Content uploaded by Kerim Hodžić
Author content
All content in this area was uploaded by Kerim Hodžić on Dec 25, 2018
Content may be subject to copyright.
978-1-5386-6974-7/18/$31.00 ©2018 IEEE
Abstract — Distribution companies use complex software
systems called WMS (Warehouse Management System). The
WMS is an important part of the company’s business and it
can make processes simple to keep track of. Smart WMS
optimizes processes to save resources and to create a more
efficient working place. This paper describes the concept of
a smart WMS that is implemented in one of the largest
distribution companies in Bosnia and Herzegovina. The
system uses artificial intelligence and optimization
algorithms to improve working process. The paper describes
the complete warehouse workflow that includes stock
planning, initial product placement, transfer from stock to
pick zone, order picking process, transport and tracking.
The anomaly detection is used in some processes to improve
the whole system. The main contribution of this paper is the
presentation of an efficient and in the real world used smart
WMS concept.
Keywords — Smart Warehouse Management System,
Stock Planning, Order Picking Process, Location
Management, Vehicle Routing Problem
I. INTRODUCTION
HE basic purpose of the warehouse is to store
products in one place and transfer them forward to
company customers. Depending on the type of the
company, those products can be ordered and shipped from
manufacturers or manufactured in-house. The complete
distribution process consists of a number of smaller ones
and they must be properly monitored. Therefore, the
software solutions called Warehouse Management
Systems (WMS) are created. If the Warehouse
Management System is implemented well, it can save
large amounts of resources to distribution/manufacturing
companies by improving the complete working process
and providing useful business analysis. Many large
manufacturing companies require that their distributors
have implemented WMS. Complete product distribution
process is usually consisted of five chronologically
connected steps shown in Fig. 1 and each step can be
Emir Žunić is with the Info Studio d.o.o. Sarajevo and Faculty of
Electrical Engineering, University of Sarajevo (e-mail:
emir.zunic@infostudio.ba).
Sead Delalić is with the Faculty of Science, University of Sarajevo
and Info Studio d.o.o. Sarajevo (e mail: delalic.sead@pmf.unsa.ba).
Kerim Hodžić is with the Faculty of Electrical Engineering,
University of Sarajevo and Info Studio d.o.o. Sarajevo (e-mail:
kerim.hodzic@etf.unsa.ba).
Admir Beširević is with the Faculty of Science, University of
Sarajevo and Info Studio d.o.o. Sarajevo (e-mail:
besirevic.admir@pmf.unsa.ba).
Harun Hindija is with the Faculty of Science, University of Sarajevo
and Info Studio d.o.o. Sarajevo (e-mail: hindija.harun@pmf.unsa.ba).
analysed and optimized separately, but it is also important
to analyse and optimize them as connected parts of the
whole process.
Fig. 1. Main WMS processes
When the goods arrive, they need to be initially placed
in the warehouse in a proper position. Warehouses can
have many different layouts. Warehouse layout mostly
depends on the building shape. Often, the complete layout
is divided into sectors of the same or different shape.
Sectors are created to separate different brands or to
create different temperature conditions. Each sector
contains shelves that can be arranged in different shapes.
Warehouses can have separate entrances and exits. The
entrance is a part of the warehouse where the goods are
brought from the manufacturers and the exit location is
the part of the warehouse from where the orders are
transferred to the final customer. Shelves are organized by
arranging many pallet places in a line and also in levels to
enlarge warehouse capacity. Shelves are divided into
stock and picking zone. Picking zone, sometimes referred
to as the golden zone, is part of the warehouse racks from
where products can be handpicked. Higher racks are
called the stock zone and products from the stock zone are
usually picked using forklifts.
It is important to plan which and how much of every
product to store in the warehouse. Storing quantities of
products that are larger than the quantity that a company
can sell can cause unnecessary costs in terms of
warehouse space, warehouse worker paying hours and
electricity costs. On the other hand, not storing enough
quantities of product can lead to the company not being
able to deliver ordered quantities, which can lead to profit
and customer loss.
Customer orders are coming during the day and
warehouse workers need to collect ordered products. The
workers are going through the aisles between shelves to
Smart Warehouse Management System Concept
with Implementation
Emir Žunić, Sead Delalić, Kerim Hodžić, Admir Beširević, Harun Hindija
T
pick the ordered items. The described process is called the
order picking. The workers usually use the picking cart to
make the process faster and easier and to be able to travel
between aisles and cross-aisles. Orders usually contain
multi-brand items that are placed in different sectors.
Products stored on the top levels not available in the
picking zone are not reachable for workers to take them
by hand and they have to wait for a forklift to come to
transfer goods from the stock to the picking zone. That
consumes an additional time for order to be picked.
When all ordered items are collected, they have to be
transferred to customers, usually by transporting vehicles.
A set of transporting vehicles is usually called the
transport fleet. Orders are grouped by customers, usually
by their location and distance. Each group is assigned to a
vehicle in the fleet.
Each described process contains the set of smaller
processes that can be optimized. This paper describes a
concept of the smart warehouse management system
implemented in one of the largest distribution companies
in Bosnia and Herzegovina. The concept includes many
algorithms for optimizing the complete warehouse
management system.
This paper consists of five sections. In the first section,
an introduction is given and the complete warehouse
working concept is described. In the second section, a
literature review of related work is given. In the case
study section, proposed concept is described. Then, the
results are given and described for optimized processes.
In the final section, the conclusion and future work ideas
are presented.
II. RELATED WORK
Every aspect of the warehouse management system can
be analysed and optimized separately. That is conducted
in Section III. In this section, a few publications that are
analysing warehouse management systems in general are
mentioned and briefly described.
First step for improvement of the traditional warehouse
management system can be achieved by automating
processes. Article [1] investigates the impact of a
warehouse management system on supply chain
performance. The supply chain procedures carried out in
the warehouse were reviewed before and after
implementing software that handles warehouse
transactions.
More improvement can be achieved by implementing
more advanced data mining and artificial intelligence
techniques in warehouse processes. Paper [2] proposes
similar alternative called intelligent WMS (i-WMS). That
system consists of five subsystems: intelligent logistic
system, intelligent warehouse system, real-time
transportation monitoring, sales forecasting system, and
intelligent sales summary system. Research conducted in
that paper integrates state-of-the-art results in the field of
intelligent systems-neural network, bee colony
optimization, fuzzy control, and decision support system -
together with the Radio Frequency Identification (RFID)
technology. Many similar aspects are also covered in this
paper, analysed from the different perspective and with
different algorithms and methods implemented.
Barcodes or RFIDs are often used in WMS for product
tracking. In [3], it is described how the RFID can improve
the efficiency of the warehouse management and make a
rapid self-recording of receiving and delivery.
The objective of paper [4] was to propose and analyse
results of an Internet of things (IoT)-based warehouse
management system with analytical approach using
computational intelligence techniques to enable
warehouse smart logistics. Because of the wide range of
approaches and algorithms that can be used in WMS,
evaluating their effectiveness is an important aspect to be
covered. In paper [5], a simulation model is proposed to
evaluate the performance of a warehouse system for
perishable products taking into account the effect of the
uncertainties which typically affect the supply chain.
III. CASE STUDY
In this section, every cardinal warehouse operation is
described and a concrete solution for the operation
improvement is proposed. To implement smart warehouse
management system, prerequisite is to have developed a
warehouse management system that includes a database
with history of transactions and product locations. Then,
collected information are used for constructing algorithms
and solutions around it. In this particular case, before the
implementation of the smart WMS was conducted,
information system for distributing company was
implemented.
Every rack was labelled by barcode which represents a
location unique identifier. Some important functionalities
that information system provided was storing history
about every internal and external order, transaction and
product location change. As a part of the system, the
Android application whose main goal was to provide
warehouse workers easier way to register important
information to the system in a way that does not affect
their work was developed. Application screens for
registering new product and product location change are
shown in Fig 2.
Fig. 2. Android application screens
A. Stock planning
For every distributing company, key for success is to
provide more quality service to every customer while
spending fewer resources. In order to serve every
customer order, there has to be enough units of every
ordered product in stocks. The solution is not just to store
more and more product units, because space and human
resources are limited.
Stock has to be planned accordingly for the future. This
is an old problem in economy known as demand
forecasting and many different algorithms and approaches
are used for solving it. Study explained in [6] attempts to
identify the main determinants of forecasting accuracy
and gives directions which method to use on what kind of
data through simulations and empirical investigations
involving 14 popular forecasting methods.
An important aspect in demand forecasting is product
seasonality. Different products are showing different
seasonal patterns. Another important aspect that must be
considered are discounts. In solution described in this
paper, the module for demand forecasting is integrated
and it uses X-13 seasonality filter and anomaly detection
algorithm for detecting and filtering values that do not fit
into the pattern.
Algorithm is used to forecast monthly demand for every
product based on twenty years history of data and can also
cope with discounts if they are announced at least one
month in advance.
B. Receiving/Location Management
When products arrive to the warehouse, they need to be
placed in appropriate location. The effective product
positioning can improve the order picking process by
shortening walking routes for workers, which gives them
more time to handle other warehouse operations. Solution
for this problem can differ for an empty and already
operative warehouses not being able to stop their work for
some period of time and reorganise product locations.
As a part of the smart WMS, location management
module is implemented as set of SQL procedures that
suggest a list of sorted positions to which every product
should be placed when it comes into the warehouse. The
main goal of the algorithm is to provide convergence of
more frequent products to locations that are closer to
location where prepared orders are stored. In most cases it
is the exit location. The algorithm uses historical data for
calculation. More details about the algorithm
implementation, results and how the distances are
calculated can be found in papers [7] and [8]. Another
important aspect in product location management is
picking zone planning. If there are enough product units
in the picking zone, then workers collecting orders do not
have to use forklifts to get items from higher rack floors
and that leads to less fuel and time consumption.
For this purpose, the picking zone prediction algorithm
is implemented and it, by using the historical transaction
data, predicts which products and how many product units
are going to be ordered tomorrow, so the workers can
prepare the picking zone a day before. The algorithm is
implemented as a set of SQL procedures and is based on
shifted (0.9) median value of historical data describing
orders and adaptive coefficients. More details about the
algorithm implementation in six variants and comparative
results can be found in paper [9].
C. Order picking
Customers are making orders on a daily basis.
Information about orders are stored by the warehouse
management system as a list of products and numbers of
product units to be picked by the warehouse workers.
Usually, in the mid-sized warehouses, the worker is
walking through the warehouse and collecting product
units in a picking cart. Orders can contain dozens of
different products so determining the best path for
collecting order is not an easy task for a human,
especially if the important information like product unit
expiry date is considered. So, as part of the smart
warehouse management system, a module that gives
workers a sorted list of products and their locations that
he needs to collect in a suggested order with the earliest
expiration dates first is implemented.
As it is described in a paper [8], the order picking
problem can be easily transformed to the traveling
salesman problem (TSP). It can be stated that the TSP is
solved for practical cases by using the 2-opt and 3-opt
algorithms. Those algorithms require the ability to
calculate the distance between two items. The algorithm
that uses the dynamic programming is used to calculate
the shortest distances between two pallet places where the
items are placed. The generalization of the algorithm for
distance calculating is given in [10]. This distance
calculating generalization enabled the generalization of
the algorithm for order picking. The implemented
algorithm can make an optimal path for a warehouse of an
arbitrary and non-standard design.
The described concept requires the worker to scan a
picked article, so the WMS can update the stock data and
order data.
D. Shipping/Transport
The Vehicle Routing Problem (VRP) and optimal
utilization of the transport fleet has been well explored
and constantly improved in the last few decades.
The VRP is researched in the literature dealing with the
operational research. The problem could be described
briefly in this way: for a given set of customers, for a
known amount of goods to be delivered to each customer
at a precisely determined time interval, and a known fleet
of available vehicles together with their constraints, an
optimal set of routes should be determined, where each
route starts from the warehouse, each customer is served
with exactly one vehicle, all customers are served, and the
given set of routes has the smallest resource cost of all
possible route combinations. It should be emphasized that
the set of routes can be optimized by different criteria
(minimal fuel consumption, minimal number of vehicles,
minimal total travel route of all vehicles and similar). It
has been shown that such a problem belongs to a group of
so-called NP-hard problems, which are known to be
difficult to solve with very powerful computer systems,
and their manual solution without the use of information
technology is not a realistic option.
The transport optimization and tracking is very
important part of this smart WMS implementation and it
is used to improve the final warehouse operation, which is
delivering goods to final customers. The implemented
algorithm is consisted of three phases. The transformation
of the input data as the time window and distance of the
customers is made, which allows the unloading time for
each customer to become zero, which simplifies the
problem. In the second phase, an initial solution is
created. Then, this solution is attempted to improve
iteratively using the Tabu Search. A VRP module is
created as a part of the web application where users can
input parameters and analyse suggested transport routes.
The algorithm has been fully implemented in a way
that supports a very large number of realistic constraints.
By doing so, the realized algorithm is adapted to practical
applications and problems, including practical limitations
that the real situation undoubtedly contains. The
aforementioned system provides the ability to maximize
the cost of transport, optimize the fleet of used vehicles,
and, based on simulated situations, obtain the best
solution for future vehicle acquisitions.
One part of implemented smart WMS is the GPS
tracking module. The module is used to track vehicles by
using the data sent by the GPS tracking device. Data
gathering application, which communicates with devices,
parses the received data and saves the data do the
database. It is developed in the Java programming
language, by using multithreading to allow concurrent
execution for maximum utilization and flexibility.
Module for GPS data analysis is developed as part of
the integrated web application and it is shown in Fig. 3.
Fig. 3. GPS tracking application screen
Analysing driven routes and comparing them to the
suggested routes gives information about which
customers were served, information about the traveling
time between customers and loading time per order and
product. This information is gathered through time and
used to improve parameter values used in the
implemented vehicle routing algorithm.
E. Other improvements
The warehouse management system contains a large set
of data that are the result of the user input: orders data,
number of articles, articles description (mass, volume,
dimensions) etc. Sales prediction is based on the analysis
and usage of the previous sales data. Consequently, the
data accuracy is crucial for the correct execution of the
process and control of the stock. The proposed smart
WMS concept uses the anomaly detection algorithms for
analysis and corrections of the order data.
The type of order data is time-series. Adaptation of a
Twitter algorithm for detecting anomalies described in
[11] is used in the proposed smart WMS concept. The
Twitter algorithm reveals seasonality and trend in data,
which makes it suitable for use in the product ordering
systems. Early detection of anomalies can reduce the
number the cancelled orders. Detected anomalies that are
caused by mistake are eliminated from records used in the
sales predictions and in the stock planning module in
order to improve the prediction results.
Anomaly detection conducted for most exported
product sales is shown in Fig. 4.
Fig. 4. Anomaly detection module
IV. RESULTS
The concept described is implemented in one of the
largest distribution companies in Bosnia and
Herzegovina. The first improvement for company
business was made by implementing the warehouse
management system with a centralized database that
stores information like product location, orders and
transaction history etc. Although, there is no metric to
measure it, improvement in everyday work was
noticeable. The next part was the process of implementing
artificial intelligence algorithms to every step of the
complete process from the very beginning, which is the
process of order planning and product positioning, to the
end, which is the transporting products to the customers.
Order picking optimization module described in [8]
was implemented first and it showed the improvement of
more than 40% in average route length. After initial
success, the company decided to start using the algorithm
in other warehouses and some of them had a non-standard
layout. The algorithm was improved and adapted to work
on different warehouse layout and that process is
described in [10]. It was noticed that there is more place
for improvement if product locations were organised
better, so the module for optimal product positioning
described in [7] was implemented. One of the most
significant cost savings is that the number of items with
the expiration date is reduced to a minimum.
In Table 1, the analysis conducted one and nine months
after the algorithm is implemented are shown. It can be
noticed that average order picking route length reduced
most for smaller orders through time. Reason for that is
more space for improvement in orders containing less
products and also slow product movements to their
suggested location only when positions become free to
occupy.
TABLE I: AVERAGE PICKING ROUTE LENGTH REDUCTION
Number of
ordered items
Average length
reduction after one
month (%)
Average length
reduction after
nine months (%)
30-40 17.34 17.3
20-29 12.5 11.4
10-19 10 10.2
5-9 7.1 12.6
0-4 4.4 14.5
The next important aspect to improve was determining
which products and how much of them to put in the
picking zone so they can be easily accessed. Algorithm
for the picking zone capacity and content prediction using
order history described in [9] was implemented. It
provided more than 90% ordered product amounts to be
picked from picking zone and also reducing transfers
from stock to pick zone by 16.2 times and also reduced
the total number of transactions by more than 4%.
Anomaly detection is used to detect irregularities like
input or calculation mistakes in information about the
customer orders. Currently, every anomaly detected by
the algorithm is cross-checked by the company personnel.
From the total number of orders, approximately 3.55%
were detected by the algorithm as anomalies. From that
number, 67% were detected as correct by the company
personnel. After this phase and algorithm adjustment it is
planned to incorporate it in other parts of the WMS that
contain user inputs or calculations.
The stock planning algorithm is currently in a testing
phase. Every 2 weeks when orders are made, the
algorithm suggests the number of product units to be
ordered and a worker in charge of stock planning is
analysing the suggestion giving us a useful feedback.
Based on the current results, it is estimated that stock
supplies can be reduced by 5% immediately up to 10% in
the next 6 months.
The algorithm for vehicle routing is using well known
heuristic methods in each phase. On instances where the
algorithm is tested, the route proposals were
approximately 2% different from the optimal routes on
average. This percentage depends mostly on the number
of customers to be served, topographic layout of the
cities, established customer and vehicle constraints and
fleet of vehicles available at a given time. Routes were
created for on average over 120 customers in several
different cities. Most customers had defined time
windows and the fleet consisted of less than 10 available
heterogeneous vehicles on average. Result quality was
influenced by the fact that with the extremely complex set
of input data the primary purpose of the algorithm was to
find any solution that would meet the set limits, all in a
reasonable time interval.
The algorithm routes were compared to routes created
manually by the managers with many years of experience.
It is noticed that in cases where the number of customers
is about 100, the algorithm uses at least one vehicle less
than the manager. The implemented algorithm reduces
costs by approximately 30% for the mentioned number of
customers. The main reasons for the cost savings are the
better usage of smaller vehicles, a larger number of
articles in used vehicles, as well as an accumulatively
reduced travelled distance. In cases where the number of
customers is larger than 100, the savings are getting closer
to 40%.
V. CONCLUSION
The warehouse management system is one of the most
important parts of the working process in distribution
companies. The optimized processes can make large time
and cost savings and can make the more efficient
environment. The concept described in this paper is using
the artificial intelligence algorithms to improve standard
warehouse management system by giving optimized
solutions to users. The implemented concept optimizes
stock planning, initial product placement, stock to picking
zone transfer process, as well as order picking, transport
and tracking processes.
The future work plan is to implement optimization
techniques in other processes important for the
distribution business. The portal where customers can
make orders and get recommendations based on their
preferences will be implemented. The anomaly detection
algorithm will be improved and adjusted for the usage in
other warehouse processes.
The VRP is going to be improved by adding more
constraints to implemented optimization techniques. The
GPS tracking app will be improved with additional
analyses and algorithms for smart tracking and route
comparisons. The algorithm for order picking will be
improved by adding additional constraints as weight,
fragility, volume and other real-world constraints.
REFERENCES
[1] A. M. Atieh et al., “Performance Improvement of Inventory
Management System Processes by an Automated Warehouse
Management System,” in Procedia CIRP, 2016.
[2] R. Pulungan, S. P. Nugroho, N. El Maidah, T. B. Atmojo, P. D.
Hardo, and P. Pawenang, “Design of an Intelligent Warehouse
Management System,” in Information Systems International
Conference (ISICO), 2013.
[3] B. Yan, C. Yiyun, and M. Xiaosheng, “RFID technology applied
in warehouse management system,” in Proceedings - ISECS
International Colloquium on Computing, Communication, Control,
and Management, CCCM 2008, 2008.
[4] C. K. M. Lee, Y. Lv, K. K. H. Ng, W. Ho, and K. L. Choy,
“Design and application of Internet of things-based warehouse
management system for smart logistics,” International Journal of
Production Research. 2017.
[5] A. Giuseppe, E. Mario, and M. Cinzia, “Analysis of a warehouse
management system by means of simulation experiments,” in 10th
International Conference on Modeling and Applied Simulation,
MAS 2011, Held at the International Mediterranean and Latin
American Modeling Multiconference, I3M 2011, 2011.
[6] F. Petropoulos, S. Makridakis, V. Assimakopoulos, and K.
Nikolopoulos, “‘Horses for Courses’ in demand forecasting,” Eur.
J. Oper. Res., 2014.
[7] E. Zunic, H. Hasic, K. Hodzic, S. Delalic, and A. Besirevic,
“Predictive Analysis based Approach for Optimal Warehouse
Product Positioning,” 41st International Convention on
Information and Communication Technology, Electronics and
Microelectronics (MIPRO), 2018.
[8] E. Zunic, A. Besirevic, R. Skrobo, H. Hasic, K. Hodzic, and A.
Djedovic, “Design of optimization system for warehouse order
picking in real environment,” in ICAT 2017 - 26th International
Conference on Information, Communication and Automation
Technologies, Proceedings, 2017.
[9] E. Zunic, K. Hodzic, H. Hasic, R. Skrobo, A. Besirevic, and D.
Donko, “Application of advanced analysis and predictive
algorithm for warehouse picking zone capacity and content
prediction,” in ICAT 2017 - 26th International Conference on
Information, Communication and Automation Technologies,
Proceedings, 2017.
[10] E. Zunic, A. Besirevic, S. Delalic, K. Hodzic, and H. Hasic, “A
Generic Approach for Order Picking Optimization Process in
Different Warehouse Layouts,” 41st International Convention on
Information and Communication Technology, Electronics and
Microelectronics (MIPRO), 2018.
[11] O. Vallis, J. Hochenbaum, and A. Kejariwal, “A Novel Technique
for Long-term Anomaly Detection in the Cloud,” Proccedings 6th
USENIX Work. Hot Top. Cloud Comput. (USENIX ’14), 2014.
