International Journal of Production Research, 2018

Vol. 56, No. 23, 7103–7121, https://doi.org/10.1080/00207543.2017.1421780

Concurrent manual-order-picking warehouse design: a simulation-based design of experiments

approach
Safwan A. Altarazi* and Maysa M. Ammouri

Industrial Engineering Department, German-Jordanian University, Amman, Jordan

(Received 8 February 2016; accepted 20 December 2017)

The design of manual-order-picking warehouses is a combination of interdependent decisions with enormous possible
varieties in design components. The strong interrelationship between these components, in addition to the dynamic and
interconnected stochastic nature of the problem; necessitate the utilisation of a simultaneous simulation-based approach.
This study proposes a concurrent simulation-based design of experiments approach for the design of manual-order-picking
warehouses. The proposed approach can investigate all possible warehousing design combinations with their stochastic
nature and interactions; hence, widening the search for performance improvement. The examined design components
include warehouse throughput, size, layout, operational policies and manpower/carts. Furthermore, the presented approach
captures the probabilistic nature of all the key warehouse functions of receiving, unloading, put away, storage, preparation
and picking and shipping; and evaluates the performance of the studied designs using the cycle time for a stock keeping
unit in the warehouse. Statistical analysis of the simulation results showed several interesting findings; horizontal layout
was preferable over all other types of layouts and small size warehouses perform better than other large sizes. The study
has also recommended using high throughput for traditional layout-small size warehouses.
Keywords: warehousing; manual-order-picking; simulation; design of experiments; performance analysis

1. Introduction
Manual-order-picking warehouse is a very common warehouses class; according to De Koster, Le-Duc, and Roodbergen
(2007), 80% of warehouses in Western Europe are manually operated. In such systems, manpower plays essential role
in conducting the warehouse key functions of receiving, unloading, putting away, storing, order preparation and picking
and shipping. Particularly, the order-picking function, in which operators retrieve stock keeping units (SKUs) from stor-
age locations in single or multiple stop tour; is the most labour intensive, accordingly expensive, function (Petersen
2002; Roodbergen and Vis 2006). In fact, most of the warehouses’ operational cost is contributed to order-picking
(Tompkins et al. 2010).
Warehouse design is an interdependent cluster of decisions at the strategic, tactical and operational levels; aiming to
meet particular performance criteria (Rouwenhorst et al. 2000). Gu, Goetschalckx, and McGinnis (2010) classified these
decisions into five categories; the warehouse overall structure, sizing and dimensioning, layout design, equipment selec-
tion and selection of operational policies. Traditionally, warehouse design is implemented in a sequential manner. For
example, after deciding on throughput and size, suitable layout is selected, racking and material handling systems are
then chosen, and finally operational polices are adopted. As each design component is predetermined before choosing
its successor; this approach limits the interactions between the design components to a small set of components levels.
Consequently, only a limited spectrum of design component levels can be investigated, with limited chance of improv-
ing performance.
Manual-order-picking warehouse design, including the operational policies, has been a research topic over the past
few decades. The problem is mainly approached using analytical, simulation and mixed analytical-simulation methods
(Cormier and Gunn 1996; Gu, Goetschalckx, and McGinnis 2010). Although analytical models are usually design ori-
ented (Gu, Goetschalckx, and McGinnis 2010), and have proved successes in different cases (Roodbergen and Koster
2001; Fukunari and Malmborg 2008; Bukchin, Khmelnitsky, and Yakuel 2012; Chen et al. 2013; Li, Huang, and Dai
2017; Rao and Adil 2017); such methods can only consider specific warehouse design combinations (Roodbergen, Vis,
and Taylor 2015). Simulation on the other hand, can simultaneously consider all possible warehousing design combina-
tions. In their proposed framework for warehouse design, Baker and Canessa (2009) suggested adopting simulation in

*Corresponding author. Email: [email protected]

© 2018 Informa UK Limited, trading as Taylor & Francis Group

7104 S.A. Altarazi and M.M. Ammouri

several steps of the framework, including establishing the unit load, selection of equipment, calculating capacity and
performance evaluation. Much research has demonstrated the applicability of simulation in several warehousing design
elements including layout design (Caron, Marchet, and Perego 2000; Dukic, Vedran, and Opetuk 2010), size (Petersen
2002), manpower and equipment (Kosfeld 1998) and operational policies (Macro and Salmi 2002; Petersen and Aase
2004; Altarazi and Ammouri 2010; Chackelson et al. 2013; Emami, Arabzad, and Sajjadi 2014; Shqair, Altarazi, and
Al-Shihabi 2014). The next section presents more details about simulation-based research for warehouse design.
The enormous aspects of the warehouse design problem including decisions in different planning levels for the five
design categories, as detailed in Section 3; and the strong interrelationship between the design components require the
analysis to be conducted in a concurrent manner. Furthermore, the dynamic and interconnected stochastic nature of the
problem necessitates the utilisation of simulation modelling. This study proposes a concurrent simulation-based design
of experiments (DoE) approach for designing manual-order-picking warehouse. Considering their full combinations at
multi levels, the proposed approach considers five key manual warehouse design components; throughput, size, layout,
operational policies and number of utilised manpower/carts. Furthermore, the presented approach integrates all the key
warehouse functions (i.e. receiving, unloading, put away, storage, preparation and picking, shipping). The performance
of studied designs is evaluated using a supply chain-oriented criterion; the average cycle time for an SKU within a
warehouse, that is, the average elapsed time between entering and shipping an SKU to/from a warehouse. This cycle
time is different from the order-to-delivery time which is usually used in literature for order-picking performance mea-
surement. The adopted cycle time considers the time of all the functions in a warehouse, not just the picking, prepara-
tion and loading times; hence, it is an appropriate evaluator for warehouses performance from the entire supply chain
perspective.

2. Related research
An overview of previous simulation-based research related to warehouse design is presented in this section. Since ware-
houses are typically stochastic discrete-event systems, most simulation-related research have adopted discrete-event sim-
ulation (DES) type. Other simulation modelling types, such as agent-based modelling (ABM) and Petri nets (PN) formal
modelling were also implemented. This section also briefly overviews literature regarding adopted simulation for auto-
mated storage/retrieval systems (AS/RS) used in warehouses.
Caron, Marchet, and Perego (2000) presented a DES approach for efficient manual-order-picking warehouse layout
design using random or cube per order index storage policies. They found that the optimal number of aisles depends on
both strategic and operational decisions such as the size of the picking area in a warehouse, batch size, number of picks
in a tour, and utilised storage policy. In a comparable DES-based study, for both random and volume-based storage poli-
cies, Petersen (2002) studied the effect of the warehouse configuration, represented by the number of aisles per zone
and the length of aisles, on the total travelled time. The results showed that the warehouse shape is significantly affected
by the storage policy, aisle length and number of aisles, absence of a back cross-aisle and number of SKUs on the pick-
ing list. Dukic, Vedran, and Opetuk (2010) used DES to differentiate between different warehouse layouts; mainly tradi-
tional with/without aisles and fishbone. They found that the fishbone layout is very suitable for pallet picking, yet,
traditional layouts were found preferable in the case of multiple picking routes. Emami, Arabzad, and Sajjadi (2014)
developed a DES model for to optimise the trucks’ loading and unloading time in warehouses. The processes covered
by the model mainly include SKU’ checking, unloading, storing, delivering and loading.
Among all manual-order-picking warehouse design components, the ‘selection of operational policies’ has attracted
most simulation, particularly DES, studies. Petersen (2000) simulated five different order-picking policies: strict order,
batch, sequential zone, concurrent zone and wave picking. Petersen and Aase (2004) introduced a simulation model to
examine the effect of several picking, routing, and storage policies on the total distance travelled within the warehouse
to fulfil orders. Sensitivity analyses for the effect of order size, warehouse shape and demand distribution were also per-
formed. They found that orders batching and class-based or volume-based storage policies would provide the minimum
travelled distance. Macro and Salmi (2002) developed a universal warehouse storage simulation model to analyse the
storage capacity and rack efficiency of medium volume small-SKUs and medium volume large-SKUs warehouses. Their
results provided the type, amount and layout of pallet storage required for the two warehouses scenarios. Altarazi and
Ammouri (2010) presented a DES-based tool for selecting key warehouse resources considering both generic and case-
based warehouse characteristics. Their results showed that adopting the volume-based storage policy and changing the
manner of using the gating system can significantly affect warehouse resources utilisation. Conceptually similar to the
work of Altarazi and Ammouri (2010) but with focusing on detailed operational policy, Chackelson et al. (2013) pre-
sented a DoE-DES approach for simultaneously configuring and evaluating different warehouse policies levels. Altarazi,
Ammouri, and Alzubi (2012) proposed a DES methodology for manual-order-picking warehouse design considering key
 International Journal of Production Research 7105

design elements in various combinations and evaluated the performance of such combinations using multi performance
criteria. The suggested methodology provided an answer of what design characteristics should be adopted in order to
optimise a particular performance criterion. Roodbergen, Vis, and Taylor (2015) presented a warehouse design DES-
based method while concurrently considering scenarios of layout and operational policies. The results revealed that the
concurrent approach has a large potential in improving the warehouse performance.
Few tries can be found in literature regarding implementing ABM and PN modelling for the warehouse design prob-
lem. Ito and Abadi (2002) proposed an ABM for simulating a warehouse. The model integrated three subsystems;
agent-based communication system, agent-based material handling system and agent-based inventory planning and con-
trol system. The model was validated by comparing its simulation results with a prototype system performance. Kim
et al. (2002) presented an ABM for the operations’ control of an actual industrial warehouse. In the model, the agents
can communicate with each other in a dynamic real-time fashion and can make decisions, such as scheduling decisions,
accordingly. Shqair, Altarazi, and Al-Shihabi (2014) adopted ABM to simultaneously simulate the relations of strategic
(number of aisles, aisle length and number of storage blocks), tactical (storage assignment policy) and operational deci-
sions (routing policy and order size) pertaining to warehouse design. The study showed that having one cross-aisle only
and using a class-based storage policy can significantly decrease the distance travelled. This study also concluded that
the best routing policy is layout-dependent decision. Elbert et al. (2016) have also adopted ABM to evaluate the effi-
ciency of manual-order-picking routing policies when pickers deviate from pre-specified routes. They concluded that
even with these human-based deviations, optimal routing would remain the preferred strategy to implement. Basile, Pas-
quale, and Domenico (2012) used PN formal modelling approach for online monitoring, scheduling and rescheduling of
warehouses’ operational activities. Buil and Piera (2008) used Coloured PN (CPN) to integrate warehouse functions
(receiving, storage, picking and shipping) at the strategic, tactical and operational levels. The proposed approach resulted
in considerable savings on equipment and workers costs.
In addition to adopting simulation modelling for manual-order-picking warehouse, simulation has been widely used
also for the design and utilisation of various AS/RS types (Roodbergen and Vis 2009). Lee (1992) analysed class-based
storage for a man-on-board AS/RS through simulation. Rosenblatt, Roll, and Zyser (1993) suggested a simulation-
analytical approach to determine the size of an AS/RS system. Randhawa and Shroff (1995) extended this study, using
simulation, to investigate different inputs/outputs configurations on AS/RS multi-criteria performance including through-
put, mean waiting time and maximum waiting time. Potrč et al. (2004) implemented DES to determine travel times and
throughput for AS/RS considering different high storage racks and velocity profiles. Ekren et al. (2010) presented a
DES-DoE approach for unit load AS/RS to identify the factors affecting the system’s performance. Lerher, Šraml, and
Potrč (2011) presented DES and analytical modelling for evaluating the performance of mini-load multi-shuttle ASRS.
Dukic, Opetuk, and Lerher (2015) used simulation to test a throughput model for single dual-tray vertical lift module
(VLM) AS/RS with a human order-picker. Recently, Rosi et al. (2016) also presented a DES model for evaluating the
VLM performance while considering different configuration designs and velocity profiles.
The above review of literature indicates:
• The suitability and capability of simulation modelling in designing manual-order-picking warehouses and
automated warehouses.
• The lack of comprehensive concurrent simulation approach for designing manual-order-picking warehouses. Most
research on warehouse design via simulation considered single design component, mainly operational policies
(Rosenblatt, Roll, and Zyser 1993; Petersen and Aase 2004) and layout and/or sizing (Caron, Marchet, and Perego
2000; Dukic, Vedran, and Opetuk 2010); with less focus on manpower and equipment (Kosfeld 1998; Macro and
Salmi 2002) and throughput components. Few researches simultaneously considered two design components;
mainly the layout and operational policies (Altarazi and Ammouri 2010; Shqair, Altarazi, and Al-Shihabi 2014;
Roodbergen, Vis, and Taylor 2015; Petersen and Aase 2016). Thus, it is the synthesis of all design components
that appears to be lacking. As indicated by Gu, Goetschalckx, and McGinnis (2010), it is clear that warehouse
design-related research has focused on analysis rather than synthesis.
• Most related simulation-based research focused on the order-picking function (Caron, Marchet, and Perego 2000;
Petersen 2002; Petersen and Aase 2004; Chackelson et al. 2013; Shqair, Altarazi, and Al-Shihabi 2014) without
integrating other functions such as receiving, put away, unloading, storage, shipping functions. Accordingly, only
partial stochastic warehouse’s operational data are utilised in these models.
• By focusing on the order-picking function, most related research used the picking-route distance travelled, or its
associated time, as an evaluation metric. This metric has two drawbacks; first, it partially evaluates the outbound
part (order preparation, picking and loading) and ignores the inbound part (receiving, put away, storage, etc.) of a
warehouse functions. Second, the metric is an internal performance indicator which does not consider the supply
7106 S.A. Altarazi and M.M. Ammouri

chain perspective. Supply chain performance cares about the overall cycle time of a warehouse not its’ internal
travelling time.
The current research presents an effort to overcome the above-mentioned issues. To this end, the proposed approach
models the main components of warehouse design concurrently, considers the full probabilistic nature of warehouse
operations, models the key warehouses’ functions and adopts a supply chain-oriented performance metric.

3. Warehouse design
This section briefly explains the five key warehouse design decisions considered in this study; throughput, size, layout,
operational policies and utilised number of manpower/carts.
Warehouse throughput, which is also called warehouse SKUs flow, represents the flow of SKUs in a warehouse. It
is associated with the probabilistic nature of SKUs inflowing and departing the warehouse. Throughput volume is pro-
portionally related to the storage requirement for a warehouse (Cormier and Gunn 1996), accordingly, related to the
overall structure of a warehouse (Gu, Goetschalckx, and McGinnis 2010). In fact, different throughputs do not only
affect the internal performance of a warehouse but also affect the performance of the rest of the supply chain (Rushton,
Croucher, and Baker 2014). In the majority of throughput-considered literature, such as (Petersen and Aase 2004),
throughput is represented by the order size with deterministic values. In reality, throughput has a probabilistic nature. In
addition, throughput is not only related to the outbound side of the warehouse, which is the ordering side, but it is also
related to the inbound side, or the receiving side.
The sizing decision is mainly associated with the required capacity of a warehouse which, in its turn, is related to
potential sales. For a particular company or supply chain this decision must also be aligned with its total number of
warehouses. The importance of this decision comes from its inverse relation with the warehouse construction and inven-
tory costs (Cormier and Gunn 1996) and its correlation to the targeted customer service level; accordingly, with the
overall supply chain performance.
The warehouse layout design issue addresses the decision of locating the various warehouse departments (receiving,
sorting, storing, shipping, etc.) and more importantly, the decision of the internal layout design. The internal layout
design, also called the aisle configuration problem, mainly covers the determination of the number of blocks in a ware-
house (existence of cross-aisle/s or not); the number of aisles, orientation, length and width in each block; and the num-
ber and location of pick-up/drop-off (P/D) points (De Koster, Le-Duc, and Roodbergen 2007).
The operational policies in a warehouse can mainly be classified into storage assignment, routing and order-picking
or batching (Caron, Marchet, and Perego 1998; Petersen and Aase 2004; Roodbergen and Vis 2006). Storage assignment
is related to where SKUs are stored in the warehouse, routing is represented by the path followed by the pickers to
retrieve SKUs, and order-picking is associated with the grouping/ungrouping of customer orders in a picking list. Three
main storage assignment policies can be found in practice: random, volume-based and class-based. The warehouse rout-
ing policy is a variant of the classical travelling Salesman problem (TSP). Generally all TSP algorithms, whether opti-
mal or heuristic, can be adopted as routing policies after some customisation. Traversal routing is one example of
heuristic routing policy in which pickers must fully traverse the entire aisle once they enter it. Picking policies include
strict-order, batch, zone and wave picking policies.
Finally, the equipment selection decision determines the processes and systems to transfer and store products within
the warehouse. If properly carried out, this strategic decision can significantly increase profitability. Few researchers
addressed the equipment selection problem (Gu, Goetschalckx, and McGinnis 2010).Methods adopted in the equipment
selection include optimisation models (Telek 2013), expert systems (Kulak 2005), multi-criteria decision-making (Tuz-
kaya et al. 2010) and systematic frameworks (Hassan 2010). Mainly these systems evaluate the equipment selection
based on their total cost, fixed and variable and technical performance criteria at a given throughput. Since the current
research is studying manual warehouses, the equipment selection decision is expressed in terms of the number of man-
power/carts utilised in a warehouse.

4. The DoE-simulation framework

This section explains the details of the proposed approach experimental design, simulation modelling and performance
evaluation. A description for the adopted warehouse to implement the proposed approach is presented first.
 International Journal of Production Research 7107

4.1 Warehouse description and assumptions

The proposed approach is built based on the operations of a single-floor rectangular warehouse facility. The rectangular
shape is selected since it is the optimal geometric shape to store pallets (Berry 1968). As shown by Figure 1, four differ-
ent warehouse storage layout alternatives are investigated; the traditional one-block layout, with one cross-aisle layout
(two-block layout), the horizontal layout and the fishbone layout. The first three layouts are selected since they are com-
mon in literature (Roodbergen, Vis, and Taylor 2015; Gue and Meller 2009) while the fishbone layout is selected to rep-
resent non-traditional warehouses, that is, warehouses without parallel picking aisles (Gue and Meller 2009; Cardona
et al. 2015). For all layouts, the length/width ratio is set approximately as 0.5 since it is approximately the optimal ratio
(Bassan, Roll, and Rosenblatt 1980; Pohl, Meller, and Gue 2009). As mentioned earlier, the selection of optimal number
of cross-aisles in a warehouse depends on different characteristics (Caron, Marchet, and Perego 2000; Petersen 2002);
for comparison purposes, traditional warehouse with one cross-aisle is considered. Table 1 summarises the probabilistic
nature for the key warehouse functions; receiving, unloading, put away, storing, order preparation and picking, loading
and shipping. This data were identified in consistence with authors’ observations; particularly with three manual-
order-picking warehouses working in the Jordanian market: a warehouse for hygienic products, a warehouse for food

Traditional warehouse layout (Layout I) Traditional warehouse layout with one cross aisle (Layout II)

Fishbone warehouse layout (Layout III) Horizontal warehouse layout (Layout IV)

Figure 1. Warehouse layout alternatives (levels).

Table 1. Warehouse’s functions probabilistic data.

Function Statistical distribution

Receiving (time between Exp(x1*) h, per N(x2*,10) SKU

arrivals)
Unloading N(5, 1) min, per storage compartment
Putting away N(0.05, 0.01) min per storage compartment per metre
Storing Delay
Picking N(0.05, 0.01) min per storage compartment per metre
Preparation N(30, 5) min per order
Loading N(5, 1) min per compartment
Shipping (time between Exp(x3*) h per x4* orders. Order contains N(25, 1) for SKU A, N(15, 1) for SKU B, and N(10, 1)
departures) for SKU C

*x1, x2, x3, and x4 vary according to the level of the throughput design component (details in Section 4). For example, for the highest
throughput level (as seen in Table 2); x1 = 1, x2 = 75, x3 = 2, and x4 = 3.
7108 S.A. Altarazi and M.M. Ammouri

products, and a warehouse for chemical cleaning products. The warehouses’ size range between 1500 and 5500 m2, they
all use unit-load for SKUs movement, and they all are use single-deep pallet racking with various storage heights. Two
of the warehouses work for 16 h per day while the third works for 10 h per day. For two of the warehouses, the obser-
vation period for data collection was about 45 working days, 2–3 h per day distributed over the daily working time; so
data from all working conditions were collected. The data of the third warehouse was provided by the company accord-
ing to their long historical observation. Accordingly, the distributions in Table 1 were estimated based on sample sizes
with 100–150 observations.
Below are further assumptions of the adopted warehouse:
• Both the traditional layouts and fishbone layout have front and back cross-aisles while the horizontal layout has
only front cross-aisle.
• The picking aisles are two-sided and are wide enough to allow pickers to change direction and also to allow pick-
ing on both sides of the picking aisles.
• The P/D point is located in the middle of the front side of the warehouse and all picking and receiving activities
take place at the P/D point.
• All SKUs have the same size and 10 SKUs compose one compartment (unit load).
• All functions (picking, storing, etc.) are conducted using a cart that can handle a single compartment.
• All storage locations are identical in size and each location can store one compartment only.
• The storage height and the storage row depth are single compartment.
• The warehouse’s working time is 12 h. In the first 4 h, operators can work only in the receiving, unloading, put
away and storage. In the last four hours, they work on order preparation, picking and shipping. In the middle four
hours, operators can perform all warehouse functions.
• Each function in the warehouse is handled by a single operator.
• Strict-order-picking policy is adopted.
• Three different SKUs categories, A, B and C; are assumed to be stored in the warehouse with orders consist of
random mixes from these three categories. Yet, the ABC classification assumes a descending demand with highest
demand for category A (50%), moderate for B (30%) and lowest for C (20%). (Table 1 provides the probabilistic
details for the received orders (demand) for the three categories)

4.2 Experimental design

Table 2 shows the five studied warehouse design components and their levels. The ‘throughput or SKU flow’ compo-
nent has four levels where the second level, for example, implies that SKUs are received and shipped to and from the

Table 2. Warehouse design components and their levels.

Component Levels

Throughput*(SKUs 1. Receiving is Exp(2) h per N(50,10) SKU, and shipping every Exp(2) h per single order (low flow)
flow) 2. Receiving is Exp(2) h per N(100,10) SKU, and shipping every Exp(1) h per single orders (medium
flow)
3. Receiving is Exp(1) h per N(75,10) SKU, and shipping every Exp(2) h per three orders (high flow)
4. Receiving is Exp(1) h per N(125,10) SKU, and shipping every Exp(2) h per five orders (very high
flow)
Size 1. 300 storage locations (small size)
2. 400 storage locations (medium size)
3. 500 storage locations (big size)
Layout 1. Traditional layout
2. Traditional layout with one cross aisle
3. Fishbone layout
4. Horizontal layout
Operational policies 1. Random storage policy with no routing policy (no policies)
2. Volume-based storage policy with traversal routing policy (standard policies)
Manpower/carts 1. Six operators/carts are available (low manpower)
2. Twelve operators/carts are available (intermediate manpower)
3. Eighteen operators/carts are available (large manpower)

*Note that the receiving and shipping functions which are explained in Table 1 are both represented in the throughput component.
 International Journal of Production Research 7109

warehouse according to the exponential distribution with mean times between receiving and shipping equals to two
hours and one hour, respectively. In addition, this level assumes that the amount of received SKUs is normally dis-
tributed with mean of 100 SKUs and standard deviation of 10 SKUs while the shipping is conducted for individual
orders. The second component is related to the warehouse size with three levels: small, medium, and big; having storage
capacity of 300, 400 and 500 storage locations, respectively. Because of the interaction between the diagonal aisles and
rectangular shape the storage locations in the fishbone layout were exactly 292, 404 and 484 for the small, medium and
big sizes, respectively. These differences are within the ±4% used by Pohl, Meller, and Gue (2011). The third compo-
nent is related to the warehouse layout with four levels indicating the different warehouses layout alternatives described
in Section 4.1. The fourth component has two levels: the first level represents lack of operational policies (random stor-
age policy with no routing policy) while the second level represents common practical operational policies (volume-
based storage policy with traversal routing policy) (Caron, Marchet, and Perego 1998; Petersen and Aase 2004; Rood-
bergen and Vis 2006). Finally, the fifth component corresponds to the utilised number of manpower/carts with low,
intermediate and large levels. As a result of the above arrangement, 288 (4 × 3 × 4 × 2 × 3) combinations/experiments
have emerged. Simulation experiment for each combination was conducted.

4.3 Simulation modelling

Simulation modelling can be implemented through formal modelling such as PN (Basile, Pasquale, and Domenico
2012), CPN (Buil and Miguel 2008), matrix-based framework (Giordano et al. 2008); commercial simulation tools such
as ARENA (Altarazi and Ammouri 2010; Ekren et al. 2010; Chan and Hing 2011; Emami, Arabzad, and Sajjadi 2014;
Peixoto et al. 2016), ProModel (Dukic, Vedran, and Opetuk 2010), AutoMod (Hwang and Gyu 2006), Enterprise
Dynamic (Chackelson et al. 2013), NetLogo (Shqair, Altarazi, and Al-Shihabi 2014); or appropriate programming lan-
guage such as visual basic (Roodbergen, Vis, and Taylor 2015) and Delphi (Basile, Pasquale, and Domenico 2012). PN
is one of the most resourceful modelling tool for discrete-event systems (Mueller 2007), its’ models are usually modular
and scalable (Basile, Pasquale, and Domenico 2012), and suits large systems with many subsystems (Zhou and
Venkatesh 1999). However, researchers on warehouse design via simulation usually prefer to implement commercial
simulation tools; implementation of formal modelling approaches, in warehouse design, has been mainly associated with
operational control issues. This can be attributed to the higher complexity of the formal model in comparison with the
friendly-use of commercial simulation packages. According to Baker and Canessa (2009), more than half of warehouse
designers use simulation software during the warehouse design process. ARENA 15.0 software was used in this
research. During the last years, ARENA had established a proven track record of enabling companies to model and
evaluate supply chains and warehouses (ARENA 2017). Several ARENA-adopted case studies covering supply chain,
logistics and retail case studies can be found in (ARENA 2017).

4.3.1 Model building

Several modules were developed and integrated together to form the overall ARENA model; entities, resources and stor-
age queues. The entities module consists of three types: stored SKUs entities, order list (picking list) entities and logic
entities. The stored SKUs entities describe the SKUs to be stored. The order list entities specify the order with the
amount, as in Table 1, of each SKU category to be picked to fulfil an order. The logic entities are used to compute the
evaluation performance measure and to initialise the definition of warehouse layout capacity. Attributes are assigned to
each SKU entity and order-list entity for identification purposes and to save the entrance time of each SKU to the ware-
house. The resources module consists of four types: unloading manpower at each receiving dock, loading manpower
who loads the picked SKU to the transporting carts, storing manpower who drives the cart and searches for the best
storage location based on the routing policy adopted and picking manpower who receives the order and delivers it from
its storing locations to the shipping dock. It is worth mentioning that a resource is capable of handling one compartment
per task and that the number of resources varied according to manpower/carts factor levels. Finally, storage queues rep-
resent storage locations where compartments are stored.
The model can be viewed in two main parts. The first part models the receiving, unloading, put away, and storage
functions. The second part is for order preparation and picking, loading and shipping functions. Figures 2 and 3 present
the flowcharts for these two parts, respectively. In Figure 2, stored entity represents the arrival of trucks to the P/D point
according to the statistical distribution given in Table 1. The unloading manpower resource unloads the received SKUs,
classifies them into the main categories and batches 10 SKUs of the same category together to form a compartment.
The compartment remains at the receiving dock, until a storing manpower resource became idle to pick it and store it in
a suitable location according to first-come first-serve (FCFS) dispatching rule. If random policy is implemented, the stor-
7110 S.A. Altarazi and M.M. Ammouri

Figure 2. The simulation flow chart of inbound warehouse operations (receiving, unloading, putting way and storing).

ing manpower resource starts searching from the nearest empty location then advances to next locations. When an
empty location is found, the compartment is placed there and storing manpower resource returns to the P/D point to
pick another compartment. On the other side, if volume-based storage policy is selected; the search starts from the first
location of the storage compartment’s category zone.
In the second part of the simulation model (Figure 3), according to the throughput level an outbound transaction is
created by the arrival of orders. Next, an order list of the required quantities of each of the main categories (A, B, C) is
identified as attributes and assigned to the order entity. Order list is picked by an available picking manpower resource
from the P/D point according to FCFS dispatching rule. Based on the adopted operational policies; the picking man-
power resource searches for the appropriate SKUs in order to fulfil the order list. If random policy is adopted, the
resource starts searching from the nearest storage location. The resource picks the SKUs according to their attribute
(quantity of each category). The resource continues searching for SKUs in different location until fulfilling the whole
order; accordingly, a single order can be picked from one or more than one storage location. In a volume-based storage
policy, the manpower picking resource starts searching for a certain SKU category from the first location of the cate-
 International Journal of Production Research 7111

Figure 3. The simulation flow chart of outbound warehouse operations (picking, order preparation and loading).

gory’s zone until the whole order is fulfilled. After fulfilling the order, the resource returns to the P/D point for order
shipping.
7112 S.A. Altarazi and M.M. Ammouri

Random variants have been generated for each distribution in each replicate, and stored in external files to be read
by the model during the simulation experiments. These files include the replicate input; the time between truck arrivals,
the load carried by each truck, the number of arriving items in the order list and times. These files are unique for each
simulation replicate and throughput level. These files are used to guarantee identical replicate’s random variants in the
simulation runs for different experiment scenarios; which insures that in the same replicate number, for an experiment
scenario, the same random variates are used. Other experiment scenarios variables include warehouse characteristics
such as number of aisles and storage location. These variables vary depending on the experimented warehouse size in
the simulation run experiment. Warm-up period has been defined as 10% of simulation run and the model performance
was evaluated during the simulation run time after the warm-up period. Throughput levels were defined after finishing
the warm-up period and entering steady-state period.

4.3.2 Model verification and validation

The model is built as small sub-models and additional logical variables were incorporated accordingly. The model logic
was verified through debugging and tracking variables’ values. Each sub-model was built and debugged separately, then,
the sub-models were integrated and rechecked. Integration continued until the whole model was created.
To validate the model, its behaviour was compared with a real-life warehouse presented in a previous work for the
same authors of the current research (Altarazi and Ammouri 2010). In comparison with the design components pre-
sented in Table 2, the characteristics of the real-life approximately-rectangular warehouse would be: very high through-
put, large size, horizontal layout, random storage policy, strict-order-picking policy, traversal routing and intermediate
manpower/carts. The simulation modelling methodology explained in Section 4.3.1 was run for this real-life warehouse
considering all the warehouse functions presented in Table 1 (data about these functions for the real-life warehouse and
other details about it can be found in (Altarazi and Ammouri 2010). The simulated average cycle time (ACT) per SKU
was estimated by the simulation model, then, this simulated ACT was compared with the actual ACT per SKU. Very
comparable values were found between the actual and simulated ACT. As a result, the proposed modelling methodology
was validated.

4.4 Performance evaluation

The ACT for an SKU is the criterion used to assess the performance of the simulation experiments. Equation (1)
expresses how the simulation model computes the ACT. Logic entities, assigned attributes and different model variables
were used to calculate the ACT.
Pi¼SKUs
CT i
ACT ¼ i¼1 (1)
n
and
CT i ¼ ST i  ET i
where
CT i is the cycle time of SKUi
STi is the time once SKUi is ready for shipment
ETi is the time when SKUi enters the warehouse
n is total number of SKUs picked during a simulation run

As can been seen, the ACT is calculated as the average time difference between receiving and shipping SKUs in a
simulation run. This time aggregates the times used by all functions listed in Table 1; including SKU unloading, putting
away, storing, picking, preparation and loading times. It is worth noticing that most related research includes only the
times of picking, preparation and loading to calculate the order-to-delivery time and ignores the times of other
functions.

5. Analysis of results and discussions

This section reports the results and analyses generated for the 288 simulation experiments. Table 3 summarises the sim-
ulation results. Each result is an average of 20 simulation replications and each experiment consists of 100 h runtime.
To interpret the obtained results, several types of analyses were conducted. Analysis of Variance (ANOVA) was used to
 International Journal of Production Research 7113

Table 3. Simulation-obtained results for ACT (in hours).

Operational policies
No policies Standard policies
Manpower/carts Manpower/carts
Throughput Size Layout Low Intermediate Large Low Intermediate Large

Low Small Traditional 11.76 10.54 11.02 10.68 9.94 9.67

Trad. with cross aisle 10.68 10.15 9.91 9.60 9.12 9.94
Fishbone 10.58 11.18 10.46 9.19 8.74 8.02
Horizontal 10.78 10.97 11.30 9.10 8.50 9.10
Medium Traditional 11.98 11.45 12.58 11.02 10.78 9.60
Trad. with cross aisle 10.97 11.14 9.72 9.60 9.53 9.43
Fishbone 11.98 11.23 11.50 9.17 9.70 9.34
Horizontal 12.14 10.97 10.08 9.34 9.12 8.18
Big Traditional 10.78 10.27 11.02 10.01 10.42 9.94
Trad. with cross aisle 11.64 10.78 10.92 10.92 10.10 9.24
Fishbone 12.48 11.59 11.35 11.02 10.27 9.36
Horizontal 10.97 10.85 10.82 9.36 9.17 9.07
Medium Small Traditional 10.92 9.19 10.08 9.29 7.61 7.90
Trad. with cross aisle 9.07 9.05 9.60 8.35 7.25 7.01
Fishbone 10.82 9.72 9.58 8.59 8.45 8.18
Horizontal 9.24 9.72 9.72 8.52 8.33 8.38
Medium Traditional 11.16 10.75 10.13 9.86 8.28 8.14
Trad. with cross aisle 10.92 10.92 9.77 10.01 9.29 7.73
Fishbone 10.66 10.51 9.96 10.51 9.07 8.66
Horizontal 10.99 10.08 10.42 8.93 8.06 7.39
Big Traditional 11.02 10.25 9.91 9.07 8.64 8.40
Trad. with cross aisle 10.85 10.58 10.13 8.47 8.28 7.63
Fishbone 11.71 10.08 10.27 9.19 8.26 8.28
Horizontal 10.25 9.62 10.15 9.14 7.49 8.02
High Small Traditional 6.34 10.20 9.48 7.51 10.82 7.10
Trad. with cross aisle 7.32 8.93 8.38 7.27 5.59 6.74
Fishbone 12.36 10.01 8.54 9.34 9.22 7.87
Horizontal 9.42 9.31 9.24 9.14 7.48 6.91
Medium Traditional 10.18 10.08 9.10 13.34 13.34 12.46
Trad. with cross aisle 8.62 7.97 8.93 10.37 9.79 9.14
Fishbone 10.61 10.30 9.26 8.78 9.50 8.23
Horizontal 9.91 9.70 8.74 8.83 6.86 9.46
Big Traditional 13.10 9.65 9.21 11.47 9.02 8.78
Trad. with cross aisle 11.95 10.61 8.26 9.34 8.86 6.91
Fishbone 11.40 10.10 9.10 8.86 8.28 8.38
Horizontal 9.06 8.50 8.26 5.88 5.78 4.61
Very High Small Traditional 7.92 9.05 7.25 7.25 7.13 6.82
Trad. with cross aisle 8.41 8.42 13.92 8.57 6.32 5.70
Fishbone 10.75 9.31 11.88 10.61 7.58 7.66
Horizontal 10.85 9.07 7.46 12.17 6.50 8.47
Medium Traditional 10.13 9.84 4.78 11.08 9.12 8.18
Trad. with cross aisle 8.54 8.04 7.34 9.34 8.71 8.13
Fishbone 10.49 8.98 9.44 10.13 7.32 7.13
Horizontal 9.29 8.81 6.77 8.90 8.21 6.11
Big Traditional 15.71 12.31 8.98 8.66 10.54 6.34
Trad. with cross aisle 15.02 7.30 7.20 9.53 7.90 6.65
Fishbone 10.61 8.83 8.47 9.54 8.59 7.43
Horizontal 9.05 8.95 5.35 9.86 7.97 4.27

identify which warehouse design components (main factors) or components’ interaction significantly affects the ACT.
Main effect plots and two-way interaction plots were generated to illustrate how levels’ means differ and which levels
can achieve the best ACT performance. The effects of some of the studied parameters were intuitive. The effects of the
7114 S.A. Altarazi and M.M. Ammouri

throughput interactions with other components, however, were not clear; hence, multi-comparison Tukey’s tests were
conducted to analyse the effects of these interactions. Minitab statistical package was used for analysing the results.
In addition to the results of the primary 288 experiments, this section also presents the results and analyses of
another 288 experiments which were conducted as a sensitivity analysis regarding the key assumption of single SKU in
each SKU category. Warehouses’ design principles are drawn at the end of the section.

5.1 ANOVA
Table 4 shows the ANOVA results. The F-ratio for the five main components, their two-way, and three-way interactions
(four-way and five-way interaction effects were neglected) was found by dividing the mean square of the effect of inter-
est by the error mean square. Setting type one error at 0.05, any effect with a P-value less than 0.05 indicates signifi-
cance. P-values reveal that all single effects, most the two-way interactions, and about half the three-way interactions;
are significant. In fact, this two-way and three-way interactions’ significance support the argument regarding the neces-
sity for a synthesis warehouse design perspective which, as evidenced by this research, can be enabled by simulation.
Figure 4 (a) and (b) further explains the ANOVA results. It is worth mentioning that several residual checking were
conducted to validate the ANOVA model (not provided for brevity reasons). The normal probability plot and the his-
togram for standard residuals have not indicated any deviation from normality, thus, normality assumption was vali-
dated. The plots of the ‘fitted values’ and ‘observation order’ against the standard residuals demonstrated the
randomness of residuals. Hence, it could be assumed that variability is independent for both ‘fitted values’ and ‘observa-
tion order’. Finally, from last two plots, variability was verified constant with respect to both variables.
Explaining the effects of operational policies and utilised number of manpower/carts is straightforward. Adopting
volume-based storage and traverse routing, in comparison with random storage and no routing policy; is translated into
a decrease in the ACT. In fact, several previous studies have resulted in similar results including (Petersen and Aase
2004). Similar to the effect of adopting ‘standard policies’, an increase in the number of manpower/carts directly turned

Table 4. Full factorial ANOVA for the ACT results.

Source DoF Seq SS Adj MS F ratio P-value

Throughput 3 100.5 33.5 35.69 0.000

Size 2 14.6 7.3 7.82 0.001
Layout 3 41.0 13.6 14.55 0.000
Operational policies 1 122.4 122.4 130.32 0.000
Manpower/carts 2 81.1 40.5 43.21 0.000
Throughput × Size 6 15.1 2.5 2.68 0.017
Throughput × Layout 9 15.8 1.7 1.87 0.062
Throughput × Operational Policies 3 8.3 2.7 2.97 0.034
Throughput × Manpower/carts 6 27.7 4.6 4.92 0.000
Size × Layout 6 30.7 5.1 5.46 0.000
Size × Operational Policies 2 12.8 6.4 6.82 0.002
Size × Manpower/carts 4 16.3 4.0 4.35 0.002
Layout × Operational Policies 3 6.7 2.2 2.39 0.072
Layout × Manpower/carts 6 3.3 0.5 0.60 0.729
Operational Policies × Manpower/carts 2 0.8 0.4 0.45 0.637
Throughput × Size × Layout 18 42.8 2.3 2.53 0.001
Throughput × Size × Operational Policies 6 19.3 3.2 3.44 0.003
Throughput × Size × Manpower/carts 12 21.6 1.8 1.92 0.037
Throughput × Layout × Operational Policies 9 19.7 2.1 2.33 0.018
Throughput × Layout × Manpower/carts 18 23.4 1.3 1.38 0.149
Throughput × Operational Policies × Manpower/carts 6 0.7 0.1 0.14 0.991
Size × Layout × Operational Policies 6 12.1 2.0 2.15 0.052
Size × Layout × Manpower/carts 12 12.9 1.0 1.15 0.328
Size × Operational Policies × Manpower/carts 4 9.6 2.4 2.57 0.041
Layout × Operational Policies × Manpower/carts 6 3.3 0.5 0.59 0.741
Error 132 124.0 0.9
Total 287 787.7
S = 0.969286 R2 = 84.26% R2(adj) = 65.77%
 International Journal of Production Research 7115

Figure 4. (a) Main effects’ and (b) Two-way interaction effects’ plots for the ACT. (Components’ levels are similar to what listed in
Table 2, for example level 3 for the throughput means ‘high’ throughput).

into a decrease in the ACT. This was expected since more resources means less number of order to pick per resource,
or less average number of SKUs in orders; which means better warehouse performance or less ACT.
Regarding the throughput (SKUs flow) component, Figure 4(a) shows that as throughput increases the ACT
decreases. This is mainly because high throughput may cause high SKUs percentage to be cross-docked. This cross-
docking behaviour implies going directly from put away to loading, skipping the storage delay, and reducing the time
of the picking function; accordingly, reducing the ACT. With lower throughput levels, SKUs go through all warehousing
functions resulted in higher ACT.
Considering the size component, Figure 4(a) shows that the best ACT can be achieved by the smallest warehouse
size regardless, mostly, the status of other components (as can be seen from Figure 4(b)). Intuitively, since increasing
the warehouse size results in an increase in the putting away and picking routes’ distances; it will cause ACT to
increase. Medium and large warehouse designs resulted in higher ACT (this is also shown in Figure 6 for the 10 SKUs
scenario presented in Section 5.3). In fact, Figures 4(a) and 6 show that the superiority in ACT is not always toward
one of these two sizes. This contradiction is revealed in Figure 4(b); the ACT performance of medium and large sizes
differs based on the status of other components levels. For example, large size results in better ACT when standard
operational policies are implemented. In real life, adopting the above results can help the proper fit of supply chains on
the efficiency–responsiveness spectrum. That is, a small size warehouse is recommended for achieving better supply
chain responsiveness.
For the warehouse layout design component, traditional layout produced the worst ACT; yet adding a cross-aisle,
that is the two-block layout, has enhanced the ACT. Actually, this result confirms literature findings (Roodbergen and
Koster 2001; Shqair, Altarazi, and Al-Shihabi 2014); however the current study confirmed that this result is still valid
taking into consideration more design components than those considered in literature and regardless of the levels of
these design components. Statistically, fishbone layout shows a comparable performance with the traditional layout as
seen from Figures 4(a) and 6. Interestingly, the horizontal layout generated the best ACT regardless of the status of
7116 S.A. Altarazi and M.M. Ammouri

other design components, as seen in Figure 4(b), and even with the 10 SKUs scenario as in Figure 6. This result can be
allied to the fact that the distance travelled in the horizontal layout is less than that in other layouts, that is, cross-aisle
travelling in the horizontal layout allows the picker to start picking items in the second section immediately while extra
travelling would be required, for example, in traditional layout (Caron, Marchet, and Perego 2000).
Figure 4(b) plots the interaction effects between each pair of the five components. As also indicated by the P-values
in Table 4, some pairs involve no significant interaction. For instance, no interaction exists between operational policies
and manpower/carts, that is, standard operational policies are always better than random policies regardless of the man-
power/carts level. Similarly, no interaction exists between layout and manpower/carts. This was expected since changing
the manpower/carts won’t affect the travelled distance if the layout has changed. On the other hand, most of the interac-
tions are significant; such as the throughput-size, throughput-layout, throughput-manpower and size-layout interactions.
Some interactions’ effects is understood, for example, the reduction in travelling distance associated with horizontal lay-
outs is more significant with large warehouses sizes than with small and medium sizes; hence interaction effect exists
for the size-layout resulting in different ACT performance of small size comparing to other sizes when layout varies.
However, the effect of other interaction is not clear. Examples of such cases include the throughput-size and through-
put-layout interactions. Tukey’s tests were implemented to reveal these issues.

5.2 Tukey’s multi-comparisons tests

Tukey’s tests were performed to reveal what throughput’s levels do interact with the levels of other design components.
Before presenting the test results, it is worth mentioning that the Tukey’s-tested simulated data were verified to confirm
the test’s assumptions including data independency within and between samples (this was guaranteed by the simulation
experiments), normally distributed data and homogenous variance across the samples (normal probability plots were
generated to validate these two assumptions).
Figure 5(a) and (b) summarises the results of Tukey’s tests. Bold lines were to present the test results with two
bases: (1) the compared configurations are ranked in ascending order with respect to the ACT performance (lower ACT

Figure 5. Tukey’s tests for (a) throughput by size and (b) throughput by layout.
 International Journal of Production Research 7117

indicates better performance) and (2) if the difference between two configurations (with respect to ACT) is insignificant
the bold line will be straight. For example, in Figure 5(a) for the ‘throughput small size’ comparisons: (1) the ascending
order for the throughput levels according to the ACT is ‘very high’, ‘low’, ‘medium’, and ‘high’; (2) the ‘very high’
throughput level has significantly higher ACT performance than all the other three levels while all the other three levels
are statistically indifferent from each other, hence, their tests are represented by a straight bold lines.
Figure 5(a) shows that small warehouses can significantly reduce the ACT only when the throughput is ‘very high’.
That is, it seems that the cross-docking practice is more applicable in small warehouses sizes; on the other hand, with
medium and large warehouses high inventory would be available and stored SKUs would be used to fulfil the orders
instead of the just-arrived SKUs.
Figure 5(b) illustrates that no difference in ACT performance is achieved for different throughputs levels for the hor-
izontal and fishbone warehouse layouts. However, traditional layouts, either with or without cross-aisle, performs better
when the throughput level is very high. In other words, traditional layouts performance is enhanced with very high-
throughput level while other layouts’ performance does not. Because traditional layout encloses the highest travelled dis-
tance among other layouts, it is the most affected by high throughput. On the other hand, high throughput would have
lower effect on other layouts since their travelled distance, comparing to traditional layout, are originally lower.

5.3 Single vs. multiple SKU per each SKU category

Sensitivity analysis was conducted to test the effect of number of SKUs within each SKU category. The original model,
called here the three SKUs scenario, explained in Section 3 assumes that each SKU category (A, B and C) consists only
of a single SKU type. In practice, a number of SKUs types can be found within each SKU category, hence requiring
different storage locations for each type but within the category’s specified space (as shown in Figure 1). Simulating all
scenarios of ‘what SKU types can be included in each category’ is not viable. Hence, for demonstration purposes a sce-
nario, called the 10 SKUs scenario, of 2, 3 and 5 SKUs types in categories A, B and C, respectively; was simulated for
the whole 288 combinations. The simulation results (not provided for brevity reasons) for the new 288 experiments of
the 10 SKUs scenario showed higher ACT values than the three SKUs scenario. This was expected since more routing
in the ‘picking’ and ‘putting away’ functions are required for the 10 SKUs scenario.
Full ANOVA analysis was performed for the 10 SKUs scenario. Main effects plots are given by Figure 6. Except
for some of throughput’s interactions; significant main factors, interactions and optimum levels were found similar to
what found in the three SKUs scenario. The dominance performance for very high throughput and small size’s ware-
houses are still valid, however, Figure 6 clearly shows that there is a throughput limit above which different throughput
levels become insignificant in affecting the ACT. As explained before, cross docking can be behind the low ACT for
the very high-throughput small size design configuration. The ACT performance enhancement of the low throughput
level in the 10 SKUs model, which is very comparable to medium and high levels while it was much worse in the
3SKUs scenario; can be attributed to the larger number of SKUs a picker can pick in a single trip comparing to the
three SKUs scenario.

Figure 6. Main effects’ plots for the ACT for the 10 SKUs scenario.
7118 S.A. Altarazi and M.M. Ammouri

5.4 Warehouses’ design principles

Based on the above analyses, the following principles for manual-order-picking warehouse design can be suggested.
• In general, the following design components produce the best ACT performance: horizontal layout, very high
throughput, standard operational policies, large manpower/carts and small size.
• In supply chains with uncertain throughput scenarios, the ACT performance can be enhanced by going to very
high-throughput level if the warehouse size is small and/or layout is traditional with or without aisle.
• If it is viable in the case of medium throughput levels, it is recommended either to sum more throughputs into
one big size warehouse or to split the throughput and allocate it to two small warehouses.
• For traditional warehouse layout, adding a cross-aisle will improve the warehouse performance.
• Even for low throughput-small size warehouses, volume-based storage policy with traversal routing policy is better
than ‘random storage policy with no routing policy’.
• The above guidelines are valid for single or multiple SKU per each SKU category. Yet, as the number of SKUs
per SKU category increases the ACT increases.
It is worth noticing that these guidelines are only applicable within the assumptions and considerations of the current
study. For example, in case of changes in the levels of the design components, different statistical distributions of ware-
house functions, non-rectangular shape of the warehouse, multiple-floor warehouse, double-deep racks and any violation
of assumptions stated in Section 4.1; then the resulted guidelines could be infeasible. Additionally, as those guidelines
were resulted based on the ACT performance, they could be invalid if other performance criterion, such as space utilisa-
tion, was utilised.

6. Conclusions and future work

This study presented a concurrent DoE-simulation modelling approach for the manual-order-picking warehouse design
problem. Warehouses’ throughput, size, layout, manpower/carts, and operational policies were the five design compo-
nents incorporated in the simulation model. Warehousing key functions of receiving, unloading, put away, storing, order
preparation and picking, loading and shipping; were all included also. The total time an SKU spends in the warehouse
was utilised to evaluate the warehouse performance. SKUs were divided into three categories with different demand
levels. Two scenarios of ‘number of SKUs per an SKU category’ were simulated, each with 288 design combinations.
Thorough statistical analyses for the simulation results of the two scenarios were conducted and the following key find-
ings were concluded:
• Horizontal layout is preferable over all other studied layouts.
• In general, small size warehouses perform better than larger sizes.
• Adding an aisle to a traditional layout improves its performance regardless of other design components’ situation.
• Higher throughput is recommended for traditional layout-small size warehouses.
• Compromising should be implemented between the number of SKUs in a warehouse and its ACT performance.
Finally, the current research emphasises the need for concurrent warehouse design-oriented studies taking into con-
sideration the probabilistic nature of all warehouses’ functions and an external customer performance evaluation. Direc-
tions for future research include extending the design components levels; for example to consider more size and
throughput levels, other operational policies, and other layouts alternatives. Incorporating more warehouse design issues
such as the storage method utilised, and the size of unit load could also be done. It is also worth investigating to con-
sider multi indicators for the warehouse performance, for instance, to consider in addition to cycle time the utilisation of
warehouse space, manpower and material handling equipment. A multi-objective analytical optimisation modelling for
the warehouse design problem is another area to be investigated. Finally, it would be interesting to explore implement-
ing PN as a simulation formal model for the concurrent warehouse design problem.

Disclosure statement
No potential conflict of interest was reported by the authors.

ORCID
Safwan A. Altarazi http://orcid.org/0000-0002-0299-012X
 International Journal of Production Research 7119

References

Altarazi, S., and M. Ammouri. 2010. “A Simulation-based Decision Making Tool for Key Warehouse Resources Selections.” Paper
presented at the World Congress on Engineering conference, London, June 30–July 2.
Altarazi, S., M. Ammouri, and E. Alzubi. 2012. “Concurrent Manual-order Picking Warehouses Design via Simulation.” Paper pre-
sented at the Institute of Industrial Engineers Annual Conference, Orlando, May 19–23.
Arena Simulation Software. 2017. Accessed 11 August 2017. https://www.arenasimulation.com.
Baker, P., and M. Canessa. 2009. “Warehouse Design: A Structured Approach.” European Journal of Operational Research 193 (2):
425–436. doi:10.1016/j.ejor.2007.11.045.
Basile, F., C. Pasquale, and D. Domenico. 2012. “A Control Oriented Model for Manual-pick Warehouses.” Control Engineering
Practice 20 (12): 1426–1437. doi:10.1016/j.conengprac.2012.08.008.
Bassan, Y., Y. Roll, and M. Rosenblatt. 1980. “Internal Layout Design of a Warehouse.” American Institute of Industrial Engineers
Transactions 12 (4): 317–322. doi:10.1080/05695558008974523.
Berry, J. 1968. “Elements of Warehouse Layout.” International Journal of Production Research 7 (2): 105–121. doi:10.1080/
00207546808929801.
Buil, R., and M. A. Piera. 2008. “Warehouse Redesign to Satisfy Tight Supply Chain Management Constraints.” WSEAS Transactions
on Information Science and Applications 5 (3): 286–291.
Bukchin, Y., E. Khmelnitsky, and P. Yakuel. 2012. “Optimizing a Dynamic Order-picking Process.” European Journal of Operational
Research 219 (2): 335–346. doi:10.1016/j.ejor.2011.12.041.
Cardona, L. F., D. F. Soto, L. Rivera, L., and H. J. Martínez. 2015. “Detailed Design of Fishbone Warehouse Layouts With Vertical
Travel.” International Journal of Production Economics 170: 825–837. doi:10.1016/j.ijpe.2015.03.006.
Caron, F., G. Marchet, and A. Perego. 1998. “Routing Policies and COI-based Storage Policies in Picker-to-part Systems.” Interna-
tional Journal of Production Research 36 (3): 713–732. doi:10.1080/002075498193651.
Caron, F., G. Marchet, and A. Perego. 2000. “Layout Design in Manual Picking Systems: A Simulation Approach.” Integrated Manu-
facturing Systems 11 (2): 94–104. doi:10.1108/09576060010313946.
Chackelson, C., A. Errasti, D. Ciprés, and F. Lahoz. 2013. “Evaluating Order Picking Performance Trade-offs by Configuring Main
Operating Strategies in a Retail Distributor: A Design of Experiments Approach.” International Journal of Production Research
51 (20): 6097–6109. doi:10.1080/00207543.2013.796421.
Chan, F., and K. Hing. 2011. “Improving the Productivity of Order Picking of a Manual-pick and Multi-level Rack Distribution Ware-
house Through the Implementation of Class-based Storage.” Expert Systems with Applications 38 (3): 2686–2700. doi:10.1016/
j.eswa.2010.08.058.
Chen, F. Y., H. W. Wang, C. Qi, and Y. Xie. 2013. “An Ant Colony Optimization Routing Algorithm for Two Order Pickers With
Congestion Consideration.” Computers and Industrial Engineering 66 (1): 77–85. doi:10.1016/j.cie.2013.06.013.
Cormier, G., and E. Gunn. 1996. “Simple Models and Insights for Warehouse Sizing.” Journal of the Operational Research Society
47 (5): 690–696. doi:10.1057/jors.1996.80.
De Koster, R., T. Le-Duc, and K. Roodbergen. 2007. “Design and Control of Warehouse Order Picking: A Literature Review.” Euro-
pean Journal of Operational Research 182 (2): 481–501. doi:10.1016/j.ejor.2006.07.009.
Dukic, G., V. Vedran, and T. Opetuk. 2010. “Order-picking Methods and Technologies for Greener Warehousing.” Strojarstvo 52 (1):
23–31.
Dukic, G., T. Opetuk, and T. Lerher. 2015. “A Throughput Model for a Dual-tray Vertical Lift Module With a Human Order-picker.”
International Journal of Production Economics 170: 874–881. doi:10.1016/j.ijpe.2015.04.009.
Ekren, B. Y., S. S. Heragu, A. Krishnamurthy, and C. J. Malmborg. 2010. “Simulation Based Experimental Design to Identify Factors
Affecting Performance of AVS/RS.” Computers and Industrial Engineering 58 (1): 175–185. doi:10.1016/j.cie.2009.10.004.
Elbert, R. M., T. Franzke, C. H. Glock, and E. H. Grosse. 2016. “The Effects of Human Behavior on the Efficiency of Routing Poli-
cies in Order Picking: The Case of Route Deviations.” Computers and Industrial Engineering 111: 537–551. doi: 10.1016/
j.cie.2016.11.033.
Emami, S. B., S. M. Arabzad, and S. M. Sajjadi. 2014. “A Simulation Study on Warehouse Loading System: The Case of Poultry
Feed Production Factory.” International Journal of Logistics Systems and Management 19 (3): 347–355. doi:10.1504/
IJLSM.2014.065501.
Fukunari, M., and C. J. Malmborg. 2008. “A Heuristic Travel Time Model for Random Storage Systems Using Closest Open Loca-
tion Load Dispatching.” International Journal of Production Research 46 (8): 2215–2228. doi:10.1080/00207540601118462.
Giordano, V., J. Zhang, D. Naso, and F. Lewis. 2008. “Integrated Supervisory and Operational Control of a Warehouse with a
Matrix-based Approach.” Institute of Electrical and Electronics Engineers Transactions on Automation Science and Engineer-
ing 5 (1): 53–70. doi:10.1109/TASE.2007.891472.
Gu, J., M. Goetschalckx, and L. McGinnis. 2010. “Research on Warehouse Design and Performance Evaluation: A Comprehensive
Review.” European Journal of Operational Research 203 (3): 539–549. doi:10.1016/j.ejor.2009.07.031.
Gue, K., and R. Meller. 2009. “The Aisle Configurations for Unit-Load Warehouses.” Institute of Industrial Engineers Transactions
41 (3): 171–182. doi:10.1080/07408170802112726.
Hassan, M. M. 2010. “A Framework for Selection of Material Handling Equipment in Manufacturing and Logistics Facilities.” Jour-
nal of Manufacturing Technology Management 21 (2): 246–268. doi:10.1108/17410381011014396.
7120 S.A. Altarazi and M.M. Ammouri

Hwang, H., and S. Gyu. 2006. “A Performance Evaluation Model for Order Picking Warehouse Design.” Computers and Industrial
Engineering 51 (2): 335–342. doi:10.1016/j.cie.2005.10.002.
Ito, T., and S. M. Abadi. 2002. “Agent-based Material Handling and Inventory Planning in Warehouse.” Journal of Intelligent Manu-
facturing 13 (3): 201–210.
Kim, B., J. Graves, S. Heragu, and A. Onge. 2002. “Intelligent Agent Modeling of An Industrial Warehousing Problem.” Institute of
Industrial Engineers Transactions 34 (7): 601–612. doi:10.1080/07408170208928897.
Kosfeld, M. 1998. “Warehouse Design through Dynamic Simulation.” Paper presented at the Winter Simulation Conference, Wash-
ington, DC, December 6–9. doi:10.1109/WSC.1998.745852.
Kulak, O. 2005. “A Decision Support System for Fuzzy Multi-Attribute Selection of Material Handling Equipments.” Expert Systems
with Applications 29 (2): 310–319. doi:10.1016/j.eswa.2005.04.004.
Lee, M. K. 1992. “A Storage Assignment Policy in a Man-Onboard Automated Storage/Retrieval System.” International Journal of
Production Research 30 (10): 2281–2292.
Lerher, T., M. Šraml, and I. Potrč. 2011. “Simulation Analysis of Mini-load Multi-shuttle Automated Storage and Retrieval Systems.”
The International Journal of Advanced Manufacturing Technology 54 (1–4): 337–348. doi:10.1007/s00170-010-2916-8.
Li, J., R. Huang, and J. B. Dai. 2017. “Joint Optimization of Order Batching and Picker Routing in the Online Retailer’s Warehouse
in China.” International Journal of Production Research 55 (2): 447–461. doi:10.1080/00207543.2016.1187313.
Macro, J., and R. Salmi. 2002. “Warehousing and Inventory Management: A Simulation Tool to Determine Warehouse Efficiencies
and Storage Allocations.” Paper presented at the Winter Simulation Conference, San Diego, December 8–11. doi: 10.1109/
WSC.2002.1166388.
Mueller, R. 2007. “Specification and Automatic Generation of Simulation Models with Applications in Semiconductor Manufactur-
ing.” PhD thesis. Georgia Institute of Technology.
Peixoto, R., L. Dias, M. S. Carvalho, G. Pereira, and C. A. Geraldes. 2016. “An Automated Warehouse Design Validation using Dis-
crete Simulation.” Paper presented at the Institute of Electrical and Electronics Engineers 19th International Conference on
Intelligent Transportation Systems. Rio de Janeiro, November 1-4. doi:10.1109/ITSC.2016.7795554.
Petersen, C. 2000. “An Evaluation of Order Picking Policies for Mail Order Companies.” Production and Operations Management 9
(4): 319–335.
Petersen, C. 2002. “Considerations in Order Picking Zone Configuration.” International Journal of Operations and Production Man-
agement 22 (7): 793–805. doi:10.1108/01443570210433553.
Petersen, C., and G. Aase. 2004. “A Comparison of Picking, Storage, and Routing Policies in Manual Order Picking.” International
Journal of Production Economics 92 (1): 11–19. doi:10.1016/j.ijpe.2003.09.006.
Petersen, C., and G. Aase. 2016. “Improving Order Picking Efficiency With the Use of Cross Aisles and Storage Policies.” Open
Journal of Business and Management 5 (1): 95–104. doi:10.4236/ojbm.2017.51009.
Pohl, L., R. Meller, and K. Gue. 2009. “An Analysis of Dual-command Operations in Common Warehouse Design.” Transportation
Research Part E: Logistics and Transportation Review 45 (3): 367–379. doi:10.1016/j.tre.2008.09.010.
Pohl, L., R. Meller, and K. Gue. 2011. “Turnover-based Storage in Non-Traditional Unit-load Warehouse Design.” Institute of Indus-
trial Engineers Transactions 43 (10): 703–720. doi:10.1080/0740817X.2010.549098.
Potrč, I., T. Lerher, J. Kramberger, and M. Šraml. 2004. “Simulation Model of Multi-shuttle Automated Storage and Retrieval Sys-
tems.” Journal of Materials Processing Technology 157–158: 236–244. doi:10.1016/j.jmatprotec.2004.09.036.
Randhawa, S. U., and R. Shroff. 1995. “Simulation-based Design Evaluation of Unit Load Automated Storage/Retrieval Systems.”
Computers and Industrial Engineering 28 (1): 71–79. doi:10.1016/0360-8352(94)00027-K.
Rao, S. S., and G. K. Adil. 2017. “Analytical Models for a New Turnover-based Hybrid Storage Policy in Unit-load Warehouses.”
International Journal of Production Research 55 (2): 327–346. doi:10.1080/00207543.2016.1158428.
Roodbergen, K., and R. Koster. 2001. “Routing Methods for Warehouses With Multiple Cross Aisles.” International Journal of Pro-
duction Research 39 (9): 1865–1883. doi:10.1080/00207540110028128.
Roodbergen, K. J., and I. F. Vis. 2006. “A Model for Warehouse Layout.” Institute of Industrial Engineers Transactions 38 (10):
799–811. doi:10.1080/07408170500494566.
Roodbergen, K. J., and I. F. Vis. 2009. “A Survey of Literature on Automated Storage and Retrieval Systems.” European Journal of
Operational Research 194 (2): 343–362. doi:10.1016/j.ejor.2008.01.038.
Roodbergen, K., I. Vis, and G. Taylor. 2015. “Simultaneous Determination of Warehouse Layout and Control Policies.” International
Journal of Production Research 53 (11): 3306–3326. doi:10.1080/00207543.2014.978029.
Rosenblatt, M. J., Y. Roll, and V. Zyser. 1993. “A Combined Optimization and Simulation Approach for Designing Automated Stor-
age/Retrieval Systems.” Institute of Industrial Engineers Transactions 25 (1): 40–50. doi:10.1080/07408179308964264.
Rosi, B., L. Grasic, G. Dukic, T. Opetuk, and T. Lerher. 2016. “Simulation-based Performance Analysis of Automated Single-tray
Vertical Lift Module.” International Journal of Simulation Modelling 15 (1): 97–108. doi:10.2507/IJSIMM15(1)8.328.
Rouwenhorst, B., B. Reuter, V. Stockrahm, G. van Houtum, R. Mantel, and W. Zijm. 2000. “Warehouse Design and Control: Frame-
work and Literature Review.” European Journal of Operational Research 122 (3): 515–533. doi:10.1016/S0377-2217(99)
00020-X.
Rushton, A., P. Croucher, and P. Baker. 2014. The Handbook of Logistics and Distribution Management: Understanding the Supply
Chain. 5th ed. London: Kogan Page.
 International Journal of Production Research 7121

Shqair, M., S. Altarazi, and S. Al-Shihabi. 2014. “A Statistical Study Employing Agent-based Modeling to Estimate the Effects of
Different Warehouse Parameters on the Distance Traveled in Warehouses.” Simulation Modelling Practice and Theory 49: 122–
135. doi:10.1016/j.simpat.2014.08.002.
Telek, P. 2013. “Equipment Preselection for Integrated Design of Materials Handling Systems.” Advanced Logistics Systems 7 (2):
57–66.
Tompkins, J. A., J. A. White, Y. A. Bozer, and J. M. Tanchoco. 2010. Facilities Planning. 4th ed. Hoboken, NJ: John Wiley & Sons.
Tuzkaya, G., B. Gülsün, C. Kahraman, and D. Özgen. 2010. “An Integrated Fuzzy Multi-criteria Decision Making Methodology for
Material Handling Equipment Selection Problem and an Application.” Expert Systems with Applications 37 (4): 2853–2863.
doi:10.1016/j.eswa.2009.09.004.
Zhou, M., and K. Venkatesh. 1999. Modeling, Simulation, and Control of Flexible Manufacturing Systems: A Petri Net Approach.
Vol. 6. World Scientific.
Copyright of International Journal of Production Research is the property of Taylor & Francis
Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv
without the copyright holder's express written permission. However, users may print,
download, or email articles for individual use.