�ݺ�ߣ

Simulation of an Order Picking System in a Pharmaceutical
Warehouse
Jo?o Pedro Jorge1, Zafeiris Kokkinogenis3,4, Rosaldo J. F. Rossetti2,3, Manuel A. P. Marques1

1
Department of Industrial Engineering and Management, Faculty of Engineering, University of Porto, Portugal
2
Department of Informatics Engineering, Faculty of Engineering, University of Porto, Portugal
3
Artificial Intelligence and Computer Science Laboratory (LIACC), Faculty of Engineering, University of Porto, Portugal
4
Institute of Mechanical Engineering (IDMEC), Faculty of Engineering, University of Porto, Portugal
{deg11008, pro08017, rossetti, pmarques}@fe.up.pt

Abstract - The paper presents an agent-based simulation model which imposes higher requirements on order processing time
of an order picking system. A pharmaceutical warehouse is at warehouses [2].
used as case study with the purpose of improving the Order picking operations often consume a large part of
implemented picking processes. Warehousing activities affect the total labour activities in the warehouse ([3], even claims
the total logistic costs of a company or supply chain. The up to 60%), and for a typical warehouse, order picking may
optimization of the required picking operations is one of the account for 55% of all operating costs [4]. Most of the
most important objectives when attempting to reduce the warehouses employ humans for order picking.
operative costs. This study intends to provide a proof-of- According to the movement of human and products,
concept agent-based model for scheduling the number of
order picking is organized into picker-to-parts and parts-to-
human resources required for picking activities. The
picker systems. In a picker-to-parts system the picker (the
improvement in the service and the planning of the manpower
used in the warehouse, thus achieved, leads to operation-cost person that performs the order picking operation) walks
reductions. The goal is accomplished by using the NetLogo along the aisles to pick items. In this system is used the pick
agent-based simulation framework. The simulation outcomes by order. During a pick cycle, pick information is
suggest that dimensioning human resources is a means to communicated by a handheld terminal or a voice picking
satisfy the desirable level of customer��s service. system. No paper pick lists are needed. The parts-to-picker
systems are usually implemented by the usage of
Keywords - Warehouse Simulation; Order Picking system; ��Automated Storage and Retrieval Systems�� (AS/RS).
Agent-Based model; NetLogo. In the present case study we use a pharmaceutical
warehouse that has four different storage areas depending on
I. INTRODUCTION the type of product stored: products with low rotation rate,
Recent trends in the warehouse planning have resulted in products with high rotation rate, big and fragile products and
order picking design and management being more important special products (inflammable). The maximum number of
and complex. Order picking operation is one of the logistic pickers is 15.
warehouse��s processes. It comports the retrieval and To simulate the order picking operations, an agent-based
collection of articles from a storage location in a specified model of the warehouse is used. The agent-based model
quantity into a box to satisfy a customer��s order. represents the real order picking entities and simulates the
Customers tend to order more frequently, in smaller customer service indicators. For this work, we used the
quantities, and they require customized service. On the other NetLogo modelling framework [5] to rapidly prototype
hand, companies tend to accept late-arrival orders while they simple, yet realistic, ��what-if�� order picking scenarios and
need to provide rapid and timely delivery within tight time analyse the system performance under different real set-ups.
windows [1]. In general, lead times are under pressure. This NetLogo is a free open-source programmable modelling and
is particularly true for pharmaceutical distribution centres. simulation platform, appropriate for modelling complex
In this business, pharmacies can order at the click of a systems. One of its main advantages is the ease of
button and expect inexpensive, rapid and accurate delivery. programming. The language is very intuitive and specifically
Obviously, managing order picking operations effectively designed for agent-based modelling, thus the user needs only
and efficiently is a challenging process for the warehouse to program agent behaviour, not the agents themselves.
function. A key objective is to shorten throughput times for Moreover, the researcher community extensively supports
order picking, and to guarantee the meeting of due times for the modelling platform and regularly develop a number of
shipment departures. In order to offer high customer service tools useful to the modeller.
level and to achieve economies of scale in transportation to This paper has the following research objectives: (i) to
support the related costs, these small size, late-arrival orders assign the correct number of pickers for a certain average of
need to meet the tight shipment time fence. Hence the time served orders; (ii) to provide a tool based on a simulation
available for picking orders at warehouses becomes shorter, model to analyse the performance of the order picking

process; (iii) to calculate the demand of each type of product picker-to-product system. Pickers work to fulfil orders. The
based on real data; (iv) to calculate the orders rate that enter number of order lines in an order is referred to as order size.
in the warehouse; (v) to create a framework with the capacity Order sizes may vary significantly.
of generating orders randomly. The pharmaceutical distribution center has four different
This paper is organized as follows. In Section II the storage areas depending on the type of product stored:
existing literature is reviewed, the real system is presented in products with low rotation rate, products with high rotation
Section III. A modelling framework, the NetLogo rate, big and fragile products and special products
implementation and the validation experiments are shown in (inflammable) and a maximum number of 15 pickers.
Sections IV, V and VI. Conclusion and future work follow in The storage areas are arranged in a pre-defined layout
Section VII. and there is a common conveyor to transports the order
boxes between them. A customer order may require products
II. RELATED WORK
from one or more storage areas and the time to collect the
The two major types of order picking systems can be products is different for each case.
distinguished into parts-to-picker and picker-to-parts At the order picking workstation, orders arrive
systems. De Koster et al. [1] have provided an extensive sequentially using boxes on the conveyors. Once the picker
literature review of the order picking operations and their and the required order product box are available, the picker
implementations. One of their conclusions was the lack of
moves to the product rack, then picks a number of required
attention from the researcher community for the pickers-to-
items and place it on the conveyor to be transported to the
parts order picking systems despite them being the dominant
implemented approach. position where the order product box waits. The Figure 1
Picker-to-parts systems occur in two types: pick by order shows a scheme of this process. The picker works on one
and pick by article (batch picking). It is also possible to order at a time until all lines of the order have been picked
distinguish picker-to-parts systems by the order arrival and and the order is said to be finished. When an order is
release. This can be either deterministic or planned [6] or finished, the system moves the finished order box to the
stochastic [7]. dispatch area.
A polling model can describe the order arrivals and
processing; a system of multiple queues accessed in a
speci?ed sequence by a multiple servers [8]. Hwang et al. [9]
use clustering-based heuristic algorithms for the batching of
orders for order picking in a single-aisle automated storage
and retrieval system. Daniels et al. [10] consider the
warehouse in which goods are stored at multiple locations
and the pick location of a product can be selected
dynamically.
There are also relevant applications in operation
management. For instance, Koenigsberg and Mamer [11]
consider an operator who serves a number of storage
locations on a rotating carousel conveyor. Bozer and Park
[12] presented a single-device polling-based material
handling systems. Although these systems have been widely
researched, they have not yet received systematic treatment
and application in the picking process organized in a picker-
to-parts system. The same situation occurs with the agent-
based models. The literature mainly presents agent-based
approaches to solve order picking problems where goods are
stored at multiple locations or warehouse control solutions Figure 1. A scheme of order picking process
[13], [14]. The study described in this paper is applied to a
real case, which the picking is organized in a picker-to- In this type of system, the picker can only pick items
product system. from the product racks that belong to the order currently
This work contributes to the literature by exploring the being processed.
agent-based metaphor to simulate an order picking system in If for any reason the product is not available, the picking
a realistic scenario using real data of a pharmaceutical time is above a limit or some error occurs the order is
warehouse. deviated and need a special procedure is used to finish it.

III. REAL SYSTEM DESCRIPTION IV. SIMULATION MODEL DESCRIPTION
Figure 1 shows the layout of the order picking In the following section we discuss the concepts of a
workstation under study. This system can be classified as a pharmaceutical order picking system focusing on the picker-

to-parts system. We describe the processes taking place Table I. Demand production and time preparation
during the operation of the warehouse as well as the data
Frequency per Average Preparation
used to model the system. Type of order
Type of order (%) Time (min)
The workstation is modelled using an agent-based
approach. Several pickers, products stored in racks, boxes
PROD1ZONE 30.46% 5'
with the products to complete the orders and a server-based
management information system, compose such a system. PROD2ZONE 27.89% 15'
The user can adjust this model changing the number of PROD3ZONE 22.91% 25'
pickers. The orders arrive randomly at the workstation with a PROD4ZONE 18.74% 43'
rate of 7, 8 orders per minute and they have random lines of
products based on the real data. This value is obtained by the This means that in 10 orders the amount of each type of
analysis of the data of one entire month (65 000 records). final product is:
The result is depicted in Figure 2 in a Poisson distribution. - PROD1ZONE: 3.1 orders
Orders are created based in this real rate. - PROD2ZONE: 2.7 orders
- PROD3ZONE: 2.3 orders
0.16
? - PROD4ZONE: 1.9 orders
For this model it was assumed that products are always
0.14
? available in the warehouse, the pickers are equally skilled
(homogeneous agents) for the order picking operations, the
0.12
? workspace is considered an open space and the time to pick a
product from the shelf is standard and do not vary with the
0.1
? product.
Probability
?

The orders can have many different statuses: ��arriving��,
0.08
? ��queuing��, ��placing��, ��preparation��, ��finishing�� and
��leaving�� as it can be seen in Figure 3. The process time
0.06
? used in this paper represents an aggregation of all
components that contribute to the processing time at the
0.04
? order picking workstation, the Effective Process Time (EPT).
Initially orders move towards the warehouse entrance and
0.02
? stand in a sequential order until a place is chosen. If a free
place exists, then the order must navigate to that place. Once
0
? the order has reached the place the status is changed to
��waiting�� and a counter is started. After the pre-defined
0
? 1
? 2
? 3
? 4
? 5
? 6
? 7
? 8
? 9
? 10
? 12
? 14
? 16
?
11
? 13
? 15
?
finishing time has passed, the order changes its status to
Number
?of
?orders
? ��leaving��.

Figure 2. Poisson distribution of the real data

In this model, customer orders can require one or more
product lines from one or more storage areas. To describe
this situation it was assumed four types of final product:
- if an order requires one or more products lines
from one area, we call it a PROD1ZONE;
from two areas, we call it a PROD2ZONE;
Figure 3. Orders status
from three areas, we call it a PROD3ZONE;
- if an order requires one or more products lines Jacobs et al. [15] presented an algorithm to compute the
from four areas, we call it a PROD4ZONE. EPT realizations directly from arrival and departure events.
The frequency and the average preparation time per each An order picking workstation is characterized by several
type of order are calculated using the real data from entire process time components (see Figure 3). At the core of the
days. And the result is depicted in Table I. process is the time required for picking items (preparing and
finish time), that is the raw pick time. Next to the raw pick
time, pickers may require some setup between processing of
orders. Conveyor systems may break down, causing
unavoidable delays. Picker availability is also an issue since
it is likely that a picker is sometimes not present at the

workstation. These components are aggregated into a single 2. The picker:
EPT. The idea is then to reconstruct the EPT directly from The user before setup can define the number of pickers
order arrival and departure times registered at the operating and their initial location is randomly generated in the
order picking workstation under consideration with the workspace. Simple reactive agents based on simple ��if-
obvious advantage that it?s not need to quantify each Then�� rules implement the pickers. The picker collects all
component that contribute to the process time. the products to finish the order. There are four types of final
Figure 3 shows an example of arrivals and departures of products: PROD1ZONE, PROD2ZONE, PROD3ZONE and
four orders at an order picking workstation. An arrival Ai PROD4ZONE. Each final product has different picking time
occurs at the moment an order i is prepared at the order
to be prepared (defined in the source code). Once the
picking workstation. A departure Di occur when the picking
products have been collected, the picker moves to the
has been finished and the respective order i is finished.
EPT realizations are calculated using the following sending area and places it on the conveyor. The picker
equation: restarts the cycle. In the proposed model, pickers are
represented by agents.
EPTi = Di ? Ai (1)
3. The server:
where Di denotes the time of ith departing order and Ai The server is an agent responsible for the managing and
denotes the arrival of the corresponding ith order. The bottom dispatching of the orders.
part of Figure 3 illustrates how EPT realizations are obtained If all positions for preparing the orders are full, the server
using Equation (1). does not allow orders to enter into the system to do the order
The picker agent has various statuses: ��selecting the picking operation; this causes a sequential order (��queuing��
product��, ��getting products from the rack��, ��going to the of orders).
workspace��, ��preparing the product�� and ��sending the The only interface variable that the user must set before
product��. the model runs is the number of pickers. All the other
variables can be changed, allowing a dynamic observation.
V. NETLOGO IMPLEMENTATION
The user can change also the following variables:
In this paper, we describe the use of NetLogo as a rapid - The demand for each type of final product;
prototyping tool for an agent-based simulation framework to - The chance that orders are generated;
evaluate the setup of a pharmaceutical warehouse order - The speed of the conveyor.
picking system.
We present how our problem was modelled and which Various monitors and plots allow the user to display the
abstractions were used to achieve the outlined objectives. result of these dynamics:
Furthermore, complexities and constraints inherent to this - The total number of orders;
problem were identified. From that, a simplified model of - The number of free places;
an abstraction of the application domain was created - The number of instances of each type of
without losing key aspects. Our purpose is to simulate the product;
activities and operation taking place in an order picking - A chart plotting the number of orders served
system in way (i) to assign the correct number of pickers for every 100 ticks (NetLogo unit of time);
a certain average of served orders; (ii) to simulate the orders - A chart plotting the number of diverted orders
behaviour: served and diverted; (iii) to calculate the orders every 100 ticks;
rate that enter in the warehouse. - An average number of orders served;
- An average number of diverted orders;
There are several concepts and agents involved in this
model: The user can experiment to change the values and seeing
the result through the monitors and charts. It is possible to
1. The Orders: observe the visual phenomena that are developed such as
Orders are randomly generated to ��arrive�� in the warehouse bottlenecks, queues and the spatial distribution of diverted
following a probability distribution according to the orders.
historical data distribution. A preparation place is assigned There are important aspects in this model:
(��placing��) to each order. Here, the order assumes the state - Queuing at the warehouse entrance;
��waiting�� until the order picking operations are finished - Bottlenecks at the exit;
(��finishing��). Upon conclusion, the order assumes the state - Congestion on the conveyor;
��leaving�� and is forwarded to the dispatching area. The - The stochastic aspects are inherent in the model;
demand for each type of product is based on data from a - Agents don't always move in exactly the same
pharmaceutical company. way;
If the waiting time is too long, (for any reason: product not - Demand of different final products may vary
found, place not available) the order is diverted. naturally.

Although NetLogo is a simple simulation framework, it
proves to be a very useful tool for creating this type of
agent-based model of real scenario. With respect to other
simulation paradigms, the agent-based approach offers the
users the possibility to observe not only the dynamic of the
system but also the interaction of the situated entities in the
system. One key point is also the agent movement.
Specifically creating a realistic system where agents (e.g.,
pickers) can move having specific goals and destinations.
The limitation in allowing agents to move dynamically in
the NetLogo environment is that they are constrained
moving discretely on a grid-based space patch-by-patch
rather having smooth trajectories.

Figure 5. Order behaviour in simulated system
VI. VALIDATION EXPERIMENT
In this section the simulation experiments are discussed Analysing the order plot over a complete working day��s
to validate the proposed model. First, a simulation scenario operations, we get a good idea of the order average of
has been created to use as a test case representing the ��real served orders and diverted orders. By looking at the
life�� operating order picking workstation. Then, the model historical company��s data, we can see that the diverted
has been simulated at a real utilization level (using different orders are in line with the current observations.
number of picker-agents, the real demand and the real order
picking times) to generate order arrival and departure
events. Subsequently, these events are used as variables for VII. CONCLUSION AND FUTURE WORK
the global model. Next, the model was simulated at various This paper presented an order picking model
utilization levels (varying the number of pickers and implemented in the NetLogo agent-based platform.
conveyor belt speed) to measure the number of orders Although the model may appear to be simplistic, its
served versus the number of diverted orders. conceptualisation encompasses many aspects of the
observed system in the real world. The model manages to
predict the average number of served orders using a certain
amount of human resources by means of an agent-based
simulation model with real data from a pharmaceutical
distribution center. Arrival and departure time data are the
only input required to calculate the time to complete the
orders. The validation of the simulation study demonstrated
that the data used are adequate to the required results. It was
found that the proposed model accurately predicts the
defined goal in a satisfactory degree.
In practice, the actual pick rate does not deviate from the
expected rate. With this regard, the EPT represents the
Figure 4. Orders behaviour in real system
actual pick time of an order picking workstation. This will
allow the identification of possible improvements for order
For the aforementioned experimental set-up, a picking activities.
comparison has been made over the average number of The proposed model has practical use because collecting
served and diverted orders of this simulation model with arrival and departure data of orders is relatively simple in
those of the real system. In this way, it?s possible to assess warehouses and the output is easily perceived.
the accuracy of the predictions of the average number of In future work, the model��s layout will be improved and
served orders. The real system��s data are depicted in Figure demonstrate how the presented approach in this paper can
4 (in green the average number of served orders) while the be applied to a more detailed order picking workstation (i.e.,
simulation obtained results are depicted in Figure 5. The heterogeneous agents to represent differently skilled pickers,
word ��tick�� in the graph of Figure 5 is a measure for the automation mechanism in the warehouse, etc.).
time that is used by NetLogo. Besides that, the current advancements in order picking
technology have allowed multiple orders to be processed
simultaneously by a picker, which is often the case in large-
scale warehouses. Thus, orders routings in large-scale

warehouses would not be processed in a FIFO sequence at 6. N. Gademann and S. Van de Velde, ��Order batching to
the workstation. Performance analysis, in such a context, of minimize total travel time in a parallel-aisle warehouse,��
IIE Transactions vol. 37, issue 1, pp. 63-75, 2005.
order picking workstation will be also subject to future 7. T. Le-Duc, ��Design and control of efficient order picking
work. Therefore, it is important to improve and enhance the process,�� Rotterdam School of Management, Erasmus
attributes of the agent-based model to tackle these issues. University Rotterdam, the Netherlands, Ph.D.
dissertation, 2005.
8. M. M. Srinivasan, ��Nondeterministic polling systems,��
ACKNOWLEDGMENT Management Science, vol. 37 no. 6, pp. 667�C681, June
1991.
This project has been supported by the FEUP project 9. H. Hwang, W. Baek, and M. Lee, ��Clustering algorithms
��Support system for availability and location of products for for order picking in an automated storage and retrieval
picking�� and FCT (Funda??o para a Ci��ncia e a Tecnologia), system,�� International Journal of Production Research,
the Portuguese Agency for R&D, under the grant, vol. 26, issue 2, pp. 189�C201, 1988.
SFRH/BD/67202/2009 10. R. L. Daniels, J. Rummel, and R. Schantz, ��A model for
warehouse order picking,�� European Journal of
Operational Research, vol. 105, pp. 1�C17, February
1998.
REFERENCES
11. E. Koenigsberg and J. Mamer, "The analysis of
1. R. de Koster, T. Le-Duc, and K. J. Roodbergen, ��Design production systems,'' International Journal of Production
and control of warehouse order picking: A literature Research, vol. 20, issue 1, pp. 1-16, 1982
review,�� European Journal of Operational Research, vol. 12. Y.A. Bozer and J.H. Park, ��Expected move request
182, issue 2, pp.68�C73, October 2007. waiting times in single-device, polling-based material
2. M. Yu, ��Enhancing warehouse performance by efficient handling systems,�� IIE Trans vol. 33, issue 1, pp. 55�C63,
order picking,�� Netherlands Research School for 2001.
Transport, Infrastructure and Logistics, Doctoral Thesis, 13. B. Kim, J. G. Robert, and S. H. Sunderesh, ��Intelligent
October 2008. agent modeling of an industrial warehousing problem,��
3. J. Drury, ��Towards more efficient order picking,�� IMM IIE Transactions, vol. 34, issue 7, pp. 601-612, 2002.
Monograph No. 1. Report, The Institute of Materials 14. B. Kim, S. H. Sunderesh, J. G. Robert, and A. St. Onge,
Management Cranfield, U.K, 1988. ��Intelligent agent based framework for warehouse
4. J. A. Tompkins, J. A. White, Y. A. Bozer, and J. M. A. control,�� Proceedings of the 37th Hawaii International
Tanchoco, ��Facilities planning,�� New York, John Wiley Conference on System Sciences, January 2004.
and Sons, 2003. 15. J. H. Jacobs, L. F. P. Etman, E. J. J. van Campen, and J.
5. U. Wilensky, ��NetLogo,�� Center for Connected Learning E. Rooda, ��Characterization of operational time
and Computer-Based Modeling, Northwestern variability using effective processing times,�� IEEE
University, Evanston, IL, Transactions on Semiconductor Manufacturing, vol. 16,
no. 3, pp. 511�C520, 2003.
http://ccl.northwestern.edu/netlogo/, (accessed
September 2012).

�ݺ�ߣ

Paper 50100 simulation of an order picking system in a pharmaceutical warehouse - camera ready

More Related Content

Paper 50100 simulation of an order picking system in a pharmaceutical warehouse - camera ready