Sie sind auf Seite 1von 11

International Journal of Production Economics 183 (2017) 710–720

Contents lists available at ScienceDirect

Int. J. Production Economics


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

A robust design for a Colombian dairy cooperative's milk storage and MARK
refrigeration logistics system using binary programming

Rafael Tordecilla-Maderaa, ,1, Andrés Poloa,1, Dairo Muñoza,2, Leonardo González-Rodríguezb,3
a
Innovar Tecnológico Research Group, Industrial Engineering Program, Fundación Universitaria Agraria de Colombia, Calle 170 #54A-10, Bogotá,
Colombia
b
Logistic Systems Research Group (Grupo de Investigación en Sistemas Logísticos), Faculty of Engineering, Universidad de La Sabana, Chía, Colombia

A R T I C L E I N F O A BS T RAC T

Keywords: This article describes a robust design for a milk storage and refrigeration logistics system for a dairy cooperative
Robustness located in the Ubaté region of Colombia using a binary programming model. Such design included cooling tank
Cost location, assigning producers to the tanks and calculating system capacity. This research was aimed at
Binary programming identifying the most robust and cost-efficient configuration (given the design's parameters).
Milk cooling tanks
Perturbation parameters, robustness requirements and performance features were thus qualitatively and
Location
quantitatively identified; the perturbation parameters’ impact on robustness requirements and performance
features was also determined. A supply chain involves a variety of configurations, along with variations in milk
supply (perturbation parameter). The most robust configuration and associated costs were thus identified
through robustness-cost characterization and used for designing the dairy cooperative's logistical storage and
refrigeration system. A tactical plan was also drawn up for optimizing cooling tank use.
Our main theoretical conclusion regarding this matter (depending on how robustness requirements and
perturbation parameters are defined and calculated) suggested that the most robust system does not necessarily
represent higher costs because the most robust configurations were in fact the least costly concerning high milk
supply values.

1. Introduction public goods and services; however, the rural sector has yet to take
suitable advantage of these resources.
According to Colombia's 2011–2014 Development Plan (Prosperity The Colombian dairy sub-sector faces a series of challenges regard-
for all), the country must have an annual 6% socially and environmen- ing its successful entry to the international market and consolidation
tally sustainable growth rate during the aforementioned period within the internal market. The National Council for Economic Policy
(Departamento Nacional de Planeación DNP, 2011). The initiatives (CONPES) has introduced policies for improving this sector's competi-
adopted for achieving this goal included Pacific Alliance and Andean tiveness (Consejo Nacional de Política Económica y Social, 2010).
Community of Nations free trade agreements which were aimed at the CONPES document 3675 was aimed at improving the Colombian dairy
growth of small- and medium-sized businesses’ exports by stimulating sector's competitiveness by developing strategies aimed at reducing
association-based collaboration. Colombia's agricultural sector reflects production costs, increasing productivity, promoting collaborative
its government's commitment to exploiting Colombian plantations’ schemes, and strengthening the sector's institutional administration.
inherent wealth and harnessing their potential for consolidating growth Such policy also sought to improve sanitary and safety measures for
and improving competitiveness, diversifying production, expanding strengthening competitiveness, improving public health, and gaining
markets, and improving farmers’ working conditions and income. The access to national and international markets; analysis has thus been
Colombian government has implemented commercial policies aimed at undertaken at each level in the production chain, including associa-
improving farmers’ income to stimulate internal markets and allocate tions/cooperatives representing the production, storage, transporta-


Corresponding author.
E-mail addresses: tordecilla.rafael@uniagraria.edu.co, rafaeldtm@hotmail.com (R. Tordecilla-Madera), polo.andres@uniagraria.edu.co (A. Polo),
munoz.dairo@uniagraria.edu.co (D. Muñoz), leonardo.gonzalez1@unisabana.edu.co (L. González-Rodríguez).
1
Full-time professor, Fundación Universitaria Agraria de Colombia.
2
Young Researcher, Industrial Engineering Program.
3
Full-time professor, Universidad de La Sabana.

http://dx.doi.org/10.1016/j.ijpe.2016.09.019
Received 30 December 2014; Received in revised form 18 September 2016; Accepted 25 September 2016
Available online 28 September 2016
0925-5273/ © 2016 Elsevier B.V. All rights reserved.
R. Tordecilla-Madera et al. International Journal of Production Economics 183 (2017) 710–720

tion, manufacturing and commercialization sectors, as well as those location problem in which a minimal cost, robust design was offered
entities responsible for inspecting production, overseeing and control- to a dairy cooperative. This methodology had originally been devised for
ling the processes so involved, and the sale of milk and other dairy a theoretical model for characterizing the relationship between robust-
products. ness and cost concerning capacity planning and warehouse location
Specialized companies produce most of the commercially available within supply chains; however, this paper focuses on its practical
milk in Colombia; this is beneficial for small- and medium-sized application to a real supply chain design-related problem.
businesses, primarily in the country's cold regions close to consumer This article is organized as follows: Section 2 reviews related works,
centers. Current government policies for consumer groups require in- Section 3 gives the methodology and explains the binary programming
depth analysis of milk production and distribution in certain regions of model used as design tool, Section 4 gives the results and analysis of the
the Boyacá and Cundinamarca departments, especially on the savannah methodological implementation based on real data provided by the
of Bogotá and in the Ubaté region (the target area for this study). cooperative and Section 5 contains our conclusions and suggestions for
Cooperatives have been established in these regions, acting as inter- future research.
mediaries between local farmers and the dairy industry. Supply chain
operators sometimes provide cooperatives with the required physical 2. Literature review
resources, such as goods, equipment and transport. The target area
typically consists of small properties in rural areas having limited access Uncertainty is an important factor when planning supply chains; it
to vehicles and poor raw milk storage conditions/availability; moreover, may influence configuration and coordination efficiency and signifi-
these dairy cooperatives lack production goals, good production prac- cantly affect performance (Peidro et al., 2009); however, few studies
tices and performance records for their cows’ milk production. have considered this factor. A review by Melo et al. (2009) found that
However, the greatest obstacle regarding raw milk supply is the lack 20% of all publications have considered stochastic variables, whereas a
of appropriate transport for sending the product to the processing review by Mula et al. (2010) stated that 30% of studies have considered
centers. A cold chain for slowing down or preventing microbiological uncertainty.
(microorganism growth), physiological (ripening), biochemical (brown- Such apparent lack of research indicates that uncertainty is fre-
ing, lipid oxidization) and/or physical changes (e.g., moisture loss) in quently ignored in studies investigating supply chain-associated pro-
the product is currently not available. Actual collection conditions in the blems. Blackhurst et al. (2004) considered supply chains’ dynamic
Ubaté region prevent raw milk from being used in pasteurized milk nature and corresponding uncertain demand as causes of the inherent
production and such milk may be rejected or used for producing low– complexity within such chains; several authors have suggested that
quality cheese, thereby constituting economic loss for the cooperatives. ignoring uncertainty may have a negative effect on supply chains (Gupta
The diversity of conditions represents an obstacle for supply chain and Maranas, 2003; Jung et al., 2004).
operators because it complicates how their recommended techniques One method of managing such uncertainty is by considering the
might become adopted by local farmers to meet set quality and volume concept of robustness. A survey of the literature concerning supply
requirements and avoid seasonal fluctuations. chain robustness (Vlajic et al., 2010) revealed that the organizational
Seminars on good farming practice have been organized to improve design of robust supply chains is not guided by an integral framework;
these conditions and contribute towards increasing daily milk produc- they also indicated that robustness can be considered at conceptual,
tion quotas. Economic assistance has been provided so that farmers can qualitative and quantitative model levels. Vlajic et al. (2012) have
invest in equipment for handling raw milk and milking machines. The suggested that robustness is an important qualitative supply chain
government and private entities have offered raw milk cooling tanks to property which can be used for improving resilience. Robustness is thus
improve milk collection and storage and preserve milk quality and mainly related to overcoming a chain's vulnerabilities caused by a
safety. variety of general interruptions (Tang, 2006; Wieland and Wallenburg,
An efficient cold chain is designed to provide the best possible 2013) and uncertainties. Quantitatively, robustness is primarily con-
conditions for avoiding or delaying changes in milk for a sufficiently sidered according to supply chain problems, such as planning (Zapfel,
long period of time (James and James, 2010) and such design must 1998; Van Landeghem and Vanmaele, 2002; Goetschalckx and
consider intrinsic dairy production variations in mountainous regions Fleischmann, 2005; Leung et al., 2007; Komoto et al., 2011), program-
similar to those in the Ubaté region, including the daily production of ming (Kutanoglu and Wu, 2004; Adhitya et al., 2007), network design
milk per cow, the distances to be covered regarding milk delivery, and (Peng et al., 2011; Mo and Harrison, 2005; Meepetchdee and Shah,
the necessary amount of cooling tanks and their corresponding 2007), and inventory management (Tee and Rossetti, 2002; Ouyang,
capacity. Such parameters are fundamental and must be carefully 2007; Kastsian and Mönnigmann, 2011).
considered for suitably placing tanks in a production network. The Santoso et al. (2005) have stated that unless supply chain design is
greatest challenge in designing a raw milk distribution network invol- robust regarding uncertain operational conditions, then the impact of
ving a cold chain is related to the seasonal and geographical fluctuations operational inefficiency, such as delays and interruptions, will always be
of milk production in the target area (i.e. Ubaté) as well as variability increased. Genin et al. (2008) focused on robustness regarding deci-
regarding product demand because of unstable economic conditions. sion-making aimed at delivering more stable results accompanied by
An inability to manage such uncertainties may lead to either an low variability concerning key performance indicators (KPIs).
unfeasible design or sub-optimal supply chain performance (Tong Bertsimas and Sim (2004) indicated that a robust solution is immune
et al., 2014). to uncertainty regarding available data. Klibi et al. (2010) focused on a
The strategic goal of supply chain design is to determine a specific definition of robustness for supply chains, asserting that
configuration which optimizes long-term performance within a specific robustness is supply chain ability to remain effective in all potential
planning horizon (Huang and Goetschalckx, 2014). This article de- scenarios. Vlajic et al. (2012) defined supply chain robustness as a
scribes how a raw-milk supply chain was configured for efficiently desirable property which is reflected in its performance; robustness can
managing a variety of future scenarios, involving decision-making in thus be viewed as the degree to which a supply chain shows acceptable
line with strategic planning under uncertainty. The supply chain has KPI values during and after an unexpected event that perturbs one or
been configured through a robust design and future uncertainty has more logistical processes. Ali et al. (2004) defined robustness as the
been modeled using the aforementioned scenarios (Peterson et al., preservation of certain of a system's desired characteristics despite
2003). fluctuations in the behavior of its components or environment.
This paper's novel contribution lies in adopting Tordecilla-Madera, Ali et al. (2004) proposed a methodology referred to as the features,
& González-Rodríguez's methodology (2014) regarding a strategic perturbation, impact, analysis (FePIA) procedure for use as a measure

711
R. Tordecilla-Madera et al. International Journal of Production Economics 183 (2017) 710–720

of robustness (i.e. this procedure can quantitatively determine how 3.1. Mathematical model
robust a system is). The present article describes how the methodology
proposed by Tordecilla-Madera and González-Rodríguez (2014), based The mathematical model used here was based on work by Sánchez
on Ali et al. (2004), was applied to a real-life situation. Tordecilla- and Muñoz (2014), adapted to the present study's specific character-
Madera and González-Rodríguez (2014) originally used a theoretical istics (i.e. suitably locating cooling tanks so that the producers could
model regarding supply chain design; however, such methodology has store the milk they produced daily). The dairy cooperative involved 41
been used in the present work for locating a dairy cooperative's cooling milk producers and the tanks had to be placed on the producers’
tanks in the Ubaté region, taking an uncertain raw milk supply into properties (not in uninhabited locations) to ease handling and main-
consideration. Our work thus contributes to the field of determining tenance. The system network thus consisted of only one type of node
supply chain robustness because, to the best of our knowledge, the (producers’ nodes). The system's optimum number of tanks and the
FePIA procedure has not been applied in this context. total capacity to be installed were determined by the model to minimize
total cost, meaning that any producers on whose property a cooling tank
3. Methodology was not located had to transport their milk. Several producers had
therefore to transport their milk (on foot or by donkeys) to the same
Tordecilla-Madera and González-Rodríguez (2014) introduced a tank, this group of producers being referred to as a “cluster” in our
methodology based on work by Ali et al. (2004). They used a theoretical work.
model for characterizing the relationship between robustness and cost Our model only contained binary variables, certain variables
regarding supply chain capacity planning and warehouse location, determining whether a cluster was created, whereas other variables
which involved four stages and was applied in the present article to specified whether a producer belonged to a created cluster. The location
the specific problem of the cooperative. However, since this was a real- model sought to minimize costs and optimize cooling tank position, one
world problem, a fifth stage had to be included to complete the system's for each created cluster. Fig. 1 shows producers’ geographical distribu-
design; this last stage was very specific to the matter in hand, tion, including their respective latitudes and longitudes. The number
concerning a tactical plan for optimizing cooling tank use. Binary assigned to each node was chosen by the authors for labeling each
programming was used for designing this system; the input for the producer and the coordinates were collected in situ using a global
model was based on real data provided by the dairy cooperative. positioning system (GPS).

Based on the above distribution, the authors designed the following location model:

• Sets
i = indicates the producer ; i = 1, 2, …, 41
j = indicates the possible cluster; j =1, 2,…,41
k = indicates cooling tank type according to its capacity; k =1, 2,…,10

• Parameters
SDailyi = daily milk supply [liters] of producer i
AHi = abscissa of producer i
OHi = ordinate of producer i
Xj = abscissa of the location of the storage center corresponding to
cluster j
Yj = ordinate of the location of the storage center corresponding to
cluster j
Dij = linear distance [km] between producer i and storage center j :
Dij =110. 6* (Xj − AHi )2 + (Yj − OHi )2
CPCk = capacity [liters] of a tank of type k
CAk = daily depreciation cost (Col $/day) of a tank of type k
CINk = depreciation cost (Col $/day ) of the infrastructure of a tank type
k
CRFk = refrigeration cost (Col $/liter ) of tank type k
SD = daily wage (Col $/day ) of an employee maintaining a cooling tank
and its infrastructure
CTR = transportation cost of one liter of milk (Col $/(km. liter ))

• Variables
Zjk = binary variable specifying whether cluster j of capacity k is
created (Zjk =1) or not (Zjk =0)
Wijk = binary variable specifying whether producer i belongs (Wijk =1)
or not (Wijk =0) to cluster j of capacity k

712
R. Tordecilla-Madera et al. International Journal of Production Economics 183 (2017) 710–720

Fig. 1. The milk producers’ geographical location.

• Objective function
Min TC = total cost (Col $/day ) Min TC Fig. 2. Flow diagram regarding the methodology.
41 41 10
∑ ∑ ∑ (CTR*SDailyi*Dij*Wijk )
• Stage 2. The impact of π on Γ
=
i =1 j =1 k =1 s and φr is determined to identify the
41 10 basic configurations that should be adopted by the chain for the
+ ∑ ∑ [(CAk + CINk + SD )*Zjk ] instances being studied;
j =1 k =1
41 41 10
• Stage 3. The impact of π on Γs and φr is determined once again for
each configuration;
+ ∑ ∑ ∑ (CRFk*SDailyi*Wijk )
i =1 j =1 k =1 (1)
• Stage 4. The robustness-cost relationship is characterized from the
results and a configuration is chosen according to this characteriza-
• Constraints tion. The chosen configuration must be used for locating and
constructing the cooperative's tanks; and
41 • Stage 5. A tactical plan is designed for tank use once their locations
∑ SDailyi*Wijk ≤ CPCk*Zjk ∀ j, k have been chosen and they have been constructed; this plan also
i =1 (2) depends on milk supply level.
41
∑ SDailyi*Wijk ≥ 0.7*CPCk*Z jk ∀ j, k 3.2.1. Stage 1. Preliminary description of the supply chain
i =1 (3) This stage consists of the qualitative and quantitative description of
41 10 robustness requirements, performance features and perturbation para-
∑ ∑ Wijk = 1 ∀ i meters. Robustness requirement is an interval associated with one or
j =1 k =1 (4) more indicators and it indicates whether a supply chain is operating
10 correctly. Many performance measures can be found in the literature
∑ Zjk ≤ 1 ∀ j for generic supply chains (Chan, 2003; Gunasekaran et al., 2004) and
k =1 (5) agri-food supply chains (Aramyan et al., 2007); the latter have proposed
that shelf life be a performance measure for agri-food supply chains,
∀ Zjk , Wijk ∈ {0, 1} (6)
this being the time interval between harvest or packaging and the
moment the item becomes unfit for consumption. The scope of the
Eq. (1) indicates the objective function which seeks to minimize present investigation means that shelf life refers to the time elapsing
supply chain total cost, including transport, operation and refrigeration between milking cows and refrigerating their milk in the corresponding
costs. These three cost terms have similar magnitude; therefore, there is tanks. Two robustness requirements associated with this indicator must
not imbalance between them. Eq. (2) states that the total amount of thus be defined: average distance (Γ1 (km)) traveled by each producer
milk transported to a cluster cannot exceed the capacity of the when transporting milk from their properties to the cooling tanks and
corresponding tank. Eq. (3) sets the minimum capacity required per the distance weighted by the volume of milk transported (Γ2 (km*liter))
cooling tank (a cooperative management requirement), meaning that at by each producer from their property to a cooling tank. Time has not
least 70% of the capacity of any tank placed in this scheme must be been treated directly to simplify handling the variables (as defined by
used. Eq. (4) ensures that each producer belongs to only one cluster. Eq. Aramyan et al., 2007); distances have been used instead. The system
(5) ensures one cooling tank for each cluster. Eq. (6) indicates that all has been considered more robust for smaller distance values due to the
the model's variables are binary. relationship between time and distance. Eqs. (7) and (8) illustrate
robustness requirements mathematically:
3.2. Proposed methodology 41 41 10
∑i =1 ∑ j =1 ∑k =1 Dij *Wijk
Γ1 =
The proposed methodology was divided into five stages (Fig. 2). 41 41
∑i =1 ∑ j =1 ∑k =1 Wijk
10
(7)

• Stage 1. Robustness requirements (Γs), performance features (φr) 41 41 10


∑i =1 ∑ j =1 ∑k =1 SDailyi*Dij *Wijk
and perturbation parameters (π) are qualitatively and quantitatively Γ2 = 41 41 10
described; ∑i =1 ∑ j =1 ∑k =1 Wijk (8)

713
R. Tordecilla-Madera et al. International Journal of Production Economics 183 (2017) 710–720

Performance features had to have limited variation so that the the mathematical model when calculating the perturbation parameter,
described robustness requirements were met (Ali et al., 2004); the would thus range according to Eq. (12), in which δ is a parameter
following performance features were considered based on the model varying from zero (0) to ten (10). Total input for the model consisted of
described above: 11 SDailyi( ∀ i ) values, varying linearly and proportionally through all
the possible values, according to the pertinent literature:
• Number of created clusters (φ1): indicates the number of ⎡
SDailyi = ⎢6 +
δ*(25 − 6) ⎤
clusters that must be created according to each supply scenario. ⎣ ⎥⎦*NCi ∀i
10 (12)
The number of created clusters is defined in Eq. (9):
41 10
3.2.3. Stage 3. Determining the behavior of each identified
φ1 = ∑ ∑ Zjk
j =1 k =1 (9) configuration
The results from Stage 2 specified a chain configuration for each
• Total installed capacity (φ2 ): refers to the sum of the total considered value regarding average supply π (11 values in total);
capacities of all tanks installed in the system. Total installed capacity however, the same configurations were observed for certain values of
is defined in Eq. (10): π, a total of eight distinct configurations thus being obtained. Additional
restrictions forced the model to adopt each configuration and, for each
41 10 configuration, π was varied once more using the 11 values considered in
φ2 = ∑ ∑ CPCk*Zjk the previous stage. Furthermore, installed capacities were not fixed, this
j =1 k =1 (10) being consistent with the definition of “configuration” provided in
Section 3.2.1. The following modifications were thus made to the model:
• Supply chain configuration (φ3): indicates the number of
1) A new known parameter aj , which is a binary matrix indicating
clusters to be created, along with the precise location of each cooling
tank and information regarding whether a tank belongs to each whether a tank is (aj =1) or is not (aj =0 ) located in the cluster j, was
cluster after the model algorithm has been run (with corresponding introduced; and
modifications in input variables). Such definition does not include 2) Eq. (5) was replaced by Eq. (13), forcing the model to adopt a
installed capacity. Section 4 gives examples of what a configuration specific configuration and allowing for behavioral analysis of each
is; compared to the previously defined features, supply chain configurations in terms of Γs, φr and total cost (TC) as supply varied
configuration is a qualitative feature. [Eqs. (11) and (12)]:
10
Perturbation parameters refer to uncontrollable variables which are ∑ Zjk = aj ∀ j
subject to uncertainty and affect a system's performance features and, k =1 (13)
therefore, its robustness (Ali et al., 2004). According to Peidro et al.
(2009), there are three sources of uncertainty in a supply chain: supply,
3.2.4. Stage 4. Characterizing the robustness-cost relationship
processing-manufacturing, and demand. The only source of uncertainty
The different parameters’ relationships were graphically analyzed
considered in the current work was associated with processing because
once Γs, φr and TC had been determined separately for each config-
of great variability concerning milk production by the cattle in the target
uration according to controlled variation in total daily milk supply (π).
area. Total daily milk supply was thus taken as the only perturbation
The most relevant parameters for characterizing the robustness-cost
parameter for the system, as shown in Eq. (11):
relationship were as follows:
41
π= ∑ SDailyi • Total daily supply (π) compared to average distance covered (Γ1);

i =1 (11)
Total daily supply (π) compared to average weighted distance
Table 1 summarizes the aspects discussed in Stage 1. covered (Γ2); and
• Total daily supply (π) compared to total cost (TC)
3.2.2. Stage 2. Identifying the impact of perturbation parameter π on
robustness requirements Γs and performance features φr These plots determined the most robust configurations or config-
Supply (π) is varied in this stage in a controlled manner to identify urations having the greatest capability to maintain average distance and
robustness requirements Γs and performance characteristics φr pat- average weighted distance within acceptable limits, along with the cost
terns. Work by Sánchez and Muñoz (2014) was considered for such of achieving these configurations, thereby indicating the cost of
variation; the number of heads of cattle owned by each producer in the guaranteeing the greatest robustness for a given configuration.
cooperative (NCi ) was determined through surveys, whereas work by
Fonseca Carreño et al. (2013) was used for establishing that one cow's 3.2.5. Stage 5. Tactical plan regarding tank use
daily production could range from 6 to 25 l, depending on conditions Once the best configuration had been selected in terms of robust-
such as climate, diet, technological infrastructure and production ness-cost relationship, the cooperative would have had to install the
system. Supply (SDailyi ) in the mathematical model, used as input for infrastructure on each selected property and locate the cooling tanks.
This stage also involved determining the methods by which the tanks
Table 1 would be used for the 11 scenarios regarding different values of π.
Aspects defined in Stage 1. Compared to Stage 3, installed capacities were fixed. The following
modifications were thus made to the model:
Robustness Performance Perturbation
requirement Γs characteristics φr parameter π
1) A new known parameter bjk was introduced, being a binary matrix
Γ1: average distance φ1 = amount of created Total daily milk indicating whether a tank of capacity k was (bjk =1) or was not (bjk =0 )
clusters supply located in cluster j;
Γ2: average weighted φ2 = total installed
2) A new variable BAj was introduced, being a binary variable indicat-
distance capacity
φ3 = supply chain ing whether a tank located in cluster j was used (BAj =1) or not
configuration (BAj =0 );

714
R. Tordecilla-Madera et al. International Journal of Production Economics 183 (2017) 710–720

3) Eq. (1), the objective equation, was replaced by Eq. (14):


41 41 10
MinTC = ∑ ∑ ∑ (CTR*SDailyi*Dij *Wijk + CRFk*SDailyi*Wijk )
i =1 j =1 k =1
41 10 10
+ ∑ ∑ (SD*Zjk ) + ∑ (CAk + CINk )
j =1 k =1 k =1 (14)
Compared to Eq. (1), the costs associated with tank and infrastruc-
ture depreciation were always incurred in Eq. (14) and did not depend
on whether the tank was used or not;
4) Eq. (5) was replaced by Eq. (15), allowing a tank to be used
depending on supply level on a given day
Zjk = bjk *BAj ∀ j, k (15)
and
5) Eq. (3) was discarded as minimum tank capacity use was no longer Fig. 3. Configuration for π=2622 l.
tactically required because during low milk production periods
certain tanks would not be used or only filled to a small percentage
of their real capacity.

4. Results and discussion

The problem regarding different scenarios was solved on a computer


(Intel Core i3 2.20 GHz processor and 2.00 GB RAM) using GAMS and
CPLEX solver and binary programming. Stage 2 runs were based on
Stage 1 characterization (Table 1) and equivalent values between δ and
π (Table 2) were obtained and remained constant throughout the
remainder of the experiment. Given the number of producers and their
heads of cattle, the minimum volume of milk which could have been
produced on any given day would have been 2622 l, whereas the
maximum volume might have reached 10,925 l. As supply grew, the
average number of kilometers traveled (Γ1) would have become
reduced because more clusters (φ1) would have been created and the
tanks would have become closer to the milk producers; total cost (TC)
and total installed capacity (φ2) of the installed cooling tanks would Fig. 4. Configuration for π=4282.6 l.

also have increased.


Figs. 3–10 give the eight configurations adopted by the system after tank could not be met when many clusters were created. Producers
the runs. Each tank's location in a cluster is indicated by a red dot and a would therefore have to travel greater distances.
producer by a blue dot and shown with its considered total supply value. Fig. 12 shows π compared to Γ2 and, for all configurations, the
There are only eight configurations because certain supply values gave average weighted distance grew almost linearly with total supply,
identical configurations, indicating that the tanks were located in areas whereas the slope was not the same for all configurations (while
having the same properties, even if the assignment of the producers was configuration A had the greatest slope, configurations G and H had
different. These configurations are labeled A–H in Table 3 along with the smallest). This result was caused by configuration A only creating
their relationship to parameter π. four clusters, whereas G and H created eight; producers would thus
At this point, each configuration (A to H) was analyzed individually have had to travel shorter average distances..
and π was initially plotted against Γ1 for each identified configuration. Figs. 11 and 12 show that configurations G and H were the most
Fig. 11 shows this plot; each configuration had a slight decrease in robust for almost all scenarios except the lowest supply scenario, for
average distance because tank locations were fixed for each configura- which both plots gave the lowest values for Γ1 and Γ2. However, these
tion; significant variation was thus not observed for the distance results did not indicate that either configuration should be adopted by
traveled by the producers. However, this slight change was not observed the cooperative because the cost pattern for each configuration still had
for the scenario having the lowest supply for configurations E, F, G and to be analyzed. Fig. 13 shows π plotted against TC and indicated that G
H, having high Γ1 value because the minimum 70% volume for each and H configurations’ greater robustness only produced a higher cost

Table 2
Results of the first runs performed with the proposed methodology.

Output variable Parameter δ

0 1 2 3 4 5 6 7 8 9 10

π (liters/day) 2622.0 3452.3 4282.6 5112.9 5943.2 6773.5 7603.8 8434.1 9264.4 10,094.7 10,925.0
Γ1 (km) 1.45 1.41 1.26 1.06 1.08 1.05 0.93 1.04 0.95 0.88 0.85
Γ2 (km*liter) 87.33 112.76 117.82 113.88 135.71 154.06 147.06 186.37 180.55 174.13 187.49
TC (Col$/day) 476,998 562,870 651,193 711,694 783,233 847,442 901,647 961,550 1,016,199 1,078,124 1,109,953
φ1 (no. of created clusters) 4 4 5 6 6 6 7 6 7 8 8
φ2 (liters) 3500 4500 5500 6500 8000 8500 9500 11,000 12,000 12,500 15,000

715
R. Tordecilla-Madera et al. International Journal of Production Economics 183 (2017) 710–720

Fig. 8. Configuration for π=9264.4 l.


Fig. 5. Configuration for π=5112.9 l.

Fig. 6. Configuration for π=6773.5 l. Fig. 9. Configuration for π=10,094.7 l.

Fig. 7. Configuration for π=7603.8 l. Fig. 10. Configuration for π=10,925.0 l.

for small supply values (between π=2622.0 and 5943.2 l). Regarding and improved milk quality due to prolonged shelf life.
higher supply values, G and H configuration costs became reduced as Figs. 11, 12 and 13 show that the differences between configurations
the highest demand was reached, producing the lowest cost for all G and H were virtually imperceptible; thus, an additional criterion was
configurations.. required to choose between them. Table 2 thus shows that φ2 values
Based on these results, it was suggested to the cooperative that the were then specifically considered (the last two columns in the table are
cooling tanks should be located according to either configuration G or for configurations G and H); φ2 =12,500 l for configuration G, whereas
configuration H, having the highest robustness and involving shorter φ2 =15,000 l for configuration H. These results indicated that config-
average distances and weighted distances from the producers’ proper- uration G should be recommended because configuration H implied a
ties to the cooling tanks, as well as lower management costs for most larger amount of non-utilized capacity for a maximum total supply of
supply scenarios. Improved robustness implied shorter transport times 10,925 l, this being undesirable for the dairy cooperative in question.

716
R. Tordecilla-Madera et al. International Journal of Production Economics 183 (2017) 710–720

Table 3
Configurations adopted by the system.

Configuration A B C D E F G H

Total supply (π) 2622.0 4282.6 5112.9 6773.5 7603.8 9264.4 10,094.7 10,925.0
3452.3 5943.2
8434.1

Table 4
Capacity to be installed at the producers’ properties according to configuration G.

Producer number Installed capacity (liters)

10 1500
12 3000
24 2000
30 1000
31 1000
37 500
38 1500
39 2000

Fig. 11. π compared to Γ1 for each configuration.


assignment was not fixed and could vary depending on supply levels for
a given period of time.

4.1. Tactical plan for tank use

The last stage of this study consisted of drawing up a plan for


installed tank use regarding the different supply scenarios. The model
produced three scenarios after the 11 runs described in Section 3.2.5;
Table 5 summarizes the results. The first scenario concerned the lowest
supply and involved using five tanks; the second scenario concerned the
next three supply values and involved using seven tanks whilst the third
scenario dealt with the seven largest supply values and included the use
of eight installed tanks. Greater supply required more tanks due to
necessarily larger capacities and the cost structure. Thus, for scenarios
involving lower demand, it would be less costly for producers to travel
longer distances than to use some of the tanks; these tanks would thus
not be used in such scenarios.
Fig. 12. π compared to Γ2 for each configuration. Table 6 summarizes the tank utilization plan. An approximate range
was established according to the π values in Table 5. Tanks 10, 12, 24,
38 and 39 were thus used for any supply value; tanks 30 and 31 were
only used when supply was medium or high, and tank 37 was only used
in the high supply scenario. The tanks that were not used in low and
medium supply scenarios were those having the lowest capacity,
according to Table 4 which indicates that the model tended to involve
using the higher-capacity tanks because of the cost structure and lower-
capacity tanks being used for storing milk from fewer producers. The
latter scenario would inevitably require the use of larger tanks, thereby
increasing long-term management costs.
Assigning producers to each tank or cluster also had to be
considered. Although the model produced three scenarios, the high-
supply scenario contained three sets of assignment (Figs. 14–18), along
with the sets for scenarios 1 and 2. Figs. 14 and 15 show that five and
seven clusters were formed, respectively. A red dot identifies a tank in
use and a dark blue dot identifies tanks which have been installed but
are not being used because of low supply values......
Fig. 13. π compared to TC for each configuration. Figs. 16–18 show the three sub-scenarios regarding the high-supply
scenario. They show all eight tanks being used but different producer
assignments. This situation would reflect installed capacity because
Table 4 gives the details regarding configuration G, indicating the certain tanks would no longer be able to receive milk as supply
producers and capacity associated with each tank. Fig. 9 gives the increased; however, since total system capacity would be sufficient,
locations and also shows the producers assigned to each cluster. This the producers could be reassigned.

717
R. Tordecilla-Madera et al. International Journal of Production Economics 183 (2017) 710–720

Table 5
Tactical plan results.

Output variable Scenario

1 (Low supply) 2 (Medium supply) 3 (High supply)

π (liters/day) 2622.0 3452.3 4282.6 5112.9 5943.2 6773.5 7603.8 8434.1 9,264.4 10,094.7 10,925.0
Γ1 (km) 1.35 0.92 0.92 0.92 0.83 0.83 0.83 0.83 0.83 0.86 0.91
Γ2 (km*liter) 72.99 66.96 83.06 99.16 99.74 113.68 127.61 141.55 155.48 172.65 203.75
TC (Col$/day) 465,176 543,960 608,656 673,352 736,821 796,868 856,915 916,962 977,008 1,044,360 1,142,795
φ1 (no. of tanks to be used) 5 7 7 7 8 8 8 8 8 8 8

Table 6
Utilization plan for tanks located according to supply range.

Producer Supply range (liters/day)


number
Low supply Medium supply High supply

(2662– (3037.2– (5528.1–


3037.2) 5528.1) 10,925)

10 X X X
12 X X X
24 X X X
30 X X
31 X X
37 X
38 X X X
39 X X X

Fig. 16. Assignments for the first high-supply sub-scenario.

Fig. 14. Assignments for the low-supply scenario.


Fig. 17. Assignments for the second high-supply sub-scenario.

Fig. 15. Assignments for the medium-supply scenario. Fig. 18. Assignments for the third high-supply sub-scenario.

718
R. Tordecilla-Madera et al. International Journal of Production Economics 183 (2017) 710–720

5. Conclusions managing abnormal supply chain events. Comput. Chem. Eng. 31, 496–518. http://
dx.doi.org/10.1016/j.compchemeng.2006.07.002.
Ali, S., Maciejewski, A.A., Siegel, H.J., Kim, J.K., 2004. Measuring the robustness of a
The present investigation involved using a binary programming resource allocation. IEEE Trans. Parallel Distrib. Syst. 15, 630–641. http://
model for designing a milk storage and refrigeration logistics system for dx.doi.org/10.1109/TPDS.2004.24.
Aramyan, L.H., Oude Lansink, A.G.J.M., van der Vorst, J.G.A.J., van Kooten, O., 2007.
a dairy cooperative in the Ubaté region in Colombia's Cundinamarca Performance measurement in agri-food supply chains: a case study. Supply Chain
department. This design had to include locating cooling tanks which Manag.: Int. J. 12, 304–315. http://dx.doi.org/10.1108/13598540710759826.
could store milk from several producers and decisions had to be made Bertsimas, D., Sim, M., 2004. The price of robustness. Oper. Res. 52, 35–53. http://
dx.doi.org/10.1287/opre.1030.0065.
regarding the number of tanks to be placed, each tank's installed Blackhurst, J., Wu, T., O'Grady, P., 2004. Network-based approach to modelling
capacity and which producers had to be assigned to each tank to uncertainty in a supply chain. Int. J. Prod. Res. 42, 1639–1658. http://dx.doi.org/
minimize total cost. A set of clusters was thus created, one for each 10.1080/0020754030360001646064.
Chan, F.T.S., 2003. Performance measurement in a supply chain. Int. J. Adv. Manuf.
tank. A variety of scenarios were considered for the design because real-
Technol. 21, 534–548. http://dx.doi.org/10.1007/s001700300063.
life daily milk supply is variable in the target area. The supply was thus Consejo Nacional de Política Económica y Social, 2010. CONPES 3675 Política Nacional
varied in a controlled manner by spanning the various ranges found in Para Mejorar la Competitividad del Sector Lácteo Colombiano. Consejo Nacional de
the pertinent literature, resulting in eight configurations which were Política Económica y Social. Departamento Nacional de Planeación, Bogotá,
[Available in]〈http://www.ica.gov.co/getattachment/f74ec780-6456-431d-b292-
individually evaluated for each supply scenario considered for deter- 0aff856388d9/2010cp3675.aspx〉.
mining each configuration's robustness and cost. Robustness was Departamento Nacional de Planeación DNP, 2011. Plan Nacional de Desarrollo
associated with the following parameters: “average distance (km) “Prosperidad Para Todos” Tomo 1. Departamento Nacional de Planeación, Bogotá,
Colombia, [Available in]〈https://colaboracion.dnp.gov.co/CDT/PND/PND2010-
traveled by each producer for transporting milk from their property 2014%20Tomo%20I%20CD.pdf〉.
to a designated cooling tank” and “average weighted distance (km*liter) Fonseca Carreño, J.A., Cleves Leguízamo, J.A., Páez Barón, E.M., 2013. Integración de
covered by each producer when transporting milk from their property microempresas lácteas del corredor central del departamento de Boyacá (Colombia).
Rev. De. Invest. Agrar. Y. Ambient. 4, 117–133.
to a designated cooling tank”. It was determined that the best config- Genin, P., Lamouri, S., Thomas, A., 2008. Multi-facilities tactical planning robustness
uration in terms of robustness and cost would include eight tanks with experimental design. Prod. Plan. Control 19, 171–182. http://dx.doi.org/
located in eight clusters, having an average distance ranging from 10.1080/09537280801896250.
Goetschalckx, M., Fleischmann, B., 2005. Strategic Network Planning Supply Chain
0.83 km/producer (high-supply scenario) to 1.35 km/producer (low-
Management and Advanced Planning, Concepts, Models, Software and Case Studies
supply scenario). The cost associated with this configuration ranged (III edition). Springer, 117–138. http://dx.doi.org/10.1007/b106298.
from a minimum of 495,176 Col$/day for the low-supply scenarios to a Gunasekaran, A., Patel, C., McGaughey, R.E., 2004. A framework for supply chain
performance measurement. Int. J. Prod. Econ. 87, 333–347. http://dx.doi.org/
maximum of 1,142,795 Col$/day for the high-supply ones. Total
10.1016/j.ijpe.2003.08.003.
installed capacity for this configuration was 12,500 l. A tactical plan Gupta, A., Maranas, C.D., 2003. Managing demand uncertainty in supply chain planning.
was then developed for tank use in the low-, medium- and high-supply Comput. Chem. Eng. 27, 1219–1227. http://dx.doi.org/10.1016/S0098-1354(03)
scenarios; it was thus determined that five out of the eight installed 00048-6.
Huang, E., Goetschalckx, M., 2014. Strategic robust supply chain design based on the
tanks had to be used for the low-supply scenario, seven out of the eight Pareto-optimal tradeoff between efficiency and risk. Eur. J. Oper. Res. 237, 508–518.
had to be used for the medium-supply scenario, and that all the tanks http://dx.doi.org/10.1016/j.ejor.2014.02.038.
had to be used in the high-supply scenario. James, S.J., James, C., 2010. The food cold-chain and climate change. Food Res. Int. 43,
1944–1956. http://dx.doi.org/10.1016/j.foodres.2010.02.001.
Regarding putting this tactical plan into effect, the dairy cooperative Jung, J.Y., Blau, G., Pekny, J.F., Reklaitis, G.V., Eversdyk, D., 2004. A simulation based
has been advised to use forecast models for scientifically and efficiently optimization approach to supply chain management under demand uncertainty.
establishing the supply scenario (low, medium or high) expected during Comput. Chem. Eng. 28, 2087–2106. http://dx.doi.org/10.1016/
j.compchemeng.2004.06.006.
each season, thereby enabling optimized usage of installed cooling Kastsian, D., Mönnigmann, M., 2011. Optimization of a vendor managed inventory
tanks. supply chain with guaranteed stability and robustness. Int. J. Prod. Econ. 131,
Academically, this investigation involved using a novel methodology 727–735. http://dx.doi.org/10.1016/j.ijpe.2011.02.022.
Klibi, W., Martel, A., Guitouni, A., 2010. The design of robust value-creating supply chain
for solving a real-world strategic location problem. The robust design networks: a critical review. Eur. J. Oper. Res. 203, 283–293. http://dx.doi.org/
arising from it (which can withstand variations in milk supply at a 10.1016/j.ejor.2009.06.011.
minimal cost without disturbing normal operations) was offered to the Komoto, H., Tomiyama, T., Silvester, S., Brezet, H., 2011. Analyzing supply chain
robustness for OEMs from a life cycle perspective using life cycle simulation. Int. J.
dairy cooperative in question. The most robust design did not necessa-
Prod. Econ. 134, 447–457. http://dx.doi.org/10.1016/j.ijpe.2009.11.017.
rily imply the highest cost; in-depth studies regarding the robustness- Kutanoglu, E., Wu, S.D., 2004. Improving scheduling robustness via preprocessing and
cost relationship must thus be conducted. dynamic adaptation. IIE Trans. 36, 1107–1124. http://dx.doi.org/10.1080/
It should be noted that, in our model, Euclidean distance was used 07408170490500681.
Leung, S.C.H., Tsang, S.O.S., Ng, W.L., Wu, Y., 2007. A robust optimization model for
for calculating the distance between producers and tanks. This has multi-site production planning problem in an uncertain environment. Eur. J. Oper.
provided a good approach because our scope is regional; however, Res. 181, 224–238. http://dx.doi.org/10.1016/j.ejor.2006.06.011.
distance must be calculated more accurately for designs on a larger Meepetchdee, Y., Shah, N., 2007. Logistical network design with robustness and
complexity considerations. Int. J. Phys. Distrib. Logist. Manag. 37, 201–222. http://
scale. dx.doi.org/10.1108/09600030710742425.
Other sources of uncertainty (in addition to that of supply) should Melo, M.T., Nickel, S., Saldanha-da-Gama, F., 2009. Facility location and supply chain
be included in future research; for example, rural roads can be affected management - A review. Eur. J. Oper. Res. 196, 401–412. http://dx.doi.org/
10.1016/j.ejor.2008.05.007.
by rain, thereby preventing the efficient transport of milk. Further Mo, J., Harrison, J.P., 2005. A conceptual framework for robust supply chain design
investigation is thus needed regarding the effect of such parameters on under demand uncertaintySupply Chain Optimization. Springer, 243–264. http://
robustness and cost. dx.doi.org/10.1007/0-387-26281-4_8.
Mula, J., Peidro, D., Diaz-Madronero, M., Vicens, E., 2010. Mathematical programming
models for supply chain production and transport planning. Eur. J. Oper. Res. 204,
Acknowledgments 377–390. http://dx.doi.org/10.1016/j.ejor.2009.09.008.
Ouyang, Y., 2007. The effect of information sharing on supply chain stability and the
bullwhip effect. Eur. J. Oper. Res. 182, 1107–1121. http://dx.doi.org/10.1016/
The authors would like to thank Fundación Universitaria Agraria de
j.ejor.2006.09.037.
Colombia, Universidad de La Sabana and the unnamed cooperative for Peidro, D., Mula, J., Poler, R., 2009. Quantitative models for supply chain planning under
their support without which it would have been impossible to conduct uncertainty: a review. Int. J. Adv. Manuf. Technol. 43, 400–420. http://dx.doi.org/
the present study. 10.1007/s00170-008-1715-y.
Peng, P., Snyder, L.V., Lim, A., Liu, Z., 2011. Reliable logistics networks design with
facility disruptions. Transp. Res. B 45, 1190–1211. http://dx.doi.org/10.1016/
References j.trb.2011.05.022.
Peterson, G.D., Cumming, G.S., Carpenter, S.R., 2003. Scenario Planning: a Tool for
Conservation in an Uncertain World. Conserv. Biol. 17, 358–366. http://dx.doi.org/
Adhitya, A., Srinivasan, R., Karimi, I.A., 2007. A model-based rescheduling framework for

719
R. Tordecilla-Madera et al. International Journal of Production Economics 183 (2017) 710–720

10.1046/j.1523-1739.2003.01491.x. cost relation in capacity planning and warehouse location problem in supply chains
Sánchez D.L., and Muñoz D.S. (2014). Estudio de reconfiguración de la cadena de through a methodology based on the FePIA procedure. Universidad de la Sabana,
suministro para productores lecheros en el municipio de Lenguazaque mediante el Chía, Colombia, [Unpublished paper].
uso de modelación matemática. Undergraduate thesis, Fundación Universitaria Van Landeghem, H., Vanmaele, H., 2002. Robust planning: a new paradigm for demand
Agraria de Colombia, Bogotá. Available in 〈http://issuu.com/maosabo/docs/ chain planning. J. Oper. Manag. 20, 769–783. http://dx.doi.org/10.1016/S0272-
tesis0188ii〉 6963(02)00039-6.
Santoso, T., Ahmed, S., Goetschalckx, M., Shapiro, A., 2005. A stochastic programming Vlajic, J.V., van der Vorst, J.G.A.J., Hendrix, E.M.T., 2010. On robustness in food supply
approach for supply chain network design under uncertainty. Eur. J. Oper. Res. 167, chain networks. Towards Effective Food Supply Chains. Wageningen Academic
96–115. http://dx.doi.org/10.1016/j.ejor.2004.01.046. Publishers, Netherlands, 63–82. http://dx.doi.org/10.3921/978-90-8686-705-9.
Tang, C.S., 2006. Robust strategies for mitigating supply chain disruptions. Int. J. Logist.: Vlajic, J.V., van der Vorst, J.G.A.J., Haijema, R., 2012. A framework for designing robust
Res. Appl. 9, 33–45. http://dx.doi.org/10.1080/13675560500405584. food supply chains. Int. J. Prod. Econ. 137, 176–189. http://dx.doi.org/10.1016/
Tee, Y.S., Rossetti, M.D., 2002. A robustness study of a multi-echelon inventory model via j.ijpe.2011.11.026.
simulation. Int. J. Prod. Econ. 80, 265–277. http://dx.doi.org/10.1016/S0925- Wieland, A., Wallenburg, C.M., 2013. The influence of relational competencies on supply
5273(02)00259-1. chain resilience: a relational view. Int. J. Phys. Distrib. Logist. Manag. 43, 300–320.
Tong, K., You, F., Rong, G., 2014. Robust design and operations of hydrocarbon biofuel http://dx.doi.org/10.1108/IJPDLM-08-2012-0243.
supply chain integrating with existing petroleum refineries considering unit cost Zapfel, G., 1998. Customer-order-driven production: an economical concept for
objective. Comput. Chem. Eng. 68, 128–139. http://dx.doi.org/10.1016/ responding to demand uncertainty? Int. J. Prod. Econ. 56/57, 699–709. http://
j.compchemeng.2014.05.003. dx.doi.org/10.1016/j.mcm.2011.02.036.
Tordecilla-Madera, R., González-Rodríguez, L., 2014. Characterization of the robustness-

720

Das könnte Ihnen auch gefallen