International Journal of Production Research

ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/tprs20

Simulated-based methodology for the interface

configuration of cyber-physical production
systems

Abderrahim Ait-Alla, Markus Kreutz, Daniel Rippel, Michael Lütjen & Michael
Freitag

To cite this article: Abderrahim Ait-Alla, Markus Kreutz, Daniel Rippel, Michael Lütjen & Michael
Freitag (2021) Simulated-based methodology for the interface configuration of cyber-physical
production systems, International Journal of Production Research, 59:17, 5388-5403, DOI:
10.1080/00207543.2020.1778209

To link to this article: https://doi.org/10.1080/00207543.2020.1778209

© 2020 The Author(s). Published by Informa

UK Limited, trading as Taylor & Francis
Group

Published online: 17 Jun 2020.

Submit your article to this journal

Article views: 1677

View related articles

View Crossmark data

Citing articles: 12 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tprs20
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH
2021, VOL. 59, NO. 17, 5388–5403
https://doi.org/10.1080/00207543.2020.1778209

Simulated-based methodology for the interface configuration of cyber-physical

production systems
Abderrahim Ait-Alla a , Markus Kreutz b , Daniel Rippela , Michael Lütjena and Michael Freitaga,b
a BIBA - Bremer Institut für Produktion und Logistik GmbH, University of Bremen Bremen, Germany; b Faculty of Production Engineering,
University of Bremen, Bremen, Germany

ABSTRACT ARTICLE HISTORY

Besides managing accruing data in production systems, Digital Twins provide additional services like Received 14 January 2020
the simulation or control of production systems. The Digital Twin and its real-world physical coun- Accepted 28 May 2020
terpart need to be interconnected using sensors and actuators to enable such services. This article KEYWORDS
proposes a simulation-based method to design and evaluate such an interface. Applying concepts Digital Twin; discrete event
from the area of software-in-the-loop, it proposes the use of a so-called Physical Twin: a simulation simulation; production
model which mimics the abilities of the physical system but allows simulation-based experiments control; autonomous
to optimise the interface. After presenting the general approach, the article provides an application transportation system;
example of the proposed procedure. A Digital and a Physical Twin are implemented for an applica- cyber-physical system
tion scenario and connected using a simple TCP/IP interface. By varying the number of sensors as
well as considering component breakdowns in the Physical Twin, this setup allows evaluating differ-
ent configurations in terms of production performance. The evaluation confirms that configurations
with more sensors result in higher production performance of the system. However, a saturation
of the performance gain is also observable when exceeding a certain number of sensors. Therefore,
determining the optimal configuration is vital to optimise the interface from a financial point of view.

Generally, such systems require at least a representation

Introduction of the physical system’s capabilities and restrictions (Lee
Over the last decade, concepts like Digital Twins and 2008), which corresponds to what nowadays is com-
cyber-physical systems (CPS) have evolved to core con- monly referred to as Digital Twins. Most of the research
cepts of the fourth industrial revolution (Industry 4.0). in Digital Twins deal with the development and integra-
Both core technologies rely on and exploit the increasing tion of such models for various use-cases, ranging from
information density in industrial processes, which results single machining processes via Digital Twins for specific
from the increasing integration of sensors and intercon- machines to complete production systems. In particu-
nectedness of resources (Leitão et al. 2015; Ivanov, Dol- lar, when dealing with Digital Twins for logistics systems,
gui, and Sokolov 2019). CPS combine several sensors multi-agent systems provide a suitable means to represent
and actuators with data processing capabilities, allow- the involved actors and their behaviour. While the devel-
ing such systems to act intelligently, depending on their opment of such a (digital) system can draw from a long
current state and the state of their environment and history of tools and methods, the two systems (physical
to render decentralised decisions (e.g. (Vogel-Heuser, and digital) need to be synchronised and connected using
Lee, and Leitão 2015; Leitão, Colombo, and Karnouskos a well-defined network of sensors and actuators. This is of
2016; Lee and Seshia 2017)). Examples for the applica- particular interest if considering logistics systems where
tion of CPS can be found in various industrial disciplines, sensors are not inherently part of the systems but need to
from condition monitoring and maintenance planning be introduced. Such a scenario is given by the application
(Boulekrouche, Jabeur, and Alimazighi 2016; Wang et al. of autonomous transport systems (ATS). These systems
2019) to production planning and scheduling (Toader usually rely either on a centralised or decentralised rout-
2017; Wiesner and Thoben 2017). CPS thereby rely on ing, which both require different degrees of supervision,
intelligent programming for decision-making and their i.e. of interconnectedness between the physical and the
interaction with other CPS in a shared environment. digital system.

CONTACT Abderrahim Ait-Alla [email protected] BIBA - Bremer Institut für Produktion und Logistik GmbH, University of Bremen,
Hochschulring 20, 28359 Bremen, Germany
© 2020 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group
This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License (http://creativecommons.org/licenses/
by-nc-nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited, and is not altered, transformed, or
built upon in any way.
 INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 5389

Consequently, this article presents a methodology for report from 2010 (Piascik et al. 2010). In terms of CPS
the configuration of the interconnection between the in general, Lee, Bagheri, and Kao (2015) defined a layer-
physical and the digital (cyber) systems with a focus based architecture to define the abilities of CPS. This
on logistics systems. The methodology applies the basic architecture sub-classified the structure of CPS into five
notion of a software-in-the-loop approach, where it rep- layers of increasingly complex abilities: from simple sen-
resents the physical system in terms of a simulation sory input and communication capabilities on the first
model. Using this model, system designers can easily level, over complex data-processing tasks on the third
experiment with different configurations in terms of layer to capabilities of understanding its environment
sensors and actuators to determine a suitable solution and acting towards a globally optimal goal system on
between the conflicting targets of logistic efficiency and the fifth level (Ait-Alla et al. 2019). Uhlenkamp et al.
costs. (2019) provide a comparable systematization of Digi-
While previous work already described a preliminary tal Twins and their abilities by mainly focusing on the
application of the concepts proposed in this article to a criteria Goal (information acquisition, information sys-
simplified use case (Ait-Alla et al. 2019), this article pro- tem, decision-making, and action implementation), Sys-
vides the methodological foundations. In extension of tem Focus (component, sub-system, system, a system of
this previous work, this article presents a detailed and systems) and the Data Integration Level (manual, semi-
structured procedure of the proposed simulation-based automated, fully automated). Especially the last criterion
method to configure the interfaces between the Digi- seems relevant, as it describes not only the data but also
tal Twin and the real-world system, using a so-called the overall integration of the Digital Twin. While historic
Physical Twin as an intermediary. Moreover, this article production systems usually only had a physical mate-
evaluates the proposed procedure model by applying it to rial flow, current systems tend to store their informa-
an extended use-case. tion in digitalised systems (Parente et al. 2020). In the
The paper is organised as follows. Section 2 presents future, Digital Twins will be capable of simulating dif-
a review of relevant literature on the topic of CPS and ferent actions and decisions in advance, using real-time
Digital Twins and introduces the research gap. Section 3 data and up-to-date information of the physical system,
introduces the proposed research methodology in the and reflecting the most suitable course of action into the
form of a procedure model. Section 4 then presents an physical system (Ait-Alla et al. 2019; Uhlenkamp et al.
exemplary application of the procedure model for an 2019; Kumar et al. 2020).
(extended) production system, for which several different
evaluations are conducted. Finally, the article closes with
Methods for the design of cyber-physical systems
a summary, discussions, contribution, limitation, and a
discussion of further research directions. The design and development of Digital Twins and, more
generally, cyber-physical systems, comes with various
challenges (Parente et al. 2020). On the software side, Lee
Literature review (2008) describes the problem of how concurrency and
execution time predictability is a crucial requirement for
Cyber-physical systems and Digital Twins
CPS. However, these are commonly not part of program-
During the early times of CPS, several terms for their ming language semantics (code execution without errors
digital (cyber) representation have emerged in the lit- is considered to be correct code regardless of execution
erature, e.g. Digital Twin (Lee, Bagheri, and Kao 2015), duration). He suggests that a rethinking of the abstrac-
digital shadow, or digital avatar (Hribernik, Wuest, and tion of computing is required to make full use of the
Thoben 2013). Today, literature mostly uses the term potential of CPS. A practical way to model Digital Twins
Digital Twin, while other terms are sometimes used syn- is by using a model of computation (MoC) as a basis.
onymously. The first documented mention of the concept An MoC defines a framework for how code is executed
of a Digital Twin dates back to a presentation at the Uni- in a programme and thereby helps the programmer to
versity of Michigan by M. Grieves in 2002. The main develop the intended functionalities. A typical example of
elements of a Digital Twin such as the real space, the an MoC is the finite state machine. Park, Easwaran, and
virtual space, and the exchange of data and informa- Andalam (2019) analyse different MoCs regarding their
tion between these two entities were shown at this event, features relevant for modelling CPS, among them con-
although the term ‘Digital Twin’ was not yet used to currency and timing capabilities, and conclude that every
describe the concept (Grieves and Vickers 2017). The MoC has its disadvantages. Therefore, different MoCs
earliest mention of the term itself in the CPS context must be combined to achieve a highly accurate, time-
that we found in the literature was in a NASA road map synchronised, and scalable Digital Twin. Tao et al. (2019)
5390 A. AIT-ALLA ET AL.

present a systematic description of how a functional Dig- the actual Digital Twin model. It introduces an interme-
ital Twin is created. While this description is not the diate model, later-on called Physical Twin, to optimise
main focus of the paper and the steps were formulated in the connection interface between the Digital Twin and
rather general terms, it represents one of very few step by the real-world system.
step guides for the creation of Digital Twins found in the
literature. However, this guideline focuses on the devel-
Research gap, definitions and contribution
opment of a Digital Twin for a specific product. It only
shortly mentions this article’s topic, i.e. the interconnec- While SiL and Digital Twins have been used extensively
tion between Digital Twin and its Physical Twin and the as tools for product development, the state of the art
dimensioning of the physical system, as part of its sixth given above shows only a few efforts to apply SiL or HiL
step concerning selecting a suitable technology. Regard- concepts to the development and optimisation of Digi-
ing testing of cyber-physical systems in the automobile tal Twins themselves. The concept of EDTs provides the
industry, the idea to reduce road tests and instead to per- closest research, which can be attributed to this topic and
form them in the lab or on computers has been around for already shows promising results. Moreover, the state of
some time, as Isermann (2006) explains in his book. Well the art provided above shows that currently, no struc-
before the term ‘Digital Twin’ became popular, simula- tured method exists, which allows experimentation or
tion models of partial components, like, e.g. controllers even optimisation of the interconnection between the
for a car’s electronic stability control (ESC), were used real-world system and its Digital Twin. Current proce-
to test the function of the component on a computer dure models note that the interconnection is a vital part
before the actual hardware is tested at all. Such a way of the process, but do not provide closer information
of testing, where a software model substitutes a hard- on how the suitability of this interconnection can be
ware component, is called software-in-the-loop (SiL). evaluated.
The development roadmap idea here is to start with SiL, Consequently, this article proposes a structured proce-
then continue with hardware-in-the-loop testing (HiL), dure model, which applies concepts from the Software-
where an actual piece of hardware is tested in a simu- in-the-Loop testing by introducing an intermediate
lated environment, before finally moving to a complete model of the physical system. Following the assumption
hardware prototype and tests on the road. Beyond the that a Digital Twin should add additional functionality,
automotive context, SiL has also been employed for other which is not provided by the real world, physical sys-
applications, such as power plants (Nguyen et al. 2019; tem. The proposed methodology refers to three different
Rossi, Reveillere, and Traverso 2019) or planetary rovers systems:
(Hellerer, Schuster, and Lichtenheldt 2016). The use of
Digital Twins for product design and development (i.e. • The real-world system: This system describes the
Digital Twin prototypes) involves testing the twin, which actual physical system existing in the real world. In the
acts as the substitute for the future product (Polini and use-case introduced later, this system comprises the
Corrado 2020). This can be considered a form of SiL, as a physical monorail transport system, its shuttles, and
software model (the Digital Twin) substitutes a hardware the required machinery. This system already provides
component. Schluse et al. (2018) present a framework several base-line capabilities, e.g. collision avoidance
for such a Digital Twin-aided product development com- and low-level routing capabilities.
bined with ideas from model-based systems engineering • The Physical Twin: The Physical Twin constitutes
(MBSE). They extend a Digital Twin by the simulated a bare-bones but realistic simulation model of the
environment as well as simulated inputs, allowing for a real-world system. This simulation model must not
modular approach where these modules, dubbed ‘Exper- extend the functionalities of its real-world counter-
imental Digital Twins’ (EDTs) by the authors, can be part. Instead, it should replicate the logic and physical
interconnected, allowing for a scalable simulation solu- properties of the physical counterpart to the degree
tion. This concept also falls into the SiL category. Zhang that is sufficient to test the intended functionalities of
et al. (2017) and Liu et al. (2019) explore the use of Dig- the Digital Twin. This Physical Twin would be con-
ital Twins for individualised product development and sidered a digital model according to the classification
how production optimizations determined in the Digital made by (Kritzinger et al. 2018).
Twin are fed back to the real-world system. The Digi- • The Digital Twin: The Digital Twin is the digital repre-
tal Twins presented in these works contain physics-based sentation of the real-world system but, in contrast to
representations of the real-world system that are used to the Physical Twin, adds additional and desired func-
determine these optimizations. In contrast, the article at tionalities, which are not covered by the real-world
hand separates these physics-based representations from system. Again, referring to the use-case provided later
 INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 5391

in this article, these could be functionalities like the of software and logistics experts, the proposed procedure
detection of transport malfunctions, advanced rout- follows a standard quality development model as, e.g.
ing, or order management. described in the ISO 9001 standard. While this procedure
model is formulated very generally, it is well established
and can be adapted easily to specific tasks. The remainder
Research methodology
of this section presents first the adapted procedure model.
In this paper, we propose a methodology for the configu- Afterward, this section provides a detailed description
ration of interconnections between a Digital Twin and its of the respective tasks involved and proposes suitable
real-world system. While such a method may not be nec- methods or classes of methods for them.
essary for Digital Twins of single processes or machines, The general approach follows the quality deployment
larger-scale logistic systems require a dedicated configu- model as given in Figure 2 and thus splits into the main
ration to ensure a viable trade-off between logistic perfor- phases Plan, Design, Code, Test, and Deploy. This arti-
mance and economic efficiency. The proposed method cle focuses on the first four phases and proposes a more
presents a structured procedure model to determine an detailed description of these phases in terms of tasks and
improved configuration for the interaction and thereby suitable methods.
can reduce cost and improve the performance of the CPS.
To our knowledge, such a structured method does not
Phase 1: plan
currently exist in literature.
Therefore, we propose to adopt the software-in-the- The first phase of the procedure model refers to the speci-
loop testing principles to allow for purely virtual exper- fication of the involved systems (physical system and Dig-
imentation in terms of the interconnection of Digital ital Twin) and of the study itself. As the overall process
Twins and their physical counterparts. The proposed focuses on achieving a simulation study in a software-
approach follows a setup as given in Figure 1. in-the-loop manner, the study’s aims have to be defined
This setup relies on an additional simulation model in- clearly beforehand. Thus, we propose to split this phase
between the real-world system and the Digital Twin. This into two sub-tasks, i.e. the specification of the simulation
so-called Physical Twin represents the real-world system study and the situational analysis of the involved systems.
with all its constraints and capabilities. However, in con- The specification of the simulation study should
trast to the Digital Twin, it does not add any additional clearly state which parts of the real-world system will be
capabilities or logic to it. Thereby, the Physical Twin rep- investigated, which situations need to be covered, and,
resents just the physical system in virtual space, along most importantly, the objectives and conditions of the
with all its hardware logic and physical phenomena and simulation study. Accordingly, this specification needs to
allows for virtual experimentation. This section proposes be developed in cooperation between the teams work-
a procedure model that describes a possible method to ing on/with the physical system and the Digital Twin.
conduct such studies to achieve an analysis in this fash- This specification defines the scope of the later Physical
ion. As such projects will generally be conducted by a mix Twin model. For example, this use-case provided later

Figure 1. Sketch of the proposed approach.

5392 A. AIT-ALLA ET AL.

Figure 2. Procedure Model.

in this article will focus on the number of synchronisa- proposed in this article focuses on the first case. While
tion points between the Digital Twin and the real-world it is possible to specify the required (or expected) func-
system. Thus, a comparably coarse Physical Twin of the tionality of a Digital Twin during this stage, its later
real-world system, mainly depicting the general manu- implementation as part of this procedure model should
facturing and transportation processes, is sufficient. In only be considered as a prototypical implementation.
contrast, if the simulation study would focus on charac-
teristics of, e.g. sensors, a more detailed model of these
components would be required.
Phase 2: design
The situational analysis uses these specifications to
determine which parts of the system need to be modelled During the second phase, software and logistics experts
later to which level of detail. At this stage, several meth- will use the first phase’s specifications to design the mod-
ods for the analysis of complex systems, specifically for els, which will be used in the later simulation study.
transportation or production systems, can be applied to This includes the definition of relevant sub-systems to
obtain the required information about the Digital Twin be implemented as part of the Physical Twin later on.
and the physical system. Again, experts from both sides Thereby, the team needs to specify which capabilities of
of the project need to collaborate to ensure a suitable the real-world system need to be modelled. This could,
analysis of the system. for example, be the capabilities for collision preven-
During this first phase, it is essential to differentiate tion already present in an already used transport sys-
between projects in which the Digital Twin is already tem or already given synchronisation points between
present or at least designed, and projects, which aim the physical system and the digital world, e.g. given by
to develop a Digital Twin first-hand. The procedure mandatory RFID-Gates or similar technology.
 INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 5393

Moreover, following the focus of the initially defined Phase 4: test

simulation study, the corresponding experts need to
The final stage of the design focuses on conducting
define the level of detail required for the Physical Twin.
the specified experiments using the developed models.
While several studies might require Physical Twins on
Depending on the chosen objectives, several methods
the level of processes and production resources, other
could apply for the evaluation of the results. Examples
studies might require very detailed models of single
for such methods are the application of sensitivity analy-
sub-systems, e.g. regarding the communication regarding
sis, Pareto analysis, or regressions to evaluate the system’s
specific characteristics of selected RFID-Readers. In such
responses to specific configurations.
cases, the team will need to be extended by appropriate
Depending on the selected scope of the simulation
experts for the corresponding technologies.
study, these tests can range from classic simulation stud-
In addition to the design of the Physical Twin, this
ies, as, for example, presented in the use-case application
stage incorporates the development and definition of
in the next section, to virtualisation-based evaluations for
suitable simulation experiments to achieve the study’s
studies focussing on specific hardware components, e.g.
objectives. Therefore, we propose to rely on methods
specific RFID-gates -readers, or communication tech-
from the Design of Experiments to determine which
nologies. In the latter cases, dedicated methods for the
factors should be evaluated to which degree to ensure
evaluation may be required, such as tools for tracking
a correct evaluation later on. Alternatively, the replace-
software behaviour, histograms and the occurrence of
ment of the real-world system with the Physical Twin
errors, or software to analyse network traffic.
allows the application of simulation-based optimisa-
Moreover, depending on the scope of the Digital Twin,
tion techniques. Such techniques usually apply meta-
i.e. if it represents a single component, resource, or a com-
heuristic search algorithms to find optimised configu-
plete system, the testing phase might additionally employ
rations using the results of simulation models to assess
approaches from the area of distributed testing. In such
the performance of each configuration. Therefore, the
cases, several Digital Twins might connect to a single
Physical Twin can act as a base-line simulation sce-
Physical Twin. In this case, again, several additional tools
nario, while the meta-heuristic can evaluate different
like network analysis tools or approaches from analysing
configurations as parametrizations of the corresponding
multi-agent systems can be beneficial to assess the overall
Digital Twins.
performance of the evaluated system.
Finally, this phase comprises the selection of suit-
able software environments to implement the Physi-
cal Twin and the corresponding interactions with the Phase 5: Deploy
Digital Twin. In general, we propose the use of avail-
able software environments for material-flow simula- The fifth and final phase of the quality deployment model
tions (e.g. Plant Simulation or AnyLogic) or software covers the deployment of the developed products, pro-
suits for the virtualisation of hardware components. cesses, and activities. In this context, this phase should
Such environments usually facilitate the implementa- focus on the implementation of the most desirable con-
tion and result in easy to understand models which figuration to the real-world system. In this context, it has
increase maintainability. Nevertheless, the selection of to be noted that the proposed procedure model aims at
the environment highly depends on the implementation an offline reconfiguration of existing, already running
of the Digital Twin and its connectivity requirements. systems or an offline configuration of future systems.
E.g. if the Digital Twin connects to the real-world sys- For this reason, this article does not discuss methods
tem via TCP/IP, the Physical Twin must support the same for an online reconfiguration of existing systems. How-
protocol. ever, it proposes the use of the introduced Physical Twin
to replace a running system during the experimenta-
tion. In cases of already existing systems, methods from
the area of change management might prove suitable for
Phase 3: code this task.
The third stage comprises the implementation and vali-
dation of the Physical Twin and the corresponding com-
Analysis: application of the procedure model
munication protocols. In terms of the validation, there
exist several methods to verify and validate simulation To evaluate the proposed approach, we applied it to an
models using experimental design or historical data in exemplary production system. The study presented in
the literature for simulation design and model-in-the- this section aims to determine an optimal interfacing
loop testing. strategy between the real-world system and its Digital
5394 A. AIT-ALLA ET AL.

Twin. Therefore, it presents a step-by-step application of adaptation of the assembly route is possible, e.g. first
the proposed procedure model and presents each step’s having the seal assembled when the bulb stations are
results. occupied or currently out of order, which reduces pro-
duction time (Veigt et al. 2011; Ganji, Veigt, and Scholz-
Phase 1: plan Reiter 2012).
Description of the real-world system: situational
analysis Simulation goals
The production system we are modelling is a real-world For the real-world system, we want to investigate how
demonstrator platform. This system was originally built many sensors; in particular, how many RFID read/write
in the context of production logistics to illustrate the devices are required for the system to ensure the best
advantages of autonomous control to traditional cen- possible performance together with the Digital Twin. As
tralised control (Veigt et al. 2011; Ganji, Veigt, and a secondary parameter, the simulation evaluates differ-
Scholz-Reiter 2012). An image of the demonstrator is ent levels of failure rates for the stations and shuttles
shown in Figure 3. to investigate how the system with a particular sensor
This demonstrator assembles rear lights for cars in configuration performs at different failure rates. This
a job shop manner and consists of three main physi- becomes important, as we assume that additional stations
cal components: Shuttles, stations, and a monorail con- will extend the real-world system during the overhaul.
veying system. The shuttles move along this rail-system Thereby, different stations and variants of the transport
autonomously and have a built-in collision prevention system might be applied, resulting in different levels of
system. It performs the individual assembly at designated reliability.
assembly and testing stations.
Figure 4 shows the five components of such a rear light:
Phase 2: design
The base metal cast, electronics, bulbs, a seal, and a dif-
fuser, usually assembled in this succession (the assembly System description: physical Twin
order is exchangeable for the electronics, bulbs, and the Compared to the real-world counterpart presented in
seal). This demonstrator includes different variants of the previous work (Ait-Alla et al. 2019), this article extends
bulbs (plain, coloured, etc.) and the diffuser (coloured, the Physical Twin in certain aspects to represent a more
dark, etc.), which result in a variety of different rear light realistic and complex system with a higher potential
variants. Every cast contains a radio frequency identifi- for dynamic assembly route adaptation. It increases the
cation (RFID) transponder, and accordingly, the stations number of assembly stations to seven, resulting in the
contain RFID read/write devices, which notify the shut- following list of stations and tasks:
tles of their next target location. Traditionally, each order
would follow a pre-determined assembly route for the (1) One station for task 1 (install electronics into the
shuttle. However, with autonomous control, a dynamic metal cast)

Figure 3. The real-world demonstrator on which the Digital Twin is based on.
 INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 5395

Figure 4. Components of a rear light in the real-world system.

(2) One station for task 2 (install bulbs) which is a task for the Digital Twin. As its last functional-
(3) One station for task 3 (install the seal) ity, the Physical Twin is also able to issue repair orders for
(4) Two stations for task 4 (install the diffuser) the shuttles and stations on request. This function simu-
(5) Two stations for task 5 (final quality check) lates the error messages of the real-world system, which
a human operator sees and reacts to.
Figure 5 shows a sketch of the production system, Note that the Physical Twin itself does not contain any
where the stations and their corresponding tasks are information on the assembly rule for the product and
shown. Every station is placed in a loop, can perform thus will not perform any actions on its own. The Dig-
only one task, and cannot be changed over. The default ital Twin controls the planning of incoming orders and
order of the tasks is consecutive (i.e. 1, 2, 3, 4, 5), but the assembly process order.
the autonomous control can exchange tasks 1–3 in any
succession order, which provides flexibility to the system.
System description: digital Twin
In the real-world system, the shuttles and the monorail
Since the Physical Twin only contains elementary logic
turnouts have a logic embedded to realise the move-
on an embedded level, the Digital Twin acts as the ‘brain’
ments. The shuttles keep moving forward until they
and contains the high-level logic of the system. This logic
arrive at a crossroad or curve. At a crossroads, the loca-
includes the planning of incoming orders and determin-
tion code is read, and depending on the code, the movable
ing the assembly order and, thus, the routes for the shut-
crossroad guides the shuttle towards a loop with a sta-
tles. In line with the real-world system, the exchange of
tion or lets the shuttle keep moving forward to the next
information occurs using an established socket commu-
crossroad or curve. After entering a station loop, the cor-
nication API. The Digital Twin tracks the current system
responding assembly is performed. The shuttle cannot
state of these systems using events. For example, each
leave the loop until the assembly is completed, which
time a shuttle passes an RFID sensor, the Physical Twin
holds significance in case of station failure. At a curve,
sends a message to the Digital Twin, and the location of
the identification code is read, so that only one shuttle
the shuttle is updated. Based upon the available informa-
can pass the curve at a time. Additionally, the shuttles
tion, the Digital Twin decides and writes the next target
contain a laser-based collision prevention system, so that
location code for each shuttle onto its respective RFID
two shuttles will not collide. The laser only detects obsta-
chip.
cles in a line, so the previously mentioned turnout logic
To detect possible shuttle failures, the Digital Twin
is used to prevent collisions on curves.
compares the time ta a shuttle needs to reach the next
Next to these logics, the Physical Twin needs to rep-
RFID sensor with an expected time-of-arrival te (r) for
resent failures of the stations and the shuttles to repro-
the shuttle, which is calculated as follows:
duce realistic behaviours. Here, the stations can detect
their failure state autonomously and report it directly.
In contrast, the shuttles cannot detect breakdowns and, d
therefore, their failures need to be determined indirectly, te (r) = + tw (r)
v
5396 A. AIT-ALLA ET AL.

Figure 5. Sketch of the production system.

Here, d is the distance to the target RFID sensor, v the The expectation regarding the results for the number
shuttle speed, and tw is an additional term to account for of sensors is that a higher number of sensors is benefi-
the idle time of the shuttle waiting at a station that is only cial: With more sensors, the deduction of shuttle failures
non-zero when the next RFID sensor for the shuttle is can be performed timelier, leading to faster issuing of a
located at a station. In that particular case, tw is calculated repair order and thus decreasing overall throughput time.
as follows: The failure rates for the shuttles and stations represent the
 average failure rate per hour. The actual failures occur in
0 if r is not located at a station a stochastic manner. Three levels of failure rate degrees
tw (r) = were chosen to investigate the effects of higher and lower
nr · pr otherwise
reliability of the production system at different numbers
of sensors, resulting in 12 different simulation scenarios.
Here, nr is the number of shuttles already waiting at
To include the dynamics of the real-world system in
the target station with the RFID sensor r, and pr is the
the Physical Twin simulation, the process parameters
processing time at the particular station.
for processing and repair times, as well as the order
When ta exceeds twice the value of te (twice the
arrival frequency, are modelled dynamically. The process
amount is used to account for any factors that might
parameters considered in the simulation are summarised
lengthen the arrival time such as speed downs to pre-
in Table 2. Values given as intervals are modelled as either
vent collisions)The Digital Twin deducts a failure of the
a triangular (T) or a uniform (U) distribution. The trian-
shuttle and sends a repair order to the Physical Twin. In
gular distribution approximates a process with a skewed
summary, Figure 6 provides an overview of the described
normal distribution.
interactions between the two models.

Simulation scenarios, assumptions, and parameters Software environments

Table 1 shows the simulation scenarios and their respec- For the implementation of both twins, the simulation tool
tive parameter values. Due to time constraints for the AnyLogic was used since modelling in the software is
simulations, four values were chosen for the number of done using the Java programming language and allows
sensors: Note that there are always at least eight sensors the inclusion of external Java libraries.
placed in the system, one at every station (including the
start/endpoint). In addition to that, a minimum num-
ber of ten sensors (i.e. 18 sensors in total) were identified
Phase 3: code
in preliminary tests to be the minimum requirement to
ensure reasonably timed synchronisation with the Dig- As part of the implementation, we implemented two dis-
ital Twin. On a higher-end, 60 sensors were chosen as tinct simulation models, one each for the Physical Twin
the maximum value since more sensors would not be and the Digital Twin. For the interaction between the
financially reasonable. Then, linearly spaced values were Digital and the Physical Twin, a communication plat-
chosen for the intermediate values. form had to be established. Therefore, we implemented
 INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 5397

Figure 6. Interactions between the Physical and the Digital Twin.

Table 1. Simulation scenarios. Table 2. Process parameters.

Number of Station and shuttle Process parameter Value
Scenario identifier sensors failure rate (per hour)
Station processing time Tasks 1, 2, and 3: 60–90 sec. (T)
S1 18 1 Tasks 4 and 5: 160–210 sec. (T)
S2 18 2 Task 5, in the case of a product
S3 18 4 defect: 300–420 sec. (T)
S4 32 1 Station repair time 120–180 sec. (U)
S5 32 2 Shuttle repair time 60–90 sec. (U)
S6 32 4 Order arrival frequency Every 120–150 sec. (U)
S7 46 1 Number of generated orders 40
S8 46 2 Size of the simulated monorail system 228 m × 96 m
S9 46 4 Speed of shuttles 1 m/s
S10 60 1 Number of shuttles 10
S11 60 2
S12 60 4
5398 A. AIT-ALLA ET AL.

Figure 7. Model design of a shuttle performing an assembly task in the Physical Twin – state model.

a socket-based TCP/IP protocol to interconnect the two with each other by sending specific messages that initiate
different instances of AnyLogic. and trigger events.
In AnyLogic, the shuttles, stations, and the other enti- Both the physical and the digital twin, have clocks that
ties are modelled as agent objects provided by the simula- synchronise the data-exchanges and start synchronously,
tion tool. Each agent has its logic programmed into it and so that system time always remains the same. Via socket
is modelled using state diagrams or a process-modelling as communication API using Transmission Control Pro-
library to describe the event-driven and time-controlled tocol (TCP), we develop a process for transferring the
behaviour of the agents. Figure 7 shows an example of data. To handle the communication between the Physi-
such a state diagram for the shuttle. The agents interact cal and Digital Twin, a data model and format has been
 INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 5399

developed. In this case, if the Physical Twin sends the results of the scenarios with 35 and 50 sensors shows that
actual system state, this data will be matched to the the reduction is quite minimal when using more than
defined model and format before being sent. Each time 35 sensors (i.e. comparing S3 (7387) to S4 (7359), S7
a data is received, the system checks the data format (7620) to S8 (7536) and S11 (8014) to S12 (8018)). So,
and accordingly forwarded the received data to the cor- while having fewer sensors leads to delayed updates and
responding process. Once the Digital Twin has made a thereby reactions in the Digital Twin regarding changes
decision, it is adapted to the data model and format and and disturbances in the Physical Twin. Having more sen-
sent to the Physical Twin. sors than a certain amount leads to diminishing returns
regarding performance since not all reported states from
Phase 4: test the Physical Twin need a real-time reaction. We conclude
that the deployment of a high number of sensors only
To evaluate the stated objectives from Phase 1 using makes sense in scenarios with a very high probability of
the designed scenarios from Phase 2, we considered the disturbances.
following performance measures: As a standard perfor- Figure 9 shows the number of received and emitted
mance measure, namely, the mean throughput time was messages for the Digital Twin. A comparison regarding
considered to measure the production performance of the number of exchanged messages shows that increasing
the system. Additionally, we recorded the number of the number of sensors automatically leads to an increase
emitted and received messages between Physical and of the exchanged messages, which is to be expected.
Digital Twin, as these are relevant measures to describe Increasing failure rates strongly affect the number of
the quantity of communication. Finally, the shuttle and emitted messages. This is because each time the Digi-
the station utilisation were considered. tal Twin expects a shuttle failure, a notification is sent
Figure 8 shows the throughput time for the assembly to the Physical Twin. In cases with higher failure rates,
of 40 rear-light (40 orders processed in each scenario) more notifications are emitted (i.e. comparing S1 (345)
for all scenarios. In order to obtain meaningful results, to S2 (400) to S3 (484) to S4 (571), S5 (350) to S6 (406)
the number of repetitions per simulation scenario is set to S7 (689) to S8 (1178) and S9 (520) to S10 (603) to S11
to four replications to account for the stochastic nature (1087) to S12 (1805)). We found out that the number of
of the scenarios. From Figure 8 it turns out that the messages increased linearly to the number of installed
mean throughput time is reduced independently from sensors.
the selected failure rate when using more sensors (i.e. Finally, Figure 10 depicts the mean station utilisation.
comparing S1 (7387) to S3 (7389), S5 (8102) to S7 (7620) According to the results, the station utilisation increases
and S9 (8946) to S11 (8014)). However, comparing the

Figure 8. Mean throughput time of the scenarios.

5400 A. AIT-ALLA ET AL.

Figure 9. Total number of received and emitted messages between the frontend and backend model.

with the number of sensors. However, the station utili- an optimal interconnection. Nevertheless, at this stage,
sation decreases with the failure rate, which has a bigger the additional costs of more sensors should be con-
impact, considering the scenarios (S1 to S5 to S9, S2 to sidered. Each additional sensor requires investments,
S6 to S10, S3 to S7 to S11 and S4 to S8 to S12). This is and thus, it would be well advised to evaluate those
because, in the scenarios with high failure rates, many investments against the achievable performance. Figure 8
disturbances result from shuttle defects, which leads to clearly states that the increase in performance only comes
unoccupied stations, and, thus, a low station utilisation. at diminishing returns.
This explains why the station utilisation in the same Moreover, the results show that increasing the number
failure rate configuration increases with the number of of sensors also increases the number of received mes-
deployed sensors (S1 to S2 to S3 to S4, S5 to S6 to S7 to sages. While this result has been expected, it indicates
S8, S9 to S10 to S11 to S12). Utilisation of more sensors that for larger systems with many sensors, the reliability
leads to a quick detection of shuttle defect, a quick main- of the communication technology has to be considered.
tenance of defect shuttles, and, therefore, to an increase of In the case of this simulation study, both instances of
the station utilisation. Although the results of the sensor AnyLogic were executed on the same computer using the
configurations with 35 and 50 sensors show that the sta- loopback interface (localhost) for the TCP/IP connec-
tion utilisation remains quite the same, this fact leads to tion. Nevertheless, during the tests, it became apparent
the conclusion that a further decrease in failure detection that increasing the simulation speed beyond a certain
times does not further improve the overall performance threshold resulted in lost messages, as the interface was
of the system. not capable of handling the quick succession of mes-
sages for larger scenarios. The current implementation
allows executing both instances on different computers
Phase 5: deployment to evaluate possible communication technologies. This
becomes important in case of deployment of the pro-
This article does not focus on the actual implemen- posed changes, as different technologies could be avail-
tation in a real-world system. Nevertheless, this sub- able, again at different costs.
section provides information on further studies that
would be required before implementation could take
place. Conclusion and future work
In conclusion, the results confirm the general assump-
Summary
tion that more sensors also result in higher produc-
tion performance of the system. Nevertheless, the results This article presents a procedure model to evaluate
demonstrate that somewhere between 35 and 50 RFID and configure the interconnection between a physical,
sensors, a saturation of the system occurs. In the next real-world system and its Digital Twin using concepts
step, a more detailed simulation study could be applied of software-in-the-loop testing. Therefore, it proposes
to determine the exact number of sensors required for to represent the real-world system using a simulation
 INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 5401

Figure 10. Mean station utilisation of the scenarios.

model, denoted as Physical Twin. Thereby, this Physical Limitations

Twin should not include any additional logic and mimic
While the presented methodology allows configuring the
the physical system as close as it is necessary. Using this
interface between the digital and the physical system, it
simulation, the synchronisation between the Digital and
also implies some limitations and risks. The introduction
the Physical Twin can be evaluated and configured using
of the Physical Twin simulation model requires the
a series of simulation studies. To evaluate the proposed
design and implementation of an additional simulation
procedure mode, this article presented an application
model. Fortunately, it can be assumed that the required
for the extension of an autonomous transportation and
model is already (partly) implemented as part of the Dig-
assembly system.
ital Twin. A Digital Twin requires a representation of the
system it is designed to control and monitor. By remov-
ing the additional logic, a Physical Twin model could be
Discussion and contribution deducted with only few additional efforts. Nevertheless,
The application of the procedure model shows that the in cases where the study aims to evaluate or configure
use of a so-called Physical Twin provides several advan- new (aspects of) production systems, the Physical Twin
tages. On the one hand, it facilitates simulation studies requires additional information, e.g. about the reliability
for running or not-yet-existing systems, as the Physi- of components or similar technical characteristics. This
cal Twin replaces the actual real-world system. On the information demand implies additional risks compared
other hand, the proposed setup allows conducting several to a classic, prototype-based evaluation. If such infor-
distinct studies to plan different aspects of the (future) mation is not accurate, the simulation study can deliver
real-world system. While this article mainly focused on unreliable or even false results.
the interaction of the Digital Twin with its physical coun-
terpart, it already included systemic characteristics like
Future work
failure rates. Using a simulation model instead of a real-
world system allows much more freedom of choice and Future work will focus on extending the procedure with
induces fewer costs than actually modifying the physical more detailed methods for the proposed steps. On the
system. In conclusion, the evaluation shows that the pro- one hand, this will focus on extending the deduction
posed methodology is suitable to evaluate and configure of the Physical Twin model, by providing guidelines on
the interaction between the physical and the digital sys- important aspects to include. On the other hand, the cost-
tem, as highlighted in the subsection on the research gap. ing and investment aspect mentioned in the use-case’s
Moreover, the proposed methodology can be applied to fifth phase will be integrated into the procedure in a more
other (simulative) evaluations as well. These include, e.g. structured way. Another stream of future research will
the evaluation of different control strategies or different evaluate the deployment phase in more detail. There-
layouts for the production system. fore, the proposed methodology can be extended to cover
5402 A. AIT-ALLA ET AL.

a comparable process found in the area of Hardware- International Journal of Production Research 57 (3): 829–846.
In-The-Loop testing. Therefore, the proposed simulation doi:10.1080/00207543.2018.1488086.
models could be gradually refined to emulate/virtualize Kritzinger, Werner, Matthias Karner, Georg Traar, Jan Henjes,
and Wilfried Sihn. 2018. “Digital Twin in Manufacturing:
certain hardware aspects of the system and thus facili- A Categorical Literature Review and Classification.” IFAC-
tate deployment. Nevertheless, this stream of future work PapersOnLine 51 (11): 1016–1022. doi:10.1016/j.ifacol.2018.
requires additional research in terms of suitable methods 08.474.
and feasibility. Kumar, Ramesh, L. Ganapathy, Ravindra Gokhale, and Manoj
K. Tiwari. 2020. “Quantitative Approaches for the Integra-
tion of Production and Distribution Planning in the Supply
Disclosure statement Chain: a Systematic Literature Review.” International Journal
No potential conflict of interest was reported by the author(s). of Production Research, 1–27. doi:10.1080/00207543.2020.
1762019.
Lee, Edward A. 2008. “Cyber Physical Systems: Design Chal-
ORCID lenges.” 11th IEEE International Symposium on Object
Abderrahim Ait-Alla http://orcid.org/0000-0001-7031-0969 and Component-Oriented Real-Time Distributed Comput-
Markus Kreutz http://orcid.org/0000-0002-1147-467X ing (ISORC), 363–69.
Lee, Edward A., and Sanjit A. Seshia. 2017. Introduction to
Embedded Systems: A Cyber-Physical Systems Approach. 2nd
References ed. Cambridge, MA: MIT Press.
Lee, Jay, Behrad Bagheri, and Hung-An Kao. 2015. “A Cyber-
Ait-Alla, A., M. Kreutz, D. Rippel, M. Lütjen, and M. Fre-
Physical Systems Architecture for Industry 4.0-Based Man-
itag. 2019. “Simulation-based Analysis of the Interaction
ufacturing Systems.” Manufacturing Letters 3: 18–23.
of a Physical and a Digital Twin in a Cyber-Physical Pro-
Leitão, Paulo, Armando W. Colombo, and Stamatis Karnouskos.
duction System.” IFAC-PapersOnLine 52 (13): 1331–1336.
2016. “Industrial Automation Based on Cyber-Physical Sys-
doi:10.1016/j.ifacol.2019.11.383.
tems Technologies: Prototype Implementations and Chal-
Boulekrouche, Boubaker, Nafaâ Jabeur, and Zaia Alimazighi.
lenges.” Computers in Industry 81: 11–25.
2016. “Toward Integrating Grid and Cloud-Based Concepts
Leitão, Paulo, Nelson Rodrigues, José Barbosa, Claudio Turrin,
for an Enhanced Deployment of Spatial Data Warehouses
and Arnaldo Pagani. 2015. “Intelligent Products: The Grace
in Cyber-Physical System Applications.” Journal of Ambi-
Experience.” Control Engineering Practice 42: 95–105.
ent Intelligence and Humanized Computing 7 (4): 475–487.
Liu, Qiang, Hao Zhang, Jiewu Leng, and Xin Chen. 2019.
doi:10.1007/s12652-016-0376-1.
“Digital Twin-Driven Rapid Individualised Designing of
Ganji, Farideh, Marius Veigt, and Bernd Scholz-Reiter. 2012.
Automated Flow-Shop Manufacturing System.” Interna-
“Prototyping in Research Domains: A Prototype for
tional Journal of Production Research 57 (12): 3903–3919.
Autonomous Production Logistics.” The Impact of Vir-
doi:10.1080/00207543.2018.1471243.
tual, Remote, and Real Logistics Labs: First International
Nguyen, Ha T., Guangya Yang, Arne H. Nielsen, and Peter H.
Conference, ImViReLL 2012, Bremen, Germany, February
Jensen. 2019. “Hardware- and Software-in-the-Loop Sim-
28–March 1. Communications in Computer and Informa-
ulation for Parameterizing the Model and Control of Syn-
tion Science. Vol. 282, edited by Dieter Uckelmann, Bernd
chronous Condensers.” IEEE Transactions on Sustainable
Scholz-Reiter, Ingrid Rügge, Bonghee Hong, and Antonio
Energy 10 (3): 1593–1602. doi:10.1109/TSTE.2019.2913471.
Rizzi, 81–89. Berlin: Springer.
Parente, Manuel, Gonçalo Figueira, Pedro Amorim, and
Grieves, Michael, and John Vickers. 2017. “Digital Twin:
Alexandra Marques. 2020. “Production Scheduling in the
Mitigating Unpredictable, Undesirable Emergent Behavior
Context of Industry 4.0: Review and Trends.” International
in Complex Systems.” In Transdisciplinary Perspectives on
Journal of Production Research, 1–31. doi:10.1080/00207543.
Complex Systems. Vol. 89, edited by Franz-Josef Kahlen,
2020.1718794.
Shannon Flumerfelt, and Anabela Alves, 85–113. Cham:
Park, Heejong, Arvind Easwaran, and Sidharta Andalam.
Springer International Publishing.
2019. “Challenges in Digital Twin Development for Cyber-
Hellerer, Matthias, Martin Schuster, and Roy Lichtenheldt.
Physical Production Systems.” Cyber Physical Systems.
2016. Software-In-The-Loop Simulation of A Planetary Rover.
Model-Based Design. Lecture Notes in Computer Science.
Hribernik, Karl, Thorsten Wuest, and Klaus-Dieter Thoben.
Vol. 11615, edited by Roger Chamberlain, Walid Taha, and
2013. “Towards Product Avatars Representing Middle-of-
Martin Törngren, 28–48. Cham: Springer International Pub-
Life Information for Improving Design, Development and
lishing.
Manufacturing Processes.” In Digital Product and Pro-
Piascik, R., J. Vickers, D. Lowry, S. Scotti, J. Stewart, and
cess Development Systems: IFIP TC 5 International Confer-
A. Calomino. 2010. “Technology Area 12: DRAFT Mate-
ence, NEW PROLAMAT 2013, Dresden, Germany, October
rials, Structures, Mechanical Systems, and Manufactur-
10–11. IFIP Advances in Information and Communication
ing Road Map.” NASA Office of Chief Technologist.
Technology. Vol. 411, edited by George L. Kovacs and Detlef
https://www.nasa.gov/pdf/501625main_TA12-MSMSM-
Kochan, 85–96. Berlin: Springer.
DRAFT-Nov2010-A.pdf.
Isermann, Rolf. 2006. Fahrdynamik-Regelung. Wiesbaden:
Polini, Wilma, and Andrea Corrado. 2020. “Digital Twin of
Vieweg.
Composite Assembly Manufacturing Process.” International
Ivanov, Dmitry, Alexandre Dolgui, and Boris Sokolov. 2019.
Journal of Production Research, 1–15. doi:10.1080/00207543.
“The Impact of Digital Technology and Industry 4.0
2020.1714091.
on the Ripple Effect and Supply Chain Risk Analytics.”
 INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH 5403

Rossi, Iacopo, Adrien Reveillere, and Alberto Traverso. 2019. Veigt, Marius, Farideh Ganji, Ernesto M. Kluge, and Bernd
“Flexibilization of Gas Turbine Combined Cycle via Heat Scholz-Reiter. 2011. “Autonomous Control in Production
Pump: Development of Control Logics via Software-in-the- Planning and Control: How to Integrate Autonomous
Loop Application.” E3S Web of Conferences 113: 1004. Control Into Existing Production Planning and Con-
Schluse, Michael, Marc Priggemeyer, Linus Atorf, and Juer- trol Structures.” In Autonomous Cooperation and Control
gen Rossmann. 2018. “Experimentable Digital Twins- in Logistics: Contributions and Limitations – Theoretical
Streamlining Simulation-Based Systems Engineering for and Practical Perspectives, edited by Michael Hülsmann,
Industry 4.0.” IEEE Transactions on Industrial Informatics 14 Bernd Scholz-Reiter, and Katja Windt, 313–329. Berlin:
(4): 1722–1731. doi:10.1109/TII.2018.2804917. Springer-Verlag.
Tao, Fei, Fangyuan Sui, Ang Liu, Qinglin Qi, Meng Zhang, Vogel-Heuser, Birgit, Jay Lee, and Paulo Leitão. 2015. “Agents
Boyang Song, Zirong Guo, Stephen C.-Y. Lu, and A. Y. C. Enabling Cyber-Physical Production Systems.” At - Automa-
Nee. 2019. “Digital Twin-Driven Product Design Frame- tisierungstechnik 63 (10). doi:10.1515/auto-2014-1153.
work.” International Journal of Production Research 57 (12): Wang, Jinjiang, Lunkuan Ye, Robert X. Gao, Chen Li, and
3935–3953. doi:10.1080/00207543.2018.1443229. Laibin Zhang. 2019. “Digital Twin for Rotating Machin-
Toader, Florentina A. 2017. “Production Scheduling in Flexible ery Fault Diagnosis in Smart Manufacturing.” Interna-
Manufacturing Systems: A State of the Art Survey.” Journal tional Journal of Production Research 57 (12): 3920–3934.
of Electrical Engineering, Electronics, Control and Computer doi:10.1080/00207543.2018.1552032.
Science 3 (7): 1–6. Wiesner, Stefan, and Klaus-Dieter Thoben. 2017. Cyber-
Uhlenkamp, Jan-Frederik, Karl Hribernik, Stefan Wellsandt, Physical Product-Service Systems. New York: Springer Inter-
and Klaus-Dieter Thoben. 2019. “Digital Twin Applications: national Publishing.
A First Systemization of their Dimensions.” In Proceedings, Zhang, Hao, Qiang Liu, Xin Chen, Ding Zhang, and Jiewu Leng.
2019 IEEE International Conference on Engineering, Technol- 2017. “A Digital Twin-Based Approach for Designing and
ogy and Innovation (ICE/ITMC): Sophia Antipolis Innovation Multi-Objective Optimization of Hollow Glass Production
Park, France, June 17–19, 1–8. Piscataway, NJ: IEEE. Line.” IEEE Access 5: 26901–26911.