Logistics Network Configuration
Designing & Managing the Supply Chain
Chapter 2
Byung-Hyun Ha
bhha@pusan.ac.kr
Outline
 Case: Bis Corporation
 What is Logistics Network Configuration?
 Methodology
 Data Collection
 Modeling and Validation
 Solution Techniques
 Features of Network Configuration DSS
 Summary
Case: Bis Corporation
 Background






Produce & distribute soft drinks
2 manufacturing plant
120,000 account (retailers and stores), all over the US
3 existing warehouse (Chicago, Dallas, Sacramento)
20% gross margin
$1,000 for each SKU (stock-keeping unit) for all products
 Current distribution strategy (designed 15 years ago)




Produce and store at the manufacturing plant
Pick, load, and ship to a warehouse/distribution center
Unload and store at the warehouse
Pick, load, and deliver to store
Case: Bis Corporation
 You, consulting company
 Proposal as reengineering the sales and distribution functions
 First phase, identifying 10,000 direct delivery account, based on
•
•
•
•
•
•
•
•
Dock receiving capabilities
Storage capability
Receiving methodologies
Merchandising requirements
Order-generation capabilities
Delivery time window constraints
Current pricing
Promotional activity patterns
Case: Bis Corporation
 Redesign distribution network
 Grouped accounts into 250 zones, products into 5 families
 Data collected
•
•
•
•
•
Demand in 1997 by SKU per product family for each zone
Annual production capacity at each manufacturing plant
Maximum capacity for each warehouse, new and existing
Transportation costs per product family per mile for distributing
Setup cost for establishing a warehouse
 Customer service level requirement
 No more than 48 hours in delivery
 Additionally,
 Estimated yearly growth, variable production cost, cost for
increasing production capacity, …
Case: Bis Corporation
 Issues





How can Bis Corporation validate the model?
Impact of aggregating customers and products
Number of established distribution centers and their locations
Allocation of plant’s output between warehouses
When and where should production capacity be expanded?
Introduction
 Issues of this chapter
 Development of a model representing logistics network
 Validation of the model
 Aggregation of customers and products and accuracy of the
model
 Number of distribution centers to be established
 Location of distribution centers
 Allocation of output of each product in plants among distribution
centers
 Decision about whether, when, and where to expand production
capacity
Introduction
 Components of logistics network
 Facilities
• Suppliers, warehouse, distribution centers, retail outlets
 Flows
• Raw material, work-in-process inventory, finished products
Sources:
plants
vendors
ports
Regional
Warehouses:
stocking
points
Field
Warehouses:
stocking
points
Customers,
demand
centers
sinks
Supply
Inventory &
warehousing
costs
Production/
purchase
costs
Transportation
costs
Inventory &
warehousing
costs
Transportation
costs
Network Design
 Strategic level – decisions that typically involve
major capital investments and have a long term
effect
 Number, location and size of new plants, distribution centers and
warehouses
 Acquisition of new production equipment and the design of
working centers within each plant
 Design of transportation facilities, communications equipment,
data processing means, …
 Tactical level
 Determine optimal sourcing strategy (strategic?)
• Which plant/vendor should produce which product
 Determine best distribution channels (strategic?)
• Which warehouses should service which customers
 Selection of transportation mode (e.g. rail, truck)
Network Design
 Network design or reconfigure problem
 Objective
• Minimize annual system-wide costs
 Subject to
• Variety of service level requirements
 The objective is to balance service level against




Production/purchasing costs
Inventory carrying costs
Facility costs (handling and fixed costs)
Transportation costs
Network Design
 Tradeoffs
$90
$80
Cost (millions $)
$70
$60
Total Cost
Transportation Cost
Fixed Cost
Inventory Cost
$50
$40
$30
$20
$10
$-
0
2
4
6
Number of Warehouses
8
10
Network Design
 Increasing number of warehouse typically yields





improvement in service level
increase in inventory cost
increase in overhead and setup cost
reduction in outbound transportation costs
increase in inbound transportation costs
Network Design
 Industry benchmarks: average # of warehouses
Pharmaceuticals
3
- High margin product
- Service not important (or
easy to ship express)
- Inventory expensive
relative to transportation
Food Companies
14
Chemicals
25
- Low margin product
- Service very important
- Outbound transportation
expensive relative to inbound
Sources: CLM 1999, Herbert W. Davis & Co; LogicTools
Outline
 Case: Bis Corporation
 What is Logistics Network Configuration?
 Methodology
 Data Collection and Aggregation
 Modeling and Validation
 Solution Techniques
 Features of Network Configuration DSS
 Summary
Data Collection
 Data for network design









Location of customers, stocking points and sources
A listing of all products (volumes, transportation modes)
Demand for each product by customer location
Transportation rates
Warehousing costs
Shipment sizes by product
Order patterns by frequency, size, season, content
Order processing costs
Customer service requirement and goals
Data Aggregation
 Optimization model for the problem?
 Typical soft drink distribution system: 10,000~20,000 accounts
 Wal-Mart or JC Penney: hundreds of thousands!
 Too much
 Data aggregation
 Customer aggregation
 Product aggregation
 Why?




Cost of obtaining and processing data
Form in which data is available
Size of the resulting location model
Accuracy of forecast demand
Data Aggregation: Customer
 Customer aggregation
 Aggregating customers located in close proximity
• Using a grid network or clustering techniques
 All customers within a single zone
• Replaced by a single customer located at the centroid of the zone
 Aggregation by classes
• Service levels, frequency of delivery, …
 Customer zone balances
 accuracy loss due to over aggregation  needless complexity
Data Aggregation: Customer
 Experimental results: cost difference < 0.05%
 Considering transportation costs only
 Customer data
• Original data had 18,000 ship-to locations
• Aggregated data had 800 ship-to locations
• Total demand was the same in both cases
Total Cost:$5,796,000 Total Customers: 18,000
Total Cost:$5,793,000 Total Customers: 800
Data Aggregation: Product
 Product aggregation
 Hundreds to thousands of individual items in production line
• Variations in product models and style
• Same products are packaged in many sizes
 Collecting all data and analyzing it is impractical
 Aggregation by distribution pattern
 Place SKU’s into a source group
• A source group is a group of SKU’s all sourced from the same place
to the same customers
 Aggregate SKU’s by similar logistics characteristics
• Weight, volume, holding cost, …
 Aggregation by product type
 Different products might simply be variations in product style or
differ only in type of packaging
Data Aggregation: Product
 Aggregation by distribution pattern
70.0
60.0
Weight (lbs per case)
50.0
40.0
30.0
20.0
10.0
0.0
0.000
0.010
0.020
0.030
0.040
0.050
0.060
Volume (pallets per case)
0.070
0.080
0.090
0.100
Data Aggregation: Product
 Test case for product aggregation






5 plants
25 potential warehouse locations
Distance-based service constraints
Inventory holding costs
Fixed warehouse costs
Product aggregation
• 46 original products
• 4 aggregated products
• Aggregated products were created using weighted averages
Data Aggregation: Product
 Experimental results: cost difference < 0.05%
Total Cost:$104,564,000
Total Products: 46
Total Cost:$104,599,000
Total Products: 4
Data Aggregation
 Recommended approach
 Aggregate demand points for 150 to 200 zones
• e.g. if customers are classified into classes according to their
service levels or frequency of delivery, each class will have 150-200
aggregated points
 Make sure each zone has an equal amount of total demand
• Zone may be different geographic size
 Place the aggregated point at the center of the zone
 Aggregate products into 20 to 50 product groups
 In this case, the error is typically no more than 1%
 Variability reduction
 Even if technology exists to solve problem with original data,
forecast customer demand at account and product level is
usually poor
Impact of Aggregation on Variability
 Measure of variability?
 Standard deviation (SD)
• Enough?
X
 Xi
n
SD2 
2
(
X

X
)
 i
n
 Which one has bigger SD than the other?
30
400
15
200
0
0
Impact of Aggregation on Variability
 Measure of variability
 Coefficient of variation
CV 
SD
X
 CVA  CVB
30
400
A
B
15
200
0
0
Impact of Aggregation on Variability
 Historical data for the two customers
Year
1992
1993
1994
1995
1996
1997
1998
Customer 1
22,346
28,549
19,567
25,457
31,986
21,897
19,854
Customer 2
17,835
21,765
19,875
24,346
22,876
14,653
24,987
Total
40,181
50,314
39,442
49,803
54,862
36,550
44,841
 Summary of historical data
Statistics
Average
annual demand
Standard deviation
annual demand
Coefficient
of variation
Customer 1
24,237
4,658
0.192
Customer 2
20,905
3,427
0.173
Total
45,142
6,757
0.150
Transportation Rates
 Constructing effective distribution network model
 We should consider reasonable transportation rates
 Important characteristics of most rates
 Rates are almost linear with distance but not with volume
 Rates of internal fleet
 Transportation cost for company-owned trucks
 Calculation of cost per mile per SKU
• Annual costs per truck, annual mileage per truck, annual amount
delivered, truck’s effective capacity
 Rate of external fleet
 Distinguish between truckload (TL) and less than truckload (LTL)
Transportation Rates
 TL carriers
 Subdivision of country into zones
 Zone-to-zone table for cost
 Cost structure is not symmetric (why?)
• e.g. Shipping Illinois  NY is more expensive than in reverse way
 LTL industry
 Types of freight rates
• Class rate (standard)
• Classification tariff based on density, ease of handling, liability
for damage
• Rate base number based on distance
• Exception rate
• Less expensive rate
• Commodity rate
Transportation Rates
 Mileage estimation
 Straight line distance Dab in US from a to b
• Let lonx and latx be longitude of x and latitude of x
Dab  69 (lona  lonb ) 2  (lat a  latb ) 2
 For long distances by correcting for earth’s curvature
Dab  2(69) sin 1
2
  lat  latb  
  lon  lonb  
  sin a
   cos(lata ) cos(latb ) sin a
 
2
2


 
 
2
Warehouse Costs
 Three main components
 Handling costs
• Labor costs, utility costs
• Fairly can be estimated
 Fixed costs
• Cost components that are not proportional to the amount of material
the flows through the warehouse
• Typically proportional to warehouse size (but not linear way)
 Storage costs
Inventory turnover ratio =
annual sales
average inventory level
fixed cost
• Inventory holding cost that are proportional to average positive
inventory level
warehouse size
Warehouse Capacities
 Capacity estimation
inventory
level
 Calculating peak level by assuming regular shipment and
delivery – twice average inventory level
order
size
average
time
 Space for access and handling
• For aisles, picking, sorting, processing facilities, AGVs, …
• Represented as a factor (>1)
Other Issues for Data Collection
 Potential warehouse locations




Geographical and infrastructure conditions
Natural resources and labor availability
Local industry and tax regulations
Public interest
 Service level requirements
 e.g. 95% of customers be situated within 200 miles of the
warehouses serving them
 Future demand
 Network design is at strategic level and impacts on next 3~5
years
 Using scenario-based approach incorporating net present value
Other Issues for Data Collection
 Example of scenario-based approach
 Determine demand and marketing cost of new product
Don’t market
nationally
.60
High local demand
.85
High demand
Market
nationally
.15
Low demand
Don’t market
nationally
Test market
.10
High demand
.40
Low local demand
Market
nationally
.55
High demand
Don’t test
market
Market
nationally
.45
Low demand
Don’t market
nationally
.90
Low demand
Outline
 Case: Bis Corporation
 What is Logistics Network Configuration?
 Methodology
 Data Collection and Aggregation
 Modeling and Validation
 Solution Techniques
 Features of Network Configuration DSS
 Summary
Model and Data Validation
 Model?
 Data validation
 Ensuring data and model accurately reflect the network design
problem
 Done by reconstructing the existing network configuration using
the model and collected data  comparing the output of the
model to company’s accounting information
 Can identify errors in the data, problematic assumptions,
modeling flaws, …
• e.g. transportation cost estimated by model consistently
underestimating actual cost  become to find that effective truck
capacity was only about 30%
 Thus, validation process not only help calibrate parameters but
also suggest potential improvement of existing network
Model and Data Validation
 Sensitivity analysis
 Make local and small changes in model, and estimate impact on
costs and service level
• Positing a variety of what-if question
• e.g. closing the existing warehouse, changing flow of materials
 Can have good intuition about what the effect of small-scale
changes
 Can identify errors in model
 In summary, model validation process involves
answering the following questions:




Does the model make sense?
Are the data consistent?
Can the model results be fully explained?
Did you perform sensitivity analysis?
Solution Techniques
 Techniques for optimizing configuration of logistics
network
 Mathematical optimization techniques
• Exact algorithms: find optimal solutions
• Heuristics: find good solutions, not necessarily optimal
 Simulation models
• provide a mechanism to evaluate specified design alternatives
created by the designer
Heuristics and Exact Algorithms
 E.g. a distribution system





Single product
Two plants p1 and p2
Plant p2 has an annual capacity of 60,000 units
The two plants have the same production costs
There are two warehouses w1 and w2 with identical warehouse
handling costs.
 There are three markets areas c1, c2 and c3 with demands of
50,000, 100,000 and 50,000, respectively
 Distribution cost per unit
Facility
warehouse
p1
p2
c1
c2
c3
w1
0
4
3
4
5
w2
5
2
2
1
2
Heuristics and Exact Algorithms
 A distribution system
$0
D = 50,000
$3
$4
$5
D = 100,000
$2
$4
Cap = 60,000
$5
$2
$1
$2
D = 50,000
Production costs are the same, warehousing costs are the same
Heuristics and Exact Algorithms
 Heuristic 1
 For each market, choose the cheapest warehouse to source
demand. Then, for every warehouse, choose the cheapest plant.
D = 50,000
$5 x 140,000
D = 100,000
$2 x 50,000
Cap = 60,000
$2 x 60,000
$1 x 100,000
$2 x 50,000
Total Costs = $1,120,000
D = 50,000
Heuristics and Exact Algorithms
 Heuristic 2
 For each market area, choose the warehouse such that the total
delivery costs to the warehouse and from the warehouse to the
market is the smallest. (i.e. consider inbound and outbound costs)
$0
D = 50,000
$3
P1 to WH1
P1 to WH2
P2 to WH1
P2 to WH 2
$4
$5
D = 100,000
$2
$4
Cap = 60,000
$5
$2
$3
$7
$7
$4
$1
$2
P1 to WH1
P1 to WH2
P2 to WH1
P2 to WH 2
$4
$6
$8
$3
D = 50,000
P1 to WH1
P1 to WH2
P2 to WH1
P2 to WH 2
$5
$7
$9
$4
Heuristics and Exact Algorithms
 Heuristic 2
 For each market area, choose the warehouse such that the total
delivery costs to the warehouse and from the warehouse to the
market is the smallest. (i.e. consider inbound and outbound costs)
$0 x 50,000
D = 50,000
$3 x 50,000
P1 to WH1
P1 to WH2
P2 to WH1
P2 to WH 2
$5 x 90,000
Cap = 60,000
$2 x 60,000
D = 100,000
$1 x 100,000
$2 x 50,000
Total Cost = $920,000
$3
$7
$7
$4
P1 to WH1
P1 to WH2
P2 to WH1
P2 to WH 2
$4
$6
$8
$3
D = 50,000
P1 to WH1
P1 to WH2
P2 to WH1
P2 to WH 2
$5
$7
$9
$4
Heuristics and Exact Algorithms
 Exact algorithm (linear programming)
 xij: the flow from i to j
min. 0 x p1w1  5 x p1w2  4 x p2 w1  2 x p2 w2
 3xw1c1  4 xw1c2  5 xw1c3  2 xw2c1  xw2c2  2 xw2c3
s.t. x p2 w1  x p2 w2  60,000
x p1w1  x p2 w1  xw1c1  xw1c2  xw1c3
x p1w2  x p2 w2  xw2c1  xw2c2  xw2c3
xw1c1  xw2c1  50,000
xw1c2  xw2c2  100,000
xw1c3  xw2c3  50,000
xij  0 i, j
Total Cost = $740,000
Heuristics and Exact Algorithms
 Network configuration problem is generally
formulated as integer programming
 Hard to obtain the optimal solution
 Some typical types of network design model
 Uncapacitated facility location model
 Capacitated facility location model
 Network optimization model
Source: Camm et al. 1997
Heuristics and Exact Algorithms
 Uncapacitated facility location model
 Example
• Which DC will open and which customer zone will assign to which
DC?
• cij: total cost of satisfying customer zone j demand from DC i
• k: number of DCs allowed
• I: index set of DCs
min.  cij xij
• J: index set of customer zones
iI jJ
• xij = 1 if customer zone j is
s.t.  xij  1 j  J
iI
assigned to DC i, 0 if not
xij  yi i  I , j  J
• yi = 1 if DC i opens, 0 if not
 yi  k
iI
xij , yi  {0, 1} i  I , j  J
Source: Camm et al. 1997
Heuristics and Exact Algorithms
 Capacitated plant location model
 Example: SunOil, a global energy company
• The world is divided into 5 different regions: N. America, S. America,
Europe, Asia, Africa
• SunOil has regional demand figures, transport costs, facility costs
and capacities
• We will ignore tariffs and exchange rate fluctuations for now, and
assume all demand must be met (so we can focus on minimizing
costs)
 Question:
• Where to locate facilities to service their demand
• What size to build in the region (small or large), should they locate a
facility there
Source: Chopra and Meindl 2004
Heuristics and Exact Algorithms
 Capacitated plant location model
 n: number of potential plant location
• As we are considering two different type
plants (small, large) for each region, n = 10





m: number of markets
Dj: demand from market j
Ki: capacity of plant i
fi: fixed cost of keeping plant i open
cij: variable cost of sourcing market j
from plant i
 yi = 1 if plant is located at site i,
= 0 otherwise
 xij: quantity shipped from plant i to
market j
n
n
m
min  f i yi   cij xij
i 1
i 1 j 1
s.t.
n
x
i 1
ij
m
x
j 1
ij
 Dj
 K i yi
for j  1,  , m
for i  1,  , n
yi  {0,1} for i  1,  , n
Heuristics and Exact Algorithms
 Network optimization model
 Example: TelecomOne merged with High Optic
• They have plants in different cities and service several regions
• Supply cities
• Baltimore (capacity 18K), Cheyenne (24K), Salt Lake City (27K),
Memphis (22K) and Wichita (31K)
• Monthly regional demands
• Atlanta (demand 10K), Boston (6K), Chicago (14K), Denver
(6K), Omaha (7K)
• They will consider consolidating facilities
Source: Chopra and Meindl 2004
Heuristics and Exact Algorithms
 Network optimization model





n: number of plant location
m: number of markets
Dj: demand from market j
Ki: capacity of plant i
cij: variable cost of sourcing
market j from plant i
 xij: quantity shipped from plant i to
market j
n
m
min  cij xij
i 1 j 1
s.t.
n
x
i 1
ij
m
x
j 1
ij
 Dj
for j  1,, m
 Ki
for i  1, , n
xij  0
Heuristics and Exact Algorithms

Assignment #3
 Build an MIP model and solve it for the following problem using solver (either
CPLEX or LINDO). Submit the model, code, and solution in printed form.
 DryIce Inc. is a manufacturer of air conditioners that has seen its demand grow
significantly. They anticipate nationwide demand for the year 2010 to be 180,000
units in the South, 120,000 units in the Midwest, 110,000 units in the East, and
100,000 units in the West. Mangers at DryIce are designing the manufacturing
network and have selected four potential sites – New York, Atlanta, Chicago, and
San Diego. Plants could have a capacity of either 200,000 or 400,000 units. The
annual fixed costs at the four locations are shown in the table below, along with
the cost of producing and shipping an air conditioner to each of the four markets.
Where should DryIce build its factories and how large should they be?
New York
Annual
fixed cost
Production &
transportation
cost
Atlanta
Chicago
San Diago
200,000 plant
$6 million
$5.5 million
$5.6 million
$6.1 million
400,000 plant
$10 million
$9.2 million
$9.3 million
$10.2 million
East
$211
$232
$238
$299
South
$232
$212
$230
$280
Midwest
$240
$230
$215
$270
West
$300
$280
$270
$225
Simulation Models
 Limitation of mathematical optimization technique
 Only dealing with static models – cost and demand do not
change over time
 Simulation-based tools
 Taking into account the dynamics of system
 Being capable of characterizing system performance for a given
design
 Simulation for micro-level analysis including
 individual ordering pattern
 specific inventory policy
 inventory movement inside warehouses
Simulation Models

Limitation of simulation


Only evaluate costs associated with a pre-specified logistics
network design
That is, simulation is not an optimization tool
•

Some ways to use simulation for optimization
•

Not useful in determining an effective configuration from a large set
of potential configurations
Employing search technique of determining good parameter for
simulation model
Two-stage approach
1. Use optimization model to generate a number of least-cost
solution at macro-level
2. Use simulation model to evaluate solutions generated in the first
phase
Features of Network Configuration DSS
 Flexibility
 Ability of system to incorporate a large set of preexisting network
characteristics
 One of key requirements of decision-support system (DSS) for
network design
 Necessary to incorporate the following features






Customer-specific service level requirement
Existing warehouses (if lease have not expired, it cannot close)
Expansion of existing warehouses
Specific flow patterns should not be changed
Warehouse-to-warehouse flow
Bill of materials (BOM) (e.g. final assembly is done at a certain
warehouse)
Features of Network Configuration DSS
 Robustness of DSS
 Capability to deal with all issues with little or no reduction in its
effectiveness
 That is, relative quality of the solution generated by DSS should
be independent of specific environment, variability of data, or
particular setting
 Reasonable running time of DSS
 Also have to be robust
Summary
 Issues important in design of logistics network
 Data collection, validation, solution techniques
 Aggregation of data
 Problem size
 Forecast accuracy
 Optimization-based decision-support system
 Considers complex transportation cost structure, warehouse size,
manufacturing limitations, inventory turnover ratios, inventory
cost, service level
 Can solve large-scale problem efficiently
Assignment #4
 Discussion questions 3, 7 (pp. 41–42)