ARTICLE IN PRESS

Int. J. Production Economics 107 (2007) 555–571

www.elsevier.com/locate/ijpe

A mixed-integer linear programming model for bulk grain

blending and shipping
Bilge Bilgen, Irem Ozkarahan
Department of Industrial Engineering, Dokuz Eylul University, 35100, Izmir, Turkey
Received 30 December 2004; accepted 6 November 2006
Available online 2 January 2007

Abstract

This paper addresses a blending and shipping problem faced by a company that manages a wheat supply chain. The
problem involves the delivery of bulk products from loading ports to destination ports, which may be served by different
vessel types. Since the products demanded by customers are mainly exported in bulk to overseas customers, the shipment
planning is of great economic importance. The problem is formulated as a mixed-integer linear programming model. The
objective function seeks to minimize the total cost including the blending, loading, transportation and inventory costs.
Constraints on the system include blending and demand requirements, availability of original and blended products; as
well as blending, loading, draft and vessel capacity restrictions. When solved, the model produces: (1) the quantity of each
original product to be used to make blended products, (2) the quantity of each product to be loaded at each port and to be
transported from each port to each customer, and (3) the number of vessels of each type to be hired in each time period.
Numerical results are presented to demonstrate the feasibility of the real world bulk grain blending and shipping model.
r 2006 Elsevier B.V. All rights reserved.

Keywords: Distribution planning; Blending; Maritime transportation; Supply chain management; Mixed integer linear programming

1. Introduction result in significant improvements in economic perfor

This paper was motivated by a problem faced by a
company that manages wheat distribution planning.
In the specific problem, which is the focus of this
paper, products are loaded on bulk vessels of various
capacities for delivery to overseas customers. A vessel
involves a major capital investment, and its daily
operating cost often amounts to several thousands of
dollars. Proper shipment planning may, therefore,
Corresponding author. Tel.: +90 232 3887869;
fax: +90 232 3887864.
E-mail address: bilge.bilgen@deu.edu.tr (B. Bilgen).
mance, which means survival in an increasingly
competitive market. The distribution network con-
sists of a number of loading ports and destination
ports (customers) that may be served by different
vessel types. The vessels under consideration are
hired for a single voyage at a time. In each voyage the
assigned vessel may be loaded in at most two ports,
and discharged in a single destination port, where the
customer is located. Different original products may
be blended at ports prior to loading. The problem is
to assign an appropriate type and number of vessels
to each customer order, while determining the
quantities to be blended, and loaded at each port,
and transported from each port to each customer.

0925-5273/$ - see front matter r 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijpe.2006.11.008
 ARTICLE IN PRESS
556 B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571
The planning process for blending and shipping
should take into account the original and blended
product availability at ports, vessel capacity and
blending requirements, loading, blending and draft
capacity restrictions at ports, and demand satisfac-
tion. This problem is solved to minimize the total
cost, which includes the blending, loading, transpor-
tation and inventory costs. A planning horizon of up
to 3 months has been considered.
A mixed-integer linear programming (MILP)
model is formulated to represent the bulk grain
blending and shipping problem. The model can be
used both as a tool for tactical planning, and a
strategic tool to analyze the effects of cost compo-
nents on the model in various situations. This
paper points out the unique and practical contribu-
tion of the proposed optimization model for
solving a real life bulk grain blending and shipping
problem.
Our mathematical model includes several compo-
nents from traditional optimization models. The
shipping aspect of the distribution system can be
considered as a specialized version of a transship-
ment model. The blending aspect of the problem
falls into the category of a capacity allocation
problem. Finally, we also have a time-expanded
model, as we deal with a multi-period problem.
The paper is organized in the following manner.
The next section provides a review of the related
research. Section 3 describes the distribution-plan-
ning problem in detail. The problem is formulated
as an optimization model in Section 4. In Section 5,
a small example problem is presented to illustrate
the applicability of the model. Computational
results using the data from a real-life case study
are reported in Section 6. Finally, the last section
gives some concluding remarks.
2. Related research
Maritime transportation is the major transporta-
tion mode for international trade. Previous surveys
by Ronen (1983, 1993) indicate that optimization
models for maritime transportation have not been
widely used. In the last decade, a growing body of
advancements concerning several aspects of mar-
itime transportation has appeared in the operations
research literature. In general, the literature has
shown an increase in the number of papers on
ship scheduling and routing (Christiansen et al.,
2004). The problem treated in this paper is not
directly concerned with ship scheduling and routing
problem. For an extensive modeling-oriented dis-
cussion of ship scheduling and routing problems,
the interested reader is referred to several research
works by Ronen (1986), Brown et al. (1987), Sherali
et al. (1999), Christiansen and Fagerholt (2002).
Few studies integrating ship routing with inventory
can be found in Christiansen and Nygreen
(1998a, b), Christiansen (1999), and Ronen (2002).
Little work has been done on maritime shipping
problems with focus on the whole supply chain
(Christiansen et al., 2004; Bilgen and Ozkarahan,
2004). At this point, it is suitable to mention the
literature where maritime shipping is an important
issue within supply chain optimization. Mehrez et
al. (1995) report the modeling and solution of a real
industrial ocean cargo-shipping problem, which
involves the delivery of bulk products from an
overseas port to transshipment ports on the US
Atlantic Coast, and then over land to customers.
The decisions made include the number and size of
ships to be hired in each time period during the
planning horizon, the number and location of
transshipment ports to use, and transportation
quantities from ports to customers. In addition to
solving ocean-shipping problem, the system they
present includes the warehousing decisions, complex
cost structures, and land transportation. Shih (1997)
presents a mixed integer programming (MIP) model
for the planning and scheduling of coal imports
from multiple suppliers for the Taiwan Power
Company. The objective is to minimize total cost
including the procurement, transportation, and
holding costs. Constraints on the system include
company procurement restrictions, power plant
demand, harbor unloading capacity, inventory
balance equations, blending, and safety stock
requirements. The model aims to determine the
amount and shipment schedules of the fuel coal
imports. Liu and Sherali (2000) extended the
problem studied by Shih (1997) and included coal
blending and environmental issues in the mathema-
tical model. They present a MIP zero/one model to
find optimal shipping and blending decisions of coal
fuel from each overseas supplier to each power
plant. They seek to make cost-effective distribution
and allocation decisions while considering supply,
quality and price from each overseas supplier; as
well as demand, quality requirements, supply
resources capacities and presence of blending
facilities at each power plant.
Recently, in the supply chain literature different
industrial problems including decisions concerning
 ARTICLE IN PRESS
B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571
ship routes are studied by Persson and Göthe-
Lundgren (2005), Gunnarsson et al. (2006), Higgins
et al. (2006). Table 1 summarizes the related
literature discussed in this paper.
Of all the previously published literature, the
models developed by Mehrez et al. (1995) and Liu
and Sherali (2000) are the closest to the problem
studied in this paper. As the industry progresses,
better models need to be constructed to incorporate
real world aspects of its operations. In this study, a
multi-index transportation model is developed to
solve the actual case for the blending and shipping
problem. The model developed in this paper differs
from the models presented by those authors in that it
incorporates integrated decisions on blending, load-
ing and shipping for the distribution planning into
one model. A main contribution of the paper is the
development of a novel multi-period optimization
model for solving a real life bulk grain blending and
shipping problem within a supply chain. The follow-
ing section describes the problem considered in detail.
3. Problem description
In this paper, an integrated model of blending
and shipping problem with a focus on the whole
supply chain is studied. This study is concerned with
the shipment planning of products from loading
ports to customers. Original products are supplied
to the loading ports. They may be blended accord-
ing to the blending formulas at ports. They are then
loaded on bulk vessels of various capacities for
delivery to overseas customers. This is a tactical,
deterministic, multi-period bulk grain blending and
shipping problem.
The following considerations further define and
delimit the problem:
 It is assumed that there is an unlimited number of
vessels of each type available. Maritime transport
routes and their corresponding voyage costs by
vessel type are known.
 When necessary, original products are blended at
ports prior to loading to produce new products
as required by customers. Blending cost differs
from port to port.
 Customers are allowed to demand original
products as well as blended products.
 It is assumed that a vessel can pick up products
from at most two loading ports (1 or 2 loading
ports) in a voyage. A vessel is discharged in a
single destination port, where the customer is
557
located. Each vessel is hired for delivering one
shipment of products to one customer, and is not
allowed to discharge cargo at loading ports.
 There is no restriction as to the number of product
types carried on each vessel. It is assumed that
there are enough compartments on vessels to
maintain separation among the products.
 It is assumed that customer’s demand for a
product and the shipment of that product must
take place in the same period.
 Ports have draft constraints, which are repre-
sented in the model as the maximum tonnage
that can be safely carried in and out of a loading
port, and carried into a destination port for a
given vessel class.
 There are many types of vessels with various
sizes. There is a known limited vessel capacity for
each vessel type. For example HandySize 25
carries up to 25 000 Metric Tonnes (MT) and
Panamax60 carries up to 60 000 MT.
 The inventory positions at each loading port are
reviewed at the end of each time period.
A customer order can be more than a full shipload.
Therefore, in the model formulation capacity
parameters are multiplied by the number of vessel
types used. The choice of vessel type to customers
with high demand affects the total cost of distribu-
tion planning, and the size of shipments left for
those customers to be supplied by other vessel types
(Ronen, 1986). For this reason, the variety of vessels
is one of the major components in modeling and
solving this problem.
The model’s input and output are represented in
Fig. 1.
4. Mathematical model
In this section, we present a mathematical model
of the integrated blending and shipping problem.
We first describe the notation used, which is
followed by the description of objective function
and constraints.
4.1. Parameter notation and definitions
The blending and shipping problem can be
formulated as a specialized transshipment model
having some additional side constraints. Before
formulating the model, the basic parameter nota-
tion and definitions are introduced. The notation
given in Table 2 is used for the model development.
 558

Table 1
Summary of the literature

Reviewed literature Objective function Model type (b) The type of Problem type Solution method Decision levels (c)
(a) commodity

Brown et al. (1987) C D Crude oil Routing-scheduling Set partitioning O

Mehrez et al. (1995) C D Bulk Location-allocation- MIP S/T
distribution
Shih (1997) C D Fuel coal Distribution-routing MIP S/T
Christiansen (1999) C D Bulk ammonia Inventory-routing Dantzig Wolfe O
decomposition
Sherali et al. (1999) C D Crude oil Routing-scheduling MIP-based rolling O
horizon heuristic
Liu and Sherali (2000) C D Coal Distribution-routing MIP+heuristic S/T
Ronen (2002) C D Bulk oil Inventory-routing MIP+heuristic S/T
Persson and Göthe-Lundgren (2005) C D Oil Distribution-routing Column generation O
Higgins et al. (2006) C D Sugar Routing-scheduling Meta-heuristics O
Gunnarsson et al. (2006) C D Pulp products Location-allocation- Heuristics S/T
distribution
ARTICLE IN PRESS

(a) C, Cost.
(b) D, Deterministic.
(c) S, Strategic; T, Tactical; O, Operational.
B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571
 ARTICLE IN PRESS
B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571
Demand
Blending Formulas
Blending Capacity
Loading Capacity
Product Routing
Draft Capacity at Loading and
& Blending
Destination ports
Model
Vessel capacity
Blending Cost
Loading Cost
Transportation Cost
Inventory Cost
Model Input
Fig. 1. Model input and output.
The quantity loaded at each port on each vessel
type, on each route, and in each time period must be
kept track of. This requires the development of a
novel variable structure for the model. The variable
xpikqlvt has been used for this purpose. The novelty of
this variable comes from incorporating the informa-
tion into both the port in which the products are
loaded and the port’s location on a route. While the
p (either k or q), indicated with superscript index,
represents the port in which the products are loaded
on vessel type v, the k, indicated with subscript
index, represents the port’s location within a given
marine transportation route. A route starting from
port_A, then visiting port_B and heading for
customer CTZA can be imagined. For this example
port_A is the first loading port, and port_B is the
second loading port, CTZA is the destination port,
on route port_A– port_B– CTZA. A vessel can pick
up products from both loading port_A and port_B,
on that route. To indicate the amount of product
AH loaded on vessel Panamax60 at loading port_A
on route port_A– port_B– CTZA, during time
period 1, the variable xport_A
AH;port_A;port_B;CTZA;Panamax60;1
is used. For this variable, the superscript indexed
‘‘port_A’’ indicates the port, where products are
loaded on vessel ‘‘Panamax60’’. The subscripted
index ‘‘port_A’’ indicates the location of port_A on
559
The amount of product i blended
at port k, during time period t
The amount of original product j
used to make blended producti, at port k,
During time period t
The amount of original and blended
product loaded at port p (k or q), on vessel
v, on route k-q-l, during time period t
The number of vessels of each type v
to be hired, on route k-q-l, during time period t
The amount of remaining stocks of
original product j, at portk, at the end of
time period t
Model Output
that route. port_A is the first loading port on route
port_A– port_B– CTZA. The same vessel ‘‘Pana-
max60’’ can pick up product AH from the second
loading port_B on this route, as well. To indicate
the amount of product AH loaded from port_B,
on route port_A– port_B– CTZA, the variable
xport_B
AH;port_A;port_B;CTZA;Panamax60;1 is used.
4.2. Verbal model
A verbal statement of the model is presented as
follows:
Minimize:
Total Cost ¼ Blending Cost þ Loading Cost
þ Transportation Cost þ Inventory Cost,
Subject to:
Blending constraints,
Availability constraints for original products and
blended products,
Loading capacity,
Vessel capacity constraints,
Draft capacity constraints at loading port(s), and
destination ports for vessel classes,
Customer demand satisfaction.
The problem considered in this paper is concep-
tually simple, but mathematically complex and
 ARTICLE IN PRESS
560 B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571

Table 2
Notation

Notation Remark

Index sets
J Set of original products {1, 2, y, J}
I Set of blended products {1, 2, y, I}
K Set of ports {1, 2, y, K}
L Set of destination ports (customers) {1, 2, y, L}
V Set of vessel types {1, 2, y, V}
T Set of time periods {1, 2, y, T}
M Set of original and blended products {I}+{J}
Input parameters
Dilt Demand for original and blended product i (iAM), by customer l, in time period t
Supplyjk The amount of original products j supplied to port k at the initial period
LCapk Maximum loading capacity at port k (Metric Tonne-MT)
BCapk Maximum blending capacity at port k (MT)
FCapv Maximum vessel capacity for each type v (MT)
DCappkv Draft capacity for vessel type v at port k (MT)
DCapclv Draft capacity for vessel type v at destination port l, where customers are located
TCostkqlv Fixed transportation cost on route kql for vessel type v ($/Vessel)
LCostk Unit loading cost at port k ($/MT)
BCostk Unit blending cost at port k ($/MT)
HCostjk Unit inventory cost for original product j at port k ($/MT)
MinBlndij Minimum ratio of original product j in blended product i
MaxBlndij Maximum ratio of original product j in blended product i

Decision variables
xpikqlvt The amount of original and blended product i loaded at port p (either k or q) on vessel type v on route k– q– l
during time period t
nkqlvt The number of vessels of type v loaded both at port k and q to go to customer l during time period t
wijkt The amount of original product j used to make blended product i at port k during time period t
bikt The amount of product i blended at port k during time period t
Ijkt The amount of remaining stocks of original product j at port k at the end of time period t
IBikt The amount of remaining stocks of blended product i at port k at the end of time period t

challenging. In order to solve such a complicated products are blended according to some formulas.
problem, the formulation of the problem has vital The cost of the blending products at port k per MT
importance. For this reason, the mathematical is denoted by BCostk. The total blending cost of the
formulation of the problem has been emphasized. system is expressed by
The cost components of the objective function and XXX
BlendingCost ¼ BCostk bikt , (1)
each constraint of the model will be discussed in
i2I k2K t
detail in the following section.
where bikt denotes the amount of product i ði 2 IÞ
4.3. Objective function blended at port k during time period t.

The objective function is to minimize the total 4.3.2. Loading cost

cost. The total cost is computed as the sum of the
blending cost, loading cost, transportation cost, and
inventory cost. The different components of the
objective function are expressed below.
4.3.1. Blending cost
Blending occurs only at loading ports. Blending
cost depends on the port in which the original
Loading cost depends on the ports from which
vessels pick up products (original and/or blended).
Vessels can visit a maximum of two loading ports in
a voyage. The loading cost occurs at port p (k and/
or q). The port can be either the first port (k) from
which products are loaded on vessel v on route
k– q– l or the second port (q). Sometimes, vessel v is
forced to travel to the second port due to the
 ARTICLE IN PRESS
B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571 561

unavailability of full tonnage or draft constraints at The total transportation cost of the system is
the first port on the same route k– q– l. The loading obtained by
cost per MT at port p is denoted by LCostp. The XXXXX
TCostkqlv nkqlvt , (4)
total loading cost of the system is expressed by
k q l v t
XXXXXX
LoadingCost ¼ LCostk xkikqlvt where nkqlvt denotes the number of vessels of type v
i2M k q l v t
XXXXXX on route k– q– l during time period t.
þ LCostq xqikqlvt ,
i2M k q l v t
4.4. Constraints
ð2Þ
where the first term expresses the loading cost at the The bulk grain blending and shipping problem
first port, and the second term expresses the loading involves the consideration of:
cost at the second port.
The same cost term can be written in a more (1) blending formulas,
compact way: (2) availability of products (original and blended) at
each port in each time period,
LoadingCost (3) maximum loading capacity at each port,
X XXXXXX
¼ LCostp xpikqlvt , ð20 Þ (4) draft capacity for each loading port and vessel
p2fk;qg i2M k q l v t class, and for each destination port and vessel
class,
where xpikqlvt
denotes the amount of product i ði 2
(5) maximum vessel capacity,
MÞ loaded at port p (either k or q) on vessel type v
(6) each customer’s demand.
on route k– q–l during time period t.

4.3.3. Inventory cost These are discussed under the constraints stated
The inventory cost at each port and for each below.
original product must be expressed. Let HCostjk be
the unit inventory cost of original product j per time 4.4.1. Blending constraints
period at port k. The objective for a blending problem is to find the
XXX best blend of ingredients into final products to meet
InventoryCost ¼ HCostjk I jkt , (3) certain specifications. Several research efforts have
j k t
concentrated on different blending problems using
where Ijkt the amount of original product j that linear programming and MIP (see, e.g. Shih, 1997,
remains at port k at the end of time period t. Liu and Sherali, 2000; Göthe-Lundgren et al., 2002;
Pongsakdi et al., 2006).
4.3.4. Transportation cost In the problem considered in this paper, the
The model takes into account voyage cost, and all product specifications to make the blended product
vessel cost components, including ballast leg cost, that is, the minimum and maximum ratios of
and port charge per call costs. Ballast leg cost is the original product j in blended product i, are given.
cost for vessel traveling from bunker port to the first The constraints below are required to meet the
loading port, including daily hire cost and fuel blended product specifications.
consumption. Port charge per call covers all costs X
against vessel during the time vessel stays at port, wijkt ¼ bikt ; 8i 2 I; k 2 k; t 2 T, (5)
j2J
including daily hire cost, fuel consumption, port
authority and government charges. Voyage cost for wijkt  MaxBlndij bikt p0; 8i 2 I; j 2 J; k 2 K; t 2 T,
a specific route is fixed, regardless of the amount of
products carried. If a vessel travels between ports k, (6)
q and l, the transportation cost consists of ballast wijkt  MinBlndij bikt X0; 8i 2 I; j 2 J; k 2 K; t 2 T.
leg cost to port k, port cost for k, port cost for q,
and voyage cost from k to q and q to l. The sum of (7)
these costs is called the transportation cost. The The capacity restriction regarding blending
transportation cost per route k– q– l per vessel class at ports must be considered. Let the blending
v is denoted by TCostkqlv . capacity at port k be denoted by BCapk. Then
 ARTICLE IN PRESS
562 B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571

the constraint Constraint (11) ensures that there is no initial

X amount of blended product i at port k at the end of
bikt pBCapk ; 8k 2 K; t 2 T, (8) time period 0.
i2I
To ensure the availability of the blended product i
ensures that the amount of products blended must for each port k and each time period t, we have the
be smaller than the blending capacity at each port in following constraints:
each time period. XXX XXX
xkikqlvt þ xkiqklvt
q l v q l v
4.4.2. Availability constraint for original products
and blended products  IBi;k;t1 þ IBi;k;t  bikt ¼ 0,
The constraints stated below ensure the avail- 8i 2 I; 8k 2 K; t 2 T, ð12Þ
ability of products (either original or blended where IBikt represents the amount of blended
product) at ports. product i that remains at port k at the end of time
Supplyjk ¼ I jk0 ; 8j 2 J; k 2 K. (9) period t.
Constraint (9) guarantees that the initial amount
4.4.3. Capacity constraints
of original product at each port is equal to the
Capacity is the upper bound on the total
inventory for that original product j at port k, at the
resources that can be consumed to meet the
end of the time period 0.
XXX XXX customer demand. A number of capacity restric-
xkikqlvt þ xkiqklvt tions, namely loading, draft and vessel capacity
q l v q l v constraints, must be considered.
X
þ wijkt  I j;k;t1 þ I j;k;t ¼ 0,
i2I 4.4.3.1. Loading capacity. The loading capacity
8i 2 J; j 2 J; 8k 2 K; t 2 T. ð10Þ constraint at each port k and in each time period t
can be formulated by the constraint below.
Constraint (10) guarantees the availability of XXXX XXXX
original product j at port k during time period t. xkikqlvt þ xkiqklvt pLCapk ,
The first term ðxkikqlvt Þ in this formulation represents i2M q l v i2M q l v

the amount of original product loaded at port k on 8k 2 K; 8t 2 T,

vessel type v on route kql during time period t.
The second term ðxkiqklvt Þ denotes the amount of
original product loaded at the same port k on
vessel type v but on route q– k– l during time period
t. k is the first loading port on one route (k– q– l), on
the other hand, the same port k is the second
loading port on another route (q– k– l). Two
different routes can be examined as an example,
namely port_B– port_C– CITA and port_A–
port_B– CTZA. While port_B is the first port on
route port_B– port_C– CITA, it is the second
port on route port_A– port_B– CTZA. To calculate
the availability of any original product loaded
from port_B,
P both of the routes must be considered.
The term i2I wijkt denotes the total amount of
original product j used to make different blended
products. Ijkt denotes the amount of original
product j that remains at port k at the end of time
period t.
A similar constraint can be generated for the
products blended.
IBi;k;0 ¼ 0; 8i 2 I; k 2 K.
ð13Þ
where LCapk denotes the maximum loading capa-
city at port k.
Constraint (13) ensures that, at each port and in
each time period, whether this port (k) is the first
port ðxkikqlvt Þ or the second port ðxkiqklvt Þ, the amount
of original and blended products loaded from the
port k cannot exceed the loading capacity for port k.
4.4.3.2. Draft capacity at each loading port for each
vessel class. Ports have draft constraints, which is
represented in the model as the maximum tonnage
that can be safely carried in and out of a loading
port for a given vessel class. As stated earlier, vessels
can visit a maximum of two loading ports in a
voyage. This constraint will be given in two steps.
The constraint at the first port:
X
xkikqlvt  DCappkv nkqlvt p0; 8k; q; l; v; t, (14)
i2M
where DCappkv denotes draft capacity for vessel
(11) type v and at port k.
 ARTICLE IN PRESS
B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571
For each route k– q– l, vessel type v, and time
period t, the amount of original and blended
products loaded onto vessel type v at port k must
be smaller than the draft capacity for that port k
and vessel type v multiplied by the number of type v
vessels used on that route.
4.4.3.3. The constraint at the second port. For each
route k– q– l, vessel type v, and time period t, if
vessel v picks up tonnage from two loading ports,
the amount of original and blended products loaded
at the second port, as well as the first port must be
considered.
The draft capacity constraint at the second port
can be formulated as
X X q
xkikqlvt þ xikqlvt  DCappqv nkqlvt p0,
i2M i2M
8k; q; l; v; t,
where the first term expresses the amount of
products loaded at the first loading port (k) on
route k– q– l during time period t, and the second
term expresses the amount of products loaded at the
second loading port (q) on the same route k– q– l
during time period t. When vessel type v arrives at
the second loading port, it already has a tonnage
loaded from the first loading port.
This constraint can be written in a more compact
way:
X X p
xikqlvt  DCappqv nkqlvt p0; 8k; q; l; v; t.
p2fk;qg i2M
(150 )
The amount of original and blended products
loaded on vessel type v at the first and second
loading ports on the same route must be smaller
than the draft capacity for the second port q and
vessel type v multiplied by the number of type v
vessels used on that route. DCappqv denotes draft
capacity for vessel v at port q.
4.4.3.4. Draft capacity for the destination port
(discharging port), where customers are located.
The draft capacity at the destination port is
expressed as
XXX XXX q
xkikqlvt þ xikqlvt
i2M k q i2M k q
XX
 DCapclv nkqlvt p0;
k q
8l 2 L; 8v 2 V ; t 2 T.
563
It can be written in a more compact way:
X XXX p XX
xikqlvt  DCapclv nkqlvt p0,
p2fk;qg i2M k q k q
8l 2 L; 8v 2 V ; t 2 T, ð160 Þ
where DCapclv denotes draft capacity for vessel type
v, at destination port l.
For each destination port l, vessel type v, and time
period t, the amount of original and blended
products loaded from ports (regardless of which
ports they are loaded from) carried to the destina-
tion l, by means of vessel type v, must be smaller
than the draft capacity for destination port l and
vessel type v multiplied by the number of type v
vessels going to destination l. This constraint
considers all marine transport routes heading for
the same destination port l by using the same vessel
ð15Þ type v.
4.4.4. The required number of vessels of each type
(Vessel capacity)
The capacity constraint on a vessel can be
formulated as
X X
xkikqlvt þ xqikqlvt  FCapv nkqlvt p0; 8k; q; l; v; t.
i2M i2M
(17)
This constraint can be written in a more compact
way:
X X p
xikqlvt  FCapv nkqlvt p0; 8k; q; l; v; t,
i2M p2fk;qg
(170 )
where FCapv denotes the maximum vessel capacity
for vessel type v.
For each given route k– q– l, vessel type v, and
time period t, an integer number of type v vessels are
required to deliver the products loaded from the
first port (k) and the second port (q). The amount of
original and blended products loaded at both ports
must be smaller than the vessel capacity multiplied
by the number of type v vessels used on that route.
4.4.5. Demand constraints
Since customer service requires complete demand
satisfaction, there must be a constraint to make sure
that the customer demand is satisfied. Customer
demand for a product and the shipment of that
product must take place in the same period.
Customers are allowed to demand both original
ð16Þ and blended products. For each original/blended
 ARTICLE IN PRESS
564 B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571

product, customer and time period, the amount of Table 3

product i loaded from the first and second ports Marine transport routes by vessel type
should be equal to the quantity of product i Candidate First Second Customer Vessel type
demanded by customer l, in time period t. The routes port port
demand constraint can now be expressed as
XXX XXX q 1 port_A port_B CKOR Panamax60
xkikqlvt þ xikqlvt ¼ Dilt , 2 port_C port_A CKOR Panamax60
k q v k q v 3 port_B null CKOR Panamax60
4 port_A null CKOR Panamax60
8i 2 M; 8l 2 L; 8t 2 T. ð18Þ
5 port_C null CKOR Panamax60
6 port_A port_B CTZA Panamax60
Similarly this constraint can also be written in a 7 port_B null CTZA Panamax60
more compact way: 8 port_C port_A CTZA Panamax60
X XXX p 9 port_A null CTZA Panamax60
xikqlvt ¼ Dilt , 10 port_C null CTZA Panamax60
p2fk;qg k q v
‘‘null’’ indicates that there is no second loading port on a route
8i 2 M; 8l 2 L; 8t 2 T, ð180 Þ (e.g. products are transported from the first port directly to the
customer).
where Dilt denotes demand for product i, customer l,
in time period t.
The complete model is given in the Appendix A.
(CKOR, CTZA), one vessel type (Panamax60),
Three activities appear to be important in the
problem’s logistical flow: blending, loading, and two original products (APH, AH), one blended
transportation. These activities are defined as product (APH_B1), and only one time period.
follows:
5.1. Objective function
Blending decisions: The amount of products
blended at each port and during time period t 5.1.1. Blending cost
must be determined. Blending decision at each According to the data given above, the blending
port certainly affects the shipping and loading cost can be written as follows:
decisions.
BCostport_A bAPH_B1;port_A;1 þ BCostport_B bAPH_B1;port_B;1
Loading decisions: Vessels can visit a maximum
of two loading ports in a voyage. It has to be þ BCostport_C bAPH_B1;port_C;1 .
decided how much products are to be loaded at
each port on each vessel type on a route, and
during time period t. 5.1.2. Loading cost
Transportation decisions: The number and type of Using the data given in Table 3, the loading cost
vessels to be hired in each time period during the term at port_A, during time period 1 can be written
planning horizon on a route selected out of the as
candidate routes must be determined.
LCostport_A xport_A
AH;port_A;port_B;CKOR;PM60;1
The model formulation used in this study þ LCostport_A xport_A
AH;port_C;port_A;CKOR;PM60;1
integrates these related activities into one model.
þ LCostport_A xport_A
AH;port_A;null;CKOR;PM60;1

5. An illustrative example for model formulation of þ LCostport_A xport_A

AH;port_A;port_B;CTZA;PM60;1
the multi-period blending and shipping problem
þ LCostport_A xport_A
AH;port_C;port_A;CTZA;PM60;1

To provide a better understanding of the pro- þ LCostport_A xport_A

AH;port_A;null;CTZA;PM60;1
posed model, a sample problem is generated for
þ LCostport_A xport_A
APH;port_A;port_B;CKOR;PM60;1
illustration. For the sample problem, marine trans-
port routes for each vessel type are given in Table 3. þ LCostport_A xport_A
APH;port_C;port_A;CKOR;PM60;1
Example: It can be assumed that there are three
ports (port_A, port_B, port_C), two customers þ LCostport_A xport_A
APH;port_A;null;CKOR;PM60;1
 ARTICLE IN PRESS
B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571 565

þ LCostport_A xport_A
APH;port_A;port_B;CTZA;PM60;1
For example, for blended product APH_B1,
port_A, time period 1, the following constraints
þ LCostport_A xport_A
APH;port_C;port_A;CTZA;PM60;1 can be written:
þ LCostport_A xport_A
APH;port_A;null;CTZA;PM60;1
wAPH_B1;AH;port_A;1 þ wAPH_B1;APH;port_A;1 ¼ bAPH_B1;port_A;1 ,
þ LCostport_A xport_A
APH_B1;port_A;port_B;CKOR;PM60;1

þ LCostport_A xport_A
APH_B1;port_C;port_A;CKOR;PM60;1
wAPH_B1;AH;port_A;1  1bAPH_B1;port_A;1 p0; wAPH_B1;AH;port_A;1
þ LCostport_A xport_A
APH_B1;port_A;null;CKOR;PM60;1  0:7bAPH_B1;port_A;1 X0,
þ LCostport_A xport_A
APH_B1;port_A;port_B;CTZA;PM60;1

þ LCostport_A xport_A
APH_B1;port_C;port_A;CTZA;PM60;1 wAPH_B1;APH;port_A;1  0:3bAPH_B1;port_A;1 p0,
þ LCostport_A xport_A
APH_B1;port_A;null;CTZA;PM60;1 . wAPH_B1;APH;port_A;1  0bAPH_B1;port_A;1 X0.

The blending capacity restriction at port_A can

be written as
5.1.3. Inventory cost
The inventory cost for original product AH, at bAPH_B1;port_A;1 pBCapport_A .
port_B can be written as,

HCostAH;port_B I AH;port_B;1 . 5.2.2. Availability constraints

The availability constraint can be written for
port port_A, blended product APH_B1, and time
5.1.4. Transportation cost period 1:
The total transportation cost during time period
1, and for the all transport routes can be written as xport_A
APH_B1;port_A;port_B;CKOR;PM60;1
follows:
þ xport_A
APH_B1;port_A;port_B;CTZA;PM60;1

TCostport_A;port_B;CKOR;PM60 nport_A;port_B;CKOR;PM60;1 þ xport_A

APH_B1;port_C;port_A;CKOR;PM60;1
þ TCostport_C;port_A;CKOR;PM60 nport_C;port_A;CKOR;PM60;1 þ xport_A
APH_B1;port_C;port_A;CTZA;PM60;1
þ TCostport_B;null;CKOR;PM60 nport_B;null;CKOR;PM60;1
þ xport_A
APH_B1;port_A;null;CKOR;PM60;1
þ TCostport_A;null;CKOR;PM60 nport_A;null;CKOR;PM60;1
þ xport_A
APH_B1;port_A;null;CTZA;PM60;1
þ TCostport_C;null;CKOR;PM60 nport_C;null;CKOR;PM60;1
þ TCostport_A;port_B;CTZA;PM60 nport_A;port_B;CTZA;PM60;1  IBAPH_B1;port_A;0 þ IBAPH_B1;port_A;1
þ TCostport_C;port_A;CTZA;PM60 nport_C;port_A;CTZA;PM60;1  bAPH_B1;port_A;1 ¼ 0.
þ TCostport_B;null;CTZA;PM60 nport_B;null;CTZA;PM60;1 The availability constraint can be written for
þ TCostport_A;null;CTZA;PM60 nport_A;null;CTZA;PM60;1 original product AH, port_B, and time period 1,
þ TCostport_C;null;CTZA;PM60 nport_C;null;CTZA;PM60;1 . xport_B port_B
AH;port_B;null;CKOR;PM60;1 þ xAH;port_B;null;CTZA;PM60;1

þ xport_B
AH;port_A;port_B;CKOR;PM60;1
5.2. Constraints
þ xport_B
AH;port_A;port_B;CTZA;PM60;1

5.2.1. Blending constraints þ wAPH_B1;AH;port_B;1  I AH;port_B;0 þ I AH;port_B;1 ¼ 0.

It is assumed that there are two original products
The term wAPH_B1,AH,port_B,1 represents the
available at each port. Minimum and maximum
amount of original product AH used to make
ratios of original products AH and APH in blended
blended product APH_B1 at port_B, during time
product APH_B1 are given as follows:
period 1. If AH was used to make another blended
product (e.g. AH_B2) at port_B during time period
MinBlndAPH_B1;AH : 0:7; MaxBlndAPH_B1;AH : 1
1, the term wAH_B2,AH,port_B,1 should be added
and MinBlndAPH_B1;APH : 0; MaxBlndAPH_B1;APH : 0:3. within this constraint.
 ARTICLE IN PRESS
566 B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571

5.2.3. Loading capacity xport_A

AH;port_A;port_B;CKOR;PM60;1
The following loading capacity constraint can be
written for port_A, and time period 1. þ xport_A
APH;port_A;port_B;CKOR;PM60;1

xport_A þ xport_A
APH_B1;port_A;port_B;CKOR;PM60;1
APH;port_A;port_B;CKOR;PM60;1

þ xport_A þ xport_B
AH;port_A;port_B;CKOR;PM60;1
APH;port_C;port_A;CKOR;PM60;1

þ xport_A þ xport_B
APH;port_A;port_B;CKOR;PM60;1
APH;port_A;null;CKOR;PM60;1
þ xport_B
þ xport_A
APH;port_A;port_B;CTZA;PM60;1
APH_B1;port_A;port_B;CKOR;PM60;1
pDCappport_B;PM60 nport_A;port_B;CKOR;PM60;1 ,
þ xport_A
APH;port_C;port_A;CTZA;PM60;1

þ xport_A where the first three terms represent the amount of

APH;port_A;null;CTZA;PM60;1
products (two original products, and one blended
þ xport_A
AH;port_A;port_B;CKOR;PM60;1 product) loaded at port_A (at the first port), the last
þ xport_A three terms represent the amount of products
AH;port_C;port_A;CKOR;PM60;1
loaded at port_B (at the second port) on route
þ xport_A
AH;port_A;null;CKOR;PM60;1 port_A– port_B– CKOR.
þ xport_A
AH;port_A;port_B;CTZA;PM60;1
5.2.5. Draft capacity for the destination port l, where
þ xport_A
AH;port_C;port_A;CTZA;PM60;1 customers are located
þ xport_A The following constraint can be written, for
AH;port_A;null;CTZA;PM60;1
customer CKOR, vessel type PM60, and time
þ xport_A
APH_B1;port_A;port_B;CKOR;PM60;1 period 1: It is possible to load one type of blended
þ xport_A and two types of original products on vessel PM60.
APH_B1;port_C;port_A;CKOR;PM60;1

þ xport_A
APH_B1;port_A;null;CKOR;PM60;1 xport_A
AH;port_A;port_B;CKOR;PM60;1

þ xport_A
APH_B1;port_A;port_B;CTZA;PM60;1
þ xport_B
AH;port_A;port_B;CKOR;PM60;1

þ xport_A
APH_B1;port_C;port_A;CTZA;PM60;1 þ xport_A
APH;port_A;port_B;CKOR;PM60;1

þ xport_A
APH_B1;port_A;null;CTZA;PM60;1 pLCostport_A .
þ xport_B
APH;port_A;port_B;CKOR;PM60;1

þ xport_A
APH_B1;port_A;port_B;CKOR;PM60;1

þ xport_B
APH_B1;port_A;port_B;CKOR;PM60;1
5.2.4. Draft capacity at each loading port for each
vessel class þ xport_C
AH;port_C;port_A;CKOR;PM60;1
Vessel PM60 can pick up both original and þ xport_A
AH;port_C;port_A;CKOR;PM60;1
blended products (AH, APH, APH_B1) from the
first loading port_A on route port_A– port_B– CK- þ xport_C
APH;port_C;port_A;CKOR;PM60;1
OR. The draft capacity constraint at the first
þ xport_A
APH;port_C;port_A;CKOR;PM60;1
loading port_A can be written for route port_A–
port_B– CKOR, vessel type PM60, and time period þ xport_C
APH_B1;port_C;port_A;CKOR;PM60;1
1 as
þ xport_A
APH_B1;port_C;port_A;CKOR;PM60;1
xport_A port_A
AH;port_A;port_B;CKOR;PM60;1 þ xAPH;port_A;port_B;CKOR;PM60;1 þ xport_B
AH;port_B;null;CKOR;PM60;1
þ xport_A
APH_B1;port_A;port_B;CKOR;PM60;1 þ xport_B
APH;port_B;null;CKOR;PM60;1
pDCappport_A;PM60 nport_A;port_B;CKOR;PM60;1 .
þ xport_B
APH_B1;port_B;null;CKOR;PM60;1
The following constraint should be written to
impose a draft capacity restriction at port_B, and þ xport_A
AH;port_A;null;CKOR;PM60;1
for vessel PM60 on the same route. When vessel þ xport_A
APH;port_A;null;CKOR;PM60;1
PM60 arrives at the second port (port_B), it already
has a tonnage loaded from the first port (port_A). þ xport_A
APH_B1;port_A;null;CKOR;PM60;1
 ARTICLE IN PRESS
B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571 567

þ xport_C
AH;port_C;null;CKOR;PM60;1
in Table 3 must be selected. The last five routes
contain the destination port CTZA.
þ xport_C
APH;port_C;null;CKOR;PM60;1

þ xport_C
APH_B1;port_C;null;CKOR;PM60;1
6. Computational study and results
 DCapcCKOR;PM60 nport_A;port_B;CKOR;PM60;1
 DCapcCKOR;PM60 nport_C;port_A;CKOR;PM60;1 In this section, the computational experiments
 DCapcCKOR;PM60 nport_B;null;CKOR;PM60;1 that were carried out on a real application are
presented. The proposed multi-period bulk grain
 DCapcCKOR;PM60 nport_A;null;CKOR;PM60;1
blending and shipping model described above was
 DCapcCKOR;PM60 nport_C;null;CKOR;PM60;1 p0. tested on a real life project of a company that
exports wheat to overseas customers. An MS-
Access database was chosen to store all necessary
5.2.6. Vessel capacity constraint data. The company has eight original products, two
The following constraint should be written for blended products, four ports, two customers, three
route port_A– port_B– CKOR, vessel type PM60 vessel classes, and three time periods. Information
and time period 1 as regarding the size of the test problem is given in
Table 4. In this specific real life application, ports
xport_A
AH;port_A;port_B;CKOR;PM60;1 can accommodate all types of vessels. In general,
þ xport_B
AH;port_A;port_B;CKOR;PM60;1
not all vessel types would be compatible with all
ports due to draft and length limitations at loading
þ xport_A
APH;port_A;port_B;CKOR;PM60;1 and destination ports. There may be cases where a
þ xport_B
APH;port_A;port_B;CKOR;PM60;1
specific vessel type cannot usually enter a given port
due to draft restrictions. Yet, there is a draft
þ xport_A
APH_B1;port_A;port_B;CKOR;PM60;1 capacity constraint, which is an important con-
þ xport_B
APH_B1;port_A;port_B;CKOR;PM60;1
straint in our model. Draft constraints are also
represented in the model as the maximum tonnage
pFCapPM60 nport_A;port_B;CKOR;PM60;1 .
that can be safely carried in and out of a loading
It is possible that vessel PM60 can pick up port for a given vessel class. The maximum tonnage
tonnage (three types of product—one type blended, a vessel can carry for movement in and out of a
two types of originals) from the first loading port specific port is less than the vessel capacity. The
(port_A) as well as the second loading port (port_B) feasible number of marine transport routes is found
on route port_A– port_B– CKOR. as 22 for each vessel type.
The MILP formulation described in the previous
section was implemented using the modeling lan-
5.2.7. Demand constraints guage ILOG OPL Studio 3.7 (2003), and the
For example, the demand satisfaction constraint problem was solved with standard mathematical
for customer CTZA, original product AH, and time programming software, namely with the branch-
period 1 can be formulated as and-bound algorithm of ILOG CPLEX 8.0 (2003).
xport_A The computational tests were performed on a
AH;port_A;port_B;CTZA;PM60;1
Pentium IV notebook with a 1.4 GHz processor
þ xport_B
AH;port_A;port_B;CTZA;PM60;1 and 512 MB of RAM.
þ xport_C
AH;port_C;port_A;CTZA;PM60;1

þ xport_A
AH;port_C;port_A;CTZA;PM60;1 Table 4
The size of the test problem
þ xport_B
AH;port_B;null;CTZA;PM60;1
Number of original products 8
þ xport_A
AH;port_A;null;CTZA;PM60;1 Number of blended products 2
Number of ports 4
þ xport_C
AH;port_C;null;CTZA;PM60;1 ¼ DCTZA;AH;1 . Number of customers 2
Number of vessel types 3
To write this constraint, the routes containing for Number of time periods 3
the destination port CTZA, out of the routes listed
 ARTICLE IN PRESS
568 B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571

Table 5
The amount of products transported by route, vessel type, and period

Route (k–q–l) Vessel type (v) Period (t) The amount of products loaded on vessel
type v on route k–q–l in time period t

port_D-port_A-CTHA Panamax60 1 54 500

port_C-port_B-CKOR Panamax60 1 55 000
port_D-null-CTHA Panamax60 1 50 000
port_D-port_A-CTHA HandySize25 2 21 000
port_B-null-CKOR Panamax60 2 46 000
port_C-null-CKOR HandyMax45 2 30 000
port_D-null-CTHA HandyMax45 2 38 000
port_B-null-CKOR Panamax60 3 55 000
port_D-null-CTHA HandySize25 3 21 000
port_D-null-CKOR HandySize25 3 21 000

Table 6
Change in total cost increasing the two cost components

Percentage change Increasing loading cost Increasing transportation cost

Minimum total cost Change in total cost (%) Minimum total cost Change in total cost (%)

10 1 47 52 680 2.5 1 54 51 332 6.9

20 1 51 16 152 4.8 1 65 13 457 12.9
30 1 54 79 625 7.1 1 75 75 581 18.2
40 1 57 60 262 9.2 1 85 54 871 22.9
50 1 62 06 570 11.3 1 96 99 831 27.1

To demonstrate the scope of the information
generated by the model, the main results of the
problem are displayed in more detail. Table 5 shows
the amount of original and blended products
transported by route, vessel type, and time period.
To observe the sensitivity of the model to
different operating conditions, the blending, inven-
tory, loading and transportation costs are increased
in order to detect their effects on the total cost. The
blending and inventory costs are not discussed here,
since they are quite small relative to others. OPL
Script Language is used in solving the multiple
instances of the model to analyze the impact of the
data.
The loading and transportation costs are in-
creased by increments of 10%. The resulting total
cost and percentage change in total cost are shown
in Table 6.
It can be seen that the total cost is sensitive to
changes in the transportation cost. Changes in
loading cost have a smaller effect on total cost. A
10% increase in loading cost will cause a 2.5%
increase in total cost. On the other hand, a 10%
increase in transportation cost will cause a 6.9%
increase in total cost. From a practical perspective,
the results obtained from the model are reasonable.
It can be seen that transportation cost has more
influence on total cost than loading cost. Therefore,
transportation cost is an important cost component
in the model’s cost structure. Since ships cost tens of
thousands of dollars, proper shipment planning can
result in large monetary savings. As mentioned
before, it is clear that there is a considerable need
for and the substantial benefit from the integrated
blending and shipping model.
6.1. Input data structure
In this section, the potential complexity reduction
that can be exploited from the data structures in OPL
is determined. The design and the use of the
mathematical programming system depend to a great
extent on the description, manipulation and display
of data (Fourer, 1997). The aim is to present the
model based on different input data structures. These
differ mainly in how they describe the indexing of
model components. OPL offers more support than
other modeling languages in this respect, because it
 ARTICLE IN PRESS
B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571
can use records and arrays indexed by arbitrary finite
sets (ILOG, 2003a, b). The model proposed in this
paper assumes that only a fraction of ports are
connected and a fraction of the products are sent
along the connections. A first attempt at exploiting
sparsity available in our problem consists of repre-
senting marine transportation data as a set of
records type. The other data is mainly characterized
by indexed collection of subsets. The model is built
upon these sets of objects. This model is not
particularly efficient, since it does not exploit the
problem structure explicitly. Some statements iterate
not over actual products and periods, but over all
products and periods. The problem can be modeled
more effectively by closely reflecting the structure of
the application. The model is characterized by a
different way of representing the index sets, in which
indexed collection of subsets are replaced by ordered
pairs, triples, and more indices. Cumulative indexing
and compact data sets are used in the model.
An example can be provided to show the differences
between the two attempts. For instance, in the first
attempt, the x variable was constructed by defining
records including only marine transportation routes
(first port-second port-customer) by each vessel class.
A better data structure is constructed using records
that cluster together all closely related data.
The marine transport routes (first_port–second_
port–customer) by each vessel class have been
already defined within the first attempt. This data is
connected with the supply data that provides the
information related to the availability of products
at each port. The marine transportation routes
connected with the supply data can also be combined
with the demand information by each customer,
product, and period. So, marine transport routes by
each vessel class, port, product, and period would
be obtained. The model precomputes in a generic
way, the marine transport routes that can ship and
receive product i, and in time period t. All summa-
tions run over the allowable combinations of the
indices. The required database structure can be
easily generated by using SQL, within the Ms-Access
database.
Table 7 shows the computational results for the
multi-period blending and shipping model that uses
different input data structures. The comparison
seems to indicate a substantial advantage of the
input data structure in which whole portions of the
model is written in a concise manner. The descrip-
tion of the model by the use of more compact
indexing structure based on the above-mentioned
569
Table 7
Computational comparison of the model based on the two input
data structures
A first attempt A better model
Objective function 1 43 89 207 1 43 89 207
Number of variables 3670 510
Number of constraints 904 589
Solver memory used 52 80 264 7 58 508
Solution time in CPU 10.2 7.54
Number of nodes 305 194
Number of iterations 2387 1155
considerations brings the number of variables to 510
and the number of constraints to 589.
7. Conclusion
A comprehensive model for the real bulk grain
blending and shipping problem has been developed.
The model allows the simultaneous determination
of the number of vessels of each type to be hired by
route and period, determination of the amount of
products blended, and loaded at each port, and
transported from each port by route, vessel type,
and period. The formulation of the real blending
and shipping problem has been emphasized.
This paper makes two primary contributions.
First, and perhaps the most important, it proposes a
model that integrates blending, loading and trans-
portation decisions simultaneously into one model
in order to obtain an optimal solution. Second, a
novel variable structure is developed to represent
the amount of products loaded at each port on each
vessel on each route and during each time period.
A natural extension of the distribution-planning
model is to include land transportation from storage
sites to loading ports. The optimization of the
marine transportation simultaneously with land
transportation is an ongoing project within this
research. Another extension of the model relates to
the consideration of a limited number of vessels.
Adding this to the problem will certainly increase its
complexity and efficient decomposition techniques
are needed to solve it.
In conclusion, we believe that the proposed model
and resulting MIP approach can be used as an
important tool in the strategic and tactical decision
making in the considered industry. This model
provides a reasonable starting point for many
industries where bulk commodity logistics problem
is considered.
 ARTICLE IN PRESS
570 B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571

Acknowledgments A.1.4. Loading capacity constraint

XXXX XXXX
xkikqlvt þ xkiqklvt pLCapk ,
The authors are extremely grateful to the i2M q l v i2M q l v
anonymous referees for their insightful remarks
8k 2 K; t 2 T.
that have significantly helped in revising the manu-
script. The authors also would like to thank Dr.
Koray Dogan for providing us with all data and A.1.5. Draft capacity at first loading port
information in the test case. X
xkikqlvt  DCappkv nkqlvt p0,
i2M
Appendix A. Integrated overall model 8k; q 2 K; 8l 2 L; 8v 2 V ; 8t 2 T.
XXX
BCostk bikt A.1.6. Draft capacity at second loading port
i2I k t X X p
X XXXXXX
þ LCostp xpikqlvt xikqlvt  DCappqv nkqlvt p0,
p2fk;qg i2M k q l v t p2fk;qg i2M
XXX
þ HCostjk I jkt 8k; q 2 K; 8l 2 L; 8v 2 V ; 8t 2 T.
j k t
XXXXX
þ TCostkqlv nkqlvt . A.1.7. Draft Capacity at destination ports, where
k q l v t customers are located
X XXX XX
A.1. Constraints xpikqlvt  DCapclv nkqlvt p0,
p2fk;qg i2M k q k q

A.1.1. Blending constraints 8l 2 L; v 2 V ; t 2 T.

X
wijkt ¼ bikt 8i 2 I; k 2 K; t 2 T,
j2J
A.1.8. The number of vessels type required
X X p
wijkt  MaxBlndij bikt p0 xikqlvt  FCapv nkqlvt p0,
8i 2 I; j 2 J; k 2 K; t 2 T . i2M p2fk;qg
wijkt  MinBlndij bikt X0
8k; q 2 K; 8l 2 L; 8v 2 V ; 8t 2 T.
A.1.2. Blending capacity
X A.1.9. Demand constraint
bikt pBCapk ; 8k 2 K; t 2 T. X XXX p
i xikqlvt ¼ Dilt ,
p2fk;qg k q v
A.1.3. Availability constraints
8i 2 M; l 2 L; t 2 T.
Supplyjk ¼ I jk0 ; 8j 2 J; k 2 K.
XXX XXX A.1.10. Integrality and non-negativity contstraints
xkikqlvt þ xkiqklvt
q
X
l v q l v nkqlvt X0 and integer; xpikqlvt X0; wijkt ,
þ wijkt  I j;k;t1 þ I j;k;t ¼ 0 bikt X0 8i; j; k; q; l; v; t
i2I
8i; j 2 J; 8k 2 K; t 2 T,

IBi;k;0 ¼ 0; 8i 2 I; k 2 K, References
XXX XXX
xkikqlvt þ xkiqklvt  IBi;k;t1 Bilgen, B., Ozkarahan, I., 2004. Strategic, tactical and opera-
q l v q l v tional production and distribution models: A review. Inter-
national Journal of Technology Management 28, 151–171.
þ IBi;k;t  bikt ¼ 0,
Brown, G.G., Graves, G.W., Ronen, D., 1987. Scheduling ocean
8i 2 I; k 2 K; t 2 T. transportation of crude oil. Management Science 33, 335–346.
 ARTICLE IN PRESS
B. Bilgen, I. Ozkarahan / Int. J. Production Economics 107 (2007) 555–571
Christiansen, M., Nygreen, B., 1998a. A method for solving ship
routing problems with inventory constraints. Annals of
Operations Research 81, 357–378.
Christiansen, M., Nygreen, B., 1998b. Modelling path flows for a
combined ship routing and inventory management problem.
Annals of Operations Research 82, 391–412.
Christiansen, M., 1999. Decomposition of a combined inventory
and time constrained ship routing problem. Transportation
Science 33, 3–16.
Christiansen, M., Fagerholt, K., 2002. Robust ship scheduling
with multiple time windows. Naval Research Logistics 49,
611–625.
Christiansen, M., Fagerholt, K., Ronen, D., 2004. Ship routing
and scheduling: Status and perspectives. Transportation
Science 38, 1–18.
Fourer, R., 1997. Database structures for mathematical pro-
gramming models. Decision Support Systems 20, 317–344.
Gunnarsson, H., Rönnqvist, M., Carlsson, D., 2006. A combined
terminal location and ship routing problem. Journal of
Operational Research Society 57, 928–938.
Göthe-Lundgren, M., Lundgren, J.T., Persson, A.A., 2002. An
optimization model for refinery production scheduling.
International Journal of Production Economics 78, 255–270.
Higgins, A., Beashel, G., Harrison, A., 2006. Scheduling of brand
production and shipping with in a sugar supply chain. Journal
of Operational Research Society 57, 490–498.
ILOG, 2003a. ILOG OPL Studio Version 3.7, Language Manual,
ILOG SA.
571
ILOG, 2003b. ILOG CPLEX 8.0 Language Manual, ILOG SA.
Liu, C.-M., Sherali, H.D., 2000. A coal shipping and blending
problem for an electric utility company. Omega 28, 433–444.
Mehrez, A., Hung, M.S., Ahn, B.H., 1995. An industrial ocean-
cargo shipping problem. Decision Sciences 26, 395–423.
Persson, J.A., Göthe-Lundgren, M., 2005. Shipment planning at
oil refineries using column generation and valid inequalities.
European Journal of Operational Research 163, 631–652.
Pongsakdi, A., Rangsunvigit, P., Siemanond, K., Bagajewicz,
M.J., 2006. Financial risk management in the planning of
refinery operations. International Journal of Production
Economics 103, 64–86.
Ronen, D., 1983. Cargo ships routing and scheduling: Survey of
models and problems. European Journal of Operational
Research 12, 119–126.
Ronen, D., 1986. Short-term scheduling of vessels for shipping
bulk or semi-bulk commodities originating in a single area.
Operations Research 34, 164–173.
Ronen, D., 1993. Ship scheduling: The last decade. European
Journal of Operational Research 71, 325–333.
Ronen, D., 2002. Marine inventory routing: Shipments planning.
Journal of Operational Research Society 53, 108–114.
Sherali, H.D., Al-Yakoob, S.M., Hassan, M.M., 1999. Fleet
management models and algorithms for an oil tanker routing
and scheduling problem. IIE Transactions 31, 395–406.
Shih, L.-H., 1997. Planning of fuel coal imports using a mixed
integer programming method. International Journal of
Production Economics 51, 243–249.