Professional Documents
Culture Documents
Article information:
To cite this document:
David de la Fuente Jess Lozano, (1998),"Determining warehouse number and location in Spain by cluster analysis",
International Journal of Physical Distribution & Logistics Management, Vol. 28 Iss 1 pp. 68 - 79
Permanent link to this document:
http://dx.doi.org/10.1108/09600039810205962
Downloaded on: 13 December 2015, At: 22:10 (PT)
References: this document contains references to 7 other documents.
To copy this document: permissions@emeraldinsight.com
The fulltext of this document has been downloaded 1686 times since 2006*
Access to this document was granted through an Emerald subscription provided by emerald-srm:501757 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors service
information about how to choose which publication to write for and submission guidelines are available for all. Please
visit www.emeraldinsight.com/authors for more information.
IJPDLM
28,1
68
Received June 1996
Revised March 1997,
November 1997
Determining warehouse
number and location in Spain
by cluster analysis
David de la Fuente and Jess Lozano
University of Oviedo, Spain
Introduction
Gross production in the food industry in Spain, which has a workforce of
400,000, was estimated in 1994 at 6,372 billion pesetas, an increase of 1 per cent
over 1993. Because of stiff competition in the European market, the problem of
the best locations for distribution centers in the year 2000 may be the key to the
consolidation of dairy companies. The solution will be to find the ideal number,
geographic position, capacity and size of warehouses to be set up so that their
contribution to profits is maximum. In order to do this, the price of each product
for each market, the products and the quantities that should be supplied from
the factory, the ideal size of each distribution center, as well as the products and
quantities that should be delivered from each warehouse, all need to be
established. The key as regards deciding warehouse locations in this sector
should not only depend on criteria for minimizing costs but also for offering the
best service to the client. In this article we shall consider the former.
Methods for locating factories and distribution centers
Some of the techniques used for global analysis of location are the following:
factor qualification systems (Eilon, 1982), gravitational methods (Douglas and
Moses, 1996), and lineal programming techniques with exact (Sweersey and
Thakur, 1995) or heuristic (Hansen et al., 1994) calculations.
This article assumes that the best way to solve the location problem is by
cluster analysis because we consider the question of assigning a particular
warehouse to a group of towns less important than the question of deciding
precisely which towns should be grouped in a particular zone. The most
important problem solved is that of zoning. Only then are warehouses located
by zone. Thus this is not, nor is intended to be, a comparative study of methods.
A method was formulated that adapted itself satisfactorily to realistic data
(train, road, irregular zones and demand, subcontracting).
Therefore, the analysis does not explicitly include multi-criteria (Pliskin and
Dori, 1982) and market competition (Mateus and Bomstein, 1991)
considerations which may be desirable so as to throw light on short-term
management policy or to help in deciding the sequence to new warehouse
construction (Hansen et al., 1991).
Determining
warehouse
location
69
IJPDLM
28,1
70
Determining
warehouse
location
71
IJPDLM
28,1
72
1a Itera.
|
|
Distance
matrix
10
5.01
10.198
15.033
12.53
14.866
9.849
2 Itera.
|
|
Distance
matrix
0
ab
7.5
5.5
ab
12.59
13.51
9.849
|
|
Distance
matrix
3 Itera.
Table I.
Numerical example of
applying the cluster
algorithm
ab
7.5
5.5
ab
cd
0.5
cd
12.093
|
|
Distance
matrix
4 Itera.
abcd
2.25
2.50
co-ordinates (3 ,0.5). The process is repeated until only one point, in this case
abcd, remains. In this way, costs for different numbers of warehouses with
irregular sizes and zones are calculated.
We have proposed various measurements (for example using Manhattan
distances as opposed to rectilinear distances etc.) of individual differences and
differences between groups, (e.g. furthest neighbors or average distances) as
well as considering the possibility of moving the point that results from the
fusion towards a terminal.
Complements to the cluster analysis
Further considerations that were introduced as research developed to improve
the result of the solution are next commented on; the parameters and specific
(present day) data for entry to the program are also considered in order to
analyze the sensitivity of the result. Since distances are weighted by demand,
and since cost evaluation also needs to be taken into account, some of the
required adaptations do not permit the use of a standard cluster analysis
package.
Determining
warehouse
location
73
IJPDLM
28,1
74
(2) Long distance rail freight costs. If there happens to be a freight handling
terminal within a centroids area of influence, the cost of sending the
merchandise that corresponds to that warehouse by train will be
calculated, to compare it with the cost of road transport calculated in the
previous section (provided that the distance between the freight handling
station and the centroid is less than the maximum distance between the
warehouse and its client towns). This cost will be calculated as the sum
of two terms: the first will be the product of the freight tariff for that
terminal multiplied by the centroids consumption, and the second will
be the freight charges for this consumption from the freight handling
terminal to the warehouse. Whenever rail freight handling transport
costs can be evaluated they will be compared with road transport costs,
and whichever of the two is lower will be opted for.
(3) Storage costs. The total storage cost of a product in a centroid depends
on several factors. First, on the number of days in storage; second, on the
quantity of the product to be distributed from the warehouse (this will be
the total distribution capacity for the whole peninsula weighted
according to the consumption percentage that the warehouse covers; this
percentage will be the sum of the consumption capacity of all the client
towns that depend on that warehouse). Third, on a fixed amount,
showing the pesetas per ton per month cost of storing the merchandise.
Storage costs are thus calculated as the product of these three factors. If
a town has not been assigned to any centroid it is assumed that there will
also be a storage cost charged by an agent working on commission in the
town; in this case, demand assigned to the calculation will simply be the
consumption capacity of the town. Storage costs will be divided into two
parts: distribution to retailer and handling.
Cost of distribution to retailer. The process followed consists of
iteratively calculating the delivery costs between the centroid and the
client towns dependent on it (a correction factor of 1.24 to change
over from linear distances to road distances is applied, as are the
corresponding tariff coefficients). If the nearest town to the centroid
and a client town are in the same province, the provincial transport
tariff will apply. Average storage costs is at present 600 pesetas per
pallet per month.
Handling costs. Handling cost will be the product of the tonnage
associated to the warehouse multiplied by the tariff for handling costs:
2,500 pesetas per pallet (1 pallet = 0.864 tons).
(4) Structural costs. These are costs produced by the size of future
warehouses (connected with the amount of demand for a particular area)
and the number of warehouses considered in the temporary solution
proposed (the partial solution of an iteration). To estimate these costs, the
logical structure of relevant present-day costs in relation to the volume of
sales in litres was studied (amortization was excluded from these costs).
Although initially the adjustment of total costs (mainly staff-generated
and general expenses) was in general of little significance, once
warehouses of lower activity were eliminated results were quite precise.
In this case, 97.32 per cent of variations in expenses between warehouses
and sales volume were explained by a simple linear regression. To
eliminate the problem of linear estimation assigning a very low cost to
smaller warehouses the opportunity of cutting off the estimation
function based on two parameters should be allowed. The first of them
measures the minimum number of sales above which cost estimation by
function, expressed in units per year, will apply. The second is which cost
should be assigned as a fixed minimum when a warehouse fails to reach
a volume of activity (this can be considered as an amount equal to the
amount results from applying the function to the minimum number of
sales, which would mean cutting horizontally the ascending straight line
of costs as a function of sales).
The sum of all these costs for all the warehouses of each iteration will give us
the total cost of that iteration.
Determining
warehouse
location
75
Cost evaluation
Once made, the program provides solutions to cost evaluation for: (a) each
iteration, (b) each distribution area of the iteration. By way of example, for the
costs previously considered see Table II.
As can be seen in Table II, the iteration and the number of warehouses for the
solution are first shown. The second line, (b), shows the number of towns taken
into account and those still pending assignment (in this case none, since all of
them belong to a warehouse), as well as those considered distant, that is, those
outside a radius considered optimum for distribution from the warehouse (124
kilometers). The following line, (c), shows the merchandise ordered in a month
for the whole peninsula; (d) is the percentage cover of all the original towns in
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(h)
(i)
(j)
(k)
(l)
(m)
(n)
Table II.
Solutions to cost
evaluation
IJPDLM
28,1
76
the proposed solution, and (e) is the maximum distance, that is, the furthest
distance from a town to its warehouse. The average distances to towns and
clients then follow (the latter indicates the average distance to a client
supposing the whole population of Spain were concentrated at a single point).
(f) (g) and (h) costs are zero in this solution because no towns are pending
assignment to a warehouse Finally, in (i) structural costs, transport costs (j)
road and/or (k) rail, (i) stock, and (m) delivery to retailers are specified for all
warehouses. (the first three lines of costs would include the costs of towns not
assigned to a warehouse). One can also see the location of warehouses for each
solution in detail in Figure 1. Table III is given as an example.
30.000 t.
315
310
305
300
295
290
285
310.7
306.1
298.5
298.7
299.3
301.2
295.7
17
16
15
No of warehouses
Figure 1.
Location of warehouses
for each solution
300.4
15*
14
13
12
11
In Table III, towns grouped under a warehouse and their distance away (in
brackets) are shown (1). Then, in (2), the warehouse co-ordinates are given; then
(3), the load to be transported each month to the warehouse in question is
shown, along with the amount of cover in relation to overall demand given by
the warehouse. Costs corresponding to the particular warehouse, and whether a
rail freight terminal is advisable or not are both shown in (5),(6) and (7). Finally,
(8) and (9) show delivery costs to retailers and total costs respectively.
Determining
warehouse
location
77
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
No. of
warehouses
Cover
(%)
Towns
pending
Distant
towns
Total cost
(millions of Ptas)
17
16
15
15
14
13
12
11
99.27
99.27
99.27
100
100
100
100
100
1 (City 27)
1 (City 27)
1 (City 27)
0
0
0
0
0
8
11
14
15
17
22
31
34
295,7
298,5
298,7
300,4
299,3
301,2
306,1
310,7
Table III.
Costs relating to
warehouse location
Table IV.
Total costs calculated
from control data
IJPDLM
28,1
78
Table V.
Costs distribution
for different numbers
of warehouses
The asterisks (*) that figure in Table V indicate that in these particular
solutions there are towns pending assignation to a warehouse, so total cost is
not only the sum of what is shown in the table but rather would have to consider
three other extra costs: road or rail direct delivery to the retailer that is not
carried out by the company, and costs of storage by agents working on
commission.
As can be seen in Table V, the optimum solution is 14 warehouses. In view of
the data from the previous table, any solution with a lower cost than the one
proposed might seem advisable for example, the case of 15 warehouses with
towns pending shown in Table V, the cost of which is 298.7 million pesetas.
However, when the final decision is made, the importance of there not being any
towns pending assignment should be underlined (they would function via
agents working on commission). Thus, the above two solutions are qualitatively
different. Moreover, one of the original criteria was to achieve total distribution
through the companys own warehouse network.
In fact, in iterations previous to the ones shown in Table V, total costs are
very much lower, and the basic cause of this is the incompatibility between both
types of structure, one with and the other without total cover.
Another important aspect to consider is the number of distant towns (beyond
the optimum retail distribution radius) that solutions suggest. The lower the
number, the better the solution. Thus, if one solution shows lower costs than
another one, the lower cost one will be better, unless there are more distant
towns, in which case decisions should be taken on the basis of the importance
of demand in the distant towns and the number of them in each solution.
A further important factor lies in the decision to opt for distribution by road
or by rail. To see the recommendations for each solution in the program, an
analysis of the warehouses proposed for the optimum solution (14 warehouses)
should be carried out. It is advisable to analyze final solutions near to the
optimum (between 17 and 11 warehouses), since the final location decision
should be based on rational rather than methodological criteria. We should also
remember that to specify the definitive locations, local legislation, industrial
In-built cost
C.Road T.
C. Rail T.
C. Stock
C. Distr.
Extra C.(*)
Total cost
(Mill. Ptas)
17(*)
16(*)
No. of warehouses
15(*)
15
14
157,2
86,1
26,9
3.0
21,3
1,2
158,4
86,5
27,0
3,0
22,4
1,2
158,4
86,3
27,2
3,0
22,6
1,2
159,8
87,2
27,3
3,1
23,0
295,7
298,5
298,7
300,4
13
12
11
158,9
87,0
27,3
3,1
23,0
159,4
88,1
27,2
3,1
23,4
161,9
85,0
29,8
3,1
26,3
164,5
84,7
30,4
2,9
28,2
299,3
301,2
306,1
310,7
estates, local taxes and so on should all be analyzed. Moreover, a solution with
a larger number of warehouses might provide the customer with a better
service, despite cost increases, and this most certainly provides a major
advantage over competition.
Conclusion
As a result of research using cluster analysis and cost evaluation applied to a
business context in the north of Spain, a method for generating and evaluating
locations for future distribution warehouses is developed. This particular
application is relevant in a wider sense, in that it shows that cluster analysis can
be successfully applied to problems of this general nature.
References
Asoo, J.V. and Wemmerlov, U. (1995), A comparative investigation of hierarchical clustering
techniques and dissimilarity measures applied to the cell formation problem, Journal of
Operational Management, Vol. 13, pp. 117-38.
Douglas, L. and Moses, S. (1996), Strategic planning of transportation services for petroleum
products. An application of capacited gravity models, European Journal of Operational
Research, Vol. 88, pp. 215-30.
Eilon, S. (1982), Multi-criteria warehouse location, International Journal of Physical Distribution
& Materials Management, Vol. 12 No. 1.
Hansen, P.H., Hegedahl, B. and Obel, B. (1994), A heuristic solution to the warehouse locationrouting, European Journal of Operational Research, July 6, Vol. 76 No. 1, p. 111.
Mateus, G.R. and Bomstein, C.T. (1991), Dominance criteria for the capacitated warehouse
location, OR; Journal of the Operational Research Society, February 1, Vol. 42 No. 2, p. 145.
Pliskin, J. and Dori, D. (1982), Ranking alternative warehouse area assignments: a multiattribute
approach, IIE Transactions, Vol. 15 No. 1.
Sweersey, A. and Thakur, L. (1995), An integer programming model for locating vehicle
emissions testing stations, Management Science, March, Vol. 41 No. 3.
Determining
warehouse
location
79
1. Anna Fredriksson, Kristina Liljestrand. 2015. Capturing food logistics: a literature review and research agenda. International
Journal of Logistics Research and Applications 18, 16-34. [CrossRef]
2. T.G. Bosona, G. Gebresenbet. 2011. Cluster building and logistics network integration of local food supply chain. Biosystems
Engineering 108, 293-302. [CrossRef]
3. Karen M. Spens, Gyngyi Kovcs. 2006. A content analysis of research approaches in logistics research. International Journal
of Physical Distribution & Logistics Management 36:5, 374-390. [Abstract] [Full Text] [PDF]