Computers & Industrial Engineering 137 (2019) 106035

Contents lists available at ScienceDirect

Computers & Industrial Engineering

journal homepage: www.elsevier.com/locate/caie

Design and optimization of order picking systems: An integrated procedure T

and two case studies
Eleonora Bottani , Andrea Volpi, Roberto Montanari
⁎

Department of Engineering and Architecture, University of Parma, Viale delle Scienze 181/A, 43124 Parma, Italy

ARTICLE INFO ABSTRACT

Keywords: This paper outlines an integrated order picking system (OPS) design procedure for the optimization of manual
Warehouse management picking activities and shows its practical application by means of a software tool in two case studies. The pro-
Order picking systems posed OPS design procedure takes into account the factors typically suggested in literature for the optimization
Simulation software of the picking process, e.g. warehouse layout, item allocation policy, picking strategy and type, and routing. For
Design procedure
each factor, the various design options are listed and described, starting again from the available literature.
Optimization
Then, two real cases are presented to show how the proposed approach can be applied in practice by im-
plementing it in an ad hoc software tool designed in Microsoft Excel™. The application of the software tool
allowed to evaluate the current picking performance of both companies and to test alternative configurations of
their order picking process, in the attempt to improve it. In both cases, the framework proved to be effective to
redesign the picking process and allowed deriving practical suggestions for the optimization of the process.

1. Introduction several factors affect the travel distance and should, therefore, be taken
into account to optimize the picking process; these factors include:
Order picking, i.e. the process of selecting a set of items, retrieving
them from their storage locations and transporting them to a sorting/ 1. the structure of the warehouse in terms of size and layout (Parikh &
consolidation process for order fulfilment (Rouwenhorst et al., 2000), is Meller, 2010; Roodbergen & Vis, 2006);
recognized as one of the most expensive warehouse activities, because 2. the use of high-level vs. low-level picking (van Gils, Caris,
it is either very labour intensive or capital intensive (Frazelle, 2002). A Ramaekers, & Braekers, 2019);
manual picking process is estimated to account for more than 55% of 3. the operational strategy, e.g. order picking vs. batch picking vs. zone
the total cost of warehouse operations (Bottani, Montanari, Rinaldi, & picking (de Koster, Le-Duc, & Zaerpour, 2012; Le-Duc & de Koster,
Vignali, 2015; Tompkins et al., 1996). This is the main reason why 2007; Petersen, 2002; Van Nieuwenhuyse & de Koster, 2009);
researchers and logistics managers consider order picking as a pro- 4. the storage assignment policy (Bottani, Cecconi, Vignali, &
mising area for productivity improvement (de Koster, Le-Duc, & Montanari, 2012; Manzini, Accorsi, Gamberi, & Penazzi, 2015;
Roodbergen, 2007). Petersen & Schmenner, 1999);
In manual picker-to-parts order picking systems (OPSs), the cost of 5. the routing policy (De Santis, Montanari, Vignali, & Bottani, 2018;
picking is mainly due to the so called “travel time”, i.e. the time spent Petersen & Aase, 2004).
by pickers when travelling (unproductively) in the warehouse. This
quota accounts for approximately 50% of the total order picking time All the factors listed above contributes, to various extent, to the
(Tompkins et al., 1996). The travel time is an increasing function of the efficiency of OPSs. However, although the number of publications
travel distance the picker should cover to retrieve the items requested dealing with one specific aspect of OPSs is extensive, there is a very
in a customer’s order. Therefore, minimizing this distance is typically limited number of studies that evaluate more than one design factor at a
suggested as potential leverage for optimising the total picking time time (van Gils, Ramaekers, Braekers et al., 2018; van Gils et al., 2019).
(Hall, 1993; Jarvis & McDowell, 1991; Petersen & Aase, 2004; Petersen, Typically, researchers have focused on a specific topic to be analysed
1999; Roodbergen & de Koster, 2001). and optimised (de Koster et al., 2007; Bottani et al., 2012, 2015), while
Researchers (e.g. Gu, Goetschalckx, & McGinnis, 2007) agree that the remaining factors are fixed and kept unchanged during the

⁎
Corresponding author.
E-mail addresses: [email protected] (E. Bottani), [email protected] (A. Volpi), [email protected] (R. Montanari).

https://doi.org/10.1016/j.cie.2019.106035

Available online 07 September 2019

0360-8352/ © 2019 Elsevier Ltd. All rights reserved.
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

optimization. Nonetheless, evaluating the above factors simultaneously 2.2. Literature analysis and classification
is relevant, because of interactions among them (Davarzani & Norrman,
2015; Van Gils, Ramaekers, Caris et al., 2018). The relevant literature relating to picking systems was reviewed and
In line with the considerations above, this paper develops an in- classified according to the OPS design factors listed previously, high-
tegrated and practical framework for the design of OPSs. From a sci- lighting the interactions among the different OPS design factors. The
entific perspective, the proposed framework aims to allow almost all studies reviewed (listed and classified in Table 1) are discussed below.
the picking design factors to be taken into account (singularly but most
importantly, jointly) in the design of the OPS. Indeed, although the 2.2.1. OPS design factor 1 – warehouse layout
design factors themselves are not new, the proposed framework cate- The structure of the warehouse in terms of size and layout affects
gorizes the published literature according to these factors and embodies several aspects of warehouse performance such as material handling
them into the different phases of the OPS design, thus formalising the cost, space cost and storage capacity. To optimally design the picking
available knowledge about picking in a structured process. From a process, there are several aspects of the warehouse layout to take into
practical point of view, the framework provides the decision maker account. A key point is whether the warehouse has a conventional
(e.g. the warehouse manager) with an overview of the key decisions to (rectangular) layout with parallel aisles or a non-conventional one. In a
make at each step of the OPS design process, coupled with the relating conventional warehouse, picking items are located on parallel pick
design options and findings/approaches available in literature to sup- aisles, lying perpendicular to the front and back cross-aisles, and pos-
port these decisions. sibly cross-aisles, which allow the picker to change lane quickly.
Some procedures for the design of OPSs actually exists in literature Compared to non-conventional layouts, conventional ones are by far
(e.g., Yoon & Sharp, 1996; Dallari, Marchet, & Melacini, 2009; Manzini, more investigated in literature (41 out of 48 studies reviewed, 85.4%)
Gamberi, & Regattieri, 2005); however, these procedures have some and they reflect the scenarios typically taken into account for the ap-
limitations. First, they have been published some years ago and would plication of both heuristic routing policies (s-shape, return, aisle-by-
benefit from an update based on the most recent literature on picking. aisle or largest gap) and meta-heuristic algorithms for the minimization
Second, the existing procedures often focus on a specific type of OPS: of the travel distance. Indeed, 28 out of 48 studies reviewed (58.3%)
this is, for instance, the case for the approach of Manzini et al. (2005), propose routing algorithms in a conventional warehouse layout, while
who focused on the particular case of flexible OPSs. Third, the opera- only one study considers a non-conventional layout.
tional aspects (called “operating strategies” by Yoon and Sharp (1996) The presence of cross-aisles is a further relevant aspect of the
and Dallari et al. (2009)) are not very detailed, which could make it warehouse layout. Cross-aisles divide the warehouse into a number of
difficult to apply these procedures in practice. As a final point, although blocks; a sub-aisle is a section of an aisle within a block. Intuitively,
several design aspects are taken into account in the existing procedures, cross-aisles provide greater flexibility in the routing of order pickers
interactions among these factors are rarely evaluated. In developing the and therefore are expected to generate shorter travel distances; how-
framework proposed in this paper, instead, an attempt is made to ever, this effect may be nullified if the number of cross-aisles becomes
highlighted interdependences among the different OPS design factors. excessive, as the cross-aisles themselves must be traversed in order to
The reminder of the paper is organized as follows. The next section reach the required items (Bottani, Montanari, & Rinaldi, 2019; Vaughan
describes the methodological approach followed in this study to de- & Petersen, 1999). Moreover, the presence of narrow aisles is another
velop the framework for OPS design, provides details about the fra- important point, as congestions among pickers can originate with this
mework itself and illustrates its software implementation. In Section 3, layout configuration (Klodawski, Jachimowski, Jacyna-Golda, &
the framework is applied in two case studies, referring to as many Izdebski, 2018); this issue has been investigated in 3 out of 48 studies
warehouses of real companies, with the aim to optimize the picking reviewed (6.3%). Heuristics and exact methods are available for this
process. Section 4 concludes by summarizing the contribution of this warehouse layout (Pan & Wu, 2012; Chabot, Coelho, Renaud, & Côté,
work, discussing practical and scientific implications and suggesting 2018).
future research directions. The location of the depot and the warehouse shape factor are fur-
ther aspects that is worth taking into account when designing the
warehouse layout (Bottani et al., 2019; Petersen, 1997). Examples of
2. Methodology studies that determined the effect of the warehouse shape on the travel
distance have been by Hall (1993) and Francis (1967).
The research methodology followed in this study is depicted in
Fig. 1 and detailed in the subsections that follow. 2.2.2. OPS design factor 2 – picking type
From an operational perspective, picker-to-part systems can be
distinguished into low-level picking and high-level picking systems (de
2.1. OPS design factors Koster et al., 2007). In the former systems, the storage aisles have
single-level (ground level) storage racks and the employee picks the
Optimizing a manual picking process requires the careful con- requested items from these racks. To utilize the space efficiently, high-
sideration of operational planning problems (Yu & De Koster, 2009), level (man-onboard) storage systems, where all storage racks can be
structural aspects (e.g. the warehouse layout; Çelik & Süral, 2018) and used for picking, are often implemented in practice. In these systems,
interdependencies between these factors. In line with this considera- the picker travels to the picking locations on board of a lifting order-
tion, the first step of this study was to identify the relevant factors to be pick truck, that stops in front of the appropriate locations and waits for
taken into account in the design of an OPS. Following a deductive ap- the picker to perform a pick (Pan, Wu, & Chang, 2014). While low-level
proach, five key OPS design factors were derived from the literature picking systems have been widely investigated in literature (35 out of
and include: (1) the warehouse layout; (2) the choice of the picking 48 studies reviewed, 72.9%), the optimal design of high-level picking
type; (3) the choice of the appropriate picking strategy; (4) the item systems has been analysed in four studies only (8.3% of the studies
allocation logic; and (5) the routing policy. reviewed). Obviously, the travel distance changes significantly if high-
level picking is used instead of low-level picking: in this latter situation,

Fig. 1. Scheme of the research methodology.

2
 Table 1
Summary and classification of the studies reviewed.

Warehouse layout Picking type Routing Storage allocation policy Picking strategy
E. Bottani, et al.

Authors Rectangular Fishbone Presence of Depot Shape Narrow High-level Low-level Heuristic Meta- Exact Random Class- Correlated Order Batch Zone
cross-aisles location factor aisles heuristic based picking picking picking

Berglund and Batta (2012) x x x x x

Bottani et al. (2019) x x x x x x x
Caron et al. (1998) x x x x x
Çelk and Süral (2014) x x x x x
Chabot et al. (2018) x x x x x x x
Chackelson et al. (2013) x x x x x x
Cheng et al. (2015) x x x
de Koster and Van der Poort (1998) x x x x x x
de Koster et al. (2012) x x x
De Santis et al. (2018) x x x x x x
Francis (1967) x x
Gibson and Sharp (1992) x x
Hall (1993) x x x x x x x
Ho and Lin (2017) x x
Ho and Tseng (2006) x x x x
Hsieh and Tsai (2006) x x x x x
Hsu et al. (2005) x x x x
Hwang et al. (2003) x x
Jarvis and McDowell (1991) x x x
Klodawski et al. (2018) x x x x x x x x
Kuo et al. (2016) x
Le-Duc and De Koster (2005) x x x x x

3
Le-Duc and de Koster (2007) x x x
Lu et al. (2016) x x x x x x
Manzini et al. (2007) x x x x x x
Manzini et al. (2015) x x
Muppani and Adil (2008) x x x x
Pan and Wu (2012) x x x x x x x
Pan et al. (2014) x x x x x x
Parikh and Meller (2010) x x x x
Petersen and Aase (2004) x x x x x
Petersen and Schmenner (1999) x x x x x x x
Petersen (1997) x x x x x x x
Petersen (1999) x x x x x x
Pferschy and Schauer (2018) x x x
Rao and Adil (2013) x x x x x x
Ratliff and Rosenthal (1983) x x x x x
Roodbergen and de Koster (2001) x x x x x x
Roodbergen and Vis (2006) x x x x x x
Roodbergen et al. (2008) x x x x x x
Theys et al. (2010) x x x x x x x
Valle et al. (2017) x x x x x
van Gils, Ramaekers, Braekers et al. x x x x x x x x
(2018)
Vaughan and Petersen (1999) x x x x x
Xiao and Zheng (2010) x x x x
Xiao and Zheng (2012) x x x x
Yu et al. (2015) x x x
Zhang (2016) x x x x
Computers & Industrial Engineering 137 (2019) 106035
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

only horizontal picking efforts should be minimized, while in case of reserves a physical storage capacity to each class, in which items are
high-level picking, the vertical travel should be taken into account in located randomly. An increase in the picking performance is generally
the optimization of the travel distance. However, routing issues in high- observed when using class-based storage compared to random storage;
level picking systems have been investigated in only one of the studies however, a high number of classes could cause difficulties in its man-
reviewed. High-level order picking systems also require decisions on agement, as it limits the flexibility of the allocation system; hence, the
how to locate items on the vertical storage location: fast moving items number of classes usually ranges from 2 to 5 (see, e.g., Rao & Adil,
are typically placed at lower levels to reduce the traveling and blocking 2013). In class-based storage, items are generally assigned to a class
of order pickers (Pan et al., 2014). Two out of 48 studies reviewed have according to the cube per order index (COI) rule (Haskett, 1963), which
evaluated storage assignment policies in high-level picking systems. measure the number of accesses to an item in a given time period.
More precisely, Parikh and Meller (2010) provided an analytic model However, other classification criteria also exist, such as turnover based
that determines the optimal system configuration when considering storage classes (Van Gils, Ramaekers, Caris et al., 2018), popularity of
vertical travel. The model estimates the throughput of the warehouse the item over its life-cycle (Manzini et al., 2015), or density-turnover
assuming random storage. The key point of the model is to determine index (Hwang, Yong, & Cha, 2003). As far as the arrangement of the
the optimal storage system configuration, in terms of the number, classes in the warehouse is concerned, the possible approaches for low-
length, and height of the storage aisles. Pan et al. (2014) presented an level picking include (Yu, De Koster, & Guo, 2015): (1) within-aisle
evaluation method for high-level picker-to-part systems in the case of allocation, where items in a single pick aisle belong to the same storage
class-based storage and applying three different routing policies. class; (2) across-aisle allocation, where each class can be located across
Finally, a trade-off exists between the warehouse layout and the use several pick aisles; (3) diagonal allocation, where classes are located
of low-level vs. high-level picking, as high-level picking systems, by with respect to the distance to the depot; (4) perimeter allocation,
utilizing the vertical space, can typically reduce floor space needs where classes are located around the periphery of the warehouse. Si-
(Parikh & Meller, 2010). milar but slightly more articulated logics exist for high-level picking
systems (Pan et al., 2014). “Correlated” storage assignment policies
2.2.3. OPS design factor 3 – picking strategy (Frazelle & Sharp, 1989) instead, take into account the correlation
A particular order picking strategy is zone picking: using this among items in the allocation: items often picked together should be
method, items are divided into a series of different zones and pickers allocated close to each other, to reduce the travel time of pickers. The
are assigned to a zone and trained to pick within that zone, while they problem typically consists of two parts: the identification of clusters of
are not allowed to move to different zones (Ho & Lin, 2017). Two types items and the assignment of clusters to locations for storage (Zhang,
of zone picking strategies exist (Tompkins et al., 1996): sequential and 2016). Procedures for the implementation of correlated storage as-
synchronized. In sequential zone picking, picking is carried out in one signment policy were developed by Zhang (2016), Xiao & Zheng (2010,
zone and one order at a time, after which the order is passed to the next 2012), and Muppani and Adil (2008).
zone. In synchronized zone picking, instead, all items corresponding to Routing and storage allocation policies have been frequently eval-
batched orders are picked simultaneously from all the zones, and then uated in conjunction, and in a very limited number of cases, they were
orders are consolidated through a sorting system. A typical problem of found to be unrelated (Chackelson, Errasti, Ciprs, & Lahoz, 2013; Ho &
zone picking involves determining the optimal number of zones (de Tseng, 2006). Conversely, most of the available literature indicates a
Koster et al., 2012). Moreover, algorithms exist for the optimal im- statistically significant interaction between storage and routing plan-
plementation of zone picking strategies (Kuo, Kuo, Chen, & Zulvia, ning problems (Manzini, Gamberi, Persona, & Regattieri, 2007;
2016). Conversely, zone picking is rarely associated with other design Petersen & Schmenner, 1999; Theys, Bräysy, Dullaert, & Raa, 2010).
factors; the only exception is the study by the van Gils, Ramaekers, Caron, Marchet, and Perego (1998) evaluated and compared different
Braekers et al. (2018), who took into account storage assignment po- routing policies with class-based storage allocation by COI; they found
licies, zoning, routing and batching simultaneously. that this allocation used in conjunction with the S-shape routing policy
Batch picking is an alternative to order picking, where a picker generates better results than the basic return policy for a range of
collects all of the items for more orders. Two types of batch picking scenarios. Petersen (1999) studied the combined effects of routing po-
strategies exist: pick-and-sort and sort-while-pick. In pick-and-sort licies, pick list sizes and storage policies. Petersen and Aase (2004)
batch picking the pickers do not sort (while picking) the items into carried out a similar study for different combinations of order-batching
customer orders, while in sort-while-pick, the products are sorted by policies, storage policies and routing policies. Petersen and Schmenner
order during picking. Batch picking has been frequently investigated (1999) evaluated different patterns of class-based storage in combina-
together with routing problems or storage assignment policies, as tion with different routing policies, pick list sizes, and depot locations.
batching should obviously take into account item location rules to Le-Duc and de Koster (2007) evaluated the effect of warehouse shape
create effective batches (Ho & Tseng, 2006; Hsieh & Tsai, 2006; and storage policy on picking efficiency, assuming a warehouse with a
Petersen & Aase, 2004). Some algorithms are available to solve jointly single central cross-aisle, class-based storage policy and return routing
the batching and routing problem, including heuristics (Cheng, Chen, policy. Rao and Adil (2013) have determined the optimal class
Chen, & Yoo, 2015; Gibson & Sharp, 1992; Hsu, Chen, & Chen, 2005) boundaries and number of aisles for a low-level picker-to-parts system
and exact approaches (Pferschy & Schauer, 2018; Valle, Beasley, & da in a two-block rectangular warehouse, where pickers adopt a return
Cunha, 2017). policy. Warehouse layout and storage allocation policies also have in-
terdependences. Berglund and Batta (2012) have proposed an analytic
2.2.4. OPS design factor 4 – items allocation model to optimally placing the cross-aisle in a warehouse as a function
Storage assignment rules attempt to provide an effective way of of the storage assignment policy.
locating products in order to improve the efficiency of the picking
process. Random storage, where the location for each item can be se- 2.2.5. OPS design factor 5 – routing
lected randomly from all eligible empty locations, is widely used be- “Routing” order pickers means determining the routes of the pickers
cause it is simple, often requires less space than other storage methods, through the warehouse when retrieving picking items. Optimal routing
and results in a better utilization of all picking locations (Petersen & algorithms, where the order picking problem is approached as a special
Schmenner, 1999). However, research shows that alternative ap- case of the travelling salesman problem (TSP), are available for con-
proaches can significantly reduce the travel time (Jarvis & McDowell, ventional warehouse layouts with up to one cross-aisle. An exact al-
1991; Petersen & Schmenner, 1999). Class-based storage policies divide gorithm for solving the TSP exists for rectangular warehouses, which
all items into a number of different classes (typically from 2 to 5), and only have crossovers at the ends of the aisles (Ratliff & Rosenthal,

4
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

1983). de Koster and Van der Poort (1998) have proposed a polynomial For each decision, several options can be available. For instance,
algorithm for a 1-block warehouse, extending that created by Ratliff about the previous point (i), the decision maker has 3 main design
and Rosenthal (1983) to solve the routing problem for a non-central options, i.e.: (a) no cross-aisles; (b) 1 cross-aisle; (c) 2 or more cross-
location of the depot. Another extension of the Ratliff & Rosenthal’s aisles. As discussed previously, the warehouse layout has direct im-
algorithm exists for the dynamic order picking problem, in which the plications on the routing of pickers; hence, choosing one of these design
picker might receive instructions for a new picking mission where he/ options lead to different performance in terms of routing, which are
she is located in a random position in the warehouse (Lu, McFarlane, highlighted as suggestions and results available in literature in the last
Giannikas, & Zhang, 2016). part of Fig. 3. To be more precise, choosing option (a), i.e. a warehouse
Roodbergen and de Koster (2001) have extended Ratliff & Ro- with no cross-aisles, allows for the application of an exact algorithm for
senthal’s algorithm to 2-block warehouses, considering again a parallel- routing; conversely, option (c), i.e. a warehouse with 2 or more cross-
aisle rectangular warehouse, where order pickers can change corridor aisles, would force the decision maker to set a heuristic or metaheuristic
either at the end of the aisle or at the cross-aisle halfway along the aisle. algorithms for routing, because of the lack of exact algorithms. This
For non-conventional warehouses, the only example of exact algorithm implicitly means that optimizing (in literal sense) the routing of pickers
is that of Çelk and Süral (2014), which is valid for fishbone layout is not feasible with this layout option.
warehouses. Some authors have tried to minimize the travel distance of Once the warehouse layout has been defined, the subsequent step is
pickers by identifying the optimal layout of the warehouse (in terms of the choice of the picking type (Fig. 4); this is an operational decision
number of aisles and depot location), given the routing policy that will among two main options, i.e.:
be adopted, both for 1-block (Roodbergen & Vis, 2006) and multiple-
block warehouses (Roodbergen, Sharp, & Vis, 2008); this approach (i) low level picking; or
suggests that interactions exist between warehouse layout and routing (ii) high level picking.
policies when trying to minimize the travel distance.
Low-level picking reflects the situation most frequently investigated
2.3. OPS design framework in literature and most of the results available (in terms, for instance, of
routing or storage allocation) refer to this scenario. If instead the de-
The relevant literature analyzed in the previous section has been cision maker selects the high-level picking, he should apply dedicated
organized into a structured framework, whose scheme is proposed in procedures for routing and class assignment in the warehouse.
Fig. 2. The framework is intended as a practical tool that collects the The decision maker is now asked to select the picking strategy
findings from the literature about the OPS design factors listed pre- (Fig. 5), which means choosing among:
viously and organizes these findings in a guided procedure, useful to
help the decision maker in the design of OPSs. To this end, each step of (i) order picking;
the OPS design framework is further detailed in Figs. 3–7; on the basis (ii) batch picking;
of the literature reviewed, these figures list for each design step: (iii) zone picking.

1. the key decisions to make (main decisions); This choice is typically driven by economic considerations and can
2. the alternative design options for each decision or the solutions be supported by a cost model available in literature. Let us suppose that
typically adopted (main design options); the decision maker selects the batch picking strategy. Two further de-
3. the algorithms, results or suggestions available to effectively make sign options are available, namely: (a) pick-and-sort; and (b) sort-while-
each decision and the possible interactions with the remaining de- pick. If option (a) is selected, the decision maker could apply both
sign steps (key available results, solutions or algorithms). heuristic and exact approaches for the joint optimization of batching
and routing; alternatively, he could evaluate whether the chosen design
To clarify its practical usage, an example of OPS design process is option is more effective than zone picking.
proposed in the following; for brevity, the example is limited to one key The next step is to evaluate the item allocation policy (Fig. 6). The
decision per design step and to one design option, but similar con- first key decision to make involves choosing among:
siderations can be easily extended to the remaining decisions and de-
sign options. (i) random allocation;
The framework starts by distinguishing the cases of an existing (ii) class-based allocation; or
warehouse and of a new warehouse. “New warehouse” denotes the si- (iii) correlated allocation.
tuation where the warehouse either does not exist or has not yet started
its activities; in both cases, the OPS should be fully designed. “Existing Random storage does not require specific algorithms to be applied;
warehouse” denotes instead the situation of an active warehouse, where also, it is the typical scenario considered by most of the routing opti-
the picking process should be redesigned. The rationale for this dis- mization algorithms, which, consequently, can be directly used when
tinction is that, compared to a new warehouse, an existing warehouse implementing this storage allocation policy. If a class-based storage is
has typically a defined structure, so that probably less opportunities instead chosen, further decisions concern the number of classes, the
exist for optimizing the warehouse layout; in line with this considera- parameter used to group items into classes and the way classes are lo-
tion, for an existing warehouse the decision maker is asked to evaluate cated in the warehouse. About the number of classes, a suitable range,
whether any option for layout optimization is available. Looking at according to the literature, is to set from 2 to 5 classes. Dedicated
Fig. 3, when building a new warehouse, or whether there are oppor- studies have evaluated the impact of the chosen number of classes on
tunities for a warehouse redesign, a decision maker has to face the the performance of the pickers routing; relating results can drive the
following key design issues: decision maker in the choice of the most appropriate number of classes.
Moreover, in case cross-aisles have been included in the warehouse
(i) the possible inclusion of cross-aisles (and how many); layout, their optimal location can be determined as a function of the
(ii) the use of a conventional vs. non-conventional layout; number of classes.
(iii) the location of the input-output depot; The items allocation policy has strong interactions with the routing
(iv) the choice of the warehouse shape factor; and policy, which is the fifth step of the OPS design process. The key de-
(v) the use of wide vs. narrow aisles. cision for routing optimization concerns the approach to be used
(Fig. 7):

5
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

Fig. 2. Scheme of the procedure for OPS design and optimization.

(i) heuristic; and warehouse layout, or as a function of the storage allocation policy.
(ii) meta-heuristic; Related results are suggested as suitable findings to guide the decision
(iii) exact. maker in the choice of the heuristic routing strategy.
The performance of the chosen combination of OPS design factors
Sometimes the decision maker cannot properly choose among the should be possibly tested (by simulation or “in field”), to evaluate the
approaches listed, as the applicability of some routing approaches effectiveness of the picking system. If the performance is judged as
(exact approaches in particular) depends on the warehouse layout (cf. satisfactory, the OPS design process ends. Otherwise, the decision
Fig. 3). In case the decision maker chooses a heuristic approach for maker will have to reconsider his decisions and repeat the process.
routing, further aspects he must take into consideration include the
storage allocation policy and the length of the order picking list.
2.4. Software tool and simulation
Dedicated studies available in literature have evaluated the effect of
heuristic or optimal routing strategies as a function of the pick list size
The OPS design framework was implemented in an ad hoc software

6
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

Fig. 3. Key decisions and available solutions/algorithms for OPS design step 1.

tool designed in Microsoft Excel™, to automate some of the framework performance and test alternative combinations of OPS design factors.
steps in real applications and tested in two case studies (called Case Indeed, the use of simulation is particularly effective when:
study A and B – see Section 3). The basic structure of the software tool is
the same in both case studies, even if some aspects (e.g. the warehouse • the system under examination is complex or has a complex behavior
layout) are, obviously, case-specific. The choice of the software plat- (Harrison, Lin, Carroll, & Carley, 2007). An OPS, given the number
form was motivated by the fact that Microsoft Excel™ is one of the most of factors affecting its performance, can be right considered as a
powerful and widespread software packages for small and medium complex system, where more elements have to be taken into account
scale quantitative analyses (Hesse & Scerno, 2009) and is almost uni- in the design of the process and often the optimization of one ele-
versally used as a software package by companies. Moreover, Microsoft ment does not lead to the optimization of the whole system;
Excel™ is effective in modelling uncertainty of empirically derived data: • it is unfeasible, difficult, dangerous or time-expensive to change the
for instance, it can be used to generate random picking lists of variable physical structure of the system under examination (Harrison et al.,
length for simulation purpose. Also, it can be effectively used to input 2007). Again, this is the typical case of the OPS design.
formulae or constants values or to carry out simple statistical analyses
of the results (Farrance & Frenkel, 2014). A generic ready-to-use ver- Outcomes of the simulations were processed using the generalized
sion of the software tool is made available as supplementary material linear model (GLM) for repeated measures of Statistical Package for the
#1 of this paper. For the sake of confidentiality, relevant data (e.g., Social Science (SPSS) release 25 for Windows, to identify the significant
picking lists, items names, and so on) have been obfuscated. A video effects of the OPS design factors on the picking time (dependent vari-
tutorial, explaining the tool functioning, is also provided as supple- able) as well as the significant two-factor interactions (Montgomery &
mentary material #2. Runger, 2003).
Simulation was chosen as the methodological approach to re-
produce the current OPSs of the two case studies, evaluate the relating

Fig. 4. Key decisions and available solutions/algorithms for OPS design step 2.

7
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

Fig. 5. Key decisions and available solutions/algorithms for OPS design step 3.

3. Framework application and results lists (batch picking);

(iii) a manual low-level picking process is carried out by two employees
3.1. Case study A equipped with an electric fork lift truck;
(iv) during sorting activity, products are divided according to the four
3.1.1. Overview and AS IS scenario delivery areas, thereby preparing the loads for the various trucks;
Company A is a food retailer operating in the UK. Its product (v) the loading activity ends early in the morning and then products
portfolio consists mainly of sweet frozen products shipped through di- are delivered to the customers during the day.
rect deliveries. This distribution strategy requires an intense logistics
flow to provide the customers with a wide product catalogue and vo- The company’s warehouse includes 851 pallet locations at −35 °C.
lume in short delivery lead times, as the product should be shipped and The warehouse layout is unusual (Fig. 8): the warehouse has only one
received by the day after the order. To better serve them, customers cross-aisle at the front of the shelves and no cross-aisles at the end of the
have been grouped into four sets, according to geographical areas of rack, which, instead, ends in a blind wall. Because of this layout, most
London; each set is served by dedicated transport means (Bottani, of the traditional routing policies (e.g. S-shape or largest gap) are not
Panciroli, Montanari, & Volpi, 2018). applicable; rather, the picker is forced to follow a return policy for each
The current order fulfilment process is as follows: picking task.
Pallet locations are divided into three categories, i.e.: picking lo-
(i) the order cut-off time is set at 19:00 for deliveries in the next day; cations – used for picking activities; bulk locations – for storage and
(ii) the orders received are aggregated by creating common picking replenishment of picking locations; and general storage locations – for

Fig. 6. Key decisions and available solutions/algorithms for OPS design step 4.

8
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

Fig. 7. Key decisions and available solutions/algorithms for OPS design step 5.

delivery area of the customer and could be immediately loaded in a

truck for shipment. On the other hand, ABP would cause the items
ordered in more picking lists to be split, generating possible in-
efficiencies in term of the total picking time. It is therefore paramount
to carry out a detailed analysis of the two picking strategies to select the
most efficient one. As far as the storage assignment is concerned, class-
based storage is a possible way in which the storage locations can be
rearranged: this implies determining the number of classes, the criteria
for items grouping and the arrangement of classes in the warehouse
(Fig. 6). Overall, the following configurations were evaluated:

1. The number of classes was varied from 2 to 5, according to the

company’s evaluation and the scientific literature;
2. The inventory turnover index (ITI) and popularity (POP) were used
to form the classes. Other possible logics (e.g. COI) were neglected,
as locations all have the same volume and there is a strict one to one
correspondence between the number of locations in the warehouse
and the number of picking items;
Fig. 8. Warehouse configuration (Company A). 3. The arrangement of classes in the warehouse was performed using
the “within-aisle”, “across-aisle” and “diagonal” logics. Items are
other departments of the company delivering high volumes to big first distributed in the location at the ground floor (low-level) and
customers. Picking locations are obviously the most easily accessible then on the locations at higher storage levels (high-level), in line
storage locations (i.e. low-level locations); high-level racks are not used with the fact that low-level picking is expected to be always more
for picking because the number of locations at low-level is sufficient to efficient, in terms of time, than high-level picking.
store all picking items. As far as the storage assignment policy is con- 4. The two picking policies simulated are the current BPS and the new
cerned, a random policy is applied. ABP.

The combination of the above factors leads to 2 × 2 × 3 × 4 = 48

3.1.2. Definition of the TO BE scenarios TO BE configurations (Table 2).
Company A was interested in analysing the current picking pro-
cesses to identify alternative configurations that would optimize the
order fulfilment in terms of total time. From the above description and
looking at the scheme in Fig. 2, it is easy to see that the OPS design Table 2
factors that can be taken into account are the picking strategy and the Design factors and TO BE configurations for Company A.
storage assignment policy. Indeed, the AS IS situation is that of an ex-
Number of Classes arrangement Items grouping Picking
isting warehouse with defined layout, which has no potential to be classes criterion criterion strategy
modified significantly; moreover, the routing policy is forced. As far as
the picking strategy is concerned, the current approach is a batch 2 Within-Aisle Inventory ABP
picking followed by sorting (“batch picking & sorting”, BPS). A possible Turnover
3 Cross-Aisle Popularity BPS
alternative approach would be to group the items to pick on the basis of 4 Diagonal
the customer’s geographic area; this approach, referred to as “area 5
batch picking” (ABP), would avoid the downstream sorting activity, as
the products would already be sorted and aggregated according to the

9
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

Fig. 9. Structure of the simulation model for

Company A (second module).

3.1.3. Model development empirical measurements and the simulation outcomes over the two
The simulation tool developed in Microsoft Excel™ was customized sample weeks (Table 3) are very close to this value.
with the relevant data of Company A. The resulting simulation model
consists in two main modules. The first one defines the allocation of 3.1.4. Simulation results and analysis
items in the warehouse based on the chosen policy (random vs. class- The simulation model was launched to reproduce the 48 config-
based) and on the values set for the remaining parameters. The second urations (with N = 15 replicates per configuration) and evaluate their
module (Fig. 9) generates random picking lists on the basis of the his- main performance over a one-year time horizon. Fig. 10(a–d) graphi-
torical data, and simulates the picking process with the chosen settings, cally shows the results of the different scenarios. A first consideration
assuming the return policy for routing. It is worth remarking that his- from the simulation outcomes is that implementing the ABP policy
torical data does not show seasonality trend across the year, but without reorganizing the allocation of products in the warehouse would
nonetheless highlights different trend depending on the weekday. This lead to a relevant (−32%) decrease in the performance of the system.
is taken into account in the simulation to generate different orders as a Indeed, both the random allocation of items and the ABP have some
function of the weekday. Outputs of the model include the total time of inefficiencies, that cause an increase in the picking time which is not
the picking process per picking list, the picker route, the graphical re- compensated by the avoided sorting activity.
presentation of the item allocation and the amount of product picked. Looking at the specific design factors, the following considerations
The model was validated by taking two sample weeks of activity of emerge. In general, the performance of the picking system improves
the warehouse and measuring the total time of picking activities with the increase in the number of classes; the effect of this factor on
throughout these weeks. From the direct observation, it emerged that the picking time is statistically significant at p = 0.00. By observing the
some aspects could worsen the picking performance, causing a picking deviations from the overall average (Delta average TO BE), it can be seen
time higher than that resulting from the simple computation of the that setting the number of classes at 2 worsens by 12% the average
travel distance. One of these aspects refers to the possible unavailability performance of the system, while the configuration with 5 classes im-
of the picker or the electric fork lift truck during the day; 85% avail- proves up to 5% the average picking time. The improvement decreases
ability was set in the simulation model to take into account this aspect. progressively as the number of classes increases, becoming almost
Moreover, the picker could experience difficulties in properly placing negligible when moving from 4 to 5 classes (marginal benefit = 0.4%).
the cardboard boxes on the pallet as the number of boxes increases in From a practical point of view, this result suggests to set the number of
the pallet and also when he/she has to pick many (> 10) boxes of the classes at 4, which is the best compromise between the picking per-
same item, as these boxes need to be stabilized once placed in the pallet formance and the ease of management of the items allocation.
and before moving to the next picking location. This manual activity In terms of the criterion for classes arrangement in the warehouse,
was taken into account in the model by adding a fixed time for boxes the best configuration is the across-aisle shape. This result was expected
consolidation and by increasing (+15%) the picking time of items because of the imposed routing policy and is also consistent with si-
whose quantity is greater than 10. With these adjustments, the simu- milar findings in literature (e.g. van Gils, Ramaekers, Braekers et al.,
lation model was launched with the picking list processed in the sample 2018). However, in absolute terms, this configuration impacts on the
weeks and the total picking time was derived as output. The comparison average system’s performance by 0.4% only; moreover, the effect of
between the simulated and real outcomes (Table 3) showed good cor- classes arrangement on the picking time is not statistically significant
respondence (< 1% error); therefore, the model was considered as (p = 0.24).
validated. The effect of the criterion used to form the classes (POP vs. ITI) on
A further validation of the model came from the comparison of the the picking time it significant too (p = 0.00) and POP appears as the
outcomes with the contractual performance declared by the 3PL service best criterion to group items into classes. Conversely, the use of ITI
provider that manages the warehouse. This company indicated that the worsens the average system’s performance by approximately 4.2%. This
picking performance is approximately 175 boxes/hour. Both the result could be justified on the basis of the fact that items’ popularity,

Table 3
Performance of the picking process for Company A.

Week1 Week2

Productivity [cases/h] Picking time [min] Productivity [cases/h] Picking time [min]

Actual Simulated Actual Simulated Actual Simulated Actual Simulated

Monday 132.2 128.5 188.0 179.5 136.1 142.4 157.0 161.5

Tuesday 195.1 199.8 294.0 323.5 167.1 151.7 260.0 254.0
Wednesday 196.1 206.5 314.0 328.2 191.8 183.3 279.0 265.0
Thursday 222.0 210.8 355.0 342.5 194.9 202.0 279.2 294.0
Friday 220.3 231.3 341.0 333.1 201.7 206.6 276.2 263.0
Saturday 112.5 111.2 184.0 185.9 119.4 118.5 161.0 163.0

Average 179.7 181.4 279.3 282.1 168.5 174.2 235.4 233.4

10
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

Fig. 10. Picking performance deviation from the average solution as a function of the OPS design factors (a = number of classes; b = classes arrangement criterion;
c = items grouping criterion; d = picking strategy) for Company A.

calculated as the probability that each item will be picked at least once considerations. The top-4 scenarios (i.e. configurations 1, 2, 3 and 4) all
per day, is particularly suitable for batch picking in the context under show good performance. Moreover, the two configurations obtained
examination. using either 4 or 5 classes, arranged with the cross-aisle logic, with
Finally, as far as the picking policy is concerned, the results ob- items grouped using the popularity index and picked according to ABP
tained show that in general splitting the picking list by delivery zone policy return very similar results in terms of the picking time.
(ABP) is the preferred solution. This configuration leads to an average Compared to the optimal setting, the use of 4 classes causes a slight
3.9% saving in time compared to the BPS; the saving compared to the decrease in the system’s performance, accounting for 0.87% on average
average solution is 1.9%. Again, the effect of the picking policy on the and reaching a peak of approximately 2% on Saturday. On the other
performance of the picking system is significant at p = 0.00. The hand, setting 4 classes instead of 5 makes it easier to implement and
combined effects of batching type and number of classes and of update the class-based storage policy. In terms of productivity, this
batching type and criterion for items grouping are also significant at configuration allows 123.5 min/day to be saved in the picking process,
p = 0.01 and p = 0.00 respectively. which results in approximately 450 h/year saved. This amount of time
To sum up, all 48 combinations returned a lower total time than the could be dedicated to different warehouse activities or could justify a
AS IS scenario, proving the effectiveness of the warehouse reorganiza- shift in the cut-off time of the orders later during the day, which would
tion. To determine the best TO BE configuration, in terms of the saving contribute to enhance the service level delivered to customers. A fur-
in the picking time compared to the AS IS scenario, two further analyses ther relevant aspect of this configuration is that the ABP policy does not
were made. First, the best configuration was identified for each require sorting; hence, the warehouse area currently dedicated to this
weekday (Table 4), focusing on the top-8 configurations (i.e., scenarios process could be freed up and used for storing additional products.
with 3 or more classes, obtained using the popularity criterion and
using the ABP approach). Results show that ABP coupled with a class-
based storage with 5 classes formed according to the items’ popularity 3.2. Case study B
and located in the warehouse using the across-aisle layout returns the
best performance both in general and for each weekday simulated. 3.2.1. Overview and AS IS scenario
Second, the percentage difference, in terms of the picking time, be- Company B operates since 1947 in Parma (Italy) as a manufacturer
tween the optimal configuration and the AS IS scenario, was evaluated of high-pressure homogenizers and plunger pumps for dairy, food and
for each day of the week. Results, shown again in Table 4, highlight beverage, biotechnology, pharmaceutical, cosmetic, and chemical
that, although the scenarios with 5 classes return the best performance, markets. Homogenizers are manufactured in different assembly lines
the result obtained using ABP with 4 classes formed according to the where operators (one in each station) assemble components brought
items’ popularity and located in the warehouse using the across-aisle from the main company’s warehouse after picking.
layout is very close to the optimal one (0.87% difference). Moreover, The warehouse stores all the components needed to assemble the
this configuration would be easier to implement and manage in prac- whole assortment of machines; according to the picking lists (about 8
tice. per machine), the operators prepare the assembly kits for sub-func-
tional groups (e.g., compression head, mechanical transmission, etc.)
and bring the kits to the station in which they need to be assembled on
3.1.5. Discussion the machine. Depending on the specific part or component, the main
The results presented above lead to the following practical warehouse can be fed by purchase orders (for standardized parts,

11
 E. Bottani, et al.

Table 4
Ranking of the top-8 results of the TO BE scenarios across the weekdays and percentage difference compared to the optimal configuration for Company A.

Scenarios Ranking of the scenarios Average saving

in picking time
[min]

Combination ID Picking Items Number of Classes Average Monday Tuesday Wednesday Thursday Friday Saturday Average day
strategy grouping classes arrangement day
criterion criterion

1 ABP POP 5 Across-aisle 1 1 1 1 1 1 1 125.4

2 ABP POP 5 Diagonal 2 2 4 3 3 3 2 124.3
3 ABP POP 4 Diagonal 3 3 3 2 4 2 3 124.2
4 ABP POP 4 Across-aisle 4 4 2 4 2 4 5 123.5
5 ABP POP 5 Within-aisle 5 6 7 5 6 6 4 118.3
6 ABP POP 4 Within-aisle 6 7 8 6 5 5 6 118.2

12
7 ABP POP 3 Diagonal 7 5 5 7 8 7 7 116.9
8 ABP POP 3 Across-aisle 8 8 6 8 7 8 8 116.3

Scenarios Average saving in picking time [min] Percentage difference with the optimal configuration

Combination ID Monday Tuesday Wednesday Thursday Friday Saturday Average day Monday Tuesday Wednesday Thursday Friday Saturday

1 97.7 132.7 154.6 136.9 139.6 91.1 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% 0.00%
2 97.2 132.1 152.2 134.7 138.9 91.0 0.55% 0.63% 0.16% 0.92% 1.34% 0.14% 0.12%
3 96.4 132.3 152.6 134.0 139.3 90.9 0.50% 0.27% 0.29% 1.12% 1.01% 0.30% 0.05%
4 96.3 132.7 150.8 136.3 138.3 86.8 0.87% 0.64% 0.01% 1.77% 0.27% 0.57% 1.97%
5 92.5 126.5 145.9 126.6 131.5 87.0 3.26% 2.38% 2.83% 4.02% 4.72% 3.69% 1.91%
6 91.2 126.4 145.8 127.4 132.0 86.7 3.30% 3.01% 2.89% 4.04% 4.38% 3.48% 2.02%
7 93.9 129.3 142.6 120.0 130.0 85.7 3.91% 1.76% 1.57% 5.51% 7.76% 4.37% 2.51%
8 90.8 126.8 141.4 125.0 128.5 85.0 4.21% 3.16% 2.70% 6.09% 5.45% 5.10% 2.79%
Computers & Industrial Engineering 137 (2019) 106035
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

Fig. 11. Warehouse configuration (Company B).

purchased to order) or production orders (for customized parts, most of the fork lift truck will be used by two dedicated employees, which
which are standardized and thus kept in stock with some make-to-order should avoid bottlenecks due to waiting times for the availability of the
exceptions). The warehouse has an area of 1.500 square meters and is lifting equipment. Nonetheless, employees will have to manage com-
directly connected to the factory; the overall height is about 8 m, with ponents belonging to different kinds of picking lists (i.e. assembly line
7 m useful for storage activities. Three automatic doors and a fire door and spare parts), which could be stored as different levels in the
have been installed respectively on the main side of the warehouse and warehouse; a precise disposition and assignment of the items in the
on the short side, facing the production factory. A pedestrian tunnel has warehouse is therefore necessary.
been built at ground level in the warehouse for safety reasons, to allow Looking at the OPS design framework, the AS IS situation reflects
the warehouse employees to quickly reach the fire door. This tunnel is again the case of an existing warehouse with defined layout. Because
useful also for manual picking activities, being used as a cross-aisle to the routing is already effective and the picking strategy is difficult to
generate a tour which combines the S-shape and return policies de- change because the picking lists are directly generated by the com-
pending on to the items to be picked. Conversely, fork lift trucks cannot pany’s information system, the key design factor that can be leverage
cross the tunnel because of the limited width (Ponzi Artusi, 2017). for the optimization of the process is the storage allocation. The fol-
The storage and picking area consists of single-depth shelves, whose lowing configurations were therefore evaluated (Table 5):
load capacity differ depending on the items stored (heavy vs. light).
Heavy items are stored on pallets in the green area of Fig. 11; light 1. Class-based storage with 3 classes, reflecting the setting currently
items are instead stored in containers in the yellow area of Fig. 11. This used by Company B, was evaluated in conjunction with:
latter forms the focus of the study and consists of 3796 storage loca- 1.1. inventory turnover index (ITI) and popularity (POP) as criteria
tions. The light items area was selected for the analysis because picking to form the classes;
activities carried out on light items cover more than 95% of the total 1.2. “within-aisle”, “across-aisle” and “diagonal” logics for locating
picks of the warehouse. As the warehouse is a new building that has the classes in the warehouse.
been recently purchased by Company B, at the time of writing it is not The number of classes of this policy was kept unchanged
fully saturated: only 2278 picking locations of this area are occupied by compared to the AS IS scenario. Indeed, this number was de-
items. Two level of storage (ground level and first level) are used for termined by the warehouse manager in the attempt to increase
picking activities, which therefore consists in a combination of low- the flexibility of the allocation system and make it easy to
level and high-level picking. An order-pick truck is used for high-level implement and update this storage policy;
picking; however, the same equipment is also used for storage activ- 2. Correlated storage was considered as an alternative storage assign-
ities, and is therefore, not always available for picking. Hence, picking ment policy. In this logic, items are grouped according to the bill of
and storage tasks should be properly scheduled to avoid bottlenecks materials (BOM) they belong to. Moreover, BOMs belonging to the
and inefficiencies. same kit are located as close as possible to each other in the same
The AS IS picking process follows an order picking strategy where aisle. For items belonging to more than one BOM, the BOM with
picking lists are generated by the ERP system; a picking list usually highest quantities was selected to determine the item’s membership.
contains parts of a sub-functional group of the machine or a spare parts
order. Picking activities involves seven employees overall; four of them The combinations of the above design factors lead to 3 × 2 + 1 = 7
pick up the components for the assembly line of machines while the TO BE scenarios.
remaining ones are specialized in spare parts picking. In terms of sto-
rage assignment, the logic currently adopted by the company is the 3.2.3. Model development
class-based storage with 3 classes, determined using the popularity The simulation tool was customized to reproduce the OPS of
criterion and located in the warehouse according to the diagonal logic.
Table 5
3.2.2. Definition of the TO BE scenarios Design factors and TO BE configurations for Company B.
Company B was interested in test different item allocation policies
Item allocation policy Classes arrangement Items grouping criterion
and evaluate their impact on the picking process, both for low-level and criterion
high-level picking. As a first point, the tasks of the order pickers were
revised as follows: (i) one operator carries out high-level picking only, Class-based (3 classes) Within-aisle Inventory Turnover
by means of the order-pick truck; (ii) one operator carries out low-level Across-aisle Popularity
Diagonal
picking of heavy items using the fork lift truck; (iii) five operators pick
Correlated (BOM) Within-aisle –
up the machine components and spare parts at low-level, placing them
in a trolley. With this new work organization, the order-pick truck and

13
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

Company B. To this end, the following constraints were introduced: which increases the total process time.
The best criterion for grouping items in classes is POP, which shows
• each item is assigned to a single location and can be stored either in a statistically significant effect on the picking time (p = 0.00). This
a container (light item) or in a pallet (heavy item), depending on the result is probably due to the fact that popularity, computed as the
product. Light item and heavy items should be located in the re- probability that an item will be picked at least once per day, allows to
spective warehouse areas, according to the previous description; assign low-level storage locations to frequently used items, no matter
• the threshold between low-level and high-level picking is set at 2 m; their presence in each picking list. On the contrary, with the ITI logic,
• none of the equipment can cross the pedestrian tunnel. the most easy-to-reach locations are assigned to items having a high
consumption coupled with a low average stock level; this does not
The resulting model consists again of two modules. In the first one, necessarily identify the most requested items in absolute terms.
the key parameters of the picking process (i.e., class arrangement and For a direct comparison of the scenarios simulated, Table 6 shows
items’ grouping criteria) are set. According to the grouping criterion the average total time per picking tour and the average travel distance
chosen (ITI or POP), the model determines the class of each item. As the for all the TO BE configurations. Looking at the picking time, the most
number of classes is fixed, a typical ABC rule (80–95–100) was im- effective configuration is obtained using the correlated storage; in terms
plemented in the simulation model to distribute the items in the classes. of travel distance, instead, the use of class-base storage with diagonal
Then, the simulator assigns a location to each item, according to its arrangement and popularity index generates the best result.
storage medium (box or pallet) and class. On the basis of these findings, a more detailed investigation was
The second module generates random picking lists, in which each made for configurations 4, 5, 6 (where the popularity criterion is used)
item is associated with the following data: (i) storage location number; and 7 (correlated storage). For these configurations, the percentage of
(ii) aisle, shelf and level of the storage location; (iii) (x, y and z) co- picks that required the order-pick truck was evaluated, as this equip-
ordinates of the location with respect to the I/O depot. Because the OPS ment is a shared resource and compatibly with the need for high-level
consists in a combination of low-level and high-level picking, the si- picks, its usage should be limited as much as possible. Results, shown in
mulation model splits each picking list into two sub-lists, one for low- Fig. 13, highlight that most of the picking tours are carried out
level picking and one for high-level picking, using the z-coordinate of manually, while approx. 20% makes use of the order-pick truck. This is
the item location. The routing is determined applying the mixed S- an interesting outcome, as the order-pick truck is used also for storage
shape and return policy previously mentioned, separately for low-level and cannot be always available for picking. Taking into account not
and high-level picks. To optimize the process, it is paramount to de- only the picking time but also the resource scarcity, the most effective
termine the best way to assign storage locations to classes. To be more configuration is class-based storage with diagonal arrangement coupled
precise, it should be checked whether it is more useful to locate items of with POP index. This is consonant with the fact that this is the setting
class A saturating ground level and first level or to locate the same items currently implemented by Company B.
only in the shelves close to the I/O, regardless of the storage level. This
check is made by the simulation model by comparing the time required 3.2.5. Discussion
for the picker to retrieve an item located at ground floor in the storage For Company B, the top-4 scenarios are configurations 4, 5 and 6
location farthest from the I/O and that required the same item placed in which all make use of the POP criterion and configuration 7, which
the first storage location close to the I/O. This module also carries out adopts the correlated storage (BOM) policy. The best configuration
the simulation of the picking process on the basis of the configuration overall in terms of picking time is obtained setting correlated storage,
obtained and returns the relevant KPIs. with items arranged in the warehouse taking into account the BOMs.
However, the class-based storage (already used by Company B) with
3.2.4. Simulation results and analysis diagonal arrangement of classes obtained using POP as criterion returns
The simulation model was launched to reproduce the TO BE con- very similar results in terms of total time (+1.17%) with a shorter
figurations (again N = 15 replicates) and evaluate their main perfor- travel distance. This configuration also minimizes the usage of the
mance indexes over a one-year time horizon. For each scenario the si- order-pick truck in terms of time. The use of this logic could allow
mulation computes the total picking time and the total picking distance, saving approximately 2–3 h of picking for each machine manufactured
by adding up the contribution of the manual activities (low-level) and by Company B; assuming an annual production of 700 machines, this
of the order-pick truck (high-level). The structure of the TO BE con- would lead to 1750 h/year saved compared to the AS IS scenario.
figurations (cf. Table 5) as it stands does not allow for a full effect A further important point is that the results obtained were not
analysis, as the correlated storage (BOM) is not obtained as combina- known to Company B: indeed, the total picking time was known as-a-
tion of OPS design factors and therefore effects cannot be evaluated. whole for the AS IS scenario, without distinguishing, e.g., between low-
Effect analysis is therefore limited to the six configurations resulting level or high-level picking activities.
from the combination of the criteria for items grouping and for classes
arrangement. 4. Conclusions and future research directions
The results obtained are shown in Fig. 12. As far as the storage al-
location criterion is concerned, the best results are obtained setting the This paper has proposed a structured framework for the design of
correlated logic (2770 s per picking tour). This was somehow expected OPSs, by organizing the relevant literature into a five-step operational
because of the order picking method chosen by the company: indeed, procedure, encompassing the layout design, the choice of the picking
with this logic, the picker finds all the parts belonging to the same kit strategy and type, the item allocation and the routing optimization. For
very close to each other. Conversely, the worst configuration makes use each step, the key decisions to make have been presented, together with
of class-based storage, arranged using the diagonal criterion (2880 s per the relevant options and the findings, approaches and suggestions
picking tour). This result was not fully expected, because BOMs form available in literature. The framework has a practical orientation, as it
only a quota of the picking lists, which also include different items to has been designed to guide the warehouse manager in making effective
pick (e.g. spare parts). The correlated storage also returns the minimum decisions having a clear overview of the possible solutions. As it has
distance; interestingly, the diagonal configuration performs well in been derived from a comprehensive review of the literature, it is rea-
terms of travel distance, but generates high picking time (although the sonable to expect that the proposed framework covers all the key as-
effect is not statistically significant). This is probably due to the fact pects of OPSs design and optimization. As such, it also contributes to
that the diagonal logic, by exploiting the first storage level to store the scientific knowledge by proposing an overview of the up-to-date
picking items, involves a more frequent use of the order-pick truck, findings relating to picking optimization. This overview could be a

14
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

(a)

(b)

Fig. 12. Picking performance as a function of the different design factors (a = storage assignment policy; b = items grouping criterion) for Company B.

Table 6
Comparison of the results of the TO BE scenarios for Company B.

Configuration ID Items grouping criterion Item allocation policy Classes arrangement criterion Average total time [s] Average total distance [m]

1 ITI Class-based Across-aisle 3078 1783

2 ITI Class-based Within-aisle 2991 1718
3 ITI Class-based Diagonal 3102 1733
4 POP Class-based Across-aisle 2650 1332
5 POP Class-based Within-aisle 2660 1377
6 POP Class-based Diagonal 2679 1325
7 – Correlated – 2648 1372

RESOURCES USAGE
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
DISTANCE DISTANCE TIME TIME
(OPERATOR) (ORDER PICKER) (OPERATOR) (ORDER PICKER)

Correlated POP + POP + POP +

Fig. 13. Travel composition for TO BE scenarios 4, 5, 6 and 7 for Company B.

valuable basis for a more comprehensive review and classification of optimization process. On the contrary, picking design is a complex task,
the literature about picking. where several factors should be taken into account and in which a
Moreover, compared to the available procedures for OPS design, the perfect design of an individual element does not ensure the perfect
proposed approach takes a different perspective, by emphasizing the design of the whole system.
interdependencies between the key design factors, e.g. layout design From a practical perspective, the OPS design framework has re-
and routing or item allocation and routing. This is an important point, presented the basis for the optimization of the OPS in two real ware-
as the typical approach followed by researchers is to evaluate only one houses, referring to a UK company operating in the food wholesale
picking design factor at a time, keeping the remaining factors un- (Company A) and to an Italian company manufacturing homogenizers
changed; hence, interactions are rarely taken into account in the (Company B). The analysis of Company A appears as particularly

15
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

interesting because of the unusual warehouse layout (ending with a Production Research. https://doi.org/10.1080/00207543.2018.1489154.
blind wall), which does not allow implementing most of the traditional Çelk, M., & Süral, H. (2014). Order picking under random and turnover-based storage
policies in fishbone aisle warehouses. IIE Transactions, 46(3), 283–300.
routing policies; also, this layout has never been investigated in the Chabot, T., Coelho, L. C., Renaud, J., & Côté, J. (2018). Mathematical model, heuristics
scientific literature. Nonetheless, a wide number of alternative config- and exact method for order picking in narrow aisles. Journal of the Operational
urations for the OPS were evaluated. Company B is instead interesting Research Society, 69(8), 1242–1253. https://doi.org/10.1080/01605682.2017.
1390532.
because of the presence of a mixed high-level/low-level picking Chackelson, C., Errasti, A., Ciprs, D., & Lahoz, F. (2013). Evaluating order picking per-
strategy and because high-level picking requires a special equipment, formance trade-offs by configuring main operating strategies in a retail distributor: A
which is also used for storage and could not be always available for Design of Experiments approach. International Journal of Production Research, 51(20),
6097–6109.
picking. This situation could be frequent in practice, but again, it is Cheng, C.-Y., Chen, Y.-Y., Chen, T.-L., & Yoo, J.-W. J. (2015). Using a hybrid approach
typically neglected in the scientific literature. based on the particle swarm optimization and ant colony optimization to solve a joint
In both cases, the framework was implemented in a software tool order batching and picker routing problem. International Journal of Production
Economics, 170(1), 805–814.
developed in Microsoft Excel™ to evaluate, by means of simulation,
Dallari, F., Marchet, G., & Melacini, M. (2009). Design of order picking system.
different solutions for OPS optimization, ranging from item allocation International Journal of Advanced Manufacturing Technology, 42, 1–12.
strategies to picking strategies. The implementation of the framework Davarzani, H., & Norrman, A. (2015). Toward a relevant agenda for warehousing re-
in Microsoft Excel™ is original as well, as to the best of the authors’ search: Literature review and practitioners’ input. Logistics Research, 8(1), 1–18.
de Koster, R., Le-Duc, T., & Roodbergen, J. (2007). Design and control of warehouse order
knowledge, there are no available simulation tools for the analysis of picking: A literature review. European Journal of Operational Research, 182(2),
the picking process. In addition, for both companies the analyses made 481–501.
with the simulator were very useful to evaluate alternative configura- de Koster, R., Le-Duc, T., & Zaerpour, N. (2012). Determining the number of zones in a
pick-and-sort order picking system. International Journal of Production Research,
tions of the picking process and derive an estimate of the resulting 50(3), 757–771.
picking time and highlighted many potentials for improvement of the de Koster, R., & Van der Poort, E. (1998). Routing orderpickers in a warehouse: A com-
picking process. From the outcomes obtained, suggestions were elabo- parison between optimal and heuristic solutions. IIE Transactions, 30, 469–480.
De Santis, R., Montanari, R., Vignali, G., & Bottani, E. (2018). An adapted ant colony
rated for the companies under examination, to improve their current optimization algorithm for the minimization of the travel distance of pickers in
picking process. The use of Microsoft Excel™ to implement the frame- manual warehouses. European Journal of Operational Research, 267, 120–137.
work and carry out the analysis, besides ensuring a good level of au- Farrance, I., & Frenkel, R. (2014). Uncertainty in measurement: A review of Monte Carlo
simulation using Microsoft Excel for the calculation of uncertainties through func-
tomation, is expected to encourage the use of the framework by the tional relationships, including uncertainties in empirically derived constants. The
warehouse managers according to the evolution of the product portfolio Clinical Biochemist Reviews, 35(1), 37–61.
and orders composition. Francis, R. L. (1967). On some problems of rectangular warehouse design and layout. The
Journal of Industrial Engineering, 18, 595–604.
Starting from this work, future research activities could focus on the
Frazelle, E. (2002). World-class warehousing and material handling. New York: McGraw
following points. Hill.
Frazelle, E. A., & Sharp, G. P. (1989). Correlated assignment strategy can improve any

• A structured and systematic review of the literature about picking order-picking operation. Industrial Engineering, 21(4), 33–37.
Gibson, D. R., & Sharp, G. P. (1992). Order batching procedures. European Journal of
could be carried out following the key design aspects of OPSs dis- Operational Research, 58(1), 57–67.
cussed in this paper; Gu, J., Goetschalckx, M., & McGinnis, L. (2007). Research on warehouse operation: A

• A software package for the optimization of OPSs could be devel- comprehensive review. European Journal of Operational Research, 177, 1–21.
Hall, R. (1993). Distance approximations for routing manual pickers in a warehouse. IIE
oped. This could be perfected starting from the Microsoft Excel™ Transactions, 25(4), 76–87.
model developed in this study and made available to the scientific Harrison, J. R., Lin, Z., Carroll, G. R., & Carley, K. M. (2007). Simulation modeling in
community; organizational and management research. Academy of Management Review, 32(4),

• Additional case studies, with different warehouse configurations

1229–1245.
Haskett, J. L. (1963). Cube-per-order index—a key to warehouse stock location.
(e.g. fishbone layout), could be evaluated, in the attempt to gen- Transportation and Distribution Management, 3, 27–31.
eralize the results obtained in this study or to investigate whether Hesse, R., & Scerno, D. H. (2009). How electronic spreadsheets changed the world.
Interfaces, 39(2), 159–167.
these results could change depending on the warehouse layout.
Ho, Y.-C., & Lin, J.-W. (2017). Improving order-picking performance by converting a
sequential zone-picking line into a zone-picking network. Computers & Industrial
Appendix A. Supplementary material Engineering, 113, 241–255.
Ho, Y.-C., & Tseng, Y.-Y. (2006). A study on order-batching methods of order-picking in a
distribution centre with two cross-aisles. International Journal of Production Research,
Supplementary data to this article can be found online at https:// 44(17), 3391–3417.
doi.org/10.1016/j.cie.2019.106035. Hsieh, L.-F., & Tsai, L. (2006). The optimum design of a warehouse system on order
picking efficiency. International Journal of Advanced Manufacturing Technology,
28(5–6), 626–637.
References Hsu, C. M., Chen, K. Y., & Chen, M. C. (2005). Batching orders in warehouses by mini-
mizing travel distance with genetic algorithms. Computers in Industry, 56(2),
Berglund, P., & Batta, R. (2012). Optimal placement of warehouse cross-aisles in a picker- 169–178.
to-part warehouse with class-based storage. IIE Transactions, 44, 107–120. Hwang, H., Yong, H. O., & Cha, C. N. (2003). A stock location rule for a low level picker-
Bottani, E., Cecconi, M., Vignali, G., & Montanari, R. (2012). Optimisation of storage to-part system. Engineering Optimization, 35(3), 285–295.
allocation in order picking operations through a genetic algorithm’. International Jarvis, J., & McDowell, E. (1991). Optimal product layout in an order picking warehouse.
Journal of Logistics: Research and Applications, 15(2), 127–146. IIE Transactions, 23(1), 93–102.
Bottani, E., Panciroli, A., Montanari, R., & Volpi, A. (2018). Optimization of order picking Klodawski, M., Jachimowski, R., Jacyna-Golda, I., & Izdebski, M. (2018). Simulation
activities in a wholesale food company. In M. Affenzeller, A. G. Bruzzone, E. Jiménez, analysis of order picking efficiency with congestion situations. International Journal of
F. Longo, Y. Merkuriev, & M. A. Piera (Eds.). Proceedings of the 30th European mod- Simulation Modelling, 17(3), 431–443.
elling and simulation symposium (EMSS 2018), ISBN 978-88-85741-06-5 (paper), ISBN Kuo, R. J., Kuo, P. H., Chen, Y. R., & Zulvia, F. E. (2016). Application of metaheuristics-
978-88-85741-03-4 (pdf), September 17–19, Budapest (Hungary). (pp. 129–136). . based clustering algorithm to item assignment in a synchronized zone order picking
Bottani, E., Montanari, R., & Rinaldi, M. (2019). Development and testing of software tool system. Applied Soft Computing, 46, 143–150.
for warehouse design and picking optimization. International Journal of Management Le-Duc, T., & De Koster, R. B. M. (2005). Travel distance estimation and storage zone
and Decision Making, 18(2), 119–150. optimization in a 2-block class-based storage strategy warehouse. International
Bottani, E., Montanari, R., Rinaldi, M., & Vignali, G. (2015). Intelligent algorithms for Journal of Production Research, 43(17), 3561–3581.
warehouse management. In C. Kahraman, & S. Çevik Onar (Eds.). Intelligent techniques Le-Duc, T., & de Koster, R. (2007). Travel time estimation and order batching in a 2-block
in engineering management: Theory and applications (pp. 645–667). Switzerland: warehouse. European Journal of Operational Research, 176(1), 374–388.
Springer International Publishing. Lu, W., McFarlane, D., Giannikas, V., & Zhang, Q. (2016). An algorithm for dynamic
Caron, F., Marchet, G., & Perego, A. (1998). Routing policies and COI-based storage order-picking in warehouse operations. European Journal of Operational Research,
policies in picker-to-part systems. International Journal of Production Research, 36, 248(1), 107–122.
713–732. Manzini, R., Accorsi, R., Gamberi, M., & Penazzi, S. (2015). Modeling class-based storage
Çelik, M., & Süral, H. (2018). Order picking in parallel-aisle warehouses with multiple assignment over life cycle picking patterns. International Journal of Production
blocks: Complexity and a graph theory-based heuristic. International Journal of Economics, 170, 790–800.

16
E. Bottani, et al. Computers & Industrial Engineering 137 (2019) 106035

Manzini, R., Gamberi, M., Persona, A., & Regattieri, A. (2007). Design of a class based 799–811.
storage picker to product order picking system. International Journal of Advanced Rouwenhorst, B., Reuter, B., Stockrahm, V., van Houtum, G., Mantel, R., & Zijm, W.
Manufacturing Technology, 32(7–8), 811–821. (2000). Warehouse design and control: Framework and literature review. European
Manzini, R., Gamberi, M., & Regattieri, A. (2005). Design and control of a flexible order- Journal of Operational Research, 122, 515–533.
picking system (FOPS). A new integrated approach to the implementation of an ex- Theys, C., Bräysy, O., Dullaert, W., & Raa, B. (2010). Using a TSP heuristic for routing
pert system. Journal of Manufacturing Technology Management, 16(1), 18–35. order pickers in warehouses. European Journal of Operational Research, 200(3),
Montgomery, D. C., & Runger, G. C. (2003). Applied statistics and probability for engineers 755–763.
(3rd ed.). New Jersey: John Wiley & Sons Inc. Tompkins, J., White, J., Bozer, Y., Frazelle, E., Tanchoco, J., & Trevin, J. (1996). Facilities
Muppani, V. R., & Adil, G. K. (2008). Efficient formation of storage classes for warehouse planning. New York: Wiley.
storage location assignment: A simulated annealing approach. Omega, 36, 609–618. Valle, C. A., Beasley, J. E., & da Cunha, A. S. (2017). Optimally solving the joint order
Pan, J. C.-H., & Wu, M.-H. (2012). Throughput analysis for order picking system with batching and picker routing problem. European Journal of Operational Research, 262,
multiple pickers and aisle congestion considerations. Computers and Operations 817–834.
Research, 39(7), 1661–1672. van Gils, T., Caris, A., Ramaekers, K., & Braekers, K. (2019). Formulating and solving the
Pan, J. C.-H., Wu, M.-H., & Chang, W.-L. (2014). A travel time estimation model for a integrated batching, routing, and picker scheduling problem in a real-life spare parts
high-level picker-to-part system with class-based storage policies. European Journal of warehouse. European Journal of Operational Research, 277, 814–830.
Operational Research, 237, 1054–1066. van Gils, T., Ramaekers, K., Braekers, K., Depaire, B., & Caris, A. (2018a). Increasing order
Parikh, P., & Meller, R. (2010). A travel-time model for a person-onboard order picking picking efficiency by integrating storage, batching, zone picking, and routing policy
system. European Journal of Operational Research, 200, 385–394. decisions. International Journal of Production Economics, 197, 243–261.
Petersen, C. (1997). An evaluation of order picking routeing policies. International Journal Van Gils, T., Ramaekers, K., Caris, A., & De Koster, R. B. M. (2018b). Designing efficient
of Operations & Production Management, 17(11), 1098–1111. order picking systems by combining planning problems: State-of-the-art classification
Petersen, C. (1999). The impact of routing and storage policies on warehouse efficiency. and review. European Journal of Operational Research, 267(1), 1–15.
International Journal of Operations & Production Management, 19(10), 1053–1064. Van Nieuwenhuyse, I., & de Koster, R. (2009). Evaluating order throughput time in 2-
Petersen, C. (2002). Considerations in order picking zone configuration. International block warehouses with time window batching. International Journal of Production
Journal of Operations & Production Management, 22(7), 793–805. Economics, 121(2), 654–664.
Petersen, C., & Aase, G. (2004). A comparison of picking, storage, and routing policies in Vaughan, T. S., & Petersen, C. G. (1999). The effect of warehouse cross aisles on order
manual order picking. International Journal of Production Economics, 92(1), 11–19. picking efficiency. International Journal of Production research, 37(4), 881–897.
Petersen, C., & Schmenner, R. (1999). An evaluation of routing and volume-based storage Xiao, J., & Zheng, L. (2010). A correlated storage location assignment problem in a single-
policies in an order picking operation. Decision Science, 30(2), 481–501. block-multi-aisles warehouse considering BOM information. International Journal of
Pferschy, U., & Schauer, J. (2018). Order batching and routing in a non-standard ware- Production Research, 48(5), 1321–1338.
house. Electronic Notes in Discrete Mathematics, 69, 125–132. Xiao, J., & Zheng, L. (2012). Correlated storage assignment to minimize zone visits for
Ponzi Artusi, A. (2017). Sviluppo ed implementazione di un modello per lo studio e l’a- BOM picking. International Journal of Advanced Manufacturing Technology, 61,
nalisi dei flussi di picking del magazzino presso la ditta GEA NIRO-SOAVI. Degree 797–807.
thesis (unpublished), University of Parma, Department of Engineering and Yoon, C. S., & Sharp, G. P. (1996). A structured procedure for analysis and design of order
Architecture. Supervisor: prof. A. Volpi. pick systems. IIE Transactions, 28, 379–389.
Rao, S. S., & Adil, G. K. (2013). Optimal class boundaries, number of aisles, and pick list Yu, M., & De Koster, R. (2009). The impact of order batching and picking area zoning on
size for low-level order picking systems. IIE Transactions, 45(12), 1309–1321. order picking system performance. European Journal of Operational Research, 198(2),
Ratliff, H., & Rosenthal, A. (1983). Order-picking in a rectangular warehouse: A solvable 480–490.
case of the traveling salesman problem. Operations Research, 31(3), 507–521. Yu, Y., De Koster, R., & Guo, X. (2015). Class-based storage with a finite number of items:
Roodbergen, K., & de Koster, R. (2001). Routing methods for warehouses with multiple Using more classes is not always better. Production and Operations Management, 24(8),
cross aisles. International Journal of Production Research, 39(9), 1865–1883. 1235–1247.
Roodbergen, K. J., Sharp, G. P., & Vis, I. F. A. (2008). Designing the layout structure of Zhang, Y. (2016). Correlated storage assignment strategy to reduce travel distance in
manual order picking areas in warehouses. IIE Transactions, 40, 1032–1045. order picking. IFAC-PapersOnLine, 49(2), 30–35.
Roodbergen, K., & Vis, I. (2006). A model for warehouse layout. IIE Transactions, 8(10),

17