CHAPTER 3

Logistics network planning

3.1 INTRODUCTION

As discussed in Chapter I, logistics encompasses several functions such

as vendor selection, transportation, inventory planning, warehousing
and facilities planning and location. In addition, it is also impacted by
production, marketing and product design decisions. Owing to the
many interrelated functions and interfaces it is difficult to understand
and measure the true logistics costs and to design effective and efficient
logistics systems (Goetschalckx et al., 1995). The requirements or
demands placed on the logistics function have been changing over the
past several years. These changing requirements may be grouped into
the following three categories: competitive pressures, deregulation of
transportation industry and information technology:
• Fierce competition and changing customer requirements dictate
shorter product life-cycles, a larger variety of products and
smaller manufacturing lot sizes. This results in a continuously
changing product mix and a higher inventory of a larger variety
of items, pushing up manufacturing and inventory related costs.
To maintain a lower inventory and still provide a high customer
service level one may adopt just-in-time manufacturing and distri-
bution, resulting in the frequent movement of items in smaller
quantities.
• The deregulation of the transportation industry has created several
new transportation alternatives. This has resulted in innovations in
transportation services, warehousing and storage options and in
increased competition. The deregulation coupled with the increasing
globalization of industries is leading to radically different logistics
strategies.
• The entire field of information technology has been experiencing a
phenomenal growth in terms of both hardware and software.
Computers are available to handle large amounts of data, software
packages are available to solve large and complex logistics planning
and design problems and information and data can be exchanged
and transmitted in real time through Electronic Data Interchange

R. G. Kasilingam, Logistics and Transportation

© Springer Science+Business Media Dordrecht 1998
48 Logistics network planning
(EDI). Identification of shipments and electronic transfer of ship-
ment data can be done by bar codes and radio frequency tags. The
locations of shipments and vehicles can be monitored by satellite.

3.2 LOGISTICS NETWORK MODEL

The changing requirements and environment have created the need for
logistics as a competitive tool to improve customer service and reduce
the total cost of providing customer service. The changes also have
created the need constantly to review and redesign the various logistics
systems and tools used by companies. The design of a logistics system
is based on four major planning areas: customer service levels, location
decisions, inventory planning and transportation management. Custo-
mer service in logistics includes product availability, lead time to obtain
the product, condition of the product when received and accuracy of
filling an order. Location decisions relate to the placement of facilities
such as warehouses, terminals, stores and plants and the assignment of
demands to supply points. Inventory planning encompasses setting up
inventory levels and inventory replenishment schemes. Transportation
management deals with transportation mode, fleet size, route selection,
vehicle scheduling and freight consolidation. All four areas are eco-
nomically interrelated and should be planned in an integrated manner
to achieve maximum benefit. Methodologies and systems that deal with
integrated planning typically are at an aggregate level and do not
include detailed problem definition. Systems and procedures that are
more detailed do not address all four areas simultaneously. The
primary reason is size and complexity.
Logistics network modeling tools attempt to include as much detail
as possible but still address the logistics system design problem in an
integrated manner. Some of the questions answered by an integrated
logistics network model are as follows:
• the number of warehouses, their location, ownership (private or
public) and their size;
• the allocation of customer demand to supply points (warehouses or
plants); allocation to single or multiple supply points;
• the amount of inventory to be maintained at various locations;
• the type of transportation services to use;
• the level of customer service to be provided (that can be provided).
Some of the complexities involved in designing a logistics system with
use of network modeling are as follows:
• the integration of vehicle routing and scheduling;
• the uncertainty in demand, which requires demand to be forecasted;
 Logistics network modeling approach 49
• the identification and development of the appropriate type of cost
functions;
• the dynamic nature of the demand and cost functions over a period
of time;
• dependency relationships between inventory and transportation
decisions;
• the relationship between customer service levels and key logistics
decisions;
• the size of the problem.
Determination of the optimal logistics network configuration is a
fairly complex task because of the large number of vendors and
customers, the hundreds of candidate locations for warehouses and
plants and the extremely large number of transportation options.
Despite these complexities and challenges, several large corporations
periodically analyse their distribution network in an effort to reduce
costs and/or improve customer service. Studies have indicated that
improvements to logistics networks have typically resulted in 5% to
15% savings in logistics costs.

3.3 LOGISTICS NETWORK MODELING APPROACH

Most logistics network design and planning approaches fall into two
categories. One approach is to use very flexible and generic models
which are solved by using commercially available mixed integer
programming software packages. The other approach is to develop
highly specialized models with specialized solution procedures to
arrive at the design. Under the second approach the model is specific to
a given application; hence it represents the design problem in hand
more accurately. Also, efficient solution procedures may be developed
to solve large-scale logistics network design problems in a reasonable
amount of time. However, they require a significant amount of time
and resources to build the model and the solution procedure. In
contrast, generic models are easy to understand and to apply to a
situation. However, they may not represent the problem completely
and accurately. Also, the amount of computational time and memory
may be excessive to solve industrial logistics design problems;
sometimes it may even be impossible to solve the generic models
beyond a certain problem size.
An ideal approach would be to have a logistics network model that is
generic and comprehensive enough to represent the logistics system
with sufficient accuracy but at the same time be amenable to specialized
solution procedures in order to solve large size design problems within
a reasonable amount of computer time. The model must consider the
50 Logistics network planning
movement and storage of raw material or components from raw
material sources to plants either directly or through warehouses and
the movement and storage of finished products or components from
plants to customers either directly or through distribution centers. The
network model represents the various logistics activities as nodes and
arcs. The nodes represent locations of facilities where product flows are
created such as plants, suppliers, warehouses and customers. There
may be several types of warehouses (public or private), vendors and
plants with different capacities, cost structures and technologies. The
arcs represent the flow of raw material, products or components
between the facilities. The flows include several transportation options
with varying transport characteristics and costs. Ideally, the model
must be able to handle multiple facilities (vendors, warehouses, plants
and customers), multiple commodities (raw material and finished
goods) and multiple time periods. The trade-off between inventory and
transportation costs and between customer service and transportation
or inventory costs must be implicitly considered by the model.

3.4 COMPONENTS OF THE LOGISTICS NETWORK MODEL

The basic components of a logistics network are vendors, products,

plants, warehouses or distribution centers, transportation services and
customers. A brief discussion of each of the components is given in this
section.

3.4.1 Vendors
The potential vendors and their locations is one of the inputs to a
network model. Associated with the vendors are the availability of
various products, lead time, quality and price. The locations of the
vendors also dictate the type of transportation services from vendors to
plants. There may be limits on the maximum number of vendors to be
used for a given product. Vendors may offer quantity discounts for
certain products over a certain order size and a group discount if
certain combinations of products are ordered.

3.4.2 Products
A set of products or a group of commodities that move through the
supply chain is one of the inputs to the model. Each commodity has
certain physical characteristics such as weight, size and volume. Also,
each commodity has a certain value and demand. The unit of measure-
ment may be truckload or car load, pallet load or unit load.
 Components of the logistics network model 51
3.4.3 Plants

These are locations where products or commodities are manufactured,

assembled or purchased from. Each plant has a location, the capability
to produce certain products and production capacity restrictions. There
are fixed costs for establishing the facility and variable costs for
manufacturing a product. Also, finished products stored at the plant
incur inventory costs.

3.4.4 Distribution centers

There may be several distribution centers between production facilities

and demand centers as well as between vendor locations and pro-
duction facilities. The distribution centers may also have a hierarchical
structure. For instance, products may be shipped from plants to
regional warehouses, then to smaller, local warehouses and finally to
customers close to the local warehouses. All warehouses have limit-
ations in terms of storage space, type of products that can be handled
and throughput or handling capacity. The resources required to handle
products at different warehouses may be different. The associated costs
of building and operating warehouses are the initial cost of establishing
a warehouse and .the recurring costs to maintain and operate the
warehouse. The number and size of warehouses and the amount of
inventory to be held are some of the major warehouse planning and
operational decisions.

3.4.5 Transportation services

These services include both local delivery and pick-up operations as

well as over-the-road or trunking services. Transportation decisions
include the mode of transportation, shipment size and allocation of
product flow from source to sink to various transportation modes. Each
transportation mode has restrictions in terms of capacity and avail-
ability. Other important characteristics to be considered include transit
time, transit time variability, costs and the number of carriers. Costs
include all fixed and variable transportation costs and in-transit
inventory costs. Movement of products between two places, for instance
from a plant to a warehouse, may be split among multiple transport-
ation modes. Within the same mode, it may be split among different
carriers. The transportation decisions made at the network level are
aggregate in nature and hence will not concern tactical decisions such
as vehicle routing and scheduling.
52 Logistics network planning
3.4.6 Customers
Customer service is measured in terms of product availability and the
lead time required to obtain unavailable products. A customer may be
served by only one warehouse or plant or by more than one plant or
warehouse. Travel time and travel distance are indicators of customer
service. Order processing time along with the actual transit time
determines the lead time. Customer demand is one of the key inputs;
demand must be known in terms of location, time of occurrence,
amount and the type of product.

3.5 GENERALIZED LOGISTICS NETWORK MODEL

In this section, a verbal formulation of a generic logistics network

model is presented. The objective function of a logistics network design
model minimizes the fixed facility costs for warehouses, vendors and
plants, the operating costs for plants and warehouses, transportation
costs, local delivery costs, inventory costs and in-transit inventory costs.
The objective function is optimized with respect to five major categories
of constraints. The first one is related to capacities of plants, warehouses
and transportation carriers. The second is related to the internal
consistency of the model; for example, a warehouse can serve a
customer only if it is built and a plant can supply to a warehouse only
if it is operating. The third set of constraints express customer service
requirements in terms of maximum distance between warehouse and
customer, customer demand and minimum inventory levels or safety
stock. It also includes the requirements of plants in terms of raw
materials or components. The fourth set of constraints ensure
conservation of flow between the various components of the logistics
network. Finally, there are constraints on the decision variables in terms
of their non-negativity and/or integrality. All inputs to the model
should be based on the same time period - day or week..
The modeling of a logistics network requires a complete and accurate
understanding of the various costs and the relationship between them.
Some of the costs may be linear, some may be non-linear and others
may be stepwise linear or non-linear. For example, production costs
may be linear or non-linear up to a certain number of units, after which
an additional fixed cost may have to be incurred. When most of the
production costs are variable the cost per unit will be linear within a
particular range. Non-linear transportation costs such as quadratic cost
functions or exponential cost functions are more realistic (Jara Diaz,
1982) but are much more complex and difficult to model and solve.
Hence non-linear cost functions are typically approximated by piece-
 Generalized logistics network model 53
wise linear costs functions. The precise functional form of costs may be
obtained by using historical data. For instance, regression analysis of
historical warehouse cost data may indicate the following relationship
[equation (3.1)] for warehouse operating cost. The first term on the
right-hand side is the storage cost and the second term on the right-
hand side is the handling cost (note that the warehouse operating cost
is the sum of storage cost and handling cost to meet the demand at
customer locations):
warehouse operating cost = [20 x (storage rate) x (demandt 6 ]
(3.1)
+ [(handling rate) x (demand)]

3.5.1 Model specification

The generalized logistics network model may be formulated as follows.
Minimize Z = vendor establishment cost + plant fixed cost
+ purchase cost for components or raw materials
+ production cost + plant inventory cost
+ warehouse inventory cost + in-transit inventory cost
+ transportation cost + local pick-up or delivery cost
+ warehouse handling cost
(3.2)
Subject to
vendor supply to all plants ~ vendor capacity
plant supply to all warehouses ~ plant capacity
warehouse supply to customers ~ warehouse capacity
transportation between points ~ carrier capacity (weight and volume)
vendor, plant or warehouse can supply only if open
receipt from vendors 2: plant requirements
receipt from plants or vendors 2: warehouse requirements
receipt from warehouses, plants or vendors 2: customer requirements
time for order processing and transportation ~ acceptable lead time
demand exceeding inventory ~ acceptable stock out level
conservation of flow at any point: inbound = outbound + usage
non-negativity and integrality constraints

3.5.2 Generic solution procedure

In most cases, the formulation of the model determines the type of
solution procedure that can be employed, and the type of solution
procedure determines the size of the network model that can be solved,
54 Logistics network planning
computer memory requirements, time to find a solution and the ability
to find an optimal solution (Ballou and Masters, 1993). The model
presented above falls into the category of large-scale mixed integer
programs. It is much more difficult to solve these models than linear or
network flow programs of comparable size. Appropriate preprocessing
methods may reduce some of the redundant constraints and variables.
Preprocessing may also be used to aggregate or combine variables
where feasible. In some cases, the model may be decomposed by plant,
customer, vendor or warehouse to obtain subproblems and a master
problem. In other cases, some of the constraints may be dualized and a
procedure based on Lagrangian relaxation and subgradient opti-
mization may be used. Heuristic procedures always come in handy as
the last resort. For large logistics network planning models (most real-
life problems are very large), optimality may never be achieved under
any of the above methods. Heuristic procedures typically run fast and
use less computer memory. However, with some heuristic methods it
may not be possible to know how far from optimal the solution is.

3.5.3 General guidelines in using the network model

Formulating a model and solving it to develop the logistics network does
not warrant any success. There are three major guidelines that need to be
followed (Napolitano, 1997). The first guideline is to validate the model
before performing any scenario analysis. One approach is to validate
results based on history. Another approach is to validate against
intuition and simple common sense. Sensitivity analysis of changes to
input data is also necessary to test the robustness of the model. Any
major conclusions or decisions must be made only after model
validation. The second guideline is to remember that the solution from
the logistics network model is not the final answer. Human expertise,
judgment and common sense have to be used to modify the answer for
implementation. The most common reasons are that an optimal solution
is best with respect to a chosen model and almost all models fail to
capture reality completely. When they do they are so large and complex
that they have to be solved heuristically, resulting in suboptimal
solutions. In general, it is a lot cheaper and quicker to use existing
software packages for distribution modeling. The loss of exactness is
more than compensated for by savings in cost and time. The last
guideline is that logistics network modeling may not be the best
approach to solve network design problems. In several cases the cost of
modeling may outweigh the associated benefits. In several cases, simple
rules of thumb will address location, size and assignment of warehouse
to customer decisions, mainly because the number of candidate locations
is limited or system constraints arising from contract or legal reasons
limit the available transportation and storage options.
 Problems 55
3.6 SUMMARY

An integrated approach to logistics network planning addresses the

interrelationships and trade-offs between the various functions.
Major decisions in all the functional areas are addressed simul-
taneously rather than hierarchically. This results in better overall
decisions. Logistics network planning is a very complex and difficult
problem to model and solve. Recent advances in operations research
in the areas of mixed integer programming and network modeling
enables us to solve large, real-life logistics network planning prob-
lems through optimal procedures. Also, several commercial software
packages, both generic and specialized, are available to solve large
problems efficiently. Some of the popular generic mixed integer
programming software libraries to solve large problems include
CPLEX and ILOG (details of software and software libraries are
given at the end of this chapter). Examples of specialized logistics
network design packages include the SAILS and CAPS Logistics tool
kit. Almost all of these software packages are available in PC DOS/
Windows and UNIX versions. Logistics network models analytically
represent the flow of products through the companies' supply chain.
It is to be recognized that these models are merely tools. As with
any other approach, good judgment and common sense are still
needed to select the final design.

PROBLEMS

3.1 Show graphically the relationship between inventory cost and transport-
ation cost. Assume that a lower inventory level may be compensated for
by using a premium transportation option to provide the same level of
customer service.
3.2 Sketch the various possible functional forms of the production cost of a
product. Explain under what conditions they are valid.
3.3 Discuss the advantages and disadvantages of using an integrated
logistics network planning methodology.
3.4 List the various decisions involved in designing a logistics network.
3.5 Represent graphically a logistics network with three vendors, two plants
and two warehouses to supply products to six customers.
3.6 For problem 3.5, assuming a single product and linear cost functions,
formulate a mixed integer programming model to represent the network.
State your other assumptions clearly.
3.7 What are the effects of various logistics network components on cycle time?
3.8 Discuss the impact of different logistics network decisions on product
stock out.
56 Logistics network planning
REFERENCES

Ballou, RH. and Masters, J.M. (1993) Commercial software for locating
warehouse and other facilities. Journal of Business Logistics, 14, 71-105.
Goetschalckx, M., Cole, M.H., Dogan, K. and Wei, R (1995) A generic model for
the strategic design of production-distribution systems. Research Report,
School of Industrial and Systems Engineering, Georgia Institute of
Technology, Atlanta, GA.
Jara Diaz, S.R (1982) The estimation of transport cost functions: a methodo-
logical review. Transport Reviews, 2, 257-78.
Napolitano, M. (1997) Distribution network modeling. Industrial Engineering
Solutions, 6, 20-5

FURTHER READING

Aikens, C.H. (1985) Facility location models for distribution planning. European
Journal of Operational Research, 22, 263-79.
Cole, M.H. (1995) Service Considerations and the Design of Strategic Distribution
Systems. Unpublished PhD dissertation, School of Industrial and Systems
Engineering, Georgia Institute of Technology, Atlanta, GA.
Copacino, W. and Rosenfield, D.B. (1985) Analytical tools for strategic planning.
International Journal of Physical Distribution and Materials Management, 15,
47-61.
House, RG. and Karrenbauer, J.J. (1982) Logistics system modeling. International
Journal of Physical Distribution and Materials Management, 12, 119-29.
Perl, J. and Sirisoponslip, S. (1988) Distribution networks: facility location,
transportation, and inventory. International Journal of Physical Distribution
and Materials Management, 18, 18-26.

TRAINING AIDS: SOFTWARE AND SOFTWARE LIBRARIES

Software
ADLnet Software, A.D. Little, Inc., tel. (617) 864-5770.
CAPS Logistics Tool Kit, CAPS Logistics, Inc., tel. (404) 432-9955.
NETWORK, Department of Operations Research, Case Western Reserve
University., tel. (216) 368-3808.
SAILS, Insight Inc., tel. (703) 683-3061.

Software libraries
CPLEX (1997) CPLEX Optimization Inc., Incline Village, NV, USA.
ILOG Optimization (1997) ILOG Ltd., Bracknell, Berks, UK.