Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Article history: The retail industry sector is one of the most competitive and where each company has to improve their
Received 26 August 2015 operations on a daily basis to remain competitive. The struggle to move to just-in-time delivery requires
Received in revised form 1 August 2016 distribution centres to readapt to this reality. Most of the literature in warehouse layout design is focused
Accepted 21 August 2016
in traditional warehouses, where the main focus is on product storage and picking. However, when oper-
Available online 22 August 2016
ating in a cross-docking basis, new approaches are required to plan the internal layout of the warehouse.
In this paper, it is proposed a mathematical programming approach, based on a min-max formulation
Keywords:
that returns the optimized layout of a cross-docking warehouse that feeds a just-in-time distribution
Warehouse management
Retail operations
operation. In this case, the layout requires the allocation of floor spaces to stores’ demands. Products
Layout are picked up at the receiving dock, and are transported by the worker along the warehouse up to the
Just-in-time location where the products for a given store are located. Then, the products are left in the required quan-
Cross-docking tity and the worker moves to the next store location that requires that product. To this end, we perform
Mathematical programming clusters of floor locations that may be used as locations to visit in product distribution routes. Our
approach was tested in a real world case study of a Portuguese retail chain in a fruits and vegetables
warehouse, which supplies more than 200 stores per day. We show that the distance travelled in the
warehouse can be reduced in more than 2000 km/month by just reallocating stores to different floor
locations.
Ó 2016 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.cie.2016.08.013
0360-8352/Ó 2016 Elsevier Ltd. All rights reserved.
2 M. Horta et al. / Computers & Industrial Engineering 101 (2016) 1–9
Fig. 1. Warehouse shape, dock function and reception and shipping points.
reception point and the minimum distance to one of the ship- 2.3. Model building concepts
ping points;
– (A6) The overall layout of floor locations vs. access areas is The proposed methodology will derive the allocation of stores
known and may have any configuration (i.e., it is not required to floor locations, minimizing the total distance travelled within
to have a specific orientation as in Fig. 1); the warehouse. The distance will be the sum of the picking and
– (A7) During the picking operation, each employee can visit, to shipping distance. In the shipping distance, it is considered that
distribute the contents of one pallet of one product, several each store orders’ are moved from the floor location to the shipping
warehouse locations within the same picking route, fulfilling dock independently from other stores – each store will have mul-
the demand of more than one store. tiple pallets and their movement is done in one or more pallet
truck routes.
2.2. Problem definition On the other hand, the picking distance is more elaborated.
Consider one pallet of boxes of bananas. This pallet will be dis-
The general problem at hand can be defined as stated bellow. tributed, usually, by more than one store. In this way, the picking
Given: distance is obtained by the distance travelled in a route that visits
more than one floor location. In order to overcome the routing
– A set of available floor locations in the warehouse and a set of problem, integrated with the allocation problem, our model is built
stores to allocate to those same locations; over the concept of clusters of floor locations.
– The distance from the reception point to each space floor A cluster of floor locations is a group of N floor locations, that
location; are neighbours (i.e., they are immediately located side-by-side)
– The distance from each space floor location to two shipping and that will be visited by one worker in one picking route. In this
points; way, if the picking starts in floor location a1, it will continue until
– The space capacity, expressed in logistic units (pallets for exam- floor location ax, where x is the number of floor locations in that
ple), of each space floor location for a given time period; cluster. Thus, the distance travelled in this route is the distance
– The space requirements, expressed in logistics units, for each from the reception point to location a1, the linear distance (we only
store for a given time period; assume linear clusters, i.e., all floor locations located in the same
– Number of pickings (products) required for each store for a aisle) of the cluster and the distance from location ax to the recep-
given time period; tion point, so the worker can start a new picking operation. In the
– Number of pallets shipped to each store in a given time period; example of Fig. 2, a cluster with 6 floor locations is exemplified.
– The average number of pickings per pallet for a given time In order to define the number of floor locations per each cluster,
period; we require specific data from the problem at hand. This data will
enable to calculate, for the period of data gathered, the average
Determine: number of stores whose demand is fulfilled from one pallet. With
this data, a fixed size of clusters is defined. The required data
– The assignment of the stores within the warehouse, defining the encompasses the following items: (i) the list of stores to whom
allocation of the stores to the available warehouse floor the demand has to be fulfilled and adequately numbered, (ii) the
locations; list of floor locations adequately numbered, (iii) the distance from
each floor location to the reception point, considering that each
So as to minimize the total distance travelled in the warehouse, floor space is divided in carton and plastic boxes and that this divi-
taking into account the stores shipping priorities. sion may require different distances to the reception point, (iv) the
4 M. Horta et al. / Computers & Industrial Engineering 101 (2016) 1–9
a1
a2
a3
a4
a5
a6
distance from each floor location to the shipping point, (v) the s – set of stores to allocate to the warehouse space floor loca-
aggregation of floor locations to clusters adequately numbered, tions, s 2 S;
that is obtained by aggregating neighbour floor locations on groups cl – set of clusters of warehouse floor locations, cl 2 CL.
of floor locations with a total number of y floor locations, where y
is the average number of number of stores from which the demand Subsets
can be fulfilled from one pallet, (vi) the distance to reception and
shipping points from each cluster, considering that the distance HS – subset of stores that have a high shipping priority, HS # S;
is the total distance travelled to and from the cluster (round trip), LS – subset of stores that have a lower shipping priority, LS # S;
(vii) demand per store in pallets during the time period covered by Note: HS [ LS ¼ S.
the data (e.g. monthly) that will determine the demand per cluster
in pallets, (viii) number of SKUs demanded per store during the Parameters
time period covered by the data (e.g. monthly) that will determine
the number of visits to that cluster (equal to the number of visits of capf – floor capacity (in number of pallets) of the warehouse
the store that requires a higher number of different SKUs), (ix) floor location f for given time period;
capacity in pallets of each floor location, (x) floor space require- ycl;f – indicates the allocation of warehouse floor location f to
ments in terms of pallets daily for each store and (xi) if store is con- each cluster cl;
sidered to have priority in shipping. distSf – distance from the warehouse floor location f to the ship-
In this way, as model data, it is required to feed the number of ping point;
floor locations per cluster as well as the actual allocation of floor distPcbcl – distance related to the picking operation in the clus-
locations to clusters, which is numbered beforehand. Since we only ter of areas cl (for picking of carton boxes);
consider linear clusters, we might have clusters of different sizes distPpbcl – distance related to the picking operation in the clus-
due to the reaching of aisle end. For instance, in Fig. 2, each half ter of areas cl (for picking of plastic boxes);
aisle has 10 floor locations, the cross-aisle and then more 10 floor PSs – number of pallets shipped to store s within a given time
locations. If linear clusters of 6 floor locations are used, the com- period;
plete aisle has 20 floor locations and, thus the final 2 locations will Ds – average requirements (in number of pallets) of the store s
be contained solely in a cluster. Cluster construction is done for a given time period;
beforehand and based in the linear clusters and fed as model data. Rs – average space requirement (in number of warehouse floor
location) of the store s for a given time period. Eq. (1) illustrates
2.4. Model formulation the computation of this parameter.
pexps;f – indicates the possible warehouse floor locations f tiguous space floor locations. By assigning one store to contiguous
where a store s can be allocated according to its shipping prior- warehouse areas it will avoid errors in the shipping operation
ity. This parameter equals 1 if the store s can be allocated to the (such as the loading of a wrong pallet in one truck), since all the
floor location f (otherwise equals 0); pallets of one store will be located side-by-side. Moreover, the pal-
NSs – number of pickings of the store s for a given time period. let construction is also facilitated, as it is possible to transfer boxes
from one pallet to another pallet of the same store in order to prop-
Variables erly construct the pallets and to optimize the truckload capacity.
Also, it is assumed that the set of space floor locations is organized
xs;f – binary variable that equals 1 if the store s is allocated to according to contiguity warehouse criteria.
the warehouse floor location f (otherwise equals 0); Finally, the constraints (9) and (10) define the variables domain.
NV cl – non-negative integer variable that indicates the number
of visits to the cluster cl; 3. Case study
Distance (km/month)
house occupancy rate, defined as the ratio between the number
of occupied space floor locations and the total number of locations 6000
available in the warehouse. In Section 4.3 a sensitivity analysis is
performed in order to evaluate the impact of changing the param- 4000
eters of the model that may be subject to higher uncertainty. Sec-
tion 4.4 presents the computational results. 2000
4.1. Scenarios 0
Total Picking Shipping
Three different scenarios are studied in this paper: a current distance distance distance
scenario, a best-case scenario and a future scenario (in 3 years). Fig. 4. Best-case scenario: Distance travelled indicators.
It was decided to analyse a future scenario since the company plan
to maintain the fruits and vegetables products in the same ware-
house for a minimum time period of 3 years. 10000
Distance (km/month)
4.1.1. Current scenario 8000
The first scenario aims to evaluate the layout that is currently
6000
implemented in the warehouse. In this scenario the stores loca-
tions were fixed according to their actual allocation in the 4000
warehouse.
The total distance travelled in the warehouse was equal to 2000
9428 km per month (Fig. 3). In this scenario, the distance travelled
0
during the picking operation accounts to 79.6% of the total distance Current Best-case
travelled in the warehouse. The warehouse occupancy rate was scenario scenario
equal to 83.7%.
Fig. 5. Comparison: Current vs. Best-case scenario.
4.1.2. Best-case scenario
This scenario corresponds to the application of the model to the
Table 1
case study with the assumptions made for each parameter and no Savings obtained from the best-case scenario in rela-
additional restrictions. tion to the current scenario.
The application of the model in this scenario led to a total dis-
Savings
tance of 7266 km per month (see Fig. 4). In this case, the picking
Distance reduction (km/month) 2162
operation represents 77.9% of the total distance travelled in the
Distance reduction (km/day) 83
warehouse. The warehouse occupancy rate remained equal to Reduction in daily hours of operation 20.8
83.7%. Potential reduction of employees 2
By comparing the current scenario with the best-case scenario
it can be concluded that the layout proposed in the best-case sce-
nario can lead to a reduction of the total distance travelled in the the company can also be more efficient by maintaining the same
warehouse by 23% (see Fig. 5). number of employees for an increase in the warehouse activity
In Table 1 the potential savings of the best-case scenario com- level.
pared to the current scenario are presented.
In Table 1 it is possible to see that the allocation of the stores in 4.1.3. Future scenario
this best-case scenario enables a reduction of the distance travelled This scenario corresponds to a future scenario (in 3 years)
monthly in the warehouse by 2162 km. Considering that the ware- where is considered a growth in the number of picked boxes. Based
house operates 26 day per month, a reduction of 83 km per day can on historical data, it was agreed with the company that annual
be achieved. In turn, knowing that the forklifts used in the ware- growth rate studied in this scenario was equal to 6%.
house moves at a maximum speed of 4 km per hour, it can be con- This growth rate was considered to be equal to all stores and
cluded that is possible to reduce 20.8 h of operation per day. This results from three factors. The first factor that contributes to this
reduction leads to the conclusion that it is possible maintain the growth is the growth trend in the market share of company’s retail
same warehouse activity level with less 2 employees (assuming stores. The second factor is related to the reduction of the average
that each employee works in average 7.5 h per day). However, price of the goods that are sold in the Portuguese supermarkets.
This price reduction has the objective to stimulate the consump-
10000 tion and is also a result from the growing competition in the mar-
ket. The last factor is related with the repackaging strategy adopted
Distance (km/month)
nario. This change is justified by the need of one additional ware- 10000
house location by 8 stores. In turn, the need for one additional
Distance (km/month)
space by 8 stores results from the growth in the number of pallets 8000 Current scenario
shipped by the warehouse. As mentioned previously, the model layout
ensures that if a store needs more than one space floor location 6000
Best-case scenario
that store will be assigned to contiguous space floor locations. This layout
factor contributed to change the stores disposition in the ware- 4000
house, since the locations to be assigned to these 8 stores were Future scenario
2000 layout
already occupied in the best-case scenario.
In this scenario the warehouse occupancy rate was equal to
86.9%, which represents an increase compared with the value 0
obtained for this indicator in the previous two scenarios (83.7%). Fig. 7. Comparison of the monthly distance travelled in the warehouse for the three
By comparing the best-case scenario with this scenario, it can scenarios.
be concluded that the monthly distance travelled in the warehouse
increased 10.9%.
Table 2
Savings of the layouts obtained in the best-case scenario and in the future scenario
4.2. Scenarios comparison when compared to the current scenario.
In Fig. 7 it is possible to see the comparison of the monthly dis- Savings Best-case scenario Future scenario layout
layout vs. current vs. current scenario
tance travelled in the warehouse for the layouts obtained in each scenario
scenario. In order to make a fair comparison the layout obtained
Distance reduction 2162 1554
in the future scenario was run with current data. From the analysis
(km/month)
of the figure, it can be conclude that the layout obtained in the Distance reduction 83 60
best-case scenario allows achieving a greater gain of immediate (km/day)
efficiency (reduction of the total distance in 23%). However, this Reduction in daily hours 20.8 14.9
of operation (h/day)
layout is not projected for the future, as it is not considered an
Potential reduction of 2 2
increase in the number of pickings. On the other hand, currently employees
the layout obtained in the future scenario does not lead to such a
significant gain in efficiency (reduction of the total distance in
16.5%) but it is designed considering a future growth.
The parameter analysed was the number of pickings per pallet
It should be noted that the implementation of the layout
given the high standard deviation obtained for this parameter from
obtained in the best-case scenario might imply future costs, arising
the data gathered and the impact that this parameter has on the
from possible changes in the location of the stores in the ware-
model solution. As mentioned previously, this parameter determi-
house in order to adapt the layout to an expected increase in the
nes the size of the clusters and, therefore, the locations that consti-
warehouse activity level. These changes involve modifications in
tute each cluster and distance travelled during the picking
the warehouse management system used by the company and a
operation. In this section the impact on the model resulting from
new adaptation of the employees to a different work environment.
a positive and a negative variation in the number of pickings per
In Table 2 are represented the potential savings of the layouts
pallet, in an amount equal to the standard deviation obtained for
obtained in the best-case scenario and in the future scenario com-
this parameter, is analysed.
pared to the current scenario. As it was mentioned above, the
The results of this analysis are shown in Fig. 8. It can be con-
reduction of the monthly distance travelled in the warehouse is
cluded that a positive and a negative variation in the number of
greater for the layout of best-case scenario. For each one of the lay-
pickings per pallet by approximately 31% results in a reduction
outs a reduction in daily hours of operation was obtained (follow-
of the distance travelled in the warehouse by 22% and in an
ing the same reasoning that was already presented in this paper).
This hourly reduction shows that implementation of both layouts
allows that the current warehouse activity can be maintained with
less 2 employees. 10000
Distance (km/month)
8000
4.3. Sensitivity analysis
6000
In order to assess the robustness of the results a sensitivity 4000 Shpping distance
analysis was performed. To perform this analysis the best-case sce- 2000 Picking distance
nario was used as a basis of comparison.
0
9 13 17
Number of pickings per pallet
10000
Distance (km/month)
6000
4000 Table 3
Sensitivity analysis to the number of pickings per pallet: current and best-case
2000 scenarios.
0 Number of pickings per pallet 9 17
Total Picking Shipping
Total distance – current layout (km/month) 11,602 8054
distance distance distance
Total distance – layout proposed in the best-case scenario 9499 6491
(km/month)
Fig. 6. Future scenario: Distance travelled indicators.
8 M. Horta et al. / Computers & Industrial Engineering 101 (2016) 1–9
Table 4
Computational results of the IP model.
Scenario Total variables Binary variables Constraints Iterations Computational time Relative gap (%)
Current 47,553 0 47,970 0 3.6 s 0.0
Best-case 47,553 38,186 50,369 902,888 2.5 h 76.6
Future 47,553 38,186 52,329 5,418,395 24 h 73.8
increase by 20.6%, respectively. An increase in the average number operation scenario is changed. It was concluded that with the lay-
of pickings per pallet, with the same pallet of a given product, more out obtained in the best-case scenario (Section 4.1.2) the monthly
stores can be visited in the same picking route and that an distance travelled in the warehouse can be reduced by 23%. This
employee needs to return fewer times to the receiving docks to col- reduction allows that the same warehouse activity level can be
lect new pallets. Therefore, it is easily concluded that an increase in maintained with less two employees. A future scenario (in 3 years)
the number of pickings per pallet results in a decrease in the dis- was also studied in this paper. In this scenario (Section 4.1.3) it was
tance travelled in the warehouse (and vice-versa). Nevertheless, assumed an annual growth rate of the number of boxes picked in
the distance variation is less sensitive than the variation in number the warehouse in 6%. This growth rate led to an increase in the
of pickings per pallet imposed to the model. warehouse occupancy rate and to a change in the disposition of
The variation in the number of pickings per pallet caused a the stores in the warehouse, which resulted from an additional
change in the allocation of the stores in the warehouse that was need of one space floor location by some stores. Despite the
obtained in the best-case scenario. Although there has been a increase in the distance travelled in the warehouse in relation to
change in the disposition of the stores, it turns out that the layout the best-case scenario, the distance obtained in this future scenario
that was proposed in the best-case scenario continues to lead to a was still 14.6% less than the currently distance travelled in the
reduction in the distance travelled in the warehouse with a varia- warehouse.
tion of the number of pickings per pallet (see Table 3). For any case In the sensitivity analysis the parameter that has more impact
analysed (number of pickings per pallet equal to 9 and 17) the in the model and that is more uncertain was analysed: the number
solution obtained in the best-case scenario leads to an efficiency of pickings per pallet. The variation in the number of pickings per
gain in terms of distance travelled in the warehouse. This efficiency pallet caused a change in the distance travelled in the warehouse
gain is equal to 18.1% if the number of pickings per pallet is equal and in the disposition of the stores in relation to the best-case sce-
to 9 and equals 19.4% if the parameter analysed equals 17. nario. However, it is important to state that the disposition of
stores in the warehouse obtained in the best-case scenario contin-
ues to lead to reductions in the distance travelled in the warehouse
4.4. Computational results
(between 18% and 19%) with a variation of the number of pickings
per pallet.
Table 4 presents the computational results for the implementa-
The main limitations of the proposed method are related with
tion of the model to the different scenarios. The base scenario has
the assumptions made, namely related with the number of receiv-
no binary variables since they are prefixed within the model – it
ing and shipping points. Another limitation is related with the
just calculates the distance travelled in the current solution. The
solution approach, namely by using the MinMax modelling and
computational times are quite high for the other two scenarios
implemented by an exact optimization method solved by a com-
and still presenting high relative gaps. This is mainly related to
mercial solver (CPLEX), which gives rise to long computational
the usage of the MinMax formulation, which usually leads to good
time. Distances have been considered in a straightforward manner
solutions but being hard to prove optimality (Fang & Wu, 1996).
(i.e., the shortest distance to arrive to one location, instead of using
In order to assess the quality of the solutions obtained, a simple
common rules to operate picking operations). The aggregation of
test was done. Instead of having as a variable the number of visits
products implies that demand is only managed at the pallet level,
to one cluster (NV cl ), it was assigned an average number of visits to
eliminating particularities of demand per product. These limita-
each store in each cluster. The optimal solution was obtained after
tions give rise to the need to expand the capabilities of the model.
6 s but the total distance travelled per month was 10.5% higher
Some future studies can therefore be performed regarding the
than the distance travelled in the Best-Case scenario. Thus, the
problem addressed in this paper. First it would be important to
solutions obtained with the MinMax formulation have good qual-
analyse weather a layout change may cause delays in the shipping
ity, despite the fact that CPLEX is not able to prove it.
operation and a consequent reduction of the service level provided
to the stores. It would be also interesting to perform a simulation,
5. Conclusions before the implementation of the new layout, of the aisles conges-
tion per time period. In terms of the modelling, the MinMax formu-
The problem studied in this paper is related to the layout of a lation may be revised and improved. Additionally, there is
fresh products’ warehouse that operates in a JIT environment. potential to incorporate in the model a routing problem in order
The layout of the warehouse studied in this paper consists in an to consider the different routes of the employees during the pick-
open area that is divided in several space floor locations that are ing operation. In this paper the warehouse locations were grouped
assigned to the clients (stores in this case), where the products into clusters according to the average number of pickings per pal-
are placed according to the daily orders. Based on the conducted let. However, since the distance between contiguous warehouse
literature review, a mathematical integer programming model is areas is small and since most of products are ordered by most of
proposed to solve the problem. The model aims to determine the the stores, the considered assumption may have a small impact
allocation of the stores to the available space floor locations in in the monthly distance travelled in the warehouse. Finally, as an
order to minimize the distance travelled in the warehouse (Vis & additional note, it is important to perform an effective allocation
Roodbergen, 2008) and to take into account the orders shipping of the trucks to the warehouse docks, taking into account the lay-
priorities (Sandal, 2005). out that will be implemented. Thus, it is expected that the alloca-
The results show that, according to the information used, the tion of the trucks to the docks can be performed in order to
distance travelled in the warehouse can be reduced if the current minimize the distance travelled in the warehouse to transport
M. Horta et al. / Computers & Industrial Engineering 101 (2016) 1–9 9
the pallets from its warehouse locations to the truck that will Ladier, A.-L., & Alpan, G. (2016). Cross-docking operations: Current research versus
industry practice. Omega, 62, 145–162.
transport the goods to the stores. The overall future studies require
Larson, T. N., March, H., & Kusiak, A. (1997). A heuristic approach to warehouse
further the inclusion of uncertainty in data, so as to become more layout with class based storage. IIE Transactions, 29, 337–348.
realistic and close to daily operations. Miao, Z., Lim, A., & Ma, H. (2009). Truck dock assignment problem with operational
time constraint within crossdocks. European Journal of Operational Research, 192
(1), 105–115.
References Richards, G. (2011). Warehouse management: A complete guide to improving efficiency
and minimizing costs in the modern warehouse. Kogan Page.
Bartholdi, J. J., & Gue, K. R. (2000). Reducing labor costs in an LTL crossdocking Sandal, S. (2005). Staging approaches to reduce overall cost in a crossdock environment
terminal. Operations Research, 48(6), 823–832. Master’s thesis. Columbia: University of Missouri.
Belle, J. V., Valckenaers, P., & Cattrysse, D. (2012). Cross-docking: State of the art. Shimizu, K., Ishizuka, Y., & Bard, J. F. (1997). Nondifferentiable and two-level
Omega, 40, 827–846. mathematical programming. Springer.
de Koster, R., Le-Duc, T., & Roodbergen, K. J. (2007). Design and control of warehouse Tsui, L. Y., & Chang, C.-H. (1992). An optimal solution to a dock door assignment
order picking: A literature review. European Journal of Operational Research, 182 problem. Computers & Industrial Engineering, 23(1–4), 283–286.
(2), 481–501. van den Berg, J. P., & Zijm, W. H. M. (1999). Models for warehouse management:
Fang, S. C., & Wu, S.-Y. (1996). Solving min-max problems and linear semi-infinite Classification and examples. International Journal of Production Economics, 59,
programs. Computers and Mathematics with Applications, 32(6), 87–93. 519–528.
Fernie, J., & Sparks, L. (2004). Logistics and retail management: Insights into current Vis, I. F. A., & Roodbergen, K. J. (2008). Positioning of goods in a cross-docking
practice and trends from leading experts. London UK: Kogan Page. environment. Computers & Industrial Engineering, 54(3), 677–689.
Fernie, J., Sparks, L., & McKinnon, A. C. (2010). Retail logistics in the UK: Past, present Vis, I. F. A., & Roodbergen, K. J. (2011). Layout and control policies for cross docking
and future. International Journal of Retail & Distribution Management, 38(11/12), operations. Computers & Industrial Engineering, 61(4), 911–919.
894–914. Won, J., & Olafsson, S. (2005). Joint order batching and order picking in warehouse
Frazelle, E. (2002). World-class warehousing and material handling. New York, NY: operations. International Journal of Production Research, 43(7), 1427–1442.
McGraw-Hill.
Gu, J., Goetschalckx, M., & McGinnis, L. F. (2007). Research on warehouse operation:
A comprehensive review. European Journal of Operational Research, 177(1), 1–21.