Content uploaded by Eftim Zdravevski
Author content
All content in this area was uploaded by Eftim Zdravevski on Mar 20, 2021
Content may be subject to copyright.
Making vending machines smarter with the use
of Machine Learning and Artificial Intelligence:
Set-up and Architecture
Dashmir Istrefi1and Eftim Zdravevski2
1Information Systems, College for Business and Technology, 10000 Prishtina,
Republic of Kosovo, dashmir.istrefi@ubt-uni.net,
2Faculty of Computer Science and Engineering, Sts. Cyril and Methodius University
in Skopje, Macedonia, eftim.zdravevski@finki.ukim.mk
Abstract. Vending machines conveniently provide snacks, beverages,
cigarettes, food, and other items to consumers. Management of the ma-
chine supplies and cash is a challenging logistic process. At a time when
the Internet of Things paradigm becomes omnipresent, industries are cur-
rently going through “The Fourth Industrial Revolution”. These things
combined are also being adopted in the vending machines industry, lead-
ing to the development of smart vending machines. This paper presents
our hoyo.ai chip that can be plugged into non-smart vending machines
and enable them internet connectivity and other smart features. With
this affordable chip, thousands of vending machines’ lifespan can be ex-
tended, enabling smart features such as predictive maintenance and just-
in-time restocking.
Keywords: Smart Vending Machines, Just-in-time restocking, Machine
Learning, Predictive Maintenance.
1 Introduction
Industries are currently going through “The Fourth Industrial Revolution”, a
term also known as “Industry 4.0”. What it mostly concerns is the integration
amongst physical and digital systems of the production contexts [1]. A prime
reason for that is the adoption of machine learning (ML) to provide some ad-
vantages. Those advantages include maintenance cost reduction, , machine fault
reduction, spare-part life increases and inventory reduction, operator safety en-
hancement, increased production, an increase in overall profit, to name a few.
All of them also have a tremendous and strong bond with the procedures of
maintenance [2, 3]. Moreover, fault detection is one of the critical components
of predictive maintenance, and it is necessary for industries to detect faults at a
very early stage [4, 5].
More recent technologies can enrich vending machines with smart features,
such as the sizeable digital touch display, various types of sensors, cameras, inter-
net connectivity, more cost-effective embedded computing power, various pay-
ment systems, and a wide range of identification technology (e.g., NFC, RFID)
[6]. These smart vending machines engage users in a more rich user experience,
reduce operating costs, improve the vending operations’ efficiency by remote
management and insightful analytics based on the collected data. These ma-
chines offer everything from entertainment to cashless payments through arti-
ficial intelligence, facial recognition, and transparent displays. Integrated sen-
sors and cameras provide valuable data about customer demographics, purchase
trends, and other information about the location’s specifics where the machine
is set up. Global shipments of smart vending machines are estimated to reach
around 3.6 million units in 2020 with a penetration rate of 20.3% [7]. Likewise,
the market changes so that vending machines now offer fresh food, milk, and
juices [8].
With over 2 billion transactions a day on vending machines, the variety of
data collected can be analyzed to obtain insights [9]. That vast amount of data
aggregated requires a different type of analytics that can increase the vending
machines’ productivity while offering better customer experience and simplify-
ing the maintenance and service intervals. The benefits also might include churn
prediction [10]. It has implications on profits that depend on the commercial
food supplier involvement [11]. Machine learning algorithms can be applied for
automated fault detection and diagnosis based on the collected data type. How-
ever, it is very tricky to select appropriate machine learning techniques, type
of data, data size, and equipment to apply machine learning (ML) in industrial
systems considering the vast volumes of data. Therefore efficient cloud-based
solutions for Big Data processing might be employed [12]. Selection of inappro-
priate predictive maintenance technique, dataset, and data size may cause time
loss and infeasible maintenance scheduling [13].
Even though the benefits of new smart vending machines are undoubted,
millions of legacy vending machines are in use across the world that does not
possess hardware capabilities to be “smart”. This paper presents our hoyo.ai
chip that can be plugged into non-smart vending machines and enable them
internet connectivity and other smart features. With this affordable chip, thou-
sands of vending machines’ lifespan can be extended, enabling smart features
such as predictive maintenance and just-in-time restocking. With the proposed
chip, the ultimate goal is to minimize the Total Cost of Ownership (TCO) for
vending operators while enhancing the consumer purchasing experience, driving
up adoption of the “Internet of vending machines” without having actually to
retire vending machines which are still functional but without built-in hardware
for smart capabilities.
The remainder of this paper is structured as follows. In section 2 we review
the recent related works, and then in section 3 we present the architecture of the
proposed system and the created prototypes and production-line chips. Finally,
section 4 concludes the paper.
2 Related Work
Machine Learning and robust optimization techniques can significantly improve
logistics operations and improve stock quantity and maintenance intervals. Ma-
chine Learning will be used to forecast item demands for each of the vending
machines, taking into account past demands and calendar effects. In like manner,
they can embody context-aware personalized recommender systems [14]. The ap-
plication of ML in smart vending machines has already been explored in works
like [15], where authors describe a use-case in coffee vending machines. Authors
of [16] describe a benchmark dataset for smart unmanned vending machines that
can be used to verify different machine learning approaches. It is a good rep-
resentation of the challenges commonly encountered in such multi-modal, and
multi-source datasets [17], requiring processing of nominal and categorical data
that needs to be combined with time series and other data [18, 19].
A conceptual framework for collaborative forecasting in the food supply chain
is presented in [20]. Authors of [21] present a reliable decision support system
for fresh food supply chain management. Ideas from such approaches applied
to the traditional fresh food supply chain can also be applied to smart vending
machines.
In [4] the researchers have chosen simple vibration data collected from an
exhaust fan. They have fit different unsupervised learning algorithms such as
PCA T2statistic, Hierarchical clustering, K-Means, Fuzzy C-Means clustering
and model-based clustering to test its accuracy, performance, and robustness.
As a result, they have proposed a methodology to benchmark different algorithms
and choosing the final model.
In [5] a novel multiple classifier Predictive Maintenance (PdM) system for
integral type faults is presented. The multiple Machine Learning (ML) classifiers
work in parallel to exploit the knowledge of the tool/logistic variables at each
process iteration to enhance decision making. In their work, the proposed tool
guarantees improved maintenance management decisions in terms of minimizing
operating cost and can be applied to any maintenance problems characterized
by integral type faults provided Run-to-Failure (R2F) historical data can be
collected or is available.
Technologies involved in[16] solution of unmanned vending machines include
Cloud (1) Management System, (2) Computer Vision and Deep Learning, (3) QR
code, Mobile Payment, RFID, and (4) Sensor, Camera, Electronic tag, Antenna.
In this paper, the two benchmarks of datasets the examined deep learning models
can achieve approximately 99.67% performance on accuracy according to their
experiment.
3 System Architecture
The method is data-driven and uses extensive amounts of data, either streamed,
on-board data, or even historical and aggregated data from off-board databases.
The methods rely on a telematics hub that communicates with a flespi IoT plat-
form https://flespi.com/. A knowledge base is created so that it can be used
to predict upcoming failures on other vehicles that show the same deviation.
A classifier is trained to learn patterns in the usage data that precede specific
repairs and can be used to predict vehicle maintenance. Set-up of SDK script to
get sensory data from the vending machine that are additionally attached and
Multi-Drop Bus / Internal Communication Protocol (MDB/ICP) that commu-
nicates with all peripheral devices of the vending machine.
The adoption of the proposed chip relies on a methodology consisting of four
phases:
1. Set up the ecosystem
–Setting up additional hardware on existing vending machines
–IoT integration
2. Data collection approach: Gather Empirical data to carry out an experiment
–Collecting data about usage behavior
–Collecting data with system state
–Data standardization
3. Modeling
–Feature extraction and data fusion
–Traditional Machine Learning models
–Deep learning models
–Optimization
4. Deployment
–Integrating ML/AI into the UI
–Cost/benefit analysis
–Market-basket analysis
–Iterative improvements of ML models
The foundation for the ecosystem that facilitates the aforementioned method-
ology is shown in Fig. 1.
The initial prototype developed and field-tested is shown in Fig. 2. The pro-
totype is an Arduino Uno device connected with A6 GSM/GPRS module. After
experimenting with different modules and configurations, the prototype is min-
imized only with the needed components. SIM900A model of the GSM/GPRS
module was selected, with ATMEGA328p placed on the PCB designed to fit this
modification. A connector with screws for RS232 communication is used instead
of DB9.
The miniaturized version of the prototype is shown in Fig. 3.
The following step was to finalize the PCB design so it is ready for serial pro-
duction. It includes improvements of the module’s rigidity and robustness that
it is ready for environmental factors, vibrations, and other operating conditions.
The result is shown in Fig. 4.
The prerelease version for hoyo.ai is shown in Fig. 5.
Fig. 1: Methodology of hoyo.ai
Fig. 2: Initial prototype developed and field-
tested
Fig. 3: Miniaturized version
of the prototype
Fig. 4: New PCB design for
serial production
Fig. 5: Pre-release version for hoyo.ai
4 Conclusion
As cities worldwide go into lockdown and shopping centers, retail outlets are
closed; as a result, vending operators need to quickly shift to different strategies
like cost-saving initiatives if they hope to ensure their survival on the long run.
With the help of big data and artificial intelligence, they might be able to achieve
precisely that, meaning that they at least might be able to avoid the Covid-19
induced economic fallout that looms for the rest of the business world.
By performing analytics and predictions, which are forwarded to a robust op-
timization model, whose outputs will be the cash transport, and these transports
guarantee that demand is fulfilled up to the desired confidence level, preventing
vending machines’ downtime due to unplanned maintenance and out-of-stock
situations while also satisfying additional constraints arising in this particular
domain. As a result of such operations, we expect productivity improvements
of vending machines from 20 to 40%. Machine Learning and Robust Optimiza-
tion techniques can significantly improve logistics operations and improve stock
quantity and maintenance intervals regarding vending devices. Machine Learn-
ing will be used to forecast item demands for each of the vending machines,
taking into account past demands and calendar effects.
In future work, it is planned to prepare the initial dataset and enrich it
with more diversity, like more product categories, integrate RFID/NFC cashless
payments, IoT portal that interacts with end customers.
Most importantly, the proposed hoyo.ai chip extends the life of existing vend-
ing machines by providing an affordable upgrade that connects them to the inter-
net so that vendors and clients can benefit from the IoT advances. The proposed
system relies on state-of-the-art automated feature engineering methods for se-
lecting robust features from various sensors, which are used to generate reliable
classification and prediction models. By utilizing state-of-the-art classification
and deep learning models, it provides unparalleled performance and benefits.
References
[1] Carvalho, T.P., Soares, F.A., Vita, R., Francisco, R.d.P., Basto, J.P., Alcal´a,
S.G.: A systematic literature review of machine learning methods applied
to predictive maintenance. Computers & Industrial Engineering 137 (2019)
106024
[2] Sezer, E., Romero, D., Guedea, F., Macchi, M., Emmanouilidis, C.: An
industry 4.0-enabled low cost predictive maintenance approach for smes.
In: 2018 IEEE International Conference on Engineering, Technology and
Innovation (ICE/ITMC), IEEE (2018) 1–8
[3] Wan, J., Tang, S., Li, D., Wang, S., Liu, C., Abbas, H., Vasilakos, A.V.: A
manufacturing big data solution for active preventive maintenance. IEEE
Transactions on Industrial Informatics 13(4) (2017) 2039–2047
[4] Amruthnath, N., Gupta, T.: A research study on unsupervised machine
learning algorithms for early fault detection in predictive maintenance. In:
2018 5th International Conference on Industrial Engineering and Applica-
tions (ICIEA), IEEE (2018) 355–361
[5] Susto, G.A., Schirru, A., Pampuri, S., McLoone, S., Beghi, A.: Machine
learning for predictive maintenance: A multiple classifier approach. IEEE
Transactions on Industrial Informatics 11(3) (2014) 812–820
[6] Solano, A., Duro, N., Dormido, R., Gonz´alez, P.: Smart vending machines
in the era of internet of things. Future Generation Computer Systems 76
(2017) 215–220
[7] Saunders, R.H.: American Faces: A Cultural History of Portraiture and
Identity. University Press of New England (2016)
[8] Pereira, A., Villanueva-Rey, P., Vence, X., Moreira, M.T., Feijoo, G.: Fresh
milk supply through vending machines: Consumption patterns and associ-
ated environmental impacts. Sustainable Production and Consumption 15
(2018) 119–130
[9] Lucas, A.: Big data, artificial intelligence and vending machines: How
coca-cola continues to assert dominance. https://datafloq.com/read/
future-of-snack-consumption- post-covid-world/8477/ (2020-06-05)
Accessed on 2020-09-01.
[10] Zdravevski, E., Lameski, P., Apanowicz, C., Slezak, D.: From big data
to business analytics: The case study of churn prediction. Applied Soft
Computing 90 (2020) 106164
[11] Terry-McElrath, Y.M., Hood, N.E., Colabianchi, N., O’Malley, P.M., John-
ston, L.D.: Profits, commercial food supplier involvement, and school vend-
ing machine snack food availability: implications for implementing the new
competitive foods rule. Journal of school health 84(7) (2014) 451–458
[12] Grzegorowski, M., Zdravevski, E., Janusz, A., Lameski, P., Apanowicz, C.,
Slezak, D.: Cost optimization for big data workloads based on dynamic
scheduling and cluster-size tuning. Big Data Research 25 (2021) 100203
[13] C¸ınar, Z.M., Abdussalam Nuhu, A., Zeeshan, Q., Korhan, O., Asmael, M.,
Safaei, B.: Machine learning in predictive maintenance towards sustainable
smart manufacturing in industry 4.0. Sustainability 12(19) (2020) 8211
[14] Jeong, S.Y., Ryong, K., Kim, Y.K.: A context-aware personalized recom-
mender system in smart vending machine. International Information Insti-
tute (Tokyo). Information 21(7) (2018) 1989–1998
[15] Kim, K., Park, D.H., Bang, H., Hong, G., Jin, S.i.: Smart coffee vending
machine using sensor and actuator networks. In: 2014 IEEE International
Conference on Consumer Electronics (ICCE), IEEE (2014) 71–72
[16] Zhang, H., Li, D., Ji, Y., Zhou, H., Wu, W., Liu, K.: Toward new retail: A
benchmark dataset for smart unmanned vending machines. IEEE Transac-
tions on Industrial Informatics 16(12) (2019) 7722–7731
[17] Kalajdjieski, J., Zdravevski, E., Corizzo, R., Lameski, P., Kalajdziski, S.,
Pires, I.M., Garcia, N.M., Trajkovik, V.: Air pollution prediction with multi-
modal data and deep neural networks. Remote Sensing 12(24) (2020) 4142
[18] Zdravevski, E., Lameski, P., Kulakov, A., Kalajdziski, S.: Transformation
of nominal features into numeric in supervised multi-class problems based
on the weight of evidence parameter. In M. Ganzha, L. Maciaszek, M.P.,
ed.: Proceedings of the 2015 Federated Conference on Computer Science
and Information Systems. Volume 5 of Annals of Computer Science and
Information Systems., IEEE (2015) 169–179
[19] Zdravevski, E., Lameski, P., Kulakov, A.: Weight of evidence as a tool for
attribute transformation in the preprocessing stage of supervised learning
algorithms. In: Neural Networks (IJCNN), The 2011 International Joint
Conference on, IEEE (2011) 181–188
[20] Eksoz, C., Mansouri, S.A., Bourlakis, M.: Collaborative forecasting in the
food supply chain: a conceptual framework. International journal of pro-
duction economics 158 (2014) 120–135
[21] Dellino, G., Laudadio, T., Mari, R., Mastronardi, N., Meloni, C.: A reliable
decision support system for fresh food supply chain management. Interna-
tional Journal of Production Research 56(4) (2018) 1458–1485
