Understanding Big Data Analytics for Manufacturing Processes: Insights from Literature Review and Multiple Case Studies

Journal Pre-proofs Understanding the capabilities of Big Data Analytics for manufacturing process: insights from literature review and multiple case study Amine Belhadi, Karim Zkik, Anass Cherrafi, Yusof M. Sha'ri, Said El fezazi PII: DOI: Reference:

S0360-8352(19)30568-6 https://doi.org/10.1016/j.cie.2019.106099 CAIE 106099

To appear in:

Computers & Industrial Engineering

Received Date: Revised Date: Accepted Date:

28 October 2018 3 July 2019 27 September 2019

Please cite this article as: Belhadi, A., Zkik, K., Cherrafi, A., Sha'ri, Y.M., El fezazi, S., Understanding the capabilities of Big Data Analytics for manufacturing process: insights from literature review and multiple case study, Computers & Industrial Engineering (2019), doi: https://doi.org/10.1016/j.cie.2019.106099

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2019 Published by Elsevier Ltd.

Understanding the capabilities of Big Data Analytics for manufacturing process: insights from literature review and multiple case study

Amine Belhadi* Cadi Ayyad University, Marrakech, Morocco E-Mail: [email protected] Karim Zkik International University of Rabat, Morocco Rabat, Morocco

Anass Cherrafi ENSAM- Meknes Moulay Ismail University Meknes, Morocco Yusof M. Sha’ri University of Technology of Malaysia Kuala Lumpur, Malaysia

Said El fezazi Cadi Ayyad University, Marrakech, Morocco * Corresponding Author Secondary E-mail: [email protected] Address: Route Dar Si-Aïssa BP. 89 Avenue Echahid Mbarek El Mokhtar, Safi

Understanding the capabilities of Big Data Analytics for manufacturing process: insights from literature review and multiple case study

Highlights 

Review of the ongoing research on BDA in manufacturing process,



Proposition of a framework grouping BDA capabilities in manufacturing process,



The proposed framework enables to identify tree layers of capabilities.



Identification of research trends in implementing BDA in manufacturing process



Multiple case study of the application of BDA in real-life context of the manufacturing process of a leading manufacturers of phosphate derivatives,



Drawing a set of recommendations based on the findings of the multiple case study.

Abstract Today, we are undoubtedly in the era of data. Big Data Analytics (BDA) is no longer a perspective for all level of the organization. This is of special interest in the manufacturing process with their high capital intensity, time constraints and given the huge amount of data already captured. However, there is a paucity in past literature on BDA to develop better understanding of the capabilities of the strategic implications to extract value from BDA. In that vein, the central aim of this paper is to develop a novel model that summarizes the main capabilities of BDA in the context of manufacturing process. This is carried out by relying on the findings of a review of the ongoing research along with a multiple case study within a leading phosphates derivatives manufacturer to point out the capabilities of BDA in manufacturing process and outline recommendations to advance research in the field. The findings will help companies to understand the big data analytics capabilities and its potential implications for their manufacturing process and support them seeking to design more effective BDA-enabler infrastructure. Keywords: Big Data Analytics; Manufacturing process; Big Data Analytics capabilities; business intelligence; literature review; multiple case study

1. Introduction The widespread emergence of digital technologies and advancing computing power and expansion of the industrial Internet of Things (IoT) have led to a new generation of networked, information-based technologies, data analytics, and predictive modeling (He & Wang, 2018). This new generation is providing unparalleled integrated computing capabilities to supply manufacturers with better wherewithal to extract value from an increasingly huge amount of data and gain a powerful competitive advantage (Chiang, et al., 2017; He & Wang, 2018). According to the research report published in the earlier 2011 by McKinsey Global Institute under the heading “Big Data: The next frontier for innovation, competition, and productivity”, big data has permeated every part of life, and becomes a paramount pioneer for production in the near future (Manyika, et al., 2011). According to Manyika, et al. (2011), Big Data involves datasets whose size exceeds the ability of typical database software tools to capture, store, manage, and analyze. Mainly, Big Data is characterized by the ability to handle data with four qualities: Volume (the size/scale of the data), Variety (the form/format of the data), Velocity (the rate of the data being produced), and Veracity (the uncertainty/reliability of the data). Big Data is often associated with the concept of Analytics which refers to the ability to acquire information from data by applying statistics, mathematics, econometrics, simulations, optimizations, or other techniques to support decision making processes (Arunachalam, et al., 2018; Wang, et al., 2016). Particularly for manufacturing processes, the challenge of Big Data Analytics (BDA) is even greater. Actually, the excessive use of process operation, control computers and information systems makes the existing manufacturing process operation databases huge and massive. Moreover, with the ever-increasing advancement of IoT devices from conventional process sensors to images, videos and indirect measurement technologies, it is expected that the data extracted from future smart manufacturing processes will expand drastically (Qin, 2014). Therefore, it seems to be a consensus that only manufacturers able to analyze their manufacturing processes based on this accelerating huge mass of data will survive in the next stage of the transformation of advanced manufacturing within the age of data as a decisive competitive asset. Such manufacturers predict the best proceeding process flow, and proactively control their processes with this knowledge (Krumeich, et al., 2014). Owing to the aforementioned considerations, Big Data Analytics in manufacturing processes has received increasing attention because of its considerable impact on manufacturing processes. First, the wide use of distributed control systems and the development of some typical information and communication

technologies (ICTs) has considerably evolved the mode of production. Today’s manufacturing processes are increasingly operating in an uncertain and complex environment with tricky operations and overcomplicated constraints (Cheng, et al., 2018). Therefore, it becomes more and more difficult to build first-principle models in those complex processes and plenty of processes and practices satisfying the traditional mode of production management are no longer appropriate (Cheng, et al., 2018; Ge, et al., 2017). Second, the immense need of production managers for the real-time, dynamic, self-adaptive and accurate production management has brought new challenges to the traditional methods. It becomes highly required to create manufacturing intelligence from real-time data to provide precise prediction of product quality, production and processing time. This is done by new effective techniques, within shorter computation time to control the continuous real-time production systems and to identify faults, defects and some other abnormal situations alongside supporting accurate and timely decision-making (Cheng, et al., 2018; He & Wang, 2018). Notwithstanding the enthusiasm and growth of interest in Big Data Analytics, little is known about their key capabilities for manufacturing processes. Indeed, organizations willing to adopt Big Data in their manufacturing processes are fighting to better understand its concept and then gain the business value from BDA (Wamba, et al., 2015). Moreover, quite few scholars highlight that BDA is still in its early stage and there are yet undiscovered directions to explore on BDA in manufacturing processes. To bridge the existing knowledge gap in the literature, the present paper aims at relying on previous studies on BDA in manufacturing processes along with an in-depth multiple case study within an international company using Big Data Analytics to improve its manufacturing process to achieve the following research objectives:

1. Clear up the concept of BDA in the concept of manufacturing processes, 2. Review, classify and summarize all relevant articles dealing with BDA in manufacturing processes drawing on a conceptual framework for classifying the literature,

3. Point out future research trends to enhance the capabilities of BDA in the manufacturing process, 4. Analyze in-depth the findings of a multiple case study within a leading company to provide recommendations to advance BDA implementation in manufacturing process. The remainder of this paper is arranged as follows. Section 2 provides an integrative definition of the concept of Big Data Analytics. The research scope along with the methodology perused to conduct the study are then introduced in Section 3, followed by Section 4 and Section 5, which present and deeply discuss the findings and results of the literature review and the multiple case study. Section 6 points out the implications for research, practice, and contributions of the study. Finally, the paper is concluded by the summary, limitations and suggestions for future research agenda in Section 6.

2. The concept and definitions of Big Data and Analytics Big Data Analytics is undoubtedly a major thoroughfare in the next round of information technology transformation in industry (Wamba, et al., 2015; Qin, 2014). Obviously, BDA involves two main concepts: Big Data and Data Analytics. Literature has widely discussed the benefits and outcomes of BDA in the growth and profitability of today’s companies (Gunasekaran, et al., 2018; Qin, 2014; Krumeich, et al., 2014). However, the fast-paced evolution of the concept of BDA has raised some confusion regarding its definition (Wamba, et al., 2015). Actually, there is no consensus on a clear and integrative definition of this compound concept. 2.1. Defining Big Data When attempting to define Big Data, size is the immediate attribute that intuitively comes to mind (Gandomi & Haider, 2015). However, there are other important attributes of big data, namely data variety and data velocity. The Three V’s (Volume, Variety and Velocity) constituted, erstwhile, the most commonly used framework to define big data (Chen, et al., 2012). Quite few definitions of ‘Big data’ are summarized in Wamba et al. (2015). Moreover, Beyer & Laney (2012, p. 2) propose a largely convergent definition of BD from Gartner using the three V’s as: ‘‘high-volume, -velocity and -variety information assets that demand cost-effective, innovative forms of information processing for enhanced insight and decision making” Over the time, some other attributes have been included to the basic framework such as Veracity, Variability, Volatility and Value (Tewari & Dwivedi, 2019 ). The Seven V’s can be defined as: 

Volume: refers to the ever-growing magnitude and size of data generated. Big data sizes easily reach multiple terabytes even petabytes.



Variety: represents the heterogeneity, diversity and unevenness of data types in a dataset. Advanced ICTs in today’s companies generate various types of structured, semi-structured, and unstructured data of various types such as text, sensor data, audio, video, log files and so on.



Velocity: is the frequency of data generation and the high-speed at which it should be processed, analyzed and acted upon.



Veracity: introduced by IBM as a defining attribute of big data. Veracity refers the degree of truthfulness and uncertainty related to most sources of data. Big Data proposes the use of specific tools and analytics to deal with imprecise and unreliable data.



Variability: pioneered by SAS as an additional attribute of big data. The Variability and complexity in the process of data generation implies a high need to connect, match, cleanse and transform data received from different sources.



Volatility: refers to the capacity of storage and retention of data. With the huge volume and velocity of data, the issue of storage retention along with security of data becomes significant for big data.



Value: coined by Oracle as the seventh V. The Value of data generated is insignificant in its raw form compared to its huge volume. This value can be significantly increased by processing and analyzing large volumes of such data.

In addition, a range of ‘big data’ definitions focusing on different aspects of the concept exists in the literature. While some definitions consider the ability perspective (e.g. Qin, (2014)), Authors like Arunachalam, et al. (2018) and Wamba, et al. (2015) propose more holistic definitions that encompass the technologies of process, management and analysis of data. In sum, a benshmark of a universel definition of BD do not yet exists. From the perspective of process manufacturing, organizations need to think about ‘big data’ as a mean to decode complex manufacturing system by expanding advanced skills and competencies that wield advanced IT tools to gather, store, organize, extract data from diffrent ICTs in order to analyze them and generate useful information that support the decicion making process to value creation and optimization. 2.2. Insights on Data Analytics Big data is not an end in itself. Its potential value cannot be highlighted unless when employed as a supportive tool of decision-making processes (Gandomi & Haider, 2015). To deal with this issue, quite few techniques and processes of data mining and statistical analysis under the heading “Data Analytics” have been designed by scholars and practitioners from the artificial intelligence, algorithm, and database communities in order to extract from high volumes of scalable and various data actionable grasp (Chen, et al., 2012). When tackling a big data project, a plethora of analytical techniques exists. According to (Sivarajah, et al., 2017 ), BDA can enhance the decision-making and increase organizational output by extract sense from the data for different types of analytic problems namely, descriptive analytics, predictive analytics and prescriptive analytics as presented in Figure 1. Descriptive analytics: provide hindsight on the current state of a business situation using business intelligence tools through the generation of regular reports, ad hoc reports, and alerts (Sivarajah, et al., 2017 ; Joseph & Johnson, 2013). Descriptive analytics are regarded backward looking and disclose ‘what happened’ or alert on what is going to happen through a subset of techniques. Banerjee, et al. (2013) highlight, in addition to usual reporting and scoreboards, dashboard as a form of application when an

organization systematically produces numerous indicators or metrics based on data to monitor a process over the time. Further tools such as Advanced Data Visualization (ADV), data mining and advanced statistical analysis are pointed out to leverage the descriptive analysis of processes. Other techniques emphasized to support descriptive analytics such as text, video and other multimedia analytics (Gandomi & Haider, 2015). These tools are necessary to feel the need of extracting facts from texts, unstructured audios and video streams, connecting them with the relevant decision-making process and finally nurturing a data-driven decision process. Inquisitive analytics: vet ‘why something happened’. Inquisitive analysis is generally nurtured by descriptive analysis output or additional data if needed to be gathered using descriptive analytics techniques in order to disclose the root causes of a problem (Banerjee, et al., 2013). Generally speaking, inquisitive analytics techniques seek to reveal the potential or recessive rules, characteristics and relationships (such as dependency, similarity, correlations, etc.) that exist in the data, such as generalization, association, sequence pattern mining, and clustering analytics, etc. (Cheng, et al., 2018). Other techniques exist such as Modeling Statistics, Query Tools, Spreadsheets, OLAP Tools, Decision Trees (Chen, et al., 2012). Predictive analytics: aim to provide glimpse and foresights into the future. Based on historical and current data, predictive analytics apply forecasting and statistical modelling to give insight into “what is likely to happen” in the future based on supervised, unsupervised, and semi-supervised learning models (Sivarajah, et al., 2017 ; Gandomi & Haider, 2015). Cheng, et al., (2018) talk about two categories of predictive analytics techniques. The first category is statistical analytics oriented techniques, which use mathematical models to induce and analyze existing data as well as infer and predict unknown information. These techniques include multinomial logit models (Sivarajah, et al., 2017 ), regression techniques (Gandomi & Haider, 2015), K-nearest neighbor (KNN), Bayesian (Cheng, et al., 2018). The second category is knowledge discovery KD-oriented techniques, which is data-driven that does not require to indicate assumptions and problems in advance. This category mainly includes machine learning techniques such as Neural Networks (NN), Multiple Backpropagation (MBP), Self-Organizing Map (SOM) (Sivarajah, et al., 2017 ), rough set, genetic algorithm (GA), association rule, support vector machine (SVM), generalized sequential pattern (GSP), etc. (Cheng, et al., 2018). Prescriptive analytics: optimize the process models based on the output information of predictive analytic models (Sivarajah, et al., 2017 ). Furthermore, prescriptive analytics is concerned with the definition of the set of decisions that should be done through the interpretation of cause-effect relationship among analytic results and business process policies (Banerjee, et al., 2013). In spite of its difficulty, some authors cited a set of techniques like Discrete Choice Modeling, Linear and Non-linear Programming and Value Analysis (Sivarajah, et al., 2017 ; Banerjee, et al., 2013). Moreover, “what if” simulators provide insights about what

may be the likely options that the business should implement in order to optimize its process (Banerjee, et al., 2013).

Hindsight Information Descriptive Analytics

What Happened ?

Dashboard, scoreboard, reporting… Advanced Data Visualization (ADV) Data mining, statistical analysis Text, video and other multimedia analytics

Foresight

Insight Root Causes Inquisitive Analytics

Why Did it Happen?

Clustering analytics Generalization, Association, Sequence pattern mining Modeling Statistics, Query Tools, Spreadsheets, OLAP Tools, Decision Trees

Decision Predictive Analytics

What is Likely to Happen? Multinomial logit models, Regression techniques, K-nearest neighbor (knn), Bayesian, Multiple Backpropagation (MBP) Self-Organizing Map(SOM), Rough set, Neural Network (NN), Genetic algorithm (GA), Association rule, Support vector machine (SVM), Generalized sequential pattern (GSP).

Action Prescriptive Analytics

What Should Done About It?

Discrete Choice Modeling, Linear and Non-linear Programming, Value Analysis, What if Simulators.

Figure 1. Classification of BD Analytic problems and techniques

2.3. Towards an integrative definition of Big Data Analytics “Big data Analytics” is the buzzword of the day. Actually, BDA is a new trend in operations management, which brings together a set of techniques for handling huge volumes of data sets so as to identify trends, detect patterns and draw on precious insights. In the view of process manufacturing management, Figure 2 depicts our point of view of BDA essence in the manufacturing process context. According to our point of view, the manufacturing process generates data by mean of the emergence of advanced ICTs such as sensors, RFID...etc. This data is of high-volume, large variety, high velocity, unascertained veracity, wide variability, intense volatility and low value. By using the high real-time processing ability of the theories and methods of big data, data undergoes several valuation steps, i.e. “Acquisition and Recording”, “Cleaning and Annotation” and “Aggregation and Representation” to provide reliable and complete raw data support on further data analysis. Afterwards, valuable information and

be

knowledge on four levels: descriptive, inquisitive, predictive and prescriptive can be discovered from the big data using advanced analytics techniques in order to redirect the piloting of the manufacturing process.

Manufacturing Process Data

Sensors RFID…

Volume

Big Data

Variety

Acquisition and Recording

Predictive Analytics

Velocity Veracity Variability

Inquisitive Analytics

Cleaning and Annotation

Data Analytics

Prescriptive Analytics

Manufacturing Process Challenges : Safety Energy & Envi. Quality …

Volatility Value

Aggregation and Representation

Descriptive Analytics

Figure 2. Essence of Big Data Analytics in the context of Manufacturing Process

3. Research scope and methodological approaches Papers addressing BDA in the context of manufacturing process begin to make their appearance in the literature, (e.g. Krumeich, et al. (2014), Ge, et al. (2017), He & Wang (2018), Gunasekaran, et al. (2018)). However, these papers did not discussed in-depth BDA capabilities for manufacturing processes. In addition, studies and articles dealing with BDA in manufacturing process related issues do not provide an overall understanding of its capabilities from many facets and dimensions and limit themselves to a single aspect. This suggests that BDA in the context of manufacturing process is still evolving and there are yet studies to explore in this area, hence, an intelligible understanding of the subject, its facets and classification is yet to be fully surrounded. The significant advancement of BDA not only uncovered a lack of academic and theoretical research in the field but a distinct lack of managerial insights and applications in the real-life context of manufacturing process. This paucity of literature creates a state of uncertainty among organizations about the capabilities and potential benefits that BDA could provide to their manufacturing processes. Therefore, this paper seeks to address this shortage in the literature by using a two-phase methodological research approach shown in Figure 3. The first phase is intended to conduct an extensive and comprehensive literature review of journal

articles to uncover the unexplored research trends, while the second phase is dedicated to conduct a profound analysis of a multiple case studies within a leading chemical company, which is contemporarily implementing BDA for the optimization of its manufacturing processes. The aim is to outline recommendations to fill the gaps in the research trends identified.

Figure 3. Methodological approaches of the study

3.1. Comprehensive review of literature on BDA in Manufacturing Process With a view to gain insight and provide more comprehensible understanding on the capabilities of BDA on manufacturing process, a structured review and classification of literature using bibliometric network is conducted. This methodology is perused owing to its rationality, transparency and reproducibility while gaining a meaningful understanding through the analysis of extant literature and the comparison of influential work (Sivarajah, et al., 2017 ). Moreover, Sivarajah, et al. (2017 ) argued that this methodology helps to stipulate current evidence around a technology or a treatment, such as to summarize the evidence of the benefits and drawbacks of an explicit map technique (in this example, BDA in manufacturing process). To that end, this study adopts the literature review methodology proposed by Rowley & Slack (2004). Recent papers likewise followed similar methodology in extracting insights from past BDA research papers (e.g. Arunachalam, et al. (2018), Sivarajah, et al, (2017 ) and Wang, et al. (2016)). This review methodology encompasses, in addition to the step of material collection, three data analysis steps: descriptive analysis, bibliometric analysis and network analysis. 3.1.1. Literature search and material collection Rowley & Slack (2004) propose several tools to assist in the identification and location of documents. From these tools, we opt for the search engines, which are believed to be efficient for locating web pages with

simple keyword based searches. In order to pick keywords for this study and ensure that the topic of the study was fully covered, “Big Data”, “Analytics” and “Manufacturing process” was selected as the main keywords. Other related keywords were also added such as “Operations Management”, “Process industry”, “Production process”, “predictive analytics”. The keywords were used both separately and in combination (using Boolean operators “and”/ “or”) for a comprehensive search within the following databases: Science Direct, Emerald, Springer, Taylor & Francis, IEEE Xplore, Wiley Library, Inderscience, Scopus (Elseiver) and Web of Sciences. The papers that used the abovementioned keywords and/ or their combination in the title, abstract, and full text were identified. The search started on June 12, 2018 and ended on August 15, 2018. Even though the early research was not limited with a time restriction, the final list of prescreened articles appeared in the period of 2004–2018. Prescreening has resulted in 304 papers, which were uploaded in Endnote, a reference management software package, for further analysis. Then, a step of removing of duplicated reference was carried out and followed by the elimination of out of scope and irrelevant papers through a comprehensive reading of the abstract of remaining papers (Arunachalam, et al., 2018; Wang, et al., 2016; Wamba, et al., 2015). Finally, the current review is restrained to only papers, which clearly describe the application of BDA in manufacturing process. This constituted a final list of 62 papers covering the periode from 2004 to 2018. 3.1.2. Literature Analysis The list of papers collected is analyzed in two steps. In the first step, descriptive and bibliometric analysis is complemented. Obviously, the content of the selected papers is reviewed and classified based on categories such as the distribution of publication year, research methodology, among others (Belhadi, et al., 2018). The descriptive analysis process is performed using bibliometric analysis in order to synthesize existing research, and thematic analysis for further conceptualization of the content of literature. To this end, BibExcel software is used due to its flexibility and capability to manage big volumes of data along with its compatibility with applications such as Excel, Pajek and Gephi. Afterwards, an issues analysis is used to identify research trends and construct a theoretical framework of BDA capabilities in manufacturing process. The purpose is to come across themes, topics and conceptual aspects of BDA capabilities relevant to manufacturing process context. The findings from the descriptive analysis and issues analysis are given in Section 4. 3.2. Multiple case study: BDA implementation in North African Phosphates Company In order to get practical insights of BDA in manufacturing process in real life context, in-depth data were retrieved from a number of context-specific studies in the North African Phosphates Company (NAPC) (For reasons of confidentiality, the name of the company cannot be disclosed; NAPC is a pseudonym). The aim was to conduct three case studies and collect more in-depth and practical data from the company to

draw lessons for the effective use of BDA and the associated capabilities and competitive and performance outcomes for the manufacturing process. In our case, the multiple case study as a research strategy is strongly recommended because of its perfect suitability for the building of theories during their formative stage (Gunasekaran, et al., 2018; Belhadi, et al., 2016). Actually, the multiple case study approach allows studying different facets of a phenomenon in real environment and particularly while the phenomenon and its conditions cannot be lightly explained (Ketokivi & Choi, 2014). Therefore, the multiple-case study approach was selected. 3.2.1. General situation of the study area NAPC is a leading manufacturer of phosphates and its derivatives located in North Africa. The 95% stateowned firm, charged with managing the North African country’s vast reserves of phosphate, produces a number of derivative products such as phosphoric acid, feeds and fertilizers. In 2017, NAPC was at the midpoint of a $20 billion industrial transformation program with the aim of enhancing its industrial capacity, upgrading cost effectiveness, and stimulating business competitiveness. The program hinged on outreaching the traditional activities of mining and exporting raw phosphate rock towards greater production of phosphoric acid and finished fertilizer products. In the next phase of the program, the company planned to ramp up its focus on fertilizer production, especially for markets in Africa, where fertilizer historically was not well used. Consequently, a huge opportunity arises to feed and take advantage of the growing demand of fertilizer by proposing products suitable to the needs of African farmers. In order to cope with that increase, the digital revolution and digitalization at all levels comes across as new challenge. Therefore, the different entities of the company took advantage of the industrial program to integrate the digitization component into the different manufacturing processes (Figure 4). The company is handling a variety of manufacturing processes and technologies over two production sites. In the mining site, various manufacturing processes extract and prepare the phosphate rock, which is the main input of the chemical site responsible of the manufacturing of phosphoric acid and various qualities of feeds (MCP/ DCP) a long with a wide variety of NPK-based fertilizers. NAPC aims for a qualitative leap with digital and big-data analytics to enable industrial teams to develop or even duplicate models of turnkey factories in Africa.

Mining site

Chemical site

Sulfuric acid production (SAP)

Extraction of phosphate rock

Crushing, grinding and processing of phosphate rock

Energy & Utilities production

Phosphoric acid production (PAP)

Feeds (MCP/DCP) production

Fertilizers (NPK) production

Figure 4. Flowchart of the overall manufacturing process of the company

3.2.2. Strategy of Data collection As the author is directly involved in the project of digitalization of manufacturing process of the NAPC, the case study follow the strategy of an engaged research project to collect and analyze data. In engaged or action research, the researcher and the company’s staff are working together to resolve or to improve a given issue in the company, and to contribute to the body of knowledge (Hutter, et al., 2018). Therefore, the strategy of data collection has based on direct and on-site observation along with interviews of a number of the key members of the project. Throughout the visits, the author started with observation, collection of available materials and semi structured interviews along with a conducted tour around the various entities implementing BDA project. The main interviewer was one of the key engineers directly involved in the project. 4. Findings and results from Literature review

4.1. Bibliometric analysis As mentioned earlier, the software BibExcel is used to exploit data in RIS format, which includes all the required bibliographic information for the papers. The focus is on information related to publication year, authors, title, journal, keywords and affiliations. 4.1.1. Distribution of papers per year of publication As shown in Figure 5, the distribution of papers per year indicates a clear increasing trend regarding the number of publications dealing with BDA in manufacturing process. Actually, starting from 2014, a steady rise of the number of papers can be noticed, within a range from seven articles in 2014 to 22 articles in the 2018. Moreover, most of the publications were appeared in last four years (85% of total publications). The trend highlights that there is a significant increase in the interest among researchers to investigate the topic of BDA in manufacturing process. 25 23 20

15

13 14

10 7 6

5 2 1

2

0

Figure 5. Distribution of papers per year of publication

4.1.2. Affiliation statistics Using BibExcel, the affiliations of all first authors were taken out from the RIS data file. Afterwards, the country in which the institution of the first author is located was extracted for further analysis plotted in Figure 6. Although the distribution of papers by first authors’ affiliations shows that researchers worldwide (26 countries from the five continents) are interested in research in the topic of BDA in manufacturing process, institutions in USA and China dominated the top 5 list of contributing institutions. This list contains also Germany, South Korea and Sweden.

16 14 12 10 8 6 4 2

USA China Germany Taiwan South Korea Sweden UK Austria Canada France Italy Russia Thailand Algeria Chile Finland Greece India Ireland New Zealand Serbia Singapore Slovenia South Africa Spain Tunisia

0

Figure 6. Distribution of papers by first authors' affiliations

4.1.3. Keywords statistics In a similar vein, BibExcel was used to take out the keywords utilized in the articles and the frequency of usage was calculated for further analysis given in Table 1. The top used keywords are “Big Data”, “Big Data analytics”, “Data Mining”, “Manufacturing Processes” and “Process control”. These keywords are the effective search keywords of the study. Table 1. List of the most frequently used keywords Keywords

Frequency

Keywords

Frequency

Big Data

26

Manufacturing Data Processing

4

Big Data Analytics

12

Manufacturing

4

Data Mining

8

Process Industry

4

Manufacturing Processes

8

Data Analytics

3

Process Control

5

Industries

3

Internet of Things

5

Operations Management

3

Data Analysis

5

Optimization

3

Machine Learning

5

Predictive Analytics

3

Energy Efficiency

4

Quality Control

3

4.1.4. Contributing authors and journals Table 2 illustrates the top ten contributing authors along with their number of publications. An astounding four authors, i.e. Zhong RY, Werth D, Krumeich J and Loos P tied for first place with three publications for each. Followed by six other authors with two publications for each.

Table 2. Top ten contributing authors Authors

Number of publications

Zhong RY

3

Werth D

3

Krumeich J

3

Loos P

3

Akter S

2

Chien CF

2

Chongwatpol J

2

Gunasekaran A

2

Hammer M

2

Liu Y

2

Finally, Table 3 depicts the contribution of journals according to the number of published articles. It is noticeable that eight journals shared about one-third of the reviewed articles. These journals are Computers & Industrial Engineering (CAIE), Computers and Chemical Engineering (CCE), International Journal of Production Research (IJPR), The International Journal of Advanced Manufacturing Technology (TIJAMT), Industrial Management & Data Systems (IMDS), Journal of Process Control (JPC), Journal of Cleaner Production (JCLP) and Journal of Failure Analysis and Prevention (JFAP). The two other thirds of the published articles are shared between 22 other journals and 18 international conferences. Table 3. Distribution of the reviewed paper in various journals and conferences Number of published articles 7

10.30%

Computers and Chemical Engineering

5

7.35%

International Journal of Production Research

5

7.35%

The International Journal of Advanced Manufacturing Technology

3

4.41%

Industrial Management & Data Systems

2

2.94%

Journal of Process Control

2

2.94%

Journal of Cleaner Production

2

2.94%

Journal of Failure Analysis and Prevention

2

2.94%

Others (one reference of each journal)

22

32.36%

International conferences

18

Total

68

26.47% 100.00%

Journal Computers & Industrial Engineering

%

4.2. Issues analysis Having carried out the descriptive analysis of selected papers through bibliometric analysis, the analysis of issues is used to browse BDA issues in manufacturing processes in order to conceptualize BDA capabilities. This is a philosophical conceptualization, which is considered as a first step towards theory building

(Seuring & Müller, 2008). Following the guidance of several authors (e.g. Seuring & Müller (2008); Arunachalam, et al. (2018)), a down top approach is used here relaying on reading the articles repeatedly. The deeply reading and rereading of the papers allowed to develop a conceptual framework to pave the way and circumscribe BDA capabilities in manufacturing processes (Figure 7). Manufacturing Process Challenges Quality and Process Control (Q&PC)

Energy & Environment Efficiency (E&EE)

Proactive diagnosis and Maintenance (PD&M)

Safety and Risk Analysis (S&RA)

Actions

Data Big Data Analytics Faculties

Intra-organized Data

Unstructured Data

Cleaning

Transformation

Data warehousing and management (DW&M)

Descriptive Inquisitive Predictive Prescriptive

Integration

Data Mining and Analytics (DM&A)

Data-Driven Culture (DDC)

Big Data Analytics Values

Enhancing transparency

Improving performance

Supporting decisionmaking

Enhancing Knowledge

Figure 7. Overall framework of BDA capabilities in manufacturing process

There are three layers toward conceptualizing BDA capabilities for manufacturing process. The first layer of BDA capabilities is the manufacturing process challenges, which refer to the components of the manufacturing process with the greatest potential of value to catch. These components include, in addition to Quality and Process Control (Q&PC), other high-impact components such as Energy & Environment Efficiency (E&EE), Proactive diagnosis and Maintenance (PD&M) and Safety and Risk Analysis (S&RA). All these components generate a huge stream of data and thrives on a number of BDA faculties allowing to transform data to actions. These faculties include Data Warehousing (DW), Data Aggregation and Integration (DA&I), Data Analytics and Modeling (DA&M) and, Data-Driven Culture (DDC). In the third layer concerns the values that BDA could provide to manufacturing process. Indeed, BDA embodies quite few values such as enhancing transparency, improving performance, supporting decision making and

enhancing knowledge. Further description of the different layers of the framework is given in the following sections. 4.2.1. Manufacturing Process Challenges Table 4 provides the distribution of papers by the challenges of manufacturing process addressed by BDA. Above all, it is noticeable that numerous papers dealt with more than one manufacturing process challenge addressed by BDA. Afterwards, the overwhelming majority of the papers are about Q&PC (51 papers). This is followed by E&EE, PD&M and S&RA with respectively 18, 9 and 9 papers. Table 4. Challenges of BDA related to manufacturing process Challenges

Key elements

Quality and Process  Statistical Process control  Alarm management Control (Q&PC),  Connectivity and causality analysis  Unsupervised Learning Approaches to Process Monitoring

and  Energy big data acquision and mining Environment  Energy consumption Efficiency (E&EE) patterns  Smart Grids Energy

diagnosis  Condition based Maintenance and Maintenance  Online/ Real Time (PD&M) predictive maintenance  Tele-Maintenance Safety and Risk  Bayesian based HAZOP  Advanced inherently safer Analysis (S&RA) design  Safety instrumented systems Proactive

References

Number

%

Li & Kashiwagi (2005) ; M’Sahli & Matlaya (2005) ; López-Escobar, et al. (2012) ; Krumeich, et al. (2014) ; Grossmann (2014); Qin (2014) ; Yang, et al. (2014) ; Chongwatpol (2015) ; MacGregor, et al. (2015) ; O’Donovan, et al. (2015) ; Stojanovic, et al. (2015) ; Severson, et al. (2015) ; Zhuchkov (2015) ; Eckstein, et al. (2016) ; Krumeich, et al. (2016) ; Kumar, et al. (2016) ; Li (2016) ; Smirnov & Nasonov (2016) ; Qiu, et al. (2016) ; Wang, et al. (2016) ; Weese, et al. (2016) ; Chen, et al. (2016) ; Beneventi, et al. (2017) ; Chen (2017) ; Chiang, et al. (2017) ; Chien, et al. (2017) ; Choi, et al. (2017) ; Ge, et al. (2017); Hammer, et al. (2017) ; Ji & Wanga (2017) ; Lee, et al. (2017) ; Lindström, et al. (2017) ; Wamba, et al. (2017) ; Zhang, et al. (2017) ; Zhong, et al. (2017) ; Amini & Chang (2018) ; Gunasekaran, et al. (2018); Hammer (2018); He & Wang (2018); Hu, et al. (2018); Isaksson, et al. (2018) ; Khakifirooz, et al. (2018) ; Kho, et al. (2018) ; Kozjek, et al. (2018) ; Lee, et al. (2018) ; Mehta, et al. (2018) ; Ning & You (2018) ; Sadati, et al. (2018) ; Stanley (2018) ; Wamba, et al. (2018) ; Wang, et al. (2018) ; Tewari & Dwivedi (2019 ) Grossmann (2014) ; Shin, et al. (2014) ; Shrouf, et al. (2014) ; O’Donovan, et al. (2015) ; Chongwatpol (2016) ; Favoino, et al. (2016) ; Li (2016) ; Papacharalampopoulos, et al. (2016) ; Paul, et al. (2016) ; Beneventi, et al. (2017) ; Chiang, et al. (2017) ; Ge, et al. (2017); Zhang, et al. (2017) ; Hammer (2018) ; Hidalgo, et al. (2018) ; Lee, et al. (2018) ; Shao, et al. (2018) ; Zhang, et al. (2018) ; Shukla & Mattar (2019) Qin (2014) ; O’Donovan, et al. (2015) ; Krumeich, et al. (2016) ; Beneventi, et al. (2017) ; Lindström, et al. (2017) ; Sun, et al. (2017); Zhang, et al. (2017) ; Hammer (2018); Isaksson, et al. (2018)

52

59.09%

18

20.45%

9

10.23%

O’Donovan, et al. (2015) ; Li (2016) ; Chiremsel, et al. (2016) ; Khakzad & Reniers (2016) ; Choi, et al. (2017); Zerrouki & Smadi (2017) ; Hammer (2018) ; Hu, et al. (2018) ; Wang, et al. (2018)

9

10.23%

Total

88

Note: Some articles are included more than once since they discuss more than one manufacturing process challenge.

Quality and Process Control (Q&PC): In the near future, organizations that are able to monitor their operations through the fast-paced increasing amount of data to forecast their quality fault and proactively control their processes by means of advanced analytics will be in advance of their competitors (Krumeich, et al., 2016). In doing so, He & Wang (2018) talked about a new generation of Statistical Process Control (SPC) dealing with complex and multivariable with which MSPM methods may fail or lead to misleading results (multimodal distribution, dynamics, nonlinear relationships between variables, non-Gaussianity, time-varying characteristics, other characteristics such as outliers, gross errors and/or failed sensors). This new generation of SPC implies the use of more and more Model Predictive Control (MPC) (Isaksson, et al., 2018; Krumeich, et al., 2016; Zhuchkov, 2015; Chongwatpol, 2015; Li & Kashiwagi, 2005) through the integration of advanced BDA techniques such as Control Chart Pattern Recognition (CCPR), Regression-Based Methods, Neural Networks, Support Vector Methods (Weese, et al., 2016; Lee, et al., 2017; Tewari & Dwivedi, 2019 ). Another aspect of QP&C is Alarm management mentioned by Hu, et al. (2018) which relay on a set of BDA techniques such as Run Length Distribution & Delay Timer Analysis (RLD&DTA), Chattering Index (CI), Oscillating Alarm Analysis (OAA), Alarm Flood Analysis (AFA), Causality Inference for Alarms (CIA)) and Mode-Dependent Alarm Analysis (MDAA). The aim is to predict the occurrence of abnormal situations and then prevent their propagation along the interconnected pathways to cause significant and catastrophic disruptions in the process. In another side, and either when little is known about the process or information is unavailable as to what forms an out of-control event, machine learning techniques such as Big Data Approximating Control (BDAC), PCA/PLS, k-means clustering, Self-organizing map (SOM) and manifold learning methods are used (Ge, et al., 2017; Weese, et al., 2016; Stanley, 2018). This is to spot correlations and causal relations between process variables to capture material and information flow paths in the process (Hu, et al., 2018; Lee, et al., 2018). Energy and Environment Efficiency (E&EE): Bowed to the overwhelming pressure of limited natural resources and growing serious environmental issues (Belhadi, et al., 2018), manufacturing processes are placing energy saving and emission reduction as two important challenges must be addressed by BDA (Zhang, et al., 2018). For instance, advanced analytical tool can be applied to optimize the factors that are believed to have the biggest impact on environmental performance. Zhang, et al. (2018) proved this fact by proposing a big data driven analytical framework based on two technology, i.e., energy big data acquisition and energy big data mining, are utilized to reduce the energy consumption and emission for energyintensive manufacturing industries. Advanced technologies implemented along with sophisticated operating conditions make the understanding of the energy consumption attitude very hard since very

100%

complicated, nonlinear dynamic variables are included (Chongwatpol, 2016). Therefore, BDA is vital to better understand and control the operational parameters, optimize energy management, and reduce environmental impact (Chiang, et al., 2017). In doing so, BDA has been widely applied in manufacturing processes such as predictive energy consumption models (Grossmann, 2014; Shin, et al., 2014; Eckstein, et al., 2016; Shao, et al., 2018), smart grid management

(Chiang, et al., 2017) and building energy

management (Favoino, et al., 2016; Hammer, 2018). Proactive diagnosis and Maintenance (PD&M): Besides its use in manufacturing process for quality and environment issues, BDA is also being incorporated in another critical challenge facing the manufacturing process namely, proactive diagnosis and maintenance of equipment. One of the early known industry applications of PD&M is the Condition-based Maintenance (CBM) (Krumeich, et al., 2016), which integrates present process states and events to estimate when equipment requires maintenance to minimize unplanned shutdown based on the usage, age, and performance of the equipment (Chiang, et al., 2017). According to Krumeich, et al. (2016), big data acquisition and sensor technology allow to locate and cluster equipment defects in order to easily identify, diagnose and solve equipment problems before a failure actually occurs. Other aspects of PD&M have been widely discussed in the literature. For instance, Beneventi, et al. (2017) and Zhang, et al. (2017) discussed the Online/ Real Time Predictive Maintenance recognizable by the integration of equipment behavior patterns by finding thresholds and relations between parameters that can be used to indicate potential problems occurring (i.e., diagnostics) and maintenance requirement (i.e., prognostics). Finally, BDA is also applied for remote or tele-maintenance of equipment along with some other maintenance-related functions such as management of spares inventories and consumption (Hammer, 2018). Safety and Risk Analysis (S&RA): Owing to the advance and the sophistication of the current manufacturing processes, risk and safety analysis is today more challenging and time-consuming (Zerrouki & Smadi, 2017). Undoubtedly, Safety and Risk Analysis in manufacturing process would benefit by proper application of BDA (Choi, et al., 2017). As stated by Hammer (2018), application of advanced Analytics along with rendering and acting on the given insights lead to set up strong safety requirements and thus protect the physical safety of the workers, users and the environment. Zerrouki & Smadi (2017) demonstrated the use of Bayesian Networks (BN) in HAZOP analysis. In the same vein, Khakzad & Reniers (2016) applied Advanced Analytics to risk-based design and decision making in chemical plants to employ the principles of inherently safer design (ISD) and land-use planning (LUP). In addition, Chiremsel, et al. (2016) used Advanced Analytics to diagnose the Safety instrumented systems (SISs) in order to prevent the occurrence of hazardous events and to alleviate their aftermath to workers, equipment, and environment.

Hu, et al. (2018) and Li (2016) underlined, in addition to HAZOP-based analytics, the diagnosis of abnormal working conditions and decisions on emergency treatment of major accidents. 4.2.2. Big Data Analytics Faculties Undoubtedly, the application of BDA holds enormous potential of improvements for the different challenges that manufacturing processes are facing. The driving force behind this huge potential is represented by quite few faculties of BDA that can benefit the manufacturing process. Data Warehousing and Management (DW&M): In context of a manufacturing process, data is much dispersed and can be extracted from extremely varied sources. For that, BDA offers the faculty of collecting, integrating, transforming and storing data from disparate data sources where traditional database systems are ineffective (Hu, et al., 2018). DW&M is a prerequisite therewith successful gain of data mining can be insured. This would lead to improved data quality enabling the enhancement of the accuracy and performance of the subsequent mining process (Cheng, et al., 2018). Hidalgo, et al. (2018) talked about self-adaptive stream processing systems dealing with the high frequency data streams featuring timevarying characteristics that challenge the traditional stream processing systems capacities. Krumeich, et al. (2016) outlined the ability to preprocess real-time data via the in-memory data management platform so that structured information can be extracted from unstructured multimedia data. Zhang, et al. (2017) highlighted the efficiency of the business-to-manufacturing markup language (B2MML) and extensible markup language (XML). Data Mining and Analytics (DM&A): Data mining algorithms or methods are applied to extract useful insights and knowledge from huge amount of data (Cheng, et al., 2018). Embedding big data mining result into the process monitoring system (PMS) and decision support system (DSS) creates a closed-loop system of feedback gathering and timely adjustment in order to achieve process optimization on all manufacturing process challenges (Zhang, et al., 2017). Actually, DM&A enables decision-making systems to thrive on self-learning, cognitive faculties to address real-time data and complex interrelationships. Hammer (2018) and Qiu, et al. (2016) considered DM&A as a vital nexus between operational technology and information technology to form advanced process control systems (APC) with advanced analytics. Chiang (2017) reported the successful use of Enterprise Manufacturing Intelligence (EMI) in the case of chemical sector. Techniques of DM&A can be divided according to the time until decision in two categories, viz. real-time techniques and offline techniques. Table 5 and Figure 8 depict the classification of offline/ online DM&A techniques according to the four stages of the BDA namely descriptive, inquisitive, predictive and prescriptive analytics.

Table 5. Classification of Data Mining and Analytics techniques Categories Descriptive analytics techniques

Inquisitive analytics techniques

Predictive analytics techniques

Prescriptive Analytics techniques

Offline techniques                                       

Fine-Kinney continuous quality control Data Visualisation Big data perception & acquisition data mining In-memory management & connectivity API Management Random Forest (RF) Excursions diagnosis Fault Tree Analysis heuristic algorithms for scheduling tasks:  First-in-first-out (FIFO)  Earliest-planned-time (EPT) forecasting K-means Clustering Gradient Descent Optimization Process Monitoring Statistical process monitoring Sampling Hadoop MapReduce programming data-driven modeling meta-heuristic optimization Characterization & Classification Bayesian Network linear and a nonlinear Predictive Control control chart and defect prediction Process Analytics Process Prediction Predictive control Statistical Learning Methods Regression analysis Cluster analysis neural network clustering-based on prediction scores Kernel Smoothing Methods principal component analysis Quantitative modeling prognostic analysis Proactive Alerts Bayesian belief network time-series model

Online techniques

References

 Visualization & Data Modeling  Aggregation & Contextualization  Event Bus

Wang, et al. (2018) ; Mehta, et al. (2018) ; Lindström, et al. (2017) ; Kozjek, et al. (2018) ; Kho, et al. (2018) ; Zhang, et al. (2018) ; Chongwatpol (2016) ; Krumeich, et al. (2014) ; Yang, et al. (2014) ; Tewari & Dwivedi (2019 )

 Multivariate latent variable (PCA/ PLS)  self-adaptive stream processing  Data classification

Chien, et al. (2017) ; Chiremsel, et al. (2016) ; Kozjek, et al. (2018) ; Kho, et al. (2018) ; MacGregor, et al. (2015) ; Hidalgo, et al. (2018) ; Amini & Chang (2018) ; He & Wang (2018); Tewari & Dwivedi (2019 ) ; Hu, et al. (2018) ; Kumar, et al. (2016) ; Sadati, et al. (2018) ; Shukla & Mattar (2019)

 Classifiers for pattern recognitions  Support Vector Machine (SVM)  online predictive maintenance  fault prediction  Streaming Analytics  pattern matching and approximation

Chiremsel, et al. (2016) ; Chen, et al. (2016) ; Eckstein, et al. (2016) ; Lindström, et al. (2017) ; (Khakzad & Reniers, 2016) ; Chongwatpol (2016) ; Ji & Wanga (2017) ; Krumeich, et al. (2014) ; Tewari, et al. (2019) ; Kumar, et al. (2016) ; M’Sahli & Matlaya (2005) ; Shin, et al. (2014) ; Zhuchkov (2015) ; Weese, et al. (2016) ; Khakifirooz, et al. (2018) ; Ning & You (2018) ; Stanley (2018)

Wang, et al. (2018) ; Chongwatpol (2016) ; Krumeich, et al. (2014)

Figure 8-a. distribution of DM&A techniques by categories Prescriptive Analytics Descriptive 7% analytics 22%

Figure 8-b. online vs offline techniques numbers

Online technics

Offline technics 14

14

7

Predictive analytics 41%

6 3

3

3

Inquisitive analytics 30%

0 Descriptive Inquisitive analytics analytics

Predictive Prescriptive analytics Analytics

Figure 8. Analysis of used Data Mining and Analytics techniques

As shown in Figure 8-a, apart from a quite few studies dealing with prescriptive analytics techniques (7% of studies), most research works are primarily concerned with the study and implementation of descriptive, inquisitive and predictive analytics techniques (93% of studies). On the other hand, according to Figure 8b, offline techniques are used much more than online techniques (38 offline techniques were levied against 12 online techniques), especially in papers offering the analysis of some real case studies such as Kozjek, et al. (2018) ; Shukla & Mattar (2019); and Tewari, et al. (2019). Data-Driven Culture (DDC): Data-driven culture is an inviolable faculty that BDA brings to the way of monitoring and optimization of manufacturing process. DDC is a thought pattern that summarizes a set of convictions, mindsets, attitudes, and ways towards process optimization. In traditional processes, the vast troves of data collected is typically used only for monitoring purposes and not as a basis for improving operations (Sadati, et al., 2018). However, the application of BDA participates effectively in culture change towards fact-based decision making through embedded analytics. Hammer (2018) argued that changes in decision-making culture gained by BDA could meaningfully improve process performance. 4.2.3. Big Data Analytics Values When applying BDA in manufacturing process, huge amounts of collected data can be transformed into several values and benefits for the different components of the manufacturing process. The objective is to leverage the knowledge and that to increase the value of information (Lee, et al., 2018). Enhancing transparency: The first key value realized from BDA is the improvement of intra-and interorganizational transparency and accountability. As stated by Chongwatpol (2015), process, which used to tap big data, will receive real-time information from sensors, RFID and other devices in true transparency and without external interference.

Improving performance: It is unanimously acknowledged that the application of BDA leads to a significant improvement in operational performance in all the challenges of the process (quality, energy, emission, safety …). Case studies indicate a meaningful improvement in performance (e.g. (Chiang, et al., 2017); (Chongwatpol, 2016); (Chongwatpol, 2015); (Isaksson, et al., 2018)). Supporting decision-making: According to Isaksson, et al. (2018), BDA generates new insights and knowledge-based data disposable for consultancy by the manufacturing process in order to support more decentralized decision-making. Zhang, et al. (2017) discussed the ability of BDA techniques such as Bayesian, and decision trees of enabling the decision makers to make flexible decisions in the presence of attributes such as flexibility, quality, innovativeness, pro-activity and cost. Lee, et al. (2018) explored the use of recent advances in deep learning and reinforcement learning to elaborate decision policy. Enhancing Knowledge: Besides technical values, BDA is confirmed to cover intelligence organization, including work and staff skills (Li, 2016). As shown by a case study presented by Li (2016), BDA allows to perform job skills training and to study via technological applications such as virtual reality technology, simulation platform, and 3D interaction and demonstration. Choi, et al. (2017) argued that BDA helps to understand phenomena, which cannot be theatrically explained as well. 4.3. Critical analysis and research trends The findings of our literature review stressed out how manufacturing process could use BDA capabilities, matter to extract value from data sets huge as big data through a three-layer framework. Consequently, BDA capabilities have confirmed essential to make process engineers and managers obtain drastically more information than before, to what concerns different challenges in their manufacturing process. However, a set of trends can be identified from the literature review conducted. In reality, these trends constitute research gaps that are still not fully addressed and thus they need much focus. 4.3.1. Research trend 1: BDA-enabler architecture In manufacturing process, data are streamed from multiple, heterogeneous and dispersed sources (structural and unstructured). Therefore, the implementation of data mining and BDA requires a smart architecture based on data management and storage strategies, governance and risk management (Chen, 2017) in order to deal with large volume of data from different sources. Indeed, volume, variety, velocity and the other seven V’s that characterize big data mean that using BDA needs high-performance resources such as storage space and processing modules (Chiang, et al., 2017; Isaksson, et al., 2018). According to our literature review, authors emphasizes this issue through two major approaches. First, some proposed frameworks proposed the establishment of internal data warehousing which enables to store and manage collected data (Hutter, et al., 2018; Choi, et al., 2017; Beneventi, et al., 2017). These solutions allow companies to have

full control over their data and provide more security but it requires huge investment costs in terms of implementation, engineering and maintenance. Second, some other works proposed the use of solutions based on outsourcing data such as Cloud Manufacturing (CM) (Papacharalampopoulos, et al., 2016; Kumar, et al., 2016; Qiu, et al., 2016). This solution reduces costs but poses a significant risk on data security. Although there is an increasing trend towards the implementation of BDA-enabler architecture, this is currently an underexplored area due to the several issues that arises such as data security and privacy (Hammer, 2018), crisis and risk management (Zerrouki & Smadi, 2017), implementation costs (Shukla & Mattar, 2019)... etc. 4.3.2. Research trend 2: Real-time data mining approaches The maximum potential from BDA can be derived for optimized decision making while controlling the manufacturing process ahead of time (Hu, et al., 2018; Hammer, 2018). One of the most important issue believed to be faced by the manufacturing process while using BDA is the time interval until the decision (Chiang, et al., 2017; He & Wang, 2018) because of the inevitably loss of process performance between the treatment of offline data and the proactive decision (Choi, et al., 2017). The time scale of this interval may vary depending on the computational environment, e.g., real-time vs. offline, and the application domain. Thus, the ability to gather and process instantaneously online data is of paramount importance. Our literature review shows an apparent trend towards BDA models able to be run in a real-time, streaming computational environment in the context of manufacturing process (Ji & Wanga, 2017; Li, 2016; Lee, et al., 2018; Kumar, et al., 2016). In recent years, real-time approaches have been investigated extensively for descriptive analytics (Hidalgo, et al., 2018; Tewari & Dwivedi, 2019 ) and inquisitive analytics (He & Wang, 2018; Ji & Wanga, 2017). On the other hand, and unlike descriptive and inquisitive analytics, which have gone forward towards these trends to be appropriate to use in data-intensives applications, predictive and prescriptive analytics are still at their infancy. Apart from some applications of online and real-time Predictive Maintenance (e.g. Beneventi, et al. (2017) ; Zhang, et al. (2017)), studies on predictive and prescriptive analytics are still conceptual (Chiang, et al., 2017; Cheng, et al., 2018; Hammer, 2018). When investigating the reasons behind this lateness, multiple issues arise such as the need to enhance the speed of requests processing and online model updating (He & Wang, 2018). For instance, Amini & Chang (2018) admitted that the shortage in their computational environment would prevent real-time monitoring from taking place. Chiang, et al. (2017) argued that in spite of the significant strive to ramp up storage capacities, there is an ongoing need to swiftly store and process data in order to achieve optimal decision in real-time. On the other hand, several authors have mentioned the cultural issue within process industries. Chiang, et al. (2017) argued that the process industry is in nature slow to respond to real-time customer

feedback. Therefore, real-time applications harbor many difficulties to be included in the mindset of process managers. Overall, the emergence of IoT and sensor-driven data in the process industry has led to an increasing demand for real-time systems, which need more focus in order to contribute to further advancements of BDA research in manufacturing process. 4.3.3. Research trend 3: Integrated Human-Data intelligence Human capital has been considered as a ‘must have’ for developing BDA capabilities in manufacturing process. In a company, human capital encompasses all the experience, knowledge, judgment, risk-taking propensity, and wisdom of people related to the manufacturing process (Hammer, 2018). Building upon our literature review, several studies argue that human intelligence alongside computational intelligence will experience significant evolution in the near future, reaching an advanced stage (Li, 2016; Chiang, et al., 2017). On the other hand, however, an underpinning issue is also arisen concerning the role of the human operator in the era of BDA and with more and more automated manufacturing process (Isaksson, et al., 2018). Let us say that the machine still cannot replace human intelligence and the operator is still needed in the manufacturing process. Therefore, it is of utmost importance to consider how the operator interacts with model-based control and optimization in order not to progressively weaken the operators’ skills (Isaksson, et al., 2018; Shukla & Mattar, 2019). Authors such as Li (2016) discussed automated learning systems that are important to combine human and machine intelligence and must be effectively integrated with human learning and decision environment. It is clearly observed that the strive for a data-driven system with humans and technical components working synergistically together in a socio- technical unit, providing creative value-added by working easily together is a trend in the literature (Hammer, 2018; Shukla & Mattar, 2019). 4.3.4. Research trend 4: Prescriptive analytics within the manufacturing process Prescriptive analytics is the most advanced stage of data analytics and can provide the greatest intelligence and value to the manufacturing process (Hammer, 2018). The application of prescriptive analytics has been increasingly sparking research concern within the manufacturing process in recent times (Wang, et al., 2018). For several researchers, prescriptive analytics, also called operational analytics, is increasingly sought because it constitutes the next round towards enhancing data analytics maturity and triggers optimized decision making without human intervention ahead of time (Hammer, 2018; Wang, et al., 2018; Chongwatpol, 2016). In fact, with the temporal nature of the manufacturing process, gaining business value from the large amount of data generated requires action expeditiously on real-time events before the value fades. This requires more than a prediction; it needs determining accurately what to do and when to do it.

Therefore, prescriptive analytics is expected to get a lot more common and pervasive among a large array of practitioners. In spite of this, our literature review shows clearly that prescriptive analytics within the manufacturing process is still in an early stage compared to descriptive, inquisitive or even predictive analytics. With the exception of some few initiatives (e.g. Wang, et al. (2018), Chongwatpol (2016) and Krumeich, et al. (2014)), the potential of prescriptive analytics in far from being well exploited. Therefore, further research is required towards the direction of combining the outcomes of predictive analytics with big data and advanced algorithms in order to advance the next generation of manufacturing process based on analytics systems. This will lead to not only define risks and potential abnormalities while addressing the manufacturing process challenges, but also suggest actionable levers, effectively providing accurate, reliable, real-time decision support to process managers. 5. Insights and learnings from multiple-case studies The goal of the case studies was to further explore the interdependencies of core BDA capabilities in a reallife context of manufacturing process and uncover emerging themes. 5.1. Description of the case studies We discuss three projects of BDA implementation at different plants subsidiaries to NAPC. The projects are part of a larger project of the digital transformation of the operations of the company in which the top management show clearly its commitment. We selected these cases because they are well documented, they involved BDA capabilities, they came up with implications for our research trends and they are judged as successful projects. 5.1.1. Case 1: Implementing BDA in a fertilizers plant In April 2018, a fertilizers plant in chosen to be a pilot area of the project of BDA implementation. For that, the management hired a highly qualified staff of IT engineers and data scientists to bring support to process and maintenance engineers during the implementation program. The first prerequisite is the implementation of a real-time system through the creation of a Data Hub in which multiple sources of data from different stages of the manufacturing process (structured and unstructured) across various departments (production, maintenance, laboratories…) are integrated. The aim is to combine these data with historical information for better process monitoring. In doing that, a range of IT capabilities endowed the manufacturing process of the company. For instance, Hadoop Data Lake, which ties also into HDFS (Hadoop Distributed File System), was acquired in order to ensure DW&M faculty. Actually, the Hadoop Data-Lake constitutes a solid infrastructure for real-time data management besides several faculties such as data gathering, historicizing, recovering, analyzing, presenting, and visualizing. Several difficulties motivated the use of

Hadoop Data Lake, namely: (1) The OPC server is deployed in windows machine, which present security risk, (2) Due to lack of Lora network, the context data of smart sensors are not exploited, and (3) The storage capacity of DSC historical data is limited to one year. The use of Hadoop Data-Lake enables to bring together a vast variety of data from different platforms linked to the manufacturing process such as: 

PI system: used as self-service monitoring capability of data Plant by Process Engineer,



OPM Treatment: used as a tool of performance monitoring,



MyOPS platform: already deployed as a mandatory tool for the HSE processes & the maintenance reliability processes,



Oracle EAM: addresses the comprehensive and routine asset maintenance requirements of asset intensive organizations,



LIMS: allows to effectively managing the flow of samples and associated data to improve lab efficiency through standardizing workflows, tests and procedures, while providing accurate controls of the process.



Connect and AlMaarifa: two platforms, by applying the principle of Internet of People (IoP), constitute an information space where people are interlinked and their knowledge and expertise are shared.

Besides the Data Hub, an Intelligent Monitoring System (ISM) was implemented based on two dimensions; (1) a vast campaign of data informatization using online sensors for process and environmental parameters (concentration, emission of toxic gas…etc.), (2) Complex Event Processing (CEP) using specific algorithms. The aim was to track and process streams of data about occurred events by integrating multiple sources with the aim to detect situations or patterns that comprise a particular meaning for the system, such as opportunities or threats, and to react to them. 5.1.2. Case 2: BDA implementation in a phosphoric acid plant In the early 2019, BDA implementation project was conducted in a phosphoric acid plant. The project was based on a tree maturity levels, viz. people, organization and infrastructure. People: A key activity at this level is to involve the workforce in developing the vision. Accordingly, a number of workshops have been organized to ensure that all employees have the same level of understanding of the overall project alongside BDA capabilities. Then, they were invited to propose the issues to be covered by BDA implementation. The technique of change agent was used to ensure broad communication and involvement of all employees. Afterwards, the staff has undergone multiple training sessions on BDA modules and digital interfaces in order to educate people to develop the ability to exploit connected data systems. Another key at this level is the reinforcement of the use of insight analysis and

data interpretation to streamline operational processes by creating routines for using historical and real-time data analytics in the work process. Organization: on the organization side, the focus was on introducing efficient and flexible structure to the implementation stage was by integrating a Stage-Gate project model with agile principles. In fact, a multidisciplinary team was formed including people with an understanding of both the manufacturing processes and the digital architecture. This bridges programming and manufacturing and enables the organization to access the full potential of the data stream. In parallel, the roles and responsibilities inherent in data management were defined by a “Data governance guidelines”. Bringing in the right people and defining their roles ensures the reliability of data management system and amplifies the organization’s digitalization potential. Infrastructure: in order to create an enabled architecture for BDA implementation, the first keystone was connecting existing applications across data flow to create a common platform. Afterwards, a system of real-time performance analysis was implemented based on cloud-based connected services provided by PI system. The system was based on a real-time data infrastructure that collects sensor-based data from many different sources, stores them in a secure, central location to facilitate their real-time visualization, analyze and share. The systems enabled automated analysis of operational information and gave warning signals for environment, production and maintenance departments. Alerts such as emissions, up/downtime, MTBF, MTTR and failure rates of equipment and production tools were shared on-line to be analyzed and consequential proactive maintenance boosted and operated accordingly to attain a more sustainable plant, both in economical and eco-friendly aspects as the lifetime is prolonged and durability enhanced. 5.1.3. Case 3: Implementation of an intelligent and self-controlled production unit The project was concerned the implementation of an intelligent and self-controlled unit of the concentration of phosphoric acid. In that vein, the unit needed to have four characteristics: (1) connectedness, (2) context awareness, (3) intelligence and (4) metered services. Therefore, the overall architecture of the system was based on cloud-based manufacturing equipment using CPS and BDA. The first step was to upgrade manufacturing equipment to cloud-based manufacturing equipment. For that, several modules were integrated into the manufacturing devices in order to improve the cognitive nature of the system in understanding its status and communicate this information to the external environment. 

Intelligent adaptive control module, which covers the principal control options such as interpolation calculation, input/output management, motion control, etc. suing fuzzy logic and neural network control,



Equipment monitoring module, which monitors all the data attributes (working progress of production tasks, real-time process parameters, etc.) of cloud-based manufacturing equipment. The aim is to generate production history in the cloud for subsequent data analytics in the cyberspace.



Data processing module functioning with the numerical control kernel to pre-process all the unit condition data that is collected from different sensors before the data goes to the cloud for further analysis.

The data attributes were then transferred the cloud via an OPC system. The objective is to create a permanent link with the cyberspace to maintain the digital twin of a cloud-based equipment. This stream is bidirectional since it transfer back orders from the cyberspace to smart equipment to execute actions based on MQTT protocol. The self-control of the unit of phosphoric acid concentration allowed to increase the quality of produced phosphoric acid alongside several environmental benefits such as the efficiency of water use during washing cycles and the reduction in HF emissions. 5.2. Recommendations drawn from the case studies When analyzing the experiences of BDA implementation within the manufacturing process in the case studies, several lessons can be learned. Thereafter, a couple of recommendations would be useful to address the research trends that emerge from the literature. Table 6 summarizes the proposed recommendations. Table 6. The summary of the proposed recommendations Research trend BDA-enabler architecture

Case 1

Case 2

Case 3

√

Rec1: The implementation of internal sourcing (Data Hub) to the storage and the processing of big data can resolve the issue of data security and privacy. However, the issue of implementation costs arises. √

√

√

Real-time data mining approaches

Integrated Human-Data intelligence

Recommendations

Rec2: External sourcing such as cloud computing can offer several opportunities for data management at the expense of privacy. Rec3: The definition of roles and responsibilities inherent in data management through “Data governance guidelines” may reduce the risk of data security.

√

√

√

Rec4: Intelligent Monitoring System (ISM) using sensor-driven information and Complex Event Processing (CEP) can promote the implementation of real-time approaches.

√

√

√

Rec5: Connection between the different process elements and applications is crucial to promote real-time monitoring

√

√

Rec6: Cloud manufacturing (CM) can provide solutions to promote realtime techniques application.

√

√

Rec7: Broad communication using IoP along with early involvement of all employees ensure the integration of humans in BDA

√

√

Prescriptive analytics within the manufacturing process

Rec8: Hiring skilled employees alongside advanced training sessions on data science can promote the tie of human-data intelligence connection √

Rec9: Generic BDA methods and algorithms utilizing artificial intelligence and machine learning can serve as a basis while developing prescriptive approaches.

√

Rec10: Generalization of automation and sensor-driven information is of outmost in the next generation of prescriptive analytics.

5.2.1. BDA-enabler architecture During the case 1, a Data Hub relying on internal sourcing was implemented; providing great benefits in term of the big storage capacity proposed by Hadoop Data Lake. Indeed, the huge capacity of Data warehousing allowed keeping process parameters history for more than one year available in the DCS, and then modeling more accurately trends and tendencies based on advanced analytics. Although this solution is not widespread in the literature due to the required investment and the high technical complexity, it is shown to be beneficial especially to resolve the problem of privacy and data security. Recommendation 1: The implementation of internal sourcing (Data Hub) to the storage and the processing of big data can resolve the issue of data security and privacy. However, the issue of implementation costs arises. Another solution for BDA-enabler architecture implementation is to rely on outsourcing resources provided by cloud manufacturing as explored in case 2 and case 3. This is the most popular solution in the literature (Papacharalampopoulos, et al., 2016; Kumar, et al., 2016; Qiu, et al., 2016) as it allows to get access to various functionalities while storing and processing data. Recommendation 2: External sourcing such as cloud computing can offer several opportunities for data management at the expense of privacy. As shown in case 2, the definition of roles and responsibility related to data management strategy is of outmost importance in order to mitigate the risk of privacy and data security. Recommendation 3: The definition of roles and responsibilities inherent in data management through “Data governance guidelines” may reduce the risk of data security. 5.2.2. Real-time data mining approaches Intelligent monitoring is a well-established system of technology management that combines real-time sensing with project-specific data processing such as complex event processing (CEP), predictive analytics,

and collaborative tools for data interpretation and decision-making (Chen, et al., 2016; Krumeich, et al., 2014; Hu, et al., 2018). Recommendation 4: Intelligent Monitoring System (ISM) using sensor-driven information and Complex Event Processing (CEP) can promote the implementation of real-time approaches. In all case studies, the establishment of a communication between different elements and applications in the manufacturing process, either by creating a common platform (case 1) or using virtual cyberspace (case 2 and 3) is very beneficial for the manufacturing process to develop effective real-time monitoring. Recommendation 5: Connection between the different process elements and applications is crucial to promote real-time monitoring One of the most remarkable field of interest while addressing real-time challenge if BDA is cloud manufacturing (CM), which represents a loosely-connected network of manufacturing services that can be swiftly adopted in the process (Isaksson, et al., 2018). In agreement with several studies in the literature (e.g. Papacharalampopoulos, et al. (2016); Kumar, et al. (2016); Qiu, et al. (2016)), the results of case 2 and case 3 confirm the usefulness of CM in advancing real-time control of the process. Recommendation 6: Cloud manufacturing (CM) can provide solutions to promote real-time techniques application. 5.2.3. Integrated Human-Data intelligence The implementation of IoP along with knowledge share platforms in case 1 (e.g. AlMaarifa of Knowledge Management and Connect) is believed to ensure an early involvement of human in BDA implementation. These platforms is a real tool for enhancing the knowledge of operators who now more familiarized with new technologies. Recommendation 7: Broad communication using IoP along with early involvement of all employees ensure the integration of humans in BDA The secret to leverage the upshots from BDA is to outfits managers and operators with solid professional proficiency as depicted in case 2 and case 3. Hence, it is vital that companies provide analytical training courses in fields like basic statistics, data mining and business intelligence to the process operators who will assume an important support part in the new information-rich work context. Recommendation 8: Hiring skilled employees alongside advanced training sessions on data science can promote the tie of human-data intelligence connection

5.2.4. Prescriptive analytics within the manufacturing process A starting point of switching to the new generation of prescriptive analytics is predictive BDA methods and algorithms. As described in case 3, fuzzy logic and neural network were utilized. The same conclusion can be drawn from the extant literature (e.g. Wang, et al. (2018), Chongwatpol (2016) and Krumeich, et al. (2014)). Recommendation 9: Generic BDA methods and algorithms utilizing artificial intelligence and machine learning can serve as a basis while developing prescriptive approaches. It is of paramount importance to computerize the collection of data all over the manufacturing process. To do so, sensors and intelligent devices must be generalized to enhance the ability of the process to exchange data and actions with the cyberspace. Recommendation 10: Generalization of automation and sensor-driven information is of outmost in the next generation of prescriptive analytics. 6. Discussion and implications Using BDA in an industrial context allows offering several opportunities for manufacturing process. However, quite few companies that fully benefits from this proven potential especially for their manufacturing processes (Isaksson, et al., 2018; Choi, et al., 2017). In fact, despite the various benefits of BDA in industrial fields, its deployment still very limited since there is no clear documentation that groups together big data capabilities and classifies them according to each context and to each use case. On the other hand, the extant research work discussing the transformative potential of BDA is actually dealing with different aspects of the use of BDA in manufacturing context rather than manufacturing process. Other works focuses primarily on the practical side and proposes a technical model for the use of big data and they generally deal with one capability at a time. Nevertheless, most of these researches do not discuss the theoretical aspects without proposing any model or referential that represents the different BDA insights and capabilities. The present study aimed to present an overview of the different opportunities offered by using BDA techniques and offer a repository for researchers and companies interested in the enormous potential of BDA. To do so, we proceeded on three different axes: 

Firstly, we have conduct a systematic literature review on BDA in the context of manufacturing process where we selected and considered 68 papers from the mains scientific data bases. A critical analysis of literature was made based on a bibliometric and issues analysis to classify selected papers according to BDA aspects, trends and categories.



Secondly, we propose an overall framework of BDA capabilities in manufacturing process, which present a summary of manufacturing process challenges, bid data analytics faculties and capabilities and bid data analytics values. This architecture was proposed based on the analysis of relevant works proposed in literature.



Finally, we conduct a multiple case studies on a leading company that has already implemented BDA in their manufacturing process. The choice of the case studies was not arbitrary. In fact, we conduct case studies on companies that use BDA to deal with the different identified research trends in our critical analysis. The purpose of these case studies is to compare the obtained results from implementing BDA in these companies with results, affirmations and predictions found in literature. We aimed also to analyze the cases and give new lines of research advancement.

It is noteworthy that our paper is the first that combines a systematic literature review with multiple case study to offer a holistic overview regarding implementing BDA in manufacturing process. 6.1. Theoretical Implications The present paper yields some interesting insights for theoretical implications of using BDA in manufacturing process and contributes on its methodological literature. Thus, this study provides a broader understanding of BDA implications within the manufacturing process by conceptualizing various concepts related to BDA in manufacturing process in a holistic and data-driven manner. Actually, the findings from the systematic literature review can aid academic researchers to tackle new empirical research in this field and clear up BDA concepts in the context of manufacturing processes, which is still in the initial stage. To do so, a review, a classification and a summary of all relevant paper dealing with BDA permitted to complement a descriptive and bibliometric analysis. This analysis shows an exponential growth regarding the number of publications dealing with BDA in manufacturing process. Considering research trends it was proved from our analysis that research projects in the field of big data or industry as well focuses and investigates more and more on the topic of BDA especially during this last four years which represent 85% of total founded and selected publications. In general, the present manuscript permits to foster the heated and scalable debate of BDA in manufacturing process and contributes to a deeper understanding of this relevant technology from the systematic literature review and conceptualization of main capabilities. Furthermore, this paper regroup the main BDA challenges, faculties and values related to manufacturing process in a comprehensive way based on obtained results from our systematic review and based on conducted trends analysis which will permit researches to better understand BDA capabilities and opportunities. In this paper, we made a critical analysis on selected research projects and we classified them according to each Data Mining and Analytics categories (descriptive, inquisitive, predictive and prescriptive analytics).

Then we identified the different techniques used in each paper and divided them into two categories (offline technics & real-online technics). The results of our critical analysis lead us to identify the main research trends (BDA-enabler architecture, Real-time data mining approaches, Integrated Human-Data intelligence and Prescriptive analytics within the manufacturing process). Afterward we analyzed each identified trends in order to respond to the main BDA research questions with regard to performance management, production control and maintenance in manufacturing processes. The conducted analysis and conceptualization lead us to gives rise to a theoretical framework of BDA capabilities, which might contribute to redirecting the future research in this field. The proposed frameworks can be a good basis for all researchers that aim to understand the opportunities of using BDA and its limitations. The framework is composed of three related layers: The first layers aims to regroup BDA challenges on fourth distinct categories: quality and process control, energy and environment efficiency, predictive diagnosis and maintenance and safety and risk analysis. The second layer presents the BDA faculties in a comprehensive way. The main goal from building this layer is to demonstrate that just some few faculties can represent the enormous potentials of BDA. This approach will permit researcher to focuses on these faculties instead of being dispersed by analyzing all capabilities described in literature. The third layer concern BDA values and regroup main BDA opportunities into four categories: enhancing transparency, improving performance, supporting decision-making and enhancing knowledge. The objective is to offer an overall view on BDA opportunities based on different aspects and objectives of manufacturing industry. 6.2. Practical Implications The present paper yields some interesting insights for practical implications of using BDA in manufacturing process and offers some interesting results issued by conducting and analyzing some meaningful case studies. By mingling the findings of the systematic literature review and analyzed case studies, this paper has attempted to provide to researchers and managers meaningful knowledge on the formulation and implementation of BDA-enabler infrastructure in the manufacturing process environment: 

This study gives a detailed overview of some successful case study that may help manufacturing companies to transform into an agile and fully digitalized smart manufacturing unit while using BDA as an important pillar in industry 4.0 for manufacturing process.



In addition, this study may help manufacturing companies to implement fully IT-enabled BDA architecture to increase their performance efficiency, improve production quality, improve maintenance management, foster employee engagement and empowerment, and strengthen safety culture and execution in the long term.

The case studies carried out are in perfect concord with identified BDA research trends. The analysis of these Case studies allowed to leive the main recommandations while using BDA according to each trend. In another hands, the case studies analysis enabled to conclude that in order to successfully implement BDA in manufacturing context we must rely on some key transformation levers such as process and operations automation and the use of digital services and advanced analytics techniques. This finding represents a huge contribution since it provides essential recommandations to consider while building fully BDA capabilities. In addition, companies operating in similar conditions to that of the case companies can benefit from the findings of the cases studies to design their infrastructure that enable the use of BDA for their manufacturing process. 7. Conclusion and future works In contemporary era of Big Data, the utilization of data for manufacturing process intelligence takes on increased importance in the path towards the operational excellence. On the other hand, manufacturing processes and production tools have experienced whopping evolutions over time, boosting swift technological advancement all over the industry. Therefore, researchers on the topic argued that most companies admit the prominence of their data for monitoring their manufacturing process and have supported the use of enhanced analytics and business intelligence as a top priority in the near future. In this respect, the present paper has provided a better understanding of how these companies can harness the potential capabilities of Big Data Analytics (BDA) in their manufacturing process as a way of boosting the digital transformation to gain business value. Accordingly, the research begins with the identification of most agreed capabilities of BDA in manufacturing process through the findings of a systematic literature review to develop a framework of BDA capabilities in manufacturing process. Afterwards, the research use a multiple case study to corroborate this framework in order to provide a practical way for managers to disclose the potency of BDA along the interdisciplinary capabilities identified in the study. Like any other study, ours has quite a few of limitations that must be recognized. The first limitation of this study is the selection of papers to be reviewed. Actually, the restrictions posed during the search process (e.g. use of specific keywords, specific databases…) may contribute to left out some high-quality articles on BDA in manufacturing process. Furthermore, qualitative analysis perused to conduct the multiple case study is another limitation of this study. Indeed, the insights gained from the case studies are qualitative, which makes it difficult to go deeper and draw objective conclusions for the capabilities of BDA in manufacturing process. Future research researches may be performed on the following aspects. First, the evaluation of the impact of BDA capabilities on manufacturing performance with quantitative analysis method based on primary

data must be carried out using more in-depth empirical studies. Moreover, BDA-enabler infrastructure must be explored in specific contexts such as SMEs and service companies. It is worth noting that the research on BDA in manufacturing process, and the corresponding theoretical study has just started out, as a result of which there is a lot of work to be done driven by application requirements and related technologies before the BDA is successfully built in the manufacturing process. References

Amini, M. & Chang, S. I., 2018. MLCPM: A Process Monitoring Framework for 3D Metal Printing in Industrial. Computers & Industrial Engineering, Volume 124, pp. 322-330. Arunachalam, D., Kumar, N. & Kawalek, J. P., 2018. Understanding big data analytics capabilities in supply chain. Transportation Research Part E: Logistics and Transportation Review, Volume 114, June , Pages , pp. 416-436. Banerjee, A., Bandyopadhyay, T. & Acharya, P., 2013. Data analytics: Hyped up aspirations or true potential?. Vikalpa, Volume 4, pp. 1-12. Belhadi, A., Sha’ri, Y. B. M., Touriki, F. E. & fezazi, S. E., 2018. Lean production in SMEs: literature review and reflection on future challenges. Journal of Industrial and Production Engineering, 35(6), pp. 368-382. Belhadi, A., Touriki, F. E. & fezazi, S. E., 2016. A framework for effective implementation of lean production in Small and Medium-sized Enterprises. Journal of Industrial Engineering and Management, 9(3), pp. 786-810. Belhadi, A., Touriki, F. E. & fezazi, S. E., 2018. Benefits of adopting lean production on green performance of SMEs: a case study. Production Planning & Control , 29(11), pp. 873-894. Beneventi, F., Bartolini, A., Cavazzoni, C. & Benini, L., 2017. Continuous learning of HPC infrastructure models using big data analytics and in-memory processing tools. Design, Automation & Test in Europe Conference & Exhibition (DATE), pp. 1038-1043. Beyer, M. A. & Laney, D., 2012. The importance of ‘big data’: a definition.. s.l.:Stamford, CT: Gartner. Cheng, Y. et al., 2018. Data and Knowledge Mining with Big Data towards Smart Production. Journal of Industrial Information Integration, Volume 9, pp. 1-13.

Chen, H., Chiang, R. H. L. & Storey, V. C., 2012. Business Intelligence and Analytics: From Big Data to Big Impact. MIS Quarterly, 36(4), pp. 1165-1188. Chen, Y., 2017. Integrated and Intelligent Manufacturing: Perspectives and Enablers. Engineering, Volume 3, pp. 588-595. Chen, Y.-J., Fan, C.-Y. & Chang, K.-H., 2016. Manufacturing intelligence for reducing false alarm of defect classification by integrating similarity matching approach in CMOS image sensor manufacturing. Computers & Industrial Engineering, Volume 99, pp. 465-473. Chiang, L., Lu, B. & Castillo, I., 2017. Big Data analytics in chemical engineering. Annual review of chemical and biomolecular engineering, Volume 8, pp. 63-85. Chien, C.-F., Liu, C.-W. & Chuang, S.-C., 2017. Analysing semiconductor manufacturing big data for root cause detection of excursion for yield enhancement. International Journal of Production Research, 55(17), pp. 5095-5107. Chiremsel, Z., Said, R. N. & Chiremsel, R., 2016. Probabilistic Fault Diagnosis of Safety Instrumented Systems based on Fault Tree Analysis and Bayesian Network. Journal of Failure Analysis and Prevention, 16(5), pp. 747-760. Choi, T., Wallace, S. W. & Wang, Y., 2017. Big Data Analytics in Operations Management. Production and Operations Management. Chongwatpol, J., 2015. Prognostic analysis of defects in manufacturing. Industrial Management & Data Systems, 115(1), pp. 64-87. Chongwatpol, J., 2016. Managing Big Data in Coal-Fired Power Plants: A Business Intelligence Framework. Industrial Management & Data Systems, 116(8), pp. 1779-1799. Eckstein, J. et al., 2016. A Comparison of two Predictive Approaches to Control the Longitudinal Dynamics of Electric Vehicles. Procedia Technology, Volume 26, pp. 465-472. Favoino, F. et al., 2016. Optimal control and performance of photovoltachromic switchable glazing for building integration in temperate climates. Applied Energy, Volume 178, pp. 943-961. Gandomi, A. & Haider, M., 2015. Beyond the hype: Big data concepts, methods, and analytics. International Journal of Information Management, 35(2), pp. 137-144. Ge, Z., Song, Z., Ding, S. X. & Huang, B., 2017. Data Mining and Analytics in the Process Industry: The Role of Machine Learning. IEEE Access, Volume 5, pp. 20590-20616.

Grossmann, I. E., 2014. Challenges in the application of mathematical programming in the enterprise-wide optimization of process industries. Theoretical Foundations of Chemical Engineering, 48(5), p. 555–573. Gunasekaran, A., Yusuf, Y. Y., Adeleye, E. O. & Papadopoulos, T., 2018. Agile manufacturing practices: the role of big data and business analytics with multiple case studies. International Journal of Production Research, 56(1-2), pp. 385-397. Hammer, M., 2018. Digitization Perspective: Impact of Digital Technologies in Manufacturing. Management Approach for Resource-Productive Operations, pp. 27-68. Hammer, M., Somers, K., Karre, H. & Ramsauer, C., 2017. Profit Per Hour as a Target Process Control Parameter for Manufacturing Systems Enabled by Big Data Analytics and Industry 4.0 Infrastructure. Procedia CIRP, Volume 63, pp. 715-720. He, Q. P. & Wang, J., 2018. Statistical process monitoring as a big data analytics tool for smart manufacturing. Journal of Process Control, Volume 67, pp. 35-43. Hidalgo, N., Rosas, E., Vasquez, C. & Wladdimiro, D., 2018. Measuring stream processing systems adaptability under dynamic. Future Generation Computer Systems, Volume 88, pp. 413-423. Hutter, T., Haeussler, S. & Missbauer, H., 2018. Successful implementation of an order release mechanism based on workload control: a case study of a make-to-stock manufacturer. International Journal of Production Research , 56(4), pp. 1565-1580. Hu, W., Shah, S. L. & Chen, T., 2018. Framework for a smart data analytics platform towards process monitoring and alarm management. Computers & Chemical Engineering, Volume 114, pp. 225-244. Isaksson, A. J., Harjunkoski, I. & Sand, G., 2018. The impact of digitalization on the future of control and operations. Computers and Chemical Engineering, Volume 114, pp. 122-129. Ji, W. & Wanga, L., 2017. Big data analytics based fault prediction for shop floor scheduling. Journal of Manufacturing Systems, Volume 43, pp. 187-194. Joseph, R. C. & Johnson, N. A., 2013. Big data and transformational government. IT Professional , 15(6), pp. 43-48. Ketokivi, M. & Choi, T., 2014. Renaissance of case research as a scientific method. Journal of Operations Management, 32(5), pp. 232-240.

Khakifirooz, M., Chien, C. F. & Chen, Y.-J., 2018. Bayesian inference for mining semiconductor manufacturing big data for yield enhancement and smart production to empower industry 4.0. Applied Soft Computing, Volume 68, pp. 990-999. Khakzad, N. R. & Reniers, G., 2016. Application of Bayesian network and multi-criteria decision analysis to risk-based design of chemical plants. Chemical Engineering Transactions, Volume 48, pp. 223-228. Kho, D. D., Lee, S. & Zhong, R. Y., 2018. Big Data Analytics for Processing Time Analysis in an IoTenabled manufacturing Shop Floor. Procedia Manufacturing, Volume 26, pp. 1411-1420. Kozjek, D., Vrabič, R., Rihtaršič, B. & PeterButala, 2018. Big data analytics for operations management in engineer-to-order manufacturing. Procedia CIRP, Volume 72, pp. 209-214. Krumeich, J., Jacobi, S., Werth, D. & Loos, P., 2014. Big Data Analytics for Predictive Manufacturing Control - A Case Study from Process Industry. IEEE International Congress on Big Data, pp. 530-537. Krumeich, J., Werth, D. & Loos, P., 2016. Prescriptive control of business processes. Business & Information Systems Engineering, 58(4), pp. 261-280. Kumar, A., Shankar, R., Choudhary, A. & Thakur, L. S., 2016. A big data MapReduce framework for fault diagnosis in cloud-based manufacturing. International Journal of Production Research, 54(23), pp. 70607073. Lee, J. H., Shin, J. & Realff, M. J., 2018. Machine learning: Overview of the recent progresses and implications for the process systems engineering field. Computers & Chemical Engineering, Volume 114, pp. 111-121. Lee, J. Y., Yoon, J. S. & Kim, B.-H., 2017. A Big Data Analytics Platform for Smart Factories in Small and Medium-Sized Manufacturing Enterprises: An Empirical Case Study of a Die Casting Factory. International Journal of Precision Engineering and Manufacturing, 18(10), pp. 1353-1361. Li, D., 2016. Perspective for smart factory in petrochemical industry. Computers & Chemical Engineering, Volume 91, pp. 136-148. Lindström, J., Larsson, H., Jonsson, M. & Lejon, E., 2017. Towards intelligent and sustainable production: combining and integrating online predictive maintenance and continuous quality control. Procedia CIRP of The 50th CIRP Conference on Manufacturing Systems, Issue 63, pp. 443-448. Li, Y. & Kashiwagi, H., 2005. High-order Volterra Model Predictive Control and its application to a nonlinear polymerisation process. International Journal of Automation and Computing, 2(2), pp. 208-214.

López-Escobar, C. et al., 2012. Statistical quality control through process self-induced vibration spectrum analysis. The International Journal of Advanced Manufacturing Technology, 58(9-12), pp. 1243-1259. M’Sahli, F. & Matlaya, R., 2005. A neural network model based predictive control approach: application to a semi-batch reactor. The International Journal of Advanced Manufacturing Technology, 26(1-2), pp. 161-168. MacGregor, J. F. et al., 2015. Latent Variable Models and Big Data in the Process Industries. IFAC-Papers OnLine, 48(8), pp. 520-524. Manyika, J. et al., 2011. Big data: The next frontier for innovation, competition, and productivity, s.l.: McKinsey & Company. Mehta, P., Butkewitsch-Choze, S. & Seaman, C., 2018. Smart manufacturing analytics application for semicontinuous manufacturing process–a use case. Procedia Manufacturing, Volume 26, pp. 1041-1052. Ning, C. & You, F., 2018. Data-driven decision making under uncertainty integrating robust optimization with principal component analysis and kernel smoothing methods. Computers & Chemical Engineering, Volume 112, pp. 190-210. O’Donovan, P., Leahy, K., Bruton, K. & O’Sullivan, D. T. J., 2015. Big data in manufacturing: a systematic mapping study. Journal of Big Data, 2(20), pp. 1-22. Papacharalampopoulos, A., Stavridis, J., Stavropoulos, P. & Chryssolouris, G., 2016. Cloud-based Control of Thermal Based Manufacturing Processes. Procedia CIRP, Volume 55, pp. 254-259. Paul, D., Zhong, W.-D. & Bose, S. K., 2016. Energy efficient cloud service pricing: A two-timescale optimization approach. Journal of Network and Computer Applications, Volume 64, pp. 98-112. Qin, S. J., 2014. Process data analytics in the era of big data. AIChE Journal, 60(9), pp. 3092-3100. Qiu, X., He, G. & Ji, X., 2016. Cloud manufacturing model in polymer material industry. The International Journal of Advanced Manufacturing Technology, 84(1-4), pp. 239-248. Rowley, J. & Slack, F., 2004. Conducting a literature review. Management Research News, 27(6), pp. 3139. Sadati, N., Chinnam, R. B. & Nezhad, M. Z., 2018. Observational data-driven modeling and optimization of manufacturing processes. Expert Systems with Applications, Volume 93, pp. 456-464. Seuring, S. & Müller, M., 2008. From a literature review to a conceptual framework for sustainable supply chain management. Journal of Cleaner Production, 16(15), pp. 1699-1710.

Severson, K., Chaiwatanodom, P. & Braatz, R. D., 2015. Perspectives on Process Monitoring of Industrial Systems. IFAC-PapersOnLine, 48(21), pp. 931-939. Shao, Y., Li, C. & Luo, J. G. :. J. Z. Y., 2018. Efficient jobs scheduling approach for big data applications. Computers & Industrial Engineering, Volume 117, pp. 249-261. Shin, S.-J., Woo, J. & Rachuri, S., 2014. Predictive analytics model for power consumption in manufacturing. Procedia CIRP, Volume 15, pp. 153-158. Shrouf, F., Ordieres, J. & Miragliotta, G., 2014. Smart factories in Industry 4.0: A review of the concept and of energy management approached in production based on the Internet of Things paradigm. 2014 IEEE International Conference on Industrial Engineering and Engineering Management (IEEM), pp. 697-701. Shukla, M. & Mattar, L., 2019. Next generation smart sustainable auditing systems using Big Data Analytics: Understanding the interaction of critical barriers. Computers & Industrial Engineering, Volume 128, pp. 1015-1026. Sivarajah, U., Kamal, M. M., Irani, Z. & Weerakkody, V., 2017 . Critical analysis of Big Data challenges and analytical methods. Journal of Business Research, Volume 70, pp. 263-286. Smirnov, P. A. & Nasonov, D., 2016. Quality-based Workload Scaling for Real-time Streaming Systems. Procedia Computer Science, Volume 101, pp. 323-332. Stanley, G., 2018. Big Data Approximating Control (BDAC)—A new model-free estimation and control paradigm based on pattern matching and approximation. Journal of Process Control, Volume 67, pp. 141159. Stojanovic, N., Dinic, M. & Stojanovic, L., 2015. Big Data Process Analytics for Continuous Process Improvement in Manufacturing. IEEE International Conference on Big Data (Big Data), pp. 1398-1407. Sun, B. et al., 2017. Perspective for equipment automation in process industries. IFAC-PapersOnLine, 50(2), pp. 65-70. Tewari, S. & Dwivedi, U., 2019 . Ensemble-based big data analytics of lithofacies for automatic development of petroleum reservoirs. Computers & Industrial Engineering, Volume 128, pp. 937-947. Wamba, S. F., Akter, S. & Bourmont, M. d., 2018. Quality dominant logic in big data analytics and firm performance. Business Process Management Journal. Wamba, S. F. et al., 2015. How ‘big data’ can make big impact: Findings from a systematic review and a longitudinal case study. International Journal of Production Economics, Volume 165, pp. 234-246.

Wamba, S. F. et al., 2017. Big data analytics and firm performance: Effects of dynamic capabilities. Journal of Business Research, Volume 70, pp. 356-365. Wang, C.-H., Cheng, H.-Y. & Deng, Y.-T., 2018. Using Bayesian belief network and time-series model to conduct prescriptive and predictive analytics for computer industries. Computers & Industrial Engineering, Volume 115, pp. 486-494. Wang, G., Gunasekaran, A., Ngai, E. W. & Papadopoulos, T., 2016. Big data analytics in logistics and supply chain management: Certain investigations for research and applications. International Journal of Production Economics, Volume 176, pp. 98-110. Wang, W., Liu, X. & Qin, Y., 2018. A fuzzy Fine-Kinney-based risk evaluation approach with extended MULTIMOORA method based on Choquet integral. Computers & Industrial Engineering, Volume 125, pp. 111-123. Weese, M., Martinez, W., Megahed, F. M. & Jones-Farmer, L. A., 2016. Statistical learning methods applied to process monitoring: An overview and perspective. Journal of Quality Technology, 48(1), pp. 424. Yang, H. et al., 2014. A system architecture for manufacturing process analysis based on big data and process mining techniques. IEEE International Conference on Big Data (Big Data), pp. 1024-1029. Zerrouki, H. & Smadi, H., 2017. Bayesian Belief Network Used in the Chemical and Process Industry: A Review and Application. Journal of Failure Analysis and Prevention, 17(1), pp. 159-165. Zhang, H. et al., 2017. Progress in Aluminum Electrolysis Control and Future Direction for Smart Aluminum Electrolysis Plant. JOM, 69(2), p. 292–300. Zhang, Y. et al., 2018. A big data driven analytical framework for energy-intensive manufacturing industries. Journal of Cleaner Production. Zhang, Y., Ren, S., Liu, Y. & Si, S., 2017. A big data analytics architecture for cleaner manufacturing and maintenance processes of complex products. Journal of Cleaner Production, 142(Part 2), pp. 626-641. Zhong, R. Y., Xu, C., Chen, C. & Huang, G. Q., 2017. Big Data Analytics for Physical Internet-based intelligent manufacturing shop floors. International Journal of Production Research, 55(9), pp. 2610-2621. Zhuchkov, R. N., 2015. Application of predictive control approach in stabilizing control design of networked plants. Automation and Remote Control, 76(9), pp. 1704-1712.