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

net/publication/347633153

Production Planning and Scheduling Using Machine Learning and Data

Science Processes

Chapter · December 2020

DOI: 10.3233/ATDE200153

CITATIONS READS

9 1,804

3 authors, including:

Ederson Carvalhar Fernandes Milton Borsato

Centro Universitário UNINTER Federal University of Technology - Paraná/Brazil (UTFPR)
12 PUBLICATIONS 62 CITATIONS 123 PUBLICATIONS 705 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Metodologia de Projeto de Produto View project

Método para análise de performance de linhas produtivos baseados em modelos digitais alimentados por dados em tempo real View project

All content following this page was uploaded by Milton Borsato on 22 February 2021.

The user has requested enhancement of the downloaded file.

SPS2020 155
K. Säfsten and F. Elgh (Eds.)
© 2020 The authors and IOS Press.
This article is published online with Open Access by IOS Press and distributed under the terms
of the Creative Commons Attribution Non-Commercial License 4.0 (CC BY-NC 4.0).
doi:10.3233/ATDE200153

Production Planning and Scheduling Using

Machine Learning and Data Science
Processes
Paulo Henrique DE MODESTIa,1, Ederson CARVALHAR FERNANDESa and Milton
BORSATOa
a
Universidade Tecnológica Federal do Paraná, Curitiba, Parana, Brazil

Abstract. Increasing manufacturing efficiency has been a constant challenge since

the First Industrial Revolution. What started as mechanization and turned into
electricity-driven operations has experienced the power of digitalization. Currently,
the manufacturing industry is experiencing an exponential increase in data
availability, but it is essential to deal with the complexity and dynamics involved
to improve manufacturing indicators. The aim of this study is to identify and allow
an understanding of the unfilled gaps and the opportunities regarding production
scheduling using machine learning and data science processes. In order to
accomplish these goals, the current study was based on the Knowledge
Development Process – Constructivist (ProKnow-C) methodology. Firstly,
selecting 30 articles from 3608 published articles across five databases between
2015 and 2019 created a bibliographic portfolio. Secondly, a bibliometric analysis,
which generated comparative charts of the journals’ relevance regarding its impact
factor, scientific recognition of the articles, publishing year, highlighted authors
and keywords was carried out. Thirdly, the selected articles were read thoroughly
through a systemic analysis in order to identify research problems, proposed
solutions, and unfilled gaps. Then, research opportunities identified were: (i) Big
data and associated analytics; (ii) Collaboration between different disciplines; (iii)
Solution Customization; and (iv) Digital twin development.

Keywords. Machine Learning, Production Planning, Data Science, Predictive

Analytics, Scheduling.

Introduction

Technological development in recent decades has provided a huge amount of data. This
dataset is known as Big Data [1]. The manufacturing industry is experiencing several
advantages from this data availability [2]. Some of these benefits turned out in higher
quality and productivity [3]. On the other hand, it is essential to deal with the dynamics
involved in it. Despite the benefits and opportunities, there are many risks to improve
manufacturing indicators that cover qualified professionals and complexity [4].

1
Corresponding Author. [email protected].
156 P.H. De Modesti et al. / Production Planning and Scheduling Using ML and Data Science Processes

The development of computer science tools offers high potential to change the
manufacturing domain. Some of these developments are applied to machine learning
(ML), which is the ability of computers to solve problems without being programmed
to do so [5]. Although computer engineering has been such a challenge for many
manufacturing companies, the absence of its full potential application can increasingly
affect their competitiveness level.
Strong candidates to surpass most of the major challenges of complex
manufacturing systems are ML algorithms. This data-driven approach is capable of
solving highly complex, non-linear patterns in data of several types and sources and
transforming raw data into models, which are then applied to prediction, and in this
case, production scheduling [5].
The application of ML techniques has increased over the last years due to several
factors, such as the availability of complex data with poor transparency [6] and the
increased convenience and power of ML tools. Many ML algorithms are developed to
analyze large amounts of data and are capable of handling high dimensionality [7].
This paper presents a literature review for ML methods and data science processes
applied to the production scheduling in manufacturing. The main contribution of this
research is to identify which trends and proposed solutions have been developed from
the last five years as well as to find unfilled gaps and opportunities of research in this
field.
This research is organized into five sections. The first section is the introduction;
the second section presents the methodological aspects; the third section describes the
research development; the fourth section discusses the main conclusions and, finally,
the fifth one exhibits the final considerations.

1. Methodological aspects

According to [8], a methodological framework consists of an examination and

description of the approaches used in the research, in order to define the set of methods,
procedures, and assumptions relevant to the subject. The lack of a standard that allows
the adoption a single procedure to establish the research methodology allows the choice
of the methodological framework to vary according to the researcher’s knowledge [9].
In addition, [10] stated that a research can be classified according to both its
purpose and design. Goal-based research is divided into three groups: exploratory,
descriptive and explanatory. The design-based study contains two main groups: paper-
source or people-source research. Paper-source research has two categories:
bibliographic and documentary. People-source has several categories as experimental,
ex-pos-facto, gathering, case study, action research, and participant. Considering the
goal of this research, it can be labeled as exploratory, as it aims to provide greater
familiarity with the topic, in order to make it more explicit or make hypotheses. It can
be said that exploratory research has as main objective the improvement of ideas or the
discovery of intuitions. In addition, as the source of information is based on material
already prepared, composed of scientific articles, the research was classified as
bibliographic. Then, the approach adopted proposed by [10] is summarized in Figure 1.
 P.H. De Modesti et al. / Production Planning and Scheduling Using ML and Data Science Processes 157

Figure 1. Methodological structure.

The development of the study was based on the Knowledge Development Process
– Constructivist (ProKnow-C) methodology, proposed by [8]. This methodology is
divided into four steps: (i) bibliographic portfolio selection; (ii) bibliometric analysis;
(iii) systemic analysis and; (iv) definition of the research question and objective. The
purpose of using this methodology is to facilitate the selection of relevant articles as it
aims to identify references with high academic relevance in databases accessed through
the Internet, using the bibliometric analysis to filter the selected articles.
The ProKnow-C methodology is adopted up to the systemic analysis step as the
goal of this research is to identify trends and opportunities related to the topics under
analysis. Access to research in databases was carried out through the CAPES journal
portal. The software that supported the study was Mendeley Desktop © version 1.19.4
and Microsoft Excel for Mac 2011 version 14.3.9.

2. Research development

The research process begins with a problem that will motivate the search for data to be
examined. The bibliographic portfolio along with bibliometric and systemic analysis,
are the main steps towards achieving the research goals [10]. The portfolio includes all
relevant published content for the main topic and analysis is vital to discover new
158 P.H. De Modesti et al. / Production Planning and Scheduling Using ML and Data Science Processes

opportunities and gain a deep understanding of production scheduling using machine

learning and data science processes.

2.1. Bibliographic portfolio

The bibliographic portfolio contains scientifically recognized publications from 2015 to

2019 and encompasses the most developed stage of knowledge on the subject so far.
In this study, three axes were chosen to be analyzed: Data Science, Production
Planning and Machine Learning. In order to have a precise search, 18 keywords were
defined to better match the axes analyzed.
The Boolean operators “AND” and “OR” were used to find articles associated with
the keywords. The combination used was: ("data analytics" OR "data mining" OR
"complex event processing" OR CEP OR "predictive analytics" OR "data acquisition")
AND ("Production planning” OR scheduling OR "mixed-integer program*” OR MILP
OR "monitoring system*” OR prediction OR "smart manufacturing" OR "intelligent
manufacturing”) AND ("regression learning" OR "supervised learning" OR "pattern
classification" OR "supervised classification" OR "artificial intelligence" OR "machine
learning").
There are several scientific databases for searching publications, but only five were
selected due to a preliminary analysis of both their relevance in manufacturing research
and the availability of full access. They were: Web of Science, Wiley, ProQuest,
Scopus and IEEE (Figure 2).

Figure 2. Research scheme.

The result obtained by primary research in the databases had a total of 3608
publications. For getting the best relevant articles from the bibliographic portfolio, four
consecutive filters were performed for the current raw data: Published in related
Vehicles, Duplicate Elimination, Title Reading and Abstract Reading.
Then, a search was filtered for publication in related vehicles (i.e. in each database
to select articles exclusively and in specific areas according to the research goal), such
as Production Operations Management, Operations Research, Computer Science,
Engineering and Decision Sciences. Therefore, the outcome was 1241 raw articles.
Thus, Mendeley was used to help find the most relevant articles across the
following three filters as Duplicate Elimination, Title Reading and Abstract Reading.
 P.H. De Modesti et al. / Production Planning and Scheduling Using ML and Data Science Processes 159

Mendeley is a free reference manager, which is used to facilitate access and analysis of
each article. The main information of the portfolio such as authors, title, year of
publication and journal were downloaded directly from the databases to Mendeley, in
order to obtain a single set of data.
Firstly, a duplication check was performed in Mendeley, resulting in a reduction of
37 articles. Secondly, after reading 1204 article titles, the portfolio was reduced to 86
papers related to the topic. Thirdly, abstract readings of each remaining article were
performed, resulting in 32 articles considered related to the research subject. The result
of each filter is illustrated in Figure 3.

Figure 3. Filter actions and results.

2.2. Bibliometric analysis

Bibliometric analysis is related to the understanding and evaluation of articles in the

bibliographic portfolio. This analysis was performed through charts containing
portfolio information, such as impact factor, number of citations, publications by year,
type of research and most of the published journals.
The portfolio impact factor, as shown in Figure 4, has a regular distribution,
ranging from 1 to 6. Two journals can be highlighted with a higher impact factor.

Figure 4. Impact factor of vehicle.

In Figure 5 is possible to see several articles with a large number of citations.
Through this number of citations, it is possible to conclude that the portfolio has good
relevance.
160 P.H. De Modesti et al. / Production Planning and Scheduling Using ML and Data Science Processes

Figure 5. Number of citations distribution.

These two with more than 150 citations belonging to the authors Fei Tao and
Zhiqiang Ge can be considered references in relation to the selected axes.
The increase in the number of articles per year in the last five years of research
(Figure 6) shows the trends of the most discussed and published subjects. The slight
reduction in the last year is explained by the fact this is the current year.

Figure 6. Number of articles published by year.

Analyzing the classification of the 30 types of articles, 77% of the articles are
classified as proposed solutions, with the remaining percentage divided equally in
evaluation research and philosophical paper (Figure 7).

Figure 7. Type of research.

 P.H. De Modesti et al. / Production Planning and Scheduling Using ML and Data Science Processes 161

It is usually possible to regularly check the publication of an article by journal, but

its differentiation has been identified in some journals. Figure 8 presents this statement.
IEEE Access is the journal with the largest number of publications. In addition, most of
the articles in this journal were printed in the current year.

Figure 8. Number of publications by journal and year.

Conform to the results, 63% of the 30 published articles were classified as

descriptive, and the others are opportunities for predictive and prescriptive
developments (Figure 9).

Figure 9. Classification of the research according to the result.

And last but not least, a comparative chart was created on published articles
classified by the manufacturing area to analyze how manufacturing topics were divided
between these studies. It was identified in four categories: Virtual Manufacturing,
Process and Planning, Control, and Scheduling and Quality Management. Their results
are visible in Figure 10.
162 P.H. De Modesti et al. / Production Planning and Scheduling Using ML and Data Science Processes

Figure 10. Published articles classified per manufacturing area and year.

2.3. Systemic analysis

The purpose of systemic analysis is to deeply study each article in the portfolio in order
to find gaps in the literature. The criterion for finding gaps is based on the author’s
decision about what is relevant to the research [11]. The articles were read and
analyzed in full in the following aspects: objective, context problem, methodology,
results, trends, and opportunities.
And last but not least, after reading full-text filter, 12 articles were eliminated
because they were not related to the research subject. The remaining articles resulted in
20 articles. Then, an extra review of the article references was carried out and another
17 articles were analyzed. From this review, 7 were eliminated and 10 added to the
final portfolio. The final result of the number of articles in the portfolio is 30 articles.

2.3.1. Identified research problems

The main problems found by the authors were highlighted as follows.
Large datasets: several studies in the portfolio reveal that there are many
algorithms capable of solving production planning problems [12]. On the other hand,
most of these algorithms take a long time to solve problems or fail when they work
with large datasets.
External factors: combining the use of different algorithms, it is possible to solve
many complex artificial intelligence problems, such as maintenance prediction or
digital twin. On the contrary, it is important to note that external factors are not
considered, such as economic recession, weather or an accident [13].
Find the best algorithm: artificial intelligence is a broad concept, where there are
many different strategies to be adopt. One such strategy is ML, which has several
algorithms. Therefore, an important problem published by some portfolio authors is the
difficulty in deciding which algorithm is more appropriate to use.

2.3.2. Proposed solutions

Although the models proposed in the portfolio can identify relationships between
variables, they may not indicate a causal relationship. However, causality is essential in
 P.H. De Modesti et al. / Production Planning and Scheduling Using ML and Data Science Processes 163

identifying true root causes. Therefore, due to the complexity of the topic, researchers
are challenged to investigate [5].
As stated by [14], process and planning are the most researched area of Big data in
manufacturing. Therefore, manufacturers must be able to understand the exponential
increase in demands for data production, even adopting the analytical techniques
necessary to extract the meaning of this data.
[15] stated both the digital twin and big data play important roles in promoting
smart manufacturing. The digital twin allows manufacturers to access real-time
information and validate data between physical and virtual equipment. Combined with
accurate analysis and predictive resources, this technique can provide the manufacturer
with a huge improvement in his productivity.
It is proposed by [16] that ML can act with Big Data analytics. The role of Big
Data is to retrieve the desired data from the database, and ML is to predict information.
By integrating the two techniques, it is possible to work with large datasets in different
formats and provide a suitable solution for manufacturing.
Finally [17], intelligent manufacturing means using advanced data analytics to
complement physical science and improve productivity and decision making. Deep
learning serves advanced analytics tools to process and analyze manufacturing data.
The development of deep learning technologies, a subset of artificial neural networks
with various layers, and their advantages over traditional ML is clear. Therefore,
computational methods based on deep learning improve system performance in
manufacturing.
[18] presented a paper to provide a computational methodology based on industrial
alarm analysis and an event-log database. Based on the results of frequent sequence
mining. A novel methodology and associated network-based visualization technique
have been published to learn how the alarm messages that occur represent the states of
technology and can be applied to predict future events. In the end, a neural network
was explicitly designed to capture long-term dependency on sequences, and a
sequence-to-sequence deep learning model was proposed.

2.3.3. Unfiled gaps identification

For the raised problems and their solutions previously presented, some unfilled gaps
can be identified.
• Big data and associated analytics: these will be the key technologies to extract
the required knowledge and provide intelligence in all engineering processes.
As proposed by [14], it is logical that future work should focus on the
development of systematic reviews and literature related to the manufacturing
areas identified in this study, such as creating a systematic big data review in
the manufacturing diagnosis.
• Collaboration between different disciplines: the need for interdisciplinary
studies related to artificial intelligence presents a great opportunity, but also a
significant risk, as collaboration between different disciplines, such as
Computer Science, Industrial Engineering, Mathematics, and Electrical
Engineering, is necessary to boost progress, according to [5] and [15].
• Solution Customization: [19] states that manufacturing is undergoing a
technological revolution that, in turn, should revolutionize the industry. On the
164 P.H. De Modesti et al. / Production Planning and Scheduling Using ML and Data Science Processes

other hand, users will require more personalization and electronics features for
all products. This will close the cycle between design, manufacturing,
marketing, sales, and aftermarket fields of the manufacturing industry.
• Digital twin development: the Digital twin enables manufacturers to manage
two-way real-time mapping between a physical object and its digital
representation, which opens the way for deep cyber-physical integration. [20]
In combination with a digital twin, intelligent data-driven manufacturing will
become more responsive, adaptable, and predictive.

3. Discussion chapter

Figure 11 shows the trends identified in the studies based on the selected portfolio.
Process and planning show an increasing trend of study over the next years. Scheduling
and control can be considered a stable field since the number of published studies is
constant. It is interesting to note that Virtual Manufacturing shows a decline, but it is
possible to infer that some related articles have not yet been published, justifying the
decline. There is only one article related to Quality Management, so it is not possible to
perform any analysis.
There are several machine learning technologies available to researchers. Based on
the portfolio, it is possible to assume that Artificial Neural Networks are the most used
algorithm in the research field, with k-means in second place.

Figure 11. Published articles in manufacturing across the years.

 P.H. De Modesti et al. / Production Planning and Scheduling Using ML and Data Science Processes 165

Figure 12. Types of machine learning more used in manufacturing.

4. Conclusions

The study does not attempt to exhaust the proposed subject, as there may be other
sources to be investigated. Anyway, this work used a structured methodology
designated ProKnow-C in a bibliographic and systemic analysis of production
scheduling using machine learning and data science processes.
Firstly, the bibliographic portfolio was created, composed of 3608 published
articles, found in five databases between 2015 and 2019. Then, 30 articles relating to
the topic were selected.
Secondly, through a bibliometric analysis, the bibliographic portfolio was
interpreted and evaluated. In addition, a comparison of the relevance of the journals in
relation to the impact factor, scientific recognition of the articles, publishing year,
highlighted authors and keywords was done.
Thirdly, in the systemic analysis the entire portfolio was thoroughly read, in order
to identify the research problems, the proposed solutions, and the unfilled gaps. The
research opportunities identified were: (i) Big data and associated analytics; (ii)
Collaboration between different disciplines; (iii) Solution Customization; and (iv)
Development of digital twins.
The use of the ProKnow-C methodology allows the researcher to have a deep
understanding of the problems, goals, gaps, and opportunities under analysis. However,
it depends on the researcher’s previous experience in selecting appropriate research
axes and keywords in order to provide the best result.
In future studies, the intention is to perform the Big data analysis associated with
ML, and focusing on scheduling of the production process, since the use of these tools
tends to be widely used in the industry.
 166 P.H. De Modesti et al. / Production Planning and Scheduling Using ML and Data Science Processes

References

[1] J. Lee, E. Lapira, B. Bagheri, and H. Kao, “Recent advances and trends in predictive manufacturing
systems in big data environment,” Manuf. Lett., vol. 1, pp. 38–41, 2013.
[2] S. Chand and J. F. Davis, “What is smart manufacturing?,” Time Magazine, 2010.
[3] M. Elangovan, N. R. Sakthivel, S. Saravanamurugan, B. B. Nair, and V. Sugumaran, “Machine
learning approach to the prediction of surface roughness using statistical features of vibration
signal acquired in turning,” Procedia Comput. Sci., vol. 50, pp. 282–288, 2015.
[4] T. Wuest, Identifying product and process state drivers in manufacturing systems using supervised
machine learning. New York, NY: Springer Verlag, 2015.
[5] T. Wuest, D. Weimer, C. Irgens, and K. D. Thoben, “Machine learning in manufacturing:
Advantages, challenges, and applications,” Prod. Manuf. Res., vol. 4, no. 1, pp. 23–45, 2016.
[6] A. Smola and S. V. N. Vishwanathan, Introduction to machine learning. Cambridge: Cambridge
University Press, 2008.
[7] K. Yang and J. Trew, Multivariate statistical methods in quality management. New York, NY:
McGraw-Hill, 2004.
[8] J. E. Tasca, L. Ensslin, S. R. Ensslin, and M. B. M. Alves, “An approach for selecting a theoretical
framework for the evaluation of training programs,” J. Eur. Ind. Train., vol. 34, no. 7, pp. 631–655,
2010.
[9] S.M. Petri, "Modelo para apoiar a avaliação das abordagens de gestão de desempenho e sugerir
aperfeiçoamentos: sob a ótica construtivista" Programa de Pós-Graduação em Engenharia de
Produção, Universidade Federal de Santa Catarina, Florianópolis, 2005.
[10] A. C. Gil, Como Elaborar Projetos de Pesquisa, 3rd ed. Belo Horizonte: Atlas, 1996.
[11] L. ENSSLIN, S. R. ENSSLIN, and H. M. PINTO, “Processo de investigação e análise
bibliométrica: avaliação da qualidade dos serviços bancários,” Revista de Administração
Contemporânea, pp. 325–349, 2013.
[12] M. Usman and S. Asghar, “An Architecture for Integrated Online Analytical Mining,” J. Emerg.
Technol. WEB Intell., vol. 3, no. 2, 2011.
[13] R. Bragança, F. Portela, and M. Santos, “A regression data mining approach in Lean Production,”
Concurr. Comput. Pract. Exp., p. e4449, Mar. 2018.
[14] P. O’Donovan, K. Leahy, K. Bruton, and D. T. J. O’Sullivan, “Big data in manufacturing: a
systematic mapping study,” J. Big Data, vol. 2, no. 1, 2015.
[15] Q. Qi and F. Tao, “Digital Twin and Big Data Towards Smart Manufacturing and Industry 4.0: 360
Degree Comparison,” IEEE Access, vol. 6, pp. 3585–3593, 2018.
[16] M. N. Rahman, A. Esmailpour, and J. Zhao, “Machine Learning with Big Data An Efficient
Electricity Generation Forecasting System,” BIG DATA Res., vol. 5, pp. 9–15, 2016.
[17] J. Wang, Y. Ma, L. Zhang, R. X. Gao, and D. Wu, “Deep learning for smart manufacturing:
Methods and applications,” J. Manuf. Syst., vol. 48, pp. 144–156, Jul. 2018.
[18] G. Dorgo and J. Abonyi, “Learning and predicting operation strategies by sequence mining and
deep learning,” Comput. Chem. Eng., vol. 128, pp. 174–187, Sep. 2019.
[19] P. Lade, R. Ghosh, and S. Srinivasan, “Manufacturing analytics and industrial Internet of Things,”
IEEE Intell. Syst., vol. 32, no. 3, pp. 74–79, 2017.
[20] F. Tao, Q. Qi, A. Liu, and A. Kusiak, “Data-driven smart manufacturing,” J. Manuf. Syst., vol. 48,
pp. 157–169, 2018.

View publication stats