Int. J.

Production Economics 138 (2012) 89–101

Contents lists available at SciVerse ScienceDirect

Int. J. Production Economics

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

Multi-objective integrated production and distribution planning

of perishable products
P. Amorim a,n, H.-O. Günther b, B. Almada-Lobo a
a
DEIG, Faculty of Engineering, University of Porto, Rua Dr. Roberto Frias, s/n, 4600-001 Porto, Portugal
b
Department of Production Management, Technical University of Berlin, Strasse des 17. Juni 135, 10623 Berlin, Germany

a r t i c l e i n f o abstract

Article history: Integrated production and distribution planning have received a lot of attention throughout the years
Received 24 January 2011 and its economic advantages are well documented. However, for highly perishable products this
Accepted 5 March 2012 integrated approach has to include, further than the economic aspects, the intangible value of
Available online 12 March 2012
freshness. We explore, through a multi-objective framework, the advantages of integrating these two
Keywords: intertwined planning problems at an operational level. We formulate models for the case where
Perishability perishable goods have a fixed and a loose shelf-life (i.e. with and without a best-before-date). The
Multi-objective results show that the economic benefits derived from using an integrated approach are much
Production and distribution planning dependent on the freshness level of products delivered.
& 2012 Elsevier B.V. All rights reserved.

1. Introduction of items that keep deteriorating with some probability distribu-

tion are electronic components, chemicals, and vegetables, etc.
Rapidly deteriorating perishable goods, such as fruits, vegeta- When the shelf-life is fixed the most common visual cue that
bles, yoghurt and fresh milk, have to take into account the customers rely on is the best-before-date (BBD). The BBD can be
perishability phenomenon even for the operational level of defined as the end of the period, under any stated storage
production and distribution planning, which has a timespan conditions, during which the product will remain fully market-
ranging from 1 week to 1 month. Usually these products start able and retain any specific qualities for which tacit or express
deteriorating from the moment they are produced on. Therefore, claims have been made. In this case, customers will adapt their
without proper care, inventories may rapidly get spoiled before willingness to pay for a product based on how far away the BBD
their final use making the stakeholders incur on avoidable costs. is. On the other hand, when the expiry date of a product is not
The customers of these products are aware of the intense perish- printed and then the shelf-life is loose, customers have to rely on
ability they are subject to, and they attribute an intangible value their senses or external sources of information to estimate the
to the relative freshness of the goods (Tsiros and Heilman, 2005). remaining shelf-life of the good. For example, if a banana has
To evaluate freshness customers rely on visual cues which may black spots or if flowers look wilted, then customers know that
differ among the broad class of perishable products. Nahmias these products will be spoiled rather soon.
(1982) dichotomized deteriorating goods in two categories In the case of loose shelf-life, especially in the fresh food industry,
according to their shelf-life: (1) fixed lifetime: items’ lifetime is manufacturers can make use of predictive microbiology to estimate
pre-specified and therefore the impact of the deteriorating factors the shelf-life of these kinds of products based on external control-
is taken into account when fixing it. In fact, the utility of these lable factors, such as humidity and temperature (Fu and Labuza,
items may decrease during their lifetime, and when passing its 1993). To make concepts clearer, shelf-life is defined as the time
lifetime, the item will perish completely and become of no value, period for the product to become of no value for the customer due to
e.g., milk, inventory in a blood bank, and yoghurt, etc. and the lack of the tacit initial characteristics that the product is
(2) random lifetime: there is no specified lifetime for these items. supposed to have. Thus, in our case, this period starts on the day
The lifetime for these items is assumed as a random variable, and the product is produced. The determination of shelf-life as a function
its probability distribution may take on various forms. Examples of variable environmental conditions has been the focus of many
research activities in this field and a considerable number of reliable
n
models exist, such as the Arrhenius model, the Davey model and the
Corresponding author.
E-mail addresses: [email protected] (P. Amorim),
square-root model. These models take into account the knowledge
[email protected] (H.-O. Günther), about microbial growth in decaying food goods under different
[email protected] (B. Almada-Lobo). temperature and humidity conditions.

0925-5273/$ - see front matter & 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijpe.2012.03.005
90 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101

Regarding production and distribution planning, many authors machines and shared buffers, arising in a company producing
have shown the economic advantages of using an integrated yoghurt. The problem has been formulated as a hybrid continuous
decision model over a decoupled approach (Martin et al., 1993; setup and capacitated lot sizing problem. To solve this problem a
Thomas and Griffin, 1996). These advantages are believed to be two-stage heuristic procedure based on the decomposition of the
leveraged when the product suffers a rapid deterioration process problem into a lot sizing and a scheduling sub-problem has been
and hence pushes towards a more integrated view of these developed. This model accounted for perishability by using a make-
intertwined problems. For perishable goods the final products to-order production strategy. With a similar approach, but focused on
inventory that is usually used to buffer and decouple these two batch processing, Neumann et al. (2002) decompose detailed produc-
planning decisions have to be questioned since customers distin- tion scheduling for batch production into batching and batch
guish between different degrees of freshness and there is an scheduling. The batching problem converts the primary requirements
actual risk of spoilage. In this work, we want to study the for products into individual batches, where the workload is to be
potential advantages of using an integrated approach for opera- minimized. They formulate the batching problem as a nonlinear
tional production and distribution planning of perishable goods mixed-integer program and transform it into a linear mixed-binary
compared with a decoupled one. These advantages will be program of moderate size. The batch scheduling problem allocates
analyzed through the economic and product freshness perspec- the batches to scarce resources such as processing units, workers, and
tive. We focus on highly perishable consumer goods industries, intermediate storage facilities, where some regular objective function
with a special emphasis on food processing, which have to cope like the makespan is to be minimized. The batch scheduling problem
with complex challenges, such as the integration of lot sizing and is modeled as a resource-constrained project scheduling problem,
scheduling, the definition of setup families considering major and which can be solved by a truncated branch-and-bound algorithm. In
minor setup times and costs, and multiple non-identical produc- this work some intermediate perishable products cannot be stored
tion lines (Bilgen and Günther, 2010). We are also interested in eliminating the buffer between related activities. Pahl and Voß (2010)
understanding if these potential advantages differ among the two extend well-known discrete lot-sizing and scheduling models by
distinct perishable goods classes that we have mentioned before: including deterioration constraints. Also for the special case of
with fixed shelf-life and with loose shelf-life. For both the cases, yoghurt production, Lütke Entrup et al. (2005) develop three
since we are interested in rapidly deteriorating goods, we con- mixed-integer linear programming models that integrate shelf-life
sider a customer who prefers products with a higher freshness issues into production planning and scheduling of the packaging
level. To tackle explicitly this customer satisfaction issue we stage. In this work product freshness is modeled as a linear function
embedded our integrated operational production and distribution hoping that retailers could pay the difference between different
planning problem in a multi-objective framework distinguishing deterioration stages. Cai et al. (2008) develop a model and an
two very different and conflicting objectives of the planner. The algorithm for the production of seafood products. Due to a deadline
first objective is concerned with minimizing the total costs over constraint and the raw material perishability, the manufacturer
the supply chain covering transportation, production, setup and determines three decisions: the product types to be produced; the
spoilage costs. The second one aims at maximizing the freshness machine time to be allocated for each product type; and the sequence
of the products delivered to distribution centers and, therefore, to process the products selected. It is interesting to notice that here
maximize customers’ willingness to pay. the perishability is focused on the raw materials.
The remainder of the paper is organized as follows. The next Especially suited for lot-sizing and scheduling in the consumer
section is devoted to the literature review of related topics. In goods industry, where natural sequences in sequencing products
Section 3, two integrated and two decoupled models are formu- can be found in order to minimize changeover time and to ensure
lated for the multi-objective production and distribution planning quality standards, there is the concept of block planning. A block
of the perishable products problem. First we will focus on represents a sequence, set a priori, of production orders of
products with a fixed shelf-life and then with a loose one. In variable size, where each production order corresponds to a
Section 4, an illustrative example to understand the impact of the unique product type. Hence, each product type occurs at a given
models is developed and analyzed. The paper is concluded in position in a block (Günther et al., 2006). This concept will be very
Section 5. useful in reducing the complexity and augmenting the applic-
ability of our models in the next section.
There are two papers worth of reference when modeling explicitly
2. Literature review routing for perishable products. Osvald and Stirn (2008) develop a
heuristic algorithm for the distribution of fresh vegetables in which
Taking into account that perishability in supply chains has perishability represents a critical factor. The problem is formulated as
received increased attention both in practice and in academic a vehicle routing problem with time windows (VRPTW) and with
research it urges to continue the development of models that time-dependent travel times. The model considers the impact of
incorporate explicitly this phenomenon (Ahumada and Villalobos, perishability as a part of the overall distribution costs. Hsu et al.
2009; Akkerman et al., 2010). Since in this section we only review (2007) consider the randomness of the perishable food delivery
literature directly connected to our research and explore concepts process and construct a stochastic VRPTW model to obtain optimal
and strategies used to model this problem, readers are referred to delivery routes, loads, fleet dispatching and departure times for
Karaesmen et al. (2011) for an extensive review on papers delivering perishable food from a distribution center. They take into
managing perishable inventories. First we look into what has account inventory costs due to deterioration of perishable food and
been done in production and distribution modeling from a energy costs for cold storage vehicles.
decoupled and integrated point of view with a special focus on Regarding transportation in the consumer goods industry there is
perishability and consumer goods industries. Then, based on this a significant trend in outsourcing this function to third party logistics
paper a discussion about the multi-objective approach used and service providers (3PL) (Lütke Entrup, 2005). This decision enables
some other important modeling concepts follows. Hence, the consumer goods manufacturers to move towards a more flexible cost
contribution of our work in this research field is clarified. structure and, hence, the manufacturer is able to choose generally
For tackling the operational production planning of perishable between Less-than-truckload (LTL), where the shipper only pays a
goods, Marinelli et al. (2007) propose a solution approach for a real price according to the capacity used, and full-truckload where the
world capacitated lot sizing and scheduling problem with parallel shipper pays a fixed cost per load (Günther and Seiler, 2009). The
 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101 91

temperature of the transportation, in the case of perishable goods, it is necessary to employ another method to understand the
especially food, has to be in accordance with the requirements of the complementary effects of supply chain costs and product perish-
perishable goods transported, for instance, chilled or frozen. In our ability. In order to follow our goal of understanding the impact of
case we will consider that all products have the same temperature integrating the analysis of production and distribution planning, both
requirements and hence can be grouped together in every shipment. in economic and in freshness terms, the use of a single objective
Looking at the literature on integrated production and function would hinder the important trade-off between these two
distribution models dealing with perishable products, Eksioglu conflicting objectives. Some authors have already given hints about
and Mingzhou (2006) address a production and distribution the importance of using a multi-objective framework to fully explore
planning problem in a dynamic, two-stage supply chain. Their the perishability phenomenon (e.g. Arbib et al., 1999; Lütke Entrup,
model considers that the final product is perishable and has a 2005). Nevertheless, to the best of our knowledge, this is the first
limited shelf-life. Furthermore, strong assumptions are made, work that addresses the integrated production and distribution
such as, unlimited capacity. They formulate this problem as a planning of perishable products in a multi-objective framework.
network flow problem with a fixed charge cost. Based on the
economic order quantity, Yan et al. (2011) develop an integrated
production–distribution model for a deteriorating item in a two- 3. Problem statement and model formulations
echelon supply chain. Their objective is to minimize the total
system cost. Some restrictions concerning perishability are The production and distribution planning problem considered
imposed. For example, the supplier’s production batch size is in this paper consists of a number of plants p ¼ 1, . . . ,P having
limited to an integer multiple of the discrete delivery lot quantity dedicated lines which produce multiple perishable items with a
to the buyer. In a more operational perspective, Chen et al. (2009) limited capacity to be delivered to distribution centers. It is
propose a nonlinear mathematical model to consider production relevant to understand the importance of the design choice of
scheduling and vehicle routing with time windows for perishable having such a complex supply chain instead of just considering
food products. The demand at retailers is assumed to be stochas- one plant and multiple distribution centers. As said before, we
tic and perishable goods will deteriorate once they are produced. focus on perishable consumer goods industries which are known
Thus, the revenue of the supplier is uncertain and depends on the for demanding increasing flexibility in the supply chain planning
value and the transaction quantity of perishable products when processes. Thus, to consider a network of production plants which
they are carried to retailers. The objective of this model is to can add increased flexibility and reliability to hedge against the
maximize the expected total profit of the supplier. The production complex dynamics of such industries is crucial. Therefore,
quantities, the time to start production and the vehicle routes are although we are tackling an operational level of decision making
determined in the model iteratively through a decomposition for these two planning tasks we assume a central organizational
procedure. The solution algorithm is composed of the constrained unit that makes decisions which are followed directly at a local
Nelder–Mead method and a heuristic for the vehicle routing with level. The length of the planning horizon for such planning
time windows. Recently, Rong et al. (2011) developed an MIP problem ranges from 1 week to 1 month.
model where the quality of a single product is modeled throughout All product variants k ¼ 1, . . . ,K belonging to the same family
a multi-echelon supply chain. Their model uses the knowledge of form a block. Therefore a product can only be assigned to one
predictive microbiology in forecasting shelf-life based on the block. Blocks j ¼ 1, . . . ,N are to be scheduled on l ¼ 1, . . . ,L parallel
temperature of transportation and stocking. Their objective func- production lines over a finite planning horizon consisting of
tion reflects this reality by taking into account the incremental macro-periods d ¼ 1, . . . ,T with a given length. The scheduling
cooling costs necessary to achieve a longer shelf-life. This work has takes into account that the setup time and cost between blocks is
a straight linkage to the models developed for the loose shelf-life dependent on the sequence of production (major setup). The
case of this paper. sequence of products in a block is set a priori due to natural
Based on this paper, when developing production and/or distribu- constraints in this kind of industries. Hence, when changing the
tion models applied to the type of perishable products treated in this production between two products of the same block only a minor
work, a considerable number of different approaches could be setup is needed that is not dependent on the sequence, but only
considered. We can impose a make-to-order strategy for all products on the product to be produced.
so that it is likely that no products will spoil and they will be In order to consider the initial stock that might be used to fulfil
delivered with good freshness standards. It is also possible to enforce current demand it is important to have an overview of the inventory
constraints on the number of periods that a product can stay in stock built up in each macro-period due to perishability concerns. The
or just control the number of spoiled products and penalize it in the length of the horizon that needs to be considered is related to the
objective function. A third way of taking into consideration products’ product with the longest shelf-life. One shall consider an integer
perishability is by using differentiating holding costs depending on multiple X of past planning horizons that is enough to cover the
the shelf-life. So items with a shorter shelf-life are given higher longest shelf-life, i.e. X ¼ dmax u~ k =Te, where u~ k is a conservative
holding costs and items with a longer shelf-life are given lower value for shelf-life of product k, hence t ¼ XT þ1, . . . ,0; 1, . . . ,T. Let
holding costs. Finally, from a customer value point of view, it is T  ¼ fXT þ 1, . . . ,0g and T þ ¼ f1, . . . ,Tg, thus the domain of t is
possible to either attribute a value to the different degrees of equivalent to ½T ¼ T  [ T þ .
freshness that a product has when delivered, or to assign a demand A macro-period is divided into a fixed number of non-over-
function that varies in function of items’ remaining shelf-life. This lapping micro-periods with variable length. Since the production
way we are explicitly controlling the perishability process since we lines can be scheduled independently, this is done for each line
know for every delivered product the corresponding monetary value separately. Sld denotes the set of micro-periods s belonging to
of the remaining shelf-life and, therefore, we may focus on maximiz- macro-period d and production line l. All micro-periods are put in
ing profit. However, most of these alternatives imply an exhaustive order s ¼ 1, . . . ,Sl , where Sl corresponds to the total number of
differentiation of costs, freshness values and demand functions micro-periods of line l. It is important to notice that each line is
among products. This can be very inaccurate in the consumer goods assigned to a plant. The length of a micro-period is a decision
industries that need to handle hundreds of stock keeping units. Thus, variable, expressed by the production of several products of one
without demeriting these approaches, to tackle the perishability block in the respective micro-period on a line and by the time to
phenomenon in planning tasks, such as production and distribution, set up the block in case it is necessary. A sequence of consecutive
92 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101

micro-periods, where the same block is produced on the same major setups showing that changing between different blocks triggers
line, defines the size of a lot of a block through the quantity of setups that are dependent on the sequence and, on the other hand,
products produced during these micro-periods. Therefore, a lot minor setups are fixed for a given product to be produced (in this case
may aggregate several products from a given block and may they are all equal).
continue over several micro and macro-periods. Moreover, a lot is In the remainder of this section two cases are formulated:
independent of the discrete time structure of the macro-periods. (1) when the shelf-life is fixed beforehand, and (2) when the
The number of micro-periods of each day defines the upper bound shelf-life is indirectly a decision variable influenced by the
on the number of blocks to be produced daily on each line. environmental setting. For each of these cases an integrated and
There is no inventory held at production plants. Thus, at the a decoupled production and distribution planning model is
end of each day the production output is delivered to distribution presented in order to compare the two different approaches
centers c ¼ 1, . . . ,DC, which have an unlimited storage capacity. afterwards. The environment variables considered for the loose
The delivery function is assured by a 3PL, and we assume that it shelf-life case are quite related to the fresh food industry. Hence,
charges a flat rate per pallet transported between a plant and a for the sake of correctness, we should consider that we are
DC. Moreover, it is assumed that the 3PL is able to cope with focusing on this specific perishable industry. In a latter phase
whatever distribution planning was decided beforehand and, the appropriated generalizations will be introduced.
hence, there is no capacity restriction for transportation. This Consider the following indices, parameters and decision vari-
3PL takes care of all sorts of other decisions besides the quantities ables that are applicable to both studied cases:
to be transported and it complies with the tightest recommenda-
tions and regulations on transportation of perishable goods. The Indices
distances between production plants and distribution centers are
small enough so that the product is delivered on the same day it is l parallel production lines
produced. Therefore, the decrease of freshness during the trans- i,j blocks
portation is considered to be negligible compared with the k products
decrease at the storage process. The small distance assumption t,d,h,b macro-periods: t A ½T; d,hA T þ ; b A T 
is quite realistic in supply chains of highly perishable goods s micro-periods
where the distribution centers are not very far away from the c distribution centers (DCs)
production plants. However, this poses a limitation for the p production plants
application of this model when this assumption is not verified.
For our purposes these assumptions shall not pose a problem Parameters
since we are still considering directly the most important cost
drivers for transportation services: distance, quantity and service Lp set of lines at plant p
level. The demand for an item in a macro-period at a distribution Klj set of products belonging to block j and line l
center is assumed to be dynamic and deterministic. 9K lj 9 number of products belonging to block j and line l
The problem is to plan production and distribution so as to Sld set of micro-periods s within macro-period d for line l
minimize total cost and maximize mean remaining shelf-life of ½dkdc  number of non-zero occurrences in the demand matrix
products at the distribution centers over a planning horizon. Capld capacity (time) of production line l available in
In Fig. 1 a graphical interpretation of the problem dealt with in macro-period d
this paper is presented. It represents the product flow from plants to alk capacity consumption (time) needed to produce one
DCs and the correspondent demand satisfaction. To understand that unit of product k on line l
this is just a layer of the supply network a shadowed replication was clk production costs of product k (per unit) on line l
drawn behind the main scheme. The figure depicts a weekly planning mlj minimum lot-size (units) of block j if produced on line l
horizon and the inventory that is carried linking the consecutive scblij ðstblij Þ sequence dependent setup cost (time) for a change-
planning horizons. Special care was taken when representing the over from block i to block j on line l

t=1 t=2 t=3 t=4 t=5

Demand per
1 9 2 5 3 2 7 8 1 2 4 6 3 4 6 7 2 7 8 9 2 4
product

DC(s)

Plant(s) B C 1 2 3 4 C B C A B C

s =1 s =2 A
Minor setup Major setup
Spoiled
inventory
A, B,...-Blocks

1, 2,...-Products

Fig. 1. Graphical interpretation of the problem statement.

 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101 93

scplk ðstplk Þ sequence-independent setup cost (time) for a change- Capld

qlks r p 8l,k,d,s A Sld ð4Þ
over to product k on line l alk lks
fk cost associated with the spoilage of one unit of product k X X
in inventory stblij zlijs þ ðstplk plks þ alk qlks Þ r Capld 8l,d ð5Þ
tcpc cost for transporting one item from production plant p i,j,s A Sld k,s A Sld

to DC c X
dkdc demand for product k in macro-period d at DC c (units) yljs ¼ 1 8l,s ð6Þ
j
ylj0 equals 1, if line l is set up for block j at the beginning of
the planning horizon (0 otherwise) X
I minimum inventory of product k to be held in macro- qlks Z mlj ðyljs ylj,s1 Þ 8l,j,s ð7Þ
khc
k A K lj
period h at DC c (units)
Rnkbc stock of product k at the beginning of the planning
zlijs Z yli,s1 þ yljs 1 8l,i,j,s ð8Þ
horizon that was produced in macro-period b at DC c
(units), 8bA T  X X
qlks ¼ xkhpc 8k,h,p ð9Þ
l A Lp ,s A Slh c
Decision variables
X X
xkhpc ¼ wkhdc dkdc þRkhc 8k,h,c ð10Þ
Bkdc quantity of stock of product k that spoils in macro- p d
period d at DC c (units)
X
Rkdc quantity of stock of product k produced in macro-period wkbdc dkdc r Rnkbc 8k,c,b ð11Þ
d to be used in the next planning horizon at DC c (units) d
wktdc fraction of demand in macro-period d of product k X
produced in macro-period t at DC c wktdc ¼ 1 8k,d,c : dkdc 4 0; ð12Þ
qlks quantity of product k produced in micro-period s on line t

l (units) X
plks setup state: plks ¼1, if line l is set up for product k in wktdc ¼ 0 8k,d,c : dkdc ¼ 0; ð13Þ
t
micro-period s (0 otherwise)
yljs setup state: yljs ¼1, if line l is set up for block j in micro- X X
khc r
I wktdc dkdc þ Rkdc 8k,h,c ð14Þ
period s (0 otherwise) d 4 h,t r h drh
zlijs takes on 1, if a changeover from block i to block j takes
X
place on line l at the beginning of micro-period s Bkdc Z Rnkbc  wkbhc dkhc 8k,b,d ¼ b þ uk þ1,c ð15Þ
(0 otherwise) h
xkhpc quantity of product k produced in macro-period h
shipped from production plant p to DC c (units) Bkdc , Rkdc , wktdc , qlks , xkhpc , zlijs Z 0;
plks , yljs A f0; 1g ð16Þ

In the first objective (1) total costs are minimized, namely:

3.1. Case 1: fixed shelf-life
production costs, transportation costs and spoilage costs. The
production costs include: sequence dependent setup costs
In the case where shelf-life is fixed, there is only need of
between blocks (major setup), sequence independent setup costs
adding an extra parameter to the list already defined, namely
of products (minor setup) and variable production line costs. This
objective function aggregates the measurable economic impor-
uk shelf-life duration of product k after completion of its
tance throughout the considered supply chain.
production (macro-periods)
In the second objective (2) the mean fractional remaining
shelf-life of products to be delivered is maximized. Hence, the
When we have a fixed shelf-life it is possible to admit that the
remaining shelf-life of a product is expressed by t þ uk d, where t
variable wktdc is only instantiated for d Zt4d rt þ uk to ensure
accounts for the completion date of the product, uk for the shelf-
that demand fulfilled in period d is produced beforehand with
life of the product and d for the delivery date of such order. This
units that have not perished yet. Following the same reasoning
quantity reflects the available time of product k at DC to satisfy
variable Rkdc is only instantiated for d Z Tuk .
retailers’ demand. Dividing this quantity by the initial shelf-life
we obtain the fractional remaining shelf-life. Since we are con-
3.1.1. Integrated model cerned about multiple items with different shelf-lives, by doing
The integrated production and distribution planning of perish- this operation we somehow normalize the impact of shelf-life
able goods with fixed shelf-life (PDP-FSL) may be formulated as a variation among products. To obtain the mean fractional remain-
multi-objective mixed-integer model: PDP-FSL ing shelf-life we need to divide the quantity relative to all orders
X X by the number of occurrences in the demand matrix ½dkdc . This
min scblij zlijs þ ðscplk plks þ clk qlks Þ
l,i,j,s l,k,s
cardinality, for a given input set data, is constant and easily
X X computed. It is important to note that this objective takes values
þ tcpc xkhpc þ fk Bkdc ð1Þ
between 0% and 100%. To further understand the behavior of such
k,h,p,c k,d,c
objective function let us now consider a scenario where a client
! demands two different products. One of the products is very
X t þ uk d 1
max wktdc ð2Þ perishable (1 unit of time) and the other is subject to low
uk ½dkdc 
k,t,d,c perishability (10 units of time). If, after producing and delivering
X them, the very perishable one has 2 days to be sold and the other
subject to : plks ryljs 9K lj 9 8l,s,j ð3Þ one has 5 days, they will contribute equally to this objective
k A K lj function.
94 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101

This multi-objective approach for modeling the integrated products, the storage capacity per temperature zone, also known
production and distribution planning for perishable goods has as climate zone, is the most common. In this case, the storage
an interesting aspect to consider regarding inventory costs. When space is divided in zones according to the temperature require-
maximizing freshness in the second objective we are already ments of different products, for instance, chilled, frozen and
trying to minimize stocks since we will try to produce as late as ambient (for a comprehensive discussion of this topic please refer
possible (note that tardiness is not allowed). Hence, if we had also to Broekmeulen, 2001). However, in this work we are more
included inventory costs in the first objective we would be focused on the customer satisfaction at an operational level, and
somehow duplicating the inventory carrying cost effect. There is hence looking specifically at every product inventory level to be
also a reasoning that comes more from a practical point of view to sure that the demand variability, so common in the consumer
justify the disregarding of inventory carrying costs in operational goods industry, as well as seasonality is well taken into account.
planning of perishable goods. Let us consider the holding cost Ht Finally, (16) defines the domain of the decision variables.
to carry one unit (palette) of inventory from one period t (day) to
period t þ 1. A fair assumption is that the holding cost consists
3.1.2. Decoupled production and distribution model
entirely of interest on money tied up in inventory. Hence, if i is
A decoupled model is designed to mimic a two-stage proce-
the annual interest rate and we consider periods as days (opera-
dure commonly found in industry due to practical organizational
tional planning) then Ht ¼ ick =365, where ck is the average
reasons. In the first stage, the production quantities of products in
production cost of producing a palette of product k. Looking, for
each production plant over the planning horizon, qlks, are deter-
instance, at the yoghurt production where to produce a pack of
mined. Then in the second stage, the delivery quantities, xkhpc, are
4  125 g cups would cost about h0:5 and a palette can hold 1056
found in function of product availability at each production plant.
packs, then for a 5% interest rate, one would have a holding cost
To allow an analytical comparison between the integrated and
per period of about h0:07 per pallet. This cost seems not
decoupled approach the two formulations bellow (PP-FSL and DP-
significant enough compared with the costs of major and minor
FSL) will need to fulfil the same production requirements as the
setups, as well as when compared with the different variable
integrated approach. This will happen when in the integrated
production costs between lines.
approach one does not allow carrying stocks from one planning
Regarding the constraints that bound this model, constraints
horizon to the other. Therefore, production requirements are the
(3) and (4) ensure that a product of a given block can only be
same as final demand requirements.
produced if both the block and the product are set up. Further-
The production planning of perishable goods with fixed shelf-
more, with (6) just one block can be produced on a given line and
life (PP-FSL) may be formulated as a multi-objective mixed-
in a given micro-period. Limited capacity in the lines is to be
integer model: PP-FSL
reduced by setup times between blocks, setup times between X X
products and also by the time consumed in production (5). min scblij zlijs þ ðscplk plks þ clk qlks Þ ð17Þ
Constraint (7) introduces minimum lot-sizes for each block. The l,i,j,s l,k,s

connection between setup states and changeover indicators for !

blocks is established by (8). X h þ uk d X 1
max wkhdc ð18Þ
Constraint (9) ensures that total production in each macro- uk c
½dkdc 
k,h,d
period at every production plant is to be distributed among the
different DCs. Moreover, this constraint links parallel production subject to : ð3Þ2ð8Þ
lines together. X X
Products arriving from different production plants at a DC are qlks Z wkhdc dkdc 8k,h ð19Þ
l,s A Slh d,c
used to meet future demand in the current planning horizon or to
constitute final stock to use in the next planning horizons, as X X
stated by (10). wkhdc ¼ 1 8k,d,c dkdc 40; ð20Þ
h,c c
Constraint (11) ensures that the initial stock in each DC of
products produced before the actual planning horizon can be X X
wkhdc ¼ 0 8k,d,c dkdc ¼ 0; ð21Þ
spent in the current planning horizon to fulfil demand. Never-
h,c c
theless, there is a special concern with perishability and therefore
only products which have not perished are allowed to be used. wkhdc , qlks , zlijs Z 0; plks , yljs A f0; 1g ð22Þ
Constraint (15) is used to calculate the quantity of perished
inventory. In the objective function (17) the cost minimization only
Each day demand is to be met without backlogging with reflects production costs. The objective function (18) aims at
specific production done until that day and with stock left from maximizing also the mean fractional remaining shelf-life of
the past planning horizons (12). Eq. (13) is just needed to ensure products to be delivered. However, it considers an aggregation
that production variables wktdc are zero when the demand at a DC of demand by DC because when we are just planning production
in a period d is null. it is not known which plant will serve a certain demand order at a
In (14) minimum inventory levels for each product in stock are DC. Furthermore, at this stage we cannot take directly into
set per macro-period and per DC. In fact, with the formulation account the possible initial stock at the DCs with a certain
chosen there is no need to explicitly define an inventory decision freshness level.
variable. The stock of product k in a macro-period h is equal to the Constraints (3)–(8) from the PDP-FSL model are to be included
production done until macro-period h to be used in a future in this one. Constraint (19) ensures that total production in each
period in the current planning horizon in addition to the produc- macro-period at every production plant is enough to cover the
tion to stock until macro-period h for each DC. The definition of aggregated future demand, while constraints (20) and (21) have
these bounds, made in an upper hierarchical decision level, is the same meaning as (12) and (13), respectively, but on an
quite important both regarding consumer satisfaction and the aggregated level.
!
quantity of spoiled products. Nevertheless, in terms of storage Given the production quantities, Q ¼ ðqlks Þl,k,s , from the
capacity at the DCs probably this is not the most realistic and production sub-problem, the distribution planning of perishable
important setting for practitioners. For perishable consumer goods with fixed shelf-life (DP-FSL) may be formulated as a
 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101 95

!
multi-objective linear model: DP-FSL ( Q ) u
ktdc spoilage state: u 
ktdc ¼1, if product k produced in macro-
X X period t for meeting demand in macro-period d at DC c is
min tcpc xkhpc þ fk Bkdc ð23Þ spoiled (0 otherwise)
k,h,p,c k,d,c

! Unlike the former case, in Case 2 it is not possible to bound

X t þ uk d 1 decision variable domains based on the precise shelf-life because
max wktdc ð24Þ
k,t,d,c
uk ½dkdc  shelf-life is itself, indirectly, a decision variable. The modeling of
this case grasps this characteristic by making shelf-life dependent
on the deterioration characteristics of the product itself expressed
subject to : ð9Þ2ð15Þ
by Drk and on the environment (temperature) conditions
Bkdc , Rkd , wktdc , xkhpc Z 0 ð25Þ expressed by Yrtc . Nevertheless, it is possible to limit the decision
variables based on a conservative estimation of the product’s
In the objective function (23), the cost minimization encom-
shelf-life u~ k . Hence for wktdc we have d Z t4d r t þ u~ k and Rkbc is
passes transportation and spoilage costs. The objective function
only instantiated for d ZTu~ k . The use of u~ k will be helpful to
(24) aims at maximizing the mean fractional remaining shelf-life
model the freshness objective in the decoupled production model
of products to be delivered at each DC.
dealing with loose shelf-life (PP-LSL).
Constraints (9)–(15) from the PDP-FSL model have the same
meaning as before.
3.2.1. Integrated model
The integrated production and distribution planning of perish-
3.2. Case 2: loose shelf-life able goods with loose shelf-life (PDP-LSL) may be formulated as a
multi-objective mixed-integer nonlinear model: PDP-LSL
In the case where the shelf-life of the perishable products is X X
min scblij zlijs þ ðscplk plks þclk qlks Þ
not fixed, for example, by a stamp tagging the BBD, but rather l,i,j,s l,k,s
depends on different external factors (cf. Section 1), it is impor- X X X
þ tcpc xkhpc þ fk Bkdc þ Yrdc frc ð26Þ
tant to define several other indices, parameters and decision k,h,p,c k,d,c r,d,c
variables to model the production and distribution planning. As
discussed in Section 2, we will rely on the fact that manufacturers !
X 1
þ
can estimate the shelf-life of products throughout the supply max uktdc wktdc ð27Þ
½dkdc 
chain based on external factors using the knowledge of predictive k,t,d,c

microbiology.
We will consider that the goods produced have the same subject to : ð3Þ2ð14Þ
X
temperature requirements and hence the temperature control is Bkdc Z Rnkbc u 
kbdc  wkbhc dkhc 8k,b,d,c ð28Þ
hod
done per DC. This idea follows what has been said in relation to
temperature transportation requirements and product grouping
d X
X
at distribution centers in Section 2. þ
ukhdc þ u
khdc ¼ 1 Yrtc Drk 8k,h,d,c ð29Þ
t4h r
Indices
0 X
X
þ
ukbdc þ u
kbdc ¼ 1 Ynrtc Drk
r discrete temperatures allowed for storing products
t4b r
respecting legal requirements d X
X
 Yrtc Drk 8k,b,d,c ð30Þ
Parameters t¼1 r


u
ktdc Zu ktdc ðdtÞ max Drk ð31Þ
Drk fraction of shelf-life decrease of product k when spend- r
ing a macro-period in stock at temperature r
þ
Ynrbc temperature state selected in macro-period b at DC c,
þ
uktdc ru ktdc 8k,t,d,c ð32Þ
8b A T 
þ
frc cost of keeping DC c at temperature r u ktdc þ u
ktdc r1 8k,t,d,c ð33Þ
u~ k conservative estimation of the shelf-life duration of
þ
product k after completion of its production (macro- wktdc r u ktdc 8k,t,d,c ð34Þ
periods) based on the worst storage conditions X
Yrdc ¼ 1 8d,c ð35Þ
r
Decision variables

kbdc r0;
þ
Bkdc , Rkd , wktdc , qlks , xkhpc , zlijs , ukbdc Z 0; u
Yrdc temperature state: Yrdc ¼1, if DC c is at temperature r þ
plks , yljs , Yrtc , u kbdc , u
kbdc A f0; 1g ð36Þ
in macro-period d (0 otherwise)
þ
uktdc fraction of remaining shelf-life of product k produced in In objective function (26) production, transportation and
macro-period t for meeting demand in macro-period d spoilage costs are taken into account similarly to the case of fixed
at DC c shelf-life. Beyond these costs, the energy cost of keeping the DCs
u
ktdc fraction of violated shelf-life of product k produced in at a certain temperature is also included. Objective function (27)
macro-period t if it was meeting demand in macro- aims at maximizing the mean fractional remaining shelf-life of
period d at DC c products to be delivered. In this case the remaining shelf-life of a
þ þ
u ktdc spoilage state: u ktdc ¼1, if product k produced in macro- certain quantity used to fulfil an order is explicitly expressed by
þ
period t for meeting demand in macro-period d at DC c is uktdc . Due to the uncertain nature of the shelf-life of each product,
not spoiled (0 otherwise) this second objective renders the multi-objective model nonlinear
96 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101

and non-convex, contrarily to the former case (fixed shelf-life,) developed before for the fixed shelf-life case applies to this one
which did not have nonlinearities. as well.
Concerning the constraints this problem is subject to, con- The production planning of perishable goods with loose shelf-
straints (3)–(14) are the same as in PDP-FSL and concern the life (PP-LSL) can be formulated as a multi-objective mixed-integer
production stage. model: PP-LSL
Constraint (28) is a nonlinear constraint used to calculate the X X
min scblij zlijs þ ðscplk plks þ clk qlks Þ ð37Þ
quantity of perished inventory.
l,i,j,s l,k,s
Constraints (29)–(33) define the whole set of possible fractions
þ
of remaining, ukbdc , and violated, u !
kbdc , shelf-lives. Constraints (31) X h þ u~ k d X 1
and (32) ensure that when a product perishes then u  ktdc takes max wkhdc ð38Þ
u~ k ½dkdc 
value 1 and when a product has still some remaining shelf-life k,h,d c
þ
then u ktdc takes value 1, respectively. For the ‘‘big M’’ constraints
(31), a tight value for ‘‘M’’ is given. It is calculated for every subject to : ð3Þ2ð8Þ
combination of d,t,k after getting the maximum value of Drk for a ð19Þ2ð21Þ
given k and then multiplying it by the difference between the wkhdc , qlks , zlijs Z 0; plks , yljs A f0; 1g ð39Þ
demand and the production day. Note that for (32) there is no
need to define the value of ‘‘M’’ since the constraint is already In the objective function (37) the cost minimization only
tight. Constraint (33) is used to ensure that a product either reflects production costs. The objective function (38) aims at
perishes or still has some freshness, since these are completely maximizing the mean fractional remaining shelf-life of products
disjoint physical states. Hence, constraints (29) and (30) define to be delivered but it considers an aggregation of demand by DC.
the amount of remaining or violated fractional shelf-life for However, this objective value is calculated based on a rough
products produced in the current and in the last planning horizon, estimation of the remaining shelf-life at delivery time using u~ k
respectively. For each combination of k,h,d,c (k,b,d,c) either ukhdc þ because in the decoupled approach we do not know at what
þ
(ukbdc ) takes a positive value if the deterioration process has not temperatures the storage will be done afterwards. With this
yet made the product spoiled or u  measure we are able to compute an upper bound of the mean
khdc (ukbdc ) takes a negative
value representing the violation of the shelf-life. These values are remaining shelf-life of products delivered.
defined based on the product inner characteristics and on the Constraints (3)–(8) belong to the PDP-FSL model and regard
profile of temperature storage. production restrictions, whereas constraints (19)–(21) from
Constraint (34) makes sure that only products with some PP-FSL are concerned about satisfying the demand in an aggre-
remaining shelf-life are used to satisfy demand and constraint gated fashion.
!
(35) only lets one temperature to be chosen in each macro-period Given the production quantities, Q , from the production sub-
for each DC. problem, the distribution planning of perishable goods with loose
Finally, constraints (36) define the variable domains. We note shelf-life (DP-FSL) can be formulated as a multi-objective non-
!
linear model: DP-FSL ( Q )
that u kbdc r0.
X X X
In Fig. 2 the behavior of a certain product k that has a loose min tcpc xkhpc þ fk Bkdc þ Yrdc frc ð40Þ
shelf-life is depicted. After production, the product has 100% of k,h,p,c k,d,c r,d,c
remaining shelf-life and, as stated before, throughout the trans-
!
portation process it is able to conserve that freshness. Afterwards, X 1
þ
in the distribution center, if the temperatures are set constantly at max uktdc wktdc ð41Þ
k,t,d,c
½dkdc 
r ¼ 3 or 4 then the products will perish before being delivered to
the retailers. The contrary happens when r ¼ 1 or 2. The dashed
subject to : ð9Þ2ð14Þ
line represents the case when the products are stored in the
beginning with r ¼ 3 but, during their storage, the temperature is ð28Þ2ð35Þ
kbdc r 0;
þ
cooled down to r ¼ 1 ensuring that the products have some Bkdc , Rkd , wktdc , xkhpc , ukbdc Z 0; u
remaining shelf-life before they are delivered. þ
Yrtc , u kbdc , u
kbdc A f0; 1g ð42Þ

In the objective function (40) the cost minimization encom-

3.2.2. Decoupled production and distribution model passes transportation, spoilage, and cooling costs. The objective
As done before, here we present a decoupled model designed function (41) aims at maximizing the fractional remaining shelf-
to mimic a two-stage procedure for the two planning problems of life of products to be delivered at each DC as it was done in
production and distribution planning. The overall reasoning objective function (27).

Fig. 2. Behavior of a product with loose shelf-life regarding remaining lifetime.

 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101 97

Constraints (9)–(14) and (28)–(35) belong to the PDP-FSL and Table 3

PDP-LSL models, respectively. The first group is concerned about Changeover times and costs data for illustrative example.
the distribution function of perishable goods and the second one
Block i Block j Line l stblij scblij
about the determination of the remaining shelf-life of products
delivered. 1 2 1 5 5
1 4 1 10 10
1 2 2 2.5 2.5
1 3 2 5 5
4. Illustrative example 1 4 2 7.5 7.5
2 1 1 10 10
To understand the trade-off present in the two models (fixed 2 4 1 7.5 7.5
2 1 2 7.5 7.5
and loose shelf-life) regarding total costs and product freshness as
2 3 2 2.5 2.5
well as the differences between an integrated over a decoupled 2 4 2 5 5
approach for production and distribution planning an illustrative 3 1 2 10 10
example was developed. 3 2 2 7.5 7.5
In this instance there are four products to be scheduled and 3 4 2 2.5 2.5
4 1 1 15 15
produced on two production lines that are located in two 4 2 1 12.5 12.5
different production plants. Each of these products belongs to a 4 1 2 12.5 12.5
different block and therefore there is always sequence-dependent 4 2 2 10 10
setup time and cost to consider when changing from one product 4 3 2 7.5 7.5
to another. Moreover, although the first line is able to produce
every product, the second one is not able to produce all of the
products. The production lines are considered similar and, there- Table 4
fore, variable production costs are neglected. The number of Shelf-lives and decay rates for the different scenarios.
micro-periods per macro-period was set at the constant value of
Scenario Temperatures Products
four allowing the production of all products in a macro-period.
The capacity of each line is the same in all macro-periods and Fixed shelf-life Loose shelf-life
every production plant (100 units). The planning horizon is 10
days (macro-periods) and the shelf-life of products varies con- 1 2 3 4 1 2 3 4

siderably among them, from highly perishable ones (1 day) to 1 0.111 0.111 0.125 1.000
others which can last throughout the entire planning horizon. Base 2 9 9 8 1 0.100 0.100 0.111 0.500
Demand has to be satisfied in two different DCs and products can 3 0.091 0.091 0.100 0.333
be transported between any pair production plant–DC. Initial stock 1 0.500 0.333 0.500 1.000
was set to zero in both DCs. In case shelf-life is not fixed, there are High perish. 2 2 3 2 1 0.333 0.250 0.333 0.500
three different temperature levels possible to be chosen at each DC 3 0.250 0.200 0.250 0.333
influencing its duration. Finally, a sensitivity analysis regarding 1 0.111 0.111 0.125 0.167
the perishability impact was conducted and different scenarios Low perish. 2 9 9 8 6 0.100 0.100 0.111 0.143
where shelf-lives and decay rates are varied were analyzed 3 0.091 0.091 0.100 0.125
(Table 4). In Tables 1–3 the remainder data of the illustrative
example is given.
The remainder of this section shows the results when solving To complement the explanations of the following sections about
the illustrative example for both cases: fixed and loose shelf-life. how the different problems were solved readers should refer to
Fig. 3.

Table 1 4.1. Results case 1: fixed shelf-life

Demand data for illustrative example.

DC Product Macro-period
In this section results for the case where the shelf-life is fixed
are presented. Since the models developed are either multi-
1 2 3 4 5 6 7 8 9 10 objective mixed-integer or multi-objective linear it was possible
to obtain optimal solutions for this small instance with the help of
1 1 0 40 0 0 0 0 0 0 48 0
CPLEX. For the integrated approach, merely by aggregating both
2 1 0 0 40 0 0 0 0 0 0 48
1 2 30 30 0 0 0 40 48 36 36 0 objectives with varying weights and solving the resulting single-
2 2 40 30 30 0 0 0 0 48 36 36 objective model to optimality it was possible to determine a
1 3 10 0 10 0 0 10 12 12 0 12 variety of solutions from which the Pareto-optimal front could be
2 3 10 0 10 0 0 10 12 12 0 12 constructed. On the other hand, to obtain the results for the
1 4 10 20 30 0 0 0 0 12 24 36
2 4 0 10 20 30 36 0 12 0 0 24
decoupled approach, in order to compare them with the ones of
the integrated approach, a more complex method is needed. First,
solutions on the Pareto-optimal front for the production sub-
problem were determined. Afterwards, for each of the production
Table 2 solutions a new Pareto-optimal front was computed for the
Transportation and temperature related costs data for illustrative example.
distribution sub-problem. To compute the solution of the coupled
DC Production plant Temperatures problem it is necessary to obtain the values for the two objectives
for each pair of connected production–distribution solutions. For
1 2 1 2 3 the first objective, concerning total costs, the summation of each
of the sub-problems’ first objective was considered. Then, for the
1 0.03 0.1 2 2.5 3.25
2 0.15 0.05 1.75 2.25 3
second objective, concerning the mean remaining shelf-life of the
products delivered, the retrieved value of the distribution sub-
98 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101

Fig. 3. Matrix showing a visual representation of the solving strategies for each Case–Approach combination.

Fig. 4. Pareto-optimal fronts of the illustrative example when using an integrated Fig. 5. Relative importance regarding total cost of production and transportation
and a decoupled approach (Case 1). cost using an integrated approach (Case 1).

problem is the one that represents the coupled solution value. In Furthermore, it is interesting to notice that the savings in costs
fact, ultimately, it is the distribution planning that fixes the when using an integrated approach over a decoupled one tend to
product delivery freshness. To obtain the integrated approach fade when we aim at an increased freshness. This may be
Pareto-front we needed about 30 min CPU running time while for explained by the fact that to achieve very high freshness stan-
the decoupled approach, since the problems are less complex, half dards almost no inventory is allowed since we are working under
of the computational time was enough. The two left quadrants of a JIT policy, this will constrain so much the solution space that the
Fig. 3 represent these solution procedures. integrated and coupled solutions are rather the same.
In Fig. 4 the Base scenario solutions of the Pareto-optimal fronts Figs. 5 and 6 concern the relative importance of production
for both the integrated and decoupled approach are presented. and transportation costs for both the integrated and the
It is rather clear from the comparison of the Pareto fronts that decoupled approach in the Base scenario.
the integrated approach strongly dominates the decoupled one. The relative importance of the costs in both approaches is rather
Both curves have a similar behavior, which means that for the independent of the freshness of the products delivered. Only when
lower values of freshness just a small increase in costs fosters approaching a very high performance by delivering products with a
significantly the remaining shelf-life of delivered products. Never- remaining shelf-life around 100% it is possible to see that the two
theless, when we are approaching a strict Just-in-Time (JIT) graphs converge (with an increase of the % Production Costs). When
accomplishment of the demand, touching very high freshness comparing the two approaches, it seems that having a larger view
standards, the costs start to increase in a more important way. over the information in the supply chain (integrated approach), while
 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101 99

4.2. Results case 2: loose shelf-life

In this section we focus on the case where the shelf-life is loose.

Since in this case there are models which are multi-objective mixed-
integer nonlinear it was not straightforward to obtain directly any
solution with the help of exact methods as done before. Therefore, a
simple hybrid genetic heuristic was developed to solve this problem.
This heuristic bases itself on the fact that when fixing Yrtc the model
is no longer non-linear because the fractional remaining shelf-life
þ
uktdc and also u 
kbdc are then known.
Hence, a population of individuals representing the tempera-
ture states, Yrtc , is randomly created ensuring feasibility. At each
generation these values are fed into CPLEX and optimal solutions
Fig. 6. Relative importance regarding total cost of production and transportation for the single-objective model resulting from aggregating both
cost using a decoupled approach (Case 1). objective functions with varying complementary random weights
are obtained. The population is then subject to simple cross-over
and mutation operators. In the end all the individuals are ranked
according to non-domination (Deb et al., 2000) and a Pareto front
is obtained. More details about this heuristic can be found in
Amorim et al. (2011). The number of different combinations for
Yrdc is ðDC:TÞ9r9 , where 9r9 stands for the cardinality of possible
temperatures. Therefore, for this small example, we would have
already 8000 possible solutions that would need to be evaluated
until optimality could be achieved for each of the varying
weightings given to both objective functions. To obtain the results
for the decoupled approach this heuristic had to be inserted in the
distribution planning sub-problem in a scheme very close to the
one used in the decoupled approach for fixed shelf-life (refer to
Fig. 7. Total percentage saving when using an integrated approach over a the right quadrants of Fig. 3). Before, the production planning
decoupled one for three scenarios (Case 1). sub-problem is solved with CPLEX since it has no nonlinearities.
To unveil the near-optimal Pareto front of the integrated
approach we needed about 50 min CPU running time since this
taking operational decisions of production and distribution planning, problem was considerably harder to solve than the one with fixed
leads to better overall costs by means of avoiding the greedy behavior shelf-life. Once again to obtain the Pareto front for the decoupled
of locally optimizing production and then adjust distribution. The approach less computational effort is needed.
decoupled approach has to emphasize quite heavily the costs of the In Fig. 8 the results of the Pareto fronts for both the integrated
transportation process in order to compensate the potential mistakes and decoupled approach are presented. These solutions concern
of the myopic production planning. The integrated approach increases the Base scenario.
the share of the production costs in the total costs in order to have a As it happened with the results of Case 1 the Pareto front
global decrease afterwards. related to the integrated approach strongly dominates the one
Finally, in Fig. 7 we perform a sensitivity analysis regarding the corresponding to the decoupled approach. It is interesting to note
perishability settings. The percentage saving of using an inte- that our simple heuristic was able to capture the non-convexities
grated approach over a decoupled one is plotted for the three of both Pareto fronts. The other reasoning made before for Case 1
scenarios. In order to calculate the saving, both Pareto fronts regarding the behavior of the fronts also applies to this case.
(integrated and decouple approach) were estimated through a As done before for Case 1, in Fig. 9 the results of the sensitivity
second-order polynomial regression which has a good fit to the analysis to understand the effect of different perishability settings
experimental data with all R2 above 90%. are presented. The percentage saving of using an integrated
The potential savings of using an integrated approach over a approach over a decoupled one is plotted for the three scenarios.
decoupled one are rather considerable for the fixed-shelf-life case
and, independently of the scenario, the behavior over the remaining
shelf-life is quite similar. For the scenario with highly perishable
products the savings can ascend up to 42% when aiming at 70% of
remaining shelf-life.
When comparing the three scenarios it is observable that the
advantages of using an integrated approach are leveraged by the
degree of perishability the goods are subject to. In fact, when we
are planning using a decoupled approach and the products are
subject to intense perishability, the myopic mistakes incurred in
production planning will be hardly corrected by the distribution
process because the buffer between those activities is reduced by
the small amount of time that goods can stay stored. Therefore,
the advantages of using an integrated approach are boosted
considerably for this scenario. On the other hand, when dealing
with products with low perishability the buffer enables the
possibility of correcting the potential production mistakes and Fig. 8. Pareto-optimal fronts for the illustrative example when using an integrated
the integrated approach has less comparative advantage. and a decoupled approach (Case 2).
100 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101

both the economic and the freshness perspective a multi-objective

framework was used. Since the models for the loose shelf-life case
were not possible to solve with standard solvers, even for a small
example, a simple heuristic was developed for these cases.
Computational results for an illustrative example show that
the Pareto front of the integrated approach strongly dominates
the Pareto front of the decoupled one for both classes of perish-
able products. The economic savings that this coupled analysis
entail is smoothed as we aim to deliver fresher products. Never-
theless, in the fixed shelf-life case for a 70% mean remaining
shelf-life of delivered products we may reach savings around 42%.
The explanation regarding the fact that the gap between the
Fig. 9. Relative importance regarding total cost of production and transportation
cost using a decoupled approach (Case 2). integrated and the decoupled approach tends to smooth for very
high freshness standards may be due to the reason that in the
latter case no inventory is allowed since we are working com-
Unlike Case 1 the savings are not as bold and the maximum saving pletely under a JIT policy, turning the problem at hand so
ascends to 20% for an average remaining shelf-life of about 65% in the constrained that the integrated and coupled solutions are rather
Base scenario, which is still rather remarkable. Nevertheless, the the same. The multi-objective framework proved to be essential
behavior of both saving curves (from Case 1 and Case 2) is very to draw these multi-perspective conclusions.
similar. The explanation for the difference in the amount of savings This work should be perceived as an exploratory research in
between the fixed and loose shelf-life case may lie in the fact that for this challenging field. Future work should focus on understanding
the loose shelf-life case the distribution process has much more the impact of the integrated approach at the tactical planning
freedom to influence both costs and especially product freshness. level and how these potential benefits and awareness of the cost-
Hence, for the decoupled approach even after the production process freshness relationship can be applied in industrial environments.
has fixed the production quantities, the distribution process is still More in detail, it is important to go deeper into the freshness
able to compensate potential mistakes through the decisions on objective function and evaluate what other indicators besides the
temperature of storage. mean value of the fractional remaining shelf-life can be modeled
Looking at the differences between the three scenarios it is and what are the differences in the plans that they yield.
interesting to notice that in this case the reasoning is not as
straightforward as in Case 1. Here, the two extreme scenarios have
a similar behavior for different reasons. The scenario with products Acknowledgment
subject to low perishability has a rather humble saving when using
an integrated approach for the same reasons as in Case 1. Hence, since The first author appreciate the support of the FCT Project
the time buffer between production and distribution is rather large PTDC/EGE-GES/ 104443/2008 and the FCT Grant SFRH/BD/68808/
the advantages of using an integrated approach are hindered. In the 2010.
scenario having products with a high perishability the explanation for
the relative low saving is related to the possibility of correcting
freshness problems coming from a myopic production planning in the References
decoupled approach through controlling the temperature of storage
in the distribution planning. When products are highly perishable a Ahumada, O., Villalobos, J., 2009. Application of planning models in the agri-food
small decrease in the storage temperature will entail a strong supply chain: a review. European Journal of Operational Research 196, 1–20.
Akkerman, R., Farahani, P., Grunow, M., 2010. Quality, safety and sustainability in
percentage augmentation of shelf-life. Hence, if in a product with food distribution: a review of quantitative operations management
7 days of shelf-life we are able to augment it to 8 days through approaches and challenges. OR Spectrum 32, 863–904.
storing it at cooler temperature, then the percentage increasing of Amorim, P., Antunes, C.H., Almada-Lobo, B., 2011. Multi-objective lot-sizing and
scheduling dealing with perishability issues. Industrial & Engineering Chem-
shelf-life is not very significant. But, if the product is highly perish-
istry Research 50, 3371–3381.
able, then an absolute increase of 1 day will reflect a strong Arbib, C., Pacciarelli, D., Smriglio, S., 1999. A three-dimensional matching model
percentage increase. Therefore, the scenario with products subject for perishable production scheduling. Discrete Applied Mathematics 92, 1–15.
to intermediate perishability (Base scenario) is the one which gains Bilgen, B., Günther, H.O., 2010. Integrated production and distribution planning in
the fast moving consumer goods industry: a block planning application. OR
more from an integrated approach. Spectrum 32, 927–955.
Broekmeulen, R., 2001. Modelling the Management of Distribution Centres.
Lecture Notes in Business Information Processing. Woodhead Publishing Ltd.
Cai, X., Chen, J., Xiao, Y., Xu, X., 2008. Product selection, machine time allocation,
5. Conclusions and scheduling decisions for manufacturing perishable products subject to a
deadline. Computers & Operations Research 35, 1671–1683.
In this paper, we have discussed the importance of integrating the Chen, H.K., Hsueh, C.F., Chang, M.S., 2009. Production scheduling and vehicle
routing with time windows for perishable food products. Computers &
analysis for a production and distribution planning problem dealing Operations Research 36, 2311–2319.
with perishable products. The logistic setting of our operational Deb, K., Agrawal, S., Pratap, A., Meyarivan, T., 2000. A Fast Elitist Non-dominated Sorting
problem is multi-product, multi-plant, multi-DC and multi-period. Genetic Algorithm for Multi-objective Optimisation: NSGA-II. Springer-Verlag,
London, UK.
We have developed models for two types of perishable products: Eksioglu, S.D., Mingzhou, J., 2006. Computational Science and Its Applications—ICCSA
with fixed shelf-life and with loose shelf-life, always taking into 2006. Lecture Notes in Computer Science, vol. 3982. Springer, Berlin, Heidelberg.
account that customers attribute decreasing value to products while Fu, B., Labuza, T., 1993. Shelf-life prediction: theory and application. Food Control
4, 125–133.
they are aging until they completely perish. The novel formulations
Günther, H.O., Grunow, M., Neuhaus, U., 2006. Realizing block planning concepts
allow a comprehensive and realistic understanding of these inter- in make-and-pack production using MILP modelling and SAP APO. Interna-
twined planning problems. Furthermore, the loose-shelf-life model tional Journal of Production Research 44, 3711–3726.
was able to incorporate the possibility of dealing with the underlying Günther, H.O., Seiler, T., 2009. Operative transportation planning in consumer
goods supply chains. Flexible Services and Manufacturing 21, 51–74.
uncertainty of a random spoilage process with the help of predictive Hsu, C.I., Hung, S., Li, H.C., 2007. Vehicle routing problem with time-windows for
microbiology. To understand the impact of the integrated approach in perishable food delivery. Journal of Food Engineering 80, 465–475.
 P. Amorim et al. / Int. J. Production Economics 138 (2012) 89–101 101

Karaesmen, I.Z., Scheller-Wolf, A., Deniz, B., 2011. Managing perishable and aging Neumann, K., Schwindt, C., Trautmann, N., 2002. Advanced production scheduling
inventories: review and future research directions. In: Kempf, K.G., Keskino- for batch plants in process industries. OR Spectrum 24, 251–279.
cak, P., Uzsoy, R., Hillier, F.S. (Eds.), Planning Production and Inventories in the Osvald, A., Stirn, L., 2008. A vehicle routing algorithm for the distribution of fresh
Extended Enterprise, International Series in Operations Research & Manage- vegetables and similar perishable food. Journal of Food Engineering 85, 285–295.
ment Science, vol. 151. Springer, US, pp. 393–436. Pahl, J., Voß, S., 2010. Advanced Manufacturing and Sustainable Logistics.
Lütke Entrup, M., 2005. Advanced planning in fresh food industries. Physica, Lecture Notes in Business Information Processing, vol. 46. Springer, Berlin,
Heidelberg. Heidelberg.
Lütke Entrup, M., Günther, H.O., Van Beek, P., Grunow, M., Seiler, T., 2005. Mixed- Rong, A., Akkerman, R., Grunow, M., 2011. An optimization approach for managing
integer linear programming approaches to shelf-life-integrated planning and fresh food quality throughout the supply chain. International Journal of
scheduling in yoghurt production. International Journal of Production Production Economics 131, 421–429.
Research 43, 5071–5100. Thomas, D., Griffin, P., 1996. Coordinated supply chain management. European
Marinelli, F., Nenni, M.E., Sforza, A., 2007. Capacitated lot sizing and scheduling Journal of Operational Research 94, 1–15.
with parallel machines and shared buffers: a case study in a packaging Tsiros, M., Heilman, C., 2005. The effect of expiration dates and perceived risk on
company. Annals of Operations Research 150, 177–192. purchasing behavior in grocery store perishable categories. Journal of Marketing
Martin, C., Dent, D., Eckhart, J., 1993. Integrated production, distribution, and 69, 114–129.
inventory planning at Libbey–Owens–Ford. Interfaces 23, 68–78. Yan, C., Banerjee, A., Yang, L., 2011. An integrated production–distribution model
Nahmias, S., 1982. Perishable inventory theory: a review. Operational Research 30, for a deteriorating inventory item. International Journal of Production Eco-
680–708. nomics 133, 228–232.