Challenges and research agenda for realizing a seamless integration of Digital Shadows into the production domain

One vision of industry 4.0 is an autonomous shopfloor configuration. A possibility to transfer this vision to reality is realizable with Digital Shadows. As Digital Shadows are well defined in research, challenges for seamless integration in production are still an open research topic. The main challenge is the enrichment of semantics on the domain in a standardized format that enables multilateral communication between all assets. To overcome these challenge, tools have to be developed to support domain experts in building and integrating Digital Shadows without knowledge of modeling and programming. This article demonstrates the main challenges of this objective and proposes a research agenda to close this research gap.


INTRODUCTION
The fourth industrial revolution (industry 4.0) was introduced in the last decade. One significant difference between industry 4.0 and previous industrial revolutions is autonomous production systems. Although fully autonomous production systems may be disadvantageous, for example, regarding the process understanding of humans, specific tasks and processes can be partially autonomized. 1 The central idea of industry 4.0 is to connect the physical shopfloor and the virtual world toward cyber-physical-production systems (CPPS) to be more efficient in production or its support processes. To achieve this, CPPS enables (semi-) autonomous decision-making between single assets that ideally leads to shorter response time and optimal resource utilization due to transparent and comprehensive providence of relevant information. 2 The main prerequisite for enabling CPPS is the availability of data and information and the capability for non-humans to understand and interpret those data and information to gain knowledge and derive adequate actions. 3 Therefore, essential building blocks for realizing CPPS are Digital Twins and Digital Shadows. While a Digital Twin represents an asset in the virtual world, a Digital Shadow supports decision-makers in their daily work by providing data-based decision-support to a specific task. However, integrating Digital Shadows into the production domain is linked to multiple challenges, especially considering the integration of semantics, simplicity, modularity, and scalability. This article marks visions, the state-of-the-art, and the corresponding key challenges for the seamless integration of Digital Shadows into the production domain. It presents necessary research activities to transform these visions into reality using the example of PPC in the injection molding domain.

COMPARISON OF DIGITAL TWINS AND DIGITAL SHADOWS
The main driver of industry 4.0 is data, information, and knowledge. Hence, a virtual representation of all production assets is required. This holistic virtual asset representation calls Digital Twin. The pure virtual representation of assets is not sufficient in an industry 4.0 environment. Instead, the allocated and gathered data should be used to provide decision-support for typical production-related tasks, for example, shopfloor optimization or predictive maintenance. Therefore, Digital Shadows are introduced that aims for a purpose-driven decision-support for such a specific task. Figure 1 illustrates the differences between a Digital Twin and a Digital Shadow.
The Digital Twin provides the asset's properties in the virtual world. These properties represent the assets' core data, their current status, and their capabilities. 4 While the Digital Twin reflects the relevant asset data in the virtual world, the Digital Shadow uses this data to provide decision-support. Since normally only a small set of data, compared to all data an assets' Digital Twin comprises, is needed to perform a specific action, Digital Shadows extract exactly the required data to fulfill the task set by a decision-maker. Furthermore, the Digital Shadow uses suitable models to transform the extracted raw data and information into knowledge. The decision-maker can then implement the results. An exemplary purpose could be the decision-support for optimal scheduling within production planning and control (PPC). Therefore, the Digital Shadow uses suitable models to fulfill the given purpose. Depending on the formulated purpose, multiple results can be generated since fulfilling a given purpose can be done in different ways. An example is the (optimal) scheduling for production orders on the shopfloor. Due to different optimization objectives, for example, minimization of adherence to due dates or maximization of machines utilization, different schedules that are optimal regarding their optimization objective can be calculated. 5,6 The Digital Shadow provides the decision-makers with all optimal solutions to decide which option should be realized. In contrast to Digital Twins, Digital Shadows do not directly affect the system and only use the data the Digital Twin provides. 7 Digital Shadows are not limited to a single domain or company and can be part of a worldwide, cross-domain collaboration. 8 Digital Shadows are one building block for industry 4.0 since it autonomously supports data-based decision-support. Ideally, the decision-maker only formulates the purpose of the Digital Shadow. Then, the Digital Shadow knows the corresponding data and its location to be extracted and suitable models to be used. However, integrating Digital Shadows into production that are concomitant with the requirements of industry 4.0 is linked to multiple challenges, especially considering simplicity, modularity, and scalability.

RELATED WORK ON DIGITAL SHADOWS
In this section, related work of Digital Shadows will be presented. Today, Digital Shadows and their building blocks are well-defined in the literature. For example, Kritzinger et al. characterize a Digital Model, a Digital Shadow, and a Digital Twin by their different type of data flow from the physical to the virtual object and vice versa. 7 A coincide definition of Digital Shadows was made by Bauernhansl et al. 9 and Becker et al. 5 They characterize a Digital Shadow as an information supplier for solving specific tasks within the production domain. Moreover, Becker et al. present a meta-model that consists of the Digital Shadows' building blocks. 5 Bauernhansl et al. derive challenges and propose a two-dimensional roadmap for integrating Digital Shadows. They argue that one of the key challenges for integrating Digital Shadows is to provide information of the right quality using semantic web technologies without going into detail. 9 Connecting to that, Scholz et al. propose research directions comprising the intersection of smart manufacturing and geographic information science. They point out standardization gaps within this perspective and present possible solutions to overcome them, for example, by using ontologies. 10 In addition, Zhu reflects that using knowledge graphs can be helpful for the semantic linking of data, models, resources, or standards in the domain of geospatial analytics. 11 It can be seen that considering semantics is also a highly relevant topic in domains other than the production domain. Schuh et al. identify suitable quality management of databases used by Digital Shadows as a core challenge. To overcome this challenge, they suggest a methodology that identifies the need for conducting a data quality project for the corresponding databases. 12 Research on Digital Shadow implementation points out that mainly a single, encapsulated use case was examined. In particular, industry 4.0-specific requirements, that is, integration of semantics, simplicity, modularity, and scalability for implementing Digital Shadows, are rarely addressed. Sample Digital Shadow integration can be found for pig rearing, 13 PPC in injection molding, 14 pasture management in the farming domain, 15 or for building information management. 16 Furthermore, comprehensive standards for building Digital Shadow are absent. Since a comprehensive integration of Digital Shadows into the production domain is not yet realized, a research agenda that will consider industry 4.0 requirements, especially the use of semantics and standards, will be provided in Section 5.

VISIONS, STATE OF THE ART, AND KEY CHALLENGES ON DIGITAL SHADOWS
In terms of the fourth industrial revolution, three visions regarding the application of Digital Shadows under consideration of different perspectives will be introduced in this article ( Figure 2).
From an end-user perspective, the first vision enables end-users (in most cases domain experts) to build and integrate Digital Shadows with no (or low) programming effort (e.g., by using an interactive, web-based interface) and in a modular and scalable manner. A vision from a business perspective enables new business areas in the form of ventures, for example, Digital-Shadow-as-a-Service. From a technical perspective, the second third vision is a seamless integration of Digital Shadows into the production environment. Seamless integration means an automated integration of Digital Shadows with underlying databases, vendors, and communication protocols. This contribution focuses on the seamless integration of Digital Shadows since it is the prerequisite for fulfilling the other two visions. Below, these prerequisites will be presented and analyzed concerning its state-of-the-art. Subsequently, the key challenges will be derived.

Prerequisite 1: Ensuring the availability of relevant data
Since the Digital Shadow should provide data-based decision-support for a specific task, all relevant data must be available digitally to fulfill the given purpose. It can be concluded that the technology for transferring data into the virtual world through established software systems like enterprise resource planning and technology like sensor or RFID technology is commonly present. 17 As technologies for transferring data into the virtual world are state-of-the-art, especially small F I G U R E 2 Perspectives for realizing a seamless integration of Digital Shadows into the production domain.
enterprises often lack in the usage of software systems. The main reasons are a lack of resources, that is, budget, operators, and maintainers for running such a system. 18,19 Summarized, the first challenge is that relevant data must be available in the virtual world. The second challenge is providing easy-to-use and low-cost solutions even for small companies that lack information technology engineers.

Prerequisite 2: Selecting relevant models for shopfloor optimization
Conducting shopfloor optimizations, models must be selected that transfer the gathered data into knowledge. The Digital Shadow is not restricted to the applied model type. Nowadays, multiple models for PPC purposes are still available. For example, models for scheduling (i.e., TSP, VRP) 20 or lot sizing. 21 Nevertheless, those models often are absent in the digital world, so transferring the models to the virtual world can be challenging. In addition, using self-defined models have to be accessible for the Digital Shadow. This could be challenging, especially when self-defined models are encapsulated in vendor-locked applications, for example, Excel.

Prerequisite 3: Enriching data and models with semantics
One central vision of integrating Digital Shadows is the autonomous gathering of data and selecting suitable models to perform decision-support. Therefore, the Digital Shadow must "know" context information regarding the data and models. Hence, data and models must be enriched with semantics, that is, assigned with unique identifiers and metadata (e.g., units, synonyms, model type). Furthermore, those data and models should be provided in internationally accessible dictionaries to enable a common understanding.
Existing semantic web technologies that model the assets' properties and their interconnection are ontologies. 22 A significant challenge is building such ontologies under consideration of existing ones. Domain experts should not build an ontology from scratch but (partly) reuse existing ontologies due to reduce the effort of building ontologies and improve the interoperability of existing ontologies. This can be realized with Ontology Design Patterns (ODP). ODPs are small building blocks that support ontology design by providing specific modeling solutions. 23 However, due to the large number and less documentation, ODPs are not yet widely considered in practice for building ontologies. 24 A further challenge in building ontologies is the need for software and programming skills. In general, domain experts lack programming and computer science skills but have immense knowledge of their domains' assets and processes. In particular, small and medium-sized enterprises have no resources for ontology experts. Thus, modeling ontologies is a challenging task since it requires skilled software engineers and profound collaboration with domain experts.
Research on related work regarding industry 4.0 shopfloor configuration and optimization points out that foundations for semantic enrichment are already state-of-the-art. In the field of standards and guidelines for the semantic enrichment of data, IEC 61360 (Standard data element types with associated classification scheme), 25 and IEC 62656 (Standardized product ontology register and transfer by spreadsheets) 25,26 exists. Furthermore, open dictionaries as eCl@ass *27 are established. Nowadays, the consideration of semantics in practice often is missing. The key challenge comprises establishing standardized dictionaries that include all assets and properties within the production domain. Open dictionaries are only reasonable when most companies use them. Ideally, assets and their properties are defined only once, under consideration of standards, and then reused by other organizations. Today, this definition process is complicated since no GUI-based tools exist to support the engineers, and many standards must be considered. Furthermore, adding new properties must be feasible, including verification of already existing properties to avoid multiple dictionary entries with the same meaning. Only integrated and comprehensive dictionaries support the vision of industry 4.0 and, subsequently the seamless integration of Digital Shadows.

Prerequisite 4: Establishing the flow of data
After the providence of data and models that are enhanced with meta-data, the flow of data has to be established. Therefore, using suitable communication protocols is necessary to set up interfaces to corresponding software, databases, or sensors. Transferring data across the internet is state-of-the-art. One famous example is the TCP/IP protocol. 28 Likewise, the existence of industry 4.0 qualified protocols and standards that could enable the data and information flow, for example, OPC UA, MQTT, or CoAP, are common business. 29 Challenges to enabling a data flow occur by the huge amount of available protocols, especially when different vendors use different protocols. One challenge is the establishment of a connection between the single assets that allows the data flow intuitively. Other challenges result from the character of the Digital Shadow, namely, identifying the relevant data points and sources for processing. Subsequently, the challenge is an automatic providence of these relevant data points. Connected to that, another challenge is deriving the correct model(s) from the given purpose the Digital Shadow must fulfill.

Prerequisite 5: Realizing cross-domain data transfer
A cross-domain data transfer supports the vision of a globalized network that deals with data, information, and knowledge. The data transfer within single plants and across domains, companies, and national borders needs to be established. In addition, the geographical domain should be considered, for example, localization services for shipping goods, to gain transparency of the entire supply chain. 10 Thus, a challenging task will be aligning companies (and even competitors of the same domain) with respect to a common dictionary. It can be identified that cross-domain data transfer is only enabled for specific purposes. For example, EDI (electronic data interchange) can be used for the standardized and paperless exchange of business-related documents like invoices. 30 Other results are OPC UA information models like EUROMAP, which comprises standardized properties for the plastics processing industry. Another challenge originates from the end-user perspective. Here, the challenge is to provide a tool that allows domain experts to build Digital Shadows without the need to consider underlying systems intuitively, so the vision of seamless integration of Digital Shadows can be fulfilled. In summary, the general foundations for seamless integration of Digital Shadows are mostly present, that is, existing standards, guidelines, semantic web technologies, or communication protocols. An open issue is connecting the single building blocks and creating specific instances for dictionaries and ontologies that fit the production domain. Moreover, software and domain experts need to collaborate since establishing Digital Shadows is an interdisciplinary task. The key challenges pointed out before can be transferred to the following research question: How are domain experts able to build Digital Shadows in an efficient way that can directly be integrated into the production environment and automatically interact with the underlying infrastructure? F I G U R E 3 Research agenda for the seamless integration of Digital Shadows into the production domain.

RESEARCH AGENDA FOR TRANSFORMING THE VISION TO REALITY
Using the example of PPC in the injection molding domain, the following research agenda is proposed to answer the research question pointed out before ( Figure 3): In the first step, ontologies and dictionaries have to be formulated under consideration of IEC 61360 (standard data element types with associated classification scheme) 25 and other guidelines. The scope of a first mid-level ontology is the PPC domain that mirrors the general assets on the shopfloor, for example, production order, raw material, or machinery, independent of the manufacturing process. After this, a second domain-level ontology represents domain-specific assets, for example, injection molding machines. These ontologies can be merged since injection molding machines from the domain ontology are a subclass of machinery in the mid-level ontology. For simplifying the ontology building process regarding IEC 61360, tools have to be developed that support engineers in enriching semantics to the relevant data points. Parallelly, a semantic classification of PPC models needs to be developed. In this case, suitable PPC models should be integrated into a model catalog. This catalog classifies the included models (e.g., lot-sizing), so the Digital Shadow can retrieve the correct models. One critical building block is the data and information flow to be enabled. In the PPC domain, multiple databases often exist. Since the Digital Shadow and the used models must know where the desired input data is located, an instance like an accessible repository is necessary. In industry 4.0, asset administration shells (AAS) fulfill this task. Based on the ontology, the AAS represents the instantiated asset and current conditions in the virtual world and provides the data points' endpoint (for example, an AAS of a specific mold can provide its current location). Hence, the AAS can be defined as an implementation of the Digital Twin. 31 After validating a use case inside PPC in the injection molding domain (concerning the prerequisites in Section 4), a second PPC use case, but in a different domain than injection molding, should be proven. As before, the new domain ontology can be merged with the mid-level PPC ontology since their classes act as superclasses. After that, a roll-out of the results, namely ontologies, AASs, and the IEC 61360 tool, should be labored to motivate other use cases to participate. In the final step, a tool for domain experts has to be set up that allows an accessible building and seamless integration of Digital Shadows into production. As a basis, this tool uses the developed ontologies, AASs, and the model catalog. Ideally, the tool provides domain experts with all available data points provided by machines and software systems, so they can connect them with PPC models easily and enable the desired data flow. The benefit for domain experts who use the tools is a fast setup of executable Digital Shadows.

CONCLUSION
Whereas Digital Shadows are well defined in theoretical research, a lack of seamless integration of Digital Shadows in production can be identified. Challenging tasks comprise the support of domain experts in building standardized Digital Shadows and their automatic integration into the underlying infrastructure. To overcome these challenges, domain experts should be equipped with tools for supporting the building process. Furthermore, technologies like the Asset Administration Shell or ontologies should be introduced to enable the Digital Shadows benefits: the decision-support based on data. Current and further work is the development of an ontology and AASs in the domain of PPC in injection molding as a foundation for seamless integration of Digital Shadows in production.

FUNDING INFORMATION
Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy-EXC-2023 Internet of Production-390621612.

CONFLICT OF INTEREST
Authors have no conflict of interest relevant to this article.

DATA AVAILABILITY STATEMENT
Data sharing not applicable to this article as no datasets were generated or analyzed during the current study. ENDNOTE * https://eclass.eu/en ACKNOWLEDGMENT Open Access funding enabled and organized by Projekt DEAL.