Digital Twins: Basics and Applications 9783031114007, 9783031114014

Citation preview

Zhihan Lv Elena Fersman   Editors

Digital Twins: Basics and Applications

Digital Twins: Basics and Applications

Zhihan Lv · Elena Fersman Editors

Digital Twins: Basics and Applications

Editors Zhihan Lv Department of Game Design Faculty of Arts Uppsala University Visby, Sweden Qingdao Institute of Bioenergy and Bioprocess Technology Chinese Academy of Sciences Qingdao, China

Elena Fersman Ericsson Artificial Intelligence Research Institute Ericsson, Sweden Department of Machine Design Royal Institute of Technology Stockholm, Stockholms Län, Sweden

ISBN 978-3-031-11400-7 ISBN 978-3-031-11401-4 (eBook) https://doi.org/10.1007/978-3-031-11401-4 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Contents

Digital Twins Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Carlos Henrique dos Santos and José Arnaldo Barra Montevechi

1

Digital Twins for Physical Asset Lifecycle Management . . . . . . . . . . . . . . . . 13 Daniel N. Wilke Digital Twins and Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Li Zhang, Wei Zhou, and Xiaoqi Chen Agricultural Digital Twins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 Yuhang Zhao, Zheyu Jiang, Liang Qiao, Jinkang Guo, Shanchen Pang, and Zhihan Lv The Application of Digital Twins in the Field of Fashion . . . . . . . . . . . . . . . . 45 Victor Kuzmichev and Jiaqi Yan Digital Twins Collaboration in Industrial Manufacturing . . . . . . . . . . . . . . . 59 Radhya Sahal, Saeed H. Alsamhi, and Kenneth N. Brown Social Media Perspectives on Digital Twins and the Digital Twins Maturity Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 Jim Scheibmeir and Yashwant Malaiya

v

Digital Twins Architecture Carlos Henrique dos Santos and José Arnaldo Barra Montevechi

Abstract This chapter presents an overview of the digital twin (DT) building architecture. In general, we can understand the DT as a decision support system composed of four main components: (i) Physical system, (ii) virtual system, (iii) systems data, and (iv) communication interface. The physical system is composed of people, machines, and processes and represents the main focus of DT. On the other hand, the virtual system is composed of one or more highly synchronized virtual models, which are capable of mirroring physical behaviors and, through analysis tools and techniques, provides optimized decisions. Systems data correspond to both physical and virtual data. The virtual model mirrors the physical through its data, that is, information collected over time, while the virtual model returns information to the physical systems through actions and decision guidelines. Finally, we highlight the communication interface as a link that allows the integration between physical and virtual environments. In this case, the entire structure that allows the exchange of data and communication between the systems has a fundamental role regarding the DT’s correct functioning. In this chapter, the reader will comprehensively understand the role and the main characteristics of each DT component and subcomponent. From an approach focused on conceptualization followed by practical examples, the advantages and limitations associated with current DT practices are highlighted, providing a solid basis on the main DT architectures for practitioners and researchers.

1 Why to Talk About Digital Twins? Before talking about the pillars and components that constitute the DTs, it is important to understand the context in which they were created and developed. By understanding their origin, main characteristics, and the importance of DTs for current decision systems, we also understand the reason to talk about this topic and the importance of dealing with the DT’s architecture. C. H. dos Santos (B) · J. A. B. Montevechi Production Engineering and Management Institute, Federal University of Itajubá, Itajubá, Minas Gerais, Brazil e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 Z. Lv and E. Fersman (eds.), Digital Twins: Basics and Applications, https://doi.org/10.1007/978-3-031-11401-4_2

1

2

C. H. dos Santos and J. A. B. Montevechi

Although several works indicate different origins of the term “digital twin,” the most widespread by researchers is that this concept was proposed in 2010 by Mike Shafto (and other authors), and it was initially referred to intelligent virtual copies of equipment belonging to the United States Aerospace Agency, NASA. At the time, DTs were based on computer models capable of evaluating and recommending changes to physical systems to optimize them (Shafto et al. 2010). However, if we consider the technological advances in the last decade, we noted that the DT concept has undergone several changes and evolutions since it was proposed. The so-called Industry 4.0, an allusion to what would be the fourth industrial revolution (Uriarte et al. 2018), and its wave of technological developments allowed the rise of DTs, popularizing the concept and its application in the most diverse sectors and with different objectives (Semeraro et al. 2021). In recent years, we noticed the adoption of DTs to support decisions in production systems of goods and services, covering the areas of manufacturing, health, logistics, and services. In this case, both approaches are reasonable, DTs of products and processes, with different objectives and characteristics (Wright and Davidson 2020). Based on the virtualization of physical systems, one of the main pillars of the new industrial era and a fundamental part of the cyber-physical systems, DTs are described by different authors and from different points of view. Boschert and Rosen (2016) simplify by describing the DT as a virtual representation that mirrors the behavior of a component, product, or system and allows its evaluation throughout its lifecycle. Wright and Davidson (2020) add that this mirroring is carried out through the connection of virtual models with physical systems from their data and integration systems. Regardless of their definition, DTs have become indispensable for decisionmaking and, according to Tao and Zhang (2017), their adoption by decision-makers is an inevitable trend. Finally, in terms of publications involving DTs, it is possible to note their importance in recent years from a simple exploratory research. Considering publications between 2010 and 2021 listed in the Scopus® scientific base, there are about 5815 scientific works involving the keyword “digital twin,” among which about 43% of the publications were in 2021, as shown in Fig. 1. We observe a sharp growth of works in recent years, a fact that motivates us to develop more practical and theoretical knowledge about the main topics related to DTs.

2 The Main Digital Twin’s Components As previously highlighted, there are different descriptions and classifications regarding DTs, a fact that makes it difficult to understand their building and operation architecture. In part, the differences in the components necessary to compose the DT are due to its scope of action; in other words, it is expected that DTs with different objectives and areas of activity will also have differences in the components that compose them.

Digital Twins Architecture

3

Fig. 1 Scientific papers about DTs listed in Scopus® from 2010 to 2021

Alam and Saddik (2017) report, in a simplified way, that the DT is based on two modules: (i) Physical module, composed of the main systems (processes and/or products) and communication subsystems capable of collecting information and data from these physical systems during their operation; and (ii) digital module, composed of computer models capable of processing such information and optimizing the decision-making. Rodiˇc (2017) also reports the DT as a system based on two main components: (i) Digital shadow: that refers to physical systems (such as products and processes) and their respective data, which are collected and organized to compose a kind of “shadow” of the physical environment; and (ii) digital master: composed of computer models capable of capturing the digital shadow and mirroring the behavior of physical systems to assist decisions. In more detail, Zhuang et al. (2018) describe three main levels of the architecture of a DT: (i) Element: composed of geometric models and virtual representations faithful to physical systems; (ii) behavior: composed of mechanisms capable of representing physical behaviors through dynamic computational models (including movements, flows, etc.); and finally, (iii) rule: where there is integration between physical and virtual systems to allow synchronism between both. Moreover, Tao et al. (2018) report that a DT is based on three components: (i) Physical entities: composed of the physical space that we intend to represent virtually (i.e., machines and processes); (ii) virtual models: composed of computer systems capable of mirroring the physical entities and which can simulate, monitor, analyze, predict and control them; and (iii) connection data: which allow the integration and synchronization between the physical and virtual environments. Despite the different descriptions present in the literature, based on the main scientific works in the area, including those previously presented and also other

4

C. H. dos Santos and J. A. B. Montevechi

Fig. 2 General architecture of DTs

relevant articles, such as the proposed by Tao and Zhang (2017) and dos Santos et al. (2021), we can consider the DT as a system of four main components: (i) Physical system (PS), (ii) virtual system (VS), (iii) systems data (SD), and (iv) communication interface (CI). Figure 2 illustrates the overall architecture of DTs with these four components and the following topics detail each one.

2.1 Physical System (PS) We can describe the PS as all elements that belong to the physical environment and which we intend to mirror virtually. Furthermore, when we consider the main function of DTs, that is, to virtually represent the physical behavior to better decisions, it is clear that optimizing the PS is the real objective of creating them. Thus, some considerations are important regarding the characteristics of the PS. First of all, it is important to emphasize that the DT contemplates both products and processes. In the case of products, Wright and Davidson (2020) highlight that the design and prototyping stages represent the main applications of DTs. Therefore, we can consider the DT as a key tool in the development of new products and, according to Lo et al. (2021), it can assist at all stages from product design, development, testing, and, finally, commercialization. We can observe several works in the literature adopting such an approach, as those proposed by Dong et al. (2021) and Huang et al. (2022). On the other hand, concerning the DTs of processes, Tao and Zhang (2017) reveal that the focus is on converging physical and virtual environments to solve existing problems in addition to allowing better management practices. In this case, DTs can be used in the planning, operation, and post-operation phases of the processes.

Digital Twins Architecture

5

Furthermore, the main objectives associated with the adoption of process DTs in this context are related to decision support in production planning, process evaluation and control, resource allocation, and routing (dos Santos et al. 2021).

2.2 Virtual System (VS) The VS is probably the most important component of DT considering visual characteristics. Usually, the VS may be described as a DT synonym, but we already know that this statement is wrong since there are other components necessary for the design of a DT. In this case, we can simplify the VS as a set of all the computational resources used to virtually represent the behavior of the PS over time. Regardless of whether it is a product or process DT, the software and hardware options available are wide and allow the design of virtual models with different levels of detail and functionality. In this case, as highlighted by Zhuang et al. (2018), it is important to ensure that the VS can capture both the visual characteristics of the PS (such as geometric specifications, dimensions, colors, and other details), as well as the behavioral characteristics (such as actions, movements, state changes, flows, etc.). Firstly, we highlight the widely used and well-known commercial packages, which include softwares such as CATIA® , SolidWorks® , and AutoCAD® for visual representation and FlexSim® , Tecnomatrix® , AnyLogic® , Simio® , Arena® , 3DVIA Composer® , and Unity 3D® for behavioral representation. In addition, there is a significant portion of virtual models that are based on proprietary codes and computer programs, developed especially for a given application (Tao and Zhang 2017; dos Santos et al. 2021). Zhuang et al. (2018) report that the main functions of the DT are to evaluate, predict, simulate, validate, and optimize the PS, and, in this case, it is important to ensure that the VS is capable of performing the desired functions. Finally, it is important to highlight the role of new technological developments and, in this context, functionalities linked to virtual reality and augmented reality have been highlighted as important complements to VS (Tao and Zhang 2017).

2.3 Systems Data (SD) According to Alam and Saddik (2017), what differentiates a simple virtual model from a DT is its ability to capture the dynamic behavior of physical systems over time. This capability is due, in part, to the exchange of data and information between the physical and virtual environments, and, therefore, the SD has a fundamental role in the building and operation architecture of the DTs. Thus, the SD can be seen as the entire data and information structure of both PS and VS. In the case of the PS, it

6

C. H. dos Santos and J. A. B. Montevechi

is data related to its functioning, while the VS data is related to the responses to the physical systems. The SD plays an important role from the building phase to the operation of the DTs. In the DT building stage, the data coming from the physical environment allow the creation of a virtual base with equivalent characteristics in terms of behavior. On the other hand, during the operation of the DT, that is, its use to support decisions, the data from the physical environment allows the constant and periodic updating of the previously created virtual model. In addition, data from the virtual environment are also collected during the DT operation and are the basis for the DT responses for decision-making (Montevechi et al. 2020). In addition to allowing the synchronization between the PS and the VS, the systems data also play a fundamental role in the DT operation, ensuring its validity and accuracy. Ensuring the accuracy of the DTs is vital, since such use usually involves high-impact decisions, and, in this context, we must frequently compare metrics from both systems, physical and virtual, to assess the correct correspondence between them (Tao and Zhang 2017). This fact illustrates the importance of SD not only in the building and updating of the DT but also in its maintenance as a decision support tool.

2.4 Communication Interface (CI) Once the role of the other DT’s components was understood, the importance of the fourth and final component, the CI, becomes clear. Considering that the physical and virtual environments are already implemented and the data of the systems are available, the CI allows the collection, integration, and exchange of data and information between both environments, allowing the DT operation over time. Therefore, we describe the CI as every structure (hardware and software) that allows such integration. The basis of product and process virtualization, according to Shin et al. (2018), is the “connection and conversion level.” In this case, we consider both the collection of data (through sensors, databases, intelligent devices, and management systems) and the conversion of this data to allow its use by all the subsystems that compose the DT (through all available Information Technology structure). In this way, we can simplify CI as a “two-way street” in which data flows between physical and virtual environments. In the same way as the other components, the Industry 4.0 has been revolutionizing the structure that composes the CI. In this case, in addition to the well-known information technology (IT), we highlight solutions such as: (i) Internet of things (IoT), which enable the connection of multiple processes, machines, products, and equipment; (ii) big data, focused on solutions that allow data processing on a large scale and variety; and (iii) cloud technology, based on developments focused on data flows without the need for physical resources (Zhong et al. 2017). In this way, DTs are evolving to become increasingly versatile, efficient, and accessible to decision-makers.

Digital Twins Architecture

7

Furthermore, it is important to highlight that the processing and transmission of data and information between the physical and virtual environments directly impact the validity of the DTs’ results. In this context, several communication protocols can be used to allow the sharing of information between different DT elements. Among the main protocols, we highlight: DT Definition Language (DTDL); FIWARE; Open Platform Communication Unified Architecture (OPC UA); Feature-Based DT Framework (FDTF); Constrained Application Protocol (CoAP); Message Queuing Telemetry Transport (MQTT); Modbus Transmission Control Protocol (TCP/IP); and Ultra Reliable Low Latency Communication (URLLC) (Siqueira and Davis 2021; Qian et al. 2022). Finally, the choice of hardware, software, and protocols that make up the CI depends on the objectives and characteristics of the DT. In this case, it is important to consider some fundamental characteristics related to communication between physical and virtual environments, such as the level of reliability, latency, security, and scalability of the CI structure (Qian et al. 2022). For more details and the state of art about the available CI structure, we suggest the works proposed by Tao and Zhang (2017), Xu et al. (2018), Siqueira and Davis (2021), and Qian et al. (2022).

3 Is This a Digital Twin? Even after studying the main components that make up a DT, we are often in doubt about some of its features. In this case, to classify a decision system as a DT, it is necessary to consider, in addition to the main components, some assumptions related to: (i) the level of integration and synchronism between the physical and virtual environments; and (ii) the level of autonomy of the DT considering its responses to physical systems. Regarding the synchronism between the PS and VS, some authors such as Tao et al. (2018) and Kunath and Winkler (2018) reveal that PS data collection must occur in real-time. However, the VS update in the face of PS changes can occur in real or near real-time, as highlighted by Alam and Saddik (2017) and dos Santos et al. (2020). In this case, the VS must be updated with a minimum delay so as not to negatively impact the decision-making. A significant part of applications presented in the literature explores approaches in near real time, a fact that is linked to the frequency of decision-making (dos Santos et al. 2021). On the other hand, the DT’s autonomy is directly related to its responses, which can exercise direct command on the physical system (autonomous) or just suggest actions (nonautonomous) (dos Santos et al. 2020). Despite the adoption of automated systems being a growing trend in recent years, there is still a certain limitation in the adoption of autonomous DTs, largely due to the limitations of the PS, which are mostly composed of manual processes. In this way, regardless of the degree of autonomy, if the VS provides commands or directives for making decisions, we can consider the system as a DT.

8

C. H. dos Santos and J. A. B. Montevechi

4 Practical Case Studies In order to illustrate the concepts presented in this chapter, some practical case studies will be presented, highlighting the building and operation architecture of DTs. These are two real cases where DTs were developed to support decision-making in production processes in different segments, covering both automated and manual processes and with different characteristics in terms of synchronism between the physical and virtual environments.

4.1 Case Study I Case I refers to a DT implemented in a clothes factory to optimize the operational planning of one of the production lines. The line produces three different products (clothing items) and the production flow is divided into eight workstations (Process B to Process I), in addition to the Reception (A) and Shipping (J) areas. Although the process is mostly manual, part of the product transport is carried out automatically using an automated guided vehicle (AGV). Figure 3 illustrates the floor plant of the line with the production flow and also its 3D model built using FlexSim® . We have a near real-time and nonautonomous approach. Therefore, the DT is updated weekly and provides guidelines for decision-making. In addition to the FlexSim® simulation model, the DT is also based on an artificial intelligence (AI) model and a decision-making dashboard. In this case, the AI model predicts the weekly demand behavior considering the demand history. Moreover, based on the forecasted demand, the simulation model tests different weekly resource planning strategies and indicates the best decision regarding resource sizing (physical and human). The dashboard integrates the physical and virtual environments and provides a user-friendly interface for the decision-maker through automated buttons where the user can run the AI and simulation models and also view the guidelines for decision-making. Figure 4 shows the DT architecture.

Fig. 3 Production line and its 3D model (case study I)

Digital Twins Architecture

9

Fig. 4 DT architecture (case study I)

The data involved in the DT operation are basically: (i) Demand history: which is recorded weekly and stored in the local database. Such information feeds the AI algorithm and allows for the forecast of the products’ demand for the following week; (ii) general process data: which are collected by IoT devices and sensors and which are also stored in the local database. Such data include information such as cycle times, lead time, productivity, machine availability, among others, and which are crucial for the development and validation of DT elements; (iii) forecast demand: which is the result of the AI and also the input data for the simulation model; and (iv) simulation results: which include guidelines for decision-making. The dashboard allows all communication between systems and it was automated through the VBA language. Before the DT implementation, the decisions were based on the manager’s experience, which can compromise the process performance of the process. On the other hand, based on the DT, managers can plan and decide considering the results of the previous week. Furthermore, we expect that the DT improves its results over time since the AI algorithm can capture the demand behavior over time and provide better input to the simulation model. In this case, the results start to be more reliable. If it is necessary changes in the model, the DT can be easily modified. For example, the user can include changes to the process layout or flow, new KPIs, and guidance for decision-makers.

10

C. H. dos Santos and J. A. B. Montevechi

4.2 Case Study II Case II refers to a DT implemented in an automated production cell. In this case, the selected DT is also based on a FlexSim® simulation model and it focuses on monitoring the cell. This cell operates in three working shifts and it has two workstations that manufacture two types of products. First, the raw materials arrive in the arrival area (A) and are sent to the supply areas (B and D) which supply the raw material to the workstations (C and E), respectively. The transport between the areas is carried out by a conveyor and the movement between these areas and the workstations is carried out by robotic arms (R1 and R2). Figure 5 shows the production flow of the cell and its 3D model. The DT was planned to support the production cell through key performance indicators (KPIs), such as total lead time, stop times, and production rate. The process data are collected in real-time through sensors and they are stored in databases. The current work in progress (WIP) is used to update the FlexSim® model in near realtime with a delay of a few minutes, mirroring the cell and allowing the comparison between the expected and the real behavior. In this case, we have a nonautonomous approach, since the DT is used to evaluate the process, but not to control it. A decision dashboard is also adopted to allow the integration between the physical and virtual systems. Figure 6 illustrates the DT architecture. The data involved in this DT are: (i) General process data: which include information about the parts being produced, such as lead time, process cycle time, machine availability, among others. Such information is important for the development and validation of the DT and it is collected in real-time through sensors; (ii) work in progress: which is collected by sensors and IoT devices that allows periodic (near real-time) updating of the simulation model; (iii) simulation results: which contain analyzes and experiments based on the real behavior to provide the decisionmaker with several performance indicators. In this case, the dashboard also allows communication between the DT elements and it was automated through the VBA language.

Fig. 5 Production cell and its 3D model (case study II)

Digital Twins Architecture

11

Fig. 6 DT architecture (case study II)

Although this DT does not control the process, it has a fundamental role aiming at increasingly efficient production system. In this case, without the DT, the decisionmaker needs to evaluate the process based on its current state, a fact that may impact the decisions at a strategic level, such as: significant changes in flow, layout change, production replanning, among others. On the other hand, with the support of the DT, the decision-maker can assess the impacts of their decisions through scenarios and experiments, making the decision more assertive and efficient.

References Alam KM, EL Saddik A (2017) C2PS: a digital twin architecture reference model for the cloud-based cyber-physical systems. IEEE Access 5:2050–2062. https://doi.org/10.1109/ACCESS.2017.265 7006 Boschert S, Rosen R (2016) Digital twin—the simulation aspect. In: Hehenberger P, Bradley D (eds) Mechatronic futures: challenges and solutions for mechatronic systems and their designers. Springer, pp 59–74 Dong Y, Tan R, Zhang P et al (2021) Product redesign using functional backtrack with digital twin. Adv Eng Inform 49:1–17. https://doi.org/10.1016/j.aei.2021.101361 dos Santos CH, De QJA, Leal F, Montevechi JAB (2020) Use of simulation in the industry 4.0 contex: creation of a digital twin to optimise decision making on non-automated process. J Simul 14:1–14. https://doi.org/10.1080/17477778.2020.1811172

12

C. H. dos Santos and J. A. B. Montevechi

dos Santos CH, Montevechi JAB, de Queiroz JA et al (2021) Decision support in productive processes through DES and ABS in the digital twin era: a systematic literature review. Int J Prod Res 59:1–20. https://doi.org/10.1080/00207543.2021.1898691 Huang S, Wang G, Lei D, Yan Y (2022) Toward digital validation for rapid product development based on digital twin: a framework. Int J Adv Manuf Technol 1–16. https://doi.org/10.1007/s00 170-021-08475-4 Kunath M, Winkler H (2018) Integrating the digital twin of the manufacturing system into a decision support system for improving the order management process. In: 51st CIRP conference on manufacturing systems integrating. Elsevier B.V., pp 225–231 Lo CK, Chen CH, Zhong RY (2021) A review of digital twin in product design and development. Adv Eng Inform 48:1–15. https://doi.org/10.1016/j.aei.2021.101297 Montevechi JAB, Santos CH, Gabriel GT et al (2020) A method proposal for conducting simulation projects in Industry 4.0: a cyber-physical system in an aeronautical industry. In: Proceeding of the 2020 winter simulation conference. Orlando, USA, pp 2731–2742 Qian C, Liu X, Ripley C, Qian M, Liang F, Yu W (2022) Digital twin-cyber replica of physical things: architecture, applications and future research directions. Future Internet 14(64):1–25. https://doi.org/10.3390/fi14020064 Rodiˇc B (2017) Industry 4.0 and the new simulation modelling paradigm. Organizacija 50:193–207. https://doi.org/10.1515/orga-2017-0017 Semeraro C, Lezoche M, Panetto H, Dassisti M (2021) Digital twin paradigm: a systematic literature review. Comput Ind 130:1–23. https://doi.org/10.1016/j.compind.2021.103469 Shafto M, Conroy M, Doyle R et al (2010) DRAFT modeling, simulation, information technology & processing roadmap. In: Technology area 11—National Aeronautics and Space Administration (NASA), pp 1–27 Shin H, Cho K-W, Oh C-H (2018) SVM-based dynamic reconfiguration CPS for manufacturing system in Industry 4.0. Wirel Commun Mob Comput 1–15. https://doi.org/10.1155/2018/579 5037 Siqueira F, Davis JG (2021). Service computing for Industry 4.0: state of the art, challenges, and research opportunities. ACM Comput Surv 54(9):1–38. https://doi.org/10.1145/3478680 Tao F, Zhang M (2017) Digital twin shop-floor: a new shop-floor paradigm towards smart manufacturing. IEEE Access 5:20418–20427. https://doi.org/10.1109/ACCESS.2017.2756069 Tao F, Cheng J, Qi Q et al (2018) Digital twin-driven product design, manufacturing and service with big data. Int J Adv Manuf Technol 94:3563–3576. https://doi.org/10.1007/s00170-017-0233-1 Uriarte AG, Ng AHC, Moris MU (2018) Supporting the lean journey with simulation and optimization in the context of Industry 4.0. Procedia Manuf 25:586–593. https://doi.org/10.1016/j. promfg.2018.06.097 Wright L, Davidson S (2020) How to tell the difference between a model and a digital twin. Adv Model Simul Eng Sci 7:1–13. https://doi.org/10.1186/s40323-020-00147-4 Xu LD, Xu EL, Li L (2018) Industry 4.0: state of the art and future trends. Int J Prod Res 1:1–23. https://doi.org/10.1080/00207543.2018.1444806 Zhong RY, Xu X, Klotz E, Newman ST (2017) Intelligent manufacturing in the context of Industry 4.0: a review. Engineering 3:616–630. https://doi.org/10.1016/J.ENG.2017.05.015 Zhuang C, Liu J, Xiong H (2018) Digital twin-based smart production management and control framework for the complex product assembly shop-floor. Int J Adv Manuf Technol 96:1149–1163. https://doi.org/10.1007/s00170-018-1617-6

Digital Twins for Physical Asset Lifecycle Management Daniel N. Wilke

Abstract This chapter explores the role of digital twins to support decision-making in managing physical assets. In particular, the roles of physics-based digital twins and data-driven digital twins are explored, and their complementary nature in informative decision-making highlighted. Physics-based digital twins allow for causal inference to support decisions proposed by agile data-driven deep digital twins through unsupervised learning. Deep digital twins (DDT) enable a generic framework for prognostics and health monitoring (PHM) that can be rapidly deployed to a heterogeneous fleet of assets to directly automate predictive maintenance scheduling from operational data. DDTs are constructed from deep generative models which learn the distribution of healthy data directly from operational data at the beginning of an asset’s life. This does not rely on historical failure data to estimate asset health, which is crucial in enabling practical adoption to newly commissioned high-value assets for which failure data does not exist. For physical asset lifecycle management (PALM), the importance of digital twins is evident through practical use case analysis enabled through state-of-the-art physics-based and data-driven technologies.

1 Introduction Physical asset lifecycle management (PALM) of high-risk or large capital assets needs to be conducted within a volatile, uncertain, complex, and ambiguous (VUCA) world (Mack 2016). In many instances, the number of assets under management is often unique or limited to only a few. The primary aim of PALM is to maximize asset value, which usually involves the following three focal areas: 1. maximizing profit and revenue 2. minimizing the cost of ownership 3. extending the asset’s operational life. D. N. Wilke (B) Department of Mechanical and Aeronautical Engineering, Centre for Asset Integrity Management, University of Pretoria, Pretoria, South Africa e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 Z. Lv and E. Fersman (eds.), Digital Twins: Basics and Applications, https://doi.org/10.1007/978-3-031-11401-4_3

13

14

D. N. Wilke

Fig. 1 The three parts constituting a digital twin

The management of these assets can be significantly improved given access to the means of 1. 2. 3. 4. 5.

predicting the likelihood of future incidents and events, tracking the condition of an asset in real-time, causal inference as to the actual causes of changes in a system, accurately predicting the remaining useful life of an asset, accurate what-if scenario analyses.

Computerized Maintenance Management Systems (CMMS) (Wireman 1994) were developed in the late 80s to address some of these shortcomings, which include: 1. 2. 3. 4. 5.

Preventive maintenance Unscheduled maintenance tracking Maintenance expense record keeping Audit and compliance record keeping Modern maintenance strategies.

However, these systems were essentially databases and primitive data analysis to assist aspects of PALM. Extending these ideas to fully support PALM requires a framework that seamlessly integrates the physical world and the analysis or inference conducted on the data obtained from the physical world. Digital twin technology enables the construction of a digital representation of a physical asset with real-time data exchange between the digital twin and physical asset. Hence, the three parts of a digital twin are a (i) physical system, (ii) digital (or virtual) system, and (iii) data exchange between the two systems, as shown in Fig. 1. The physical system represents a physical asset whose lifecycle needs to be managed, formally referred to as physical asset lifecycle management (PALM). The stages of an asset’s life with its associated management stage are outlined in Fig. 2. Planning in PALM is typically limited to establishing asset requirements mostly based on existing assets to establish the value-add and business case for an asset. However, a more modern approach would also involve designing an asset for PALM by designing and manufacturing for sensing to enable lifecycle management. Numerous OEMs focus on developing modern assets to be sensing and maintenance centric. Finally, the asset is procured, delivered, and installed to allow deployment of the asset, which may entail a run-in of the asset after commissioning. This enables the asset to be utilized over its useful life with continuous monitoring and maintenance to extend the end of useful life. Here, useful life refers to the value-add of the asset, typically based on a business case.

Digital Twins for Physical Asset Lifecycle Management

15

Fig. 2 Contrasting physical asset lifecycle management against physical asset life stages

2 Digital Twin Asset Lifecycle Management (DTALM) Digital twins are digital assets whose primary value is to enable quality information from data (Grieves 2016). Like a physical asset, a digital asset’s lifecycle needs to be managed to obtain the maximum value from the digital asset. The stages of digital twin asset lifecycle management (DTALM) are outlined in Fig. 3.

Fig. 3 Contrasting digital asset lifecycle management against digital asset life stages

16

D. N. Wilke

A digital twin is an asset that can add value when the needs and goals are clearly defined. The value of a digital twin includes 1. 2. 3. 4. 5.

Improve production and efficiency Improve safety and environmental impact Accelerate and inform decision-making Optimize maintenance Institutional knowledge and improved work quality.

In contrast, a digital twin is an asset that requires lifecycle management to maximize its impact and usefulness. A digital twin results from multiple assets and technologies coming together, including generative modeling, IOT, 4G/5G, databases, big data, blockchain, edge computing, cloud computing, and artificial intelligence.

3 Digital Twin Essence Before exploring various digital twin paradigms for PALM, we first need to understand the essence of a digital twin. Fundamentally, all digital twins have the common goal of enabling causal inference and uncertainty quantification for dynamic environments. Although this common goal is succinct and efficient, it hides the complexities it encapsulates to achieve it. Before we proceed, it is important to distinguish between a simulation and a digital twin of a physical asset. A simulation can infer what could happen to a prototype product, while a digital twin infers and informs what is happening to a specific and actual product. Consequently, a digital twin requires data from the physical asset to be transferred to the digital twin. In contrast, information from the digital twin needs to be transferred back to the physical asset. Therefore, a digital twin is a system that requires interaction between the physical and digital domains.

4 Digital Twin Systems A digital twin system consists of two domains that interact with each other, namely 1. Physical domain 2. Digital domain a. Data domain b. Information domain. as shown in Fig. 4.

Digital Twins for Physical Asset Lifecycle Management

17

Fig. 4 A digital twin system consists of physical and digital domains. The digital domain can be represented by a data domain and an information domain

4.1 Physical Domain The physical domain represents the physical asset, IoT sensors, actuators, and hardware to enable communication between the physical and digital domains.

4.2 Digital Domain The digital domain can be sub-classified into data and information domains. The data domain is defined by middleware to manage data storage (Yun et al. 2017), data processing, and data modeling. Middleware enables integration through third-party software. Interoperable systems share data directly between systems, i.e., the original raw data is received from another system and used as is. In essence, the physical domain and data domain within the digital domain determine the data’s quality and quantity. A project budget and physical limitations often drive the physical and digital domains. This makes the data available that can be explored to obtain insightful information about an asset’s state, condition, and remaining useful life. Hence, the importance of the information domain cannot be understated as this often allows quality information to be extracted from limited data. The information domain within the digital domain transforms supplied data into information that is then relayed to the physical asset as signals to interact with the hardware. This is achieved by updating a generative model of the physical asset to enable inference and interpretation (Booyse et al. 2020). A generative model is, in essence, some description of the input parameters through a coordinate system with likely distribution of parameters. Any supplied input is then transformed to a usually higher dimensional coordinate system with likely associated response distribution. As data is made available through the middleware, the generative model is updated. The updated model is then used to conduct inference, which is either forward inference or inverse inference. The forward inference maps data from the input parameter to the output response, while inverse inference maps data from the output response to the input parameters (Asaadi et al. 2017) (Fig. 5).

18

D. N. Wilke

Fig. 5 An overview of the essence of a generative model

Although the outlined steps regarding generative models seem simplistic, it is important to expand on the complexities of several steps as we consider the types of generative models in more detail. The two main strategies (Erikstad 2017) to obtain a generative model for digital twins are as follows: 1. Physics-based generative models 2. Data-based generative models.

4.3 Physics-Based Generative Models for Digital Twins Physics-based digital twins fuse physics-based simulation models and sensor data through modeling updating. The advantage of physics-based simulation models is that causality is readily available subject to unambiguous model updating. Physics-based generative models are rarely used in situ for digital twins but rather used to generate data using the design of experiments (DOEs) (Sacks et al. 1989; Snyman and Wilke 2018). The reason is that running a finite element, finite volume, or discrete element model is usually too time-consuming for digital twin applications unless a highly efficient model can be abstracted for the physical asset in limited cases. Depending on the data from the simulation software, the surrogate model can be constructed using the output response and/or partial derivatives of how the output response changes based on changes in the input parameters. Popular choices for surrogate models using only the output response include radial basis functions (RBF) (Hardy 1971), while Gradient-enhanced Kriging (GEK) (Bouhlel and Martins 2018; Snyman and Wilke 2018) is popular when both the output response and partial derivatives are available, while gradient-only surrogates are popular when using only partial derivatives (Snyman and Wilke 2018). There are several benefits of physics-based generative models that include the following: 1. In causal models, a change in the output response is due to a change in the input parameter. 2. The mapping from the input parameters to the output response is based on physics. 3. The input parameters are usually descriptive and independent, i.e., each parameter has a unique role and meaning.

Digital Twins for Physical Asset Lifecycle Management

19

4. To update a generative model requires only a few parameters to be estimated from data made available by middleware. The disadvantages of physics-based generative models include the following: 1. The direction of mapping from input parameters to output response is usually fixed, which requires an inverse problem to be solved to estimate the input parameters given an output response. 2. Ambiguous model updating unless parameter identifiability and identification can be established with confidence to identify shortcomings (Ben Turkia et al. 2019). 3. Parameter identifiability can be challenging, i.e., identifying all possible input parameters that give a similar output response (Ben Turkia et al. 2019). 4. Parameter identification can be challenging, i.e., identifying the “correct” input parameter given an observed output response. 5. The physics in the model is static, while only the model parameters are dynamic. Changes in physics require a new model to be constructed. The issue with unambiguous model updating is that all model parameters need to be identifiable from the sensed data. In principle, for non-linear models, this can be challenging to establish (Ben Turkia et al. 2019), as parameter identifiability and parameter identification can change as the sensor data changes over the parameter domain.

4.4 Advances in Parameter Identifiability Recent advances in parameter identifiability exploit virtual parameter updating using simulated data instead of actual data. The advantages include: 1. knowing the actual parameters to be identified, 2. exploring identifiability over the entire parameter domain. Allows for identifiability to be established at the parameter level over the parameter domain. The implication is that an input parameter that is identifiable in some part of the input parameter domain may not be identifiable over another part of the input parameter domain (Ben Turkia et al. 2019). By statistically analyzing the data used to construct a surrogate model, the ability to identify each parameter over the parameter domain can be determined, as shown in Fig. 6. Given the output response A for a state, it is evident that Parameter 2 is identified with lower variance as compared to Parameter 1, for which Parameter 1 is more uncertain. Hence, some input parameters can be identified better than others, given some output responses. Considering another state with output response B, it is evident that both Parameters 1 and 2 are identified with equal variance. Hence, given the output response from the same sensor channels for two states, the ability to identify an input parameter varies over the parameter domain (Ben Turkia et al.

20

D. N. Wilke

Fig. 6 Parameter identifiability overview. The estimated input parameters are shown given two output responses A and B, for two states. Note that the identified input parameters can change over the parameter domain due to non-linearities in the physics at different states

2019). Digital twins need to be designed to accommodate potential issues associated with this. The ability to improve the identifiability of a parameter can be addressed in one of three ways: 1. Statistical regularization of the input parameter domain, e.g., likely input parameters, can be prioritized over unlikely input parameters if this information is available. 2. Physical regularization of the input parameter domain, i.e., restricting the parameters to physically plausible parameters or parameter sets obtained from additional simulations. 3. Additional output sensor data is required to identify the input parameters.

5 Data-Driven Digital Twins Data-based digital twins simultaneously estimate a generative model and the model parameters from the sensor data through modeling training. The disadvantage of databased simulation models is that causality is generally not readily available. Supervised, semi-supervised, and unsupervised data-driven models can be constructed depending on the available data. In physical asset management, unsupervised learning is generally applicable. Hence, we will limit discussions to unsupervised learning. For unsupervised learning, the challenge is to discover the unobserved and unknown input parameter space from only observing output responses. The generative model can reconstruct observed and unobserved outputs from sampling the discovered input parameter domain. Specifically, we will focus on latent variable models (LVM), i.e., generative models that either discover or prescribe a latent space, often representative of the unknown input parameter space. In general, unsupervised latent generative modeling is based on two phases: training and evaluation, as shown in Fig. 7.

Digital Twins for Physical Asset Lifecycle Management

21

Fig. 7 Unsupervised generative modeling’s training and evaluation phases. Training is characterized by encoding–decoding, while evaluation can be generative or informative

The training phase is characterized by encoding the training data obtained from the middleware to a latent space. Specifically, the training data is usually higher dimensional than the latent space. Hence, encoding is a dimensional reduction step that transforms higher dimensional data into lower dimensional data in the latent space. The rationale for this is based on physics. Physics are additional constraints that are imposed by nature on possible behaviors that can be observed in reality. Equality constraints reduce the observed data space’s dimensionality, while inequality constraints reduce the set of parameters but not the dimensionality (Snyman and Wilke 2018). Since physics is so restrictive, the result is that a much lower dimensional space is embedded in the observed high-dimensional data. This is often referred to as the manifold hypothesis (Fefferman et al. 2013). Data in the latent space is then decoded to reconstruct the observed highdimensional output response from the lower dimensional latent space. This ensures that the generative model can generate the observed output response. Evaluation can be generative when the LVM is used to reconstruct observed signals. This looks similar to the training phase except that the generative model reconstructs samples it has not seen before. The model here is essentially static

22

D. N. Wilke

and trained offline with only the inference dynamic or online. The rationale is that the model can regenerate samples similar to the training data with high accuracy, while data that is distinct from the training data will reconstruct poorly (Booyse et al. 2020). Following this approach in physical asset, management for digital twins typically requires an asset to be trained during early life or the beginning of useful life data. As the asset ages, the ability to regenerate the observed samples will deteriorate when the model is kept fixed. Generative evaluation mainly relies on generative models that focus on encoding data effectively, i.e., models that are trained to identify a latent coordinate system that efficiently explains the variance in the output response. Models favored for this task are variance-driven models, which optimally explains the variance in the observed data with the lowest dimensional latent space. Alternatively, evaluation can be informative when the LVM encodes the observed signals. Given a quality coordinate system for the latent space, each latent parameter has a unique role and is independent of the other latent parameters (Wilke et al. 2022). Higher dimensional data is transformed into lower dimensional information that informs changes and interprets changes (Cawley 2018). Here, the model is often updated and retrained in an online learning fashion to maximize the information retention from the data supplied by the middleware. As a result, both the model training and inference are dynamic or online (Dobson et al. 2013). Informative evaluation mostly relies on informative models that focus on finding unique and independent parameters that effectively explain the variance in the observed output data, i.e., models that are trained to identify a latent coordinate system that represents sources or independent parameters that drive the output response (Wilke et al. 2022). Models favored for this task are source-driven models, which find independent sources that explain the variance in the observed data. The consequence is that a higher dimensional latent space is typically recovered compared to a typical generative model (Wilke et al. 2022). Numerous data-based unsupervised generative models and strategies have been developed from statistical to machine and deep learning.

5.1 Statistical Learning Models Statistical learning is usually limited to linear models. A linear model implies that the input and output coordinate systems do not change spatially over the input or output parameter domains (Wilke et al. 2022). As a result, these models can represent their respective coordinate systems as vectors, which makes these methods easy to investigate and interrogate (Wilke et al. 2022). In addition, training of these models is usually conducted in closed linear Algebra form by solving the respective optimization problem as a closed-form optimality criterion problem, e.g., closed-form least-squares solution (Snyman and Wilke 2018). Statistical learning LVM often employed in the generative evaluation include probabilistic principal component analysis (PCA) (Tipping and Christopher 1999),

Digital Twins for Physical Asset Lifecycle Management

23

while statistical LVM often employed in informative evaluations include probabilistic independent component analysis (ICA) (Cawley 2018; Dobson and Cawley 2015; Schmidt et al. 2022; Wilke et al. 2022).

5.2 Machine Learning Models Machine learning extends linear models to non-linear models. For non-linear models, the input and output coordinate systems change spatially over the input or output parameter domains (Wilke et al. 2022). As a result, these models cannot represent their respective coordinate systems as vectors but rather parametric or non-parametric functions, making these methods more challenging to investigate and interrogate. Hence, often the reference to black-box models. Machine learning usually requires extensive pre-processing of the data into a suitable form to achieve quality results. These models are trained iteratively to find the optimal model for the given data, which is time-consuming and expensive. The quality of the trained model is therefore highly dependent on the chosen architecture and quality of the trained model (Booyse et al. 2020), which can be tricky to achieve as complications include 1. Mode collapse. 2. Vanishing gradients. 3. Failure to converge. With generative adversarial networks (GANs) (Booyse et al. 2020; Thanh-Tung and Tran 2020), typically prone to all three. Machine learning LVM often employed in the generative evaluation includes variational autoencoders (VAE) (Diederik et al. 2013; Booyse et al. 2020), while informative evaluations include beta-VAE (Higgins et al. 2017).

5.3 Deep Learning Models Deep learning is similar to machine learning, except it aims to eliminate user preprocessing to incorporate that as part of the model training. Deep learning’s performance can be even more sensitive than machine learning, dependent on the chosen architecture and the quality of the trained model (Booyse et al. 2020; Kafka and Wilke 2021). Deep learning LVM employed in generative evaluation includes GANs (Booyse et al. 2020), while informative evaluations include adversarially learned inference (ALI)-based strategies (Dumoulin et al. 2016). The benefits of data-based unsupervised generative models include: 1. Model and model parameters are simultaneously estimated.

24

D. N. Wilke

2. Only identifiable parameters are identified from data made available by middleware. 3. Model physics and model parameters are dynamic. Changes in model physics only require new data to be trained on. 4. Changing single asset data to fleet asset data enables models to be applied to individual assets or fleets of assets. 5. The direction of mapping from input parameters to output response can be altered to map directly from the output response to input parameters, often referred to as direct inverse maps (Asaadi et al. 2018). 6. Transfer learning can be employed to mitigate the cost and time of training. The disadvantages of data-based unsupervised generative models include: 1. The input parameters are usually not descriptive nor independent. 2. To update the generative model requires the model to be trained which can be time-consuming and expensive, depending on the chosen architecture and compute resources.

5.4 Industrial Digital Twin Applications for PALM Huang et al. (2021) is a recent review of AI-Driven Digital Twins in Industry 4.0 that covers condition monitoring digital twin applications for cutter-workpiece engagement using artificial neural networks (ANN), convolutional neural networks (CNNs), and support vector data descriptors (SVDD). Roller-bearing digital twin technologies are primarily based on convolutional neural network (CNN) combined with deep long short-term memory (DLSTM), and generative adversarial networks (GANs) are outlined. Ultrasonic guided wave digital surrogates using transfer learning are employed for damage visualization of guide waves. Huang et al. (2021) also include a review on predictive maintenance digital twin applications based on the monitoring status and health indicators that include remaining useful life (RUL) for several physical assets, including drilling machines, bearings, and wind turbines. The review covers statistical, hybrid modeling, and deep learning for time-series analysis and forecasting relying on long short-term memory (LSTM), Gaussian mixture models (GMM), stacked sparse autoencoders (SSAE), GANs, and distributed k-means clustering. Madni et al. (2019) explore the benefits of digital twin technology for modelbased systems engineering (MBSE) to investigate performance, maintenance, and operational health characteristics of physical assets. The benefits of using digital twins to conduct Design Failure Mode and Effects Analysis (DFMEA) for design modification and improvement are included. The paper discusses the benefits of integrating digital twins with system simulation and Internet of Things (IoT) in support of MBSE and provides specific examples of the use and benefits of digital twin technology in different industries. It concludes

Digital Twins for Physical Asset Lifecycle Management

25

with a recommendation to make digital twin technology an integral part of MBSE methodology and experimentation testbeds. Perno et al. (2022) present a systematic survey of digital twin implementations in the process industry, focusing on enablers and barriers that include physics-driven digital twins. Glatt et al. (2021) utilize physics-based simulations to develop a digital twin of a material handling system capable of prediction, monitoring, and diagnosis. Booyse et al. (2020) work presents a Deep Digital Twins (DDT) as a rapidly deployable and generic framework for prognostics and health monitoring (PHM). The framework is rapidly deployable to individual assets or heterogeneous fleets of assets that allow for the automation of predictive maintenance scheduling directly from operational data. The DDT is constructed from deep generative VAE or GAN models that learn the distribution of healthy data directly from operational data during early life or initial useful life. Based on several datasets, the DDT has been demonstrated to detect incipient faults, track asset degradation, and differentiate between failure modes in both stationary and non-stationary operating conditions. Balshaw et al. (2022) highlight the importance of temporal preservation to further enhance latent analysis for DDTs.

References Asaadi E, Wilke DN, Heyns PS, Kok S (2017) The use of direct inverse maps to solve material identification problems: pitfalls and solutions. Struct Multi Optim 55. https://doi.org/10.1007/ s00158-016-1515-1 Balshaw R, Heyns PS, Wilke DN, Schmidt S (2022) Importance of temporal preserving latent analysis for latent variable models in fault diagnostics of rotating machinery. Mech Syst Signal Process 168:108663 Ben Turkia S, Wilke DN, Pizette P, Govender N, Abriak N-E (2019) Benefits of virtual calibration for discrete element parameter estimation from bulk experiments. Granular Matter 21. 110. https:// doi.org/10.1007/s10035-019-0962-y Booyse W, Wilke DN, Heyns SP (2020) Deep digital twins for detection, diagnostics and prognostics. Mech Syst Signal Process 140:106612 Bouhlel MA, Martins JRRA (2018) Gradient-enhanced kriging for high-dimensional problems. Eng Comput 35:157–173 Cawley P (2018) Structural health monitoring: closing the gap between research and industrial deployment. Struct Health Monit 17:1225–1244 Dobson J, Cawley P (2015) Independent component analysis for improved defect detection in guided wave monitoring. Proc IEEE 104:1–12. https://doi.org/10.1109/JPROC.2015.2451218 Dumoulin V, Belghazi I, Poole B, Mastropietro O, Lamb A, Arjovsky M, Courville A (2016) Adversarially learned inference. arXiv:1606.00704 Erikstad SO (2017) Merging physics, Big Data analytics and simulation for the next-generation digital twins. In: High-performance marine vehicles conference Fefferman C, Mitter S, Narayanan H (2013) Testing the manifold hypothesis. J Am Math Soc 29. https://doi.org/10.1090/jams/852 Glatt M, Sinnwell C, Yi L, Donohoe S, Ravani B, Aurich JC (2021) Modeling and implementation of a digital twin of material flows based on physics simulation. J Manuf Syst 58:231–245 Grieves M (2016) Origins of the digital twin concept. https://doi.org/10.13140/RG.2.2.26367.61609

26

D. N. Wilke

Hardy RL (1971) Multiquadric equations of topography and other irregular surfaces. J Geophys Res 76(8):1905–1915. https://doi.org/10.1029/JB076i008p01905 Higgins I, Matthey L, Pal A, Burgess CP, Glorot X, Botvinick MM, Mohamed S, Lerchner A (2017) beta-VAE: learning basic visual concepts with a constrained variational framework. In: ICLR (2017) Huang Z, Shen Y, Li J, Fey M, Brecher C (2021) A survey on AI-driven digital twins in Industry 4.0: smart manufacturing and advanced robotics. Sensors (Basel, Switzerland) 21(19):6340. https:// doi.org/10.3390/s21196340 Kafka D, Wilke DN (2021) An empirical study into finding optima in stochastic optimization of neural networks. Inf Sci 560:235–255 Mack O (2016) Managing in a VUCA world. Springer, Heidelberg Cham Madni A, Madn C, Lucero S (2019) Leveraging digital twin technology in model-based systems engineering. Systems 7:7 Perno M, Hvam L, Haug A (2022) Implementation of digital twins in the process industry: a systematic literature review of enablers and barriers. Comput Ind 134:103558 Sacks J, Welch WJ, Mitchell TJ, Wynn HP (1989) Design and analysis of computer experiments. Stat Sci 4:409–423 Schmidt S, Wilke DN, Heyns PS (2022) A comparison between independent component analysis and established signal processing methods for gearbox fault diagnosis under time-varying operating conditions. In: Hammami A, Heyns PS, Schmidt S, Chaari F, Abbes MS, Haddar M (eds) Modelling and simulation of complex systems for sustainable energy efficiency. MOSCOSSEE 2021. Applied condition monitoring, vol 20. Springer, Cham. https://doi.org/10.1007/978-3-03085584-0_21 Snyman JA, Wilke DN (2018) Practical mathematical optimization: an introduction to basic optimization theory and classical and new gradient-based algorithms, 2nd edn, vol 2. Springer. https:// www.springer.com/gp/book/9783319775852 Thanh-Tung H, Tran T (2020) Catastrophic forgetting and mode collapse in GANs. In: International joint conference on neural networks (IJCNN), pp 1–10 Tipping ME, Bishop MC (1999) Probabilistic principal component analysis. J Roy Stat Soc. Ser B (Stat Methodol) 61:611–622 Wilke DN, Heyns PS, Schmidt S (2022) The role of untangled latent spaces in unsupervised learning applied to condition-based maintenance. In: Hammami A, Heyns PS, Schmidt S, Chaari F, Abbes MS, Haddar M (eds) Modelling and simulation of complex systems for sustainable energy efficiency. MOSCOSSEE 2021. Applied condition monitoring, vol 20. Springer, Cham. https://doi. org/10.1007/978-3-030-85584-0_5 Wireman T (1994) Computerized maintenance management systems. Industrial Press Yun S, Park J, Kim W (2017) Data-centric middleware based digital twin platform for dependable cyber-physical systems. In: Ninth international conference on ubiquitous and future networks (ICUFN), pp 922–926

Digital Twins and Additive Manufacturing Li Zhang, Wei Zhou, and Xiaoqi Chen

Abstract Additive manufacturing has been regarded as a revolutionary technology and gained fast development in the past two decades. However, it is largely a trialand-error process to find the optimal printing strategy, and the poor reproductivity poses a problem for quality control. It would help to tackle the problems by making the additive manufacturing process digitized, visualized and predictable in real time. Digital twins are defined as a digital representation of the hardware. They have the potential to model the physical world in real time. This chapter summarizes the state-of-the-art status, current problems and prospects of digital twins for additive manufacturing.

1 Additive Manufacturing Additive manufacturing (AM), often referred to as 3D printing, is a layer-by-layer product fabrication method with formless raw materials, based on the 3D model data triple (Weißenfels 2022). AM is unique for manufacturing complex, composite and hybrid structures with design freedom and high precision, which are not achievable using conventional manufacturing processes. Additionally, it has the ability to deliver parts from conception to market with a high speed, simultaneously lower wasting of material (Gibson et al. 2021). What is more, with a near net shape formation, part count can be reduced significantly by eliminating assembling processes L. Zhang Guangdong Key Laboratory of Modern Control Technology, Institute of Intelligent Manufacturing, Guangdong Academy of Sciences, Guangzhou 510070, China e-mail: [email protected] L. Zhang · X. Chen (B) School of Engineering, Swinburne University of Technology, Hawthorn, VIC 3122, Australia e-mail: [email protected] W. Zhou Singapore Centre for 3D Printing, School of Mechanical and Aerospace Engineering, Nanyang Technological University, Singapore 639798, Singapore e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 Z. Lv and E. Fersman (eds.), Digital Twins: Basics and Applications, https://doi.org/10.1007/978-3-031-11401-4_4

27

28

L. Zhang et al.

(DebRoy et al. 2018). In the past two decades, this method has been considered as a revolutionary manufacturing technology due to its advantages over the conventional fabricating processes (Yin et al. 2021). AM processes can be classified into two categories: Directed energy deposition (DED) and powder bed fusion (PBF), as defined by ASTM Standard F2792 (Yap et al. 2015). They can also be further districted as different heat sources such laser (L), plasma (PA), electronic beam (EB), or gas metal arc (GMA), etc. PBF-L, PBFEB, DED-L, DED-EB, DED-PA and DED-GMA are frequently used by researchers (DebRoy et al. 2018). Meanwhile, they can also be differentiated by the types of feedstocks including powder and wire. Several different types of metal additive manufacturing processes are illustrated in Fig. 1, including the directed energy deposition (DED) and powder bed fusion (PBF) (Zhang et al. 2020). With the development of Industry 4.0, tremendous demand for customized components will be needed by the market. AM will be widely used to produce customized components, especially for aerospace, marine, medical, new energy and automotive applications. However, although this technology shows great potentials in different area applications, it is still a very young technology, and there are still many gaps and needs for its further applications. For example, one of the biggest drawbacks of AM is the lack of reproductivity (Weißenfels 2022). The AM processes highly depend on starting materials, machines, processing parameters, printing strategies, process sensing and controlling, inert gas protecting, etc. High consistency of all the key factors is required for the stability of AMed parts. Every fine distinction may lead to great differences that is why the properties of the printed parts often do not correspond to the planned specification. Several disciplines are involved in the AM processes, including materials, process design, process monitoring and controlling, standards and certification, even mechanical and optical design. What is more, the mechanical properties of the printed parts

Fig. 1 Schematics of main additive manufacturing processes

Digital Twins and Additive Manufacturing

29

are still unpredictable. Trial-and-error approach is still the fashion to produce components with sound microstructures and good mechanical properties by AM, which is rather time-consuming, materials wasting and expensive. To obtain optimal process conditions, numerous experiments are needed to optimize the process variables within given machines and processes. This hinders the adoption of this technology as a widely used manufacturing process. Modeling and simulation are usually used for parameter optimization and process guiding. Nevertheless, modeling and simulation usually use approximations and hypothesis method, neglecting important details of the process. For example, the fluxion in the melt pool, the change of laser reflectivity caused by the change of the material condition, also the change of the environment, shielding gas etc., are usually neglected during the modeling and simulation process. Therefore, it is arduous to simulate the true process accurately. Another main issue is time-consuming, as the modeling and simulation process highly relies on the computing power of computers. The prediction of temperature distribution inside an AM part that is being printed with non-proprietary mesh-based finite element (FE) models will take at least several hours, if not days (Gaikwad et al. 2020). It is still needed to verify the simulation results by destructive testing. Commonly used modeling and simulation methods include finite element method (FEM), finite difference method (FDM), level set method (LSM), volume of fluid (VOF) method with FDM, lattice Boltzmann method (LBM) and arbitrary Lagrangian–Eulerian (ALE) (DebRoy et al. 2017). To overcome the barriers of AM and promote its wide application, it is vital to realize the visualization and digitization of the process. Real-time monitoring and controlling even prediction of the vital factors such as temperature, stress, microstructures and mechanical properties of the printed parts are of great meaning to AM. In-depth research is underway on the technologies of digitization, simulation (Song et al. 2018a; b), big data and machine learning (Ren et al. 2020). However, the current methods only solve part of the problems. A new method that can realize the total real-time digitalization and visualization is very urgently needed. There is indeed a new method that can solve the problems—the digital twin.

2 Digital Twins Digital twins (DT) are defined as a digital representation of the hardware (Mukherjee and DebRoy 2019). Actually, a digital twin can also replica an active unique product characterized by certain properties or conditions, a production system or even a service (Zhang et al. 2020). Digital twins were firstly proposed by National Aeronautics and Space Administration (NASA). They used this method to monitor the behaviors of a satellite and simulate the possible changes in the settings. A digital twin replica of the physical system was expected to explore the space (Tuegel et al. 2011). Since then, this technology was developed in many areas. Till now, DT has already been constructed and

30

L. Zhang et al.

utilized for many manufacturing processes by many industries and government agencies. It is reported that general electric already built over 550,000 digital twins of real, physical systems from jet engines to power turbines (Mukherjee and DebRoy 2019). DecisionLab Ltd. and Siemens have built a digital twin named ATOM. This system can visualize fleet and maintenance facility operations. Key performance indicators (KPIs) of the operation system can also be captured and predicted. It even helps to make decision of investment through quickly run a virtual and detailed scenery and be driven by live data already available within the supply chain. Digital twins for manufacturing system can visualize the manufacturing processes by a model containing all three-dimensional data and physical properties, simultaneously collect and present all the relevant process data by sensors installed in the manufacturing system. These data will be collected and transmitted by Internet of things (IoT), when there is too much data, it can be stored in the cloud. Machine learning with optimized algorithm is able to deal with the big data in real time. People can monitor the conditions of the manufacturing processes and machines at any time. It is also available to get the concerned information through the DT system, without going to the factory. All the physical systems have been replicated in the cyber-world. And the motions and changes are all synchronous. With these abilities, DTs are also the potential solutions to many problems in AM processes. By visualize the AM processes, it is easy to understand the thermal behaviors, material reaction processes and microstructure formation processes. Even realizing the prediction of parts’ mechanical properties fabricated by AM is possible. Real-time sensing and monitoring will help to control the process easily. By dealing with the big data, the optimal printing strategy and parameter combinations can be calculated. The barriers of AM can be overcome if a DT of AM system can be successfully constructed and utilized. However, development of DTs for AM is still in its infancy, and it faces various research challenges (Chhetri et al. 2017).

3 DTs for AM Needs and Challenges Several research teams have already studied this topic. Notably, DebRoy and their coworkers did lots of the pioneering work in this area (DebRoy et al. 2017; Mukherjee and DebRoy 2019; Knapp et al. 2017). They proposed an overarching framework, demonstrated this concept within the AM research community and subsequently provided a perspective of the current status and research needed for the main building blocks of a first-generation digital twin of AM. A DT of AM needs including a mechanistic model, a sensing and control model, a statistical model, as well as big data and machine learning, as shown in Fig. 2. (Zhang et al. 2020). Gaikwad et al. (2020) tried to build an early digital twin example for real-time process monitoring and defect prediction. In order to get high statistical fidelity in detecting process flaws, they combined physics-driven predictions with in situ sensor data and machine learning. Laser powder bed fusion (LPBF) and directed energy

Digital Twins and Additive Manufacturing

31

Fig. 2 Logical schematic of the digital twin for additive manufacturing (AM)

deposition (DED) metal AM processes were both tested to verify the calculation and prediction, and the results substantiated their conclusion. A summary of part of previous works in the DT for AM is listed in Table 1; the contributions and limitations of the works are indicated. Based on the current status of researches on DT of AM, it can be concluded that: Real-time simulation of AM system is under developing and the calculation burden is a main problem. The prediction is mainly focused on the distortion and melt pool phenomena, mechanical properties are still unable to predict; actual tailoring of the final component’s properties based on the predictive model is still a long way in the future. Overall, the research of DT in AM is still developing, and much work (database, hardware, software, etc.) needs to be done.

3.1 Real Time Monitoring Real-time monitoring and analysis are vital for AM. The state of environment and printed parts changes all the time. Even a small change of thermal behavior or powder feeding condition or a shielding gas flow state can affect the quality of the products. Parts with stable mechanical properties can be fabricated only when the real-time monitoring and controlling of AM process can be realized. Currently, the computational burden during twining the AM process impeded realtime monitoring and controlling. As thermal behavior, solidification process, residual stress, distortion and mechanical properties have to be calculated. The common

32

L. Zhang et al.

Table 1 Prior works in the DT for AM Authors

Contribution

Limitation

Gaikwad et al. (2020)

Demonstrated an early foray into the digital twin paradigm for real-time process monitoring in AM via seamless integration of physics-based modeling (simulation), in-situ sensing, and data analytics (machine learning)

Only temperature distribution was monitored by single in-situ sensors The melt pool phenomena, such as latent heat losses and surface tension was ignored

Knapp et al. (2017)

Presented a novel framework of a mechanistic model to predict the melt pool-level phenomena

Actual tailoring of the final component’s properties based on the predictive model is still a long way in the future

DebRoy et al. (2017)

Provided a perspective of the current status and research needs for the main building blocks of a first-generation digital twin of AM

The components required to construct a digital twin of AM hardware are still developing

Mukherjee and DebRoy (2019)

A mechanistic model, sensor data and an interface between them (machine learning) are three crucial components to create a digital twin for AM

A comprehensive, open source digital twin of 3D printing is not available

Chhetri et al. (2017)

The first work that demonstrated how dynamic data-driven application systems enabled feature re-ranking method can help in keeping the digital twin up-to-date

More KPIs (mechanical properties) need to predict and the applicability for other systems need to be demonstrated

computers cannot deal with the enormous amount of calculation. Therefore, sensitive sensors enabling monitoring of the states of melt pool are needed. Monitors such as the pyrometer, high speed camera, thermal imager, etc. can monitor the temperature, temperature field, shape, flow condition in real time. With the information tested in real time, the calculation burden will be decreased. More importantly, appropriate model with optimal algorithm for the AM process is a vital factor to realize real-time simulation of the AM processes.

3.2 Database and Models As introduced in the first section, AM processes can be classified into several different types according to heat resource, feedstock, etc. Each kind of AM process has its unique characteristics, and each kind of material has its own thermophysical parameters. Building database is necessary to reduce the computational burden. There is

Digital Twins and Additive Manufacturing

33

a great need of record data for each kind of AM process and material. When the database is big enough, and all the commonly used data can be called from database, it will be easier to build digital twins for AM processes. Furthermore, the commonly used AM processes and materials can be built into models. New processes and materials can be slightly adjusted based on the existing models and database. Therefore, building database and models for AM processes are of great importance for building DTs for AM.

3.3 Machine Learning Machine learning is the brain of DT system for AM. The big data collected by sensors and achieved by the simulation process is driven by the technology of machine learning. The prediction accuracy on the AM results depends on the machine learning technology. Also, the reaction speed is highly relied on the developing degree of machine learning. Researchers have already made lots of efforts to explore the way of predicting the AM results by machine learning method. For example. Ren et al. (2020) trained a physical-based machine learning algorithm by data created from a thermal field prediction numerical model. The model had already been verified in predicting thermal field for laser-aided additive manufacturing (LAAM) processes. LAAM process is one kind of the directed energy deposition (DED) processes. Finally, after training, the RNN-DNN showed agreement of more than 95% in predicting thermal field process. However, to achieve the real-time and comprehensive prediction of the AM results, more effective machine learning algorithm needs to be developed.

3.4 Internet of Things It is clear that, to digital twin the AM process and achieve key information to help predict and control the results, accurate data is vital. During the process, mass of data generated from each part of the system. All data that may affect the printing results should be collected and analyzed in real time. Beside the correct sensors and appropriate models with optimal algorithm, the transmitting of data plays an important role. Data from each part should be transmitted to the machine learning part timely. Therefore, to build DTs for AM successfully, the linkage between each part of the AM system and linkage between cyber-space and physical space are also crucial factors. Currently, some obstacles are still existing to get the smart connection level. For example, the existing solutions are not available in dealing with the heterogeneous equipment. Flexibility in connectivity and interactive message are not available using current methods (Tao et al. 2018). Tao et al. (2018) tried to realize smart connections for various manufacturing resources by designing an Industry-Internet-of-Thing Hub

34

L. Zhang et al.

(IIhub) based on a cyber-physical system (CPS) framework. However, to achieve full compatibility of various communication protocols and interfaces, all the communication protocols applied in current industry field needs to be developed and the emerging communication protocols needs to be updated. That is obvious a huge project. With the developing of new connection technologies such as 5G technology, Internet of everything and real-time data collecting and transmitting are sure able to realize.

4 Conclusions and Outlook With the development of Industry 4.0, AM will play a more and more important role. Current AM technologies are still unable to effectively support the increasing needs of customized demands. DTs are the potential solutions to overcome many issues in AM. By visualizing and real-timely monitoring and controlling the AM processes, it will be much easier to solve the barriers of AM. Obviously, the construction of DTs for AM is still in its infancy. The first generation of DTs for AM is under exploring and developing. Several key technologies including machine learning, Internet of things, big data, smart sensors and connection, etc., are still not available to offer enough supporting. Nevertheless, DTs are significant for the further development and application of AM processes.

References Chhetri SR, Faezi S, Al Faruque MA (2017) Digital twin of manufacturing systems. Center for Embedded and Cyber-Physical Systems, pp 1–19 DebRoy T, Zhang W, Turner J, Babu SS (2017) Building digital twins of 3D printing machines. Scripta Mater 135:119–124. https://doi.org/10.1016/j.scriptamat.2016.12.005 DebRoy T, Wei HL, Zuback JS, Mukherjee T, Elmer JW, Milewski JO, Beese AM, Wilson-Heid A, De A, Zhang W (2018) Additive manufacturing of metallic components—process, structure and properties. Prog Mater Sci 92: 112–224. https://doi.org/10.1016/j.pmatsci.2017.10.001 Gaikwad A, Yavari R, Montazeri M, Cole K, Bian L, Rao P (2020) Toward the digital twin of additive manufacturing: integrating thermal simulations, sensing, and analytics to detect process faults. IISE Trans 52(11):1204–1217. https://doi.org/10.1080/24725854.2019.1701753 Gibson I, Rosen D, Stucker B, Mahyar K (2021) Additive manufacturing technologies. Springer Nature Switzerland AG. https://doi.org/10.1007/978-3-030-56127-7 Knapp GL, Mukherjee T, Zuback JS, Wei HL, Palmer TA, De A, DebRoy T (2017) Building blocks for a digital twin of additive manufacturing. Acta Mater 135:390–399. https://doi.org/10.1016/j. actamat.2017.06.039 Mukherjee T, DebRoy T (2019) A digital twin for rapid qualification of 3D printed metallic components. Appl Mater Today 14:59–65. https://doi.org/10.1016/j.apmt.2018.11.003 Ren K, Chew Y, Zhang YF, Fuh JYH, Bi GJ (2020) Thermal field prediction for laser scanning paths in laser aided additive manufacturing by physics-based machine learning. Comput Methods Appl Mech Eng 362(xxxx). https://doi.org/10.1016/j.cma.2019.112734

Digital Twins and Additive Manufacturing

35

Song J, Chew Y, Bi G, Yao X, Zhang B, Bai J, Moon SK (2018a) Numerical and experimental study of laser aided additive manufacturing for melt-pool profile and grain orientation analysis. Mater Des 137:286–297. https://doi.org/10.1016/j.matdes.2017.10.033 Song J, Chew Y, Jiao L, Yao X, Moon SK, Bi G (2018b) Numerical study of temperature and cooling rate in selective laser melting with functionally graded support structures. Addit Manuf 24:543–551. https://doi.org/10.1016/j.addma.2018.10.039 Tao F, Cheng J, Qi Q (2018) IIHub: an industrial Internet-of-Things hub toward smart manufacturing based on cyber-physical system. IEEE Trans Industr Inf 14(5):2271–2280. https://doi.org/10. 1109/TII.2017.2759178 Tuegel EJ, Ingraffea AR, Eason TG, Michael Spottswood S (2011) Reengineering aircraft structural life prediction using a digital twin. Int J Aerosp Eng 2011. https://doi.org/10.1155/2011/154798 Weißenfels C (2022) Additive manufacturing processes. In: Simulation of additive manufacturing using meshfree methods. Lect Notes Appl Comput Mech 97:7–17. https://doi.org/10.1007/9783-030-87337-0_2 Yap CY, Chua CK, Dong ZL, Liu ZH, Zhang DQ, Loh LE, Sing SL (2015) Review of selective laser melting: materials and applications. Appl Phys Rev 2(4). https://doi.org/10.1063/1.4935926 Yin Y, Tan Q, Bermingham M, Mo N, Zhang J, Zhang MX (2021) Laser additive manufacturing of steels. Int Mater Rev. https://doi.org/10.1080/09506608.2021.1983351 Zhang L, Chen X, Zhou W, Cheng T, Chen L, Guo Z, Han B, Lu L (2020) Digital twins for additive manufacturing: a state-of-the-art review. Appl Sci (Switzerland) 10 (23):1–10. https://doi.org/10. 3390/app10238350

Agricultural Digital Twins Yuhang Zhao, Zheyu Jiang, Liang Qiao, Jinkang Guo, Shanchen Pang, and Zhihan Lv

Abstract In recent years, the development of Digital Twins has advanced by leaps and bounds, and Digital Twins have gradually begun to combine various fields and apply them to the current digitalization of the physical world. Digital Twins can play an important role in the field of agriculture. Digital Twins technology can fully improve the yield and profit of crop products and alleviate food safety issues. Regarding the current common problems in the agricultural field, this chapter synthesizes the existing technologies of Digital Twins, discusses the development prospects of it in the agricultural field, and puts forward the problems that still exist in the application of Digital Twins in the agricultural field.

1 The Digital Twins of Agriculture At present, the global agricultural level is still in a generally underdeveloped state. Agriculture in most countries and regions is still relatively traditional. The application of new-generation information technology has not yet been popularized in the agricultural field. Agricultural practitioners mainly rely on their learning and experience. Guiding agricultural production activities, agricultural production is greatly restricted, and the global food problem has not been completely resolved (van Dijk et al. 2021). The emergence of Digital Twins technology enables farmers and stakeholders to deal with unexpected situations. By continuously monitoring the entire Y. Zhao · Z. Jiang · S. Pang College of Computer Science and Technology, China University of Petroleum (East China), Qingdao 266580, China L. Qiao · J. Guo College of Computer Science & Technology, Qingdao University, Qingdao 266071, China Z. Lv (B) Department of Game Design, Faculty of Arts, Uppsala University, Uppsala, Sweden e-mail: [email protected] Qingdao Institute of Bioenergy and Bioprocess Technology, Chinese Academy of Sciences, Qingdao, China © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 Z. Lv and E. Fersman (eds.), Digital Twins: Basics and Applications, https://doi.org/10.1007/978-3-031-11401-4_5

37

38

Y. Zhao et al.

Fig. 1 Current situation of agricultural digital twins

process from production to marketing and sales, it can help identify problems in advance and schedule predictive maintenance at the appropriate time as well as complex problems and provide immediate solutions. Figure 1 shows current situation of agricultural Digital Twins. Digital Twins technology can effectively promote the development of agriculture in the direction of high quality, ecology, modernization, and informatization. Digital Twins technology has been applied in farm management because it eliminates the basic restrictions related to location, time, and human observation. Agricultural production will no longer require physical proximity, allowing remote monitoring, control, and coordination of farm operations. Through the Digital Twins of farms and agricultural activities on a global scale, participants at all levels of the agricultural value chain will be able to obtain more information resources, predict crop yields more effectively, and expand the scale of production with limited resources. IBM believes that digital replication of real farms can create a “Digital Twins” or “virtual model” for global farms, realize the health management of crops, and share farm data so that all agricultural participants can share ideas, research, and materials, exchange global farm and crop growth-related data, and interconnect with the food supply chain (Jans-Singh et al. 2020). The Dutch company Connecterra has developed a digital cow assistant service based on Digital Twins technology. It monitors cows remotely and detects when the cows are in estrus, to fully understand the health of the cows, further predict the start date of the next cycle, and use materials. Networking technology sends feasible insights and smart suggestions to farmers; the BeeZon apiary monitoring system in Greece has designed a Digital Twins of

Agricultural Digital Twins

39

the bee colony to monitor and control bee diseases, pest infections, pesticide exposure, and toxicity, and help beekeepers remote control its bee farms and make wise management decisions.

2 Digital Twins Build Smart Farms From traditional artificial planting to existing mechanized planting, from hard work to automatic seeding now, building an intelligent Digital Twins farm with real-time monitoring and full-process control is no longer an unattainable dream. In recent years, Gartner has named the Digital Twins in the supply chain as the Digital Supply Chain Twin and vividly described that the digital plum is not just a model of its physical objects: every physical object there is at least one unique Digital Twins corresponding to it. It can control the object it represents and has all the data representing its object, such as identification, status, content, etc.; it can find the state of the physical object and get notifications, can simulate physical objects, events, and processes in the real world, and analyze them with rules, predictions, and algorithms. This theory has attracted widespread interest and attention from the supply chain industry, supply chain software providers, and researchers. At present, Digital Twins technology has been applied in many fields such as industry, urban design, construction, manufacturing system, community planning, energy, logistics, and supply chain. Combining the agricultural field, Digital Twins technology provides monitoring and recording of the entire process from before planting to after the sale of agricultural products, realizing an integrated automated production and sales model from sowing to crop growth and development, to harvesting and selling crops. The construction of agricultural Digital Twins relies on artificial intelligence technology, simulation technology, big data analysis technology, VR technology, blockchain technology, cloud computing technology, and other key technical support. At the same time, there is not a single linear relationship between the various technologies of agricultural Digital Twins. Instead, it relies on the integration of multiple technologies into an integrated technology solution to provide support and services for the agricultural Digital Twins. According to the Internet of things technology, cloud computing technology, MR technology, combined with data interaction, and distributed storage system, the monitoring information is digitized in the basic support layer and data layer and transmitted to the model building layer. The big data analysis technology is used to build an agricultural ecosystem that meets the requirements, combined with Simulation technology, blockchain technology, etc., to transform the deterministic law and complete mechanism model into software to simulate a three-dimensional visualization farm and is committed to a variety of application services at the application layer that complements each other concurrently (Shi et al. 2021). As show in Fig. 2, smart farm can be constructed by Digital Twins.

40

Y. Zhao et al.

Fig. 2 Smart farm constructed by digital twins

2.1 Artificial Intelligence Predicts Plant Growth Plant growth regulation mostly relies on the comprehensive development of a variety of science and technology such as organic synthesis, microanalysis, plant physiology, and biochemistry. The growth of plants from seed germination, leaf growth to flowering, and fruiting are currently mostly artificially observed. It is difficult to predict the growth trend of plants in a comprehensive, specific, and uniform manner. The Digital Twins farm can use sensors to accurately monitor the chemical composition of the soil, soil moisture, and various dynamic data in the field in real time, according to the intelligent forecast of future weather conditions, and the wind direction, wind force, rainfall, light and other influences fed back by the sensors. The climatic factors of crops are based on monitored market data, soil weather conditions, and crop growth and development in real time. In addition, it can also preciscly predict the yield growth and income of crops and realize all-around automatic control of crop products (Su et al. 2021; Kadow et al. 2020). Secondly, digital twins can provide timely early warning of natural disasters, and when natural disasters come, protect crop products as soon as possible, and take corresponding emergency measures to respond to different natural disasters such as insects, snow disasters, and floods. At the same time, digital twins can protect crop products from natural disasters through accurate detection and intelligent analysis (Richard and Dixon 2001). In the basic support layer, according to the soil, weather, wind, wind direction, light, rainfall, and other information collected by proprietary chips and sensors, various factors affecting the growth of crops and the growth and development of crops and other information are converted into digital form and transmitted to the data layer. In the data layer, it includes the actual crop growth data transmitted by the basic support layer, the environmental data affecting agricultural products from the outside, and the data calculated by the virtualization of the model

Agricultural Digital Twins

41

building layer, and it has functions such as data processing, storage, fusion, analysis, and prediction. Based on the related monitoring of crop products in the smart farm, the prediction of market big data models, and the intelligent adoption of measures to respond to natural disasters, an intelligent analysis layer for environmental perception is constructed, and the data provided by the infrastructure layer is uniformly calculated, stored, analysis, testing and acceptance, etc. (Haibe-Kains et al. 2020). Through the real-time analysis of big data, artificial intelligence technology can solve problems such as the inability to control the future growth of crops, to achieve the stage of maximizing total revenue (Wu et al. 2020).

2.2 Virtual Reality Simulation of 3D Digital Farm The existing smart farms do not only need to remotely control machines; they can sow, irrigate, and harvest their 10,000 mu of fertile land. In meeting the needs of food production, farmers still have to face various challenges every day, and the frequency of manual labor is still too numerous to list. Farms are often remote and have high asset values, which makes them vulnerable to various uncertainties for a long time. The Digital Twins smart farm uses Simulation technology to build a threedimensional Digital Twins farm that simulates the true growth of crops and establishes a digital model of the multi-faceted living environment of crops. In the threedimensional Digital Twins farm, it can realize remote control of precise insect and weed control, timely and automatically perform irrigation and fertilization based on real-time monitoring data, and timely simulate crop protection measures according to weather and seasonal changes, and respond to real-time changes in weather conditions and soil level, reduce costs and environmental pollution, while improving the effectiveness of medicines and fertilizers, and generate corresponding automatic solutions based on the growth and development of crops in real time (Arnald et al. 2021). In the model construction layer, it provides farmers with data acquisition and establishment of digital models of crop products, the establishment of digital models of market demand, and digital models of soil composition and historical information in the unit (Pirch et al. 2021). In the Simulation analysis layer, the digital model is integrated into the laws and mechanisms of market demand, weather laws, etc., to simulate and stimulate the three-dimensional visualization farm (Angelica et al. 2021; Pratviel et al. 2021). Through virtual reality technology to simulate the real farm conditions, a highly automated, integrated, mechanized, and integrated smart farm can be realized, and all-round automated sowing, irrigation, deworming, weeding, and harvesting can be combined in the field of Digital Twins and agriculture. The future is no longer an unrealistic fantasy.

42

Y. Zhao et al.

2.3 Blockchain Technology Realizes Supply Chain Management At present, food safety issues are still in a state of the urgent need to be resolved. Food needs to go through multiple processes such as harvesting, processing, packaging, storage, and transportation from the place of production to the consumer’s table. But, due to poor control of harvest and processing time and encounters bad weather, or the use of unqualified packaging materials, or improper storage, or unreasonable shipment, as well as many factors such as harvesting, processing, packaging, storage, and transportation equipment and technology, will cause the product to rot, deteriorate, or be contaminated or polluted. Product safety issues are caused by damage, etc., resulting in lower sales of agricultural products, and consumers cannot rest assured that they can eat agricultural products. As shown in Fig. 3, blockchain technology can be used to solve these issues. Digital Twins farm provides comprehensive management of the agricultural product supply chain, allowing customers to trace the growth and development process information of any agricultural product by scanning the QR code to determine whether the quality of the purchased agricultural product meets the requirements. Based on blockchain technology, a supply chain can be realized, combined with the results of big data analysis of historical data, to predict the demand for certain types of agricultural products in the future market. The farmland is divided according to the demand for agricultural products, and according to the real-time feedback of product demand and land characteristics, the most suitable type of agricultural product planting and the area of agricultural product planting in the unit are reasonably planned, so as to maximize the total income of agricultural products (Aung and Chang 2014). When sowing, the intelligent method is used for automatic sowing, and the seeds of good varieties are selected by intelligent detection and then fully automated sowing to ensure the quality of crop products. After the sale of agricultural products is completed, the customer can provide feedback information on the use of agricultural products, which can be used by the farm to improve the entire process of the next

Fig. 3 Application of blockchain to food safety issues

Agricultural Digital Twins

43

round of agricultural products, and implement the overall improvement and perfection of the farm through an iterative process. Real-time monitoring of the growth status of crop products predicts the best harvest period of crop products and adopts a fully integrated and automated harvesting method during harvest. Before the crop products are harvested, the output of the crop products is estimated, combined with the current market demand information and the market historical demand information, to estimate the future income of the crop products and maximize the benefits. At the application layer, farmers are provided with application services such as real-time calculation of profit maximization based on the conditions of agricultural products. Through blockchain technology, the supply chain management of agricultural products from seeding and growth to processing and production has greatly increased the safety of edible agricultural products during the growth and development process recorded by the QR code, protecting the rights and interests of consumers, and creating a safe food safety environment (Pouliot and Sumner 2008).

2.4 Problems that Still Exist in the Application of Digital Twins in the Agricultural Field First, virtual reality technology reflects the various information conditions of the farm through a digital form, but it is difficult to control the relationship between various plants. By the German scholar H. in the concept of allelopathy proposed by Molisch in 1937, it is believed that the allelopathy of plants is the direct or indirect effect of plants on other plants by secreting chemical substances in the metabolic process from the body. The specific extent of this direct or indirect impact is not reflected in the current Simulation farm. Second, artificial intelligence technology is relatively limited in the prediction of natural disasters. Unexpected earthquakes, fires, droughts, floods, etc., cause huge damage to crops. Natural disasters are one of the important factors affecting agricultural development. There are many types of natural disasters and the situation is complex, and their impact on agricultural production is very serious. How to improve the efficiency of disaster response when sudden disasters come and to protect crop products and the agricultural economy from damage for the first time is still a major problem at present.

3 Conclusion This chapter briefly introduces the origin and development history of Digital Twins and the key technologies used by Digital Twins. Combined with the current applications of Digital Twins in manufacturing, cities, battlefields, and agriculture, it scientifically predicts the possible development direction of Digital Twins in agriculture

44

Y. Zhao et al.

in the future. Combining the application of Digital Twins in the agricultural field, the viewpoint of building a Digital Twins smart farm is put forward, and with this as the core, the supply chain management process of constructing a Digital Twins smart farm is refined in many ways. According to the existing technology in the Digital Twins, it is applied to the pain points of agricultural planting issues, food safety issues, etc., and fully anticipates the development trend of smart agriculture in the future. In addition, there are still many difficulties in the existing Digital Twins technology. Aiming at the possible defects of the current Digital Twins, it predicts the problems that may be left by the application of the Digital Twins in agriculture.

References Aung MM, Chang YS (2014) Traceability in a food supply chain: Safety and quality perspectives. Food Control 39:172–184 Haibe-Kains B, Adam GA, Hosny A, Khodakarami F, Waldron L, Wang B, McIntosh C et al (2020) Transparency and reproducibility in artificial intelligence. Nature 586(7829):E14–E16 Jans-Singh M, Leeming K, Choudhary R, Girolami M (2020) Digital twin of an urban-integrated hydroponic farm. Data-Centric Eng 1e20. https://doi.org/10.1017/dce.2020.21 Kadow C, Hall DM, Ulbrich U (2020) Artificial intelligence reconstructs missing climate information. Nat Geosci 13(6):408–413 Pirch S, Müller F, Iofinova E, Pazmandi J, Hütter CV, Chiettini M, Sin C, Boztug K, Podkosova I, Kaufmann H, Menche J (2021) The VRNetzer platform enables interactive network analysis in Virtual Reality. Nat Commun 12(1):2432. https://doi.org/10.1038/s41467-021-22570-w Pouliot S, Sumner DA (2008) Traceability, liability, and incentives for food safety and quality. Am J Agr Econ 90(1):15–27 Pratviel Y, Deschodt-Arsac V, Larrue F, Arsac LM (2021) Reliability of the Dynavision task in virtual reality to explore visuomotor phenotypes. Sci Rep 11(1):587. https://doi.org/10.1038/s41 598-020-79885-9 Richard A., Dixon (2001) Natural products and plant disease resistance. Nature 411(6839):843–847. https://doi.org/10.1038/35081178 Shi L, Li B, Kim C, Kellnhofer P, Matusik W (2021) Towards real-time photorealistic 3D holography with deep neural networks. Nature 591(7849):234–239. https://doi.org/10.1038/s41586-020-031 52-0 Su Y, Gabrielle B, Makowski D (2021) The impact of climate change on the productivity of conservation agriculture. Nat Clim Change 11(7):628–633. https://doi.org/10.1038/s41558-02101075-w van Dijk M, Morley T, Rau ML, Saghai Y (2021) A meta-analysis of projected global food demand and population at risk of hunger for the period 2010–2050. Nat Food 2(7):494–501. https://doi. org/10.1038/s43016-021-00322-9 Wu F, Lu C, Zhu M, Chen H, Zhu J, Yu K, Li L et al (2020) Towards a new generation of artificial intelligence in China. Nat Mach Intell 2(6):312–316

The Application of Digital Twins in the Field of Fashion Victor Kuzmichev and Jiaqi Yan

Abstract Contemporary fashion is rapidly moving into virtual reality (VR) yearly, especially in pandemic period. With the attempts of internet, on the one side, smart gadgets, smart mirrors and various virtual garment-based IT (information technology), consumers are able to try on garment, evaluate the style and fit and order anytime and anywhere. On the other side, design of new virtual garments significantly reduces labor and material costs. Brands using 3D digital garment design and development processes are recorded up to a 75% reduction in sampling, and a 50–75% reduction in time spent on product development. The new demands of mass produced (ready-to-wear, R-t-W) and made-to-measure (M-t-M) virtual garments encourage huge efforts to formalize the traditional process of planning, production and sales and to innovate the new one by using advantages of virtual technologies. At the same time, digitalization has faced serious problems in obtaining digital twins (DTs) of human bodies and garments, and joining them into a completely new virtual system “virtual human body—virtual garment” due to the diversity of human morphology and physical properties of garments. In accordance with virtual design process, this chapter devotes to two primary DTs: virtual human body and virtual garment derived from virtual pattern block.

1 Digital Twins of Human Bodies DTs of human bodies, also called virtual human models (VHMs), represent humans in the virtual reality (Sjarov et al. 2020) which are used widely in gaming and fashion industries. For games, VHMs named also as avatars should be looked as real people but sometimes with stylish features in according with a hero characters. V. Kuzmichev (B) Department of Clothing Design, Ivanovo State Polytechnic University, Ivanovo, Russia e-mail: [email protected] J. Yan College of Fashion Technology, Zhongyuan University of Technology, Zhengzhou, Henan, China © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 Z. Lv and E. Fersman (eds.), Digital Twins: Basics and Applications, https://doi.org/10.1007/978-3-031-11401-4_6

45

46

V. Kuzmichev and J. Yan

For fashion industry, the VHMs of human bodies are more complex due to special features and applications. Firstly, they should reflect the attributes of physical human body based on particular external appearance, somatotype characteristics, skeleton structure and morphology characteristics (posture, shape, dimensions, etc.), structure and composition (skeletons, muscles, tissues, etc.), mobility and other properties (Nakazawa 2000) which should present in a VHM in terms of its application. The VHMs are the results of modeling system and are intended to reproduce the actual shape and size of the human body with known reliability and morphology. Secondly, VHMs should have the attributes applicable to making, representing and wearing virtual garments with different functions. Thirdly, VHM is supposed to have the essential properties as physical counterpart for digital garment fitting. Thus, the effectiveness of virtual system “VHM—virtual garment” will depend on the both components.

1.1 Virtual Human Models in Fashion Industry In digital fashion, there are two types of VHMs according to standard ISO 204971:2021 (2021) (Fig. 1): (1) Parametric human body PHB is the VHM with changeable parameters such as size, dimensions, shape, height, body mass index; (2) Virtual human body VHB (fashion avatar) is the model for digital fitting based on size, shape, cross sections, skeletal structure, on the one side, and body texture, skin color, head and so on, on the other side. VHBs are classified into three key forms. First key form is the virtual clone virtual shape, scanatar (VC) which is identical to real physical body and equal to 3D scanned point cloud. As shown in Fig. 2a, a virtual clone is created after scanning of real person by forming 3D surface data from

Fig. 1 Virtual human bodies in digital fashion

The Application of Digital Twins in the Field of Fashion

47

a

b

c

Fig. 2 Processes of creating key forms of VHM: VC (a), VT (b), VFM (c)

a 3D point cloud, using surface modeling, noise elimination, hole-filling and mesh generation processes. Second key form is the virtual twin virtual size, avatar (VT) as a PHB morphed according to body measurements and acquired through manual or automatic measurements (Fig. 2b). The virtual twin is standardized typical body lookalike with similar body dimensions and is designed by entering parameters retrieved from a national population database. The virtual twin can be transformed from virtual clone. Third key form is virtual fit mannequin VFM (Fig. 2c) which is the copy of actual human body dress form (or fit form). This mannequin is used for draping simulation, sampling silhouette, fit evaluation and garment visualization. The VHM is a fundamental object of digital fashion due to huge scale of application in terms of how to evaluate the suitability of virtual garments on virtual human body in cyberspace. Table 1 shows the use of VT in digital fashion.

1.2 Source Information for Generating Virtual Human Model To generate a VHM, the next digital information is required (as standard ISO 188252:2016 2016): (1) Body segments/regions—an intact body is usually segmented to ten main regions (neck, shoulder, chest, waist, abdomen, crotch, upper arm, forearm, thigh, calf), considering the human anatomy and functionality for garment construction.

Adequacy for human Partly anatomy and morphology

For forming an avatar library, transforming basic size in others automatically, and interactive simulation

Application

Partly or absolutely VT of standard body lookalike with similarly defined dimensions

For digital fitting through the interactive and automatic simulation of pattern strain, heat map and visual analysis of garment balance, gap between body and garment, surface wrinkles, etc.

R-t-W VR

R-t-W VR

Area

Fitting VT with full information about a size, shape, cross section, and skeletal structure

PHB (full or torso) with changeable parameters, dimensions, posture, shape, etc.

Basic

For R-t-W

Type of virtual twin

Description

Item

Table 1 Virtual twins in fashion industry

Partly

For virtual garment performance, marketplaces and online shopping

Partly. Several measurements could be corrected to proportion improving

For virtual garment performance, personal social groups such as Instagram®

R-t-W, VR + AR

R-t-W, VR + AR (augmented reality)

Sales A perfect nice-looking VT as virtual influencer AYAYI

Fashion design VT with body texture representing a basic design size, for example S,M,L

With AI

Absolutely (in the future)

For digital fitting through automatically analysis of garment balance, pattern strain, gap between body and garment, heat map, surface wrinkles, etc.

R-t-W, M-t-M VR + AR

(continued)

Absolutely

For digital fitting of customized and personalized garments

M-t-M. VR + AR

PHB with artificial VC as the graphical sensors for representation of the detecting, analyzing user and automatically correcting problems in digital fitting

For M-t-M

48 V. Kuzmichev and J. Yan

Solid surface

Flexible, any poses

Simplified head, feet, hand, skin

Looks like a physical body with perfect appearance

Looks like a physical Looks very body in total appealing like a model

Solid and deformable Solid surface (compressed) surfaces in terms of fat and muscle

Flexible, any poses

Appearance

Solid surface

Type of surface

Flexible, any poses

Sales

VT shape is equal to VT copying a real real or smoothly model shape corrected mannequin

Fashion design

Represents a defined Represents an A real size and size of standard body attractive design size shape of individual and shape person

Flexible, standing, walking poses

Ergonomic features

VT imported from the 3D avatar library by applying traditional and additional body dimensions

Fitting

Size and proportions Easy to adapt to different body shapes and dimensions

PHB designed on a database which consists of limited number of body dimensions

Basic

For R-t-W

Type of virtual twin

Method of obtaining

Item

Table 1 (continued)

Looks as standard human body

A real size and shape of typical standard body

Solid, deformable and sensory sensitive surface in terms of fat, muscle and receptors

Flexible, any poses

VT as high-technological object with integrated sensors and soft tussle and muscles deformability

With AI

Looks as real person

A real size and shape of individual person

Flexible, any poses

VC created by forming 3D scanned point cloud

For M-t-M

The Application of Digital Twins in the Field of Fashion 49

50

V. Kuzmichev and J. Yan

a

b

Fig. 3 Digital source information for VHM generating: virtual bones and joints (a), virtual body landmarks (b)

(2) Cross sections—closed contours of body surface extracted from the coronal, sagittal, transverse and other angles’ planes obtained on the basic anthropometrical levels (neck, bust, chest, waist, hip, tight, etc.); (3) Body dimensions—corresponding anthropometric dimensions measured of a human body in standard standing position; (4) Skeletal structure—frame from bones and joints which is more simplified now than those in human anatomy. Figure 3a shows the main bones of skeleton structure; (5) Body landmarks—points that defines characteristic of the human body surface in standing position. Figure 3b shows the main anthropometrical landmarks; (6) Body texture—surface appearance of VHM related to mapping elements of images, such as skin, hair, colors, tones; (7) Body poses—positions which are presenting the human ergonomic activities in garments including standing, walking, knee bending, arm rising, elbow bending, diving, etc. to evaluate the fit of virtual garments in dynamic; (8) Motion—activity of changing body poses in dynamic; (9) Flesh—representation of muscle and fat as modeling element for creating VHM and virtual motions in accordance with a national sizing system.

1.3 Tools for Virtual Body Model Digitalization 3D digitalization of the real human body demands transformation from physical information to digital source information that can be applied in digital fitting systems. A VHM is generated with the help of a series of tools: (1) Measuring tools: The required dimensions (heights, straight distances, girths) can be measured manually by anthropometrical devices or automatically by 3D body scanners. The contemporary 3D body scanners—light-based (e.g.,

The Application of Digital Twins in the Field of Fashion

51

TELMAT® , [TC]2® , SizeStream® ) and laser-based (e.g., Human Solutions® )— can capture the optical data, and the compatible software that control the scanning procedure and process data. Because VC is the digital replica of real body, is it possible to create new morphological and garment features (e.g., cross sections, volumes, proportions) instead of traditional linear measurements (Pei et al. 2019) and to design directly 3D garments on VC and to generate 2D flattened pattern by 3D-to-2D techniques (Petrak and Rogale 2006). Moreover, abundant traditional and unlimited number of self-defined body dimensions could be obtained respectively automatically or manually with VC in default or third-party software (Xia et al. 2019). (2) Standard size chart: A standard VTs specifications refer to the dimensions of particular population from national standards or enterprises’ documents. (3) VHM generating software: VHM can be usually generated automatically or manually. Figure 4 shows the VT generated in Style 3D® with different features; (4) Virtual skeleton rigging software: VCs from 3D scanning cannot instantly be applied for digital fitting owing to deficiency of virtual skeletons and joints. 3D platform or softwares can rig virtual skeletons and joints in 3D models, e.g., Adobe Mixamo® , Blender® . (5) Detailing software: some additional software are used for sculpturing the detail shape and editing the surface texture of a VC, e.g., Zbrush® , Adobe substance painter® ; (6) Postprocessing software: 3D industrial modeling CAD, such as Rhinoceros® , helps to modify and obtain necessary information with a VHM.

Fig. 4 Examples of VTs generating in Style 3D®

52

V. Kuzmichev and J. Yan

a

b

Fig. 5 VFM generating: scheme of finding axis and feature points (left) and superimposed cross sections of chest, waist and hip for calculating average information of typical VFM (a), European standard VFM from Alvanon® Body Platform (b)

1.4 Virtual Fit Mannequin Generating In order to fit the most population, R-t-W garments are usually designed with the VFM in accordance with the national sizing systems. VFM can be obtained in two ways: (1) scanning the counter part of physical standard mannequin and outputting as digital form, (2) automatically generating by inputting basic dimensions or information through existing algorithms (Li and Chen 2009). Figure 5a illustrates the algorithm of extracting cross sections from VCs belonging to similar body type and superimposing them. After averaging the cross sections and doing other calculations, new typical VFM could be generated. For instance, Alvanon® Body Platform established a database covering 6000 virtual Alvaform® VFM which met different national standards (Fig. 5b).

2 Digital Twins of Garment According to standard ISO 20947-2:2020, a digital twin of garment, namely virtual garment (VG), is a virtual model designed to accurately reflect its physical properties such as draping, gravity, surface strain, shape and other features and existing in virtual space.

The Application of Digital Twins in the Field of Fashion

53

2.1 Structure of Virtual Fitting System A virtual fitting system is the synthesis of VHM, VG and other virtual objects under modeling or simulation process. The contemporary virtual fitting systems, e.g., CLO 3D® , Style 3D® , Vstitcher® , TUKA 3D® , can represent the physical counterpart by completely design and simulate garments in virtual space with higher productivity and less resources consumption. As shown in Fig. 6, the virtual garment systems operate with a combination of several interrelated modules: (a) Virtual pattern module: drafting flat shapes expressing necessary garment features, and arranging 3D pieces around a VHM; (b) Virtual fabric module: generating virtual fabrics with digital properties, e.g., draping, tensile modulus, bending rigidity, shear resistance, thickness, weight, appearance characteristics (color, shining, transparent, facture, etc.). The virtual fabrics can be instantly selected from the built-in library, or generated through specific algorithms transforming the real properties to the digital ones (Kuijpers et al. 2020). (c) Virtual auxiliary material module: editing the specific technological and decorated elements, e.g., closures; (d) Virtual human module: generating VTs or importing VCs; (e) Virtual garment module: visualizing and evaluating 3D garment in real-time; (f) Additional modules such as design library, picture and video rendering, 3D runway.

Fig. 6 Virtual garment system and its components

54

V. Kuzmichev and J. Yan

a

b

c

Fig. 7 VG presentations in Style 3D® : a virtual women blouse (a), transparent mode (b), visualization of surface strain with heat map (c)

With the integrated virtual fitting system, the exterior appearance of the garment can be visualized objectively after sewing, and the quantitative fit indicators (e.g., gap, distortion rate, compression pressure under garment, fabric tensile), presented visually by heat map and measured on the garment surface (Yan and Kuzmichev 2020). Figure 7 shows a virtual women blouse after sewing (by stitching virtual patterns around the virtual body) in Style 3D® software and additional presentations for qualitative and quantitative evaluations in terms of criteria related to surface strain, gap, compressive pressure, surface wrinkles, balance, etc. (ISO 20947-2:2020 2020). Concretely, the basic fitting criteria is the garment balance which indicate the correct relationship between the virtual human body sizes, garment contour and body posture. It is usually determined by the horizontal hemline and vertical center lines or side seams. As shown in Fig. 7a, the virtual blouse displays overall good fit and balance with well-positioned structural lines and the seams. Gap is the distance between a virtual garment and virtual human body (Fig. 7b) and can be changed from zero to several centimeters depending on body types and garment style. Pattern strain presents the amount of deformation caused by the disproportion between the dimensions of VHM and pattern in the drape simulation process after the body covering. It can be visualized in different ways. As shown in Fig. 7a, c, the pattern strain is visualized by a surface heat map measured in kPa or by location of wrinkles and creases. There are two ways of virtual garment generating (1) 2D-to-3D, (2) 3D-to-2D.

2.2 Generating Virtual Garment from Virtual Patterns The 2D-to-3D method is similar as the real garment cutting and sewing and 3D virtual garment is formed referring to 2D patterns. Firstly, a VHM should be assigned, generated or imported according to the certain requirements. Secondly, flat patterns should be constructed based on the body dimensions and the ease allowance in pattern

The Application of Digital Twins in the Field of Fashion

55

design. Ease allowances (E) are the values of the difference between the pattern dimensions and the body dimensions, which provides physical fit and comfort, permit body movement and helped to construct different garment style (Scott et al. 2019). Thirdly, virtual shell, lining or interlining fabrics of each piece of pattern should be assigned from interior library or exterior data. Fourthly, pattern pieces should be arranged on the suitable positions around the avatar and sewed together. Fifthly, virtual garment fit can be visualized in different views by observing the 3D exterior appearance in terms of criteria chosen. Adjustments to the pattern, avatar, fabric, etc., can proceed by iterating the circulation from first to fifth steps until the final approval. The final virtual garment can be rendered as graphics or videos, and directly outputted for real production.

2.3 Generating Virtual Garment Directly on Virtual Human Model A virtual garment can be directly construct on the virtual body without drafting flat pattern, which is called 3D-to-2D design method as Dong et al. (2018) presented (Fig. 8). As shown, the pattern is constructed directly in 3D virtual environment and generated from the avatar surface. For one thing, the patterns obtained are similar to human skin and could be used for producing the close-fitting garments with zero gap. For another, with the gap as distance ease allowances added in matched landmarks on a VHM, the patterns of bigger dimensions could be obtained (Hong et al. 2017). This method can been applied with good accuracy to one-layer garments of simple styles. Fig. 8 Scheme of 3D-to-2D design method

56

V. Kuzmichev and J. Yan

3 Future Development All innovations in digital fashion are driven to get a comfortable, good-fitting and functional garments with limited resources. Under this way, there are many problems related to measuring human body, to bridge 3D and 2D technologies, and digitization of apparel aspects which should be solved. To solve these problems, the professional knowledge and workmanship should be presented in formalized forms to be easy transform in VR: “…fit will depend upon our ability to combine the knowledge and skills of the nineteenth century tailor or dressmaker with the strengths of manufacturing and information technology. The continued development and conservation of both our technological and intellectual resources is critical (Bye et al. 2006).” The technology of virtual twin generation will drive in three main directions: Firstly, developing of virtual typical standardized twins in accordance with national sizing systems for M-t-M clothes; Secondly, integrating of AI elements in virtual twin to analyze it reactions during digital fitting test of different virtual garments to solve the challenges of sizing and fit inherent in the clothing industry; Thirdly, generating of virtual twin of real person in terms of its morphological, psychological and other features which are influencing on consumer purchase. To get these results, next directions could be developed: • Increasing database about typical standard bodies and create body shape size charts with a mathematical foundation by adding to existing linear measurements (such as height, bust, waist) a new dimensions and characteristics to draw and control frontal and profile contours and position of virtual twin in 3D virtual reality (such as projection dimensions, cross sections in sagittal, coronal, and transverse planes, segmentation of full body dimensions for effective analysis of body morphology and key measurements distribution); • Developing a unique and innovative body data-driven and body-to-pattern approach about human bodies with different weight, fat tissue and muscular mass reshaping in static and dynamic poses by compression garments and modeling of virtual human body reshaping by these garments. A model should base on relations existing between the same measurements before and after shaping; • Developing an algorithm of avatar surface segmentation to comfort evaluation in accordance with reaction of real body on compression, friction and other interactions between a body and wearing garment.

The Application of Digital Twins in the Field of Fashion

57

References Bye E, LaBat KL, DeLong MR (2006) Analysis of body measurement systems for apparel. Cloth Text Res J 24(2):66–89. https://doi.org/10.1177/0887302X0602400202 Dong Z, Jiang G, Huang G et al (2018) A web-based 3D virtual display framework for warp-knitted seamless garment design. Int J Cloth Sci Tech 30(3):332–346. https://doi.org/10.1108/IJCST-052017-0060 Hong Y, Zeng X, Bruniaux P et al (2017) Interactive virtual try-on based three-dimensional garment block design for disabled people of scoliosis type. Text Res J 87(10):1261–1274. https://doi.org/ 10.1177/0040517516651105 ISO 18825-2:2016 (2016) Clothing—digital fittings—part 2: vocabulary and terminology used for attributes of the virtual human body. The International Organization for Standardization, Switzerland ISO 20947-1:2021 (2021) Performance evaluation protocol for digital fitting systems—part 1: accuracy of virtual human body representation. The International Organization for Standardization, Switzerland ISO 20947-2:2020 (2020) Performance evaluation protocol for digital fitting systems—part 2: virtual garment. The International Organization for Standardization, Switzerland Kuijpers S, Luible-Bär C, Gong H (2020) The measurement of fabric properties for virtual simulation—a critical review. IEEE SA Industry Connections, pp 5–38 Li J, Chen J (2009) A mannequin modeling method based on section templates and silhouette control. Int J Cloth Sci Tech 21(5):300–310. https://doi.org/10.1108/09556220910983795 Nakazawa S (2000) International apparel design tutorial, the human body and clothing: the beauty of human body, structure elements, pattern. China Textile & Apparel Press, pp 1–64 Pei J, Park H, Ashdown SP (2019) Female breast shape categorization based on analysis of CAESAR 3D body scan data. Text Res J 89(4):590–611. https://doi.org/10.1177/0040517517753633 Petrak S, Rogale D (2006) Systematic representation and application of a 3D computer-aided garment construction method: part I: 3D garment basic cut construction on a virtual body model. Int J Cloth Sci Tech 18:179–187. https://doi.org/10.1108/09556220610657943 Scott E, Gill S, Mcdonald C (2019) Novel methods to drive pattern engineering through and for enhanced use of 3D technologies. In: Proceeding of 3DBODY.TECH 2019—10th international conference and exhibition on 3D body scanning technologies and processing technologies. Lugano, Switzerland, pp 132–138. https://doi.org/10.15221/19.211 Sjarov M, Lechler T, Fuchs J et al (2020) The digital twin concept in industry—a review and systematization. In: 25th IEEE International conference on emerging technologies and factory automation (ETFA). IEEE, vol 1, pp: 1789–1796. https://doi.org/10.1109/ETFA46521.2020.921 2089 Xia S, Guo S, Li J et al (2019) Comparison of different body measurement techniques: 3D stationary scanner, 3D handheld scanner, and tape measurement. J Tex I 110(8):1103–1113. https://doi.org/ 10.1080/00405000.2018.1541437 Yan J, Kuzmichev VE (2020) A virtual e-bespoke men’s shirt based on new body measurements and method of pattern drafting. Text Res J 90(19–20):2223–2244. https://doi.org/10.1177/004 0517520913347

Digital Twins Collaboration in Industrial Manufacturing Radhya Sahal, Saeed H. Alsamhi, and Kenneth N. Brown

Abstract Digital twins (DTs) play a vital role in achieving the vision of Industry 4.0. They have been used to facilitate business and maximize profits by providing up-to-date operational data representation of physical assets to manage manufacturing machines’ performance, effectiveness, and quality. Creating a DT requires different elements, including IoT sensors, communications networks, connectivity, and digital platforms. Thanks to blockchain, AI, big data, computing paradigms, and automaton technologies, the DT can receive continuous, real-time data and predict the potential risks within the cyber-world. Moreover, the DTs can help in production within industrial manufacturing by easing collaboration among different things (i.e. people, nodes, machines, devices, etc.) on the factory floor. The DTs collaboration plays a significant role in achieving higher quality production, improving decisionmaking, and saving labour costs. Substantially, the DTs collaboration is still mostly at a conceptual stage to demonstrate wide industrial adoption and become well-defined engineering practice within the industry. Consequently, this chapter has introduced a lightweight framework of DTs collaboration for industrial manufacturing. The framework aims to empower more intelligent and collaborative solutions based on DTs for industrial manufacturing. Furthermore, we describe how the framework can be applied for three industrial manufacturing use cases: energy, railway, and logistics. Finally, we highlighted the future directions to guide interested researchers in this interesting area.

R. Sahal (B) · K. N. Brown School of Computer Science and IT, University College Cork, Cork, Ireland e-mail: [email protected] K. N. Brown e-mail: [email protected] S. H. Alsamhi Insight Centre for Data Analytics, National University of Ireland, Galway, Ireland e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 Z. Lv and E. Fersman (eds.), Digital Twins: Basics and Applications, https://doi.org/10.1007/978-3-031-11401-4_7

59

60

R. Sahal et al.

1 Introduction DTs are mirroring the real worlds into virtual worlds. They can describe the factory’s detailed presentation, including machines, devices, robots, production, and processes (Fuller et al. 2020; Haag and Anderl 2018). In manufacturing, DT technology can be used to represent the end-to-end manufacturing processes, starting from product design, monitoring of products manufacturing, planning and execution to prognostics and health management (Rosen et al. 2015; Fei et al. 2018). Consequently, DTs give a big picture of the product’s lifecycle from product design until they are delivered to the customers at various levels, including (i) component, (ii) process, (iii) asset, and (iv) system (Qi and Tao 2018; Kritzinger et al. 2018). For the component level, the DTs are used to represent a single component within the manufacturing process (Lu et al. 2020; Lohtander et al. 2018). The DTs represent a single piece of equipment within a production line for the asset level. For the process level, DTs describe any industrial process within the product lifecycle from design to use of the finished product by customers (Madni et al. 2019). Finally, the DTs are used at the system level to monitor and improve an entire production line. Indeed, industrial manufacturing systems are complex systems that are hard to represent and monitor by a single DT. Therefore, multiple DTs collaborate to monitor a manufacturing component, asset, process, and system to understand better what is happening in the production line and improve product quality, enhance traceability, and predict the potential risks within the manufacturing environment. In particular, DTs collaboration means sharing and exchanging information among industrial entities and sharing tasks to act accordingly (Sahal et al. 2021a, b; Alsamhi et al. 2021a, b, c, d, e). Thus, the DTs collaboration paves the way for the cyber-physical integration of intelligent industrial manufacturing. For example, the DTs collaboration provides an intelligence level of interoperability to track the information related to the operations on the factory floor. Then, the manufacturers’ management floor can make actionable decisions based on predicted impending failures and schedule prior maintenance to avoid downtime (Vathoopan et al. 2018).

1.1 Contribution Our main contributions in this chapter can be summarized as follows: • We have discussed the concept of DTs collaboration for industrial manufacturing. • We have proposed a lightweight framework of DTs collaboration for industrial manufacturing. The proposed framework empowers more intelligent and collaborative solutions based on DTs for improving industrial manufacturing. • We have described how the proposed framework mapped to three industrial manufacturing use cases, including energy industry-fault diagnosis of wind turbines, railway industry-predictive maintenance, and logistic industry-dynamic routing.

Digital Twins Collaboration in Industrial Manufacturing

61

1.2 Chapter Organization The remainder of this chapter is organized as follows: The proposed lightweight framework of DTs collaboration for industrial manufacturing is introduced in Sect. 2. The use cases of industrial manufacturing are described in Sect. 3. The future direction are discussed in Sect. 4. Finally, conclusions are presented in Sect. 5.

2 Lightweight Framework of Digital Twins Collaboration for Industrial Manufacturing DT is one of the core elements of industrial manufacturing digitalisation, which can represent a real-world industrial system such as production systems in a virtual space (Lu et al. 2020). Multiple DTs are used to describe the industrial manufacturing system in hierarchical levels (Benkamoun et al. 2014): (i) DTs in a flat network represent individual things at the machine level. They exchange information with each other on things and learn about their operation and behaviour to build a common understanding of the machine condition, (ii) DTs for things in a tree or a chain represent the subsystem level (i.e. a workstation) and the system level (i.e. shop floor) where each DT is passing on information to the next level (Catarci et al. 2019). Multiple DTs are deployed to represent the up-to-date industrial manufacturing data of the physical assets in operation, including asset status and the relevant historical data. The deployed DTs can intelligently collaborate by utilizing the intelligence of DT-driven operational data to predict the potential risks within the industrial manufacturing systems. In particular, the DTs are collaborating by applying predictive data analytics to analyse DT-based historical operating data, learn about their things using shared knowledge and real-time data, and then predict the potential risks in real-time such as COVID-19 (Sahal et al. 2022; Alsamhi and Lee 2020). A better understanding of the predicted potential risks can facilitate decision-making among participants on the management floor within the industrial manufacturing systems. Accordingly, having a DTs collaboration within a manufacturing environment can offer significant advantages in deploying industrial manufacturing solutions in a decentralization fashion such as wind energy farms and logistics industry units (Perno et al. 2022). The DTs share and exchange information among distributed entities and share tasks to act accordingly. In Fig. 1, we have described the proposed lightweight framework of DTs collaboration for industrial manufacturing. The proposed framework empowers more intelligent and collaborative solutions based on DTs for industrial manufacturing. It introduces one higher level in the cyber-physical system. The proposed framework’s merit is exploiting the emerging industrial technologies’ capabilities, including blockchain, AI, predictive analysis, cloud computing, and edge computing, to provide autonomous, secure collaborative solutions for industrial manufacturing. The framework could be developed and implemented on top of any deployed cyber-physical system based on DTs. Four layers are used to equip

62

R. Sahal et al.

Fig. 1 Proposed lightweight framework of digital twins collaboration for industrial manufacturing

the lightweight framework of DTs collaboration with operational data intelligence. As shown in Fig. 1, the four layers are the physical layer which contains shop floor participants and management participants, the DTs layer, the industrial technologies layer, and the applications layer. These layers will be elaborated flowingly.

2.1 Physical Layer The physical layer contains all nodes involved in the industrial manufacturing system. These nodes could be divided into shop floor participants and management floor participants. The shop floor participants include the factory, machines, workers,

Digital Twins Collaboration in Industrial Manufacturing

63

monitoring devices (e.g. CCTV), sensor devices, robotic devices, etc. On the other hand, the management floor participants can be the people who can contribute by using their operational data (e.g. the decision-makers, the employees in management departments, HR, security staff, and so on).

2.2 Digital Twins Layer The proposed framework introduces collaborative DTs for industrial manufacturing systems, which provides autonomous collaborative industrial manufacturing solutions based on DT-driven operational data. Therefore, this layer is responsible for representing a virtual replica of the industrial manufacturing system at different levels by providing multiple DTs with up-to-date operational data. The data-driven DTs collaborations help to understand the DT status, interact with other DTs at the edge level, learn from other DTs, and share common semantic knowledge within industrial manufacturing systems. The DTs provide real-time data of the production system for the processes, equipment sensors, outputs from management units (e.g. decisionmaking, customer service and purchasing). The DT-driven data are used as inputs for machine learning systems to predict the potential risks within the product lifecycle. The intelligence of DT-driven operational data help makes a timely decision to avoid downtime risks and maximize profits.

2.3 Industrial Technologies Layer This layer briefly highlights the role of emerged industrial technologies to build a concert industrial manufacturing solution based on DTs collaboration. • The blockchain network is used to connect multiple DTs through using distributed ledger technology (DLT). The DT-based blockchain network offers secure distributed operational data management and analytics across multiple participants (Hasan et al. 2020; Sahal et al. 2021a, b; Lee et al. 2021). • Pairing AI with DTs technologies creates new efficiencies for the industrial plant. Applying predictive data analytic techniques (e.g. machine learning and deep learning) using data-driven DTs provides predicted potential risks with production systems such as early detecting faults indicated by degraded performance or damaged physical counterpart (e.g. node, device) (Kapteyn et al. 2020). • Due to the generation of large volumes of industrial data, the predictive data analysis using DTs operational data could be performed on computing paradigms such as cloud and edge computing to leverage extra computing capabilities for real-time analysis (Brovkova et al. 2021; Borodulin et al. 2017). • The industrial data visualization tools provide valuable dashboards to visualize the DTs operational data in real-time for a quick, clear understanding of physical

64

R. Sahal et al.

assets within manufacturing and allow decision-makers to conclude and insights more quickly.

2.4 Application Layer DT can be used in all industrial manufacturing systems (e.g. energy industry, rail industry, logistic industry, mould industry and healthcare industry) at diffident levels, from product design, manufacturing planning and execution to prognostics and health management (Perno et al. 2022).

3 Digital Twins Collaboration in Industrial Manufacturing Use Cases This section discusses three use cases, including energy industry-fault diagnosis of wind turbines, railway industry-predictive maintenance, and logistic industrydynamic routing (see Fig. 2).

3.1 Energy Industry-Fault Diagnosis of Wind Turbines This subsection presents an overview of the energy industry with a detailed mapping of our proposed framework. Overview The energy industry has been growing significantly over the past decades, increasing wind farms growth. DTs represent the wind energy system composed of the DTs’ representation of the geographical wind farms. The DTs collaboration is used to optimize the operation of the wind energy system (Sahal et al. 2020). For example, the integration among DTs having real-time wind data enables diagnosis faults and then perform early maintenance of physical assets, systems, and production processes to increase the life cycle of the wind systems (Sahal et al. 2021a; b). Mapping the digital twins collaboration-based lightweight framework to fault diagnosis of wind turbines The physical asset could be an engine or a turbine with a set of sensors that can collect real-time data and operational status about the wind system (see Fig. 2). The DTs system, which represents the wind energy system, can collaborate to diagnose the fault of the wind system caused by hardware failures within a turbine. The DTs collaboration can understand each DT status, interact with other DTs, learn from each other DTs, and share common semantic knowledge across geographically wind farms. Furthermore, the DTs of the wind system are collaborated to track wind farms and then identify the potential failures by visualizing the change of the wind system over time. The blockchain network is used to

Digital Twins Collaboration in Industrial Manufacturing

65

Fig. 2 Selected industrial manufacturing use case including; energy, railway, and logistics

securely exchange wind data of DTs for the multiple turbines among geographically distributed farms. Predictive data analytics supports health analyses and maintenance decisionmaking by leveraging digital simulation and data-driven intelligence (Song et al. 2020). For instance, data analytics can be used within DTs’ interaction and communication to describe, diagnose, predict, and prescribe the physical wind system’s behaviour for fault diagnosis. For the turbine failures, the predictive data analytics techniques are applied over DTs-driven data to detect the DTs that contain the faulty operational data and then locate the sensors within a specific turbine in a particular wind farm that the human eye would miss. The data analytics outcome will be used as inputs to notify the decision-makers to make the best decision for abnormal data in case of potential failure over the wind energy industry system. Ultimately, the geographically produced wind data can help energy manufacturers perform their analysis by deploying predictive maintenance solutions based on DTs using scalable big data platforms launched on cloud or edge (Hu et al. 2018).

66

R. Sahal et al.

3.2 Railway Industry-Predictive Maintenance In this subsection, an overview of railway maintenance is presented together with a detailed mapping of our proposed framework. Overview Railway 4.0 is one of Industry 4.0 dimensions using new digital technologies including big data, IoT, DT, AI, and cloud computing (Sahal et al. 2020; Kans et al. 2016; Gupta et al. 2020). The railway companies compete to provide high and attractive service to the passengers by utilizing automation and emerging technologies. One of the rail industry’s significant challenges is avoiding delays to meet passengers’ satisfaction and maximize their profits. To do that, the rail companies start to deploy predictive maintenance applications to diagnose the fault and perform maintenance actions early. Nowadays, railway companies use DTs to improve railway performance by utilizing railway DT-based operational data analysis intelligence (Sahal et al. 2021a; b). In particular, they use the DTs collaboration to gain improved information visibility and better understand the past, present, and future predictions. With the DTbased prediction of the potential risks, the decision-makers in rail companies can support the transformation of rail track maintenance and deliver safe, reliable, and resilient service (Aivaliotis et al. 2019). Mapping the digital twins collaboration-based lightweight framework to railway industry-predictive maintenance The rail sector is a complex network of physical assets and systems that come together to enable people and goods to travel safely, in a timely way at various speeds and distances. The physical rail assets have the station, process, and individual assets such as train, switch, sensors, etc. (see Fig. 2). These physical rail assets could be represented in interoperable and collaborative DTs to show high visibility of the rail supply chain. The DTs represent the rail supply chain of the railway industry. The blockchain network is used for the secure exchange of rail data of DTs for the rail industry participants within the rail supply chain of the railway industry (Altun and Tavli 2019; Liao et al. 2021). The DTs representing the rail supply chain of the railway industry can collaborate to diagnose the railway’s fault, whether caused by hardware failures (e.g. rail, train, switch, etc.) or weather conditions that affect the rail, e.g. temperature and humidity. The DTs collaboration can understand each DT status, interact with other DTs, learn from other DTs, and share common semantic knowledge across geographical railways. The collaborative DTs-based predictive maintenance solutions that use new industrial technologies must be performed on high computing capacities such as edge computing and cloud computing. Predictive data analytical models are used to support decision-making by utilizing the intelligence of the DTs-based operational data. For instance, data analytics can be used within DTs’ interaction and communication to describe, diagnose, predict, and prescribe the behaviour of the physical rail system for fault diagnosis. The outcome of the data analytics will be used to make the best decision for abnormal data or warn the decision-makers in case of potential failure over the rail industry supply chain. In addition, the rail industry faces significant challenges regarding scalability,

Digital Twins Collaboration in Industrial Manufacturing

67

reliability, and flexibility. So, using cloud computing services for the rail industry can provide rail companies with better operations efficiency.

3.3 Logistics Industry-Dynamic Routing This subsection presents an overview of the logistics industry with a detailed mapping of our proposed framework. Overview Recently, Logistics 4.0 technologies have emerged as one of the dimensions of Industry 4.0, including intelligent robotics, self-driving vehicles, and automated systems for managing the movement of products among warehouses and factories (Jabeur et al. 2017). Logistics 4.0 solutions aim to create interoperable and connected logistics chains to become more innovative. Moreover, the logistics process generates big data generated by tracking the movement of goods. Mapping the digital twins collaboration-based lightweight framework to logistic industry-dynamic routing According to the Industry 4.0 principles, Logistics 4.0 can be described as collaborative cyber-physical systems. Therefore, collaborative DTs are used to represent the logistics data (Sahal et al. 2021a; b). The DTsbased logistics data are used for potential logistics optimization by monitoring physical assets and other equipment to eliminate downtime within the logistics system. The logistic physical asset could be a fleet, truck, ship, container, robots, warehouse, and people with a set of sensors that can collect real-time data and operational status about the logistics supply chain (see Fig. 2). The digital logistics supply chain twin is used for end-to-end product tracking and identifying issues by visualizing goods’ digital movements over time. Furthermore, the decentralized digital logistics supply chain twin comprises the DTs representing the geographical logistics warehouses, logistics centres, and participants. The logistic data generated from the logistics supply chain elements are represented in DTs. The blockchain network is used for secure exchange logistics data of DTs for the logistic supply chain. The collaborative DTs are adopted to visualize the logistics supply chain, which could track products and provide end-to-end service from unloading at the quayside to shipping goods to their destinations. The collaborative DTs framework provides smart logistics service for a faster flow of goods, real-time analysis of comprehensive supply chain data, better synchronization of dynamic routing logistics processes, unbroken shipment tracking to improve distribution planning and delivery reliability. With dynamic routing, logistics systems have a flexible and powerful scheduling capability to deliver the products to the customers on time to keep their reputation and quality of services aligned with the customers’ stratification. The logistics companies can use real-time information such as weather or construction delays to change carrier routes on the fly. They can plan for future shipping routes based on historical logistics operational data. Also, they can increase their profits by taking place the dynamic routing to continue delivering a flawless shipping experience.

68

R. Sahal et al.

To assess the potential risks within the logistics manufacturing (e.g. whether it is harmful to the products such as medicines and frozen food), a prediction is needed to estimate how long a refrigerated truck will require to arrive at one or more processing plants. Therefore, dynamic routing is essential to direct the fleets based on past experiences and real-time tracking of the on-road performance of the fleet. In addition, when any problem occurs due to weather or roadblocks, the dynamic rerouting feature helps decision-makers suggest alternate and efficient routes for delivery. Consequently, many variables are operating within trucks that are needed to be monitored (i.e. temperature and humidity inside the containers, driver’s road time, and the route conditions). The data generated from the elements regarding the refrigerated truck could be represented into DTs at hierarchical levels: (1) With local DT at each container, (2) intermediate more powerful DTs on the refrigerated truck at the network edge, (3) much more powerful when DTs represent the logistics units in the cloud. The refrigeration unit’s collaborative DTs system can predict the product state and queue length using the DT-driven operational data. Based on these predictions, the decision can be made to direct the truck to the best plant to avoid potential risks. In addition, with cloud computing, logistics companies can leverage cloudbased services as a way to grantee scalability, facilitate efficient logistics and data management, and support data storage.

4 Future Directions 4.1 Security and Privacy The security and privacy associated with DTs are challenging within smart transportation because of the massive amount of data and the risk of sensitive data created from smart transportation systems. Therefore, IoT devices should analyse DTs data locally using federated learning and then share only the model to the blockchain instead of sending the raw data. Thus, the issue of security can be solved by using blockchain technology, while privacy can be solved by using federated learning. The combination of both techniques can significantly enhance the security and privacy of DTs in transportation systems.

4.2 Connectivity With the growth of intelligent devices in intelligent transportation systems, connectivity is still a challenge for these smart devices to achieve real-time goals. The massive number of intelligent machines in intelligent transportation needs for advanced communication technologies like Beyond Fifth Generation (B5G) or Sixth

Digital Twins Collaboration in Industrial Manufacturing

69

Generation (6G) (Alsamhi et al. 2021a, b, c, d, e; Alsamhi and Rajput 2012, 2014; Ansari et al. 2020) in smart environment (Almalki et al. 2021a, b; Alsamhi et al. 2019a, b; Saif et al. 2021a, b, c) and risk zone (Alsamhi et al. 2018a, b, c; 2019a, b; 2021a, b, c, d, e; Saif et al. 2021a, b, c). Blockchain may help devices borrow data from neighbouring devices to keep the transportation system working efficiently if any smart devices get disconnected. Running machine learning at the edge may ensure full connectivity, high accuracy, and prevent missing data.

4.3 Timing, Speed, and Response Timing and speed are tricky for the DTs. For starters, time enhances decision-making and reaction times for customer service demands which require high accuracy and prompt replies.

4.4 Data Modelling Standardization is essential for adopting DTs in different industrial manufacturing. The fully connected DTs need to use standard models to define DTs schema based on their corresponding physical assets and the communication behaviour within the virtual world. These standards are complex to facilitate DTs interactions and collaboration in different domain areas. Furthermore, these standards can range from the file format of the data storage to the details of how the DTs are communicating to address the requirements of different collaboration within the industrial manufacturing units and departments over different geographical areas (Rasheed et al. 2020).

5 Conclusion There are some advantages of using DTs technology in manufacturing, including; (i) increasing the reliability of equipment and production lines, (ii) improving data security, productivity and quality of product, (iii) reducing maintenance costs by predicting timely maintenance before breakdowns occur, (iv) making faster decisionmaking to avoid downtime risks and maximize profits, and (v) efficient supply and delivery chains. Furthermore, DT collaboration is beneficial for smart transportation, smart cities, smart healthcare, combating COVID-19, smart logistics, and smart environments. This motivates us to introduce a lightweight framework of DTs collaboration for industrial manufacturing to empower more intelligent and collaborative solutions based on DTs for industrial manufacturing. First, we have presented the proposed lightweight framework’s multilayers: the physical layer, which contains shop floor participants and management participants, the DTs layer, the industrial

70

R. Sahal et al.

technologies layer, and the applications layer. Then, we have described three use cases to discuss how the proposed framework applied for industrial manufacturing: energy industry-fault diagnosis of wind turbines, railway industry-predictive maintenance, and logistic industry-dynamic routing. Acknowledgements This research has emanated from research supported by a research grant from Science Foundation Ireland (SFI) under Grant Number SFI/16/RC/3918 (CONFIRM), and Marie Sk-lodowska-Curie grant agreement No. 847577 co-funded by the European Regional Development Fund.

References Aivaliotis P, Georgoulias K, Arkouli Z, Makris S (2019) Methodology for enabling digital twin using advanced physics-based modelling in predictive maintenance. Procedia CIRP 81:417–422 Almalki F, Alsamhi SH, Sahal R, Hassan J, Hawbani A, Rajput N, Saif A, Morgan J, Breslin J (2021a) Green IoT for eco-friendly and sustainable smart cities: future directions and opportunities. Mob Netw Appl 1–25 Almalki FA, Soufiene BO, Alsamhi SH, Sakli H (2021b) A low-cost platform for environmental smart farming monitoring system based on IoT and UAVs. Sustainability 13(11):5908 Alsamhi SH, Lee B (2020) Blockchain-empowered multi-robot collaboration to fight covid-19 and future pandemics. LEEE Access 9:44173–44197 Alsamhi S, Rajput N (2012) Methodology for coexistence of high altitude platform ground stations and radio relay stations with reduced interference. Int J Sci Eng Res 3:1–7 Alsamhi SH, Ansari MS, Rajput NS (2018c) Disaster coverage predication for the emerging tethered balloon technology: capability for preparedness, detection, mitigation, and response. Disaster Med Pub Health Prep 12(2):222–231 Alsamhi SH, Rajput N (2014) HAP antenna radiation pattern for providing coverage and service characteristics. In: International conference on advances in computing, communications and informatics (ICACCI) Alsamhi S, Ma O, Ansari M (2018a) Artificial intelligence-based techniques for emerging robotics communication: a survey and future perspectives. arXiv preprint arXiv:1804.09671 Alsamhi S, Ma O, Ansari M (2018b) Predictive estimation of the optimal signal strength from unmanned aerial vehicle over internet of things using ANN. arXiv preprint arXiv:1805.07614 Alsamhi S, Ansari MS, Zhao L, Van SN, Gupta S, Alammari AA, Saber AH, Hebah MY, Alasali M A A, Aljabali HM (2019a) Tethered balloon technology for green communication in smart cities and healthy environment. In: First international conference of intelligent computing and engineering (ICOICE) Alsamhi SH, Almalki FA, Ma O, Ansari MS, Angelides MC (2019b) Performance optimization of tethered balloon technology for public safety and emergency communications. Telecommun Syst 1–10 Alsamhi SH, Almalki FA, Afghah F, Hawbani A, Shvetsov AV, Lee B, Song H (2021a) Drones’ edge intelligence over smart environments in b5g: blockchain and federated learning synergy. IEEE Trans Green Commun Netw Alsamhi SH, Lee B, Guizani M, Kumar N, Qiao Y, Liu X (2021b) Blockchain for decentralized multi-drone to combat COVID-19 and future pandemics: framework and proposed solutions. Trans Emerg Telecommun Technol e4255 Alsamhi S, Almalki F, Al-Dois H, Ben Othman S, Hassan J, Hawbani A, Sahal R, Lee B, Saleh H (2021c) Machine learning for smart environments in B5G networks: connectivity and QoS. Comput Intell Neurosci

Digital Twins Collaboration in Industrial Manufacturing

71

Alsamhi SH, Almalki F, Ma O, Ansari MS, Lee B (2021d) Predictive estimation of optimal signal strength from drones over IoT frameworks in smart cities. IEEE Trans Mob Comput Alsamhi SH, Almalki FA, AL-Dois H, Shvetsov AV, Ansari MS, Hawbani A, Gupta SK, Lee B (2021e) Multi-drone edge intelligence and SAR smart wearable devices for emergency communication. Wirel Commun Mobile Comput Altun C, Tavli B (2019) Social internet of digital twins via distributed ledger technologies: application of predictive maintenance. In: 27th Telecommunications Forum (TELFOR), pp 1–4 Ansari MS, Alsamhi SH, Qiao Y, Ye Y, Lee B (2020) Security of distributed intelligence in edge computing: threats and countermeasures. In: The cloud-to-thing continuum. Palgrave Macmillan, Cham, pp 95–122 Benkamoun N, ElMaraghy W, Huyet A, Kouiss K (2014) Architecture framework for manufacturing system design. Procedia CIRP 17:88–93 Borodulin K, Radchenko G, Shestakov A, Sokolinsky L, Tchernykh A, Prodan R (2017) Towards digital twins cloud platform: microservices and computational workflows to rule a smart factory. In: Proceedings of the10th international conference on utility and cloud computing, pp 209–210 Brovkova M, Molodtsov, Bushuev V (2021) Implementation specifics and application potential of digital twins of technological systems. Int J Adv Manuf Technol 1–8 Catarci T, Firmani D, Leotta F, Mandreoli F, Mecella M, Sapio F (2019) A conceptual architecture and model for smart manufacturing relying on service-based digital twins. In: 2019 IEEE international conference on web services (ICWS), pp 229–236 Fei T, Jiangfeng C, Qinglin Q, Zhang M, Zhang H, Fangyuan S (2018) Digital twin-driven product design, manufacturing and service with big data. The Inter-Natl J Adv Manuf Technol 94:3563– 3576 Fuller A, Fan Z, Day C, Barlow C (2020) Digital twin: enabling technologies, challenges and open research. IEEE Access 8:108952–108971 Gupta A, Sundhan S, Gupta SK, Alsamhi S, Rashid M (2020) Collaboration of UAV and HetNet for better QoS: a comparative study. Int J Veh Inf Commun Syst 5(3):309–333 Haag S, Anderl R (2018) Digital twin–proof of concept. Manuf Lett. 15:64–66 Hasan H, Salah K, Jayaraman R, Omar M, Yaqoob I, Pesic S, Taylor T, Boscovic D (2020) A blockchain-based approach for the creation of digital twins. IEEE Access 8:34113–34126 Hu L, Nguyen N, Tao W, Leu M, Liu X, Shahriar M, Al Sunny S (2018) Modeling of cloud-based digital twins for smart manufacturing with MT connect. Procedia Manuf 26:1193–1203 Jabeur N, Al-Belushi T, Mbarki M, Gharrad H (2017) Toward leveraging smart logistics collaboration with a multi-agent system based solution. Procedia Comput Sci 109:672–679 Kans M, Galar D, Thaduri A (2016) Maintenance 4.0 in railway transportation industry. In: Proceedings of the 10th world congress on engineering asset management (WCEAM 2015), pp 317–331 Kapteyn M, Knezevic D, Willcox K (2020) Toward predictive digital twins via component-based reduced-order models and interpretable machine learning. In: AIAA Scitech 2020 Forum, p 0418 Kritzinger W, Karner M, Traar G, Henjes J, Sihn W (2018) Digital twin in manufacturing: a categorical literature review and classification. IFAC-PapersOnLine 51:1016–1022 Lee D, Lee S, Masoud N, Krishnan M, Li V (2021) Integrated digital twin and blockchain framework to support accountable information sharing in construction projects. Autom Constr 127:103688 Liao S, Wu J, Bashir A, Yang W, Li J, Tariq U (2021) Digital twin consensus for blockchain-enabled intelligent transportation systems in smart cities. IEEE Trans Intell Transp Syst Lohtander M, Ahonen N, Lanz M, Ratava J, Kaakkunen J (2018) Micro manufacturing unit and the corresponding 3D-model for the digital twin. Procedia Manuf 25:55–61 Lu Y, Liu C, Kevin I, Wang K, Huang H, Xu X (2020) Digital Twin-driven smart manufacturing: connotation, reference model, applications and research issues. Rob Comput-Integr Manuf 61:101837 Madni A, Madni C, Lucero S (2019) Leveraging digital twin technology in model-based systems engineering. Systems 7:7

72

R. Sahal et al.

Perno M, Hvam L, Haug A (2022) Implementation of digital twins in the process industry: a systematic literature review of enablers and barriers. Comput Ind 134:103558 Qi Q, Tao F (2018) Digital twin and big data towards smart manufacturing and industry 4.0: 360 degree comparison. IEEE Access 6:3585–3593 Rasheed A, San O, Kvamsdal T (2020) Digital twin: values, challenges and enablers from a modeling perspective. IEEE Access 8:21980–22012 Rosen R, Von Wichert G, Lo G, Bettenhausen K (2015) About the importance of autonomy and digital twins for the future of manufacturing. IFAC-PapersOnLine 48:567–572 Sahal R, Alsamhi S, Brown K, O’Shea D, Alouffi B (2022) Blockchain-based digital twins collaboration for smart pandemic alerting: decentralized covid-19 pandemic alerting use case. Comput Intell Neurosci 2022:7786441 Sahal R, Breslin J, Ali M (2020) Big data and stream processing plat-forms for Industry 4.0 requirements mapping for a predictive maintenance use case. J Manuf Syst 54:138–151 https://www.sci encedirect.com/science/article/pii/S0278612519300937 Sahal R, Alsamhi S, Breslin J, Brown K, Ali M (2021a) Digital twins collabo-ration for automatic erratic operational data detection in industry 4.0. Appl Sci 11 https://www.mdpi.com/2076-3417/ 11/7/3186 Sahal R, Alsamhi S, Brown K, O’Shea D, McCarthy C, Guizani M (2021b) Blockchain-empowered digital twins collaboration: smart transportation use case. Machines 9. https://www.mdpi.com/ 2075-1702/9/9/193 Saif A, Dimyati K, Noordin KA, Alsamhi SH, Hawbani A (2021a) Multi-UAV and SAR collaboration model for disaster management in B5G networks. Internet Technol Lett e310 Saif A, Dimyati K, Noordin KA, Shah NSM, Alsamhi S, Abdullah Q (2021b) Energy-efficient tethered UAV deployment in B5G for smart environments and disaster recovery. In: 1st International conference on emerging smart technologies and applications (eSmarTA) Saif A, Dimyati KB, Noordin KAB, Shah NSM, Alsamhi S, Abdullah Q, Farah N (2021c) Distributed clustering for user devices under unmanned aerial vehicle coverage area during disaster recovery. arXiv preprint arXiv:2103.07931. Song Z, Canedo A (2020) Digital twins for energy efficient asset maintenance (Google Patents, 2020) US Patent 10,762,475 Vathoopan M, Johny M, Zoitl A, Knoll A (2018) Modular fault ascription and corrective maintenance using a digital twin. IFAC-PapersOnLine 51:1041–1046

Social Media Perspectives on Digital Twins and the Digital Twins Maturity Model Jim Scheibmeir and Yashwant Malaiya

Abstract Digital twins are virtual representations of their physical counterparts. Digital twins help us model, monitor, and predict physical things. The digital twin concept and implementations are frequently discussed on social media platforms. This chapter offers an analysis of the varying conversations of Digital Twins on social media, specifically the Twitter platform. Social media offers a platform for sharing information that can be analyzed to extract valuable information. Social media records can be analyzed to evaluate the velocity, volume, and variety of data related to a specific topic. Industry mentions, use cases, and sentiment of the associated topics and network graphs are introduced as well as supporting background information. The analysis reviews over 24,000 tweets collected between September of 2019 and July of 2021. We have identified the most mentioned industries with interest in Digital Twins. Among identified trending topics, the top three include the Internet of Things, artificial intelligence, and industrial uses. A maturity model for digital twins is introduced, informed by the identified trends and their popularity. The significance of the findings is discussed.

1 Defining Digital Twins The digital twin concept has many definitions and contributing authors. Jones et al. (2020) attribute Michael Grieves, along with John Vickers, with the origination of the concept. According to Rosen et al. (2015), the digital twin concept’s roots come from NASA’s Apollo program, twinning a spacecraft for training and mission support purposes. The term “digital twin” was coined by Shafto et al. in 2010 (Shafto et al. 2010). Grieves describes a digital twin as consisting of a physical asset, its virtual representation, and a two-way connection (Grieves 2014). Grieves is a commonly cited author and includes in his definition of a digital twin the existence of a bidirectional virtual to physical connection. The CIRP encyclopedia definition does not include a virtual to physical connection in its description. Tao et al. extend the J. Scheibmeir (B) · Y. Malaiya Department of Systems Engineering, Colorado State University, Fort Collins, USA e-mail: [email protected] © The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 Z. Lv and E. Fersman (eds.), Digital Twins: Basics and Applications, https://doi.org/10.1007/978-3-031-11401-4_8

73

74

J. Scheibmeir and Y. Malaiya

original three-component model by Grieves, the physical, virtual, and bi-directional connection between them, to one having five dimensions: the physical environment (PE), virtual environment (VE), services of both, the data of the digital twin, and the connection (Tao et al. 2018). Eckhart & Ekelhart do not define digital twins as having control over their physical counterparts (Eckhart and Ekelhart 2019). Rather they focus the definition and capability of the digital twin toward monitoring, visualization, and prediction. In their research of identifying definitions of the digital twin, Negria et al. (2017) found that the digital twin’s definition has varied and diverged away from solely modeling a physical system. Implementation definitions also range from digital twins being an augmented reality (AR) application to machine learning models (Schroeder et al. 2016). The amount and timeliness of integration that is required for a virtual instance to be considered as a digital twin have not been agreed upon. Eckhart & Ekelhart do not specify that a digital twin should secure the physical counterpart unless that is a part of the optimization. Digital twins can aid in the security of the physical counterpart using different access models and malicious activity identification techniques (Scheibmeir and Malaiya 2020). Eckhart & Ekelhart have suggested characterizing the digital twin concept based upon the level of data flow, integration, and autonomy. Other characterizations toward defining a digital twin include simulation, assisting with a physical system’s operational health, and optimizing a system process (Haag and Anderl 2018; Glaessgen and Stargel 2012; Uhlemann et al 2017). The common difference among digital twin definitions is whether the digital twin should control its physical counterpart. Digital twins increase the digital touchpoints of a cyber-physical system (CPS) and offer hackers knowledge of system integrations (Hearn and Rix 2019). Many digital twin integrations are with devices commonly referred to as the Internet of Things (IoT) technology. IoT has improved the management of homes, businesses, industries, and public sectors (Girma 2018). The information security concerns of IoT range from authorization, authentication, privacy, and access control of embedded systems. In general, IoT technology has produced a new cyber-attack surface (Atalay and Angin 2020). A study on consumer IoT, smart speakers, identified enjoyment (34.24%) as the most influential reason for the adoption of the devices (Arpnikanondt et al. 2020). While IoT consumer devices may offer enjoyment, they must also be secured. While enjoyment is a major contributing factor to IoT adoption among consumers, miniaturization, and technology price decline have attracted Industry 4.0. Industry 4.0 is the convergence of modern manufacturing and modern computing. Smart factories are building smart devices. If, or when, a smart factory is exploited, the supply chain of smart devices may generate exponential security concerns. To mitigate new threat vectors, a multi-model of security access controls can help the digital twins secure their physical counterparts. Multiple security models within the digital twin act as filters that trap malicious behavior before the physical assets executing the instruction. Control instructions, current, and predicted future states can be compared across the physical and virtual systems. Discrepancies can imply an inaccurate digital twin or indicate malicious acts. However, codified rules and advanced analysis techniques within system operations will not be enough to deter

Social Media Perspectives on Digital Twins and the Digital …

75

and prevent all risks and exploits. Security must start with the organization’s culture in a bottom-up approach (people, processes, and system inception to retirement). Social and cultural issues and complexities exist in the implementation of digital twins. Frequently, this is related to the types of data being collected, stored, and exposed by the digital twin. Digital twins have been used by device producers to understand how product use differs across cultures and locations. Even a local sports team’s home game schedule can be a factor in modeling and predicting factory production. Cybersecurity is not the only concern when implementing digital twins. Current standards and architectures for IoT, a technology that informs a digital twin, do not solve their interoperability problems (Novo O and Di Francesco 2020). Organizations contributing to IoT standards include the World Wide Web Consortium (W3C), the Internet Engineering Task Force (IETF), the Internet Research Task Force (IRTF), OneM2M, and the ETSI Industry Specification Group for cross-cutting Context Information Management (ETSI ISG CIM). Progress in IoT standardization includes these example services and protocols: • A Thing Description is a file containing semantic metadata about an IoT thing including its properties and behaviors • Resource Directories are repositories of things and their network identification • Constrained Application Protocol (CoAP) offers device communication over UDP and other transports. A CoAP datagram is illustrated in Fig. 1. The first two bytes of a CoAP datagram indicate the version of the protocol. Version is followed by two bits indicating the type of the message. The type of message could include confirmable, which requires acknowledgment of receipt. Acknowledgment becomes another message type. The token length field indicates the length of the upcoming token field. The value of the token field is used to connect request messages to their responses. The message identification field can be used to identify duplicate messages as well as to match an acknowledgment to a confirmable message. After message options, a byte of all one’s indicates the start of the message payload. CoAP messages are asynchronous and use unreliable transports such as UDP but offer mitigating features such as retransmission of confirmable messages (Shelby et al. 2014).

Fig. 1 The message format of constrained application protocol

76

J. Scheibmeir and Y. Malaiya

Standards and interoperability among devices are important because a digital twin may have a lifecycle many decades long. During such a span of operations, many IoT devices that are informing the digital twin may be swapped in and out due to failure, enhancement, or upgrade. The lifecycle of a digital twin requires affordable and feasible interoperability of IoT devices. IoT devices should be reusable, discoverable, and adaptable. These attributes of IoT devices help a digital twin to become maintainable. To evaluate performance and scalability, tools such as CoAPBench may be utilized to evaluate implementations. The CoAPBench employs virtual clients that simulate IoT device registrations. CoAPBench can scale many concurrent clients while measuring response times from the management layers of an IoT and digital twin system. For a digital twin to achieve the characteristic of fidelity, or sameness to its physical counterpart, many IoT devices will be integrated and informing the digital twin solution. Non-functional characteristics, such as the performance and maintainability of the system will be critical in the management of the digital twin over an extended lifespan. Characteristics such as reuse and discoverability of IoT endpoints will help accelerate the maintenance and enhancements of digital twins over their lifespan. A development model and methodology for using APIs for digital twins have been put forward (Scheibmeir and Malaiya 2019). The development model begins with an objective tree and contextual diagram to cover the environment, relationships, and operations of the physical entity. The development of a digital twin must encompass the functionality of the physical counterpart, supporting and foundational data sources and integrations, as well as the context of the operating environment and culture. Using context diagrams and objectives trees are methods to explore and define the needs of a digital twin. Test-driven development was suggested as a practice for implementing APIs in a test-first approach. Utilizing OpenAPI specification aids design and test documentation and supports reuse. Traditional software development lifecycles place testing the system after its development. A better approach is to “shift left” and test during the design and development through practices such as Test-driven development and Behavior-driven development. These practices focus on the creation of unit tests and UI tests before any code being implemented. With these practices, testing comes before code development work and thus “shifts left” in a traditional development cycle. Performance engineering for digital twins must be done early, such as testing individual parts or components of the API operations. Testing for performance concerns early in the development helps avoid expensive redesign efforts. Figure 2 is from the 2019 work of Scheibmeir and Malaiya and illustrates the use of contextual diagrams, objective trees, TDD, and many more practices in the development of APIs for digital twins. The model suggests API mediation but fails to extend into concerns for the user interface. Augmented reality has been suggested as an interface modality for digital twins.

Social Media Perspectives on Digital Twins and the Digital …

77

Fig. 2 A framework for developing APIs for digital twins

2 Use of Social Media Analytics in Research Social media data is common to many research investigations. Social media data is publicly available and offers velocity, variety, and volume of data. Researchers can extract valuable conclusions from social media due to its public nature and ease of access (Cruickshank and Carley 2020). Twitter data has enhanced biased survey populations and assisted in research by aiding in the latitude and longitude of where conversations take place (Martin et al. 2020). A study by Bougie et al. found that 23% of tweets by the software engineering groups they followed were toward software engineering topics (Bougie et al. 2011). Of that 23% of tweets, 62% were toward solving software engineering problems. Software engineering practitioners use social media platforms to learn about technology trends (Storey et al. 2010). They do not cite scientific research in their blogs (Williams 2018). Beyond trend identification, social media platforms offer links to web resources, networking people, and directing our attention (Büchi 2017). Searching for and accessing information are the leading factors among college students for accessing social media platforms (Gómez-García et al. 2020).

2.1 Social Media Analytics Methodology Utilizing R programs and the Twitter API, we have collected (not exhaustively) 24,275 tweets between August 2019 and July 2021. This is not a comprehensive

78

J. Scheibmeir and Y. Malaiya

collection of all tweets referring to digital twins. Our collection of tweets is limited by unpaid access to Twitter’s API and further constrained by daily limits and the R programs collecting tweets toward many different topics. While the analysis is limited, it informs on the public discourse about digital twins and our methodology will be discussed in enough detail to enable similar research for those who want to dig deeper in this area. A content-based analysis is utilized within this research to determine themes among the tweets. Themes may include technology trends or industries where digital twin technology is frequently discussed. Time series analysis indicates ebbs and flows of the discussions and helps identify when peaks or lulls in the discussions are occurring. Sentiment analysis provides a numerical approach to how positive or negative the meaning of a tweet’s language may be. Network graphs help identify relationships. This chapter will utilize network graphs to detect relationships between the industry discussions of digital twins and which technology trends are included and omitted from the discussion. When confronted with large amounts of free-form text, it may be useful to utilize clustering techniques to determine the distinct topics and conversations occurring. The cluster sizes are determined by the within-cluster sum of squares (WSS) and the average silhouette methods. A dendrogram is a data visualization object and a type of tree graphic. Dendrograms depict the closeness or sameness of objects after they have been clustered. These methods are useful when analyzing social media and other data sources and will be utilized throughout this chapter.

2.2 Time Series Analysis of Tweets About Digital Twins Twitter supplies a created date field that identifies when Twitter users posted their communication. The earliest tweet within our collection was posted on August 29th, 2019. The last tweet in our collection is dated July 31st, 2021. Figure 3 is a time series chart identifying the date the tweets from our data set were posted to the Twitter platform and the number of tweets per day. A smoothed line is positioned along the time series to indicate the overall trend in the volume of tweets. The chart identifies a peak in the discussions of collected tweets during January 2020. To determine the trends driving up this peak, we isolate by the posted date and identify tweets having the highest retweet counts. Retweets are a feature and behavior among Twitter users who can repost a tweet to propagate the message through their network. Within the January 2020 peak period of digital twin tweets, a tweet by Stephane Nappo was the most retweeted with eighty retweets (Nappo 2020). The tweet’s message is like many of the definitions reviewed earlier in this chapter and describes a digital twin as a virtual model that can bridge the physical and digital worlds. The image represents a virtual replication of a city. Smart cities are a popular form of digital twins. The tweet utilizes many hashtags, such as #AR

Social Media Perspectives on Digital Twins and the Digital …

79

Fig. 3 Time series chart of our collection of tweets referring to digital twins with a peak in January of 2020

(augmented reality), #IoT (Internet of Things), and #AI (artificial intelligence), that draw Twitter users’ attention and help gain more attention to the tweet based upon platform algorithms.

2.3 Unsupervised Clustering of the Digital Twin Tweets We utilize a document term matrix as input into an unsupervised cluster analysis. The document term matrix is a large object that contains an identifier of each tweet, the words used within the tweet’s text, and the frequency of the words. The clustering algorithm searches through the document term matrix and groups the tweets based upon patterns in the utilized words and their frequencies. To determine an appropriate number of groups, or clusters, to be created, we utilize the within-cluster sum of squares (WSS) and silhouette methods. There are other methods that can help with clustering and determining cluster sizes, such as DBSCAN, HDBSCAN, or gap statistic methods (Burkov 2019). The WSS method will iterate through generations of clustering incrementing the number of individual groups with each generation. During each iteration, the squared distance between all the observations within the cluster and its center are summed. This is done for all clusters and the total WSS is then compared with the other generations each having an increasing number of clusters. The ideal number of clusters is frequently determined visually, known as the “elbow method.” The

80

J. Scheibmeir and Y. Malaiya

Fig. 4 Within-cluster sum of squares indicates that the proper number of clusters, identified as “elbows” in the line, maybe two, four, or six groups

“elbow” is visually identified when the WSS decreases rapidly in initial generations of smaller n number of clusters and the decrease flattens as n increases. The WSS output is plotted in Fig. 4 with a few potential “elbows” in the line occurring at two, four, and six clusters generated. The silhouette method also strives to find the proper number of clusters in a collection of data. The method is like the WSS method in that it will iterate through generations of cluster creation and compare each generation. The comparison is performed across the distance between observations in a cluster and observations in the neighboring cluster. If many clusters exist within a small dimension, observations will be near neighboring observations, and this may indicate that too many clusters have been generated for the dataset. We utilize R libraries of nbclust and factoextra to quickly implement the WSS and silhouette methods. The output of the silhouette method is found in Fig. 5 and identifies four clusters as the appropriate amount for our collection of digital twin tweets. Another helpful data visualization graphic when performing text analysis and hierarchical clustering is the dendrogram. Dendrograms are tree-based graphics that indicate relationships. Dendrograms are frequently created when observing the distance between observations in document term matrices and help visualize cluster distribution. The problem with dendrograms is that they do not scale well when the number of observations approaches many thousands. In these cases, the graphics become either quite large or very densely populated making discernment difficult. Because dendrogram diagrams do not scale well with large observations, we have cast only a

Social Media Perspectives on Digital Twins and the Digital …

81

Fig. 5 Average silhouette method indicates four clusters as the appropriate grouping size for our collection of tweets

sample of 1% of our 24,000 tweets. Dendrograms can be customized with specific visual formats such as the typical tree diagram and circular and in our case, we are utilizing the phylogenic shape. Phylogeny is the development of traits or taxonomic grouping. It can help discuss biology and the evolution of species. Here, we utilize a phylogenic dendrogram to illustrate the evolution of the conversations within the digital twin tweets, illustrated in Fig. 6. Trends were extracted from the four clusters by the frequency of mention. The largest cluster in the volume of tweets is the first group, magenta in the phylogenic dendrogram (only a sample of 1% of tweets were used to generate the graphic), and the top seven trends by mention come from this first cluster: • • • • • • •

the Internet of Things Artificial Intelligence industry use collaboration the virtual world novelty data.

The remaining three clusters then provide three other trends to round out the top ten: machine learning, blockchain, and augmented reality. The largest cluster of tweets is displayed by word cloud in Fig. 7, further illustrating many of the top trending concepts.

82

J. Scheibmeir and Y. Malaiya

Fig. 6 Phylogenic dendrograms can be created using distance calculations from document term matrices but dendrograms do not scale well with large numbers of observations

Fig. 7 Word cloud graphic of the most frequent terms from cluster one

Social Media Perspectives on Digital Twins and the Digital …

83

Tweets can be retweeted by users to further promote the message content. The most retweeted post from the first cluster of tweets references technology predictions by Global NTT (Rai 2019). Digital twins are one of the emerging technologies that the predictions include. The tweet’s embedded link is to an online article that summarizes the predictions and mentions that digital twins can collect data from instrumented assets, model behavior, identify patterns, and create more accurate conclusions (BW Online Bureau 2021). From the second cluster, the most retweeted post references the 44th episode of IoT Coffee Talk, an online webinar by Tiffany (2021). This tweet merges business concerns such as the conversion of manual, human-driven, or paper processes into optimized and automated processes via digital transformation. These technologies may improve the efficiency of industry and consumer behaviors to also solve sustainability concerns. Some of the hashtags in the tweet by Tiffany are like those within the most retweeted tweet of the first cluster, #AI, #AR, and #DigitalTwins (Rai 2019). However, Tiffany introduces additional trends in his tweet including 5G, edge, cloud, sustainability, and digital transformation (Tiffany 2021). The most retweeted post from the third cluster is again a reference to trends and predictions, this time the trends listed were identified by the research and advisory organization, Gartner. The tweet links to an article that identifies eight trends in three categories with two additional cross-cutting trends. The three categories include Intelligent, Digital, and Mesh. Digital twins are identified as the fourth 2019 technology trend by Gartner (Panetta 2021). The article by Panetta further states that digital twins have: • • • •

robustness in their modeling profile to support business outcomes link to physical assets to potentially model and control drive new business opportunities when big data analytics and AI are applied interaction to help evaluate future states such as modeling and simulation.

The most retweeted post in the fourth cluster offers some distinction from the previous three (RolSOuLi 2021). This tweet references an open-source distributed ledger system that is like standard blockchain but utilizes a different algorithm requiring less energy (Ullah et al. 2021). Because IOTA can run on devices having less computational power and bandwidth, it enables the value and security of distributed ledger in the realm of IoT devices. The tweet mentions the tangle algorithm, which is used by the IOTA distributed ledger and could be utilized to secure digital twins by creating more trust in the IoT ecosystem.

2.4 Twitter Analysis by Industry Content-based analysis of the tweets has identified the mentions of specific industries. The International Labor Organization maintains a curated list of industries and descriptions (International Labor Organization 2021). This curated list can be utilized in a labeling algorithm to identify industry mentions within the tweets. The health

84

J. Scheibmeir and Y. Malaiya

Fig. 8 The health industry had the most mention within the collection of digital twin-related tweets

industry is the most mentioned within this collection of tweets, followed by entertainment and utilities. The textile industry was not mentioned within our collection of tweets, illustrated in Fig. 8. Considering the sentiment and emotions that are prevalent in the tweets is an interesting research angle. Sentiment analysis typically reviews content on a continuum of negative to positive. Our sentiment analysis will review the tweets by industry and for specific emotions that may be felt or influenced by the message of the tweets, including sentiments such as anticipation or fear among others. This analysis will utilize the NRC lexicon to label the tweet’s sentiment. The labels having the most tweets were the health and entertainment industries (shown in Fig. 8). It is more probable that a tweet using words that convey anticipation will reference the health industry (31.0%) compared to the entertainment industry (8.7%). The naïve Bayes algorithm was utilized to determine these probabilities. The formula for naïve Bayes is explained for our classification problem and data set in Eq. 1. The naive Bayes equation explained P(A|B) =

P(B|A) × P(A) P(B)

where P( A|B) Probability of industry mention given a specific sentiment P(B|A) Probability of sentiment given a specific industry in mentioned

(1)

Social Media Perspectives on Digital Twins and the Digital …

85

P( A) Count of industry observations divided by total number of observations P(B) Number of instances of a sentiment, divided by all observations. To calculate these probabilities, we may start with the numerator, which is the product of the probability of positive sentiment given a specific industry and the probability of a tweet having a relationship to the same specific industry. This example will focus on the sentiment of anticipation (it has many observations) and the health industry as it was identified as having the most labeled tweets. Table 1 identifies the sentiment factors across all industry classes and will inform our formulas for the health industry. In this training data set of the classification model, 780 tweets reference health and within those tweets, 157 have the sentiment of anticipation. The conditional probability of anticipation sentiment (B in Eq. 1) given an agricultural tweet (A in Eq. 1) is presented in Eq. 2: Conditional probability of positive sentiment given a tweet references the agriculture industry: P(Anticipation|Health) = 157 ÷ 780 P(Anticipation|Health) = 0.201

(2)

The conditional probability of anticipation sentiment given a tweet referencing health is 20.1%. To complete the numerator, we need the product of the P(B|A) and the a priori, or the number of health-related instances divided by the total number of size in the data set. Equation 3 determines the a priori. The a priori is the probability of a tweet referencing health and is found by dividing the count of health instances by the total training data set count. P(Health) = 780 ÷ 3411 P(Health) = 0.229

(3)

The numerator is divided by the probability of a tweet having the sentiment of anticipation. This is determined by dividing the number of anticipation instances for all classes (507) by the total amount of training data instances (3411). Probability of a tweet having the sentiment of anticipation is determined in Eq. 4. The probability of the sentiment anticipation occurring in the data set P(Anticipation) = 507 ÷ 3411 P(Anticipation) = 0.149

(4)

The posterior or the probability of a tweet referencing the health industry if we know that the message contains the sentiment of anticipation can be determined using Eq. 5. The probability a tweet’s message is referring to the health industry given the sentiment includes anticipation

2

Energy

6

33

2

10

18

Food

Forestry

Health

Public

1

3

Mining

23

4

4

2

1

Metal

Postal/Telecom

4

5

Mechanical

Media

2

Hotel

1

22

Financial

157

44

6

Entertainment

9

4

4

1

1

7

5

2

137

10

7

50

2

2

7

3

Education 1

25

7

Construction 2

2

20

Commerce

18

1

1

1

Fear

1

Disgust

13

Anticipation

Automotive

Anger

Agriculture

Industry

Table 1 Counts of sentiments by classes of industry

7

8

7

5

7

3

14

30

4

17

38

21

11

11

6

1

11

Joy

5

4

1

3

1

11

7

1

4

45

6

2

11

5

Negative

49

30

47

14

6

22

5

255

120

8

80

137

75

62

165

46

8

42

Positive

1

19

3

2

1

1

Sadness

2

7

2

2

2

3

9

11

43

1

4

3

Surprise

8

8

21

1

1

10

3

162

64

4

33

3

29

16

56

19

8

8

Trust

102

57

93

28

11

50

15

780

290

25

181

360

148

116

289

101

20

75

Totals

86 J. Scheibmeir and Y. Malaiya

Social Media Perspectives on Digital Twins and the Digital …

P(Health|Anticipation) = 0.310 =

0.201 ∗ 0.229 0.149

87

(5)

These equations could be performed a second time with the numerator updated for the entertainment industry to prove the statement, given a tweet has the sentiment of anticipation, there is a greater probability that the tweet references the health industry (31.0%) compared to the entertainment industry (8.7%). Given a tweet references the food industry, there is an 8.0% probability that the sentiment of the tweet will be anger. The probability of the sentiment of trust occurring is highest for the industry of automotive; however, automotive referencing tweets only convey trust with a probability of 40.0%. The sentiment of disgust is rarely found in the tweet messages, and the highest probability of disgust was found in messages labeled toward the industry of forestry (1.4%). The R library e1071 offers a naïve Bayes function that eases the implementation of the algorithm. Unfortunately, given the quantity of data we have, and the factors supplied to the model, we only achieve an accuracy of 25.5%. To increase the accuracy of this model, first, increase the number of tweets in the collection and, second, improve the factor selection beyond only utilizing the factor of sentiment. Network graphs visually identify relationships. Within the analyzed conversations, not all industry-related tweets reference the top trends. Tweets that reference the food or hotel industries have very little relationship to trends. This is visible in Fig. 9, a network graph where the industry nodes are yellow, the trend nodes are green, and the relationships between these labels are red lines. The construction industry tweets are the most inclusive to top trends.

3 Background on Maturity Models While we have noted the many definitions of the digital twin, determined popular industries in the public discussion, and uncovered the sentiment in the conversations, we have not uncovered what a good digital twin is. To help organizations determine the level of value, and to further improve and enhance their development process, we suggest a digital twin maturity model. Maturity models help organizations achieve capability and capacity within a discipline or process (Mittal et al. 2018). To increase the capability or capacity, an organization first places itself along a trajectory that is determined by current performance (De Jesus and Lima 2020). Achieving a state of greater capability along the same course becomes the goal. A maturity model establishes the milestones of capability and the distance between current and goal states. Assessments of maturity inform organizations and their leadership teams about their current capability and readiness. Organizations frequently utilize a questionnaire to place their competencies or system capabilities along the path of the maturity model. These can be self-assessments or utilize consultants. The questions and the

88

J. Scheibmeir and Y. Malaiya

Fig. 9 This network graph illustrates the relationships found in tweets between industries and trends

maturity model assessment effort evaluate Key Performance Indicators (KPIs) to position the organization and system capabilities. Organizations have two options when requiring a maturity model. The first option is to apply a generic model, and the second option is to build a specific and contextual model to a problem domain. To build a specific model, five factors must be considered: context, conceptual characteristics, interaction with experts, the use of surveys, and qualitative research.

4 The Digital Twin Maturity Model The creation of a maturity model for digital twins requires defining the benefit that would come from using the model. Kluth et al. describe a maturity model as a representation to evaluate business processes (Kluth et al. 2014). Kohlegger et al.

Social Media Perspectives on Digital Twins and the Digital …

89

describe a maturity model as one that represents distinct stages of increasing capability (Kohlegger et al. 2009). A maturity model for digital twins is a tool and associated process to measure increasing and distinct milestones of value derived from a digital twin by its capabilities. After defining the benefit, the next step in creating a maturity model is to determine the characteristics and parameters of digital twins for distinction along a path of increasing system capability or capacity. Some foundational and general parameters have been established; governance, supportive technology, connectivity, value generation, and competence of the organization (Colli et al. 2018). Dimensions of maturity models frequently include high-level concerns of people/culture which includes the skills, organizational structures, and processes, as well as technology (Cognet et al. 2020). A digital twin maturity model can be informed by existing models for Industry 4.0. Industry 4.0 describes the integration of people, objects, and equipment to allow flexibility and autonomous decision-making in manufacturing (Agostini and Filippini 2019). Industry 4.0 can be described as a transformation from predominantly mechanical to predominantly digital manufacturing (Oztemel and Gursev 2018). Given a digital twin is of a factory, it would be integral to Industry 4.0. Digital manufacturing is aided by these technologies and principles (Bakkari and Khatory 2017): • Changeability—The manufacturing equipment and product will evolve, requiring a capacity for change. • Decentralized Decisions—Smart factory systems are composed of smart machines. Smart describes the optimal condition of equipment making decisions autonomously. Although autonomous, smart systems may be informed by centralized data sources, control units, or human workers. • Interoperability—When such change to the system or environment occurs, components will require updating and enhancement to support the adaptation. Thus, interoperability of equipment will be a necessity. • Real-time Reaction—Based upon the capacity for decentralized decisions and guided by IoT such as sensors and actuators, smart components can make corrections in real time. • Simulation—IoT devices such as sensors and actuators can be emulated so that entire behaviors of smart systems become virtualized. Other technology trends and principles that are key to Industry 4.0 include big data, cloud, additive manufacturing, AR, robotics, and machine–machine–human integration (Crnjac et al. 2017). The IoT technology is foundational for these mechanisms. IoT provides big data, may include additive manufacturing and robotic instrumentation, and can inform both machines and humans in the loop. A digital twin of an Industry 4.0 plant is the composition of these mechanisms for modeling, monitoring, simulating, and securing the physical plant relative to its environment. One successful digital twin implementation will not simply be copied by other organizations. However, a maturity model can help guide the capabilities and improvements of a digital twin along a path to implement these mechanisms.

90

J. Scheibmeir and Y. Malaiya

There are many different maturity models. Some of the models are well known and not specific to digital twins, such as the CMMi model. Other models have a higher correlation to digital twins based upon their focus on digitalization, such as the SMSRA and M2DDM. Further models have been created by technology and consulting companies that offer solutions or expertise. Those models include examples such as Rockwell Automation, Price Waterhouse Coopers, and Siemens. Data is a core component of many models, including the Maturity Model for Data-Driven Manufacturing (M2DDM). These models and others are listed in Table 2. A good maturity model removes confusion by isolating the factors and priorities that will help an organization achieve the next level of capability. Parente and Federo suggest removing conjunction, equifinality, and asymmetry for causality in models to be clear to organizations (Parente and Federo 2019). Asymmetry is a characteristic of causality that may explain one result and then fails to explain another result. Asymmetry can create doubt in the accuracy of a model. Equifinality implies that similar benefits and capabilities may be the outcome of more than one level of maturity. When equifinality exists in models, organizations will cease to increase the risk or cost in implementation as the value may not increase. A conjunction is a relationship between technologies, processes, or culture that holds back value creation until all related tenets increase in maturity together. If such related conditions are spread across maturity levels, intermediary benefits offered at lower levels would not become actual value until much higher levels of maturity are achieved. Maturity models should not suffer conjunction, equifinality, or asymmetry. The ERP 4.0 maturity model by Basl and Novakova (2019) has six levels across dimensions of business model, technology, data, and processes. To construct the model, Basl et al. analyzed trends from survey data and layered the trends into the maturity model levels based upon their frequency found from the survey. The survey was completed by 26 ERP system suppliers (Novakova 2019). Trends having the most frequency of being acknowledged by the system suppliers were positioned higher into the levels of the maturity model. The trends were identified through the survey included cloud, IoT, blockchain, digital twins, edge computing, AI, big data, social networks, and AR/VR. These trends are very similar to those identified through social media analytics and are illustrated as green network nodes in Fig. 9. The most frequent trends include cloud, IoT, and AR. Trends with lesser frequency include extending asset life, optimizing performance, and implementing blockchain. Other trends included big data, mobile ERP apps, and in-memory computing (IMC). A segment of Basl et al. ERP 4.0 maturity model is illustrated in Table 3. The digital twin maturity model has been informed according to the guidance by de Jesus and Lima of using context, characteristics, expertise, survey (social media analysis), and qualitative research. Academic research, commercial solution and providers’ models, and social media analytics were input factors for the creation of the digital twin maturity model. Basl’s method utilized in the ERP 4.0 maturity model creation uses trend popularity to determine the maturity levels. While not the final version of our model, the approach by Basl does offer insight into public opinion and the volume of driving trends (as illustrated in Fig. 10).

Social Media Perspectives on Digital Twins and the Digital …

91

Table 2 Example maturity models and descriptions Example models

Short description

Maturity model for data-driven manufacturing (M2DDM)

The Maturity Model for Data-Driven Manufacturing (M2DDM) contains six levels of maturity (begins at level 0). The 4th level is digital twin and characterized by smart systems, decentralized decisions, and centralized intelligence to keep humans in the loop. The 5th, and highest level, is the self-optimizing factory (Weber et al. 2017)

Smart manufacturing systems readiness assessment (SMSRA)

The Smart Manufacturing Systems Readiness Assessment (SMSRA) provides manufacturing organizations with an indication of their current factory state compared against a reference model of capabilities. The last stage is transformed implying the business has executed a change to its business model (Jung et al. 2016)

Complexity management maturity

The first level of the Complexity Management Maturity is initial and represents that an organization has not yet quantified the amount of complexity at hand

C3M

The C3M model presents five levels of maturity for IT-based case management systems (CSM) across three phases of CSM adoption; pre-CSM, CSM, post-CSM. C3M is novel as it presents levels of capabilities and the risks that may be associated with the levels of benefits (Koehler et al. 2012)

Capability maturity model integration (CMMI)

Capability Maturity Model (CMM) was constructed in 1986 and updated in 2006 to include tech and process as the CMMI model. CMMI includes the phases of initial, repeatable, defined, managed, and optimizing

Test Maturity Model Integration (TMMi)

TMMi utilizes the same structure as CMMi and helps organization gauge and improve their software testing practices (TMMi Foundation 2020)

Industry 4.0/digital operations self-assessment

PWC’s self-assessment places an organization’s Industry 4.0 capability concerning their target state and offers them a benchmark to the positions of industry competition. Cognet et al. compared the PwC and IMPULS models and found that the IMPULS model has 84% coverage of the PwC digital maturity model’s KPIs

The connected enterprise maturity model

Created by Rockwell Automation, this five-stage maturity model offers best practices for modernizing culture and technologies when networking operational technology (OT) and information technology (Parkinson 2015) (continued)

92

J. Scheibmeir and Y. Malaiya

Table 2 (continued) Example models

Short description

Digitization roadmap

The digitization roadmap by Siemens is constructed to help organizations transform their business. Six areas, such as process, security, and collaboration, are reviewed and benchmarked and an associated ROI study is completed to evaluate financial consequences of improvement activities (Siemens 2020)

Table 3 A subset of Basl’s ERP 4.0 maturity model Level

Description and inclusion

0

Traditional RDBMS system, with basic ERP process automation, and no cloud adoption

1

Mobility, additional automation, and digitization of processes

2

The complexity, digital capabilities, and analysis all increasing

3

Initial migration to cloud services, business intelligence efforts underway, continued increase in process automation

4

As-a-service implementations, IoT integration, digital twin capabilities

5

AI, RPA, all cloud deployment, all business processes automated

Trends identified from the Industry 4.0 models and academic research, such as decentralized and interoperability, have little mention within our collection of tweets. Collaboration has many more references compared to the two least mentioned topics. Another jump exists between the trends of changeability and predictability; however, the topics of fidelity and autonomy retain the most conversation found in the collection of tweets. The digital twin maturity model has been constructed based upon the characteristics found in literature review, social media analytics, and based upon input from existing maturity models. The digital twin maturity model is composed of six levels: initial, managed, integrated, immersive, autonomous, and ubiquitous (illustrated in Fig. 11). The lowest maturity level is the initial digital twin. The initial digital twin is limited in scope, such as only instrumenting a few parts and components. The initial digital twin offers limited insight and is far from being complete, integrated, or even secured. The second level of maturity is the managed level. At the managed level, the twin has a prioritized roadmap and coverage beyond ad hoc parts and includes parity with system components. The third level of maturity increases the digital twin’s capability via integration and interoperability. At this third level of integration, the digital twin can model, monitor, and predict many of the physical subsystems. The fourth level of maturity, immersive, is mostly defined by its human interface. At the fourth level of maturity, the digital twin is assessable using immersive interfaces, such as augmented or virtual reality. The autonomous level of the digital twin has

Social Media Perspectives on Digital Twins and the Digital …

Fig. 10 Amount of maturity level/trend mention in our social media analysis

Fig. 11 The six levels of the digital twin maturity model

93

94

J. Scheibmeir and Y. Malaiya

the cybersecurity, integration, and authority, among other characteristics, to selfoptimize its physical counterpart. Finally, the sixth level of maturity of a digital twin is when it consumes the context of its environment. This is the ubiquitous level. This level requires investment and technology that will be beyond the scope of most organizations. Achieving the ubiquitous level requires instrumenting the physical world, beyond the immediate assets, to understand global weather patterns, political, social, and economic phenomena, as well as other growing concerns. Table 4 provides the capabilities and their descriptions. Referring to Parente and Federo’s guidance for models to be effective for organizations, we test our digital twin maturity model for conjunction, equifinality, and asymmetry. Conjunction in a maturity model exists when the benefit promised by achieving a lower level, such as the integrated digital twin, cannot be reaped until the digital twin reaches an advanced level, such as immersive. In our model, for example, value is delivered to an organization at the integrated digital twin level, as that level of maturity allows the twin to grow from modeling individual parts or components into modeling entire subsystems. Furthermore, value is achieved at the integrated level through engaging with users with wearable technology, such as understanding the physical location of system operators for safety reasons. It is clear then that value arrives at the integrated level without requiring the immersive level to have been met. Table 4 The six levels of capability and a short description of their enablement Capability

Description of enablement

Initial

At this level of maturity, the digital twin can model a selection of parts or a few components of the system. The digital twin can inform human operators and offers a viewpoint toward collaboration. It is far from a smart or autonomous capability

Managed

Digital twins increase the cybersecurity risk footprint by increasing integration touchpoints and consuming data in transit, storage, and processing. A managed digital twin is measured for its ability to secure itself and the physical asset. The managed digital twin has moved beyond ad hoc instrumentation of parts into a prioritized roadmap that incorporates cybersecurity concerns

Integrated

A complex system is composed of many systems and subsystems. At this level, the digital twin incorporates all targeted data sources into a unified virtual instance of the physical counterpart

Immersive

A digital twin at this level offers a modern and immersive interface with AR or VR capabilities. Beyond monitoring the components, the immersive interface may offer simulated experiences of the components

Autonomous

Once the digital twin is integrated, informed, and secured, it may become smart or optimize without decisions from a human control interface

Ubiquitous

Complex systems operate within the context of their environment. A ubiquitous digital twin of a physical asset would integrate with a digital twin of the physical world, such as climate models. This level of maturity requires investment and integrations that organizations will scope out of their implementation for years to come

Social Media Perspectives on Digital Twins and the Digital …

95

Similarly, asymmetry would damage the trust in the digital twin model when a characteristic fails to explain its importance in each of its succeeded levels. For example, the integrated digital provides an API that can be consumed by a headset offering an immersive experience. The integrated digital twin also provides interfaces to the sensors and actuators that will be utilized to digitally annotate the physical world through augmented reality. Furthermore, an integrated digital twin is required for the autonomous digital twin to exist. The autonomous digital twin requires integrations to the many parts, components, and subsystems to control and optimize the physical asset. Even the autonomous digital twin requires integration to the digital twin of the physical world. If we moved a step down in maturity, down from the integrated digital twin to the managed, all the previous features and benefits would exist in a product roadmap but not in the implementation. The managed digital twin is more than a simple roadmap and vision, it offers an implementation whose limited existence is now counted (managed and measured) so that vulnerabilities, risk, and remediation are a part of the planning and implementation. Without applying cybersecurity early into the maturity model, future benefits would have a greater risk. Any future maturity state beyond the initial digital twin will always offer the original benefit of the digital point of view into a limited part or component. The last area to defend the digital twin maturity model includes the characteristic of equifinality. Equifinality implies that similar benefits and capabilities may be the outcome of more than one level of maturity. If a digital twin were at the maturity level of initial, we would not want to allow the twin to become autonomous nor would the benefits of an autonomous system be reached at the initial level. The small scoped system could likely ruin many integrated parts and components, as it is not yet informed of the entire system’s states, such as whether dependencies are operating, within appropriate thresholds, failed, or shutdown. The initial twin would need to reach the integrated phase to have this knowledge and should not have widespread integrations with other systems without first safely being counted, measured, and secured in the managed level. The benefit at each phase of our model can be reached at that level, without delaying the benefit until future phases. It is important to note that while cybersecurity is a component of the managed level, cybersecurity must be addressed throughout later phases.

5 Conclusion and Future Work This chapter introduced findings from social media analytics on digital twins as well as a new maturity model. From the social media analytics, the top three trends identified included the IoT, AI, and industrial uses. An analysis into the industrial uses found the health industry as the most mentioned, followed by entertainment and utilities. The textile industry was not mentioned within the collection of tweets used in this research. Sentiment analysis was performed on the messages within the tweets and a comparative analysis was offered across industries. Given a tweet references the

96

J. Scheibmeir and Y. Malaiya

food industry, there is an 8.0% probability that the sentiment of the tweet will be anger. The probability of the sentiment of trust occurring is highest for the industry of automotive; however, automotive referencing tweets only convey trust with a probability of 40.0%. The sentiment of disgust is rarely found in the tweet messages, the highest probability of disgust was found in messages labeled toward the industry of forestry (1.4%). Given a tweet has the sentiment of anticipation, there is a greater probability that the tweet references the health industry (31.0%) compared to the entertainment industry (8.7%). Network graphs were utilized to visually identify relationships. Within the analyzed conversations, not all industry-related tweets referenced the top trends. Tweets that reference the food or hotel industries had very little relationship to top trends. The collection of tweets identifies a peak in the discussions during January 2020. The tweet having the most retweets was retweeted eighty times. That popular tweet’s message was like many of the academic definitions reviewed in this chapter, as a virtual model that can bridge the physical and digital worlds. To help organizations determine the level of value, to further improve, and to enhance their development process, we suggest a digital twin maturity model. The digital twin maturity model is composed of six levels: initial, managed, integrated, immersive, autonomous, and ubiquitous. The maturity model was discussed in terms of conjunction, equifinality, and asymmetry. These three characteristics should not exist in maturity models as they reduce trust in the accuracy and the value that maturity models offer. Future research should focus on case studies, implementing the maturity model, and further evaluating it for accurate causality of benefits achieved in the various phases.

References Agostini L, Filippini R (2019) Organizational and managerial challenges in the path toward Industry 4.0. Eur Innov Anag 22:406–421 Atalay M, Angin P (2020) A digital twins approach to smart grid security testing and standardization. In: IEEE International workshop on metrology for Industry 4.0 & IoT, pp 435–440. https://doi. org/10.1109/MetroInd4.0IoT48571.2020.9138264 Bakkari M, Khatory A (2017) Industry 4.0: strategy for more sustainable industrial development in SMEs. In: Proceedings of the IEOM 7th international conference on industrial engineering and operations management Basl J, Novakova M (2019) Analysis of selected ERP 4.0 features and proposal of an ERP 4.0 maturity model. In: Research and practical issues of enterprise information systems. Springer International Publishing, Cham, pp 3–11. Web Bougie G, Starke J, Storey M, German DM (2011) Towards understanding twitter use in software engineering: preliminary findings, ongoing challenges and future questions. In: Web2SE’11. ACM, pp 31–36. https://doi.org/10.1145/1984701.1984707 Büchi M (2017) Microblogging as an extension of science reporting. Public Underst Sci pp 953–968. https://doi.org/10.1177/0963662516657794 Burkov A (2019) The hundred-page machine learning book. Andriy Burkov

Social Media Perspectives on Digital Twins and the Digital …

97

BW Online Bureau (2021) ‘Future disrupted’ predictions for 2020: data, automation and IoT will enable virtual societies. Smart Cities, Nov-2019. [Online]. Available: http://bwsmartcities.bus inessworld.in/article/-Future-Disrupted-predictions-for-2020-Data-automation-and-IoT-willenable-virtual-societies-/13-11-2019-178854/. Accessed 18-Jul-2021 Cognet, B et al. (2020). Towards a novel comparison framework of digital maturity assessment models. Product lifecycle management in the digital twin era. Springer International Publishing, pp 58–71. Web Colli M, Madsen O, Berger U, Møller C, Wæhrens BV, Bockholt M (2018) Contextualizing the outcome of a maturity assessment for Industry 4.0. IFAC-Pap. 51:1347–1352 Crnjac M, Veža I, Banduka N (2017) From concept to the introduction of industry 4.0. Int Ind Eng Manag 8:21–30 Cruickshank IJ, Carley KM (2020) Characterizing communities of hashtag usage on twitter during the 2020 COVID-19 pandemic by multi-view clustering. Appl Netw Sci. https://doi.org/10.1007/ s41109-020-00317-8 De Jesus C, Lima RM (2020) Literature search of key factors for the development of generic and specific maturity models for industry 4.0. Appl Sci 10(17):5825. Web Eckhart M, Ekelhart A (2019) Digital twins for cyber-physical systems security: state of the art and outlook. In Security and quality in cyber-physical systems engineering. Springer International Publishing, pp 383–412. https://doi.org/10.1007/978-3-030-25312-7_14 Girma A (2018) Analysis of security vulnerability and analytics of internet of things (IoT) platform. In: Information technology—new generations. Advances in intelligent systems and computing, 738, https://doi.org/10.1007/978-3-319-77028-4_16 Glaessgen E, Stargel D (2012) The digital twin paradigm for future NASA and U.S. air force vehicles. In: 53rd AIAA/ASME/ASCE/AHS/ASC structures, structural dynamics, and materials conference Gómez-García M, Matosas-López L, Ruiz-Palmero J (2020) Social networks use patterns among university youth: the validity and reliability of an updated measurement instrument. Sustainability. https://doi.org/10.3390/su12093503 Grieves M (2014) Digital twin: manufacturing excellence through virtual factory Replication Haag S, Anderl R (2018) Digital twin—proof of concept. Manuf Lett 15:64–66. https://doi.org/10. 1016/j.mfglet.2018.02.006 Hearn M, Rix S (2019) Cybersecurity considerations for digital twin implementations report. IIC J Innov International Labor Organization. Industries and sectors. Accessed 16 May 2021. Available: https:// www.ilo.org/global/industries-and-sectors/lang--en/index.htm Jones D, Snider C, Nassehi A, Yon J, Hicks B (2020) Characterising the digital twin: a systematic literature review. CIRP J Manuf Sci Technol 29:36–52. https://doi.org/10.1016/j.cirpj.2020. 02.002 Jung K, Kulvatunyou B, Choi S, Brundage MP (2016) An overview of a smart manufacturing system readiness assessment. In: Nääs I et al (eds) Advances in production management systems. Initiatives for a sustainable world. APMS 2016. IFIP advances in information and communication technology, 488. https://doi.org/10.1007/978-3-319-51133-7_83 Kluth A, Jäger J, Schatz A, Bauernhansl T (2014) Method for a systematic evaluation of advanced complexity management maturity. Procedia CIRP 19:69–74. https://doi.org/10.1016/j.procir. 2014.05.041 Koehler J, Hofstetter J, Woodtly R (2012) Capabilities and levels of maturity in IT-based case management. In: Barros A, Gal A, Kindler E (eds) Business process management. BPM 2012. Lecture Notes in Computer Science, 7481. https://doi.org/10.1007/978-3-642-32885-5_4 Kohlegger M, Maier R, Thalmann S (2009) Understanding maturity models—results of a structured content analysis. In: Proceedings of the 9th international conference on knowledge management (I-KNOW’09), pp 51–61

98

J. Scheibmeir and Y. Malaiya

Martin Y, Cutter S, Li Z (2020) Bridging twitter and survey data for evacuation assessment of Hurricane Matthew and Hurricane Irma. Nat Hazard Rev. https://doi.org/10.1061/(ASCE)NH. 1527-6996.0000354 Mittal S, Khan M A, Romero D et al (2018) A critical review of smart manufacturing & Industry 4.0 maturity models: implications for small and medium-sized enterprises (SMEs). J Manuf Syst 49:194–214 Nappo S (2020) (@stephanenappo) Virtual models, aka #digitaltwins, further bridging #physical & #digital worlds. 21-Jan-2020. Tweet Negria E, Fumagallia L, Macchia M (2017) A review of the roles of digital twin in cps-based production systems. In: International conference on flexible automation and intelligent manufacturing, pp 939–948 Novakova M (2019) Trends of enterprise information systems in 4.0 conditions. Diploma Thesis, University of Economic Novo O, Di Francesco M (2020) Semantic interoperability in the IoT. ACM Trans Internet Things 1(1):1–25. https://doi.org/10.1145/3375838 Oztemel E, Gursev S (2018) Literature review of Industry 4.0 and related technologies. J Intell Anuf 31:127–182 Panetta K (2021) Gartner Top 10 Strategic Technology Trends for 2019. Smarter with Gartner, 15-Oct-2018. [Online]. Available: https://www.gartner.com/smarterwithgartner/gartner-top-10strategic-technology-trends-for-2019/. Accessed 18-Jul-2021 Parente TC, Federo R (2019) Qualitative comparative analysis: justifying a neo-configurational approach in management research. RAUSP Anag 54:399–412 Parkinson B (2015) The Connected Enterprise® maturity model: metrics that matter. Retrieved 24 July 2021, from https://www.rockwellautomation.com/en-us/company/news/blogs/the-con nected-enterprise-maturity-model--metrics-that-matter.html PWC (n.d.) Industry 4.0 self-assessment. Retrieved 24 July 2021, from https://i40-self-assessment. pwc.de/i40/landing/ Rai O (2019) (@Omkar_Raii), Predictions by @GlobalNTT reflect ho #emergingtech such as #AI, #ML, #IoT, #RPA #Cybersecurity, #DigitalTwins, #AR, #VR, #Blockchain & #BigData will significantly improve productivity, growth & innovation across entire work, live and play environments. 13-Nov-2019. Tweet RolSOuLi (@M6sp2004), #IOTA #iotatoken #coloredcoins #DigitalTwins $IOTA $TANGLE $MIOTA #MachineLearning #iotcommunity #iottrends #iottechnology #MIOTA #TANGLE #iotdevices #iotworld #iotsolutions #BigData #CloudComputing Ya!!! Fecha Oficial para la migración a Chrysalis. 18-Mar-2021. Tweet Rosen R, Von Wichert G, Lo G, Bettenhausen KD (2015) About the importance of autonomy and digital twins for the future of manufacturing. In: 15th IFAC symposium on information control problems in manufacturing INCOM 48(3):567–572 Scheibmeir J, Malaiya Y (2019) An API development model for digital twins. In Proceedings— Companion of the 19th IEEE international conference on software quality, reliability and security, Institute of Electrical and Electronics Engineers Inc., pp 518–519. https://doi.org/10.1109/QRSC.2019.00103 Scheibmeir J, Malaiya YK (2020) Multi-model security and social media analytics of the digital twin. ASTEJ. 5:323–330 Schroeder G, Steinmetz C, Pereira C, Muller I, Garcia N, Espindola D, Rodrigues R (2016) Visualising the digital twin using web services and augmented reality. In: 2016 IEEE 14th International conference on industrial informatics (INDIN), pp 522–527 Shafto M, Conroy M, Doyle R, Glaessgen E, Kemp C, LeMoigne J, Wang L (2010) Draft modeling, simulation, information technology & processing roadmap. Technology Area 11 Shelby Z, Hartke K, Bormann C (2014) The constrained application protocol (CoAP). https://doi. org/10.17487/RFC7252

Social Media Perspectives on Digital Twins and the Digital …

99

Siemens (2020) https://www.plm.automation.siemens.com/media/global/pt/Siemens SW Digitalization roadmap Fact Sheet_tcm70-71287.pdf. Retrieved 24 July 2021, from https://www.plm.aut omation.siemens.com/media/global/pt/Siemens SW Digitalization roadmap Fact Sheet_tcm7071287.pdf Storey M et al (2010) The impact of social media on software engineering practices and tools. In: Proceedings of the FSE/SDP workshop on future of software engineering research. ACM, pp 359–364 Tao F, Zhang M, Liu Y, Nee AYC (2018) Digital twin driven prognostics and health management for complex equipment. CIRP Ann Manuf Technol 66:169–172 Tiffany, R (@RobTiffany), Welcome to #IoT Coffee Talk #44 where we chat about #Digital #IIoT #Automation #DigitalTwins #Edge #Cloud #DigitalTransformation #5G #AI #Data & #Sustainability over a cup of coffee. 24-Mar-2021. Tweet TMMi Foundation (2020) Model aims and objectives. Retrieved 24 July 2021, from https://www. tmmi.org/model-aims-and-objectives/ Uhlemann T, Lehmann C, Steinhilper R (2017) The digital twin: realizing the cyber-physical production system for industry 4.0. Procedia CIRP 61:335–340. https://doi.org/10.1016/j.procir.2016. 11.152 Ullah I, de Roode G, Meratnia N, Havinga P (2021) Threat modeling-how to visualize attacks on IOTA? Sensors (basel, Switzerland) 21(5):1834. https://doi.org/10.3390/s21051834 Weber C, Königsberger J, Kassner L, Mitschang B (2017) M2DDM—A maturity model for datadriven manufacturing. Procedia CIRP 63:173–178. https://doi.org/10.1016/j.procir.2017.03.309 Williams A (2018) Do software engineering practitioners cite research on software testing in their online articles? A preliminary survey. ACM, pp 151–156. https://doi.org/10.1145/3210459.321 0475