Robotics and Computer Integrated Manufacturing 61 (2020) 101837

Contents lists available at ScienceDirect

Robotics and Computer Integrated Manufacturing

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

Review

Digital Twin-driven smart manufacturing: Connotation, reference model, T

applications and research issues
Yuqian Lua, Chao Liub, Kevin I-Kai Wangc, Huiyue Huanga, Xun Xua,
⁎

a
Department of Mechanical Engineering, The University of Auckland, New Zealand
b
School of Engineering, Cardiff University, United Kingdom
c
Department of Electrical, Computer and Software Engineering, The University of Auckland, New Zealand

ARTICLE INFO ABSTRACT

Keywords: This paper reviews the recent development of Digital Twin technologies in manufacturing systems and processes,
Smart manufacturing to analyze the connotation, application scenarios, and research issues of Digital Twin-driven smart manu-
Digital Twin facturing in the context of Industry 4.0. To understand Digital Twin and its future potential in manufacturing, we
Industry 4.0 summarized the definition and state-of-the-art development outcomes of Digital Twin. Existing technologies for
Cyber-physical System
developing a Digital Twin for smart manufacturing are reviewed under a Digital Twin reference model to sys-
Big Data
tematize the development methodology for Digital Twin. Representative applications are reviewed with a focus
Standard
on the alignment with the proposed reference model. Outstanding research issues of developing Digital Twins for
smart manufacturing are identified at the end of the paper.

1. Introduction (see Section 2). This is followed by an in-depth discussion on the con-
notation of Digital Twin-driven smart manufacturing in Section 3,
Digital Twin has gained significant impetus as a breakthrough highlighting how Digital Twin will transform the future manufacturing
technological development that has the potential to transform the landscape. Section 4 details a Digital Twin reference model and en-
landscape of manufacturing today and tomorrow [1]. Digital Twin [2], abling technologies for developing a Digital Twin-driven smart manu-
acting as a mirror of the real world, provides a means of simulating, facturing solution. An overview of existing Digital Twin applications
predicting and optimizing physical manufacturing systems and pro- and some typical application scenarios are presented in Section 5.
cesses. Using Digital Twin, together with intelligent algorithms, orga- Section 6 discusses the critical research issues for future research.
nizations can achieve data-driven operation monitoring and optimiza- Section 7 concludes the research work.
tion [3], develop innovative product and services [4], and diversify
value creation and business models [5]. 2. Digital Twin overview
Though studies have reported the potential application scenarios of
Digital Twin in manufacturing, we identified that current approaches to This section traces the history of the Digital Twin concept, clarifies
the implementation of Digital Twin in manufacturing lack a thorough its relations with several other tropical concepts in the manufacturing
understanding of Digital Twin concept, framework, and development domain, summarizes its research and development progress, and
methods, which impedes the development of genuine Digital Twin highlights the research gaps.
applications for smart manufacturing. In this study, we discussed the
connotations of Digital Twin-driven smart manufacturing in the context 2.1. Definition
of Industry 4.0. The objectives and the contributions of this paper are to
provide comprehensive discussions on the impact, reference model, Digital Twin was conceived in [6] as a method to predict the
application scenarios and research issues of Digital Twin for achieving structural behavior of an aircraft by analyzing and simulating the air-
smart manufacturing. craft's behavior on its digital model in 2011. A year later, NASA defined
The remainder of the paper starts with tracing the vision of Digital Digital Twin as “an integrated multi-physics, multi-scale, probabilistic
Twin and the development to date based on studies from the literature simulation of a vehicle or system that uses the best available physical

Conflict of interest. None.

⁎
Corresponding author.
E-mail addresses: [email protected], [email protected] (X. Xu).

https://doi.org/10.1016/j.rcim.2019.101837
Received 29 April 2019; Received in revised form 6 July 2019; Accepted 7 July 2019
Available online 08 August 2019
0736-5845/ © 2019 Elsevier Ltd. All rights reserved.
Y. Lu, et al. Robotics and Computer Integrated Manufacturing 61 (2020) 101837

models, sensor updates, fleet history, and so forth, to mirror the life of through which data can flow between themselves. The connections are
its flying twin [2].” There was limited exploration since then, until made possible by the secure implementation of computer networks, the
2015, the explosion of machine learning, wireless communication and Internet, and communication protocols. However, despite the con-
cloud computing boosted the research activities on Digital Twin. Sev- nectivity, IoT does not include the idea of digital models in the cyber-
eral definitions of Digital Twin came out afterwards. For instance, Di- space. The IoT is the infrastructure in the physical space for connecting
gital Twin was seen as the next generation of simulation [7]. Tao and physical assets [11].
Zhang believed that Digital Twin is a method of achieving the con-
vergence between physical and virtual spaces [8]. 2.3. Research activities
Now, Digital Twin has evolved into a broader concept that refers to
a virtual representation of manufacturing elements such as personnel, The research activities on Digital Twin have gained hyper growth
products, assets and process definitions, a living model that con- during the past three years driven by the strategic implementation of
tinuously updates and changes as the physical counterpart changes to Industry 4.0 from world-leading research organizations and tech giants.
represent status, working conditions, product geometries and resource
states in a synchronous manner [9]. The digital representation provides 2.3.1. Academic research outcomes
both the elements and the dynamics of how a physical ‘thing’ operates Digital Twin related engineering research is in its infancy with
and lives throughout its life cycle. significant growth during the past three years as shown in Fig. 2(a). The
number of publications on this topic in 2018 tripled that in 2017. A
2.2. Concept clarification large percentage of the research outcomes come from the US, Germany,
and China, who are leading the race to Industry 4.0. A small number of
There exist diverse viewpoints on the relationships between Digital researchers and research organizations contributed nearly 40% of the
Twin and other concepts, such as simulation, Cyber-physical Systems total number of articles on this topic.
(CPSs) and Internet of Things (IoT). Though these concepts are closely
related, they, by their nature, are different on the concept, core ele- 2.3.2. Industry research outcomes
ments, and application. Digital Twin has attracted strong interests from industry practi-
tioners too. The Digital Twin market is forecasted to reach $15.66
2.2.1. Digital Twin and simulation billion by 2023 at an annual growth rate of 37.87% according to a
A Digital Twin is a digital replica of a real-world ‘thing.’ While this market research in 2017 [13]. GE developed a Digital Twin platform –
looks close to simulation, Digital Twin is much more. A Digital Twin is a PREDIX that can better understand and predict asset performance [14].
high-fidelity representation of the operational dynamics of its physical SIEMENS's focus covers smart operations during the complete process
counterpart, enabled by near real-time synchronization between the of product design, production and operation [15]. ABB emphasizes on
cyberspace and physical space [10]. The operational dynamics are enabling data-driven decision makings [16]. Microsoft also geared up
critical elements of a Digital Twin because a twin's behavior is based on its Digital Twin product portfolio, providing a ubiquitous IoT platform
near real-time data coming from the actual physical counterpart. Si- for modeling and analyzing the interactions between people, spaces,
mulation focuses on what could happen in the real world (what-if sce- and devices [17]. Initiatives from these tech leaders have significantly
nario), but not what is currently happening. In the manufacturing pushed the boundary of Digital Twin for engineering applications.
context, a Digital Twin can be used for monitoring, control, diagnostics,
and prediction, other than just simulation [10]. 2.4. Research challenges

2.2.2. Digital Twin, CPS, and IoT Though some early adopters have demonstrated some applications
Though Digital Twin, CPS, and IoT all use networking and sensors, of Digital Twin for manufacturing, current implementation limitations
Digital Twin is a different but interrelated concept with CPS or IoT, as are (1) inadequate understanding of the connotation of Digital Twin-
shown in Fig. 1. driven smart manufacturing, with the current focus mostly on product
A CPS is characterized by a physical asset and its Digital Twin. In operation and maintenance, (2) the lack of reference models for Digital
contrast, a Digital Twin is limited to the digital model, not the twinning Twin, and (3) superficial knowledge of the research questions and
physical asset, though a Digital Twin cannot live without its twining challenges of Digital Twin, with current research outcomes mostly
asset in the physical space. In other words, Digital Twin represents the showing preliminary application examples.
prerequisite for the development of a CPS [12]. The sustainable development of Digital Twin-driven smart manu-
IoT refers to connections between a network of physical assets facturing needs critical analysis on the above aspects based on the de-
velopment trend of smart manufacturing, which the research in this
paper aims to address.

3. Digital Twin-driven smart manufacturing

Manufacturing is becoming smart at all levels from the physical

device, through factory management, to production networks, gaining
abilities to learn, configure and execute with cognitive intelligence.
This section outlines the trend of smart manufacturing and discusses the
connotation of Digital Twin-driven smart manufacturing, highlighting
the impact that Digital Twin may have for future manufacturing.

3.1. Smart manufacturing

Smart manufacturing is coined by several agencies, such as the

Department of Energy (DoE) and the National Institute of Standards and
Fig. 1. The relationship between Digital Twin, CPS and IoT (adapted from Technology (NIST) in the United States. According to Davis et al., smart
[11]). manufacturing is the dramatically intensified application of

2
Y. Lu, et al. Robotics and Computer Integrated Manufacturing 61 (2020) 101837

Fig. 2. Statistics from Scopus database (TITLE-ABS-KEY (“digital twin”) AND (LIMIT-TO (SUBJAERA, “ENGI”)), Date: 2 July 2019). (a) Document per year; (b)
Documents by author; (c) Documents by country/territory; (d) Documents by affiliation.

‘manufacturing intelligence’ throughout the manufacturing and supply Smart factories that are fully responsive to changes and demands
chain enterprise [18]. It comprises the real-time understanding, rea- from the factory, supply chain, and customer side can achieve batch-
soning, planning and management of all aspects of manufacturing size-of-1 production with high efficiency and flexibility. The ubi-
processes, facilitated by the pervasive use of advanced sensor-based quitous manufacturing intelligence in distributed factories and
data analytics, modeling, and simulation. NIST defines smart manu- production systems can sense, configure and collaborate by them-
facturing systems as “fully-integrated, collaborative manufacturing systems selves based on near real-time production status and demands,
that respond in real time to meet changing demands and conditions in the which therefore provides the required agility for producing highly
factory, in the supply network, and customer needs. [19]” personalized products.
In smart manufacturing, a physical ‘thing’ in a factory is connected
to the Industrial Internet via standard cyber gateways and abstracted as 3.2. Digital Twin for smart manufacturing
a Digital Twin in the cyberspace. Each Digital Twin in the cyberspace is
an abstraction of its counterpart in the physical world by reflecting its Digital Twin plays a pivot role in the vision of smart manufacturing.
physical status. The cyberspace stores and processes the streamed data It enables the shift from analyzing the past to predicting the future. The
from connected physical objects. These data are used to model, simulate live representation of reality via Digital Twins allows us to evolve from
and predict the status of each physical thing under dynamic working ex-post data gathering and analytics towards real-time and ex-ante
conditions. The pervasive use of smart technologies, such as Big Data business practices. Mirroring the vision of smart manufacturing in
Processing and Artificial Intelligence enables the extraction of manu- Fig. 3, Digital Twin can influence future manufacturing from the fol-
facturing intelligence at every single moment of manufacturing activ- lowing aspects.
ities. The collective intelligence in locally connected factories and the
cyberspace paves the way for some dramatic changes from the aspects • Digital Twin for manufacturing assets: A manufacturing asset can
of intra-business operation, inter-business collaboration and production be connected and abstracted to the cyberspace via its Digital Twin.
model, as shown in Fig. 3. Manufacturers can gain a clearer picture of real-world performance
and operating conditions of a manufacturing asset via near real-time
• Smart Production: Manufacturing systems augmented with cogni- data captured from the asset and make proactive optimal operation
tive intelligence [20] can take over more and more production jobs. decisions. With truthful information flowing from a manufacturing
Connected and self-organizing manufacturing systems will tackle asset, manufacturers can improve their situational awareness and
new manufacturing tasks with high efficiency and flexibility. The enhance operation resilience and flexibility, especially in the context
relationship between humans and machines will also change; one of mass personalization.
direction is a factory will become fully-immersed human-machine • Digital Twin for people: Digital Twins can also connect workers at
collaboration space [21]. the shop floor. The representation of a person, including personal
• Smart Production Network: Connected cyber-physical production data like weight, health data, activity data, and emotional status can
systems will form a global production network that can respond in help to establish models to understand personal wellbeing and
almost real-time to dynamic changes in local production systems working conditions of humans in a factory. The understanding of
and external supply chain [22]. A production network of adaptive human state at workforce can help design human-centered human-
and self-optimizing production systems can enable autonomous machine collaboration strategies to increase the physical and psy-
configuration and planning of production activities for production chological health of workers, as well as achieving best production
jobs at changing scales to achieve sound economic, environmental performance. Workers can also upskill themselves via ultra-realistic
and social impacts. training programs which blend physical factory setups with virtual
• Mass Personalization: Production model will move from a push- what-if scenarios. The ability to set up personalized virtual training
type mass production model to pull-type mass personalization [23]. programs based on Digital Twins of workers and factories can lead

3
Y. Lu, et al. Robotics and Computer Integrated Manufacturing 61 (2020) 101837

Fig. 3. Smart manufacturing vision.

to tremendous resource optimization and operational efficiency.

• Digital Twin for factories: Digital Twins can also work for fac-
tories, making a replica of a live factory environment. Digital Twin
and data-driven production operations can allow the establishment
of a self-organizing factory environment with complete operational
visibility and flexibility. Connectivity and data tracking throughout
the complete manufacturing process enable factory operations to be
transformed into data-driven evidence-based practices, offering the
capabilities of tracing product fault sources, analyzing production
efficient bottlenecks and predicting future resource requirements.
• Digital Twin for production networks: By connecting manu-
facturing assets, people and service via Digital Twin, every aspect of
business can be virtually represented. Connecting distributed Digital
Twins between companies will allow companies to build virtually
connected production networks. Leveraging Big Data capabilities,
this strategy provides unprecedented visibility into operation per-
formance and creates the possibility of predicting future needs in a
network of Digital Twins.
Fig. 4. A Digital Twin reference model.

4. Digital Twin reference model

model, the connection and reflection between these two ends will dis-
Digital Twin reflects the two-way dynamic mapping between a connect, and the information model becomes a one-off snapshot of its
physical object and its virtual model in the cyberspace [24]. A Digital physical counterpart. High-performance data processing is the key to
Twin presents a middleware architecture that abstracts its physical bridge the gap between the heterogeneous data stream and the Digital
counterpart for high-level engineering management systems to make Twin information model.
near real-time decisions [25]. Fig. 4 shows a Digital Twin reference
model. At the technical core, the development of Digital Twin needs
three components: (1) an information model that abstracts the speci- 4.1. Information model
fications of a physical object, (2) a communication mechanism that
transfers bi-directional data between a Digital Twin and its physical A physical object is abstracted with a pre-defined information model
counterpart, and (3) a data processing module that can extract in- that represents its specifications of concern. Standard plays a critical
formation from heterogeneous multi-source data to construct the live role in providing the information model for describing various physical
representation of a physical object. These three components must work objects in the manufacturing domain. Fig. 5 lists well-recognized
together for constructing a Digital Twin. Without an information model standards that provide standard information models for describing
to abstract the features of a physical entity, data transmitted to the physical objects in the manufacturing domain. These information
cyberspace will lose its meaning and context. Without a data synchro- models are classified into two subtypes: information models for product
nization mechanism between a physical model and its information Digital Twin and information models for production Digital Twin.

4
Y. Lu, et al. Robotics and Computer Integrated Manufacturing 61 (2020) 101837

Fig.5. Timeline-based depiction of standards for Digital Twin in the manufacturing domain (Enriched from [26]).

4.1.1. Information models for product Digital Twin semantic vocabulary for manufacturing equipment to provide struc-
The most predominant standards for developing product Digital tured, contextualized data with no proprietary format [33]. It is de-
Twin are ISO 10303 [27] and ISO 14649 [28] standards. ISO 10303 veloped to translate manufacturing data into a common, internet-based
[27], known as STEP, provides a neutral data structure for exchanging language interpretable by software applications. MTConnect defines a
product data between CAD systems. Its latest development of AP242 hierarchical information model for machine tools. The information
[29] for ‘Managed Model Based 3D Engineering’ by merging AP203 and model represents the logical structure of a machine tool, including the
AP204, with a focus on the representation of 3D model data, geometric components, the available data and the relationships between them.
tolerance and PMI (Product Manufacturing Information), provides a OPC Unified Architecture (OPC UA) [34] is another open standard
sound technological base for global design and manufacturing colla- that specifies information exchange for industrial communication on
boration. Geometric tolerance and PMI information can now be read by devices within machines, between machines and from machines to
machines directly from product design files in STEP AP242 model systems. A widely-recognized OPC-UA information model is MTCon-
without the need of interpreting 3D drawings. This change closes the nect-OPC-UA companion specification, aiming at improving the inter-
communication gaps between various stages of the product lifecycle, operability between these two standards. MTConnect-OPC UA compa-
resulting in autonomous process planning, manufacturing, inspection, nion specification ensures interoperability and consistency between
and so forth. MTConnect specifications and the OPC UA specifications, as well as the
ISO 14649 [28] and ISO 10303-238 [30] (also known as STEP-NC) manufacturing technology equipment, devices, software, or other pro-
are proposed to replace the RS274D (ISO 6983) M and G code via a ducts that implement those standards.
modern associative language that directly connects the CAD design data It is common that the information model from a single standard
used to determine the machining requirements for operation with the cannot meet the application requirements because of the breadth of the
downstream fabrication processes. STEP-NC allows manufacturing or- potential application of Digital Twin. Past studies suggest a systematic
ganizations to share machining information between machines seam- information model development process to ensure maximum standard
lessly via the exchange of semantic-enriched ‘what-to-do’ information. conformance and usability [35]. MTConnect and OPC-UA community
It relies on intelligent machine tools to interpret ‘how-to-do’ instruc- also recommend a bottom-up approach to expanding existing in-
tions adaptive to the local machining conditions. The shift of inter- formation models to suit new application needs, especially when now
preting local machining instructions into the machine controller level IT disruptions outpace the manufacturing standard development.
maximizes the interoperability between distributed machine tools.
4.2. Industrial communication – twinning tools
4.1.2. Information models for production Digital Twin
ISO 13399 [31] is an international standard by ISO for the com- A communication network is another critical factor for enabling the
puter-interpretable representation and exchange of industrial product establishment of Digital Twins. State synchronization between a Digital
data about cutting tools and tool holders. It provides a neutral me- Twin and its counterpart in the physical space relies on bi-directional
chanism capable of describing product data regarding cutting tools. and real-time data communication. State changes to a physical object
This information model has been used for CAD/CAM/CNC integration, are detected by sensors and transmitted to its Digital Twin in the cy-
tool management, product data management and manufacturing re- berspace. In this regard, industrial communication protocols can help
source planning. Similarly, ISO 14649-201 [32] defines a model for collect data from physical devices.
specifying machine tool data required for cutting processes. Table 1 presents a list of industrial communication protocols used
Targeting at describing machine tools, MTConnect standard offers a for industrial process monitoring and control. These protocols are

5
 Y. Lu, et al.

Table 1
Popular industrial communication protocols mapped to the OSI model ([36–44].

6
Robotics and Computer Integrated Manufacturing 61 (2020) 101837
Y. Lu, et al. Robotics and Computer Integrated Manufacturing 61 (2020) 101837

mapped to the ISO Open Systems Interconnection (OSI) model but very information exchange. This is evident from the early development of
often modified (or simplified) to satisfy the real-time and reliability the Fieldbus systems. Since 2000, the concept of IoT and Wireless
requirements of industrial processes. Existing industrial systems are Sensor Networks (WSN) are also impacting the industrial network field.
typically implemented with heterogeneous networks. These industrial Most of modern approaches are adopting existing standards such as
networks can be classified into three different categories. The first ca- IEEE 802.11 [46] (e.g., WiFi-based), IEEE802.15.1 (e.g., Bluetooth-
tegory consists of the earliest development of industrial networks, or based), and IEEE 802.15.4 [47] (e.g., Zigbee-based). While wireless
the so-called Fieldbus, which represents the common legacy networks networks have the intrinsic benefit of ease of installation due to no
in existing industrial automation systems. The second category is the wiring and low cost, existing approaches are still limited due to the lack
next generation of industrial networks, which are typically Ethernet- of reliability and potentially long latency for safety-critical and real-
based protocols but with modifications to satisfy the real-time and re- time data.
liability requirements. The third category is the recent development
following the trend of the Internet of Things that typically makes use of 4.3. Big data processing
wireless network technologies.
The data gathered from various sources to construct a Digital Twin
4.2.1. Fieldbus networks will be Big Data [48], if not now. Efficient processing of Big Data
In the late 1970s to late 1980s, several dedicated industrial net- gathered from the physical space is the third pillar of developing a
works (or Fieldbus), such as PROFIBUS [39] and Modbus [38], were Digital Twin.
developed to support the machine to machine communications and the Data processing methods that use statistical analysis and prediction
remote terminal control of programmable logic controllers, for process models while ignoring noise and conflicts between single data records
and peripheral control/automation. As shown in Table 1, many do not apply to Digital Twin development by default. The following
Fieldbus protocols were designed to operate on different physical media unique features need to be considered for Big Data analysis solutions
and have widely incompatible communication stacks across different targeting Digital Twin industrial applications.
layers of the OSI model. This has thus lead to closed-loop silos which
prevents data exchange and communication between standards. The • Hidden Meaning – Feature extraction in industrial Big Data ana-
current trend is moving towards adopting Ethernet-based standard, lysis needs to analyze the meaning of a feature and the relations
such as Modbus/TCP [41], in order to facilitate inter-communication at between features in the real world, in addition to statistical analysis
a higher level (e.g., the Internet or the enterprise control system). of feature relations.
Table 1 also provides a high-level comparison between different • Timeliness – Industrial data analysis requires low-latency data
types of industrial networks in terms of their communication data rate processing to enable time-sensitive applications, such as cloud-based
and the number of supported devices on a single network. While the industrial control [49].
performance varies significantly among the three categories, it should • High Quality – Data quality is sometimes more important than its
be noted that the data rate and the number of supported devices in a volume. Industrial Big Data applications need high-quality data that
specific network are heavily influenced by the selected physical covers the full spectrum of the system/process to be analyzed. Noise
medium, operation mode (and hence communication overhead), net- and data conflicts can directly break data analysis and result in
work topology, and the length of the physical medium (e.g., cable unusable results.
length). While Fieldbus networks seem to offer a slower data rate and
fewer devices, their key advantage is usually the deterministic com- Therefore, there exists a demand for a low-latency data processing
munication time for safety-critical operations. system that can integrate domain knowledge verification for data pro-
cessing. To this end, we propose the following general data processing
4.2.2. Ethernet-based industrial networks framework for constructing Digital Twins as shown in Fig. 6.
An increasing number of manufacturers are using Industrial
Ethernet-based solutions to connect systems. This is driven by the need 4.3.1. Data acquisition and cleansing
of high-performance integration between factory installations and the Real-world data collection comes with noise and missing data. It is
Industrial Internet of Things [45]. The advantages of Industrial Ethernet essential to clean low-quality raw data into ordered, meaningful and
over traditional Fieldbus systems are its homogenous network infra- simplified forms.
structure, ease of integration with the Internet, greater bandwidth to A missing value is a datum that has not been stored or gathered due
transmit safety-critical data, and the ability to communicate over longer to a faulty sampling process, cost restrictions or limitations in the ac-
distances. Even with the adoption of a common Ethernet standard, quisition process. Inappropriate handling of the missing values will
devices that support different Industrial Ethernet standards are not easily lead to poor knowledge extraction and wrong conclusions [19].
compatible or interoperable with each other because of the unique One option is to discard the instances that may contain a missing value.
protocol stacks in different Industrial Ethernet standards. The future However, this approach is rarely beneficial, as eliminating instances
Ethernet IEEE 802.1 TSN (Time Sensitive Networks) standard could may produce a bias in the data processing process, and a Digital Twin
eventually make time-critical and deterministic network communica- can miss some critical snapshots. Another method is to use a statistical
tion via standard Ethernet components possible, thus facilitating wider method to ‘guess’ an approximate value to fill the missing values. This
adoption and better interoperability. method can be a good choice if integrated with domain knowledge
reasoning. There are physical models behind activities occurring in the
4.2.3. Industrial wireless networks manufacturing environment. The domain knowledge can be used as the
One of the key driving features in Digital Twin and throughout the base rules for making a reasonable prediction of the missing value.
automation and manufacturing industries is the need for data/ Conflicting or redundant data records can introduce bias in the data

Fig. 6. Data processing for industrial

big data.

7
Y. Lu, et al. Robotics and Computer Integrated Manufacturing 61 (2020) 101837

analysis results. Two main approaches can be used to solve the pro-
blem. The first one is to correct the noise by using data polishing [50],

MySQL, PostgreSQL, Microsoft SQL Server

Accumulo, Cassandra, HBase, Hybertable,
primarily when the data affects the annotation of a physical-related

Couchbase Server, CouchDB, MarkLogic,

status. The second method is to use noise filters [51], which identifies
Aerospike, DynamoDB, Redis, Riak

and removes the noisy instances in the data while maintaining the re-
presentation of the physical state of the object to be modeled. Both
methods require extensive use of domain knowledge to drive the pol-
AllegroGraph, IBM, Neo4j

ishing and filtering decisions to ensure corrective actions can be taken

so that the nature of the data set is not changed.

4.3.2. Data storage

MongoDB

SimpleDB
Example

There is a multitude amount of data storage options tailored to

different kinds of data formats and application requirements, which are
classified as relational databases and non-relational databases (as
shown in Table 2). Relational databases are good choices for applica-
Used for storing clickstream data and log files, such as CPS

tions that involve the management of complex database transactions

Internet search and other large-scale Web applications
Content management and monitoring web and mobile

Complicated query, database transactions, and data

and heavy data analysis, because of referential integrity.

Non-relational databases are geared towards managing large sets of
Recommendation engines and knowledge base

varied and frequently updated data, often in distributed systems. They

avoid the rigid schemas associated with relational databases. The ar-
chitectures vary and are separated into four primary classifications,
each of which is suitable for different application scenarios as listed in
Table 2.
In the context of smart manufacturing, key-value databases can be
used for storing software log files from connected CPSs. Similarly,
documents can be stored in document databases to ensure high flex-
ibility and durability. Engineering knowledge can reside in graph da-
applications
Application

tabases to accelerate querying and reasoning. Column stores allow for

log files

analysis

fast querying and processing and it is heavily used for big-data analysis
where speed is critical. In practice, these databases will need to com-
plement each other to ensure low latency.
A data model that pairs a unique key and its associated value in storing data

Stores data in document-like structures that encode information in formats

Emphasizes connections between data elements, storing related ‘nodes’ in

4.3.3. Time-sensitive data processing

Digital Twin applications, such as real-time monitoring, prediction,
Stores data across labels that can have a huge number of columns

and control impose a stringent latency requirement for the data pro-
Stores information in structured tables with rows and columns

cessing architecture [52]. A ‘designed for latency’ data processing ar-

chitecture becomes a critical criterion.
Parallel computing technologies can ensure low-latency data pro-
cessing. The essence of data parallelism paradigm is to divide a com-
putational task into a cluster of similar sub-tasks that can be processed
independently and whose results are combined afterwards, upon com-
pletion. MapReduce is one such technology that has been widely used.
Recognizing its limitation for stream analytics [53], some competitors
emerged as an alternative, capable of performing stream analysis.
graphs to accelerate querying

Apache Spark is a large-scale data processing engine for both batch and
stream processing. Apache Flink later emerged as a better option be-
cause of its built-in streaming processing, instead of using micro
batching to handle stream processing. Apache Storm is an open-source
distributed real-time processing platform. Besides, some other
such as JSON
Description

streaming engines can be used where appropriate, such as Kafka

elements

Streams and Google Dataflow.

Though the above technologies provide the foundation for Big Data
processing, high-performance processing of time-series data is the key
Document database
Key-value database

to the successful implementation of a Digital Twin. This is due to two

Graph Database

reasons: (1) data gathered from the physical world are mostly discrete
Column stores

SQL database
Representative data storage options.

time data, and (2) there is timeliness requirement for a Digital Twin,
Database

regardless of the latency requirement between a Digital Twin and its

physical counterpart. The effective handling of time-series data to
construct a Digital Twin can ensure a required latency between a Digital
Non-relational database

Twin and its physical counterpart.

Relational database

5. Application scenarios

Although Digital Twin is a relatively new concept, some practical

Table 2

applications of Digital Twin have already been developed and reported

Type

in the literature. This section briefs the current status of Digital Twin

8
 Y. Lu, et al.

Table 3
Overview of existing Digital Twin applications.
Source Digital Twin type Physical object Information model Data communication standard Benefit(s)/Purpose(s)

[54] Manufacturing asset Grinding wheel Numerical models Universal Asynchronous Receiver Improve energy and resource efficiency of grinding process
Transmitter (UART)
[55] Satellite assembly system CAD model + Behavior model + Rule model – Smart production management and control
[56] Manufacturing cell – – Comprehensive simulation-based system engineering
[57] 3D printer MTConnect MTConnect Digital Twin-enabled cyber-physical cloud manufacturing
[58] 3D printer Proprietary ontology HTTP Cloud-based bidirectional communication between machine and Digital
Twin
[59] Welding production line CAD model + OPC-UA OPC-UA Digital Twin for product life cycle management
[60] Engine parts – – Digital Twin-based smart process planning
[61] Core making machine CAD model + Kinematic model OPC-UA Digital Twin for machine reconditioning (significant reduction of the
commissioning time)
[62] Hot rolling production line – – Optimization of hot rolling production line scheduling
[63] Drilling system and crane – – Monitor machine conditions and operator performance at a low cost
[64] Production system – – Tutoring service, augmented assistance, in-line diagnostics, and
condition monitoring service
[65] Factory – – Digital Twin to improve factory design
[66] Rotating machinery Finite element model – Digital Twin for rotating machinery fault diagnosis
[67] Waste electrical and electronic Extended ISO 10303 standard – Digital Twin-based system for the WEEE recovery to support the

9
equipment manufacturing/ remanufacturing operations throughout the product's
life cycle
[68] Machine tool MTConnect MTConnect Comprehensive machine tool and machining process monitoring,
decision-making support for humans
[69] Machine tool – – Machine tool operation monitoring, surface roughness prediction
[70] Automotive production cell – – Digital Twin for monitoring energy consumption
[71] Multi-robot systems – – Digital Twin for just-in-time planning of intelligent multi-robot systems
with improved execution time
[72] Machine tool FDI (factory design and improvement) model – Machine tool status monitoring
[24] 3D printer – – Digital Twin of 3D printers for rebuilding digital part and data
visualization
[73] Industrial valve – PLCopen Exchange data between Digital Twin and other systems
[74] Robots Multibody dynamics model – Digital Twin enabled VR testbed, robot control, and manual guidance
[75] Part STEP/ G code MTConnect Linking as-planned to as-fabricated product data
[76] Factory Production line Proprietary 3D model Ethernet + OPC Digital Twin to support decision-making over the system design and
solution evaluation
[77] Warehouse Proprietary 3D model – Digital Twin enabled VR testing environment
[8] Shop floor – – Digital Twin of shop floor to support smart operations in the
manufacturing process
[78] Industrial work cell Unified Robot Description Format(URDF) + Simulation Modbus TCP/IP + Ethernet IP A simulator based on Digital Twin for a flexible framework of work cell
Description File (SDF)
[79] People Employee – – Digital Twin of employees to support Intelligent control of an assembly
station
Robotics and Computer Integrated Manufacturing 61 (2020) 101837
Y. Lu, et al. Robotics and Computer Integrated Manufacturing 61 (2020) 101837

applications. First, an overview of existing Digital Twin applications is setups, tool paths, tool requirements, and in-process tolerances, are
provided, and the current status of Digital Twin applications is dis- communicated via STEP-NC AP238 protocol [30]. MTConnect is used to
cussed. Second, three representative Digital Twin applications are in- monitor the machining results. It allows the machine tool status and
troduced to demonstrate the advantages and potential of Digital Twin. coordinates to be streamed to the Digital Twin in real time (100 Hz).
The MTConnect data stream also includes measurement results as re-
5.1. Overview of existing digital twin applications ported by touch probes to enable tolerance evaluation. QIF is used to
report the results of the quality evaluations. The utilization of these
Existing Digital Twin applications reported in the literature have standards provides the Digital Twin Machining solution with great in-
been reviewed and summarized in Table 3. The details of each Digital teroperability.
Twin application are also briefly mentioned, including (1) the type of The Digital Twin functions as a server that allows Web-based ap-
the Digital Twin, i.e., manufacturing asset, human, factory or produc- plications to access all the data in the Digital Twin. In the Digital Twin
tion network), (2) the physical object of the Digital Twin, (3) the in- Machining application developed by STEP Tools Inc., the models of
formation model of the Digital Twin, (4) data communication standard workpieces, cutters, fixtures, as well as operations and tool paths are
used in the application, and (5) the benefit(s) or purpose(s) of the Di- fully assembled to perform real-time machining simulation. During real
gital Twin application. machining processes, the assembled model is updated in real time to
It can be seen that compared to the total number of publications on show the machining results on the Digital Twin. Operators can remotely
Digital Twin, most of the existing research on Digital Twin is conceptual monitor the machining processes using mobile devices that support
work, development of practical Digital Twin applications is still at an standard Web browsers. Measurements can be made on the Digital twin
early stage. Key findings are as follows: and alerts can be sent if tolerances are not being met. Thus, the Digital
Twin Machining enables “build it here, build it now and build it right,”
• Digital Twin type: 85% of Digital Twin applications are developed as claimed by STEP Tools Inc.
for manufacturing assets; 11% are developed for factories; only one
Digital Twin application for people has been identified, and there is 5.2.2. Digital Twin for a rotor system
no Digital Twin application for production networks. This shows Wang et al. [66] developed a Digital Twin application for rotating
that prior Digital Twin research mainly focused on manufacturing machinery fault diagnosis that can identify the fault parameters of a
devices; the importance of the involvement of human in the Digital rotor system and perform quantitative diagnosis of the rotor system.
Twin environment has been overlooked. Besides, the lack of appli- The overall system architecture of the developed application is shown
cations for production networks indicates that research on com- in Fig. 8.
munication/interactions between Digital Twins has not attracted Since the main purpose of this application is the fault diagnosis for
much attention. rotating machinery, modeling of the Digital Twin mainly considers the
• Information model: For the Digital Twin for manufacturing assets, dynamic behavior of the rotor system. Hence, the Digital Twin of the
information models that describe the data structure and semantics rotor is constructed using a finite element model that includes the
are mostly used, including different types of data models geometry, dynamics and material properties of the rotor. The critical
(MTConnect, OPC-UA, AutomationML and so forth) and databases. speed and unbalance response of the rotor under different conditions
However, the information model for a factory Digital Twin has not are obtained by finite element analysis. Four displacement sensors and
been explored to depth. Whether this should be an integration of a data acquisition system were implemented to collect the vibration
existing information models for manufacturing assets or should be a signals from the rotor system.
standard that governs all still needs to be addressed. The rotor unbalance fault quantification and localization were
• Data communication standard: Only a few applications have used performed to realize the fault diagnosis. Compared with traditional
unified data communication standards for modeling a Digital Twin. fault diagnosis methods, the developed Digital Twin application enables
This issue can severely limit the interoperability and accessibility of unbalance quantification and localization for fault diagnosis, which
a Digital Twin. It may also be the main reason why no Digital Twin further enables accurate diagnosis and adaptive degradation analysis of
application for production networks has been developed so far. rotating machinery.
• Purpose/benefit: Most applications are developed to provide
monitoring functions (status monitoring, process visualization, fault 5.2.3. Digital Twin enabled Cyber-Physical Machine tool
diagnosis, and so forth) and prediction functions (fault prognosis, Aiming at advancing legacy machine tools to Cyber-Physical
product lifecycle management, process optimization, and so forth). Machine Tools (CPMT) [83], Liu et al. [68] developed an MTConnect-
Most applications can be seen as decision-making support applica- based CPMT where the Digital Twin of the machine tool is a core
tions for humans; while very few of them have included direct/au- component. Fig. 9 shows the system architecture of the developed
tonomous feedback control from Digital Twin to a physical object. MTConnect-based CPMT. Real-time machining data were collected
from the CNC controller and embedded sensors and communicated
5.2. Representative applications through to the Digital Twin via MTConnect standard.
The implementation of MTConnect standard significantly improved
Though the development of Digital Twin applications is still at a the interoperability of the machining data and hence the accessibility of
very early stage, several full-fledged Digital Twin applications have the Digital Twin. A prototype of a machine tool monitoring system was
emerged. Here, we discuss some of the important application scenarios. developed to enable near real-time remote machine monitoring.

5.2.1. Digital Twin machining 6. Research issues

STEP Tools Inc. developed a Digital Twin Machining application
[80] that enables real-time quality inspection of machining results. Based on the discussions in the above sections, we summarize the
Fig. 7 shows the system framework of Digital Twin Machining. The following key research issues for advancing the research of Digital
Digital Twin is enabled by the utilization of four standards: 1) STEP Twin-driven smart manufacturing.
[27], 2) STEP-NC [28], 3) MTConnect [33], and 4) Quality Information Research issue 1: architecture pattern for a Digital Twin
Framework (QIF) [81]. There exist two system architecture patterns, namely server-based
STEP AP242 [29] is used to describe the design information of a and edge-based. In server-based architecture, the data acquired from a
workpiece. The manufacturing solutions, including all the operations, physical device is routed back to a centralized server that performs the

10
Y. Lu, et al. Robotics and Computer Integrated Manufacturing 61 (2020) 101837

Fig. 7. System framework of the Digital Twin Machining [82].

data analysis and Digital Twin construction. This pattern provides Two common methods can be used to gather data from physical
economies of scale and facilitates easy maintenance. In edge-based ar- devices, i.e., capturing changes and taking snapshots. There is extensive
chitecture, some data analysis is applied at the ‘edge’ of the system. use in large scale computer systems for both methods, and sometimes a
That is, data pre-processing is performed locally and on the raw data system uses a mix of them. Both methods need to be validated for
captured from a physical device. As a result, edge-based architecture specific application cases.
should be more effective on low-latency data processing if designed Research issue 4: standards for Digital Twin
well. However, this pattern is more complicated to maintain. Though anyone can develop a Digital Twin solution using common
Research issue 2: communication latency requirement for a Digital technologies, standards will facilitate the longevity of a Digital Twin
Twin solution. Standard-compatible Digital Twin solutions can inherit the
Latency requirement is application-driven. The application scenario flexibility, interoperability and scalability of existing and new standards
determines the required communication latency between a physical for information model and communication protocols. This is especially
device and its Digital Twin. This is because the system development important when a Digital Twin will be deployed in an open network of
costs and difficulties increase significantly as the communication la- Digital Twins. Recognizing the need for standardization, ISO is actively
tency requirement becomes stringent. In practice, Digital Twin-based developing a dedicated standard for Digital Twin manufacturing [9].
shop floor monitoring can accept higher latency than cloud-based in- Research issue 5: functionalities of a Digital Twin
dustrial control. BMWi, Germany [85] specifies the nominal commu- Existing Digital Twin applications are mainly developed for mon-
nication latency for various manufacturing applications, which can be itoring and prediction purposes and used as decision-making support
used as a guideline for designing the system architecture of a Digital applications for humans. Though human involvement in the smart
Twin application. manufacturing environment is essential, direct/autonomous feedback
Research issue 3: data capture mechanism control from the Digital Twin to the physical world should be

Fig. 8. Digital Twin for rotating machinery fault diagnosis [66].

11
Y. Lu, et al. Robotics and Computer Integrated Manufacturing 61 (2020) 101837

Fig. 9. MTConnect-based Cyber-Physical Machine Tool [84].

developed. Hence, the Digital Twin application can endow the physical 7. Conclusions
objects with a certain degree of autonomy.
Research issue 6: Digital Twin model version management This paper presents the current status and advancement of Digital
A Digital Twin model can evolve over time as a result of engineering Twin-driven smart manufacturing. The core concept, reference model,
changes to its physical counterpart, changes to the modelling interests enabling technologies, application scenarios, and research issues of
throughout the lifecycle of the physical counterpart, or other cases. In Digital Twin-driven smart manufacturing are discussed in detail.
these cases, the different versions of a Digital Twin models over time With the rapid growth of integrating information technologies and
should be captured, stored and integrated. Snapshot-based and change- operation technologies in the industry, significant efforts have been
based version management principles can be applied for the effective made to make manufacturing smart. As a core element of future man-
management different versions of a Digital Twin model. ufacturing, Digital Twin-driven applications are going to challenge and
Research issue 7: humans in Digital Twin applications change the fundamentals of manufacturing systems and operations. The
Humans play an important role in the Digital Twin-driven smart convergence of the digital world and physical world enables smart
manufacturing environment. While some low-level operations can be decisions to be made at every single point of manufacturing operations,
autonomously achieved without human intervention, many decision- thus can foster a data-driven smart manufacturing environment.
making activities have to be performed by humans. Though some new As can be seen from the literature, nearly 500 articles related to
interaction technologies such as AR have been studied and im- Digital Twin in the engineering domain have been published since
plemented in a manufacturing environment to improve human-machine 2016, and the number is proliferating, together with huge interest from
interactions, humans are still not considered as an integral part of the the industry. R&D in this area needs to follow a common reference
smart manufacturing system. Significant research effort needs to be model. The authors believe that constructing a Digital Twin needs a
made on the topic of Digital Twin for people in the smart manufacturing standardized information model, high-performance data processing,
environment in the future. and industrial communications to work together. Existing standards in
the manufacturing and industrial control domain need to be used where

12
Y. Lu, et al. Robotics and Computer Integrated Manufacturing 61 (2020) 101837

appropriate. The dedicated standard for Digital Twin smart manu- 1016/j.jmsy.2018.05.003.
facturing being developed by ISO can be the starting point for con- [26] W.Z. Bernstein, T.D. Hedberg Jr, M. Helu, A.B. Feeney, Contextualising manu-
facturing data for lifecycle decision-making, Int. J. Product Lifecycle Manag. 10
solidating research efforts in this space. (2018) 326.
The research activities are going to stay active due to the challen- [27] ISO, ISO 10303-1: iIndustrial automation systems and integration-product data
ging issues of constructing a reliable Digital Twin in practice. This is representation and exchange-part 1: oOverview and fundamental principles.,
(1994).
especially true for manufacturing applications, which pose stringent [28] ISO, ISO 14649-1: iIndustrial automation systems and integration - Physical device
requirements on timeliness, accuracy, and reliability. The authors be- control - Data data model for computerized numerical controllers - Part 1:
lieve researches on standards, communication protocols, time-sensitive oOverview and fundamental principles, (2003).
[29] ISO, ISO 10303-242: iIndustrial automation systems and integration - Product data
data processing, and reliability need to be the priorities for the next representation and exchange - Part 242: aApplication protocol: mManaged model-
stage of the research while focusing on application scenarios of Digital based 3D engineering, (2014).
Twin. [30] ISO, ISO 10303-238: iIndustrial automation systems and integration - Product data
representation and exchange - Part 238: aApplication protocol: aApplication in-
terpreted model for computerized numerical controllers, (2007).
References [31] ISO, TC 29, ISO 13399-3: 2014 cutting tool data representation and exchange - Part
3: reference dictionary for tool items, Int. Org. Standard. (2014).
[1] R. Rosen, G. von Wichert, G. Lo, K.D. Bettenhausen, About the importance of au- [32] ISO, ISO/TS 14649-201:2011 - Industrial automation systems and integration –
tonomy and digital twins for the future of manufacturing, IFAC-PapersOnLine 48 Physical device control – Data data model for computerized numerical controllers –
(2015) 567–572, https://doi.org/10.1016/J.IFACOL.2015.06.141. Part 201: mMachine tool data for cutting processes, (2011).
[2] E.E.H. Glaessgen, D.D.S. Stargel, The digital twin paradigm for future NASA and US [33] MTConnect Institute, Standard — MTConnect, (n.d.). https://www.mtconnect.org/
air force vehicles, in: 53rd AIAA/ASME/ASCE/AHS/ASC structures, Structural standard20181(accessed April 24, 2019).
Dynamics and Materials Conference 20th AIAA/ASME/AHS Adaptive Structures [34] OPC Foundation, Unified architecture - OPC Foundationfoundation, (n.d.). https://
Conference 14th AIAA, 2012, p. 1818. opcfoundation.org/about/opc-technologies/opc-ua/(accessed April 25, 2019).
[3] F. Tao, J. Cheng, Q. Qi, M. Zhang, H. Zhang, F. Sui, Digital twin-driven product [35] Y. Lu, H. Wang, X. Xu, ManuService ontology: a product data model for service-
design, manufacturing and service with big data, Int. J. Adv. Manuf. Technol. 94 oriented business interactions in a cloud manufacturing environment, J. Intell.
(2018) 3563–3576, https://doi.org/10.1007/s00170-017-0233-1. Manuf. 30 (2019) 317–334, https://doi.org/10.1007/s10845-016-1250-x.
[4] R. Söderberg, K. Wärmefjord, J.S. Carlson, L. Lindkvist, Toward a digital twin for [36] ODVA, ControlNet - CIP on CTDMA technology Technology technology Overview
real-time geometry assurance in individualized production, CIRP Ann. - Manuf. overview series, 2016.
Technol. 66 (2017) 137–140, https://doi.org/10.1016/j.cirp.2017.04.038. [37] Performance evaluation of control networks: ethernet, ControlNet, and devicenet,
[5] Y. Lu, X. Xu, Cloud-based manufacturing equipment and big data analytics to enable IEEE Control Syst. 21 (2001) 66–83, https://doi.org/10.1109/37.898793.
on-demand manufacturing services, Robot. Comput. Integr. Manuf. 57 (2019) [38] Modbus, MODBUS over serial line specification and implementation guide V1.02,
92–102, https://doi.org/10.1016/j.rcim.2018.11.006. 2006.
[6] E.J. Tuegel, A.R. Ingraffea, T.G. Eason, S.M. Spottswood, Reengineering aircraft [39] PROFIBUS Nutzerorganisation e.V., PROFIBUS technical description, 1999.
structural life prediction using a digital twin, Int. J. Aerospace Eng. 2011 (2011) [40] R. Schlesinger, A. Springer, T. Sauter, Automatic packing mechanism for simplifi-
1–14, https://doi.org/10.1155/2011/154798. cation of the scheduling in profinet IRT, IEEE Trans. Industr. Inform. 12 (2016)
[7] S. Weyer, T. Meyer, M. Ohmer, D. Gorecky, D. Zühlke, Future modeling and si- 1822–1831.
mulation of CPS-based Factories: an example from the automotive industry, IFAC- [41] Modbus Org, MODBUS messaging on TCP/IP implementation guide V1.0b Modbus-
PapersOnLine 49 (2016) 97–102, https://doi.org/10.1016/j.ifacol.2016.12.168. IDA, 2006.
[8] F. Tao, M. Zhang, Digital twin shop-floor: a new shop-floor paradigm towards smart [42] P. Danielis, J. Skodzik, V. Altmann, E.B. Schweissguth, F. Golatowski,
manufacturing, IEEE Access 5 (2017) 20418–20427, https://doi.org/10.1109/ D. Timmermann, J. Schacht, Survey on real-time communication via ethernet in
ACCESS.2017.2756069. industrial automation environments, Proceedings of the 2014 IEEE Emerging
[9] ISO, ISO/AWI 23247 - Digital Twin manufacturing framework (Under develop- Technology and Factory Automation (ETFA), IEEE, 2014, pp. 1–8, , https://doi.org/
ment), (2019). 10.1109/ETFA.2014.7005074.
[10] B. Schleich, N. Anwer, L. Mathieu, S. Wartzack, Shaping the digital twin for design [43] Hart Communication Foundation, HART communication application guide, 2013.
and production engineering, CIRP Ann. 66 (2017) 141–144, https://doi.org/10. [44] J. Song, S. Han, A. Mok, D. Chen, M. Lucas, M. Nixon, W. Pratt, WirelessHART:
1016/J.CIRP.2017.04.040. applying wireless technology in real-time industrial process control, 2008 IEEE
[11] B. Bagheri, J. Lee, Big future for cyber-physical manufacturing systems, Design Real-Time and Embedded Technology and Applications Symposium, IEEE, 2008,
World (2015), https://www.designworldonline.com/big-future-for-cyber-physical- pp. 377–386, , https://doi.org/10.1109/RTAS.2008.15.
manufacturing-systems/ (accessed April 24, 2019). [45] Industrial ethernet is now bigger than fieldbuses, (n.d.). https://www.anybus.com/
[12] T.H.-J. Uhlemann, C. Lehmann, R. Steinhilper, The digital twin: realizing the cyber- about-us/news/2018/02/16/industrial-ethernet-is-now-bigger-than-fieldbuses(ac-
physical production system for industry 4.0, Procedia CIRP (2017) 335–340, cessed April 23, 2019).
https://doi.org/10.1016/j.procir.2016.11.152. [46] F. Tramarin, S. Vitturi, M. Luvisotto, A. Zanella, On the use of IEEE 802.11n for
[13] Digital twin market worth 15.66 billion USD by 2023, (n.d.). https://www. industrial communications, IEEE Trans. Industr. Inf. 12 (2016) 1877–1886, https://
marketsandmarkets.com/PressReleases/digital-twin.asp(accessed April 24, 2019). doi.org/10.1109/TII.2015.2504872.
[14] Predix platform | GE Digital, (n.d.). https://www.ge.com/digital/iiot-platform(ac- [47] E. Toscano, L. Lo Bello, Comparative assessments of IEEE 802.15.4/ZigBee and
cessed April 22, 2019). 6LoWPAN for low-power industrial WSNs in realistic scenarios, 2012 9th IEEE
[15] Digital twin | Siemens, (n.d.). https://www.plm.automation.siemens.com/global/ International Workshop on Factory Communication Systems, IEEE, 2012, pp.
en/our-story/glossary/digital-twin/24465(accessed April 24, 2019). 115–124, , https://doi.org/10.1109/WFCS.2012.6242553.
[16] ABB Ability, (n.d.). https://new.abb.com/abb-ability(accessed April 24, 2019). [48] A. Gandomi, M. Haider, Beyond the hype: big data concepts, methods, and analy-
[17] Overview of azure digital twins | Microsoft Docs, Microsoft. (n.d.). https://docs. tics, Int. J. Inf. Manage. 35 (2015) 137–144, https://doi.org/10.1016/j.ijinfomgt.
microsoft.com/en-us/azure/digital-twins/about-digital-twins(accessed January 1, 2014.10.007.
2019). [49] J. Pizoń, J. Lipski, Perspectives for fog computing in manufacturing, Appl. Comput.
[18] J. Davis, T. Edgar, J. Porter, J. Bernaden, M. Sarli, Smart manufacturing, manu- Sci. (2016) 12.
facturing intelligence and demand-dynamic performance, Comput. Chem. Eng. 47 [50] Choh Man Teng, Polishing blemishes: issues in data correction, IEEE Intell. Syst. 19
(2012) 145–156, https://doi.org/10.1016/J.COMPCHEMENG.2012.06.037. (2004) 34–39, https://doi.org/10.1109/MIS.2004.1274909.
[19] Smart manufacturing operations planning and control program | NIST, (n.d.). [51] A. Bosco, M. Mancuso, Noise filter for Bayer pattern image data, 2008.
https://www.nist.gov/programs-projects/smart-manufacturing-operations- [52] C. Weber, J. Königsberger, L. Kassner, B. Mitschang, M2DDM - a maturity model for
planning-and-control-program(accessed April 7, 2018). data-driven manufacturing, Procedia CIRP (2017) 173–178, https://doi.org/10.
[20] E. Wenger, Etienne, Artificial Intelligence and Tutoring Systems : Computational 1016/j.procir.2017.03.309.
and Cognitive Approaches to the Communication of Knowledge, Kaufmann [53] J. Lin, Mapreduce is good enough? If all you have is a hammer, throw away ev-
Publishers, Morgan, 1987. erything that's not a nail!, Big Data 1 (2013) 28–37.
[21] A. Bauer, D. Wollherr, M. Buss, Human–robot collaboration: a survey, Int. J. [54] K. Kannan, N. Arunachalam, A digital twin for grinding wheel: an information
Humanoid Rob. 5 (2008) 47–66. sharing platform for sustainable grinding process, J. Manuf. Sci. Eng. 141 (2019)
[22] Y. Lu, T. Peng, X. Xu, Energy-efficient cyber-physical production network: archi- 21015.
tecture and technologies, Comput. Industr. Eng. 129 (2019) 56–66, https://doi.org/ [55] C. Zhuang, J. Liu, H. Xiong, Digital twin-based smart production management and
10.1016/j.cie.2019.01.025. control framework for the complex product assembly shop-floor, Int. J. Adv. Manuf.
[23] A. Kumar, From mass customization to mass personalization: a strategic transfor- Technol. 96 (2018) 1149–1163, https://doi.org/10.1007/s00170-018-1617-6.
mation, Int. J. Flexible Manuf. Syst. 19 (2007) 533. [56] M. Schluse, M. Priggemeyer, L. Atorf, J. Rossmann, Experimentable digital twins-
[24] A. Angrish, B. Starly, Y.-S.Y.-S. Lee, P.H.P.H. Cohen, A flexible data schema and streamlining simulation-based systems engineering for industry 4.0, IEEE Trans.
system architecture for the virtualization of manufacturing machines (VMM), J. Industr. Inf. 14 (2018) 1722–1731, https://doi.org/10.1109/TII.2018.2804917.
Manuf. Syst. 45 (2017) 236–247. [57] L. Hu, N.-T. Nguyen, W. Tao, M.C. Leu, X.F. Liu, M.R. Shahriar, S.M.N. Al Sunny,
[25] Y. Lu, X. Xu, Resource virtualization: a core technology for developing cyber-phy- Modeling of cloud-based digital twins for smart manufacturing with MT connect,
sical production systems, J. Manuf. Syst. 47 (2018) 128–140, https://doi.org/10. Procedia Manuf. (2018) 1193–1203, https://doi.org/10.1016/j.promfg.2018.07.
155.

13
Y. Lu, et al. Robotics and Computer Integrated Manufacturing 61 (2020) 101837

[58] M.R. Shahriar, S.M.N.A. Sunny, X. Liu, M.C. Leu, L. Hu, N.-T. Nguyen, MTComm Engineering and Applications, ICIEA 2017, 2017, pp. 90–94, , https://doi.org/10.
based virtualization and integration of physical machine operations with digital- 1109/IEA.2017.7939185.
twins in cyber-physical manufacturing cloud, Proceedings - 5th IEEE International [72] S. Choi, G. Kang, C. Jun, J.Y. Lee, S. Han, Cyber-physical systems, A case study of
Conference on Cyber Security and Cloud Computing and 4th IEEE International development for manufacturing industry, Int. J. Comput. Appl. Technol. 55 (2017)
Conference on Edge Computing and Scalable Cloud, CSCloud/EdgeCom 2018, 289–297, https://doi.org/10.1504/IJCAT.2017.086018.
2018, pp. 46–51, , https://doi.org/10.1109/CSCloud/EdgeCom.2018.00018. [73] G.N. Schroeder, C. Steinmetz, C.E. Pereira, D.B. Espindola, Digital twin data mod-
[59] Y. Zheng, S. Yang, H. Cheng, An application framework of digital twin and its case eling with AutomationML and a communication methodology for data exchange,
study, J. Ambient Intell. Humaniz. Comput. (2018) 1–13, https://doi.org/10.1007/ IFAC-PapersOnLine 49 (2016) 12–17, https://doi.org/10.1016/j.ifacol.2016.11.
s12652-018-0911-3. 115.
[60] J. Liu, H. Zhou, G. Tian, X. Liu, X. Jing, Digital twin-based process reuse and [74] E.G. Kaigom, J. Roßmann, Toward physics-based virtual reality testbeds for in-
evaluation approach for smart process planning, Int. J. Adv. Manuf. Technol. 100 telligent robot manipulators - an erobotics approach, IEEE International Conference
(2019) 1619–1634. on Intelligent Robots and Systems, 2016, pp. 1000–1005, , https://doi.org/10.
[61] M. Ayani, M. Ganebäck, A.H.C. Ng, Digital Twin: applying emulation for machine 1109/IROS.2016.7759171.
reconditioning, Procedia CIRP (2018) 243–248, https://doi.org/10.1016/j.procir. [75] M. Helu, A. Joseph, T. Hedberg, A standards-based approach for linking as-planned
2018.03.139. to as-fabricated product data, CIRP Ann. 67 (2018) 487–490, https://doi.org/10.
[62] L.-L. Liu, X. Wan, Z. Gao, X. Li, B. Feng, Research on modelling and optimization of 1016/J.CIRP.2018.04.039.
hot rolling scheduling, J. Ambient Intell. Humaniz. Comput. (2018), https://doi. [76] H. Zhang, Q. Liu, X. Chen, D. Zhang, J. Leng, A digital twin-based approach for
org/10.1007/s12652-018-0944-7. designing and multi-objective optimization of hollow glass production line, IEEE
[63] S. Tang, R. Wang, X. Zhao, X. Nie, Building cloud services for monitoring offshore Access 5 (2017) 26901–26911, https://doi.org/10.1109/ACCESS.2017.2766453.
equipment and operators, Proceedings of the Annual Offshore Technology [77] T. Petković, D. Puljiz, I. Marković, B. Hein, Human intention estimation based on
Conference, 2018, pp. 852–864. hidden markov model motion validation for safe flexible robotized warehouses,
[64] A. Padovano, F. Longo, L. Nicoletti, G. Mirabelli, A digital twin based service or- Robot. Comput. Integr. Manuf. 57 (2019) 182–196, https://doi.org/10.1016/j.rcim.
iented application for a 4.0 knowledge navigation in the smart factory, IFAC- 2018.11.004.
PapersOnLine 51 (2018) 631–636, https://doi.org/10.1016/j.ifacol.2018.08.389. [78] P. Tavares, J.A. Silva, P. Costa, G. Veiga, A.P. Moreira, Flexible work cell simulator
[65] J. Guo, N. Zhao, L. Sun, S. Zhang, Modular based flexible digital twin for factory using digital twin methodology for highly complex systems in industry 4.0, Iberian
design, J. Ambient Intell. Humaniz. Comput. (2018), https://doi.org/10.1007/ Robotics Conference, Springer, 2017, pp. 541–552.
s12652-018-0953-6. [79] I. Graessler, A. Poehler, Intelligent control of an assembly station by integration of a
[66] J. Wang, L. Ye, R.X. Gao, C. Li, L. Zhang, Digital twin for rotating machinery fault digital twin for employees into the decentralized control system, Procedia Manuf.
diagnosis in smart manufacturing, Int. J. Prod. Res. (2018) 1–15. (2018) 185–189, https://doi.org/10.1016/j.promfg.2018.06.041.
[67] X.V. Wang, L. Wang, Digital twin-based WEEE recycling, recovery and re- [80] STEP Tools Inc., Digital Twin Machiningmachining, (n.d.). http://www.steptools.
manufacturing in the background of industry 4.0, Int. J. Prod. Res. (2018), https:// com/blog/20171011_twin_machining/(accessed March 27, 2019).
doi.org/10.1080/00207543.2018.1497819. [81] DMSC, Home - QIF Standardstandard, (n.d.). http://qifstandards.org/(accessed July
[68] C. Liu, H. Vengayil, R.Y.R.Y. Zhong, X. Xu, A systematic development method for 7, 2018).
cyber-physical machine tools, J. Manuf. Syst. 48 (2018) 13–24, https://doi.org/10. [82] Digital thread and digital twin demonstrations at future of Flightflight, (n.d.).
1016/j.jmsy.2018.02.001. http://www.steptools.com/blog/20161005_digital_thread_demo/(accessed March
[69] Y.Y. Cai, B. Starly, P. Cohen, Y.-S.Y.-S. Lee, Sensor data and information fusion to 27, 2019).
construct Digital-Twins virtual machine tools for cyber-physical manufacturing, [83] X. Xu, Machine tool 4.0 for the new era of manufacturing, Int. J. Adv. Manuf.
Procedia Manuf. 10 (2017) 1031–1042, https://doi.org/10.1016/j.promfg.2017. Technol. 92 (2017) 1893–1900, https://doi.org/10.1007/s00170-017-0300-7.
07.094. [84] C. Liu, X. Xu, Q. Peng, Z. Zhou, MTConnect-based cyber-physical machine Tool: a
[70] D. Hauf, S. Sus, A. Strahilov, J. Franke, Multifunctional use of functional mock-up case study, Procedia CIRP 72 (2018) 492–497, https://doi.org/10.1016/J.PROCIR.
units for application in production engineering, Proceedings - 2017 IEEE 15th 2018.03.059.
International Conference on Industrial Informatics, INDIN 2017, 2017, pp. [85] L. Rauchhaupt, G. Kadel, F. Mildner, B. Kärcher, R. Heidel, U. Graf, F. Schewe, D.
1090–1095, , https://doi.org/10.1109/INDIN.2017.8104925. Schulz, Network-based communication for industrie 4.0-Proposal for an adminis-
[71] N. Wantia, J. Roßmann, An online task planning framework reducing execution tration shell, 2016.
times in industrial environments, 2017 4th International Conference on Industrial

14