 1 I2.t2 – – 0 1 – – – – – –
J2.M2 1 1 1 0 I2.t3 – – 1 1 – – – – – –
J2.M3 0 1 0 0 I2.t4 – – 1 1 – – – – – –
J2.M4 0 0 1 0 I3.t1 – – – – 1 1 – – – –
J2.M5 1 0 0 1 I3.t2 – – – – 1 1 – – – –
J3.M1 0 1 0 1 I3.t3 – – – – 0 1 – – – –
J3.M2 0 1 0 1 I3.t4 – – – – 0 1 – – – –
J3.M3 1 0 1 0 I4.t1 – – – – – – 1 0 – –
J3.M4 1 0 0 1 I4.t2 – – – – – – 1 1 – –
J3.M5 1 0 1 0 I4.t3 – – – – – – 1 1 – –
J4.M1 1 0 0 1 I4.t4 – – – – – – 1 1 – –
J4.M2 0 0 1 0 I5.t1 – – – – – – – – 0 0
J4.M3 0 1 0 0 I5.t2 – – – – – – – – 1 1
J4.M4 1 0 0 0 I5.t3 – – – – – – – – 1 1
J4.M5 1 0 0 1 I5.t4 – – – – – – – – 0 1
J5.M1 0 0 1 0
J5.M2 1 0 0 1
J5.M3 1 0 0 1
materials is received by the factory after 1.5 days equal to
J5.M4 0 1 1 0
36 h. So, the amount of lead time for all the raw materials is
J5.M5 0 1 0 1
36 h. Also, the amount of average daily sales rate can be
calculated according to Table 4. As is clear, the amount for
each product will change according to the time periods. So,
the amount of average daily sales rate will change according to
10 vehicles to carry raw materials from supplier to factory. the time period and the amounts of ROP and buffer will also
Also, two vehicles are considered for each supplier. Vehicles
1 and 2 are for supplier 1, vehicles 3 and 4 are for supplier 2,
Table 18 ROP and
vehicles 5 and 6 are for supplier 3, vehicles 7 and 8 are for Product ROP Buffer
buffer
supplier 4, and vehicles 9 and 10 are for supplier 5. For ex-
ample, according to Table 17, in the first period, vehicle 5 P1.t1 37,500 4,300
moves from the third retailer, and in the fourth period, vehicle P2.t1 36,000 4,000
8 moves from the fourth retailer in order to carry raw materials P3.t1 38,450 4,260
toward the factory. P4.t1 41,000 4,390
In manufacturing, the concept of buffering is defined as P1.t2 38,660 4,330
maintaining enough supplies to keep operations running P2.t2 36,550 4,050
smoothly. These supplies often include the raw materials P3.t2 38,900 4,290
needed for production, and also the inventories of finished P4.t2 42,500 4,420
products waiting for shipment. Too much buffer stock can P1.t3 39,120 4,350
result in high inventory carrying costs. Too less stock can P2.t3 36,760 4,200
cause repeated occurrences of stock-outs. Also, increase in P3.t3 39,200 4,350
buffer size increases inventory holding cost and storage facil- P4.t3 43,050 4,470
ity cost. This inventory cost decreases net profit. Hence, busi- P1.t4 39,860 4,400
nesses need to maintain a fine balance and decide on the P2.t4 37,000 4,250
amount of buffer inventory to be held. P3.t4 40,000 4,390
According to the situation of the factory and suppliers and P4.t4 43,990 4,490
the agreement that exists between them, the order of raw
Int J Adv Manuf Technol

Table 19 optimal solutions of the


centralized and decentralized Solution no. Decentralized objective Absolute gap Gap% CPU time (s)
model
1 895,796,671 423,479,843 0.472741 3s
2 625,974,611 153,657,783 0.24547 3s
3 534,137,259 61,820,431 0.115739 5s
4 525,937,138 53,620,310 0.101952 5s
5 494,438,592 22,121,764 0.044741 7s
6 494,048,930 21,732,102 0.043988 18 s
7 484,372,190 12,055,362 0.024889 49 s
8 484,305,007 11,988,179 0.024753 54 s
9 475,099,583 2,782,755 0.005857 58 s
Average 557,123,331 84,806,503 0.120014 22.44

Centralized method solution time 14 s. Centralized method objective function 472,316,828

change according to the type of the product and the time difference from the objective function of the centralized model
period of the product. It should be noted that the amounts of is selected as the best objective function in the decentralized
ROP and buffer should always be less than the amount of model.
capp = 150,000 (product storage capacity of the factory). The values of the gap for these two objective functions
Equation 45 shows how to calculate the amount of buffer were 0.47 and 0.005, respectively. Also, the solving time,
which varies with the number of iterations, indicates that the
Buffer ¼ ROP−ðaverage daily sales rate  lead timeÞ ð45Þ
more the solutions are closer to the optimal answer and the
Table 18 shows the optimum amounts of ROP and buffer. lower the values of the gap are, the greater the solving time
These amounts create an optimum balance between mainte- will be. The average objective function of the decentralized
nance costs and shortages in each period. model is 557,123,331 and the average solving time is 22.44 s.
According to Table 19, the value of the objective function As expected, the solution of the centralized model is far
of the centralized model is 472,316,828 and the time of solv- better than the solutions of the decentralized model. Also,
ing this model is equal to 14 s. The various values of the the time of solving a centralized model, because of being
objective functions of the decentralized model by the effect single stage, is better than the average of solving times of
of various iterations are shown in the second column of this the decentralized model.
table. These values are given in the table from the worst to the
best. The value of the absolute gap, which includes the differ-
ence between the values of the objective function of the 6 Sensitivity analysis
decentralized model and the centralized model, appears in
the third column. Also, the values of the gap in percent and In this section, we investigate the Puya Company problem
the solving time in seconds are given in the fourth and fifth again to illustrate the solutions’ sensitivity in terms of impor-
columns. Accordingly, the objective function of 895,796,671 tant and changeable parameters of the problem. We try to
as the worst objective function in the decentralized model and demonstrate how changing these parameters may affect the
the objective function of 475,099,583 with respect to its little network structure of this chain. This also validates the

70000000
70000000
60000000
60000000
50000000
50000000
40000000
Cost

40000000
Cost

Retailer Cost
Retailer Cost
30000000 30000000
Manufacturer Cost Manufacturer Cost
20000000 20000000
Total Cost Total Cost
10000000 10000000
0 0
-30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 30
% change from initial value %change frome initial value

Fig. 4 The influence of vcir on cost Fig. 5 The influence of vcjp(2) on cost
Int J Adv Manuf Technol

800000000 the factory to the distributor increases distributor cost and


700000000 manufacturer cost. As can be seen, the slope of the diagram
600000000
500000000
of manufacturer is more than the distributor, and by increasing
Cost

400000000 Retailer cost the transportation cost from the factory to the distributor, the
300000000 Manufacturer cost manufacturer cost has increased from 11,110,000 up to
200000000 29,415,500 and distributor cost has increased from
Total cost
100000000
28,189,900 up to 35,105,000.
0
-30 -20 -10 0 10 20 30 Also, we incrementally increased the transportation cost of
% change from initial value product from distributor to customer from −30 to +30%. By
doing so, the total cost starts to rise. Other parameters of the
Fig. 6 The influence of vcjkp(3) on cost model are assumed to be fixed.
It is observed from Fig. 6 that the range of the objectives
correctness of the proposed model because it reacts in a rea- increases consistently with the increase in vcjkp(3). Regarding
sonable way to the parameter changes. transportation cost of product from distributor to customer
We perform sensitivity analyses in this section. In this anal- (see Fig. 6), the objective function in manufacturer model
ysis, the influence of transportation cost on the general model, generates higher cost than the retailer model due to the change
manufacturer’s model, and retailer’s model is explored. The in transportation cost level. Also, the slope of retailer cost
results of the three analyses are presented in Fig. 4. graph is greater than manufacturer graph. This is because the
We incrementally increased the transportation cost of retailer’s model has vcjkp(3) parameter in its objective function,
raw material from supplier to factory from −30 to +30%. and it implies more transportation of product from distributor
This cost includes vcir. The rest of the parameters have been to customer.
considered fixed. As can be seen, by increasing the trans-
portation cost of raw material, the costs of manufacturer
and distributor will both increase, but the slope of the graph 7 The analysis of relationship between the percentile
of manufacturer is greater than the slope of the graph of gap and the number of iterations
distributor. With 30% tolerance in transportation cost, the
costs of the manufacturer have been increased from As shown in Fig. 7, the number of iterations is greatly
11,374,000 to 27,415,000 and the costs the distributer have affected by the percentile gap, and this is also intuitive
been increased from 30,150,000 to 34,105,000. This shows since large size of data exchanged might require more
that by increasing the transportation cost, the costs of the intensive communications to reach coordination. In order
manufacturer will increase more comparing to the distrib- to view the comparison in this graph better, we multiplied
utor and the manufacturer will be affected more. The reason the average percentile gap by 100. The two most impor-
for this change can be found in the manufacturer’s objec- tant observations come from the iterations. In some cases,
tive function (Eq. (19.2)). This part of manufacturer’s ob- one iteration is only required because it is optimal for the
jective function is directly related to the transportation cost. manufacturer to deliver everything requested. This hap-
We incrementally increased the transportation cost of prod- pens with a small percent of gap. In most of the other
uct from the factory to the distributor from −30 to +30%. runs, two iterations are enough to generate a feasible so-
According to Fig. 5, an increase in transportation cost from lution to the problem. The retailers never order less than
the available supply quantities. This is due to the decom-
position choices made for the centralized model, which
100
requires some further discussion.
80
In summary, the proposed coordination provides very
good solutions in reasonable number of iterations for this
% Gap *100

60 problem using in decentralized planning model with only


Cenralized-
Decentralized minimal information sharing. Also, the proposed coordi-
40
Manufacturer-Retailer nation mechanism is always terminated with a feasible
20 solution plan.
Two models of manufacturer and distributor interact
0 with each other. The distributor model puts the amount
1 2 3 4 5 6 7 8 9
Iteration
of demand into the manufacturer’s model. In the state of
the absence of a shortage, the algorithm will end, but if
Fig. 7 Relationship between the percentile gap and the number of there is a shortage, the model repeats to a point where the
iterations shortage does not occur and the demand is satisfied.
Int J Adv Manuf Technol

Suppose that in the first iteration of the model, SHkpt ≥0, & Extension of both CDM and DDM approaches in the
then the quantity of product p transferred from the factory hardboard production industry.
to the distribution center will be calculated. This value is & Introducing a coordination mechanism in the proposed
calculated by the constraint (29) in the manufacturer’s DDM approach in order to handle problems related to
model. In fact, this difference in demand is caused by production and distribution in a multi-echelon supply
the shortage. In the second iteration, the manufacturer’s chain network.
model should satisfy the demand difference, and the un- & Real data is used to show the application of model and
satisfied demand (deficiency) in the manufacturer’s model validation of the model results.
by section (Eq. (19.6)) in the objective function will lead
to worsening the model compared to the first iteration. Average percentile gap in the objective value between the
Therefore, each round worsens the model due to increas- CM and the DM and the average number of iterations are
ing the amount of shortage and its effect on the objective 1.2% and 8, respectively. An integrated optimization model
function of the manufacturer. In the case of the distributor has been formulated as an MILP model via GAMS software.
model, due to the shortage in all iterations, the product is Computation times were all very extremely long. One of
shipped less than the ideal model. So, maintenance costs the runs of the manufacturer’s model took 933 s. These did not
become lower and fewer vehicles move toward customers, occur for the same iteration again, which seems to suggest that
resulting in reduced costs for the distributor. As a result, some combinations of data make the problem more difficult to
due to the reduction in distribution costs and the increase solve optimally, because solutions are closer together. The
in the production cost, the gap between production cost worst solution is obtained when a demand pattern is increas-
and distribution cost increases. ing, the production capacity of the manufacturer is high, and
Also, the results obtained from the centralized model are the storage capacity of the retailer is low (the percentile gap is
considered to be optimum solutions and the decentralized 6.3%). The proposed coordination provides very good solu-
model will never have better answers than the centralized tions in reasonable number of iterations for this problem using
model due to the addition of constraints, variables, and com- in decentralized planning model with only minimal informa-
plexity of the solution process. It is therefore obvious that in tion sharing. Also, the proposed coordination mechanism is
each of the iterations, the decentralized model tries to get always terminated with a feasible solution plan.
closer to the centralized model and be better. Therefore, with Therefore, new strategies and policies could be taken into
increasing number of iterations, the gap between the central- account by the Puya Company to sustain their competitive
ized and decentralized model decreases. position.
An interesting direction for the future is to examine the
solutions given from centralized and decentralized method
8 Conclusion under different objective functions, such as maximization
of profit and minimization of the inventory. Also, future
This paper provides a decentralized supply chain planning researches could focus on development of efficient profit
framework for Puya Wood Company. Especially, only distribution strategies in case of centralized optimization
minimal-information sharing between the manufacturer and and may investigate the application of meta-heuristic ap-
the retailer is considered as the most important factor to im- proaches and other soft computing techniques in order to
plement fully decentralized planning environment. We have handle large-scale problems. The separation of operations
explored a “locally optimized” supply chain procedure of and production details of hardboard in centralized and
Puya Company. Then, we compared this decentralized opti- decentralized models can be considered in future research.
mized model with centralized optimization where a single For example, joinery can be considered in manufacturing
decision-maker plans the production quantities of the supply operations because in the joinery sector, several tons of
chain to minimize total costs. wood are processed every month. Moreover, this type of
The contribution of this paper is presenting a decentralized planning can be applied in other supply chains such as
(distributed) approach in order to address the production and natural gas and oil, water resource management, and crisis
distribution problem in supply chain including manufacturer and disaster planning.
and retailers. The main highlights of this paper can be sum-
marized as follows:

& Proposing a DDM approach for a multi-product supply


Acknowledgements The authors would like to thank the anonymous
chain planning problem while the production and distri-
reviewers and the Editor for their constructive comments and suggestions.
bution problems in supply chain are addressed indepen- The authors acknowledge the support from Puya Hardboard Company in
dently in manufacturer and retailers. Mazandaran Province in Iran.
Int J Adv Manuf Technol

References 23. Spitter JM, Hurkens CAJ, de Kok AG, Lenstra JK, Negenman EG
(2005) Linear programming models with planned lead times for
supply chain operations planning. Eur J Oper Res 163:706–720
1. Baghalian A, Rezapour SH, Zanjirani Farahani R (2013) Robust 24. Lin, J. Chen, Y. (2004) A supply network planning problem in a
supply chain network design with service level against disruptions multi-stage and multi-site environment. Presented at the 35th
and demand uncertainties: a real-life case. Eur J Oper Res 227:199– International Conference on Computers and Industrial Engineering
215 25. López De Lacalle LN, Fernandez A, Rodriguez A, Lopez R,
2. Lambert DM, Cooper MC (2000) Issues in supply chain manage- Fernadez-Larrinoa J, Azkona I (2014) On the cutting of wood for
ment. Ind Mark Manag 29:65–83 joinery applications. Proc Inst Mech Eng B J Eng Manuf. doi:10.
3. Dudek G, Stadtler H (2007) Negotiation-based collaborative plan- 1177/0954405414534431
ning in divergent two tier supply chains. Int J Prod Econ 45:465– 26. López De Lacalle LN, Lamikiz A, Salgado MA, Herranz S, Rivero
484 A (2002) Process planning for reliable high-speed machining of
4. Verwaal E, Hesselmans M (2004) Drivers of supply network gov- moulds. Int J Prod Res 40(12):2789–2809. doi:10.1080/
ernance: an explorative study of the Dutch chemical industry. Eur 00207540210140068
Manag J 22(4):442–451 27. Song C, Zhuang J (2017) Modeling a government-manufacturer-
5. Lejeune N, Yakova N (2005) On characterizing the 4C’s in supply farmer game for food supply chain risk management. Food Control.
chain management. J Oper Manag 23(1):81–100 doi:10.1016/j.foodcont.2017.02.047
6. Caridi M, Cigolini R, De Marco D (2005) Improving supply-chain 28. Toptal A, Çetinkaya B (2015) How supply chain coordination af-
collaboration by linking intelligent CPFR. Int J Prod Res 43(20): fects the environment: a carbon footprint perspective. Ann Oper
4191–4218 Res. doi:10.1007/s10479-015-1858-9
7. Hegeman J, Peidro D, Alemany M, Díaz-Madroñero M (2014) A 29. Baboli A, Pirayesh M, HAJI R (2008) An algorithm for the deter-
decentralized production and distribution planning model in an un- mination of the economic order quality in a two-level supply chain
certain environment, Studies in Fuzziness and Soft Computing. with transportation costs: comparison of decentralized with central-
Springer-Verlag, Berlin, p 313 ized decision. J Syst Sci Syst Eng 17(3):353–366
8. Jung H, Chen F, Jeong B (2008) Decentralized supply chain plan- 30. Bilir C, Ekici S, Ulengin F (2017) An integrated multi-objective
ning framework for third party logistics partnership. Comput Ind supply chain network and competitive facility location model.
Eng 55(2):348–364 Comput Ind Eng 108:136–148
9. Schneeweiss C (2003) Invited review: distributed decision making- 31. Nematollahi M, Hosseini-Motlagh S, Heydari J (2017)
a unified approach. Eur J Oper Res 150:237–252 Coordination of social responsibility and order quantity in a two-
10. Huang G, Lau J, Mak K (2003) The impacts of sharing production echelon supply chain: a collaborative decision-making perspective.
information on supply chain dynamics: a review of the literature. Int Int J Prod Econ 184:107–121
J Prod Res 41(7):1483–1517 32. Proch M, Worthmann K, Schlüchtermann J (2017) A negotiation-
11. Kumar RS, Pugazhendhi S (2012) Information sharing in supply based algorithm to coordinate supplier development in
chains: an overview. Procedia Eng 38:2147–2154 decentralized supply chains. Eur J Oper Res 256:412–429
12. AkbariJokar R, Frein Y, Dupont L (2002) Évolution du concept de 33. Geunes J, Edwin Romeijn H, Heuvel W (2016) Improving the
logistique. Rev Fr GestionIndustrielle 21(3):5–22 efficiency of decentralized supply chains with fixed ordering costs.
13. Naso D, Surico M, Turchiano B, Kaymak U (2007) Genetic Eur J Oper Res 252:815–828
algorithms for supply-chain scheduling: a case study in the 34. Claudia C, Taraneh S (2016) Incorporating social benefits in multi-
distribution of ready-mixed concrete. Eur J Oper Res 177(3): objective optimization of forest based bioenergy and biofuel supply
2069–2099 chains. Appl Energy 178:721–735
14. Phillips N, Lawrence T, Hardy C (2000) Inter-organizational col- 35. Cambero, C., Sowlati, T., Pavel, (2015) Economic and life cycle
laboration and the dynamics of institutional fields. J Manag Stud environmental optimization of forest-based biorefinery supply
37(1):23–44 chains for bioenergy and biofuel production, Chem Eng Res Des,
15. Seifert D (2003) Collaborative planning, forecasting, and replenish- http://dx.doi.org/10.1016/j.cherd.2015.10.040
ment: how to create a supply chain advantage. AMACOM, New 36. Peter B, Niquidet K (2016) Estimates of residual fibre supply and
York the impacts of new bioenergy capacity from a forest sector trans-
16. Soosay C, Hyland P, Ferrer M (2008) Supply chain collaboration: portation model of the Canadian prairie provinces. Forest Policy
capabilities for continuous innovation. Supply Chain Manag : Int J Econ 69:62–72
13(2):160–169 37. Zhang F, Dana MJ, Jinjiang W (2016) Integrating multimodal trans-
17. Cannella S, Ciancimino E (2010) On the bullwhip avoidance phase: port into forest-delivered biofuel supply chain design. Renew
supply chain collaboration and order smoothing. Int J Prod Res Energy 93:58–67
48(22):6739–6776 38. Alam M, Shahi C, Pulkki R (2014) Economic impact of enhanced
18. Bhatnagar, R. Chandra, P. Goyal, S.K. (1993), Models for multi forest inventory information and merchandizing yards in the forest
plant coordination, Eur J Oper Res 67, 141–160 product industry supply chain. Socio Econ Plan Sci 48:189–197
19. Shin H (2007) Collaborative production planning in a supply-chain 39. Ghaffariyan M, Acuna M, Brown M (2013) Analysing the effect of
network with partial information sharing. Int J AdvManufTechnol five operational factors on forest residue supply chain costs: a case
34:981–987 study in Western Australia. Biomass Bio Energy 59:486–493
20. Pibernik R, Sucky E (2007) An approach to inter-domain master 40. Weintraub, A., Epstein, R. (2002), The supply chain in the forest
planning in supply chains. Int J Prod Econ 108:200–212 industry: models and linkages, chapter, supply chain management:
21. Anderson, C. Bartholdi, J.J. (2000) Centralized versus models, applications, and research directions, volume 62 of the
decentralized control in manufacturing: lessons from social insects. series. Appl Optim 343-362
Complexity and Complex Systems in Industry, The University of 41. Sanei O, Kazemi Zanjani M, Nourelfath M (2016) The value of
Warwick, UK, p 652 integrated tactical planning optimization in the lumber supply
22. Bonabeau E, Dorigo M, Theraulaz G (1999) Swarm intelligence: chain. Int J Prod Econ 171:22–33
from natural to artificial systems. Santa Fe Institute on the Sciences 42. Shabani N, Sowlati T (2015) A hybrid multi-stage stochastic
of Complexity. Oxford University Press, New York programming-robust optimization model for maximizing the
Int J Adv Manuf Technol

supply chain of a forest-based biomass power plant considering 45. Forget P, Amours S, Frayret J (2008) Multi-behavior agent model
uncertainties. J Clean Prod. doi:10.1016/j.jclepro.2015.09.034 for planning in supply chains: an application to the lumber industry.
43. Gaudreault J, Frayret J, Rousseau A, Amours S (2011) Combined Robot Comput Integr Manuf 24:664–679
planning and scheduling in a divergent production system with co- 46. Ozdamar L, Yazgac T (1999) A hierarchical planning approach for
production: a case study in the lumber industry. Comput Oper Res a production-distribution system. Int J Prod Res 37:3759–3772
38:1238–1250 47. Schneeweiss C, Zimmer K, Zimmermann M (2004) The design of
44. Gaudreault, J., Forget, P.,Frayret, J., Rousseau, A., Lemieux, S., contracts to coordinate operational interdependencies within the
Amour, S. (2010) Distributed operations planning in the lumber supply chain. Int J Prod Econ 92:43–59
supply chain: models and coordination. Int J Ind Eng : Theory
Appl Pract 17(3)

Das könnte Ihnen auch gefallen