Adoption of knowledge-graph best development practices for scalable and optimized manufacturing processes

Abstract

Using data analytics to properly extracting insights that are in-line to the enterprises strategic goals is crucial for the business sustainability. Developing the most fitting context as a knowledge graph that answer related businesses questions and queries at scale. Data analytics is an integral main part of smart manufacturing for monitoring the production processes and identifying the potentials for automated operations for improved manufacturing performance.

This paper reviews and investigates the best development practices to be followed for industrial enterprise knowledge-graph development that support smart manufacturing in the following aspects:

• •Decision for intelligent business processes, data collection from multiple sources, competitive advantage graph ontology, ensuring data quality, improved data analytics, human-friendly interaction, rapid and scalable enterprise's architectures.
• •Successful digital-transformation adoption for smart manufacturing as an enterprise knowledge-graph development with the capability to be transformed to data fabric supporting scalability of smart manufacturing processes in industrial enterprises.

Graphical abstract

Specifications table

Subject area:	Computer Science
More specific subject area:	Knowledge-Graph for Smart Manufacturing in the context of Industry 4.0
Method Name:	Adoption of Best Development Practices of Enterprise Knowledge-Graph for scalable Manufacturing Processes
Name and reference of original method:	- The article reviews and combines the different features of enhancements from different methods existing in the researches in literature for constructing, visualizing, scaling and friendly interacting enterprise knowledge-graph for optimized and scalable manufacturing processes. - Original method can be can be represented with the method name: Acknowledge-Graph for Industry 4.0. Reference: Bader, S. R., Grangel-Gonzalez, I., Nanjappa, P., Vidal, M. E., & Maleshkova, M. (2020, May). A knowledge graph for industry 4.0. In European Semantic Web Conference (pp. 465–480). Springer, Cham. - Also, the method can be represented with the method name: A knowledge-graph based data representation for IIoT-enabled cognitive manufacturing. Reference: Liu, M., Li, X., Li, J., Liu, Y., Zhou, B., & Bao, J. (2022). A knowledge graph-based data representation approach for IIoT-enabled cognitive manufacturing. Advanced Engineering Informatics, 51, 101,515.
Resource availability:	- Hubauer, T., Lamparter, S., Haase, P., & Herzig, D. M. (2018, October). Use Cases of the Industrial Knowledge Graph at Siemens. In ISWC (P&D/Industry/BlueSky). - Garofalo, M., Pellegrino, M. A., Altabba, A., & Cochez, M. (2018). Leveraging knowledge graph embedding techniques for industry 4.0 use cases. In Cyber Defence in Industry 4.0 Systems and Related Logistics and IT Infrastructures (pp. 10–26). IOS Press.

Introduction

Data-Driven is a wide concept and can be understood as business management based on information. A detailed analysis of dynamically and statically acquired data around a particular problem should be performed in best practices in order to increase the probability of success for the company's decisions. Recently, data-driven modelling has shifted towards the utilization of statistical inferences, machine learning to learn from previous experiences deep learning algorithms and other data-driven modelling frameworks, there are some good advantages of such shifting as design-optimization and real-time controlling with minimized computational complexities but the lack of the big data around specific business problem sometimes can lead poor generalization of the data-driven developed models. To overcome this problem, a prior domain-knowledge can be incorporated into the modern data-model development such as the deep neural network. Also, considering the data-driven graph construction as a graph can be considered as important mathematical tool to describe ubiquitous relations however, graph in many real-world scenarios can often suffer from uncertainty based on the fact that data associated with the vertices in the graph can be highly dimensional, noisy and not well-distributed with even no clear definition. Here also incorporating prior domain knowledge practices to the design of the data-driven graph will increase the efficiency of the subsequent analysis tasks and constructing a high-quality graph which is increasingly important in today's research in complex business analysis and decision-making.

The good understanding and successful adoption of knowledge graph implementation at certain manufacturing domain can be come in several benefits and added-value services. These include speed-up manufacturing processes with more accuracy and at scale also, allowing more friendly human interaction and continuous monitoring for improvement and adaptation. At the other hand, construction of reliable knowledge graph system is never being an easy task as it highly integrated system with database storage and SQL enquiry tools as well as visualization tools to simplify the ontology structure consisting of the vertices and nodes of the knowledge graph. Reaching mature level of adopting knowledge graph in manufacturing allows advanced analytics that facilitate scalability, interoperability, ease of answering critical business's performance related question and nicely visualizing the trends-of-interests for better business decision-making and strategic planning.

The coming sections in this paper are exploring and describing how to achieve intelligence manufacturing through optimized processes in the context of industry 4.0 and industrial successful digital transformation adoption. Different challenges need to be considered in order to reach to successful adoption of knowledge graph that allow friendly human interaction and optimized processes even in large scale. These highlighted challenges briefly introduced as interoperability issue as collecting data from multiple sources, the design of the graph ontology, allowing data profiling that ensure data quality and content and, the ease of obtaining business's insights based on continuous monitoring and alerting as well as visualizing in real-time the trends of interests. By successfully addressing these challenges for best practices adoption, the whole scale of manufacturing business under analysis can be transformed as data-fabric answering related questions, quickly visualize trends to conclude insights and at the end effective decision-making and business strategic planning can be achievable addressing the successful manufacturing digital transformation towards industry 4.0.

Decision support system for intelligent manufacturing businesses

Strategic goals and the businesses requirements of an enterprises are considered as the business path that is needed to be followed to ensure business sustainability. The main purposes of the data analytics are to extract insights that highlight potential opportunities or finding the useful business information by answering business questions and queries. DSS systems aid top management executives to quickly decide based on their observations of the trends of certain data of interests. Determining clearly these trends can be considered as complex and challenging tasks, especially in the cases of semi-structured or unstructured multiple source data environments. The visualization and clear 2D/3D (or even higher dimensions) data representation is also needed for effective DSS systems.

In modern DSS systems for smart manufacturing, the business environments can be identified as Cyber-Physical Social-System (CPSS) where all the aspects of the decision-support needed can be considered [1]. Information can be seamlessly transferred between the physical environment (sensors, actuators, robotics and computational engines) and the CPSS world with the aid of the sematic integration. The smart manufacturing communities can be considered as expert engineers/managers and machines/manufacturing tools and the communication between these communities can be happened vertically (different levels of expertise and applications) or horizontally [2]. The following figure illustrates the semantic of the CPSS smart manufacturing environment. It is always desirable to have CPSS environment as autonomous, sustainable and intelligent system.

.Also, this system has the ability of mapping resources as semantically rich forms to allow efficient communication and integration between machines and decision makers. Besides that, the capability of self-evaluation in each space should be presented.

Importance of building viable knowledge graph

Despite that modern organizations have undergone several technological transformations, developing a business strategy, cases and long-term roadmap still prone to be failed or embrace the change [3]. The most common obstacles facing enterprises in this regard can be explained as the following: unclear understanding for the business applications and use cases which limit clear vision and strategy, confusion on how and where to start regarding the most cost-effective and relevant use cases, multiple non-streamlined and optimized initiatives across the organization, redundant and not readily available enterprise data also, lack of required skills and training. Building the enterprise knowledge graph can enhance significantly the organization's capability to embrace long term-vision and roadmap as well as be adaptable to changes and future demands through the extraction of the valuable business's insights [4].

In today's complex business environments especially in large organization, the multiple IT systems, multi-system storage and taxonomical data architecture, all can lead to data inconsistency and duplication. This can make it very difficult to have a trusted view of customers and business. Building knowledge graph can be highly beneficial to provide a unified view of the enterprise and customer data arranged as a global schema [5]. This allows the capabilities for capturing the interrelationships of different data items coming from different databases. Moreover, the added-advantage can be obtained is the ability to gain insights about business environments and customers’ demands which can be utilized for analytics of multiple business use cases.

Understanding data collection from multiple source systems

Data collection from distributed multiple sources is the most important step in data integration and fusion towards building knowledge graphing [6]. Two important considerations should be taken into account here, first mapping that outline the structure of each source and how the data residing in. Another important aspect to be considered is the business requirements and metrics that should be simultaneously highlighted and confirmed as they important in determining how the data will be used from the beginning [7]. The scope of the knowledge graph can be easily understood in the line of the combination of the business requirements and mapping of the multiple data sources [8]. Based on this combination and with the power of advanced business analytics, insights can be extracted from collected data and these insights deriving and ensuring implementable business solutions as outcomes. The following figure depicts the manufacturing enterprise's data sources variety and the achieved data insights variety.

Designing the competitive advantage graph ontology

Ontology is mainly about how the data is contextualized. It can be considered as a set of vertices and edges that can map the data attributes to their relevant data schema [9]. Here, vertices representing the real-world entities and relationships between these entities are representing through the edges and with these vertices and entities, a comprehensive context start to be growing and forming the knowledge graph. The sooner this is can be done, the better it is as the right-contextual schema can be written. This is highly collaborative and communicative efforts across the team. Designing a powerful knowledge graph, that aids the easy to access hidden interactions between the different data entities and that allows interesting self-learning properties empowered by AI/M [10]. With competitive and robust knowledge graph design, hundreds of connections can be comfortably surveyed and different insights aspects related to operational processes, products, customers, events, 3rd parties and suppliers can be easily extracted. This can be depicted as inFig. 2 shows.

Fig. 1.

Smart factory hierarchical architecture (Baotong Chen, 2017)

Fig. 3.

Automotive supply chain data in a visual graph from Neo4j (Source: AI Journal).

Fig. 4.

a. Data quality components. b: different stages for polishing and enhanced data quality.

Fig. 2.

Data harmonizing to form enterprise knowledge graph for multiple business insights.

In brief; if ontology as the knowledge graphs entity relationship schema done properly then it can support unmanned and long-term self-learning evolving as linkage between multiple data domains. As conclusion remarks from the literature, there are some best practices to be followed in designing efficient enterprise knowledge graph. These practices are focusing on the processes of the design and implementation can be followed to leverage the enterprise knowledge graph and they can be explained as the following: First; define clearly the business purposes of developing the enterprise knowledge graph. Secondly; explore the different aspects of the three layers’ views as data ingestion, data storage and knowledge consumption and then tune them to the specific business requirements. The first layer as data ingestion and integration there are two pieces to be considered here as importing data and content into the graph database and this called as ETL processes (extract, transform and load) or keep the data in-place in its source system. This approach establishes a repeatable process that extracting relevant data from multiple sources at a regular basis and then populate and update the enterprise knowledge graph according to that. Additional steps can be applied also for ETL approach as sematic tagging and applying the ontology before loading data to the graph databases. ETL approach is suitable for extracting data from unstructured content.

Data profiling ensuring data quality and content

Data profiling is happening quickly after the data start flooding into the knowledge graph to ensure data integrity, accuracy and completeness. Data profiling can be considered as validation and quality assurance to make sure that the predefined rules and standards are consistently preserved and allow discovered anomalies [11]. When data is gathered from multiple source systems, data profiling become more important issue to ensure that the quality and consistency are not being compromised [12]. The following figure, illustrates the big data quality components.

After data polishing and optimized quality according to end-client's needs then, data profiling take place and can be considered in different profiling aspects and stages as the following Fig. 5 shows.

Fig. 5.

Aspects of big data profiling challenges.

It is recommended practice to be followed is that, to develop a data profiling solution in the phase of the knowledge graph construction that quickly validate the pre-designed ontologies to make sure that once the graph is built then it can be flooded with quality data.

Knowledge graph insights to enhance the performance of manufacturing use cases

To achieve highly useful knowledge graph insights to accelerate manufacturing intelligence, these insights should be easily feedback to the current environment (under analysis). knowledge graphing is evolutionally development and the first set of the ontologies design is pivotal [13]. These first ontologies should be moved forward and tuned to address the right use cases of the manufacturing business under analysis.

A good example of how industrial knowledge graph is very effective in enhancing the manufacturing performance in varied and diverse business cases, Siemen's industrial knowledge graph applications are highly worthy to be investigated in this context. Siemens developed a general graph infrastructure and platform for implementation called metaphactory supported with graph databases as Amazon Neptune. This complete package solution that can support scalability of the graph database backend to enable querying processes and analytics, data integration at both level of physical and virtual as a dedicated data integration layer also, components for end-user interactions (as virtualization, semantic searching and data exploration) as well as for expert user components to maintain the knowledge assets such as ontologies, data instances and queries. The knowledge graph middleware component contains the components and services for rapid applications’ development for specific solution with interfaces ability to third-parties. This industrial knowledge graph complete package solution can be utilized for various user groups and applications’ use cases as building digital twin considered as integrated information space for building management and services. The solution also can be utilized for risk management as planning and constant monitoring of all possible risks. Processes monitoring and machine services operations are great capabilities enabling optimized production with clear machine maintenances strategies and cases [13]. The good design of the knowledge graph can automate the extraction of use data content for fast as real-time reporting and delivery for actionable-insights.

Usually organizations have been built as the management of the data is done by relevant departments. Also, with the capability to view and connect data as same way human brains perform inferences to conclude about specific business queries or problems. It is recommended here that organizations ask themselves on how much their data and information are distributed across business's units and departments and also how many clicks and reports are needed to be navigated to find an answers for specific business problem or to get the relevant results for particular search. Figuring out that, helps organization to understand how much a good design for the enterprise knowledge graph can add the advantage for their operation by keeping them almost updated on real-time basis about different business cases scenarios and problems.

Enterprise knowledge graph integrated digital twin in context of smart manufacturing

The processes of constructing knowledge-graph based data integration in the vision of industry 4.0 can be considered as compromised of four different steps. These steps can be highlighted as data sources definition, ontology creation, data profiling as mapping data sources with ontology and then harmonizing data in knowledge graph [14]. According to the vision of industry 4.0, smart factories involve different communications as intelligent human-to-machine and machine-to-machine [15]. To achieve maturity in such types of communication, devices such as actuators, sensors and cyber-physical with their associated data need to be defined and described. Interoperability conflicts resulted from the different semantic representations of these devices can be considered as a challenge to be solved. For effective and efficient communications in smart factories, frameworks for different standards and standardization need to be proposed and utilized to enable the description of the main properties of systems, components, processes and the interactions between them [16].

In industry 4.0 standards there are numerous traditional as well as new and upcoming standards. Traditional standards as simply for design, manufacturing automation, quality control, … etc. newly standards for industry 4.0 digitalization for technologies such as digital-twin, industrial IoT, asset managements and others. The standards landscape is huge and somehow is difficult to be understood therefore, standards classification frameworks are needed.

In brief; standardization frameworks classify, align and interaction with different industrial standards according to industrial enterprises purposes and their features, RAMI 4.0 or IICF are examples of standardization frameworks by the industrial 4.0 communities [17]. To tackle the problem of standard operability amongst different standardization frameworks and device knowledge, a knowledge-driven approach is developed as industrial 4.0 knowledge-graph (I40KG) as universal description of standards and standardization framework as I40KG integrate more than 200 standards and standardization frameworks. Ontotext I40KG platform is a sematic approach as integration of standardization efforts and reference initiatives into enterprise industrial knowledge graph [18]. The industry 4.0 knowledge graph ontology is utilized for the representation of main characteristics of industry 4.0 as standards, norms and reference frameworks and the relationship between them. SPARQL and GraphQL are used to explore Ontotext's I40 KG (as a semantic approach for the integration of standardization efforts and references). First, semantic object model schema (SOML) as first draft schema from the I40KG ontology need to be generated [19]. This drafted SOML schema contains a number of additions and fixes to be activated to generate the I40KG schema.

Closed-Loop digital twin for continuously improving data analytics

Closed-loop digital twin or a fully integrated digital twin empowered by industrial internet of things (IIoT) platform, is a crucial step toward enhancing each aspect of product development and production processes [20]. The closed-loop digital twin is capable to transform the entire manufacturing value chain starting from product design to production, to performance and monitoring services, to logistics. Also, that can be extended to consider the entire manufacturing facilities and even the supply chain. Translating the different processes to knowledge-base can be obtained from the fully integrated closed-loop digital twin architecture into business analytics crucial to enhance the return of investment for manufacturing enterprises [21]. The benefits of such achievements can be highlighted as the following: Feedback loop which can be utilized as the feature to allow real-time performance data in productions lines to be immediately and continuously applied for product design and manufacturing processes [22]. In integrated digital-twin the feedback from each process can be conveyed in real-time through the system to other parts of related processes, resulting in highlighting areas to be improved for increased competitive advantage.

This is particularly important in modern market-places which can be described as fast-changing customers’ demands. This helps the manufacturing enterprises to be more agile and flexible so that they can be more sustainable in dynamic and complicated markets environments.

Increased innovation and product quality as fully-integrated digital twin infrastructure helps manufacturers to introduce new products faster to markets [23]. The real-time sensors collected data can deliver vital data allowing engineers to quickly identify ways to improve existing products or even create new ones before even the end-clients realize that they need it. Also, these data empower end-customers with the required information on most efficiently use the product thus improve their uses.

Production quality and compliances in terms of designs and production processes can be significantly enhanced with the proper adoption of closed-loop fully integrated digital twins as the data produced by machines can be instantaneously gathered to compare the expected and actual performance. This can help significantly to pinpoint anomalies, defects, design flaws and take countermeasures.

Closed-Loop Digital Twin enhances processes’ efficiency on the plant floor environment as well as across the entire manufacturing process. This can be achieved as simulating every aspect of the plant reduces the time consuming for physical prototyping by ensuring the optimal configuration of the plant's layout, control system and process at the design stage of a factory [24]. In brief; production at the level of digital twins accelerate and validate manufacturing processes, planning and commissioning. With closed-loop digital twin, it is easier and more quickly to optimize assets and work at the factory floor level enabling more efficient processes, reducing costs and downtime and optimizing the workflow with the capabilities to predict when predictive maintenance is needed.

Enhanced processes standardization and compliances as the closed-loop digital twin enable the establishments of detailed processes standards which aids the efforts on how the products built and closely monitored in real-time [25].

As a conclusion; integrated digital-twin systems and processes convey a competitive advantage and can be considered as the backbone for future intelligent manufacturing for delivering innovative products efficiently and quickly to markets.

Human friendly industrial knowledge graph based deep learning

The recent advancements of deep learning allow wide-range of interesting applications for human-machine real time interactions and collaboration. More specifically, deep learning as a natural language processing allowing the development of question & answer knowledge graph and that can be utilized with great potential for intelligent conversational agent and recommender systems. Connecting deep learning to knowledge graph empower natural query with specified data entities and attributes to retrieve answers [26]. At the stage of building out a knowledge graph team, it is important to look to the adjacent skills and experiences of the standards in the semantic web and knowledge representation space (such as RDF, OWL, and SPARQL) or graph databases (such as Neo4j and Amazon Neptune) [27].

In their paper titled “FAIR and Interactive Data Graphics from Scientific Knowledge Graph” [28], Authors describe a virtual exploration way for knowledge graph which can be accomplished by interactive data graphics (bar charts, scatter plots, etc.). This accomplishment done as modeling a chart representing a set of metadata describing visual exploration (combination of SPARQL & queries) as separately from visual context (Vega-lite specifications) then, the interactive data graphics can be synchronized to knowledge-graph in web-based environment.

Machine learning for rapid & scalable enterprise knowledge-graph architectures

Innovations related to business data analytics is expected to be significantly increased and fuelled by 70% with the power of knowledge graph in 2025. It's also expected that, 30% of organizations will utilize knowledge graph for their decision-making process in 2023 [29].

Considerable benefits can be obtained by using machine learning algorithms (as useful format for feeding training data) into knowledge graph as quick building and structuring of graph databases, drawing connections between data points to make these data noticed and visualized. These enhanced capabilities of the knowledge graph, make it the suitable development methodology for recent applications for internet searching, recommendation systems, intelligent conversational agents (chatbots) and e-commerce sites. The capabilities of the knowledge graph can be used for product description, enhanced customers’ experiences through virtual assistants for booking and delivery scheduling. It is obvious that, knowledge graph with machine learning accelerate business analytics and enable the speed-to-insights and this will lead to significant increase of the adoption of Intelligent knowledge graph technologies in near future.

Modern knowledge graphs are well-suited for scalable enterprises architectures [30]. This can be understood upon knowing that the knowledge graphs usually are built upon ontological concepts as part of the semantic web with the flexibility to be adopted for more expressive hierarchical systems. Based on that, modern knowledge graph is fit for AI-driven tasks and become widely adopted in big organization such as Google, Apple, Uber and others. The knowledge graph and machine learning are intertwined at the stage during the construction of a knowledge graph and this leads to the enhanced outputs of the data-driven machine learning models as knowledge extracted from the graph. New insights can be extracted from the knowledge graph by learning to classify/cluster nodes and prediction of the missing connections. Machine learning can help in building and maintaining knowledge graphs while knowledge graphs provide context to improve machine learning.

The main question here to be considered is how organizations can achieve accelerated time-to-insights and improved decision-making process with knowledge graph [31]. To answer this important question, first; it is important to understand that, the knowledge graph can describe, represent and link all enterprise data regardless of the data sources and structures, this can be considered as added-value services for data consumers in terms of automated services to enjoy access, visibility and understanding all the available information in the enterprise. Knowledge graphs can be built straight from relational, semi-structured and unstructured data allowing the initial data integration to be achievable as click of button to form the interconnected knowledge graph and make it ready for analysis. Usually to build a robust data integration, that requires a hybrid approach as automation and human interactions are used to knit the data fabric.

From knowledge-graph to data fabric for enterprise-scalable data integration

Nowadays, the enterprise data landscape is increasingly getting hybrid and varied and this exposing the challenge of getting relevant data from hugely unstructured data volume from the external data sources [32]. This is getting more and more due to the emerging technologies of IoT and the recent trends of hybrid multi-cloud environments. Old data strategies focusing on relational data systems are not feasible or practical. There is a need to develop a data fabric architecture where a network of information can be created from the internal and external enterprises’ data sources to gain the power of developing businesses added-values applications for analytics also for AI and decision-support systems. Data fabric allows accessing and sharing of data in a distributed network environments and in another way is the development of platforms for decentralizing data assets management [33]. With the capabilities of data fabric to combine various kinds of data and to allow them to be accessed and governed effectively and efficiently then, the data fabric architecture can be considered as an enterprise most valuable asset.

Empowering data fabrics with knowledge graphs can be good practical solution to achieve agile and scalable enterprise's data integration [34]. The knowledge graph facilitates the collection of interlinked descriptions combining entities, objects of events and concepts in certain context by linking the semantic metadata. This can be utilized later by data fabric as modern data management architecture allowing organizations to flexibly use their data to gain quick data analytics, data-driven insights. The added-advantage can be obtained is considered as proactive preparing, integration and modelling from data coming from numerous and diverse enterprise sources in an integrated platform.

Conclusions

In today complex manufacturing businesses and marketing environments, emerging technologies such as IoT, digital twin, artificial intelligence and cloud computing, connect organizations with multiple sources data. Driving business analytics as data-driven decision-making to properly set strategic planning and solving business problems is a huge challenge to be achieved. To overcome this challenge to automatically obtain relevant information, data fabric empowered knowledge-graph can be customized to support the architecture managing data integrations and accessibly at large-scale. It is highly important recently for manufacturing enterprises to adopt best practices for competitive knowledge-graph design and implementation to be highly agile in achieving high products and productions performance.

CRediT authorship contribution statement

M.S. Jawad: Conceptualization, Methodology, Software, Writing – original draft. Chitra Dhawale: Data curation, Writing – original draft. Azizul Azhar Bin Ramli: Visualization, Investigation. Hairulnizam Mahdin: Writing – review & editing, Writing – original draft.

Declaration of Conflicting Interest

The authors that they have no known competing financial interests or personal relationship that could have appeared to influence the work presented in this paper

Data availability

• No data was used for the research described in the article.