Artificial Intelligence and Machine Learning Applications in Smart Production: Progress, Trends and Directions

The history of Artificial Intelligence (AI) development dates to the 40s. The researchers showed strong expectations until the 70s, when they began to encounter serious difficulties and investments were greatly, reduced. With the introduction of the Industry 4.0, one of the techniques adopted for AI implementation is Machine Learning (ML) that focuses on the machines ability to receive data series and learn on their own. Given the considerable importance of the subject, researchers have completed many studies on ML to ensure that machines are able to replace or relieve human tasks. This research aims to analyze, systematically, the literature on several aspects, including publication year, authors, scientific sector, country, institution, keywords. Analyzing existing literature on AI is a necessary stage to recommend policy on the matter. The analysis has been done using Web of Science and SCOPUS database. Furthermore, UCINET and NVivo 12 software have been used to complete them. Literature review on ML and AI empirical studies published in the last century was carried out to highlight the evolution of the topic before and after Industry 4.0 introduction, from 1999 to now. Eighty-two articles were reviewed and classified. A first interesting result is the greater number of works published by USA and the increasing interest after the birth of Industry 4.0.


Introduction
Smart production systems require innovative solutions to increase the quality and sustainability of manufacturing activities while reducing costs. In this context, artificial intelligence (AI)-driven technologies, leveraged by I4.0 Key Enabling Technologies (e.g., Internet of Thing, advanced embedded systems, cloud computing, big data, cognitive systems, virtual and augmented reality), are ready to generate new industrial paradigms [1].
In this regard, it is interesting to remember that the father of artificial intelligence, John McCarthy [2], in the 1990s, defined artificial intelligence as "artificial intelligence is the science and engineering of making intelligent machines, especially intelligent computer programs". Generally, the term "AI" is used when a machine simulates functions that humans associate with other human minds, such as learning and problem solving [3].
The paper is organized as follows. Section 2 presents the proposed methodology and details the research methodology adopted for the literature survey. Section 3 analyzes the main results of the bibliometric analysis. Finally, in Section 4, the main contribution of the research is summarized.

Methodology
The methodological approach used mixes bibliometric, content analysis, and social network techniques. In this study, a state-of-the-art research was conducted through the SCOPUS and Web of Science databases. For the publication time span, the time from 1999 to 2019 was considered with the intent to understand how the level of attention towards the topic has changed before and after the introduction of Industry 4.0. The research methodology chosen for this study was a systematic literature review [25]. The main phases of the study were as follows: 1.
In phase 1, bibliometric data was collected (step 1). Then, a screening of the overall result was carried out to identify which documents can be taken into consideration, in line with the research areas deemed interesting and relevant (step 2). At the end of this step, the last step (step 3) aimed to select the documents to be analyzed in detail.

2.
Phase 2: Analysis. Once phase 1 was completed, the next phase was phase 2, which was the analysis of the results. The approach used for the bibliometric analysis included: • The use of indicators for the parameters studied; and • SNA (social network analysis) for the keywords.
The indicators chosen to perform the analysis were total papers (TPs), which is the total number of publications, and total citations (TCs), which is the total number of citations.
SNA finds application in various social sciences, and has lately been employed in the study of various phenomena, such as international trade, information dissemination, the study of institutions, and the functioning of organizations. The analysis of the use of the term SNA in the scientific literature has undergone exponential growth in the use of this mode of computable representation of complex and interdependent phenomena. For the purpose of the study, UCINET, NetDraw software was used, which was expressly designed for the creation and graphic processing of networks, and was used to represent the keywords in the network, and Excel for data input.
The software UCINET, NetDraw returned a sociometric network that describes the relationships between the classes, that is, data entered as input.
Furthermore, NVivo 12 software, the leading program for computer-assisted qualitative analysis (CAQDAS), was used to analyze keywords of all documents. In this specific case, it was used to identify the possible links between the keywords of the various documents examined, developing conceptual schemes from which to make interpretative hypotheses.

3.
Phase 3: Discussion. At the end of the second phase, a third and final one followed, where the results were discussed, and conclusions were drawn.
In Figure 1, the main phases and steps followed for the analysis are shown.

Phase 1: Research and Classification
The first phase consisted of the search for documents, which included the activities of collecting the material belonging to the academic universe. This first phase was divided into three steps as follows.

Identification (Step 1)
For a comprehensive survey of the phenomenon, an investigation on the Scopus (SCP) and Web of Science (WoS) databases was carried out using Boolean operators. We began by making a search query on the Scopus and WoS databases with the general keywords "artificial intelligence" AND "machine learning" AND "application", as shown in Table 1.
In order to maintain the consistency of the results, the same keywords were used in both databases and a time horizon of 20 years was chosen, from 1999 to 2019.
The choice of keywords for performing the survey was based on the awareness that AI and ML can be an important tool in the effort to adopt responsible business practices in the context of smart production. In this regard, it is worthy to note that with the increasingly urgent discussions of climate change, it seemed appropriate to focus our research on the topic of sustainability. Thus, the selection of papers also considered applications on sustainability.

Phase 1: Research and Classification
The first phase consisted of the search for documents, which included the activities of collecting the material belonging to the academic universe. This first phase was divided into three steps as follows.

Identification (Step 1)
For a comprehensive survey of the phenomenon, an investigation on the Scopus (SCP) and Web of Science (WoS) databases was carried out using Boolean operators. We began by making a search query on the Scopus and WoS databases with the general keywords "artificial intelligence" AND "machine learning" AND "application", as shown in Table 1.
In order to maintain the consistency of the results, the same keywords were used in both databases and a time horizon of 20 years was chosen, from 1999 to 2019.
The choice of keywords for performing the survey was based on the awareness that AI and ML can be an important tool in the effort to adopt responsible business practices in the context of smart production. In this regard, it is worthy to note that with the increasingly urgent discussions of climate change, it seemed appropriate to focus our research on the topic of sustainability. Thus, the selection of papers also considered applications on sustainability.  The result is not entirely unexpected, and the reason is to be found in the fact that Scopus, being an Elsevier product, collects data from all the other databases, in particular Science Direct and those queried by the Scirus search engine, while Web of Science (WoS) collects fewer documents.
From the documents extracted in Scopus, it was found that most of them are conference papers (57.28%) and, subsequently, articles (33.85%).
On the contrary, the research on Web of Science (WoS) underlines that most of the documents are articles (46.12%) and, subsequently, proceedings papers (42.86%).
All the document types are filled in Table 3. AI began working in the 1940s and researchers showed strong expectations until the 1970s when they began to encounter serious difficulties and investments were greatly reduced.
Since then, a long period began, known as the "AI winter" [26]: Despite some great successes, such as IBM's Deep Blue system, which in the late 1990s defeated the then chess world champion Garri Kasparov, the study of solutions for AI has only come back for a few years. The push for a new technological development has been given by the I4.0, which considered AI as one of the primary key enabling technologies (KETs).
From this period onwards, the literature has been enriched with documents, as shown in Figure 2. Growth is apparent after 2011 when new technologies began to be implemented more frequently. In fact, the Industry 4.0 term first appeared at Hannover Messe in 2011 when Professor Wolfgang Subsequently, the increase in the adoption of these ones has led researchers to keep pace with the growth of I4.0 [27].

Screening (Step 2)
Trying to give an overview of the topics and areas interface, in the screening phase, an analysis of documents characterized by free access was chosen, excluding those that have restrictions, and to restrict the field to the thematic areas of scientific interest.
With this in mind, the number of open access items has been drastically reduced (1288 results for Scopus and 149 for WoS) and, also applying the filter related to the thematic areas (Table 4), it determined a further reduction: 947 for Scopus and 60 for WoS. Note how the number of filters applied is different. The databases, in fact, offer the same search options, but, in the specific case of the thematic areas, the latter are more numerous and structured on Web of Science (WoS) compared to Scopus. Subsequently, the increase in the adoption of these ones has led researchers to keep pace with the growth of I4.0 [27].

Screening (Step 2)
Trying to give an overview of the topics and areas interface, in the screening phase, an analysis of documents characterized by free access was chosen, excluding those that have restrictions, and to restrict the field to the thematic areas of scientific interest.
With this in mind, the number of open access items has been drastically reduced (1288 results for Scopus and 149 for WoS) and, also applying the filter related to the thematic areas (Table 4), it determined a further reduction: 947 for Scopus and 60 for WoS. Note how the number of filters applied is different. The databases, in fact, offer the same search options, but, in the specific case of the thematic areas, the latter are more numerous and structured on Web of Science (WoS) compared to Scopus.

Inclusion (Step 3)
At the end of the screening process, the inclusion step was started, which consisted in the selection of documents, which was extracted from the last passage, destined to be included in the sample on which bibliometric analysis was performed. In this review step, for the purposes of eligibility, we examined the complete text of each document independently. For each article, we examined whether there was interest from the academic world, and if it contained case studies or real applications, proposals for new AI and ML algorithms, or possible future scenarios.
Therefore, the final sample to be analyzed consisted of 60 documents for Scopus and 22 for WoS.

Phase 2: Analysis
This section presents and discusses the findings of this review. First, an overview of the selected studies is presented. Second, the review findings according to the research criteria, one by one in the separate subsections, are reported.

Top Highly Influential Analysis
This section lists the most highly cited documents in WoS and Scopus. The list is structured by research source, date, title, authors, source title, and top citation (TP) in WoS or Scopus, according to the research source. The whole list is available in the Appendix A. Looking into the Appendix A, it is possible underline that the document by Larrañaga, Calvo, Santana et al. in 2006 [28] has the highest citation count of 298. This article reviews machine learning methods for bioinformatics and it presents modelling methods. Moreover, the document year is 2006, so before I4.0 was introduced. Therefore, having more years than today has an advantage in terms of diffusion. This means that it is one of the most influential documents in the academic world, as it proposes some of the most useful techniques for modelling, giving the document the opportunity to become a pioneer in the computer science research area.
Obviously, all documents before I4.0, in general, have more citations than the most recent documents. However, it is significant to note that even recent documents have a very high number of citations compared to the year of publication. This denotes the interest in the topic from the scientific community.
The citation analysis revealed that the first article that we can identify among the most cited in the I4.0 period dates to 2016. The work, published by Krawczyk [29], proposes application models to further develop the field of unbalanced learning, to focus on computationally effective, adaptive, and real-time methods, and provides a discussion and suggestions on the lines of future research in the application subject of the study. It received 119 citations. Moreover, an article published by Wuest, Weimer, Irgens et al. [30] received much attention among the scientific community. It contributes by presenting an overview of the available machine learning techniques.
Finally, the citation analysis pointed out that the average number of citations of all documents is 16.58. This value is expected to increase rapidly considering the interest in the issues of ML and AI.

Publications by Years
Consistent with what is defined in Section 3.1.1., the study shows that the number of items included in the analysis is definitely low for the entire period before I4.0 and then suddenly increases, starting in 2012. The data shown in Figure   With reference to 2019, the figure refers to the first months of the year, so it is plausible that during the year, there will be a further increase in the documents in the literature. Furthermore, an increase is expected in the coming years, in parallel with the growth of I4.0

Most Collaborative Authors
The analysis highlighted that most of publications have more than one author. From this point of view, it is possible to identify the number of authors for each document. As shown in Figure 4, most of the manuscripts were produced by groups ranging from two to five authors. The indicators chosen to perform the analysis were total papers (TPs), which is the total number of publications.  With reference to 2019, the figure refers to the first months of the year, so it is plausible that during the year, there will be a further increase in the documents in the literature. Furthermore, an increase is expected in the coming years, in parallel with the growth of I4.0

Most Collaborative Authors
The analysis highlighted that most of publications have more than one author. From this point of view, it is possible to identify the number of authors for each document. As shown in Figure 4, most of the manuscripts were produced by groups ranging from two to five authors. The indicators chosen to perform the analysis were total papers (TPs), which is the total number of publications. With reference to 2019, the figure refers to the first months of the year, so it is plausible that during the year, there will be a further increase in the documents in the literature. Furthermore, an increase is expected in the coming years, in parallel with the growth of I4.0

Most Collaborative Authors
The analysis highlighted that most of publications have more than one author. From this point of view, it is possible to identify the number of authors for each document. As shown in Figure 4, most of the manuscripts were produced by groups ranging from two to five authors. The indicators chosen to perform the analysis were total papers (TPs), which is the total number of publications.

Research Areas Analysis
The total research area analysis collected from the 82 papers was 164 because each paper can be considered as more than one research area analysis. Given the small number of documents identified in the period before I4.0, the ranking refers mostly to the current industrial revolution. Also, in this case, the result is consistent with the introduction of paradigm 4.0, which has intensified research and the adoption of technology.
The first thematic areas and disciplines that are at the top of the ranking are computer science, engineering and biochemistry, genetics, and molecular Biology, respectively, with 29%, 23%, and 6% of publications. Furthermore, the other disciplines identified for which applicative findings are found are considered transversal to the first three disciplines and this is a consequence of I4.0. In terms of the percentage contribution, the first three areas cover about 60% of the papers considered.
Considering the top 20 research areas, given the frequency of the research areas' distribution, Figure 5 shows a higher level of concentration in the disciplines indicated above.
The total research area analysis collected from the 82 papers was 164 because each paper can be considered as more than one research area analysis. Given the small number of documents identified in the period before I4.0, the ranking refers mostly to the current industrial revolution. Also, in this case, the result is consistent with the introduction of paradigm 4.0, which has intensified research and the adoption of technology.
The first thematic areas and disciplines that are at the top of the ranking are computer science, engineering and biochemistry, genetics, and molecular Biology, respectively, with 29%, 23%, and 6% of publications. Furthermore, the other disciplines identified for which applicative findings are found are considered transversal to the first three disciplines and this is a consequence of I4.0. In terms of the percentage contribution, the first three areas cover about 60% of the papers considered.
Considering the top 20 research areas, given the frequency of the research areas' distribution, Figure 5 shows a higher level of concentration in the disciplines indicated above.
In fact, in terms of the percentage contribution, the first five areas cover about 70% of the papers considered. Regardless, by only counting research areas found once, there is a total of 27.
This means two things: • The large number of fields in which this kind of research is involved; and • Most papers have a transversal approach, that is, the object of each research crosses more than one field of application, thus involving more research areas.
This confirms the wide interest in these subjects from several fields.

Top Source Journals Analysis
In this section, the top 20 sources or journals that were published most frequently were extracted. A journal is a time-bound publication with the objective of promoting and monitoring the progress of the discipline it represents.
In this specific case, the total source journals detected from the documents is 74, but, considering the top 20, given the frequency of the source journals' distribution, only the first 13 sources have more than one paper published, with a total percentage contribution of 43% of the total.
After analyzing the sources separately, the results obtained in the two databases were found to not be the same. In WoS, the top source journal was IEEE Access with two publications while in Scopus, the top source journals are Procedia Computer Science, Matec Web of Conferences, and Machine Learning with four publications, which contribute 5% of the total. In fact, in terms of the percentage contribution, the first five areas cover about 70% of the papers considered. Regardless, by only counting research areas found once, there is a total of 27.
This means two things: • The large number of fields in which this kind of research is involved; and • Most papers have a transversal approach, that is, the object of each research crosses more than one field of application, thus involving more research areas.
This confirms the wide interest in these subjects from several fields.

Top Source Journals Analysis
In this section, the top 20 sources or journals that were published most frequently were extracted. A journal is a time-bound publication with the objective of promoting and monitoring the progress of the discipline it represents.
In this specific case, the total source journals detected from the documents is 74, but, considering the top 20, given the frequency of the source journals' distribution, only the first 13 sources have more than one paper published, with a total percentage contribution of 43% of the total.
After analyzing the sources separately, the results obtained in the two databases were found to not be the same. In WoS, the top source journal was IEEE Access with two publications while in Scopus, the top source journals are Procedia Computer Science, Matec Web of Conferences, and Machine Learning with four publications, which contribute 5% of the total.
Aggregating the data collected from the two databases, the ranking moves to that obtained by Scopus, making sure that IEEE Access is no longer first in the standings, but only eighth, and that the former are precisely those of Scopus: Procedia Computer Science, Matec Web Of Conferences, and Machine Learning, with the same number of publications. Next, the 10 source journals have a 3% publication contribution while the rest have a one-to-one relationship (1%) with the corresponding source journal.
The low level of concentration of the sources suggests that there is a great deal of interest in these topics from several scientific journals. As a matter of fact, it is foreseeable that specialized sector sources (AI Magazine and Machine Learning) are among the first 13; however, it is interesting to note that other sources are involved, such as Sustainability Switzerland or BMC Bioinformatics and Nuclear Engineering and Design. Figure 6 shows the top 20 source journals contributions.
Sustainability 2020, 12, x FOR PEER REVIEW 10 of 24 Aggregating the data collected from the two databases, the ranking moves to that obtained by Scopus, making sure that IEEE Access is no longer first in the standings, but only eighth, and that the former are precisely those of Scopus: Procedia Computer Science, Matec Web Of Conferences, and Machine Learning, with the same number of publications. Next, the 10 source journals have a 3% publication contribution while the rest have a one-to-one relationship (1%) with the corresponding source journal.
The low level of concentration of the sources suggests that there is a great deal of interest in these topics from several scientific journals. As a matter of fact, it is foreseeable that specialized sector sources (AI Magazine and Machine Learning) are among the first 13; however, it is interesting to note that other sources are involved, such as Sustainability Switzerland or BMC Bioinformatics and Nuclear Engineering and Design. Figure 6 shows the top 20 source journals contributions.

Country Analysis
The results that emerged through research on the two databases are consistent with each other. In both cases, in fact, the countries that give the greatest contribution to the research are China and the United States (Figure 8). The result is obvious since in China and the United States, more than 1.3 billion and 0.3 millions of people live, respectively, and so there are more researchers than in the single European nations. Focusing on Europe, Germany published more papers than any other European country. This is not a random result: I4.0 was born in Germany, so this outcome was expected. However, the following observation cannot be ignored from this data: The USA and China carry the first two places in the list while it is not the same for European countries. Europe, despite its talents and resources, has lost ground. Presenting its report on artificial intelligence, the French deputy and mathematician Cédric Villani declared that, "Europe must be able to compete with China and the United States while protecting its citizens and pointing the way to go on ethical issues". If we are not careful, the 21st century rules will not be defined in Brussels, but in Shanghai. Artificial intelligence is also a land marked by intense geopolitical rivalry that could redefine global power relations.

Country Analysis
The results that emerged through research on the two databases are consistent with each other. In both cases, in fact, the countries that give the greatest contribution to the research are China and the United States (Figure 8). The result is obvious since in China and the United States, more than 1.3 billion and 0.3 millions of people live, respectively, and so there are more researchers than in the single European nations. Focusing on Europe, Germany published more papers than any other European country. This is not a random result: I4.0 was born in Germany, so this outcome was expected. However, the following observation cannot be ignored from this data: The USA and China carry the first two places in the list while it is not the same for European countries. Europe, despite its talents and resources, has lost ground. Presenting its report on artificial intelligence, the French deputy and mathematician Cédric Villani declared that, "Europe must be able to compete with China and the United States while protecting its citizens and pointing the way to go on ethical issues". If we are not careful, the 21st century rules will not be defined in Brussels, but in Shanghai. Artificial intelligence is also a land marked by intense geopolitical rivalry that could redefine global power relations.
Even so, regarding Europe, it is worthy to also note that since 2017, France, Germany, and Italy have intensified their trilateral cooperation to promote digitizing the manufacturing industry. In this regard, in the near future, we expect a significant evolution of smart production initiatives and therefore an increase in scientific research. Figure 7 shows the country contribution distribution. Even so, regarding Europe, it is worthy to also note that since 2017, France, Germany, and Italy have intensified their trilateral cooperation to promote digitizing the manufacturing industry. In this regard, in the near future, we expect a significant evolution of smart production initiatives and therefore an increase in scientific research. Figure 7 shows the country contribution distribution.

Affiliation Analysis
The total number of affiliation detected from the 82 papers is 153. Also, in this case, considering the top 20, the frequency of the affiliation distribution shows that most papers have a one-to-one relationship with the corresponding affiliation. Only the first four affiliations have three papers (2% of the contribution) and the second four have two papers (1.3% of the contribution). This result gives us information about the wide interest on this subject from several universities and research centers all over the world. Then, the affiliation analysis confirms the result of the country analysis ( Figure 8). In fact, if we try to sum the first eight affiliations by their own country, the outcome is: • Nine papers from China; • Six papers from Germany; and • Five papers from the USA.
In September 2018, the most important event on artificial intelligence was held in Shanghai. China is very determined to focus on future technologies.
For some months, China has become the world's leading power in terms of scientific publications. Late in the 20th century technologies, China chose to do what the English-speaking people call a "frog jump" and focus on 21st century technologies.
China, with its 800 million Internet users and without any privacy protection policy, has access to more personal data than the United States and Europe.

Affiliation Analysis
The total number of affiliation detected from the 82 papers is 153. Also, in this case, considering the top 20, the frequency of the affiliation distribution shows that most papers have a one-to-one relationship with the corresponding affiliation. Only the first four affiliations have three papers (2% of the contribution) and the second four have two papers (1.3% of the contribution). This result gives us information about the wide interest on this subject from several universities and research centers all over the world. Then, the affiliation analysis confirms the result of the country analysis ( Figure 8). In fact, if we try to sum the first eight affiliations by their own country, the outcome is: • Nine papers from China; • Six papers from Germany; and • Five papers from the USA.
In September 2018, the most important event on artificial intelligence was held in Shanghai. China is very determined to focus on future technologies.
For some months, China has become the world's leading power in terms of scientific publications. Late in the 20th century technologies, China chose to do what the English-speaking people call a "frog jump" and focus on 21st century technologies.
China, with its 800 million Internet users and without any privacy protection policy, has access to more personal data than the United States and Europe. Sustainability 2020, 12, x FOR PEER REVIEW 12 of 24

Top Keywords Analysis
Through NVivo 12, the top 20 keywords were extracted directly, which are those that always appear in association with each document.
Starting from this classification, the graphic representation, a word cloud shape, of the keywords (Figure 9) was extracted. It can be noted that the most used term is precisely "machine", "learning", and "intelligence", which the software represents with greater characters than all the other terms. The font size describes how much the keyword is indexed. Another mode of representation is the tree words ( Figure 10). Also, in this case, the most indexed words are those represented in the larger boxes.

Top Keywords Analysis
Through NVivo 12, the top 20 keywords were extracted directly, which are those that always appear in association with each document.
Starting from this classification, the graphic representation, a word cloud shape, of the keywords (Figure 9) was extracted. It can be noted that the most used term is precisely "machine", "learning", and "intelligence", which the software represents with greater characters than all the other terms.

Top Keywords Analysis
Through NVivo 12, the top 20 keywords were extracted directly, which are those that always appear in association with each document.
Starting from this classification, the graphic representation, a word cloud shape, of the keywords (Figure 9) was extracted. It can be noted that the most used term is precisely "machine", "learning", and "intelligence", which the software represents with greater characters than all the other terms. The font size describes how much the keyword is indexed. Another mode of representation is the tree words ( Figure 10). Also, in this case, the most indexed words are those represented in the larger boxes. The font size describes how much the keyword is indexed. Another mode of representation is the tree words ( Figure 10). Also, in this case, the most indexed words are those represented in the larger boxes. Sustainability 2020, 12, x FOR PEER REVIEW 13 of 24 As expected, the most indexed words are obviously "learning", "machine", and "intelligence", with high numbers. It is logical that among the first results, words that recall the technology itself were obtained, but it is interesting to note that words referring to other fields of AI applications are also indexed. The reason is to be found in the fact that AI and ML are technologies that cross all the sectors involved in I4.0 and that, therefore, do not remain circumscribed.
Specifically, words, such as "data", "neural", "decision", and "management", are very or average indexed, demonstrating the fact that AI also extends to many other sectors.
Another tool for the analysis for keywords is the UCINET software, through which social networks analysis is carried out.
Social network analysis (SNA), which is also often called social network theory, is a modern technology of social relations.
SNA finds application in various social sciences, and has recently been used in the study of various phenomena, such as international trade, information dissemination, the study of institutions, and the functioning of organizations. The analysis of the use of the term SNA in the scientific literature shows that in the last five years, there has been exponential growth of the use of this mode of computable representation of complex and interdependent phenomena. The software returns a graph representing a socio-metric network (Figure 11), which draws the relationships that exist within the class. Each relationship is represented by an oriented arrow. As expected, the most indexed words are obviously "learning", "machine", and "intelligence", with high numbers. It is logical that among the first results, words that recall the technology itself were obtained, but it is interesting to note that words referring to other fields of AI applications are also indexed. The reason is to be found in the fact that AI and ML are technologies that cross all the sectors involved in I4.0 and that, therefore, do not remain circumscribed.
Specifically, words, such as "data", "neural", "decision", and "management", are very or average indexed, demonstrating the fact that AI also extends to many other sectors.
Another tool for the analysis for keywords is the UCINET software, through which social networks analysis is carried out.
Social network analysis (SNA), which is also often called social network theory, is a modern technology of social relations.
SNA finds application in various social sciences, and has recently been used in the study of various phenomena, such as international trade, information dissemination, the study of institutions, and the functioning of organizations. The analysis of the use of the term SNA in the scientific literature shows that in the last five years, there has been exponential growth of the use of this mode of computable representation of complex and interdependent phenomena. The software returns a graph representing a socio-metric network (Figure 11), which draws the relationships that exist within the class. Each relationship is represented by an oriented arrow. Sustainability 2020, 12, x FOR PEER REVIEW 14 of 24 Figure 11. Keywords Network by UCINET Software.
In Figure 11, nodes and leaves can be identified. The nodes are represented by red circles and are correspond to the most common keywords, where the words "machine", "learning", "artificial", and "intelligence" have been united to form the key words "machine learning" and "artificial intelligence".
The leaves, on the other hand, are represented by blue squares and correspond to the articles. To facilitate reading, the document titles were not inserted, but the (Identification) ID count for each of them is shown in the Appendix A.
The first thing that can be noticed is the isolation of many leaves that are not connected to the nodes. This means that the corresponding documents are not described by the keywords represented by the nodes. Really, they are characterized by keywords that have a frequency of the order of units.
Another thing that easily jumps to the eye is a density that is larger around the keywords "machine learning", "decision", "data", "algorithm", "system", "artificial intelligence", "method", and "optimization". This density is reflected in the cloud and the box chart produced by NVivo 12. Therefore, we can say that those are the words that most often appear in the documents analyzed, emphasizing, once again, that they include terms that do not just refer to the technology object of study but also to other fields of application.

Benefits of Artificial Intelligence and Machine Learning in Industrial Contexts
From the analysis of the research carried out, the first information that emerged is that there is a growing importance of innovation and digitalization in products, services, and processes. Consequently, it means that the adoption of advanced manufacturing technologies, such AI and ML, is an emerging issue. In other words, AI/ML algorithms represent an opportunity to handle high dimensional problems and data. The interest in the subject is extended to all scientific sectors, but with a focus on computer science and engineering.
The most significant benefits of using AI and ML in industrial sectors include: 1) Greater innovation, 2) process optimization, 3) resources optimization, and 4) improved quality.
After all, AI with ML is one of the most important technologies today and is transforming the economy and society, as demonstrated by the over 340,000 patent applications filed since the 1950s. In Figure 11, nodes and leaves can be identified. The nodes are represented by red circles and are correspond to the most common keywords, where the words "machine", "learning", "artificial", and "intelligence" have been united to form the key words "machine learning" and "artificial intelligence".
The leaves, on the other hand, are represented by blue squares and correspond to the articles. To facilitate reading, the document titles were not inserted, but the (Identification) ID count for each of them is shown in the Appendix A.
The first thing that can be noticed is the isolation of many leaves that are not connected to the nodes. This means that the corresponding documents are not described by the keywords represented by the nodes. Really, they are characterized by keywords that have a frequency of the order of units.
Another thing that easily jumps to the eye is a density that is larger around the keywords "machine learning", "decision", "data", "algorithm", "system", "artificial intelligence", "method", and "optimization". This density is reflected in the cloud and the box chart produced by NVivo 12. Therefore, we can say that those are the words that most often appear in the documents analyzed, emphasizing, once again, that they include terms that do not just refer to the technology object of study but also to other fields of application.

Benefits of Artificial Intelligence and Machine Learning in Industrial Contexts
From the analysis of the research carried out, the first information that emerged is that there is a growing importance of innovation and digitalization in products, services, and processes. Consequently, it means that the adoption of advanced manufacturing technologies, such AI and ML, is an emerging issue. In other words, AI/ML algorithms represent an opportunity to handle high dimensional problems and data. The interest in the subject is extended to all scientific sectors, but with a focus on computer science and engineering.
The most significant benefits of using AI and ML in industrial sectors include: (1) Greater innovation, (2) process optimization, (3) resources optimization, and (4) improved quality.
After all, AI with ML is one of the most important technologies today and is transforming the economy and society, as demonstrated by the over 340,000 patent applications filed since the 1950s.
Other information that emerged is about the authors and affiliation. Many of these are in a 1:1 ratio compared to the selected documents and this supports the fact that there is no interest in technological applications in one direction, but that, once again, the interest is very wide in the scientific community.
Furthermore, it can be said that the countries most interested in scientific research are the USA, China and European countries. This result is not a surprise.
In terms of investment, the effort currently being deployed by the United States and China to acquire dominance in the AI sector is far superior to that of other countries. More specifically, China has clearly stated its ambition to become a world leader in AI by 2030 [31]. Among the Chinese plans, of absolute interest is the "Made in China 2025" plan, dedicated to the manufacturing sector; the "Internet +" plan is also dedicated to smart manufacturing and innovation.
A direct consequence of the above considerations could be having new generations of researchers who will contribute to future comparisons, accompanied by new questions for investigations.

Emerging Trends of Artificial Intelligence and Machine Learning in Sustainable Manufacturing
From the perspective of sustainability, the analysis highlighted that the new paradigm of smart manufacturing has the potential to bring fundamental improvements in the industry by addressing the issue of scarce resources and improving productivity.
In fact, the survey pointed out a growing interest on applications related to green manufacturing and sustainable development, proving that AI/ML play an important role in increasing sustainability through the intelligent utilization of materials and energy consumption (i.e., reduction of energy consumption and pollutant emissions, environmental footprint monitoring and evaluation, etc.).
Furthermore, it emerged that AI/ML algorithms present a wide array of applications that provide an opportunity for sustainable development, which will involve several stakeholders from different countries and sectors, including inventory and supply chain management, predictive maintenance, and production.
In particular, Pérez-Ortiz, Jiménez-Fernández, Gutiérrez et al. [32] reviewed the most important classification algorithms applied to renewable energy (RE) problems. The main use of algorithms is as a tool for predictive analysis and consequently for data preprocessing, result interpretation, or evaluation in order to improve energy and resource management.
In this context, it also emerged that AI/ML have been successfully utilized in various processes' optimization, applications in manufacturing, and predictive maintenance in different industries.
The work published by Lieber, Stolpe, Konrad et al. [33] represents a good research within steel industry production. It proposes an approach for automatically preprocessing value series data to improve the quality of the process and products. It means that AI/ML techniques were found to provide promising potential for improved quality control optimization in manufacturing systems.
Appropriate adoption of AI/ML technologies will promote sustainable manufacturing and the formation of a new generation of intelligent manufacturing, including all areas that characterize a sustainable process, ranging from the supply chain management to quality control, to predictive maintenance, to energy consumption. Table 5 summarizes the main areas in sustainable manufacturing, their respective key objectives, and the main AI/ML applications.
However, the relationship between I.4 technologies, AI/ML, and sustainability demands a more conceptual and empirical investigation. This is corroborated by an article recently published in Nature Sustainability by the director of the Earth Institute at Columbia University, Jeffrey Sachs, and other experts, and the so-called Fourth Industrial Revolution (made of artificial intelligence and other digital technologies) is even cited as one of the six transformations necessary to achieve the sustainable development goals [34].

Supply Chain Management
Ready product available in the appropriate place at a specific time Improves transparency, accelerates decision-making, and produces accurate demand forecasting

Quality Control
Recognize the early signs of potential production failures within the shortest terms in order to save resources and sustain operational efficiency Improves the response time and allows eliminating possible failures Predictive Maintenance Detects possible production malfunctions that may cause product quality issues Creates accurate forecasts as to when the machinery must be repaired Energy consumption Recommendations that will strike a balance in energy use Improves excessive use of certain materials, redundant production scrap waste, inefficient supply chain management, logistics, and unequal distribution of energy resources.

Conclusions
This research focused on the study of the state of the art of AI and ML applications, selecting literature on what has now become a particularly hot topic in scientific research. The literature available on any subject is now wide and a complete coverage of all the documents published with respect to a particular topic can be challenging or even impossible. Therefore, a systematic selection of the most relevant literature was implemented. This document provides a systematic review of applications in various scientific fields using ML techniques. For the selection of documents, objective and clear methods of investigation were used, independent of the experience of the researchers. Among the objectives of the document, it aimed to not only provide a comprehensive framework on the literature on the research of AI and ML but also a starting point for integrating knowledge through research in this area and to suggest future research paths. It is important to underline that this document was produced using only two databases, i.e., WoS and Scopus, in which only documents with open access were included. There are, therefore, many other documents with restricted access and other indexing databases, such as Google Scholar, that could be integrated for future research.
Author Contributions: All authors contributed equally to this work. All authors have read and agreed to the published version of the manuscript.