See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/328290180

Machine Learning Techniques for Smart Manufacturing: Applications and

Challenges in Industry 4.0

Conference Paper · October 2018

CITATION READS

1 10,197

5 authors, including:

Bojana Bajic Ilija Cosic

University of Novi Sad Faculty of Technical Sciences University of Novi Sad
23 PUBLICATIONS 461 CITATIONS 98 PUBLICATIONS 957 CITATIONS

SEE PROFILE SEE PROFILE

Milovan Lazarevic Nemanja Sremčev

University of Novi Sad University of Novi Sad Faculty of Technical Sciences
49 PUBLICATIONS 479 CITATIONS 37 PUBLICATIONS 232 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Aleksandar Rikalovic on 16 October 2018.

The user has requested enhancement of the downloaded file.

Machine Learning Techniques for Smart
Manufacturing: Applications and
Challenges in Industry 4.0
B. Bajic a, I. Cosic a, M. Lazarevic a, N. Sremcev a, A. Rikalovic a

a
Faculty of Technical Sciences, University of Novi Sad, Trg Dositeja Obradovića 6,
21000 Novi Sad, Serbia, E-mail: [email protected],
{ilijac, laza, nextesla, a.rikalovic}[email protected]

Abstract
The Industry 4.0 is now underway, changing traditional manufacturing into smart manufacturing and
creating new opportunities, where machines learn to understand those processes, interact with
environment and intelligently adapt their behaviour. Big data and artificial intelligence (AI) make
machines in industrial production smarter than before addressing the question of how to build
computers that improve automatically through experience. Machine learning (ML), as a subfield of
AI, has become the main driver of those innovations in industrial sectors, which provides the
opportunity to further accelerate discovery processes as well as enhancing decision making. However,
ML algorithms learn directly from the examples, data and experience and are able to figure out how
to perform important tasks by generalizing from them. This paper summarizes challenges and future
trends of ML applications for smart manufacturing and provides an overview of several ML
algorithms (e.g. support vector machine, k-nearest neighbour, neural network etc.) that are able to
give the answers to those issues and avoid the potential problems in the future.

Keywords: artificial/computational intelligence, deep learning (DL), industry 4.0, machine learning (ML)

1. INTRODUCTION [5]. ML uses increased computing power and

various software for gaining the meaningful
The Industry 4.0 lays on the increment of
information and knowledge from the big data,
available data sets. Different types of available,
which are collected from the environment, but
large and complex sets of data, namely big data,
also, has ability to learn from those data by
cannot be processed by using existing
getting the artificial/computational intelligence
conventional technologies. Advanced methods,
[5], [6]. For some specific tasks, ML is able to
technologies, algorithms and software must be
achieve a higher level of requirements than
used in order to collect and extract data from the
human. This highlights the importance of the big
manufacturing environment [1]–[4]. Big data
data from which the information is obtained.
changes the way decisions are made inside the
However, the balance has to be found. Too
manufacturing environments based on different
much information can lead to delay of the
scientific areas such as computer science,
actions or the wrong conclusions of the certain
mathematics and advanced statistics [4]–[6].
problem and the lack of the information may not
The field, that combines together all of these lead to problem solving [4]. Another great issue
sciences is machine learning (ML) [4], [5]. ML is related to security aspect of the data [7].
is becoming the most important method that is Having that issue in mind, ML has to utilize the
used for predicting and classifying the difficulty different techniques and algorithms in order to
solving problems inside the production systems achieve maximal benefit from the data [4], [6].

29
The most important techniques that are used for data and its corresponding output while
learning, classified by the available feedback, determining the structure in the machine so that
are supervised, unsupervised and reinforcement mapping can be learned [5].
learning methods [8]. This paper focuses on the
Different researches have different
challenges and applications that ML faces with
approaches to structure the field of machine
in today’s manufacturing systems. Also, the
learning, but the structure that is most widely
accent is put on future trends of ML in
used is classified by learning processes, and that
manufacturing applications where the primary
is supervised, unsupervised and reinforcement
objective lays behind the utilization of big data
learning, shown in Figure 1 [4], [8], [14].
in order to accomplish cost efficient, fault-free
and optimal quality manufacturing process [9].

2. BACKGROUND
This section presents an overview of the
related studies of Industry 4.0 and ML, as well
as the ML techniques.
2.1. Industry 4.0 and Machine
Learning Fig 1. ML techniques and algorithm (adapted from [4])

The rapid development of technologies al of Precision Eng

interconnected with ICT and internet of things The supervised learning is machine learning
enables the growth of manufacturing which has technique, specified for large amount of input
led to the Industry 4.0 [10]. The implementation data (training sets), are applied to the systems
of CPS combined with IoT can provide where the correct response is provided by the
intelligent, flexible systems capable of self- knowledgeable expert [4], [5].
learning which presents the core of Industry 4.0
In supervised learning, a system is trained
[1]. In order to achieve intelligent and flexible
with data that has been labelled. The labels
systems, the big data is required. In knowledge
categorise each data point into one or more
discovery in databases (KDD) of big data,
groups. Then the system learns how this training
machine learning plays an important role along
data is structured and uses this to predict in
with data mining, statistic, pattern recognition
which categories to classify new output data [3].
and other methods [11], [12]. ML, as a part of
The final goal of completed supervised learning
intelligent system in Industry 4.0, is broadly
process is that the outputs are close enough to be
implemented in various fields of manufacturing
useful for all given input sets [15].
where its techniques are designed to extract
knowledge out of existing data [13]. The new The most common supervised machine
knowledge (information) supports the process of learning assignments are classification and
decision-making or making prediction of regression [4], [16]. In classification
manufacturing system. But the end goal of the assignments, the program has to learn how to
ML techniques is detection of the patterns among predict the most likely category, class or label
the data sets or regularities that describe the for discrete output values from one or more
relationships and structure between those sets [4]. input data sets [16]. Similar to classification,
regression problem, also requires supervised
2.2. Machine Learning techniques
learning techniques. The difference in regression
ML system, that maps an input into an problems is that programs must foresee and
output, needs to be trained in order to learn. The predict the value of a continuous output by
system training is achieved by giving an input themselves [17].

30
 According to Wuest et al. [4] as well as Reinforcement learning system receives
Jordan and Mitchell [18], the supervised inputs while interacting with manufacturing
learning is the most commonly used ML environments and making sequential decisions
technique, because majority of applications can in order to maximise future rewards [6]. It
provide labelled data. addresses assignments where some of the data
UNSUPERVISED LEARNING
presents labelled training sets and other data
does not [5]. In reinforcement learning, instead
The unsupervised learning represents the of training datasets that indicate the correct
intelligent learning where evaluation of the output for a given input, the training datasets are
action is not dependent, provided nor presumed to provide only an indication whether
supervised, because there is no knowledgeable an action is correct or not. If an action is not
expert [4], [5], [15], [19]. Unlike supervised correct, there still remains the problem of
learning, the unsupervised learning does not finding the correct action [18].
learn from labelled data. Instead of that, it
discovers patterns among the data [16].
3. ML ALGORITHMS IN
The assignment of unsupervised learning is
MANUFACTURING
to discover groups of related observations of the
input data, namely clusters [6]. Such This section provides an overview of ML
observations within groups have cognation algorithms used in manufacturing processes.
based on some similarly measurements where
3.1. Support Vector Machine
similar points are grouped together [6], [16].
The main goal of unsupervised learning is to Support vector machine (SVM), a supervised
discover the unknown relationships between learning algorithm, is used for linear as well as
classes using the clustering analysis [4]. for non-linear problems, such as classification
According to Jordan and Mitchell [18] and and regression [5], [16], [20]. SVM was formed
Hackeling [16] the another unsupervised based on the idea of creation the flat hyperplane
learning task is dimensionality reduction. It or set of hyperplanes [21]. The hyperplanes
represents the process of discovering the divide the high-dimensional or infinite-
relationships between input data sets and can be dimensional vector space into distinguished
used for visualising. Considering that some parts with maximum margin distance between
problems might contain thousands and the two nearest training data points of classes
thousands of input data, problem with big data [20], [21]. SVM test point of data is said to
becomes impossible to visualize [16]. belong to one class if it is located below the
hyperplane, and the other way around [21]. The
REINFORCEMENT LEARNING mapping into an infinite-dimensional vector
Reinforcement learning presents another ML space is preformed by kernel functions which
technique that has focus on learning from led to the main goal of SVM and that is fixing
experience [4]–[6]. According to report of The the computational predicting problems [21],
Royal Society [6] and Jordan and Mitchell [18], [22]. The goal of an SVM is to produce a model,
reinforcement learning presents combination of based on the training data, that predicts the test
unsupervised and supervised learning point of data where is needed a subset of the
techniques, where the information available in training data [5]. In order to achieve the goal,
the training data is intermediate between i.e. to achieve prediction of maximum accuracy,
supervised and unsupervised learning, while SVM requires a large sample size [23].
Wuest et al. [4] and Hinton [5] consider it as “a However, the advantage of SVM is its ability
special form of supervised learning”, known as to work with incomplete data as well as the
semi-supervised learning. speed of classification. The biggest drawbacks

31
of SVM is slow speed of learning and its lack of 3.4. k-Nearest Neighbour
explanation ability to humans [23]. The k-nearest neighbour (k-NN) is the ML
3.2. Decision Tree algorithm used for non-linear problems, i.e.
classification and pattern recognition that
Decision tree is machine learning algorithm
requires computation of distances between
that is easily understandable and humans
sample of data where each input data is labelled
interpretable due to its graphical representation
of its k closest neighbouring samples [26].
[5], [24]. The challenging issue is finding an
optimal type of decision tree for training data Ismail et al. [25] consider that k-NN as a
sets. Two types of decision trees are mentioned prediction algorithm, similar to clustering due to
by Hinton [5]. The first type is classification tree its ability to predict and compare similarities of
which gives a categorical output, while the the big data. According to Kotsiantis [23], k-NN
second type is regression trees, which gives algorithm is very sensitive to data values that are
numerical output. Another disadvantage of missing, noisy, fuzzy, irrelevant and redundant,
decision tree is the inability to solve non-linear and the speed of classification is slow. That
problems unlike the SVM algorithm, but it has leads to accuracy in general which is quite low.
quite high speed of learning [23]. Decision tree Also, the k-NN is hard to interpret due to its
has wide utilization in exploration and unstructured collection of training big sets of
prediction problems due to its ability to score so data. Contrary to that, k-NN has a great speed of
highly on critical features of data mining [25]. learning which makes it one the fastest learning
ML algorithms [23]. Figure 4 represents an
3.3. Rule – Based Learners
example of k-NN where the space is divided in
Rules-based learners, also known as expert four different areas by coloured circle. An
systems [5], are considered to be one of the empty circle presents a new data, which falls in
major forms of machine learning in combination green area.
with data mining. Also, rule-based learners are
used for extraction the information based on
statistical significance with the help of “if-then”
rules [11]. Although they have application in
both, supervised and unsupervised learning, they
have the greater utilization in unsupervised
learning environments for KDD due to its
comprehensibility [23], [25]. As Hinton [5] and
Kotsiantis [23] noticed, the drawback of rule-
Fig 4. An example of k-NN algorithm
based learners is classification accuracy due to
acquisition of the knowledge. However, 3.5. Naive Bayesian
classification accuracy can be improved by Naive Bayesian (NB) represents ML
automatic feature construction algorithms and algorithm, which is defined as a very simple
combining different characteristics where the model composed of acyclic graphs. Acyclic
background expert knowledge is used. graphs of NB model have only one parent and
Another advantage of the rule-based learners multiple children. The parent represents the
is that the system is able to explain the process unobserved node, while the children are
how the result was generated where the learning observed nodes [4], [23].
algorithm always methodically checks the entire According to Kotsiantis [23], naive Bayesian
sets of data [5]. Considering the noise, rule- algorithms are usually less precise that some
based learners are resistant to it due to their other ML algorithm that are more sophisticated
pruning strategies which avoid overfitting the (i.e. ANN) due to the supposition of
data [23]. independence among children. The opposite of

32
Kotsiantis opinion, Hastie et al. [15] consider which it is constructed. There it can be multiple
that NB algorithm can often outperform some hidden layers, not just one as presented.
more sophisticated algorithms due to the fact However, if there is a greater number of hidden
that it cannot affect the estimation dramatically layers with very large dataset, than is talked
and that NB has significant advantages over about deep neural networks (DNN) [29].
other algorithms.
Naive Bayesian is useful for problems related
with classification, regression, clustering and
others [27], [28]. Also, it requires little storage
space during both stages: training and
classification [4]. But the major advantages of
the NB classifier are its short computational
time for training, fast process of learning and
ability of working with big, fuzzy, noisy and
incomplete data [4], [23].
3.6. Artificial Neural Network
Fig 5. An example of ANN (adapted from [30])
Artificial neural network (ANN) is ML
model that is used for nonlinear classification DNNs are based on fundamental concept of
and regression problems [11], [16], applied in ANN. A deep structure of DNN exploits
various fields of manufacturing due to the fact multiple hidden layers with multiple neurons in
that it plays an important role in today’s each layer, a nonlinear activation function, a
problem solving by parallel processing, where it cost function and a back-propagation [31]
simulates the decentralized “computation” of the algorithm for information-processing in a
humans central nervous system. The ANN hierarchical architecture for pattern
allows an artificial system to perform classification problems [32].
supervised, unsupervised and reinforcement In comparison with ANN architectures,
learning assignments [4]. The challenging task DNNs are able to learn high-level features,
for ANN is achieving the high accuracy where which are more complex and abstract, by
the big data is required. Another issues are integrating feature learning [31] and model
related to dealing with overfitting, missing data construction into a single model. That model is
values, speed of learning and complexity of the created by selecting different kernels or tuning
models they produce [23]. the parameters by end-to-end optimization
Similarly to SVM, ANN is capable of where the parameters of DNN model are trained
handling the high-dimensional data, continuous jointly withоut human supervision [33], [34].
features and high-variance, but the process of
learning is slow [23], [25]. The ANN is good for 4. APPLICATIONS,
KDD for hidden patterns or trends due to its
CHALLENGES AND FUTURE
abilities of processing which has similarities
TRENDS OF MACHINE
with human brain [25]. In contrary to that, АNN
requires complete records of data to do their LEARNING
work. In other words, the data must not be Machine learning techniques bring many
fuzzy, noisy or incomplete. Another drawback is improvements inside the manufacturing
that the large size of sample is required in order environments. The conducted review led to
to achieve its maximum prediction accuracy identification of application, challenges and
[23]. The example of ANN algorithm is shown future trends of ML which are discussed in the
on Figure 5, where is presented the layers of following subsections.

33
4.1. Application building a qualitative knowledge base where the
The application of ML techniques is in results of a simulation experiment are used.
Also, the supervised learning is employed for
constant enlargement over the last couple of
investigating the decision-making and process-
decades [35]. Growing application of ML has
led to the availability of complex big data that planning problems in manufacturing [38].
has certain patterns [36], database technologies, Nowadays, the ML algorithms have wide
computer power [37] and usability of ML utilization in different manufacturing areas such
techniques [4]. as optimization, troubleshooting and quality
From the perspective of manufacturing, control [36]. The result of the scientific research
various types of big data sets can be captured, has shown that ML techniques is consider to be
collected, extracted and analyzed in order to a powerful tools for permanent quality
improvement in a large and complex processes,
improve the traditional manufacturing systems
e.g. semiconductor manufacturing [38].
[1], [37]. Finding the knowledge in big data and
transforming it into information is done by KDD Machine learning techniques have
with the help of ML techniques [1]. According application in LEAN manufacturing systems as
to Escobar Diaz and Morales-Menendez [37], well by using just-in-time (JIT) and kanban
the primary objective of ML applications in tools. The results of implementation of ML into
combination of big data analytics is the LEAN manufacturing systems show that the
achievement of defect-free and fault-free neural networks and decision trees represent two
processes. most practical algorithms with special abilities
Most of the manufacturing issues appertain for adjusting the number of kanbans in a
dynamic JIT manufacturing environments [38].
under classification problems, where the experts
of industrial field have to determine a label of The ML has many security applications for
the class to specific object or a situation based controlling the access and verification. On the
on the big data set [38]. Problems of one hand, security application uses pattern
classification do not have to be related only to recognition along with other methods where the
manufacturing but can relate to the entire goal is to decide if the given face needs to be
industry. As Rikalovic et al. [39] mentioned classified or labelled as unfamiliar. On the other
application of machine learning techniques in hand, the goal of verification is to decide
combination with ANFIS (adaptive neuro-fuzzy whether person is the person he claim that he is
inference systems) that rely on expert while dealing with various conditions, e.g.
knowledge has utilization in intelligent decision lighting conditions, facial expressions etc. [35].
support system for industrial site classification.
The application of ML algorithms such as k-
The applications of ML in manufacturing refer nearest neighbour and naive Bayesian have
to pattern recognition in existing sets of data. That utilization in building predicting models such as
is beneficial for development of foreseeing the predicting the failure of airplane components
future behaviour of the manufacturing system with before they stop working [38].
the end goal of detecting the present behaviour
Another very important application of ML is
patterns or regularities that describes relations
related with transparency as “the ability of an
between data [1], [4].
organization to unravel and quantify
Another advantage of ML application is its uncertainties to determine an objective
ability to further improve the extracted estimation of its manufacturing capability and
knowledge by learning from results where ML readiness” [40]. The transparency can be
has ability to create new valuable information achieved as reducing the complexity of the
[4]. As Pham and Afify [38] mentioned in their results where ML algorithms evolve the patterns
paper, the supervised ML technique is used for from existing data and extracting the knowledge

34
used for decision-making and giving an [45]. Due to such concerns, the additional
approximations about future behaviour [4]. researches are needed in order to improve the
Also, as Lee and Lapira [41] and Lee et al. [42] storage as well as the security of big data.
stated that the increment of transparency in However, sometimes the nature and size of big
manufacturing environments has a number of data does not allow the big data to be located in
advantages, but the application of ML leads to the single space, but has to be distributed over
the major cost reduction in every aspect of the the distinct physical locations. In order to
transparent manufacturing environment. achieve connection as well as the learning
system between the big data from the entire
4.2. Challenges and future trends
manufacturing environment the communications
Machine learning is still young scientific through wireless devices is required [43].
field that is growing rapidly due to its practical
Another major challenge as well as the future
applications. Having that in mind, one of the
trend is development of the ML algorithms as
major trends is related with the environment in
which a ML algorithms operate due to issues of well as the improvement of existing ones
parallelism [43]. ML systems are interconnected implemented in manufacturing systems that are
with the complex software collections which run able to handle various situations and learn from
them [1]. As the Wuest et al. [4] consider that
on computing platforms and provide a range of
algorithms and services to data analysts [43]. the question what ML technique and algorithm
Due to that, as mentioned before, trends but also to choose is a great issue due to their advantages
the challenges are related with big data from and disadvantages. Because of that the so called
which knowledge can be extracted [13]. “hybrid algorithms” that present the
combination of various ML algorithms, are
Today, manufacturing is facing an increment becoming more and more used in the flexible
of challenges related to complexity and dynamic manufacturing systems [46]. However, the
behaviours [44] while adding the fact that the biggest challenge for the development and
manufacturing is affected by uncertainty [4]. In application of ML algorithms that
other words, the constant enlargement of big manufacturing is facing today is finding the
data and its availability, high-dimensionality, human experts in data science and optimization
variety as well as homogeneity represents the scientific fields having in mind that different
main challenges in manufacturing environment skills are required [47].
because the knowledge cannot be extracted. In
order to overcome these challenges, the machine
learning techniques and algorithms are used [4]. 5. DISCUSSION AND
However, the gathered big data has tendency to CONCLUSION
contain irrelevant, missing and redundant This paper provides the preliminary literature
information which can lead to the impact on the overview of the machine learning techniques as
performance of the ML learning abilities [1]. a part of intelligent system (i.e. supervised,
The major concerns related with collecting unsupervised and reinforcement learning) and
new kinds of big data, motivated by its the mostly used algorithms as well as their
economic value, is privacy and security issues advantages and disadvantages within the
since the big data are stored in virtual cloud Industry 4.0. Due to the fact that there can be
platforms [28], [43]. The misuse of big data in confusion when it comes to machine learning
manufacturing, as well as in entire industry, is in and statistics, a detailed comparison is given and
increment, because of the wireless remote the differences are presented between these two
control of the physical devices. Therefore, it is methods. Also, the application, challenges and
possible to take control over the physical future trends of machine learning represent the
machinery if the data is not properly secured the main focus of this paper.

35
 The application of ML is closely related to development of ML and its implementation
big data without which the development and inside intelligent manufacturing environments
implementation of ML techniques would not be with focus on Industry 4.0.
possible due to the fact that extracting the
knowledge is the most important action that lead
REFERENCES
to the achievement of defect-free and fault-free
processes. Also, the further improvement of ML [1] B. Nikolic; J. Ignjatic; N. Suzic; B.
application enable the the learning process Stevanov; A. Rikalovic, Predictive
where the artificial/computational intelligence manufacturing systems in industry 4.0:
plays the most important role. Today, the ML trends, benefint and challenges, Proc. of the
algorithms have wide utilization in different 28th DAAAM Inter. Simp. on Intelligent
manufacturing areas such as optimization, Manufacturing and Automation, pp. 796-
control, troubleshooting, security and 802, (2017).
verification where the increment of transparency
[2] H.S. Kang et al., Smart manufacturing: Past
of the entire manufacturing environment is
research, present findings, and future
beneficial for cost reduction without affecting
directions. International Journal of
quality of production. Also, ML algorithms have
Precision Engineering and Manufacturing –
application in semiconductor as well as in
Green Technology 3(1) (2016) pp. 111-128.
LEAN manufacturing environments where the
results of the experiment show an improvement [3] J. Lee; H.A. Kao; S. Yang, Service
in process quality. innovation and smart analytics for Industry
4.0 and big data environment. Procedia
Challenges and future trends in machine
CIRP 16 (2014), pp. 3-8.
learning are closely related. All that represents
the current challenges also represents the future [4] T. Wuest; D. Weimer; C. Irgens; K.-D.
trends in further research on ML techniques and Thoben, Machine learning in
algorithms. The most challenging assignment as manufacturing: advantages, challenges, and
well as the biggest trend is gathering the big data applications. Production & Manufacturing
from the manufacturing environments in order Research 4(1) (2016) pp. 23-45.
to have enough extracted knowledge and [5] G. E. Hinton, Machine Learning, Apress,
information for developing the ML algorithms New Jersey, USA (2017).
that are able to learn automatically from patterns
and previous behaviour. Sometimes big data can [6] The Royal Society, Machine learning: the
be irrelevant, fuzzy, noisy and redundant which power and promise of computers that learn
additionally complicates the learning process. by example, (2017).
Also, because of the fact that the extremely large [7] L. Monostori, Cyber-physical production
amount of data is collected, the storage space systems: Roots, expectations and R&D
represents the other challenging issue which challenges. Procedia CIRP 17 (2014), pp.
leads to privacy, security and economic 9-13.
questions. However, still the biggest challenge
[8] L. Monostori, Ai and machine learning
and future trend that manufacturing is facing
techniques for managing complexity,
today is reflected in the further development and
changes and uncertainties in manufacturing.
application of ML algorithms, but also finding
IFAC Proceedings 35(1) (2002) pp. 119-
the human experts in data science and
130.
optimization scientific fields represent another
challenging issue. This paper is the initial part of [9] S. Yin; O. Kaynak, Big data for modern
the research, so in future its continuity will industry: challenges and trends. Proc. IEEE
contribute to the further exploration and 103(2) (2015), pp. 143-146.

36
[10] J. Qin; Y. Liu; R. Grosvenor, A categorical Humanized Computing 9(3) (2018), pp.
framework of manufacturing for industry 789-801.
4.0 and beyond. Procedia CIRP 52 (2016),
[21] K. Hansson; S. Yella; M. Dougherty; H.
pp. 173-178.
Fleyeh, Machine learning algorithms in
[11] S.S. Ghuman, A review of data mining heavy process manufacturing. American
techniques. Industrial Management & Data Journal of Intelligent Systems 6(1) (2016),
Systems 3(4) (2014) pp. 1401-1406. pp. 1-13.
[12] O. Doğan, A roadmap for data driven [22] J.H. Friedman, Recent advances in
decision making to improve quality a predictive (machine) learning. Journal of
roadmap for data driven decision making to Classification 23(2) (2003), pp. 175-197.
improve quality. Proc. of the 3rd Int. Conf.
[23] S.B. Kotsiantis, Supervised machine
on Engineering and Natural Sciences
learning: a review of classification
(ICENS), pp. 1-8, (2017).
techniques. Informatica 31 (2007) pp. 249-
[13] S. Sumathi; L.A. Kumar; S.P, 268.
Computational Intelligence Paradigms for
[24] C. Gröger; F. Niedermann; B. Mitschang,
Optimization Problems Using MATLAB®/
Data mining-driven manufacturing process
SIMULINK®, (2016).
optimization. Proc. of the World Congress
[14] M. K. Bharatkumar, Intelligent system’s on Engineering, vol. III, p. 7, (2012).
design approaches: a review. International
[25] R. Ismail; Z. Othman; A.A. Bakar, Data
Journal of Engineering and Management
mining in production planning and
Research 5(4) (2015), pp. 208-213.
scheduling: a review. Proc. of the 2nd Conf.
[15] T. Hastie; R. Tibshirani; J. Friedman, The on Data Mining and Optimization, pp. 154-
Elements of Statistical Learning, Springer, 159, (2009).
(2008).
[26] G.A. Susto; A. Schirru; S. Pampuri; S.
[16] G. Hackeling, Mastering Machine Learning McLoone; A. Beghi, Machine learning for
with scikit-learn, Packt Publishing, (2014). predictive maintenance: A multiple
classifier approach. IEEE Transactions on
[17] B. Ratner, Statistical and Data Mining
Industrial Informatics 11(3) (2015), pp.
Second Edition Machine-Learning, CRC
812-820.
Press Taylor & Francis Group, (2011).
[27] M. Chen; S. Mao; Y. Liu, Big data: A
[18] M.I. Jordan; T.M. Mitchell, Machine
survey. Mobile Networks and Applications
learning: trends, perspectives, and
19(2) (2014), pp. 171-209.
prospects. Science 349(6245) (2015), pp.
255-260. [28] N. Khan et al., Big data: survey,
technologies, opportunities, and challenges.
[19] Z. Sun, A framework for developing
The Scientific World Journal 2014 (2014),
management intelligent systems.
pp. 1-18.
International Journal of Systems and
Service-Oriented Engineering 6(1) (2016), [29] S. Gu; L. Rigazio, Towards deep neural
pp. 37-53. network architectures robust to adversarial
examples, no. 2013, (2014).
[20] J.A. Saucedo-Martinez; M. Perez-Lara; J.A.
Marmolejo-Saucedo; T.E. Salais-Fierro; P. [30] P. Kim, MATLAB Deep Learning: With
Vasant, Industry 4.0 framework for Machine Learning, Neural Networks and
management and operations: a review. Artificial Intelligence, (2017).
Journal of Ambient Intelligence and

37
 [31] M. Usama et al., Unsupervised machine [41] J. Lee; E. Lapira, Predictive factories: the
learning for networking: techniques, next transformation. Manufacturing
applications and research challenges, pp. 1- Leadership Journal (2013), pp. 2-9.
37, (2017).
[42] J. Lee; E. Lapira; S. Yang; A. Kao,
[32] L. Deng, A tutorial survey of architectures, Predictive manufacturing system - Trends
algorithms, and applications for deep of next-generation production systems.
learning, APSIPA Transactions on Signal IFAC 46(7) (2013).
and Information Processing 3 (2015).
[43] M.I. Jordan; T. M. Mitchell, Machine
[33] K. He; J. Sun, Convolutional neural learning: trends, perspectives, and
networks at constrained time cost, Proc. of prospects. Science 349(6245) (2015), pp.
the IEEE Conference on Computer Vision 255-260.
and Pattern Recognition (CVPR), pp. 5353-
[44] A. Gupta, Ramifications of machine
5360, (2015).
learning in the manufacturing sector.
[34] V. Sze; Y.-H. Chen; T.-J. Yang; J. Emer, Gyancity Journal of Engineering and
Efficient processing of deep neural Technology 3(1) (2017), pp. 31-37.
networks: a tutorial and survey,
[45] M. Khan; X. Wu; X. Xu; W. Dou, Big data
Proceedings of the IEEE, pp. 1-32, (2017).
challenges and opportunities in the big data
[35] A. Smola; S.V.N. Vishwanathan, challenges and opportunities in the hype.
Introduction to Machine Learning, Proc. of the IEEE ICC 2017 SAC
Cambridge University Press, (2008). Symposium Big Data Networking Track
Big, (2017).
[36] E. Alpaydin, Introduction to machine
learning, MIT Press, (2004). [46] H. Li, Improve the performance of a
complex fms with a hybrid machine
[37] C.A. Escobar Diaz; R. Morales-Menendez,
learning algorithm. Journal of Software
Machine Learning and pattern recognition
Engineering and Applications 10(3) (2017),
techniques for information extraction to
pp. 257-272.
improve production control and design
decisions, Industrial Conference on Data [47] M. Baumers; E. Özcan, Scope for Machine
Mining, pp. 286-300, (2007). Learning in Digital Manufacturing, (2016).
[38] D.T. Pham; A.A. Afify, Machine-learning
techniques and their applications in
manufacturing, Journal of Engineering
Manufactureol 219(5) (2005), pp. 395-412.
[39] A. Rikalovic; I. Cosic; R.D. Labati; V.
Piuri, Intelligent decision support system
for industrial site classification: A GIS-
based hierarchical neuro-fuzzy approach,
IEEE Systems Journal 12(3) (2018), pp.
2970-2981.
[40] J. Lee; E. Lapira; B. Bagheri; H. an Kao,
Recent advances and trends in predictive
manufacturing systems in big data
environment. Manufacturing Letters 1(1)
(2013), pp. 38-41.

38

View publication stats