Journal of Manufacturing Systems 64 (2022) 372–389

Contents lists available at ScienceDirect

Journal of Manufacturing Systems

journal homepage: www.elsevier.com/locate/jmansys

Review

Digital twin modeling

Fei Tao a, Bin Xiao a, Qinglin Qi a, b, *, Jiangfeng Cheng a, Ping Ji b
a
School of Automation Science and Electrical Engineering, Beihang University, Beijing 100191, China
b
Department of Industrial and Systems Engineering, The Hong Kong Polytechnic University, Hong Kong, China

A R T I C L E I N F O A B S T R A C T

Keywords: The digital twin is an emerging and vital technology for digital transformation and intelligent upgrade. Driven by
Digital twin data and model, the digital twin can perform monitoring, simulation, prediction, optimization, and so on.
Digital twin modeling Specifically, the digital twin modeling is the core for accurate portrayal of the physical entity, which enables the
Digital twin model
digital twin to deliver the functional services and satisfy the application requirements. Therefore, this paper
Enabling technologies
provides systematic research of current studies on the digital twin modeling. Since the digital twin model is a
Enabling tools
faithful reflection of the digital twin modeling performance, a comprehensive and insightful analysis of digital
twin models is given first from the perspective of the application field, hierarchy, discipline, dimension, uni­
versality, and functionality. Based on the analysis of digital twin models, current studies on the digital twin
modeling are classified and analyzed according to the six modeling aspects within the digital twin modeling
theoretical system proposed in our previous work. Meanwhile, enabling technologies and tools for the digital
twin modeling are investigated and summarized. Finally, observations and future research recommendations are
presented.

1. Introduction The digital twin has progressed from theoretical research to prag­
matic implementation, whereas the model is a paramount constituent of
The increasing production intricacy in a more demanding market is the digital twin and a prerequisite for successful digital twin applica­
gathering momentum for the integration of the physical and digital tions. In virtual space, based on the attributes of the physical entity, the
world. At the same time, human’s escalating practical needs for indus­ digital twin model can be expressed in four model dimensions: geome­
trial products are challenging the digital model’s capability to interact try, physics, behavior, and rule [9]. The geometric model describes the
with the physical object. The digital twin was conceived in this context geometric shape and assembly relationships of the physical entity. The
and has sparked a far-reaching industry revolution [1,2]. In 2003, the physical model reflects the physical properties, characteristics and
concept of the digital twin was originally proposed by Michael Grieves constraints of the physical entity. The behavioral model represents the
[3] in his product lifecycle management course at the University of dynamic behavior of the physical entity in response to the internal and
Michigan. In the beginning, digital twins were employed primarily in external mechanisms. The rule model incorporates historical data and
the military and aerospace. Currently, the digital twin is in a period of can exploit tacit knowledge, making the digital twin model smarter. By
rapid development [4]. As seen in Fig. 1, more than 1000 papers con­ incorporating multidisciplinary knowledge, the multidimensional digi­
cerning the digital twin have been published in each of the three tal twin model can perform functions like prognostication, optimization,
consecutive years since 2019. By the end of 2021, 2934 papers were and control in the digital world.
published in the year. Besides the surging number of papers year by However, the effective digital twin model needs to be built from
year, the digital twin was listed as top 10 strategic technology trends by multiple digital twin modeling aspects. In our previous work [10], the
Gartner company for three consecutive years, ranking No. 5 in 2017 [5], digital twin modeling theoretical system was proposed, which decon­
and No. 4 in 2018 [6], 2019 [7]. In 2020, Gartner also listed the digital structed and investigated the digital twin modeling from the six aspects
twin as an emerging technology for the next 5–10 years. At that time, of model construction, model assembly, model fusion, model verifica­
representations of the digital twin for the real and virtual worlds will be tion, model modification, and model management. Model construction
ubiquitous [8]. integrates the knowledge in related disciplines to construct the

* Corresponding author at: School of Automation Science and Electrical Engineering, Beihang University, Beijing 100191, China.
E-mail address: [email protected] (Q. Qi).

https://doi.org/10.1016/j.jmsy.2022.06.015
Received 23 January 2022; Received in revised form 9 June 2022; Accepted 29 June 2022
Available online 13 July 2022
0278-6125/© 2022 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved.
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

Meanwhile, a new digital twin model in the manufacturing context was

introduced. Berri et al. [18] provided a review of digital twin models
regarding the fluid-dynamic behavior in electro-hydraulic servo valves.
The comparison and evaluation of the precision of digital twin models
were illustrated as well.
These publications conducted detailed surveys of the functions of the
digital twin model for different objects or scenarios in manufacturing.
However, current review works focus on the functional representation of
the digital twin model at the surface layer, while overlooking the
analysis of the model attributes and features at the deep layer, which
shapes the digital twin model. Meanwhile, digital twin models have
been evolving and applied in various fields, but the current review works
focus only on the manufacturing domain. It is absent a thorough and
Fig. 1. Number of papers published per year during the digital twin growth insightful summary of the similarities, differences, and connections be­
stage [4]. tween the multiple attributes and functionalities of digital twin models
in various application fields.
elementary digital twin model, like unit level digital twin model in a Rasheed et al. [19] reviewed the current methods and techniques
specific application field. More complex digital twin model can be ob­ used in the model construction aspect of the digital twin model. The
tained through model assembly of elementary digital twin models in the existing challenges associated with the digital twin model construction
spatial dimension. In case assembled model in spatial dimension cannot were also elucidated, along with the authors’ informative proposals and
portray the internal mechanisms of the object or the production process insights. Bordeleau et al. [20] reviewed diverse model-driven engi­
comprehensively and incisively, model fusion in multiple fields, disci­ neering technologies used in model construction that potentially
plines, and model dimensions is required after model assembly. Model contribute to solving the digital twin modeling challenges mentioned.
verification for the constructed and assembled or fused digital twin Ríos et al. [21] offered a digital twin modeling review about measure­
model is also a necessity. If the digital twin model fails to address the ment uncertainty within data transfer standards and how it relates to
physical production requirements, further model modification is testing data when constructing the model. Andronas et al. [22] sum­
essential to secure the accuracy of the digital twin model. Crucial as marized and addressed the limitations of existing techniques for the
well, model management provides services, including the management digital twin modeling of the flexible material.
of digital twin models in the model library and the processing and uti­ For system-level digital twin models such as production equipment
lization of data and knowledge generated in the digital twin modeling. or the shop floor, their modeling is complex system engineering which is
Hence, as the basis for the digital twin model-driven approach and infeasible to construct the entire model in one go. Therefore, based on
imperative requirement for digital twin functionality implementation, proper model management, it is necessary to assemble and fuse unit-
the digital twin modeling drives the practical deployment of the digital level digital twin models and verify and modify the modeling validity.
twin technology in the relevant domains and industries. These papers reviewed the digital twin modeling perceptively from
In this paper, a systematic review of current research on the digital different viewpoints. However, current studies stagnated in model
twin modeling and its result, i.e., the digital twin model, will be pre­ construction and neglected other aspects of the digital twin modeling
sented mainly in terms of the four model dimensions and six modeling including model assembly, model fusion, model verification, model
aspects mentioned above. modification and management. Also, related reviews lack a summary
and analysis of the specific aspects of the digital twin modeling and the
1.1. Related review works enabling technologies and tools in each aspect. Moreover, the existing
research on technologies and tools for the digital twin modeling is
The digital twin has been an ongoing area of concern in various in­ discrete and divorced, where no universal and coherent analysis,
dustries recently. Correspondingly, a high volume of publications on the generalization, and elaboration for both is in progress.
digital twin modeling have been published, of which several reviews.
Barricelli et al. [11] reviewed current studies on the definitions of the 1.2. Purpose of this paper
digital twin and analyzed the difference between them. Lim et al. [12]
summarized the concepts of the digital twin and the related techniques The digital twin model is the foundation and container for the digital
within the full life cycle. In the design phase of the full life cycle, Leng twin functionality, while the digital twin modeling is the groundwork
et al. [13] overviewed the digital twin frameworks. Errandonea et al. for the digital twin model-driven methodology. Therefore, via an
[14] conducted further review studies in the digital twin maintenance exhaustive and incisive overview, this paper aims to:
phase of the full life cycle.
These studies reviewed the digital twin definitions, concepts, and 1) Provide a comprehensive summary and analysis of the currently
frameworks from various viewpoints, such as the full life cycle. How­ available digital twin models in terms of the application field, hier­
ever, as key elements for the digital twin applications, the modeling and archy, discipline, dimension, universality, and functionality.
model require more focused research. Therefore, a systematic review of 2) Classify and analyze the existing research on the digital twin
the digital twin modeling and model from multiple perspectives is modeling according to the six modeling aspects within the digital
needed. twin modeling theoretical system.
Melesse et al. [15] performed a systematic literature review to pro­ 3) Investigate the enabling technologies and tools accessible in the six
vide an assessment of the utility of the digital twin model in industrial digital twin modeling aspects.
operations and highlight the challenges in its implementation. In the 4) Propose future research orientations and prospective approaches to
review work of Wright et al. [16], they underlined the essence of the resolving existing problems and emerging challenges in the digital
differences between models and digital twins, overviewed some crucial twin modeling.
advantages of employing digital twin models, and recommended ori­
entations for further research to explore its maximum potential. Lechler The rest of this article is organized as follows. Section 2 introduces
et al. [17] overviewed and explained the relationship between the the literature classification criteria and review methodology. Section 3
functionality of the digital twin model and its application fields. analyzes the digital twin model from multiple perspectives. Section 4

373
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

researches the digital twin modeling from multiple aspects. Section 5 3. Multi-perspective analysis of the digital twin model
summarizes enabling technologies and tools used in different aspects of
the digital twin modeling. Several observations and future research Among these 296 academic publications, the 12 publications mainly
recommendations are given in Section 6. At last, the conclusions and focus on the macroscopic framework, modeling procedures and meth­
future work are provided in Section 7. odologies, the technologies, and the tools used in modeling, which will
be covered in later sections. In the remaining 284 publications, digital
2. Literature review methodology twin models derive from architectures or paradigms and application
examples in case studies. As shown in Fig. 2, these publications are
2.1. Classification criteria classified by the application fields of their digital twin models. Because
one reviewed publication may concern more than one application field,
The fundamental literature classification criterion in this study is the publication will be counted in each statistic data of its application
based on the application field of the digital twin model. digital twin fields which causes the total percentage of the fan diagram to be greater
models in various application fields are deconstructed, classified, and than 100 %. In the meantime, the 284 publications related to diverse
summarized from the perspective of the hierarchy, discipline, dimen­ application fields will serve as the principal basis for the statistics and
sion, universality, and functionality. As a way to realize the digital twin analytics in the subsequent sections.
model in different fields, the digital twin modeling in the publications
reviewed is deconstructed and categorized into corresponding modeling 3.1. Application field analysis of digital twin models
aspects within the digital twin modeling theoretical system. Further
dissections are taken in each aspect as well. Based on the paper classi­ Digital twin models have been successfully implemented in a wide
fication for the digital twin modeling, the enabling technologies and spectrum of manufacturing, such as from full-scale factories to precision
tools used in each modeling aspect are researched. parts, from macro manufacturing systems to specific product production
processes. From Fig. 2, nearly half of the reviewed publications are
2.2. Literature review methodology related to manufacturing. The application of digital twin models in
manufacturing is not only wide-ranging but also detailed and subtle. A
Our review is well-grounded in the scientific publications including variety of workpieces or parts make up a piece of integral equipment
journal articles, conference papers, book chapters, reviews, and edito­ with ad hoc utilities. Dissimilar equipment is used in production lines or
rials, of which the reviews with reference significance have been out­ shop floors for diverse production purposes. For robotics alone, the
lined in the introduction. The review is undertaken under four digital twin models can be categorized into several sub-applications
procedures: (1) Retrieve the Scopus database by advanced document (industrial robot [23–26], mobile robot [27], line-following robot
search with specific query strings to obtain the original retrieved [28], robot arm [29]) which are ready for specific production scenarios.
document results, totaling 331. (2) Eliminate extraneous publications by Except for manufacturing, digital twin models are working in fields like
reviewing detailed content. The pertinence of each retrieved publication energy, aerospace, engineering construction, city, healthcare, agricul­
to the research themes (e.g., attributes of the digital twin model) is ture, and so on. Second only to manufacturing, applications of digital
evaluated based on the abstract, introduction, and conclusion. For twin models in the energy field (35 papers) can be further refined into
instance, despite the keywords "digital twin" and "model" included in the energy exploitation and energy source equipment. The third major
publication title, its specific content emphasized the similarities and application field is aerospace (20 papers) including flight vehicles and
dissimilarities between the digital twin technology and model-based parts, equipment, or systems associated with them. In addition, digital
systems engineering rather than the digital twin modeling. Similarly, twin models have specific uses in other areas such as earthquake
the authors removed such irrelevant publications from further review. In monitoring [30], heritage conservation [31,32], and naval vessel repair
total, 296 academic publications are found after filtering out the [33].
extraneous ones and reviews. (3) Review the entire content of all rele­ Although the digital twin model has been effectively practiced in
vant publications filtered in step 2 and contextualize them under the many fields, in some fields, it can currently only be applied in a narrow
criteria presented in subsection 2.1. (4) Undertake the insightful analysis range of scenarios. For example, the current digital twin technology
depending on the application fields, attributes, and functionalities of the applied in the apparel field is only limited to the reconstruction of his­
digital twin model, different modeling aspects within the digital twin torical garments [34]. In future research, virtual design and try-on of the
modeling theoretical system, and related technologies and tools in the apparel can be achieved by constructing digital twin models for apparel
literature selected. The relevant data statistics and analytics will be in various styles to reduce the cost of physical design and enhance the
illustrated in later sections. Table 1 shows the database, query strings, try-on experience. In the chemical industry, the digital twin model ap­
original retrieved document results, papers filtered in step 2, and the plications are dedicated to the process and device level, i.e., phosphate
time frame of the literature review. slurry piping process [35], rectification installation [36], and combus­
tion furnace [37]. Therefore, the construction of digital twin models for
specific chemical materials or chemical molecules and atoms at a more
microscopic level is a seminal breakthrough for future digital twin
technology research. Besides narrow application scenarios, digital twin
models in some fields also suffer from uneven hierarchical distribution,
Table 1
insufficient multidisciplinary integration, one-sided study on model di­
Retrieval index of the literature review.
mensions, and a gap between the functions provided and the actual
Retrieval index Detail content needs, which will be elaborated on in later sections.
Database Scopus
Query strings TITLE ({digital twin} OR {digital twins} OR {Digital twin} OR 3.2. Hierarchical analysis of digital twin models
{Digital twins} OR {Digital Twin} OR {Digital Twins}) AND TITLE
({model} OR {models} OR {modeling} OR {modelling}OR
{Model} OR {Models} OR {Modeling} OR {Modelling}) Physical entities are hierarchical and the digital twin model is a
Document 331 virtual mapping of the physical entity. Accordingly, based on the
results concept of the digital twin shop floor hierarchy proposed by Tao et al.
Papers filtered 296 [9], digital twin models surveyed in this review can be generally clas­
Time frame 2003.1–2021.6.28
sified into unit level, system level, and system of systems (SoS) level by

374
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

Fig. 2. Application field distribution of digital twin models.

their structure and functionality. production-oriented hierarchy where it operates.

For a penetrating perception of the hierarchical structure of the The concept of the digital twin shop floor hierarchy is not limited to
digital twin models in manufacturing, the hierarchy can be divided into manufacturing, we extend the concept to fields like aerospace, city, and
production-oriented and product-oriented ones. healthcare. Table 2 shows their hierarchical distribution, and the
(1) Production-oriented hierarchy of the digital twin model: In respective analysis is performed below. In the field of aerospace, the
production-oriented hierarchy, equipment, production lines, and shop flight vehicle as a whole, we define as the SoS. The subsystem consti­
floors constitute unit level, system level, and SoS level, respectively. tuting the flight vehicle, such as the propulsion system, is defined as the
Unit-level digital twin model for individual equipment enables functions system. The component that performs the functions of the subsystem is
like monitoring, fault prediction, maintenance, etc. Production lines and defined as the unit. The current studies are not comprehensive for the
shop floors can be modeled to represent the interaction and coupling hierarchical structures of digital twin models of the flight vehicle. In
relationships between subsystems in favor of analyzing and predicting particular, system-level models require to be pioneered and well
the evolution of the whole system. Among the 284 publications, 51 investigated. In the field of city, the individual urban construction is
reflect the hierarchical theory, visualized in Fig. 3. Current research considered as the unit. The building complex, such as industrial estate
requires continued advancement not only at the unit level but also at the and residential areas, is at the system level. Finally, the whole city is at
system and SoS level. Based on an evenly distributed hierarchy, each the SoS level. The city-oriented digital twin models are majorly
level can be incorporated level by level to realistically portray the distributed at the unit level and SoS level. However, there is a complete
physical shop floor. Besides uneven hierarchical distribution, the exist­ vacuum in the study at the system level. Therefore, system-level digital
ing connections between levels mainly stay at the geometric dimension, twin models, such as commercial, industrial and residential areas, are
with weak coupling and poor linkage in physical or logical dimensions. new directions for future research.
(2) Product-oriented hierarchy of the digital twin model: In addition Besides the varying degree of success achieved in the engineering
to being a production unit in the production-oriented hierarchy, pro­ area, the digital twin model in healthcare is receiving increasing
duction equipment has its hierarchical structure as a product. There are attention [68]. With the “best-fit” feature of the digital twin technology,
several subsystems in the machine tools. For example, the digital twin digital twin models within healthcare setting will be a break from
mechanical machine tool contained the main body, spindle system, and established care paradigms [69]. It was reported that a quarter of
feed system. Objects with specific functions, like drivers, motors, etc., healthcare executives experimented with the digital twin in 2021 and
were contained in the feed system [25]. Here, the mechanical machine 66 % of them indicated their investment in the digital twin will increase
tool is at the SoS level with its subsystems at the system level. Functional over the next three years [70]. With significant application value and
execution units, such as motors, are at the unit level. Few studies now broad application prospects, the digital twin model in healthcare is
investigate a product-level digital twin model from the perspective of a
product-oriented hierarchy. The integrity of the digital twin model will
be compromised without an accurate portrayal of its sub-models and the Table 2
connections between them at each level of the product. This adversely Hierarchy of the digital twin model in other fields.
affects the functionalities of the digital twin model and consequently the Field Unit System SoS

Aerospace Aircraft tire [38] N/A UAV

Air rudder [39,40] [46–48]
Aero-engine bearing Spacecraft [49]
[41] Rocket [50]
Aircraft cabin [42]
Hydraulic valve [43]
Air bearing [44]
Turbofan [45]
City Utility pole [51] Campus [60] City
Bridge [52–54] [61–64]
Historical architecture
[31–34]
Building [55–59]
Healthcare Coronary heart vessels Cardiovascular N/A
[65] system [67]
Heart [66]
Fig. 3. Hierarchical distribution in the digital twin shop floor.

375
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

analyzed according to its hierarchical structure, which can be further maintenance from responsive to anticipatory, while cutting hospital
subdivided into human-oriented and medical resources-oriented operation and management costs. No papers have yet addressed the
hierarchies. medical resources-oriented hierarchy of the digital twin model——as a
(1) Human-oriented hierarchy of the digital twin model: We define promising research direction.
the overall human body as the SoS from a biologically and medically-
based perspective. Then the eight systems, for example, the cardiovas­
cular system, making up the human body, are at the system level. Lastly, 3.3. Disciplinary analysis of digital twin models
the human organ is defined as the unit. Although digital twins applied in
healthcare is a popular research trend nowadays, there are rarely suc­ Among the 284 publications, 137 are selected where relevant disci­
cessful implementations, with only coronary heart vessels [65] and plines were specifically considered in digital twin models. Based on
heart [66] at the unit level, and cardiovascular system [67] at the system relevant disciplinary backgrounds or oriented to relevant disciplinary
level. Even more unfortunate is the lack of exhaustive and precise digital problems, these publications utilized relevant disciplinary methodolo­
twin models for the entire human body. Since the digital twin technol­ gies, technologies, or tools in the digital twin modeling, or their digital
ogy perfectly corresponds to the brand-new health philosophy today, twin models incorporated relevant disciplinary knowledge or solved
also known as preventative medicine, the study of digital twin models in relevant disciplinary problems in practical applications. Therefore, some
the medical field will be a compelling facilitator for future medical publications included more than one discipline. These publications will
concepts. be counted in each statistic data of their disciplines which causes the
(2) Medical resources-oriented hierarchy of the digital twin model: total percentage of the fan diagram to be greater than 100 %. Collec­
For a magnetic resonance imaging machine, the entire equipment is SoS tively, those selected publications encompass 43 distinct disciplines
and its subsystems such as the radio frequency (RF) system are at the displayed in Fig. 4. It can be noted that digital twin models embrace a
system level. Correspondingly, the functional units constituting the RF wide range of disciplines and become interdisciplinary studies. For
system such as RF coil, RF generator, etc., are at the unit level. The example, electro-mechanical-hydraulic integration has always been one
digital twin model hierarchies of medical resources enable real-time of the crucial development directions of smart manufacturing. The
monitoring and interconnection of devices and even parts on each employment of digital twin models further enhances the electro-
level of the device. As a result, it is expected to transform the device mechanical-hydraulic integration within large equipment like preci­
sion machine tools [71–80], making them more automated and

Fig. 4. Disciplinary distribution of digital twin models.

376
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

intelligent. Machine learning arising from interdiscipline can predict can be found in other fields, such as engineering construction, health­
protein 3D structures in the medical field [81], and their seamless care, city, and vehicle. As an example from healthcare [68], The geo­
integration within the digital twin model allows an extension in pre­ metric model was built by the mesh according to the peculiarities of the
diction performance. coronary heart vessel. The grid structure enabled the physical model to
In general, the digital twin is of the enormous potential to innovate reflect the properties of the vessel, like the presence or absence of
traditional disciplines, empower emerging ones, and boost the research vortices, the degree of dissipation, and the blood distribution. The
in highly sophisticated ones. However, existing research remains behavioral model incorporated a system of differential equations to
concentrated on manufacturing-related disciplines, such as mechanics describe the flow process in the vessel. The rule model can predict the
and electricity. The extensions to emerging disciplines like data science heart diseases progression based on the stress-strain properties and flow
and interdisciplinary ones like molecular biology will drive the digital process. But in general, the four model dimensions are not compre­
twin technology into a new era. hensively and adequately conducted in these fields. Some dimensions
received disproportionate consideration, but others garnered limited
attention. Therefore, future research needs to be explored more inten­
3.4. Dimensional analysis of digital twin models sively and systematically based on methodical approaches towards the
four model dimensions, thus leading to a more multipurpose and accu­
Tao et al. [9] proposed five-dimensional digital twin model, rate digital twin modeling.
including physical entity, virtual model, connection, data, and service.
The virtual model is composed of sub-models in four dimensions: geo­
metric model, physical model, behavioral model, and rule model. These 3.5. Universality analysis of digital twin models
sub-models are equipped with their own characteristics and functions.
By assembling or fusing the above sub-models, the virtual mapping of Based on the applicable scope of the digital twin model in the pub­
the digital twin model for the physical entity can be achieved. Among lications reviewed, the universality of the digital twin model is studied
the 284 publications, 126 in 15 application fields, ranging from in this section. Digital twin models are further selected and summarized
manufacturing to vehicle and environment to business, are further into two categories: 1) specific one peculiar to an object/scenario; 2)
screened. In these publications, their digital twin models partially or generic one available for application fields.
completely reflected the four model dimensions. (1) Specific digital twin model: Oriented by the actual needs or
63 of the 126 publications with digital twin models correlate to problems from specific application objects and scenarios, the publica­
manufacturing. Fig. 5 presents that the digital twin model research to­ tions, as shown in Fig. 6, adopted a method of procedure with relevant
wards the four model dimensions was studied in a relatively balanced technologies or tools in the digital twin modeling, which make specific
manner. It is worth noting that there are 29 publications whose digital digital twin models meet the actual needs or solve the actual problems.
twin models incorporated the full four dimensions. As an example from Individuality, effectiveness, and applicability are the distinguishing
[96], the virtual mapping at the geometric dimension for the physical features of specific digital twin model, which were validated by the
production line was achieved by the 3D model assembly. Static struc­ application requirements from certain objects or scenarios. Neverthe­
tural modeling created physical models where the physical attributes less, there are challenges for mathematical modeling to maintain precise
and structural features of the equipment were reflected. The production consistency between the digital twin model and sophisticated physical
behavior of the equipment was described by kinematic analysis. By the system. To some extent, it is arduous for a specific digital twin model to
model fusion of the equipment behavioral models, the production line fulfill all the demands from certain application scenarios. Meanwhile,
rule model was obtained to achieve remote control in virtual space. specific digital twin models may undermine the accurate mapping of
However, there are still some application examples where the digital physical assets and desired efficacy. However, specific digital twin
twin model is not fully functional because the four model dimensions are models for different application scenarios and even distinct domains can
not completely represented. In efforts to facilitate digital transformation be assembled or fused to address such challenges based on the corre­
and upgrade in the manufacturing industry, the establishment of a sponding constraints and coupling relationships. The details regarding
complete functional digital twin model is a prime priority. Creating the assembly and fusion will be given in Section 4.
virtual model based on geometric, physical, behavioral, and rule model (2) Generic digital twin model: The generic digital twin model is
dimensions is the infrastructure for the materialization of the complete expressed as the framework or architecture at the macro level whose
functional digital twin model. concepts, paradigms, or know-what can guide the development and use
Besides manufacturing, digital twin models with full four dimensions of digital twin models in specific applications. 55 out of the 284 publi­
cations incorporated the generic digital twin models, and the 8 appli­
cation fields they covered are shown in Fig. 7. Besides the 102 for the
specific one and 55 for the generic one, there are 127 out of the 284
publications where the definitions, concepts, and frameworks of the
generic digital twin model were referenced and extended. For example,
the five-dimensional digital twin model [9] is available for all applica­
tion fields. Among these 127 papers, researchers have used it in the
aircraft assembly in aerospace [82], the pre-warning system in vehicle

Fig. 5. Dimensional distribution of digital twin models in manufacturing. Fig. 6. Distribution of specific and generic digital twin models.

377
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

[83], the industrial product in manufacturing [84], the transport ser­

vices in transportation & logistics [85], and the structural design in
engineering construction [86]. Notwithstanding the significant advan­
tages of the generic digital twin model over a specific one, such as
universality, scalability, and versatility, the standard modeling ap­
proaches are prerequisites for successfully unleashing its maximum
potential.

3.6. Functionality analysis of digital twin models

Beyond the idea or concept stage of what digital twin models could
do, 102 of the 284 publications are selected which developed specific
functions to solve problems, meet needs, or reach performance in their
case studies. From Fig. 8, it is apparent that digital twin models currently
offer an extensive functional spectrum with 18 functions. Meanwhile,
Fig. 7. Application field distribution of generic digital twin models. the digital twin models in some publications are multifunctional. These
publications will be counted in each statistic data of their functions
which causes the total percentage of the fan diagram to be greater than
100 %. The primary functions are visualization, prediction,

Fig. 8. Functional distribution of digital twin models.

378
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

optimization, simulation, and monitoring. Not restricted to the con­ optimized design, virtual interaction, and the like. Therefore, geometric
ventional functions, the digital twin model has been expanded to edu­ model fidelity is crucial for geometric model construction. For example,
cation [87–90], management [91–93], and so forth. Therefore, for the Zhang et al. [94] predicted the tool wear state by premium quality visual
practical and specific requirements of digital twin models in various display and analysis in virtual space.
application fields, it is imperative for the existing conventional functions For fast geometric model transfer, loading, and browsing, an
to be consolidated, intensified, and refined for efficacy maximization. appropriate model simplification method is expected. Geometric model
Simultaneously, further exploration of the new functions to provide simplification can use fewer data to achieve high fidelity representation
more content is also required. of physical entities with small transfer files. It also provides high
More precisely, some more targeted functions are determined by the compatibility across different platforms. For example, Li and Nan [95]
own attributes of sub-models in different dimensions. The collaboration proposed a generic model simplification method that retains meaningful
out of the fused sub-models can better respond to the demands from mesh structure details while redundant facets are reduced, enabling
specific application scenarios. So, the functionality should be analyzed efficient model edit and analysis for users.
from the unique properties of sub-models in different dimensions and
the collaboration between them. After the appropriate iterations of 4.1.2. Physical model construction
model verification and modification, more effective and efficient func­ Physical model construction provides the basis for services such as
tions can be derived. quality control and the analysis and prediction of the physical property.
In detail, physical model construction can classify into static and dy­
4. Multi-aspect analysis of the digital twin modeling namic ones.
Static physical model construction includes the quantitative
The digital twin modeling is digital modeling in virtual space for the modeling of physical property, state, and behavior, which is determined
properties, methods, behaviors, and other characteristics of the physical by physical entities solely and independent of varying physical analysis
asset. In the following section, the digital twin modeling will be illus­ methods. For today’s increasingly complex engineering applications,
trated from the six modeling aspects within the digital twin modeling multi-physics coupling analysis is required in static physical model
theoretical system with several application cases. construction. By a multi-physics field coupling modeling approach,
Ruan et al. [96] constructed a fluid-thermal-electric multi-physics field
coupling model to unveil the reinforcement of thermoelectric properties
4.1. Model construction by nanofluids.
For a dynamically varying physical system, such as thermal con­
Of the 296 publications filtered in step 2 of subsection 2.2, 90 are duction inside a mechanical part, a finite number of nodes need to be
selected that highlight the methodologies, technologies, and tools created and calculated on the Spatio-temporal solution domain to obtain
available in model construction. Incorporating knowledge from corre­ the physical state distribution of the whole system. Based on this
sponding disciplines in the particular field, the digital twin model can be methodology, Wei et al. [76] built a finite element model for real-time
accurately constructed from the four model dimensions. From Fig. 9, wear simulation to enable dynamic assessment of performance attenu­
model construction primarily concentrated on the physical dimension, ation of CNC machine tools by digital twin models.
while the other dimensions received less attention. For advancing future
model construction research in a well-balanced manner and straight to 4.1.3. Behavioral model construction
the point, this section will analyze and conclude model construction in The behavioral model is constructed to represent the sequential,
terms of four model dimensions. concurrent, linked, periodic, and random behaviors of physical entities.
An accurate behavioral model determines the exactitude of the motion
4.1.1. Geometric model construction and control of the digital twin model. However, physical entities are
Geometric model construction delineates the shape, size, internal exposed to a variety of uncertainties in practical operation, and behav­
structure, spatial position and attitude, and assembly interfaces of ioral models tend to be discrepant as a result.
physical entities. For geometric model construction, the model fidelity Abnormal values in the data typically undermine the behavioral
and simplification are worthy of attention. model accuracy or consequently deviate the digital twin model from the
The geometric model is not only for shaping, but its structural physical entity. Therefore, a precise analysis of the root of anomalous
integrity and data accuracy also underpin the motion analysis, data and its relevance to other variables is essential when constructing
behavioral models. For example, Boulfani et al. [97] constructed a more
accurate behavioral model by extracting and analyzing the amount of
abnormal temperature variation from the physical generator mis­
behaviors and removing the abnormal behaviors from the digital twin
model of the generator.
Meanwhile, adjusting the algorithm parameters to identify the
optimal value of each parameter allows for the improvement of the
model accuracy. The adjustment of the parameters can be an iterative
process on several well-run models. However, parameter tuning entails a
deep insight into the parameters’ meaning and their impact on the
model. Based on the digital twin model and data, Luo et al. [98] studied
a hybrid predictive maintenance algorithm with optimal parameters by
algorithm adjustments, which realized the prediction of dynamic be­
haviors of the cutting tool.

4.1.4. Rule model construction

Rule model construction unveils the implicit knowledge and portrays
the evolutionary trends and patterns of physical entities. Based on the
whole life cycle of physical entities, there are two principal ways of rule
Fig. 9. Dimensional distribution in model construction. model construction, i.e., mining and analysis of whole life cycle data and

379
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

formal representation of experience and knowledge. incorporates the fusion between different model dimensions within a
The whole life cycle data is the upper limit of the rule model, and the digital twin model, as well as the fusion between digital twin models in
rule model construction is the continuous approximation toward this different fields.
upper limit. Therefore, for the maximum mapping of the whole life cycle
data by the rule model, data pre-processing is required. Scheffel et al. 4.3.1. Fusion between model dimensions within a digital twin model
[99] enable the fault detection of the digital twin model based on the By clarifying the coupling mode and constructing the coupling
pre-processing of raw data from phases of the whole life cycle of the relationship beyond the geometric space, the fusion between four model
machine. dimensions within a digital twin model can be accomplished. For case in
Through means like data mining, information processing, knowledge point, Liu et al. [103] created the individual instances of ontology
measurement, and graphical mapping, complex experience and knowl­ classes in different dimensions and fused them into a semantic net with
edge are empowered with rule models. Therefore, rule models can slot values. In this way, the production system hierarchy for model
reveal dynamic evolution in experience and knowledge domains, which fusion was constructed.
eventually enable digital twin models to understand and apply human
intelligence. Lermer and Reich [100] created the digital twin model 4.3.2. Fusion between digital twin models in different fields
through the basic knowledge extraction of the dataset from the pro­ Besides the fusion within a digital twin model, the fusion of ones with
duction process and its mapping to a fuzzy rule set of expertise. Hence, various disciplinary knowledge in different fields is crucial for con­
the potential of the digital twin model for machine condition assessment structing digital twin models at the system level or SoS level. The digital
was further exploited by the additive basic knowledge denoted by twin model of the CNC machine, for example, is composed of digital
discovered fuzzy rules. twin models like electrical, hydraulic, and mechanical systems con­
taining various disciplinary knowledge in different fields. The electrical
4.2. Model assembly system controls the hydraulic system through a communication inter­
face. The hydraulic system drives the mechanical system through a drive
Model assembly is a process to realize unit-level digital twin models connection. The mechanical system in motion feeds information back
to the one on higher hierarchies. Model assembly can be realized by into the electrical system. In this manner, the digital twin models of
adding appropriate spatial constraints of physical assets based on hier­ these systems are fused to develop an intelligent closed loop. Similarly,
archical relations and assembly sequence between models. For example, Luo et al. [71] fused the electrical, hydraulic, and mechanical systems
Angjeliu et al. [31] identified and geometrically assembled the struc­ into the digital twin descriptive model where fault prediction and
tural elements of complex architecture regarding the hierarchy from diagnosis for the CNC milling machine tool were achieved.
parts to complete model. Unlike traditional CAD model assembly, digital
twin model assembly can be dynamically adjusted based on 4.4. Model verification
bi-directional data communication in real-time to ensure the
high-fidelity mapping of physical assets throughout their life cycle. More For achieving functional efficacy and meeting requirement consis­
specifically, model lightweight before model assembly and interference tency, model verification is needed, which ensures that the digital twin
detection during model assembly are considerations for effective model accurately mirrors the actual system. Model verification is to
assembly. evaluate the consistency between the digital twin model and the phys­
ical object output under the same conditions. The unit-level digital twin
4.2.1. Model lightweight for model assembly models are required to be verified first to guarantee a valid foundation
The format, information granularity, and modeling accuracy of the for a more complex one. Key points in model verification can be
models to be assembled are commonly unevenly distributed, causing concluded as follows:
increased complexity and reduced efficiency in model assembly.
Therefore, lightweight digital twin models are expected as long as in­ 4.4.1. Information reconfirmation for model verification
formation integrity, model accuracy, and functionality satisfy the Assess all information relevant to the digital twin model and iden­
scenario-specific needs. For example, Fang et al. [101] proposed a tifies what additional analysis to improve the model credibility. There­
lightweight method for model assemblies that decrease the geometric fore, Model verification and modification is an iterative process where
errors of simplified meshes and increase the frame rate. The proposed the model is continuously refined until satisfactory [10].
method was validated in several digital twin production scenarios.
4.4.2. Vertical analysis for model verification
4.2.2. Interference detection for model assembly Evaluate the model evolvement at various historical phases to assess
When conducting model assembly in the virtual space, errors are the model suitability for the prospective application. For example, by
inevitable in model construction. Consequently, interference problems utilizing the predetermined cost function, Schluse et al. [104] developed
may occur between digital twin models at the same or different levels, iterations with the verification for historical parameters.
which compromise the accuracy of subsequent model assembly or even
failure to assemble. To improve the quality and efficiency of the as­ 4.4.3. Concept and fidelity analysis for model verification
sembly, interference detection is required during model assembly. Evaluate the model’s algorithms and sub-models in different di­
Zhang et al. [102] raised the assembly accuracy and efficiency by a mensions to identify the assumptions unbefittingly and sub-model fi­
method for assembly interference detection based on MRT, where geo­ delity for the prospective application. In predictive maintenance, sub-
metric topologies and assembly constraints are no longer in demand. model parameters were verified for accurate digital twin modeling ac­
cording to the deviation exceeding between actual and simulated signals
4.3. Model fusion in the predetermined range [23].

Model fusion is a necessity when model assembly in the geometric 4.4.4. Logic tracing for model verification
spatial dimension is insufficient to depict physical objects. For the Evaluate the digital twin model in behavioral and rule dimensions to
realistic mapping of the physiological state of the cardiovascular system, identify whether individual behaviors and their combinations are
Mazumder et al. [70] fused digital twin models of organs and tissue in desirable. For example, during the tensile tests, the candidates precisely
the cardiovascular system through pulmonary and systemic blood flow replicating the reduction were identified, otherwise, they will be
regulated by hemodynamic equations. In more detail, model fusion removed. Then, among identified candidates, the set replicating the

380
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

measured force in maximum precision was selected for production digital twin modeling aspect. Rational management is required to
simulation in the fracture toughness test [105]. guarantee that modeling is satisfied promptly with a system of high-
quality, efficient, packaged technologies and tools. The management
4.5. Model modification should include the feasibility, functionality, reliability, efficiency,
maintainability, usability, and portability of the modeling technologies
If there is an unacceptable deviation between the digital twin model and tools. Moreover, it in turn facilitates the management and utiliza­
and the physical object in model verification, model modification is tion of the data and empirical knowledge generated during the digital
required. Reasonable selection of the parameters for modification is twin modeling.
paramount to an effective and efficient model modification. Simulta­
neously, based on objective functions formulated rationally, the
4.7. Statistics analysis toward the digital twin modeling
appropriate methodology for modification is required.
The 120 papers in 16 application fields are selected and classified
4.5.1. Parameter for model modification
according to the six modeling aspects within the digital twin theoretical
To prevent pathological numerical problems in model modification,
modeling system. Meanwhile, nearly half of them (58 papers) is con­
the number of parameters selected for each modification should be
cerned with the manufacturing field. From Fig. 10, model construction
adequate. However, an insufficient selection of parameters will yield
has 45 papers related and remains the highest proportion among the six
less effective modification results. For example, depending on the effect
modeling aspects at 77.59 %. However, there is still neglect of model
of each modification, a certain amount of model parameters of new
management, with only three papers [107–109] on it, a mere 5.17 % of
porous metal plasticity was selected for updating. Accordingly, the
the six modeling aspects. It is evident that the current research on
sufficient selection of parameters for modification refined the predictive
modeling for the digital twin model mainly restricts to model con­
capability of the digital twin model [105]. Meanwhile, the parameter
struction and involves less on the other modeling aspects, especially the
variables to be modified should select those that are error-prone and
fewest research on model management. For establishing a complete
physically meaningful in the structure. For example, Zhang et al. [106]
digital twin model to adapt to the corresponding intricate application
used the proxy model to identify the locations of parameters for modi­
scenarios, it is advisable to reinforce the research on the five modeling
fication in virtual materials. At the same time, their error probability
aspects besides model construction.
was evaluated by the generic algorithm to optimize the modification
Except for manufacturing, only in aerospace, all six modeling aspects
accuracy.
within the digital twin modeling theoretical system are accessible.
However, the corresponding papers are far less frequent. For other
4.5.2. Methodology for model modification
fields, all modeling aspects are not thoroughly addressed, with the
After determining the parameters, a reasonable methodology for
exception that they all shed light on model construction, e.g., model
modification is demanded. The core of the modification method is to
construction in the field of the city [51,55,60,61,110,111], and model
construct an objective function that aligns the digital twin model output
construction in the chemical industry [35,36]. Simultaneously, the pa­
as close-to-real results as possible while enabling iterative modifications
pers are attainable whose general methodologies are feasible for one and
for the parameters of the digital twin model. In [31], while considering
several modeling aspects within all application fields ([112] for model
the structural response of the system, dynamic measurements were used
fusion, [113–117] for model management, and [118] for model con­
for model modification. The modification measured how accurately the
struction, verification, modification, and management).
model matches the real building, and the iteration of verification and
modification improved the accuracy of the model inch by inch. At the
5. Enabling technologies and tools in the digital twin modeling
same time, Vrabic et al. [27] utilized gradient descent for optimizing the
digital twin model output accuracy and control parameters in each cycle
There is still a mountain of challenges for technologies and tools that
until they converged to terminal values.
remain to be resolved for the digital twin modeling. Meanwhile, current
research concentrates on frameworks, processes, and know-what at the
4.6. Model management
macro level rather than specific technologies, tools, and know-how at
the micro-level [119]. Furthermore, the digital twin modeling is a highly
Based on the above digital twin modeling aspects, model manage­
intricate process that entails a time-intensive orientation, fine-tuning,
ment refers to multidimensional and multi-field digital twin model
and refinement. This section, therefore, surveys and studies the
management and model knowledge base management. It is crucial to
enabling technologies and tools used in the six digital twin modeling
managing the data generated from the digital twin modeling as well.
aspects, which serve as technological and tool references for the digital
Therefore, users can derive relevant services from model management.
twin modeling and consequently facilitate the digital twin imple­
In [46,47], a library for component-based reduced-order models was
mentation by researchers and practitioners.
constructed, where reduced-order models can be rapidly created,
adapted, and evaluated. However, the digital twin model attributes and
digital twin modeling data are not the entirety of the management,
where extensions to the authority, technology, and tool are possible.

4.6.1. Authority management for model management

For security and convenience reasons, some functions and services
provided by model management need to be restricted and customized
for different users. Therefore, authority management is needed, which
includes the authority to browse or utilize digital twin models and their
data at each digital twin modeling aspect, the authority to monitor the
model status and receive feedback after the digital twin model is
deployed to production, and so on.

4.6.2. Technology and tool management for model management

Inevitably, a multitude of technologies and tools are involved in each Fig. 10. Distribution of digital twin modeling aspects in manufacturing.

381
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

5.1. Enabling technologies and tools in model construction complex geometric information construction and animation production.
For web-side geometry modeling, web graphics library (WebGL) [129]
With disciplinary knowledge in multiple application areas, model combined open graphics library (OpenGL) [108] and the programming
construction establishes a model foundation on four dimensions for language of JavaScript [129] to provide 3D graphics application pro­
complex digital twin models. In this part, the technologies and tools gramming interfaces (APIs). To deliver digital twin services at people’s
implemented in model construction are investigated from the four fingertips, 3D rendering of mobile web pages with better cross-platform
modeling dimensions. capabilities are imperative. By simplifying and encapsulating WebGL,
ThreeJs [103,129], with its user-friendliness and low system resource
5.1.1. Enabling technologies and tools for geometric modeling consumption, will boost the 3D modeling of web pages on mobile in the
Geometric modeling technologies portray geometric appearance, future.
which mainly address the geometric and topological information of the
physical entity. In the literature reviewed, point cloud [31,32,58,60,93, 5.1.2. Enabling technologies and tools for physical modeling
109,120] and building information modeling (BIM) [31,32,58,93] are Physical modeling technologies are used to delineate the physical
the most frequently used geometric modeling technologies. Point cloud features and constraints based on geometric modeling. For physical
is a dense collection of points collected by certain measurement means. modeling, the most commonly used technology is finite element
The point cloud derived from laser scanning [31,32] and photographic modeling [32,76,77,106,130–134], using mathematical approximations
scanning (Photogrammetry [32], Pinhole camera modeling [51]) is to develop models of materials, loads, constraints, and mechanical
called a dense point cloud, where the number of points is larger and properties of the physical entity. The core of finite element modeling is
denser. Through point cloud, the target surface can be characterized, discretization. However, discretization operations are positively corre­
and finally, 3D physical entities are represented directly and efficiently. lated with the complexity of the physical entity. Accordingly, the
BIM incorporates the technological merits of static structural modeling reduced-order modeling technology [134,135] is available to reduce the
[103] and object-oriented modeling [71,73] to create 3D models with computational effort for complex physical entities. Because physical
non-graphical features and information. Compared to GIS analysis entities operate in a complicated environment, finite element modeling,
technique [30] focusing on the macro level, BIM covers the micro-level. as the recommended physical modeling technology in this paper, de­
By integrating the efficient data acquisition of point cloud and the macro mands expansion from a single structural field to a multi-physics
geometric spatial concept of GIS, BIM will fulfill more ambitious digital coupled field and seamless integration with physical modeling tools.
twin scenarios like digital twin cities. Apart from the above, finite Other technologies for physical modeling including Denavit-Hartenberg
element modeling technology, shown in Fig. 11, was used for geometric (D-H) notation [25], object-oriented modeling method [71,73], data
modeling of the historical building [31] and machine tool [76]. augmentation [136], and computational fluid dynamics (CFD) modeling
3D geometric models are commonly constructed by specialized tools [137,138] are shown in Fig. 11.
such as 3D modeling software. Shown in Fig. 11, Pro/E [103], CATIA By integrating physical modeling technologies, the physical model is
[121], SolidWorks [82,87,122,123] are the 3D modeling software established and then analyzed in the physical modeling tools. Ansys
dedicating to industrial machinery. Autodesk REVIT [57] and Surfer [130,139] is a general-purpose finite element analysis software, which
[30] are the ones focusing on building information modeling and 3D incorporates multiple physical fields. It delivers rich components for
geographic information modeling, respectively. Reverse modeling tools, rapid finite element modeling and visualizes its analytical results
such as LiDAR scanner [52], Zenfone AR [124], Kinect sensor [29], Laser through diverse charts. Also, as the mainstream finite element analysis
scanner [93], enable geometric modeling of physical entities as well. software, Abaqus [106] is more adept at complex nonlinear problems
Furthermore, importing models constructed by the above software and and structural analysis. In addition, by creating custom physical
tools into platforms like Flexsim [91], Abaqus [106], Demo 3D [125], modeling components, Simulink [36,44,90,122,131,140–142] can
Unity3D [89,126], and Siemens NX [127,128] can perform more describe the physical concepts behind physical entities and the physical

Fig. 11. Framework of technologies and tools in model construction.

382
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

connections between variables. Besides software, OpenModelica [23], machine learning tasks. Compared to artificial neural network (ANN)
an object-oriented and multi-domain modeling language, generally [143,149], deep convolutional neural network (DCNN) [148], can bet­
constructs physical models of mechanical, hydraulic, thermal, and ter simulate the brain while reducing overfitting. Noticeably, when
electrical components. PreComp [131] is a code specializing in creating using machine learning for rule modeling, the robustness of the rule
composite blades’ structural properties like cross-coupled stiffness. The model can be further improved by data augmentation [136]. As a subset
joint modeling of Simulink and Ansys is a promising approach. For the of machine learning, deep learning is a recommended rule modeling
continuous physical modeling optimization, researchers can load the technology with enhanced learning capability and broader coverage.
physical parameters Simulink generated onto finite element models of However, the open and dynamic environments physical entities typi­
Ansys and feed the analytical results back into Simulink. Fig. 11 shows cally operate in impose challenges on deep learning technology.
other physical modeling tools such as MWorks [71,73], Simcenter Based on the logic, laws, and rules, rule modeling tools equip the
Amesim software [50], Flexsim [91], Demo 3D [125], and Mechatronic digital twin model with intelligent capabilities and boost service quality.
Concept Designer [127]. Ontology Web Language (OWL) [88,150,151] is used for a semantic
description of ontologies. It draws on research in the AI field on
5.1.3. Enabling technologies and tools for behavioral modeling knowledge representation, especially logical description, thus delivering
Behavioral modeling technologies endow geometric and static a richer expressiveness for rule models of the digital twin model. Ten­
physical models with dynamic functional behaviors. Based on the mo­ sorFlow [152] is a symbolic mathematical system based on data flow
tion evolution in time and space studied by kinematic and dynamic programming and extensively applied to implement various machine
analysis [103], each behavior of a physical entity can be described by learning algorithms [153]. Its lightweight feature allows for faster cre­
differential equations[143]. For the behavioral model construction of ation and iteration of rule models. As a recommended rule modeling
the robot, Denavit-Hartenberg (D-H) notation [25] establishes the ki­ tool, Tensorflow is available in various fields of the digital twin research
nematic equations of the robot by formulating the end-effector position and can be deployed on various servers, PC terminals, and web pages.
and attitude over the base coordinate system. In more complex sce­ Fig. 11 shows other rule modeling tools like Ansys [68], Simcenter
narios, the dynamical equations for the physical entity’s behaviors can Amesim software [50], Demo 3D [125], Protégé [91,103,150,151], and
be established based on Lagrangian mechanics [27]. Mechanical systems Simulink [44,70,140,142].
are commonly exposed to random disturbances that affect the motion
evolutions of the system. As shown in Fig. 11, Stochastic modeling 5.2. Enabling technologies and tools in model assembly
technology [144] depicts the impact of random factors through random
variables and probability distributions. Hence, it can develop the sto­ After the individual models are constructed, the geometric connec­
chastic behavioral models of the system under random disturbances. In tions between the models need to be established for more complex
summary, external random factors’ impacts and internal geometrical or digital twin model. For model assembly, it is essential to consider the
physical parameters should be considered simultaneously in future topological information between models based on the model hierarchy
research on behavioral modeling technologies. and assembly sequence. Static structure modeling technology [103] can
Behavioral modeling tools can actualize the behaviors of digital twin address each unit-level digital twin model as an analysis class. Re­
models in virtual environments. In web-based digital twin applications, lationships in model assembly between classes can be established by
JavaScript control script [103] based on ThreeJs can realize the func­ describing the properties of each class and defining its state and oper­
tionality implementation, internal adjustment, external interaction, and ations. It is a new direction for model assembly research that enables
event response of digital twin models according to the real scenario rapid automatic assembly of 3D models based on the current 2D infor­
requirements. As object-oriented data modeling language, Automa­ mation such as facility plan drawings. Computer vision technology can
tionML [108,145] can be used to establish behavioral models of discrete extract information from images and process them for actual measure­
object interoperability. Besides the programming languages above, ments. By combining it with optimization algorithms like particle swarm
Unity3D [89,126] is a real-time 3D platform for interactive content optimization algorithm [109], a rapid automatic assembly can be ach­
creation. It enables rapid behavioral modeling by mounting C#-based ieved. As shown in Fig. 12, Monte Carlo method [109] can also be used
scripts to 3D objects in virtual scenes. Similarly, behavioral modeling for model assembly.
can be implemented via other platforms shown in Fig. 11, such as In addition to technologies, some well-known commercial tools are
Simcenter Amesim software [50], Ansys [68], Flexsim [91], CarSim available in the manufacturing field, such as Flexsim [91] and Demo3D
[123], Demo 3D [125], and Mechatronic Concept Designer [127]. [122] as shown in Fig. 12. During model assembly, these platforms can
Unity3D is recommended for behavioral modeling due to its accessi­ define the relative position of facilities to each other and the lateral area
bility, versatility, and rich learning resources. However, the disadvan­ of aisles, thus maximizing space utilization. More generic model as­
tage is that the packages exported from Unity3D consume a larger sembly tools are Unity3D [154] for PC and ThreeJS [103] for the web.
portion of resources. Both can effectively address the geometric relations between digital
twin models, including overlap, intersection, adjacency, etc. Simulink
5.1.4. Enabling technologies and tools for rule modeling offers a number of toolboxes that can be used in different specialized
Rule modeling technologies mine, extract and represent rules from areas. Simscape is the one specializing in multi-physics field modeling.
historical data, operational logic, and expertise, grounding the digital With it, users can assemble components by physical connections be­
twin model capabilities such as decision making, evaluation, and opti­ tween modular block diagrams. Based on the assembly topology of
mization. To realize the above capabilities, machine learning [146] is an components, 3D model assembly can be done by importing 3D models.
effective technology that continuously improves and extends the per­ However, the tools presently accessible for model assembly are pri­
formance of the digital twin model by reorganizing the existing marily for manufacturing. The development of model assembly tools for
knowledge structure. As shown in Fig. 11, Machine learning is formu­ other fields like healthcare will broaden the potential applications of the
lated in various forms, each with its features. For instance, by combining digital twin technology.
Monte Carlo [38] and Markov chain [147], Markov Chain Monte Carlo
(MCMC) overcomes the extremely complicated calculation of 5.3. Enabling technologies and tools in model fusion
high-dimensional integration. MCMC can also realize the dynamic
modeling where the sampling distribution varies as the simulation After model assembly, technologies for model fusion specify the
proceeds. With high modeling flexibility, Gaussian process [148] is coupling relationships and modes between digital twin models. For
available for regression, classification, feature extraction, and other such example, semantic information modeling technology [13] establishes

383
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

Fig. 12. Framework of technologies and tools in different digital twin modeling aspects.

the semantics and relationships of the data. The geometric information including MWorks [71,73], Simulink [23,25,77,140], Flexsim [91], and
from model assembly and the process information from the production Demo 3D [125]. MWorks is a platform for developing the multi-field
process documentation can be intelligently integrated to construct static engineering system. It can verify the applicability of parameter values
coupling between digital twin models. Besides the static model fusion by to real scenarios by a value comparison between the model variables in
semantic information modeling, Petri net technology [109] can repre­ simulation results and the measured metric variables. Simulink can
sent the correlations between synchronous and concurrent logical be­ examine simulation results and troubleshoot abnormal model behavior
haviors to realize the dynamic fusion between digital twin models. The via its graphical debugger and profiler. Meanwhile, Simulink’s model
definition of the unidirectional and bidirectional coupling patterns be­ analysis and diagnostic tools can verify the model’s consistency and
tween digital twin models in different domains [10] is also a techno­ pinpoint errors in the model. Notably, Simulink is more suitable for
logical concern for model fusion. In Fig. 12, Kalman filter method [45] is beginners to start with its graphical programming feature.
available for model fusion as well.
The fusion between digital twin models in different dimensions is 5.5. Enabling technologies and tools in model modification
also a concern for model fusion. As shown in Fig. 12, Functional Mock-
up Interface (FMI) [155] defines a set of interfaces to enable model After model verification, the digital twin model requires modifica­
fusion in different dimensions. For example, FMI can connect the vehicle tion with adequate technologies in a targeted manner so that the digital
model created by CarSim and the functional model created by Simulink twin model meets the application requirements. For instance, since the
for joint simulation. In addition to model fusion in different dimensions, element property of the digital twin model was expressed by the nu­
Mechatronic Concept Designer [127] seamlessly integrates disciplinary merical matrix, isoparametric transformation [106] and the Gaussian
knowledge of mechanics, electricity, and automation based on unified integration method [106] were used to adjust each element of the nu­
modeling language from the beginning of machine design. In this way, merical matrix to modify the model properties. Ultrafast crack growth
functional models can be created and fused in different disciplines. As an methodology [157] computed the equivalent constant amplitude
integrated simulation tool of wind turbines, OpenFAST [131] can fuse spectra to rapidly correct a mountain of crack growth realities on a
aero, hydro, servo, and elastic models for simulation analysis. Based on single processor. Other technologies for model modification are shown
OpenModelica, OMEdit [23] can provide interactive APIs which facili­ in Fig. 12, including single step numerical integration method [158] and
tate user-defined model extensions for model fusion. The current tools particle-tracking algorithm [159]. However, the universal technological
for model fusion in different dimensions and disciplines are limited to framework for model modification is lacking, causing increased
the engineering field, and tools for model fusion within and across other trial-and-error costs and even failure to derive the expected modification
fields are scarce. results.
In Fig. 12, tools including Abaqus [25], Simcenter Amesim software
5.4. Enabling technologies and tools in model verification [50], Ansys [77,160], are viable for model modification. Besides the
powerful finite element modeling capabilities, Abaqus and Ansys can
The purpose of model verification technologies is to verify whether prioritize the model variables with higher sensitivity by calculating their
and what degree the validity, accuracy, and applicability of the digital sensitivity towards the objective functions and state variables. Then, the
twin model running results address the physical production needs. It is modification iterations are carried out with appropriate modification
worth noting that the choice of model verification technologies should technologies to obtain a more accurate finite element model. As an
be tailored to the actual application scenario. The small displacement electromechanical system simulation platform, Simcenter Amesim
torsor theory [109] and Monte Carlo method [109] can identify unsure software enables performance verification and control modification for
model assembly positions. Hence, the assessment of the assembly ac­ the digital twin model in an iterative process. This iterative process can
curacy can be performed by avoiding assembly errors. Static and dy­ also be performed jointly with Simulink through interfaces. Individual
namic equations analysis technology [106] can identify and analyze model modification tool frequently targets only one model dimension or
user-defined element errors, thus laying the groundwork for model one field. For multidimensional digital twin models in multiple fields,
modification. Lattice Boltzmann Method [156] can verify the perme­ collaborative modification across multiple tools is desired.
ability stability of porous media by analyzing models at the pixel level.
However, the existing model verification technologies for the digital 5.6. Enabling technologies and tools in model management
twin model lack mature technological systems and generic theoretical
guidance, which are advancement directions for future research. As As the digital twin modeling progress, the model data accumulated
shown in Fig. 12, other model verification technologies in the literature and the modeling knowledge base gradually formed necessitate man­
reviewed include nonlinear least square method [23], Kalman filter agement for later reference and reuse. However, multidimensional and
method [45], and Ultrafast crack growth methodology [157]. multilevel digital twin models with multidisciplinary knowledge are
In model verification, the running results of the assembled or fused unfavorable for storage, exchange, and reuse. The model order reduc­
digital twin models are required to be compared with the actual values. tion technology [107] simplifies the complexity of the digital twin
For this reason, some software is available, as shown in Fig. 12, model and conserves its essential features. As a result, it facilitates model

384
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

storage and exchange while componentizing the model for easy reuse in However, no literature researches the hierarchy of the digital twin
other digital twin modeling aspects. Because requirements for the model model. Consequently, the correlations between objects in the same
vary with each modeling aspect, it may require model reuse and scenario, let alone different scenarios and even application fields, are
fine-tuning based on existing model data in the model library. Krylov not explicit. Since the hierarchy is underlying the myriad correlations
subspace method [107] can discretize the model state equations into between digital twin models, it is crucial to establish the hierarchical
several small sets of equations that simply need targeted modifications structure for the digital twin model. In future research, the macro­
to certain sub-equations for the model reuse and fine-tuning. The scopical hierarchical system for digital twin models in various applica­
Point-to-plane method [109] develops mapping relationships between tion fields and the microcosmic hierarchical structure for model
simpler point cloud models and tanglesome geometric elements of the dimensions within an individual one are required to be established.
digital twin model. Therefore, the simplification of digital twin model
management is achievable in the geometric dimension. The model 6.2. Discipline of the digital twin model
management technology should not be restricted to models and
modeling, but also extend to the services they provide. As shown in From traditional to cutting-edge disciplines, from engineering to
Fig. 12, the meta-information driven method [114] can standardize medical disciplines, the digital twin has embraced a vast disciplinary
model exchanging services by (de-) serializing dynamic model spectrum. Nevertheless, the disciplines involved in digital twin models
information. are isolated from each other in current research. The conversion of
In model management, the service towards digital twin model for disciplinary knowledge to accurate and efficient digital twin modeling is
users derived from model management can be obtained by tools. In insufficient. Thus, future works entail gap-filling between different
Fig. 12, Protégé [24,29] is available for model exchange which can disciplines for deeper interdisciplinary integration. The seamless disci­
import and store models in various formats such as XML, UML, and RDF. plinary knowledge transformation to satisfy the practical modeling
Besides, users can utilize OWL 2 accelerated by Protégé [24] for rapid needs is also a future research direction for the digital twin technology.
virtualization and parameter configuration. Remarkably, by represent­
ing models as module diagrams, Simulink can visually manage them 6.3. Dimension of the digital twin model
through an interactive graphical editor. Simscape of Simulink can divide
the model based on its hierarchical nature into equipment level, topol­ A comprehensive study of the digital twin model from the perspec­
ogy level, and multi-physics field level for efficient management. tive of the four model dimensions of the virtual model, i.e., geometry,
Meanwhile, Simscape provides rapid modification service of part pa­ physics, behavior, and rule. By portraying physical entities in these four
rameters to respond to changing mechanical modeling needs. However, dimensions, the digital twin model can be better characterized and the
the current model management tools are not well connected with tools corresponding research can be more targeted. But for other areas, the
in the other digital twin modeling aspects. The disconnection prevents analysis of the attributes and functions of the virtual model based on the
from reusing the model data and modeling knowledge the digital twin four model dimensions is absent. In some cases, the existing portrayal of
modeling generated and hinders the extension of services digital twin the digital twin model is relatively one-sided and cannot incorporate and
models provided. reflect all dimensions of its virtual model, which prevents the digital
twin model from realistically mapping the physical entity. At the same
5.7. Conclusions for technologies and tools above time, studies that address the fusion of these four model dimensions are
not available in the literature. Future research for digital twin models
The digital twin modeling theoretical system has been proposed and should be grounded on these four dimensions, to accurately mirror the
investigated from the six modeling aspects. However, the technological physical entities and fully utilize the digital twin model effectiveness.
standards and systems for each modeling aspect and the entire modeling The fusion of different model dimensions requires consideration as well.
process are still waiting to be elaborated to address the technological
issues involved in the actual modeling procedures. 6.4. Universality of the digital twin model
At each modeling aspect within the digital twin modeling theoretical
system, there are related software tools to perform the required works. By classifying the digital twin models into specific and generic ones
Nevertheless, nowadays, no tool integrates the functionalities of the according to their characteristics and scope of application, we can
previously described modeling software in each aspect to accomplish the enable digital twin models to deliver more dedicated and distinct ser­
all-process digital twin modeling. Therefore, in future research, the vices for production and life. So far, no article has elaborated on the
development of an all-process digital twin modeling tool needs to move criteria for categorizing digital twin models based on their universality.
forward under the guidance of the digital twin modeling theoretical For digital twin models with different universality, the dedicated and
system. all-purpose modeling methodologies, modeling theoretical systems, and
modeling tools remain unavailable. At present, only the five-
6. Observations and recommendations dimensional digital twin model is a general one applicable to all appli­
cation fields. Future research requires distinguishing the distinct nature
This work provides a thorough review of over 300 publications. of digital twin models with different universality. Accordingly, more
Therein, Fastidious data statistics and analytics on various properties of effective specific and generic digital twin models will be created by
the digital twin model and dissimilar modeling aspects of the digital employing appropriate modeling methodologies and tools.
twin modeling have been conducted. Accordingly, there are some ob­
servations and recommendations expounded as follows: 6.5. Functionality of the digital twin model

6.1. Hierarchy of the digital twin model Digital twin models can provide a wide range of effective functions
based on certain production requirements. However, the current liter­
In terms of the digital twin shop floor in manufacturing, diverse ature suffers from insufficient consideration of the multidimensional
production units can be hierarchized depending on their production Spatio-temporal scales and extrinsic factors for physical objects.
characteristics and functionalities. A reasonable hierarchical structure Consequently, the inconsiderateness risks undermining the robustness of
of digital twin model contributes to the clarified and efficient organi­ the functionality of the digital twin model during the real-time dynamic
zation, coordination, and management. In other application fields be­ operation and during the evolutionary process advancing over time. To
sides manufacturing, the digital twin technology has already blossomed. take full advantage of the digital twin utility, future research on digital

385
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

twin model functions is expected to thoroughly consider the practical­ Even though research for the digital twin is surging, current works on
ities of the corresponding physical objects and operational scenarios. the digital twin modeling are still in the infancy of rapid development.
Many imperatives should be resolved to refine the capability of the
digital twin modeling in portraying the digital twin models and
6.6. Modeling aspects of the digital twin modeling
improving their feasibility in practice. For instance, an integrated soft­
ware platform incorporating all modeling aspects is in urgent demand.
The six modeling aspects within the digital twin modeling theoretical
Accordingly, this paper serves as a reference for relevant researchers to
system present an essential reference for whole-process accurate digital
explore future orientations of the digital twin modeling.
twin modeling. Unfortunately, existing digital twin modeling studies
have incomplete and unbalanced coverage of these six modeling aspects.
The digital twin modeling in some fields is merely confined to model Declaration of Competing Interest
construction. Research on the modeling aspects besides model con­
struction still stagnated at a relatively rudimentary level without well- The authors declare that they have no known competing financial
developed systematic methodologies. Moreover, no literature em­ interests or personal relationships that could have appeared to influence
powers all the six modeling aspects with concrete application cases. In the work reported in this paper.
future work, we need to strengthen the study of the six modeling aspects
intensively and extensively. It is paramount to garner attention to the Acknowledgments
consistency, coherence, and collaboration between the modeling aspects
in their implementation. In this way, the connotation and applicability This work is financially supported in part by grants from Key Project
of the six modeling aspects can be continuously refined and expanded in of International (Regional) Cooperation and Exchange Programs of
future practices. NSFC (52120105008), in part by National Natural Science Foundation
of China (NSFC) under Grant 52005024, and in part by grants from
6.7. Technologies and tools for the digital twin modeling project No. G-YZ3N.

A wide variety of modeling technologies and tools are accessible References

today for various modeling aspects. Nevertheless, a majority of present
[1] Tao F, Qi Q. Make more digital twins. Nature 2019;573:490–1.
modeling technologies and tools specialize in model construction, [2] Tao F, Qi Q, Liu A, Kusiak A. Data-driven smart manufacturing. J Manuf Syst
leaving insufficient options for subsequent aspects. There is a dearth of a 2018;48:157–69.
technology and tool system or criterion to instruct the employment of [3] Grieves M, Vickers J. Digital twin: mitigating unpredictable, undesirable
emergent behavior in complex systems. In: Kahlen FJ, Flumerfelt S, Alves A,
the technologies and tools used in each modeling aspect. The correlation editors. Transdisciplinary Perspectives On Complex Systems. Cham: Springer;
between technologies and tools utilized in the same modeling aspect is 2017. p. 85–113.
ambiguous. Furthermore, the technologies or tools deployed in different [4] Tao F, Zhang H, Liu A, Nee AYC. Digital twin in industry: state-of-the-art. IEEE
Trans Ind Inform 2018;15(4):2405–15.
modeling aspects are separated from each other and without coherence [5] Gartner’s Top. 10 strategic technology trends for 2017. Available: 〈https://www.
between them. The separation and incoherence will lead to unnecessary gartner.com/smarterwithgartner/gartners-top-10-technology-trends-2017〉.
costs in time and money and increase modeling errors. It is urgent to [6] Gartner’s Top. 10 strategic technology trends for 2018. Available: 〈https://www.
gartner.com/smarterwithgartner/gartners-top-10-technology-trends-2018〉.
develop a modeling technology system and an integrated software
[7] Gartner’s Top. 10 strategic technology trends for 2019. Available: 〈https://www.
toolkit or platform, which incorporate all the six modeling aspects. As a gartner.com/smarterwithgartner/gartners-top-10-technology-trends-2019〉.
result, researchers can perform the digital twin modeling at minimal [8] 5 Trends Drive the Gartner Hype Cycle for Emerging Technologies. Available:
cost and enable maximum access to the corresponding service provided 〈https://www.gartner.com/smarterwithgartner/5-trends-drive-the-gartner-hype-
cycle-for-emerging-technologies-2020〉.
by high-fidelity digital twin models. [9] Tao F, Liu W, Zhang M, Hu T, Qi Q, Zhang H, et al. Five-dimension digital twin
model and its ten applications. CIMS 2019;25(01):1–18.
[10] Tao F, Zhang H, Qi Q, Xu J, Sun Z, Hu T, et al. Theory of digital twin modeling
7. Conclusions and future work and its application. CIMS 2021;27(1):1–16.
[11] Barricelli BR, Casiraghi E, Fogli DA. Survey on digital twin: definitions,
Since the first introduction of the terminology and the first applica­ characteristics, applications, and design implications. IEEE Access 2019;7:
167653–71.
tion in aerospace, the digital twin has been rapidly advancing from
[12] Lim KYH, Zheng P, Chen CH. A state-of-the-art survey of digital twin: techniques,
concept to application in recent years. Along with this, the research and engineering product lifecycle management and business innovation perspectives.
applications of digital twins are increasingly in-depth and specific. J Comput Inf Sci Eng 2020;31(6):1313–37.
Therefore, the digital twin model as the cornerstone of the digital twin [13] Errandonea I, Beltrán S, Arrizabalaga S. Digital twin for maintenance: a literature
review. Comput Ind 2020;123:103316.
technology and the digital twin modeling as the golden touch for this [14] Leng J, Wang D, Shen W, Li X, Liu Q, Chen X. Digital twins-based smart
cornerstone have received more and more attention. This paper provides manufacturing system design in industry 4.0: a review. J Manuf Syst 2021;60:
a systematic review of over 300 previous publications related to the 119–37.
[15] Melesse TY, Pasquale VD, Riemma S. Digital twin models in industrial operations:
digital twin modeling and its outcome, digital twin models. The primary a systematic literature review. Procedia Manuf 2020;42:267–72.
contributions of this review can be concluded as follows: [16] Wright L, Davidson S. How to tell the difference between a model and a digital
twin. Adv Model Simul Eng Sci 2020;7:13.
[17] Lechler T, Fuchs J, Sjarov M, et al. Introduction of a comprehensive structure
1) It delivers holistic and fine-grained statistics and analytics on the model for the digital twin in manufacturing. Proc 2020 IEEE 25th Int Conf Emerg
state-of-the-art research of the digital twin model in terms of the Technol Fact Autom 2020:1773–80.
application field, hierarchy, discipline, dimension, universality, and [18] Berri PC, Vedova MDLD. A review of simplified servovalve models for digital
twins of electrohydraulic actuators. Phys Conf Ser 2020;1603(1):012016.
functionality. [19] Rasheed A, San O, Kvamsdal T. Digital twin-values, challenges and enablers from
2) It reviews and analyzes the current research on the digital twin a modeling perspective. IEEE Access 2020;8:21980–2012.
modeling at different modeling aspects. In summary, present digital [20] Bordeleau F, Combemale B, Eramo R, et al. Towards model-driven digital twin
engineering: current opportunities and future challenges. Proc 2020 Int Conf Syst
twin modeling studies are mainly staying on model construction and
Model Manag 2020:43–54.
modeling integrity is a pressing issue. [21] Ríos J, Staudter G, Weber M, et al. Enabling the digital twin: a review of the
3) It summarizes the technologies and tools used in different modeling modelling of measurement uncertainty on data transfer standards and its
dimensions and modeling aspects and provides practical use cases as relationship with data from tests. Int J Prod Lifecycle Manag 2020;12(3):250–68.
[22] Andronas D, Kokotinis G, Makris S. On modelling and handling of flexible
well as general directions for enabling ones, which are feasible and materials: a review on digital twins and planning systems. Procedia CIRP 2021;
promising in corresponding digital twin modeling aspects. 97:447–52.

386
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

[23] Aivaliotis P, Georgoulias K, Arkouli Z, Makrisa S. Methodology for enabling [52] Shim C, Dang N, Lon S, Jeon C. Development of a bridge maintenance system for
digital twin using advanced physics-based modelling in predictive maintenance. prestressed concrete bridges using 3D digital twin model. Struct Infrastruct E
Procedia CIRP 2019;81:417–22. 2019;15(10):1319–32.
[24] Shangguan D, Chen L, Ding J. A hierarchical digital twin model framework for [53] Bridge assessment for PSC girder bridge using digital twins model. In: Ha-Minh C,
dynamic cyber-physical system design. Proc 2019 5th Int Conf Mechatron Robot Dao D, Benboudjema F, Derrible S, Huynh D, Tang A, editors. CIGOS 2019,
Eng 2019:123–9. Innovation for Sustainable Infrastructure. Singapore: Springer; 2019. p. 1241–6.
[25] Ke Y, Xu W, Yao B, Zhou Z, Pham DT. Digital twin-based energy modeling of [54] Jiang F, Ding Y, Song Y, Geng F, Wang Z. Digital twin-driven framework for
industrial robots. Proc Asian Simul Conf 2018 Methods Appl Model Simul fatigue life prediction of steel bridges using a probabilistic multiscale model:
Complex Syst 2018:978–81. application to segmental orthotropic steel deck specimen. Eng Struct 2021;241:
[26] Kuts V, Cherezova N, Sarkans M, Otto T. Digital twin: industrial robot kinematic 112461.
model integration to the virtual reality environment. J Mach Eng 2020;20(2): [55] Agostinelli S., Cumo F., Guidi G., Tomazzoli C. The potential of digital twin model
53–64. integrated with artificial intelligence systems. In: Proceedings of the 2020 IEEE
[27] Vrabič R, Erkoyuncu JA, Butala P, Roy R. Digital twins understanding the added International Conference on Environment and Electrical Engineering and IEEE
value of integrated models for through-life engineering services. Procedia Manuf Industrial and Commercial Power Systems Europe. 〈https://doi.org/10.11
2018;16:139–46. 09/EEEIC/ICPSEurope49358.2020.9160810〉.
[28] Fitzgerald J, Larsen PG, Pierce K. Multi-modelling and co-simulation in the [56] Yitmen I, Alizadehsalehi S, Akıner İ, Akıner ME. An adapted model of cognitive
engineering of cyber-physical systems: towards the digital twin. In: Beek M, digital twins for building lifecycle management. Appl Sci 2021;11(9):4276.
Fantechi A, Semini L, editors. From Software Engineering to Formal Methods and [57] Agouzoul A, Tabaa M, Chegari B, Simeu E, Dandache A, Alami K. Towards a
Tools, and Back. Cham: Springer; 2019. p. 40–55. digital twin model for building energy management: case of Morocco. Procedia
[29] Li X, He B, Zhou Y, Li G. Multisource model-driven digital twin system of robotic Comput Sci 2021;184:404–10.
assembly. IEEE Syst J 2020;99:1–10. [58] Heaton J, Parlikad AK. Asset information model to support the adoption of a
[30] Song S, Jang Y. Construction of digital twin geotechnical resistance model for digital twin: west Cambridge case study. IFAC-Pap 2020;53:366–71.
liquefaction risk evaluation. Proc 2018 2nd Int Symp Comput Sci Intell Control [59] Austin M, Delgoshaei P, Coelho M, Heidarinejad M. Architecting smart city digital
2018:1–6. twins combined semantic model and machine learning approach. J Manag Eng
[31] Angjeliu G, Coronelli D, Cardani G. Development of the simulation model for 2020;36(4):04020026.
digital twin applications in historical masonry buildings: the integration between [60] Yan J., Zlatanova S., Aleksandrov M., Diakite A.A., Pettit C. Integration of 3D
numerical and experimental reality. Comput Struct 2020;238:106282. objects and terrain for 3D modelling supporting the digital twin. ISPRS Ann
[32] Santagati C, Russa FML. Historical sentient – building information model: a Photogramm Remote Sens Spatial Inf Sci 2019;IV(4/W8), pp. 147–154.
digital twin for the management of museum collections in historical architectures. [61] Ham Y. Participatory sensing and digital twin city: updating virtual city models
The International Archives of the Photogrammetry. Remote Sens Spat Inf Sci for enhanced risk-informed decision-making. J Manag Eng 2020;36(3):04020005.
2020. XLIII-B4. [62] Kim J, Kim H, Ham Y. Mapping local vulnerabilities into a 3D city model through
[33] Wang P, Yang M, Zhu J, Ju R, Li G. Dynamic data driven modeling and simulation social sensing and the CAVE system toward digital twin city. Proc ASCE Int Conf
method for digital twin. J Syst Eng Electron 2020;42(12):2779–86. Comput Civ Eng 2019. https://doi.org/10.1061/9780784482445.058.
[34] Kuzmichev V, Moskvin A, Moskvina M. Modeling of digital twins of historical [63] Antonova DP, Ilieva S. Digital twin modeling of smart cities. In: Ahram T, et al.,
fashionable bodies. Proc High Educ Inst 2020:144–50. editors. Human Interaction, Emerging Technologies and Future Applications III;
[35] Elkarii M, Bouallou C, Ratnani A. Towards modelling a diphasic flow using the 2021. p. 384–90.
CFD technique to achieve a digital twin of a phosphate slurry piping process. [64] Kanak A., Ugur N., Ergun S. A visionary model on blockchain-based
Chem Eng Trans 2020;81:757–62. accountability for secure and collaborative digital twin environments. In:
[36] Khakimov RA, Shcherbo NS. Development and creation a model of a digital twin Proceedings of the 2019 IEEE International Conference on Systems, Man and
of the cubepart rectification installation for the separation of a binary water- Cybernetics. 〈https://doi.org/10.1109/SMC.2019.8914304〉.
alcohol mixture. IOP Conf Ser: Mater Sci Eng 2018;450:062006. [65] Ilya N, Ivan Z, Dmitry P, Polina K, Anna M, Ravil M, et al. Methods of
[37] Aversano G, Ferrarotti M, Parente A. Digital twin of a combustion furnace computational modeling of coronary heart vessels for its digital twin. MATEC
operating in flameless conditions: reduced-order model development from CFD Web Conf 2018;172:01009.
simulations. Proc Combust Inst 2021:5373–81. [66] Sack K., Blauer J., Campbell M., Swenson D. Creating a digital twin to investigate
[38] Zakrajsek A.J., Mall S. The development and use of a digital twin model for tire AV block: in-sights from a validated electromechanical full-heart model. In:
touchdown health monitoring. In: Proceedings of the 2017 58th AIAA/ASCE/ Proceedings of 2020 Computing in Cardiology Conference.
AHS/ASC Structures, Structural Dynamics, and Materials Conference. [67] Mazumder O, Roy D, Bhattacharya S, Sinha A, Pal A. Synthetic PPG generation
[39] Liu S, Bao J, Lu Y, Li J, Lu S, Sun X. Digital twin modeling method based on from haemodynamic model with baroreflex autoregulation: a digital twin of
biomimicry for machining aerospace components. J Manuf Syst 2021;58:180–95. cardiovascular system. Proc 2019 41st Annu Int Conf IEEE Eng Med Biol Soc
[40] Liu S, Lu Y, Li J, Song D, Sun X, Bao J. Multi-scale evolution mechanism and 2019:5024–9.
knowledge construction of a digital twin mimic model. Robot Cim-Int Manuf [68] Corral-Acero J, Margara F, Marciniak M, Rodero C, Loncaric F, Feng Y, et al. The
2021;71:102123. ‘Digital Twin’ to enable the vision of precision cardiology. Eur Heart J 2020;41
[41] Liu Z, Chen W, Zhang C. Data super-network fault prediction model and (48):4556–64.
maintenance strategy for mechanical product based on digital twin. IEEE Access [69] Chaudhari A., Ramanathan N. Digital twins for healthcare: articulating their
2019;7:177284–96. value through a deeper understanding of the patient journey. Available:
[42] Laukotka Hanna F, Schwede M, Dieter Krause L. Use of digital twins overarching 〈https://www.luxresearchinc.com/blog/digital-twins-for-healthcare-articulating
life cycle phases: Model-based product family development using the aircraft -their-value-through-a-deeper-understanding-of-the-patient-journey〉.
cabins as example. Z fu¨r Wirtsch Fabr 2020;115:101–4. [70] Safavi K., Thompson A., Kalis B., McHugh J. Accenture digital health technology
[43] Bare Z, Gaertig E, Krebs J, Arndt C, Gensel A. A note on leakage jet forces: vision 2021. Available: 〈https://www.accenture.com/us-en/insights/health/
application in the modelling of digital twins of hydraulic valves. Int J Fluid Power accenture-digital-health-technology-vision-2021〉.
2021;22(1):113–46. [71] Luo W, Hu T, Zhang C, Wei Y. Digital twin for CNC machine tool: modeling and
[44] Gou N, Cheng K, Huo D. Multiscale modelling and analysis for design and using strategy. J Amb Intel Hum Comp 2018;10:1129–40.
development of a high-precision aerostatic bearing slideway and its digital twin. [72] Lu Y, Liu C, Wang K, Huang H, Xu X. Digital twin-driven smart manufacturing:
Machines 2021;9(5):85–100. connotation, reference model, applications and research issues. Robot Cim-Int
[45] Liu M., Wang B., Liu D. A digital twin modeling method for turbofan engine real- Manuf 2020;61:101837.
time test data analysis and performance monitoring. In: Proceedings of the 2020 [73] Luo W, Hu T, Zhu W, Tao F. Digital twin modeling method for CNC machine tool.
11th International Conference on Prognostics and System Health Management. Proc 2018 IEEE 15th Int Conf Netw Sens Control 2018:613–8.
[46] Kapteyn M.G., Knezevic D.J., Willcox K. Toward predictive digital twins via [74] Scaglioni B, Ferretti G. Towards digital twins through object-oriented modelling-a
component-based reduced-order models and interpretable machine learning. machine tool case study. Proc 2018 9th Vienna Int Conf Math Model 2018:613–8.
AIAA Scitech 2020 Forum. 〈https://doi.org/10.2514/6.2020–0418〉. [75] Liu C., Hong X., Zhu Z., Xu X. Machine tool digital twin: modelling methodology
[47] Kapteyn MG, Knezevic DJ, Huynh DBP, Tran M, Willcox KE. Data-driven physics- and applications. In: Proceedings of the 2018 48th International Conference on
based digital twins via a library of component-based reduced-order models. Int J Computers and Industrial Engineering.
Numer Methods Eng 2020:1–18. [76] Wei Y, Hu T, Zhou T, Ye Y, Luo W. Consistency retention method for CNC
[48] Kapteyn MG, Willcox KE. Predictive digital twins: where dynamic data-driven machine tool digital twin model. J Manuf Syst 2021;58:313–22.
learning meets physics-based modeling. Proc 2020 Int Conf Dyn Data Driven Appl [77] Zhao P, Sun B. Adaptive modification of digital twin model of CNC machine tools
Syst 2020:3–7. coordinately driven by mechanism model and data model. Phys: Conf Ser 2021;
[49] Wang J, Xiang Y, He Z. Models and implementation of digital twin based 1875:012003.
spacecraft system engineering. CIMS 2019;25(06):1348–60. [78] Liu J., Yu D., Bi X., Hu Y., Yu H., Li B. The research of ontology-based digital twin
[50] Mena D.J., Pluchart S., Mouvand S., Broca O. Rocket engine digital twin – machine tool modeling. In: Proceedings of the 2020 IEEE 6th International
modeling and simulation benefits. AIAA Propulsion and Energy 2019 Forum. Conference on Computer and Communications.
〈https://doi.org/10.2514/6.2019–4114〉. [79] Yu B. Optimizing machining time and oscillation based on digital twin model of
[51] Kim J, Ham Y. Vision-based analysis of utility poles using drones and digital twin tool center point. In: Proceedings of the 2020 IEEE Eurasia Conference on IOT,
modeling in the context of power distribution infrastructure systems. Proc Constr Communication and Engineering.
Res Congr 2020. https://doi.org/10.1061/9780784482865.101. [80] Xie N, Kou R, Yao Y. Tool condition prognostic model based on digital twin
system. Procedia CIRP 2020;93:1502–7.

387
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

[81] Xu M, Zhu T, Zhang JZH. Automated construction of neural network potential [110] Zhang T, Liu X, Luo Z, Dong F, Jiang Y. Time series behavioral modeling with
energy surface: the enhanced self-organizing incremental neural network deep digital twin for Internet of Vehicles. Eurasip J Wirel Commun Netw 2019:271.
potential method. J Chem Inf Model 2021;61(11):5425–37. [111] Gobeawan L, Wise DJ, Wong ST, Yee ATK, Lim CW, Su Y. Tree species modelling
[82] Hou Z, He W. Modeling and control of digital twin-based aircraft assembly state for digital twin cities. In: Gavrilova ML, Tan CK, editors. Transactions on
inspection robot. CIMS 2021;27(04):981–9. Computational Science XXXVIII. Berlin: Springer; 2021. p. 17–35.
[83] Zhu T, Wu C, Zhou Y, Zhang P, Zhang X. Construction of extended collision [112] Chakraborti A., Heininen A., Koskinen K.T., Lämsä V. Digital twin: multi-
warning system digital twin conceptual scheme based on TRIZ functional dimensional model reduction method for performance optimization of the virtual
modeling. CIMS 2021;27(02):337–51. entity. Procedia CIRP;93:240–245.
[84] Li X, Hou X, Yang M, Wang L, Wang Y, Li X. Construction and application of CMF [113] Lehner D., Wolny S., Mazak-Huemer A., Wimmer M. Towards a reference
design service model for industrial products driven by digital twins. CIMS 2021; architecture for leveraging model repositories for digital twins. In: Proceedings of
27(02):307–27. the 2020 25th IEEE International Conference on Emerging Technologies and
[85] Meng Z, Tang T, Wei G, Yuan L. Digital twin based comfort scenario modeling of Factory Automation. 〈https://doi.org/10.1109/ETFA46521.2020.9212109〉.
ATO controlled train. J Phys: Conf Ser 2020;1654:012071. [114] Priggemeyer M., Roßmann J. Meta-information driven modeling of services for
[86] Li M., Ma Y., Yin Z., Wang C. Structural design of digital twin laboratory model experimentable digital twins. In Proceedings of the 2020 IEEE International
based on instruments sharing platform. In: Proceedings of the 2022 34th Chinese Systems Conference. 〈https://doi.org/10.1109/SysCon47679.2020.9275877〉.
Control and Decision Conference. 〈https://doi.org/10.26914/c.cnkihy.2020 [115] Singh S, Shehab E, Higgins N, Fowler K, Reynolds D, Erkoyuncu JA, et al. Data
.033937〉. management for developing digital twin ontology model. Proc Inst Mech Eng B: J
[87] Lohtander M, Ahonen N, Lanz M, Ratava J, Kaakkunen J. Micro manufacturing Eng Manuf 2020;235(14):2323–37.
unit and the corresponding 3D-model for the digital twin. Procedia Manuf 2018; [116] Müller M., Talkheshtani B.A., Jazdi N., Weyrich M., Rosen R., Wehrstedt J.C.
25:55–61. Information models with intelligent evaluation for the digital twin. In:
[88] Andryushkevich S.K., Kovalyov S.P., Nefedov E. Composition and application of Proceedings of VDI-Kongress Automation 2020. https://doi.org/ 10.13140/
power system digital twins based on ontological modeling. In: Proceedings of the RG.2.2.19227.49444/2.
2019 IEEE 17th International Conference on Industrial Informatics. 〈https://doi. [117] Werner H, Jerome F, Simon S. Data-consistent model engineering of production-
org/10.1109/INDIN41052.2019.8972267〉. centric digital twins conceptualization of a data structure and model management
[89] Rasslkin A., Rjabtikov V., Vaimann T., Kallaste A., Partyshev A. Digital twin of an approach for digital twins based on semantic web technologies. 19 Leitkongress
electrical motor based on empirical performance model. In: Proceedings of the der Mess- und Autom: Seamless Converg Autom IT: Autom 2018:183–96.
International Conference on Electrical Power Drive Systems. 〈https://doi.org/1 [118] Zhang L, Zhou L, Horn BKP. Building a right digital twin with model engineering.
0.1109/ICEPDS47235.2020.9249366〉. J Manuf Syst 2021;59:151–64.
[90] Brylina O.G., Kuzmina N.N.; Osintsev K.V. Modeling as the foundation of digital [119] Qi Q, Tao F, Hu T, Anwer N, Liu A, Wei Y, et al. Enabling technologies and tools
twins. In: Proceedings of the 2020 Global Smart Industry Conference. for digital twin. J Manuf Syst 2019;59:3–21.
〈https://doi.org/10.1109/GloSIC50886.2020.9267812〉. [120] Gopalakrishnan S, Hartman NW, Sangid MD. Model‑based feature information
[91] Zhang Q., Zhang X., Xu W., Liu A., Zhou Z., Pham D.T. Modeling of digital twin network (MFIN): a digital twin framework to integrate location‑specifc material
workshop based on perception data. In: Proceedings of International Conference behavior within component design, manufacturing, and performance analysis.
on Intelligent Robotics and Applications;2017, p. 3–14. Integ Mate Manuf Innov 2020;9:394–409.
[92] Verboven P, Defraeye T, Datta AK, Nicolai B. Digital twins of food process [121] Zheng S, Zhang Y, Xie W, Fan H, Wang Q. Aircraft final assembly line modeling
operations: the next step for food process models? Curr Opin. Food Sci 2020;35: based on digital twin. J Zhejiang Univ Eng Sci 2021;55:05.
79–87. [122] Wang B., Yuan L., Yu X., Ou L. Construction and optimization of digital twin
[93] Alizadehsalehi S, Yitmen I. Digital twin-based progress monitoring management model for hardware production line. In: Proceedings of IECON 2020 the 46th
model through reality capture to extended reality technologies (DRX). Smart Annual Conference of the IEEE Industrial Electronics Society. 〈https://doi.org/1
Sustain. Built Environ 2021. https://doi.org/10.1108/SASBE-01-2021-0016. 0.1109/IECON43393.2020.9254967〉.
[94] Zhang X, Liu L, Feng B. Tool wear online monitoring method based on DT and [123] Lozoya-Santos J.D.J., Ferreira A.C., Martínez A.V. Digital twin model proposal for
SSAE-PHMM. J Comput Inf Sci Eng 2021. https://doi.org/10.1115/1.4050531. an automotive chassis. In: Proceedings of the 2019 30th International Conference
[95] Lia M, Nan L. Feature-preserving 3D mesh simplification for urban buildings. on Adaptive Structures and Technologies.
ISPRS J Photo Remote Sens 2021;173:135–50. [124] Minos-Stensrud M, Haakstad OH, Sakseid O, Westby B, Alcocer A. Towards
[96] Ruan H, Xie H, Wang J, Liao J, Sun L, Gao M, et al. Numerical investigation and automated 3D reconstruction in SME factories and digital twin model generation.
comparative analysis of nanofluid cooling enhancement for TEG and TEC Proc 2018 Int Conf Control Robot Syst Soc 2018:1777–81.
systems. Case Stud. Therm Eng 2021;27:101331. [125] Ding K, Zhang X, Zhou G, Wang C, Yang H, Zhang F, et al. Digital twin-based
[97] Boulfani F., Gendre X., Ruiz-Gazen A., Salvignol M. Anomaly detection for multi-dimensional and multi-scale modeling of smart manufacturing spaces. CIMS
aircraft electrical generator using machine learning in a functional data 2019;25(06):1491–504.
framework. In: Proceedings of the 2020 Global Congress on Electrical [126] Gu Z, Gao T. A visual particle system based on mechanism model data in digital
Engineering. 〈https://doi.org/10.23919/GC-ElecEng48342.2020.9285984〉. twin. Proc Int Conf Innov Mob Internet Serv Ubiquitous Comput 2020:183–93.
[98] Luo W, Hu T, Yee Y, Zhang C, Wei Y. A hybrid predictive maintenance approach [127] Osinde NO, Byiringiro JB, Gichane MM, Smajic H. Process modelling of
for CNC machine tool driven by digital twin. Robot Comput Integr Manuf 2020; geothermal drilling system using digital twin for real-time monitoring and
65:101974. control. Designs 2019;3(3):45.
[99] Scheffel RM, Fröhlich AA, Silvestri M. Automated fault detection for additive [128] Vachálek J, Imiová D, Vaek P, Fika I, Imovec M. Design and implementation of
manufacturing using vibration sensors. Int J Comput Integr Manuf 2021;34(5): universal cyber-physical model for testing logistic control algorithms of
500–14. production line’s digital twin by using color sensor. Sensors 2021;21(5):1842.
[100] Lermer M., Reich C. Creation of digital twins by combining fuzzy rules with [129] Fonseca I.A.and Gaspar H.M. Fundamentals of digital twins applied to a plastic
artificial neural networks. In: Proceedings of the 2019 45th Annual Conference of toy boat and a ship scale model. In: Proceedings of the 2020 34th International
the IEEE Industrial Electronics Society. https://doi.org/10.1109/IECON.2019.89 ECMS Conference on Modelling and Simulation. 〈https://doi.org/10.714
26914. 8/2020–0207〉.
[101] Luo F, Qiang L, Ding Z. A digital twin-oriented lightweight approach for 3D [130] Bolotov MA, Pechenin VA, Ruzanov NV, Grachev IA. Information model and
assemblies. Machines 2021;9(10):231. software architecture for the implementation of the digital twin of the turbine
[102] Zhang W, Ma M, Li H. Method for assembly interference detection based on MRT. rotor. J Phys Conf Ser 2019;1368(5):052013.
CIMS 2017;23(06):1253–63. [131] Chetan M, Yao S, Griffith DT. Multi-fidelity digital twin structural model for a
[103] Liu C, Jiang P, Jiang W. Web-based digital twin modeling and remote control of sub-scale downwind wind turbine rotor blade. Wind Energy 2021. https://doi.
cyber-physical production systems. Robot Comput Integr Manuf 2020;64:101956. org/10.1002/we.2636.
[104] Schluse M., Atorf L., Rossmann J. Experimentable digital twins for model-based [132] Mukherjee V., Martinovski T., Szucs A., Westerlund J., Belahcen A. Improved
systems engineering and simulation-based development. In: Proceedings of the analytical model of induction machine for digital twin application. In:
2017 Annual IEEE International Systems Conference. 〈https://doi.org/10.1109/ Proceedings of 2020 International Conference on Electrical Machines. 〈https://d
SYSCON.2017.7934796〉. oi.org/10.1109/ICEM49940.2020.9270916〉.
[105] Hochhalter JD. On the effects of modeling as-manufactured geometry: toward [133] Phelps C., Tessler A. iFEM for marine structure digital twins – effective modeling
digital twin. Int J Aerosp Eng 2014;439278:1–10. strategies. SNAME Marit Conv 2020.
[106] Zhang C, Sun Q, Sun W, Mu X, Wang Y. A construction method of digital twin [134] Lai X, Wang S, Guo Z, Zhang C, Song X. Designing a shape–performance
model for contact characteristics of assembly interface. Int J Adv Manuf Technol integrated digital twin based on multiple models and dynamic data: a boom crane
2021;113:2685–99. example. J Mech Des 2021;143(7):1–15.
[107] Hartmann D, Herz M, Wever U. Model order reduction a key technology for [135] Jatale A. Reduced order models (ROMs) for simulation democratization and
digital twins. In: Keiper W, Milde A, Volkwein S, editors. Reduced-Order digital twins. In Proceedings of the 2018 10th AIChE Southwest Process
Modeling (ROM) for Simulation and Optimization. Cham: Springer; 2018. Technology Conference.
p. 167–79. [136] Wagg D, Worden K, Barthorpe R, Gardner P. Digital twins: State-of-the-art and
[108] Schroeder GN, Steinmetz C, Rodrigues RN, Henriques RVB, Pereira CE. future directions for modelling and simulation in engineering dynamics
A methodology for digital twin modeling and deployment for industry 4.0. Proc applications. ASME J Risk Uncertain Part B 2020;6(3):030901.
IEEE 2020:1–12. [137] Aksenov A.A., Gileva L.V., Danilishin A.M., Kozhukhov Y.V.. Analysis of CFD
[109] Wang K, Liu D, Liu Z, Wang Q, Tan J. An assembly precision analysis method modeling accuracy in centrifugal compressor stages digital twins. In: Proceedings
based on a general part digital twin model. Robot Comput Integr Manuf 2021;68: of 2020 International Multi-Conference on Industrial Engineering and Modern
102089. Technologies. 〈https://doi.org/10.1109/FarEastCon50210.2020.9271363〉.

388
F. Tao et al. Journal of Manufacturing Systems 64 (2022) 372–389

[138] Wang R, Zhou X, Dong L, Wen Y, Tan R, Chen L, et al. Kalibre: knowledge-based [150] Dai S, Zhao G, Yu Y, Zheng P, Wang W. Ontology-based information modeling
neural surrogate model calibration for data center digital twins. BuildSys 2020. method for digital twin creation of as-fabricated machining parts. Robot Comput
https://doi.org/10.1145/3408308.3427982. Integr Manuf 2021;72:102173.
[139] Krukovskyi PG, Diadiushko YV, Garin VO, Tryfonov OV. CFD model as a digital [151] Bao Q, Zhao G, Yu Y, Dai S. Product information units modeling oriented to
twin of the radiation state of the new safe confinement of the chernobyl NPP. digital twin. IOP Conf Ser Earth Environ Sci 2020;726:012014.
Probl Sci Technol 2020;4(128):54–62. [152] Zhang S, Kang C, Liu Z, Wu J, Ma C. A product quality monitor model with the
[140] Song X., Jiang T., Schlegel S., Westermann D. Investigation of inventive tuning digital twin model and the stacked auto encoder. IEEE Access 2020;8:113826–36.
algorithm for the realization of digital twins of inverter model in inverter- [153] Abadi M, Barham P, Chen J, Chen Z, Davis A, Dean J, et al. Tensorflow: a system
dominated power distribution grid. In: Proceedings of NEIS 2020 Conference on for large-scale machine learning. Proc 12th USENIX Conf Oper Syst Des
Sustainable Energy Supply and Energy Storage Systems. 〈https://doi.org/10.11 Implement 2016:265–83.
09/ISGTEurope.2019.8905551〉. [154] Liu Q, Leng J, Yan D, Zhang D, Chen X. Digital twin-based designing of the
[141] Singh S, Weeber M, Birke K. Advancing digital twin implementation: a toolbox for configuration, motion, control, and optimization model of a flow-type smart
modelling and simulation. Procedia CIRP 2021;99:567–72. manufacturing system. J Manuf Syst 2021;58:52–64.
[142] Creation of a digital twin model, redesign of plant structure and new fuzzy logic [155] Perabo F., Park D., Zadeh M.K., Smogeli Y., Jamt L. Digital twin modelling of ship
controller for the cooling system of a railway locomotive. In: Saponara S, power and propulsion systems: application of the open simulation platform (OSP).
Gloria AD, editors. Applications in Electronics Pervading Industry, Environment In: Proceedings of the 2020 IEEE 29th International Symposium on Industrial
and Society. Switzerland: Springer; 2021. p. 125–35. Electronics. 〈https://doi.org/10.1109/ISIE45063.2020.9152218〉.
[143] Tarkhov DA, Malykhina GF. Neural network modelling methods for creating [156] Xiao H, He L, Li J, Zou C, Shao C. Permeability prediction for porous sandstone
digital twins of real objects. J Phys Conf Ser 2019;1236(1):012056. using digital twin modeling technology and Lattice Boltzmann method. Int J Rock
[144] Barykin SY, Kapustina IV, Sergeev SM, Kalinina OV, Vilken VV, Elena P, et al. Mech Min Sci 2021;142:104695.
Developing the physical distribution digital twin model within the trade network. [157] Ocampo J, Millwater H, Crosby N, Gamble B, Hurst C, Reyer M. An ultrafast crack
Acad Strateg Manag J 2021;20:1–18. growth lifing model to support digital twin, virtual testing, and probabilistic
[145] Bao J, Guo D, Li J, Zhang J. The modelling and operations for the digital twin in damage tolerance applications. In: Niepokolczycki A, Komorowski J, editors.
the context of manufacturing. Enterp Inf Syst 2019;13(4):534–56. Structural Integrity in the Age of Additive Manufacturing. Cham: Springer; 2019.
[146] Ritto TG, Rochinha FA. Digital twin, physics-based model, and machine learning p. 145–58.
applied to damage detection in structures. Mech Syst Signal Process 2021;155: [158] Orlov S, Burkovski L. Digital twin of continuously variable transmission for
107614. predictive modeling of dynamics and performance optimization. In:
[147] Ghosh AK, Ullah AS, Kubo A. Hidden Markov model-based digital twin Schaumburg H, Korablev V, Ungvari L, editors. Technological Transformation: A
construction for futuristic manufacturing systems. AI EDAM 2019;33(3):317–31. New Role For Human, Machines And Management. Cham: Springer; 2020.
[148] Liu J., Dong Y., Liu Y., Li P., Wang T.. Prediction study of the heavy vehicle p. 89–107.
driving state based on digital twin model. In: Proceedings of the 2021 IEEE [159] Prifling B, Neumann M, Hlushkou D, Kübel C, Schmidt V. Generating digital twins
International Conference on Power Electronics, Computer Applications. of mesoporous silica by graph-based stochastic microstructure modeling. Comput
〈https://doi.org/10.1109/ICPECA51329.2021.9362618〉. Mater Sci 2021;187:109934.
[149] Reed S, Lfstrand M, Andrews J. Modelling stochastic behaviour in simulation [160] Zou R, Liang X, Chen Q, Wang M, Zaghloul M, Lan H. A digital twin approach to
digital twins through neural nets. J Simul 2021. https://doi.org/10.1080/ study additive manufacturing processing using embedded optical fiber sensors
17477778.2021.1874844. and numerical modeling. J Light Technol 2020;38(22):6402–11.

389