Proceedings of the 2023 Winter Simulation Conference

C. G. Corlu, S. R. Hunter, H. Lam, B. S. Onggo, J. Shortle, and B. Biller, eds.

DIGITAL TWINS: FEATURES, MODELS, AND SERVICES

Andrea Matta Giovanni Lugaresi

Department of Mechanical Engineering Department of Mechanical Engineering

Politecnico di Milano KU Leuven
2023 Winter Simulation Conference (WSC) | 979-8-3503-6966-3/23/$31.00 ©2023 IEEE | DOI: 10.1109/WSC60868.2023.10407260

Via La Masa 1 Celestijnenlaan 300

20156, Milano, ITALY 3001 Leuven, BELGIUM

ABSTRACT
This work provides an overview of digital twins, digital replicas of real entities conceived to support
analysis, improvements, and optimal decisions. Specifically, it aims to better clarify what digital twins are
by pointing out their main features, what they can do to support their related physical twins, and which
models they use. An illustrative case together with a few selected application examples is used to better
describe digital twins. A discussion on the actual challenges and research opportunities is also reported.

1 INTRODUCTION
With the coming of the Industry 4.0 wave, digital representations of products and manufacturing systems
have been considered central for optimizing their development, production, and delivery phases. Digital
twins (DTs) are not simply conceived as simulation models of their physical counterparts for offline what-if
analysis. They are developed as self-adaptable and empowered decision-makers timely aligned with the
dynamics of the real entity. The global DT market size was valued at 8.6 billion USD in 2022 and it is
expected to reach 137.7 billion USD by 2030 with a Compounded Average Growth Rate (CAGR) of 42.6%
(Fortune Business Insights 2023). According to a recent survey, only around 5% of companies affirm DTs
are not part of their digital transformation strategy (Dertien and Macmahon 2022) whereas another 86%
consider DTs a crucial solution in their strategy. Also, DTs are subject to international standardization
efforts (ISO 23247:2021 2021).
DTs are conceived to mirror physical entities, independently from the domain. As a consequence,
the variety of applications surveyed by recent literature is vast (Attaran and Celik 2023). Among these,
manufacturing, transportation, agriculture, construction, and healthcare are the major domains of DT
applications (Liu et al. 2023). Depending on the application, different types of DTs can be distinguished:

• Product Digital Twin. The digital replica mirrors a physical object from its manufacturing phase to its
disposal along its whole life cycle. The DT collects and analyzes data collected from manufacturing
processes as well as from customers’ use to provide valuable feedback to improve the product
design phase. Three-dimensional representations of products are relevant to simulate the physical
behavior of products in specific situations such as machining processes and disassembly operations.
• System Digital Twin. The digital replica mirrors a complex system, i.e. a collection of parts
organized for some purpose (Coyle 1997). Examples are production lines, automated warehouses,
traffic systems, etc. The main purpose of system DTs is to support decision-makers in improving
operational efficiency, effectiveness, and costs. Since the time synchronization of activities and
resource availability are the core elements, in general, these DTs do not make larger use of geometrical
or physical models.
• Environment Digital Twin. The digital replica mirrors an environment or a place. Examples of
applications are working environments, entertainment places, etc. The main purpose of place DTs
is to provide an immersive environment in which the analyst can better evaluate the physical

979-8-3503-6966-3/23/$31.00 ©2023
Authorized licensed use limited IEEE University. Downloaded on December
to: Zhejiang 46 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

counterpart. In this case, a link with the metaverse exists and would deserve further clarification.
Virtual Reality and Augmented Reality are key technologies for developing the models used by
this type of DTs (Attaran and Celik 2023).
• Biological Digital Twin. The digital replica mirrors a human being, or part of him(her), or any other
biological system such as a plant or a fish farm. The main purpose is to provide support in medical
and life science domains for alert predictions, surgical operations, environment control, etc. Yet,
the development of human DTs is still in the early stages (Ahmed and Devoto 2021; Jimenez et al.
2020).

The variety of functionalities that DTs can potentially offer is very large encompassing different
application fields from aerospace to urban traffic. This extremely large spectrum of DT use cases has
caused a multitude of scientific contributions and market studies from several disciplines, each one with its
own terminology, models, and approaches. The scope of this work is to help readers clarify the basic concepts
of DTs as well as their underlying models and possible applications. Particular attention is dedicated to
simulation in relation to DTs by discussing its key modeling role and related research challenges.

2 DIGITAL TWIN FEATURES

The DT concept was first introduced by Grieves in the early 2000s. The DT “is a set of virtual information
constructs that fully describes a potential or actual physical manufactured product from the micro atomic
level to the macro geometrical level. At its optimum, any information that could be obtained from inspecting
a physically manufactured product can be obtained from its digital twin” (Grieves and Vickers 2017). From
this definition, it appears the DT was conceived to support product-related decisions from the beginning-
of-life phase to the end-of-life phase in a product life-cycle management approach to close the loop from
production, use, and disposal to design phases. Many other definitions have been proposed in recent literature
to emphasize specific DT features such as the integration and interconnections of DT’s elements, the digital
counterpart, the predictive capabilities, and the descriptive power. Barricelli et al. (2019) classified 75
papers according to their DT definitions from manufacturing, aviation, and healthcare domains. Other
definitions of DTs that are more customized for manufacturing, aviation, and healthcare domains can be
found in Negri et al. (2017), Xiong and Wang (2022), and Croatti et al. (2020), respectively.

2.1 Basic Features

Considering the multi-domain applications of the DT technology and its multiple uses, a definition of
DT would hardly satisfy all the requirements. Therefore, it seems more practical to list the fundamental
attributes of DTs and discuss their relevance. Figure 1 summarizes the DT features.
Digital. Mirroring physical objects within digital entities to train staff and explore new solutions is not
a new concept. Perhaps, the most famous example is the digital environment for simulating the Apollo 13
mission (Barricelli et al. 2019). The increase in computational power, data accessibility, and availability
of modeling software environments, have created the ideal conditions to develop digital representations of
physical entities affordable with cost and time. This feature is universally recognized by literature, i.e.,
the DT is a set of instructions coded in a computer program to describe the physical entity’s behavior.
Digitization allows for exploiting the high calculation speed of computers and for obtaining accurate answers
in a short time.
Descriptive. The main purpose of the DT is to provide information and knowledge about the physical
entity, e.g., its state, its emerging behavior, its trends, or anything that may help the manager of the physical
entity to improve its performance. This feature is particularly relevant when the physical entity cannot be
reached (e.g., a space shuttle, or a ship) or it is not easy to extrapolate the information (e.g., a human body
or a bulk deformation process at high strain rates), or it necessitates sophisticated models and algorithms
for knowledge extraction. Coupling the human ability to conceptualize the problem and its context from
visualization of tables of numbers, graphs, and other symbolic information, with the computer speed of

47
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

Figure 1: Basic features (inner circle) and advanced features (outer circle) of Digital Twins.

processing information when executing step-by-step operations offers unprecedented opportunities (Grieves
2014). This feature allows DTs to provide functionalities such as status visualization, monitoring, analysis
of observed behaviors, diagnosis of malfunctioning, prediction of failures, etc.
Synchronous to physical. The DT should be able to describe the physical entity at any moment.
This feature implies the state changes of the physical entity being transferred to the DT. In large and
geographically distributed systems, this communication might not be trivial. As far as physical-to-digital
communication, significant examples are from heavy industries (e.g., cement, steel, chemical, etc) or power
plants, which require continuously updated status of their high-investment equipment. Physical-to-digital
communication must be automated as widely recognized by literature, but vice versa there is no unanimous
consensus. Indeed, the descriptive information provided by the DT can be used to make a corrective action
on the physical entity; this digital-to-physical communication can be asynchronous (i.e., the implementation
of the action has a time delay with respect to the decision time), automated (e.g., the cutting parameters of a
machine tool are changed after quality inspections) or manual (e.g., the operator starts product changeover
in a manufacturing line).

2.2 Advanced Features

In addition to the fundamental features described in the previous section, DTs can be designed also
considering advanced attributes as described in the following (see also Figure 1).
Predictive. The alignment with the physical entity state together with the availability of high-speed
computational power allows DTs to numerically simulate the future periods under some well-defined
scenarios. Simulation results can be used for several purposes such as estimating system performance,
checking deliveries, supporting resource allocation, etc. This feature can be very relevant in several contexts,
but it is not a strict requirement for a DT. Indeed, for DTs representing single products such as engines,
compressors, etc, this feature might not be necessary. On the other hand, it can be relevant in all those
situations affected by large complexity and high uncertainty that necessitate experiments with stochastic
simulation models, e.g. semiconductor manufacturing systems. One of the first examples of this predictive
functionality is the Computer Numerically Control (CNC) of machine tools, which simulates online the
tool trajectory to control the velocity and acceleration of machine drives in the scenario defined by the part
program. A significant trend is also to use the simulation capabilities of DTs for virtual commissioning
of complex systems such as machine tools (Wang et al. 2023) and automotive factories (Mykoniatis and
Harris 2021).
Prescriptive. Providing automated feedback to the physical entity in a closed-loop control approach
is another advanced feature of DTs. This feature fits especially when the system complexity is high and
unmanageable by humans, in these cases, the DT can explore a large number of alternatives and select
the best one. Another situation appears when the required feedback time is short. Automation particularly

48
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

helps when the physical product is difficult to reach or when the decision is repetitive. From this point of
view, a programmable logic controller can be considered a primitive DT with simple prescriptive abilities.
Lastly, high costs or risks of performance losses may require automated feedback from digital to physical
(e.g., in power plants). When a DT does not have a prescriptive feature, it is reasonable to assume that
humans will take prescriptive actions after having acquired enough information from the DT.
Adaptive. DTs are normally coupled with physical systems of a certain relevance and costs, which tend
to have a long life-cycle during which a lot of changes happen either in the system or in the environment
and the context where they operate. DTs should keep their descriptive ability along the whole life-cycle
of the physical entity despite changes. Examples are the downgrading of equipment, the increased car
traffic in a city, and the opening of a new airport runway. As a consequence, models used to provide DT
services might also need updates and dramatic changes to be consistent with the new situation. Automated
adaptation of DTs to new system changes and new situations is considered an advanced feature of DTs.
Currently, most implementations require manual intervention to the software code, limiting the adaptability
of DTs (Tavakoli et al. 2008).
Granularity. Grieves defines DTs able to represent a product from the micro-atomic level to the
macro-geometrical level. (Grieves and Vickers 2017). The same consideration can be done for a DT of
a factory from the equipment level to the plant level. The fundamental motivation is that DTs are used
to support humans for different purposes. Therefore the descriptive ability of DTs must be cross-level
encompassing the different fidelities available. Further, DTs must also be consistent with their uses. For
instance, if the DT is used for production scheduling on a shop floor, the simulation of tool breakages
and machine tool structure deformation does not add much value. Hence, the granularity aspect should be
considered an advanced feature, not necessarily available in a DT.

2.3 Digital Models, Digital Shadows, and Digital Twins

A recent trend is to call DT any digital representation of a physical object. This large adoption of the DT
term may be a source of confusion and misleading because the functionalities provided by digital models
can vary from case to case. This section aims at clarifying the usage of different digital representations.
Since the origins of computer simulation in the ’60s (Tocher 1967), digital models are widely used to
numerically simulate complex products and systems for the estimation of their key performance measures.
Digital models are typically used offline and do not necessarily have to represent existing objects. Indeed,
digital models are widely used to evaluate detailed design alternatives during the engineering phases.
Further, there is often no automated data flow from the physical entity to the digital model. Vice versa, the
results from simulation experiments should help to make and implement decisions on the physical entity.
However, this would generally be done manually. In relation to the DT main features described in Figure 1,
digital models are digital by definition as well as descriptive. Classical examples of digital model uses are
assembly line balancing, layout planning, resource allocation, etc. Another term used in recent literature
is digital factory. This is related to digital models with extensive use of 3D representations of the whole
factory often used for design purposes.
When data management is affordable, the feeding of digital models with data coming from the physical
entity becomes natural. This automated physical-to-digital data flow started to be evident in the 80s, with
the advent of the Computer Integrated Manufacturing framework (Williams 1989). The term digital shadow
is widely used in recent literature to emphasize that the digital model is aligned with the physical entity
(Kritzinger et al. 2018). This implies that the digital model represents the actual state of the physical entity,
so querying the digital shadow or the physical entity is indistinguishable. Many important functionalities
can be provided by digital shadows such as state visualization, monitoring, alert prediction, etc. Since the
digital shadow follows the physical entity, this last must exist. In relation to the DT main features described
in Figure 1, digital shadows are digital, descriptive, and synchronous to physical. Recognizing the status
of DTs to digital shadows is an open debate (Bergs et al. 2021) yet and out of the scope of this tutorial.
We simply point out that digital shadows own the fundamental features of DTs.

49
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

3 DIGITAL TWIN ELEMENTS

According to the conceptual five-dimension model proposed by Tao and Zhang (2017), a DT can be
developed on top of different types of models that encompass five elements, namely physical entity, virtual
or digital representations of the system, data, connections, and services as represented in Figure 2. This
model can effectively convey the main topics that regard DTs, outlining the role of each dimension.

3.1 Physical Dimension

The physical entity represents the real-world physical or perceived system, for instance, a manufacturing
system or a machining process that is dynamically connected via a communication or integration medium.
In (Tao and Zhang 2017), the physical entity is summarily defined as a set of objects composing the
physical entity. However, the knowledge about the physical entity might be far from perfect, and, for
this reason, the DT with the acquired sensor data may be used to improve such knowledge. The physical
entity model is, therefore, the model of the physical entity with the current knowledge. The model can be
any representation of its components and relationships. A variety of models can be used to describe the
physical entity, from the least informative ones such as a simple list of components and sub-components
to a more formal one such as a class diagram, an Entity Relationship Graph, or the percentages of the
material chemical composition. This model can be improved when the knowledge and understanding of
the physical entity increase.

3.2 Virtual or Digital Dimension

Depending on the requirements, the virtual entity can be composed by using either discrete events simulation,
continuous simulation, or hybrid simulation (Robinson 2014). As already mentioned in section 2, the
prediction capability can also be provided by analytical models or simple formulas. However, in this case,
it would not be worth considering it as a DT.
In general, multiple digital models can be used to describe the physical entity, each dedicated to a
specific physical entity’s behavior. The choice of the model should depend on the purpose of use and,
therefore, to which service the model will be dedicated. For example, a machine reliability model can be
used to schedule maintenance operations, whereas a material flow model will be used to predict the system
service level. Another example is a simple model providing the maximum load of a product based on the
minimum section and the nominal ultimate tensile strength of the material. In contrast, more sophisticated
models would represent the product with FEM equations, material properties, and a solid model. Simulation
models offer deep descriptions of the physical counterpart with insights to support decision-making, and
predicting anomalies or future failures (VanDerHorn and Mahadevan 2021).
The virtual representations in DTs are mainly of two types: model-based providing structural information
about the physical entity, and model-free representing what was observed in the past. DTs can use both
types depending on the service requirements. The following subsections explain the main differences.

3.2.1 Model–based
Model-based digital twins rely on models to describe the behavior of the physical entity (e.g., physics-based
models, Discrete Event Simulation (DES) models, analytical models). For example, physics models are
typically expressed by mathematical equations that describe the physical laws governing the system’s
behavior. The related variables can, therefore, incorporate data from sensors to calibrate the model and
make accurate predictions. An example of a simple model is Taylor’s formula which provides the increased
age of a cutting tool based on the cutting speed used by the machine spindle (Mills and Redford 1983).
A more advanced model would be a FEM model to simulate the tool wear. Taking a production line as
an example of a physical entity, the production rate of the system in the shift can be given by a simplified

50
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

model that takes into account the status of the bottleneck machine. A more detailed model would simulate
the operations until the end of the shift to estimate the number of parts produced in the remaining period.
Model-based digital twins are particularly useful for systems with well-defined physical laws and
behavior, such as mechanical or electrical systems, as well as systems with a systematic structure, such as
discrete-part manufacturing systems. Of course, model-based digital twins are computationally demanding
and generally require much more input information and longer response times. To lighten these virtual
models, meta-modeling approaches can be used to simplify the model into a simpler and lighter model by
fitting equations from simulation output data. Examples of meta-models, also known as surrogate models,
are linear regression, or other types of non-linear regression such as Kriging, Kernel, neural networks, etc.
Such lighting techniques can be used either to simplify deterministic models or stochastic models.

3.2.2 Model–free
Model-free digital twins can rely on data-driven algorithms to detect patterns, anomalies, and correlations
in data collected from the physical entity. These DTs do not require an explicit model of the system
and rely on the data generated by it to provide valuable insights. For instance, a model-free DT can be
equipped with machine learning algorithms to identify process parameters that affect its throughput or yield
and construct recommendations to improve its efficiency. A granularity feature allows the use of different
models with different detail levels, accuracy, costs, and times.

3.3 Service Dimension

The functionalities provided by DTs are embedded in services accessible by humans or other physical
and digital objects. Service types can be different depending on the specific application. For example,
optimization methods can be combined with simulation methods in simulation-optimization models. In
this approach, the service module would utilize techniques such as gradient-based or greedy algorithms to
search for the best possible solution. For each type of service, a proper model is needed to be defined.
Section 5 presents a selection of relevant services of DTs.

3.4 Data Dimension

In order to accurately reflect the behavior of the real-world entity or to enhance its performance, a DT must
be equipped with the ability to automate data acquisition (Dittmann et al. 2021). There are three possible
data sources in a DT: the real-time (dynamic) data stream input from the physical entity, the baseline (static)
data that was used in building the model for the first time, and the human expert knowledge (optional). DTs
use various models and techniques to manage, fuse, and process the extensive and diverse data gathered
from the physical entity. The most commonly used data models are relational data models, object-oriented
data models, hierarchical data models, etc. More advanced data models such as knowledge graph models
can allow for rich and flexible data representation to connect data and infer reasoning among diverse data
(Esser and Fahland 2021). Techniques are used to provide feedback mechanisms for decision-making and
control of the physical entity. Common examples are data visualization, streaming processing, time-series
analysis, batch-oriented processing, and security analytics (Damjanovic-Behrendt and Behrendt 2019).

3.5 Connection Dimension

The bi-directional communication between the physical entity and the DT uses communication protocols.
These protocols must reflect the synchronization models adopted for communication. Simple synchronization
models are time-based protocols reading/sending a stream of data every constant time, or event-based
communicating any state change. What information and data to read/send is also part of the connection
models. Sending large amounts of data can create high latency. On the contrary, limiting communication
can cause misalignment between DT and the physical entity (Tan and Matta 2022). Depending on

51
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

the complexity of the physical entity and the geographical distribution of its components and devices,
the connection dimension can change significantly. For instance, a product DT such as an engine has
connection protocols very different from an automated warehouse DT.

4 DIGITAL TWIN ARCHITECTURE

Once the main features and models of a DT have been defined for its scope, they must be reproduced
in software components that actuate its behavior. To do so, a proper architecture is needed. Despite the
architecture can be built in accordance with the 5-dimension model (section 3), it will reasonably be adapted
to the specific use cases. We can identify some common principles for its design. Additional notes are
added to guide researchers in choosing architectural solutions for the development of DTs in research
laboratories.

4.1 Communication Technologies

Communication technologies can be categorized into low-power wide-area networks (LPWAN) and short-
range networks. LPWANs include, among others, cellular networks (e.g., 4G and 5G), and long-range wide-
area networks (Karagiannis et al. 2015). Short-range networks are used where the devices have low power
need to be connected locally, and require a gateway for wide-area connectivity. They include, among others,
RFID, Bluetooth, and Wireless Local Area Networks (Kurose and Ross 2013). For research laboratories,
short-range networks are recommended for ease of deployment and maintenance. Implementation examples
in research laboratories for DTs can be found in (Zhou et al. 2021; De Marchi et al. 2022).
Network protocols are used to establish communication between the physical devices and the digital
components. For IP-compliant communication, typical choices include Hyper-Text Transfer Protocol
(HTTP), which has a relatively higher latency, energy, and memory consumption (Naik 2017), or Data
Distribution Service (DDS), usually more suitable for real-time applications given its low latency and
higher scalability. Non-IP-compliant devices can be connected using protocols such as the Message Queue
Telemetry Transport (MQTT) or OPC-UA, among others. Each of the protocols will manifest its advantages
depending on the specific application. For instance, despite it does not guarantee high security, MQTT has
a simple and lightweight nature and a wide range of applications for embedded systems (Naik 2017). For
research laboratories, MQTT has proven to be a flexible and easy-to-deploy solution, also because of the
availability of open-source software libraries (Zhou et al. 2021).

4.2 Communication Protocols

The choice of communication protocol depends on a variety of factors such as the type and number of
sensors required, the distance among them, and the control system adopted. Local Area Network (LAN)
is a general-purpose networking technology that can be used for a wide range of applications, including
industrial control systems. However, its use in industrial environments may require additional measures to
ensure reliability and security. This solution is normally adopted in research laboratories for DT technologies
(Lugaresi et al. 2021). Profinet is a high-speed, real-time Ethernet-based protocol that is well-suited for
use in complex automation systems that require high data transfer rates and precision control. Modbus is
a widely used, open-source protocol that is relatively simple and inexpensive, making it a good choice for
smaller systems or where cost is a primary consideration.

4.3 Database Components

The application-specific data requirements of the DT drive the implementation of related software com-
ponents. For instance, the database-related components of a DT encompass the internal repositories or
any data storage mechanism that houses data. Cloud-based databases offer benefits such as accessibility,
scalability, processing power, and efficiency in data transfer (Alam and El Saddik 2017). Local-based

52
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

data management may be preferred to ensure data security. Also, hybrid approaches offer compromises
(VanDerHorn and Mahadevan 2021).

4.4 Service Components

Other specific components might need to be developed to perform specific functions and implement the
services of the DT. The service components can be in the cloud or in the local network for security reasons.
They can be grouped into unique components or kept separated, the latter facilitating the use of services
developed or provided by different companies. An example of a service component is a scenario manager
used for conducting what-if analyses based on the outcomes of DT predictive models. Specifically, if a
decline in production performance is detected over a certain period, forward-looking simulation models
might be initiated to assess the effectiveness of alternative policies under the latest parameter settings. The
scenario manager would then gather the experimental results and convert them into a set of instructions.

5 DIGITAL TWIN SERVICES

Within its life cycle, a DT relates to services that can be divided into two main sets: (1) internal services
guarantee the capability of a DT to provide its functionalities all along its expected lifespan, and (2)
external services that provide useful benefits for its physical counterpart. All the services make use of the
five-dimension model. The next sections further elaborate on the most important services for each set.

5.1 Internal Services

Figure 3 shows a simple representation of a DT life cycle, highlighting the different roles of internal and
external services. The following paragraphs further elaborate on relevant internal services.

5.1.1 Model Creation and Update

The DT uses virtual entity models to describe the real entity. These models are generated from the
knowledge described in the physical entity models. Model generation must be compliant with their use,
i.e., coarse and light models will be generated for rapid responses whereas accurate and heavy models
are created for more detailed analysis where the response rapidity is less relevant. Model generation and
update can be supported within the same DT environment or it can be done externally. In the first case,
the user creates and updates the virtual model using the same constructs present in the DT and offered by
the service; in the latter case, the service is just a model upload in the DT architecture. In both cases,
the creation and update are executed manually by experts. However, most physical entities are dynamic
and may change with a high frequency. Hence, the digital constructs need to be able to adapt to always
represent the physical entity. We may identify several types of adaptation: (1) model structure, which refers
to the adaptation of structural components, equations, and relationships describing the real entity;(2) model
level, which refers to the tuning of the model, i.e. the possibility to exclude, from the digital representation,
the components that do not significantly contribute to the system description with respect to a particular
goal and (3) model parameters, which refer to the adjustment of the input data model to reflect the current
conditions. The model update belongs to the synchronization cycle that routinely aligns the state, validates
the model, updates the model and possibly the prediction, and checks if the identified prescription can still
be implemented.

5.1.2 State Alignment

The synchronization service has the scope of aligning the states of physical and digital entities. This
service is needed even in case the virtual model is a perfect representation of the system because there
might be incorrect data and stochasticity. For instance, the number of parts in a queue might vary due
to the production rates of downstream and upstream processes. If a short-term performance estimation or

53
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

prediction via the DT services is of interest, it is essential to start all experiments considering the current
buffer level. The synchronization service retrieves the necessary data to correctly represent the system
state. An ideal real-time shadowing is practically unachievable for complex systems (Tan and Matta 2022),
hence the alignment of the system state would typically be replaced by either a cadenced service (i.e., each
fixed time period) or by an event-based trigger as decided with the design of the connection models, this
introduces the next internal service, model validation. Synchronization needs a set of traces and a correct
system logic to operate (i.e., the real part must be aligned with its virtual counterpart).

5.1.3 Model Validation

Validation refers to checking whether the DT is up to date and aligned with the physical entity, in terms
of its capability to correctly represent the system behavior (e.g., output performance metrics, parameter
profiles). Online validation guarantees that if an unpredictable change occurs in the physical entity, the
DT will be capable of replicating that change as well (Lugaresi et al. 2022; Hua et al. 2022). This service
is labeled as a synchronization service because it is necessary for detecting large deviations not imputable
to stochastic noise.

5.1.4 Prediction Update

Upon the alignment of DT with the real entity, a new prediction must be elaborated starting from the
aligned state. This prediction is generally accomplished using performance evaluation techniques either
model-based such as simulation or model-free such as neural networks.

5.1.5 Prescription Check

This service closes the synchronization cycle by checking the validity of the DT prescription (Aydt et al.
2014). For instance, while the DT is elaborating on selecting the control action, the physical entity keeps
operating and the identified action could be no more feasible.

5.2 External Services

The services offered externally by DTs are vast. The following part describes the most relevant ones
classified according to the enabling features.

5.2.1 Descriptive-based Services

A DT has the capability to monitor the condition of its physical counterpart in real-time because of the
continuous flow of information from the physical entity. For example, the use of machine tool vibration data
can be used to generate a real-time health score of the resource. Monitoring services can be implemented as
micro-services or smaller applications that allow the DT to monitor different aspects of the physical entity
(Damjanovic-Behrendt and Behrendt 2019). These services include state visualization, tracing, performance
metrics, alerting (i.e. detecting and isolating problems), as well as dashboards.
Virtual commissioning involves creating a digital representation of a system in order to test it before
construction. DTs offer a promising approach to virtual commissioning since they can provide a realistic
simulation environment. These operations reduce the risk of errors and improve the efficiency of the
commissioning process. Also, DTs can be used to train operators and maintenance personnel, allowing
them to gain familiarity with the system before it is built. In order to exploit DTs for virtual commissioning,
the physical entity and its components must be modeled including the functional relationships of sensors
and actuators, to be able to simulate their behavior in a virtual environment. A DT must then be constructed
and integrated with the control system software, allowing it to interact with the simulated system. Hence,
referring to Figure 1, the synchronous feature of a DT for virtual commissioning is referred to as a
digital-to-digital scenario.

54
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

Figure 2: DT concept (adapted from Tao and Zhang,

2017): DTs need models for each dimension. Figure 3: Internal services of DTs.

5.2.2 Prediction-based Services

A DT includes digital models that enable predictions and forward-looking analyses. For instance, the real-
time state of a manufacturing system can feed a discrete event simulation model and use it to estimate the
future production performance starting from the current conditions (Monostori et al. 2016). It is common to
use such services in combination with optimization algorithms. Indeed, simulation-optimization approaches
use simulation experiments to construct the search space.
Prognosis services are designed to identify potential issues with the entity before they occur, enabling
us to take corrective actions. This is typically done by exploiting advanced analytics and machine learning
algorithms, which use data to identify patterns and anomalies in a system’s behavior. For example, a DT
of a production line might analyze data from sensors installed on each station of the line to detect any
deviations from nominal operating conditions (e.g., chip geometry, and temperature profiles). Then, the DT
service can use real-time data to feed a simulation model and predict when a particular piece of equipment
is likely to fail or require maintenance. Prognosis services can also be used to optimize manufacturing
processes by continuously identifying opportunities for process improvement.

5.2.3 Prescription-based Services

A prescription service provides recommendations on how to improve the performance of the real entity,
based on the results of experiments performed on the DT. For example, the identification of bottlenecks and
inefficiencies in a production process can be done through analyses of digital replicas. A prescription service
must then provide counteractions for how to adapt or reconfigure the production line toward improvements.
Alternatively, shop-floor data can be used to identify when a station is not working at peak efficiency
(prognosis), followed by a prescription for how to adjust the station settings to optimize energy use (e.g.,
tool changeover, lower spindle speed). Other examples of prescriptive-based services are for planning and
control of processes and systems or optimization of some relevant variables.

6 ILLUSTRATIVE EXAMPLE
In this section, a simple production system with its DT architecture is used as an illustrative example.
The physical entity is a two-station lab-scale manufacturing system available at Politecnico di Milano,
illustrated in Figure 4a. The system is built with LEGO Mindstorm (Lugaresi et al. 2021). The processing
times on both stations follow triangular distributions with parameters (3, 8, 5) and (2, 5, 3) s, respectively.
Both buffers can hold up to 8 units and 12 pallets are circulating. The blocking after-service (BAS) policy
is applied. This setting reproduces manufacturing system dynamics, such as blocking, deadlocks, and
stochastic behaviour. Next, each of the five types of DT dimensions is described, with models and enabled
services.

55
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

Figure 4: Illustrative example – a) 2-station closed-loop lab-scale system, b) class diagram model of the
physical entity, c) class diagram model, and d) Petri Net model of the virtual entity.

6.1 Physical Dimension

The system is a closed-loop production line. It is assumed that unlimited parts to be processed are available
and that the pallet loading and unloading phases are very short. Each station processes one pallet at a time.
If a failure occurs, the pallet is held for an additional amount of time in the station until it is repaired. The
conveyors bring pallets from one station to the other and operate as buffers. A station cannot download
parts until the corresponding downstream buffer is full. The class diagram in Figure 4b represents the
physical entity model developed from the most recent knowledge about the system.

6.2 Virtual or Digital Dimension

The virtual entity of the illustrative example consists of a DES model of the system. The model is described
using the Petri Net formalism, in which the main transitions correspond to the processing of the stations
and the places represent the pallets’ locations (i.e., in the stations, on the conveyors). See also Figure 4d.
Further, the discrete event simulation model is coded in Simpy. A Unified Modeling Language (UML) class
diagram model is used to describe the main elements of the model, i.e. stations and the buffers in Figure
4c. The DES model does not need to exactly correspond to the physical model. Indeed, approximations
and abstractions are normally introduced in the modeling phase. For instance, transportation time from one
station to another is modeled using a dummy station with deterministic processing time, differently from
the knowledge described in section 6.1. The simulation model as a virtual entity enables useful services,
such as predicting the end-of-the-day performance of the system with current settings, and diagnosis, when
simulation experiments are included in a simulation-optimization approach.

6.3 Data Dimension

Data are collected from the physical entity and stored in a real-time InfluxDB database. Data are modeled
using UML class diagrams, in which each class represents a table. Such a setting enables real-time
monitoring services on raw data. For instance, the last pallet processed by each station can be retrieved
via a query to the database component. Also aggregated data resulting from data analyses and predictions
are stored in the same database (e.g., throughput, system time).

6.4 Connection Dimension

The connections between all components are achieved by means of the Secure Shell Host (SSH) protocol,
and the system is controlled using messages exploiting the MQTT protocol. With this architecture, it is
possible to exchange information and implement actions on the system online. Each message is structured

56
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

Table 1: Illustrative example: role of digital twin components and enabled services.

Dimension Components Models Enabled Services

Physical Closed-loop 2-station lab-scale model built Class Diagram Model Demonstrative System
with LEGO® components, including sen- Dynamics
sors and actuators
Virtual DES Model in Simpy Petri Net Model, Prediction, Diagnosis
Class Diagram Model
Data Real-time Database in InfluxDB Class Diagram Model Real-time Monitoring
Connections Message-based infrastructure using MQTT Message Model Prescription (actuators)
Services Internal services (creation, state alignment, Synchronization Prediction, Prognosis
validation, update); External Services (re- Validation (what-if)
maining cycle time prediction, throughput Generation
estimation, system time estimation) Statistical Model

following a specific data model. For instance, a message indicating an activity starting in a station is written
as the dictionary {”activity” : s, ”id” : id, ”ts” : time, ”tag” : ”s” }, where s is a variable indicating the
station number, id indicates the identifier of the pallet, time the event time-stamp in UNIX format, and tag
is a string indicating the activity performed in the station. Specific messages can be modeled to indicate
the collection of specific data, or controls to the physical actuators, hence enabling the online prescription
of corrective actions. Further examples are available in (Lugaresi et al. 2021).

6.5 Service Dimension

For reasons of space, we only consider the case in which a production plan already exists and, at a certain
moment, Station 2 undergoes degradation. This station becomes slower with its processing time following
a new triangular distribution with parameters (9, 14, 11). The DT assesses and evaluates counteractions to
manage the described situation.
A model generation internal service is deployed implementing the methodology described in (Lugaresi
and Matta 2021). The result is a discrete event simulation model in Simpy. The simulation model is
validated online using the technique described in (Lugaresi et al. 2022), and synchronized to the last known
system state, via the respective internal services. A scenario manager component exploits the virtual entity
to perform a what-if analysis. For this simple case, the analysis is conducted on two alternatives: (1) do
nothing, keep producing at a slower pace and repair the station at the end of the shift; (2) react, stop the
plant to allow repairing activities and then continue with the production pace before the slow-down. The
alternatives are evaluated as two separate simulation experiments are executed in order to determine which
scenario maximizes the production output until the rest of the shift. In this case, the prediction results
in an average 165 ± 3 parts in the first scenario, and 209 ± 3 parts in the second one. Hence, the second
scenario is selected and the prescription is sent for implementation.

7 APPLICATION EXAMPLES
In this section, three noteworthy use cases from the literature are selected and analyzed in light of the
identified DT characteristics in this paper. Table 2 collects these use cases and summarizes their main
features. Leng et al. (2021) built the DT of a warehouse to support packing and storage assignment
optimization. The DT gathers real-time information from a physical warehouse product-service system
and applies it to a digital model. The DT incorporates a joint optimization model to efficiently manage
both stacking and storage assignments of the warehouse in a timely manner. With real-time data, the joint
optimization model is able to make periodic optimal decisions, which are then verified through a simulation
engine. The authors developed and tested a prototype DT using a tobacco warehouse product-service
system as a case study.

57
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

Table 2: Description of selected use cases with a significant DT application.

Use Cases Leng et al. (2021) Zhang et al. (2021) Son et al. (2021)
Basic all all all
Features Advanced Prescriptive Predictive, prescriptive Predictive
Physical Warehouse Job-shop Automotive Plant
Virtual 3D model, simulation Optimization Model Information models:
(MILP) product, process, plan,
plant, and resources
Dimensions Service Packing Optimization, API interfaces: avail- Production Planning,
Coordination, Storage ability prediction, distur- Performance Prediction
Assignment bance detection, perfor-
mance evaluation
Data n.a. ODBC, JDBC interfaces Service-oriented
Connection Event-based MTConnect, OPC-UA, Web-based
TCP/IP

Zhang et al. (2021) proposed a methodology that involves the integration of a DT to enhance the
dynamic scheduling process in a machining job shop that produces hydraulic valves. The DT is exploited
within a production planning and control context. The physical entity is given an initial production plan.
The synchronization service updates in real-time the resource availability. Then, a scheduling component
elaborates the new production plan in a rolling approach.
Son et al. (2021) introduced a DT to predict if an automotive product can be manufactured according to
a predefined schedule in the presence of abnormal scenarios. The authors’ approach includes an information
model purposely designed to represent the main objects involved in automotive body production lines. The
DT is developed jointly with other components to create an integrated web-based manufacturing platform.
The authors used real production planning data to verify the DT performance through dedicated experiments.

8 FINAL REMARKS AND RESEARCH CHALLENGES

This work has summarized the current knowledge about DTs with the scope of making the readers aware
of the recent essence of this paradigm. DTs are not a single technology but a set of technologies integrated
to provide specific services in relation to a product, a system, a place, or a human. The consequence
is an increase in complexity that represents its main limitation. DTs coupled with simple products or
systems would be too expensive to develop and maintain for the expected benefits. This work has two
main limitations. The first is the space limit that forced us to summarize concepts that would deserve more
discussion. For instance, the cybersecurity of DTs is a relevant problem that is attracting the attention of
researchers and practitioners. The second is the bias of the authors, who are more involved in system DTs
rather than product, place, and human DTs.
Research is also needed to tackle the challenges that are currently limiting the DTs. Integration
challenges arise from the complexity of managing existing DTs once they are operational. Although the
interaction between the physical and digital worlds is a crucial aspect of DTs, few studies have explored
this topic in depth. Thus, there is a need to develop techniques specifically tailored to address the challenge
of physical-digital alignment. Also, the digital-to-physical alignment is not clear, a prescription selected by
DT from solving an optimization problem might not be implementable anymore in reality because this last
evolved during the optimization task execution. This is particularly important for achieving level 5 of the
DT evolution framework, which involves building and managing DTs at the federated level and integration
with information systems (e.g., ERP, MES). However, there is currently a lack of clarity on how this will
be achieved. Overconfidence in DTs might be the cause of trusting simplistic models that are not adequate
for the complexity of the real entity. How to continuously validate the models used by a DT during its

58
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

operation is another challenge. Model validation techniques are currently used offline and may require
limiting assumptions, specific experiments, and long times. Online validation of digital twin models is an
open field for novel approaches. When a DT model is not valid, a modification must be introduced to re-align
the twins. How to modify the used models is a challenge. Techniques from Artificial Intelligence can be
used to fit the behavior of the real entity or a single part of it. More advanced approaches have the potential
to automatically discover from data new knowledge about the real entity and use it for improving the DT
models. Several studies have proposed DT architectures that are largely domain- and technology-specific.
The use cases, applications, and domains vary extensively, together with the proposed application-specific
architectures. The development of a DT remains dependent on the requirements of its intended applications,
with several related complex choices (e.g., communication standards, data management system).

ACKNOWLEDGMENTS
This work was partially supported by the Horizon Europe Project ”Auto-Twin” under Grant Agreement No. 101092021.

REFERENCES
Ahmed, H., and L. Devoto. 2021. “The Potential of a Digital Twin in Surgery”. Surgical Innovation 28(4):509–510.
Alam, K. M., and A. El Saddik. 2017. “C2PS: A Digital Twin Architecture Reference Model for the Cloud-based Cyber-physical
Systems”. IEEE Access 5:2050–2062.
Attaran, M., and B. G. Celik. 2023. “Digital Twin: Benefits, Use Cases, Challenges, and Opportunities”. Decision Analytics
Journal 6:100165.
Aydt, H., W. Cai, and S. John Turner. 2014. “Dynamic Specialization for Symbiotic Simulation-based Operational Decision
Support Using the Evolutionary Computing Modelling Language (ECML)”. Journal of Simulation 8(2):105–114.
Barricelli, B. R., E. Casiraghi, and D. Fogli. 2019. “A Survey on Digital Twin: Definitions, Characteristics, Applications, and
Design Implications”. IEEE Access 7:167653–167671.
Bergs, T., S. Gierlings, T. Auerbach, A. Klink, D. Schraknepper, and T. Augspurger. 2021. “The Concept of Digital Twin and
Digital Shadow in Manufacturing”. Procedia CIRP 101:81–84. 9th CIRP Conference on High Performance Cutting.
Coyle, R. G. 1997. “System Dynamics Modelling: a Practical Approach”. Journal of the Operational Research Society 48(5):544.
Croatti, A., M. Gabellini, S. Montagna, and A. Ricci. 2020. “On the Integration of Agents and Digital Twins in Healthcare”.
Journal of Medical Systems 44:1–8.
Damjanovic-Behrendt, V., and W. Behrendt. 2019. “An Open Source Approach to the Design and Implementation of Digital
Twins for Smart Manufacturing”. International Journal of Computer Integrated Manufacturing 32:366–384.
De Marchi, M., B. G. Mark, N. Angeli, E. Rauch, and D. T. Matt. 2022. “Using RFID and Industrial IoT in Manual Assembly
for Data Collection and Processing in Learning Factories”. Available at SSRN 4074692.
Dertien, S., and C. Macmahon. 2022. “State of Digital Twin 2022”. White Paper:1–12.
Dittmann, S., P. Zhang, A. Glodde, and F. Dietrich. 2021. “Towards a Scalable Implementation of Digital Twins-A Generic
Method to Acquire Shopfloor Data”. Procedia CIRP 96:157–162.
Esser, S., and D. Fahland. 2021. “Multi-dimensional event data in graph databases”. Journal on Data Semantics 10(1-2):109–141.
Fortune Business Insights 2023. “Digital Twin Market Report”. https://www.fortunebusinessinsights.com/
digital-twin-market-106246, accessed 12.05.2023.
Grieves, M. 2014. “Digital twin: Manufacturing Excellence Through Virtual Factory Replication”. White paper:1–7.
Grieves, M., and J. Vickers. 2017. Digital Twin: Mitigating Unpredictable, Undesirable Emergent Behavior in Complex Systems,
85–113. Cham, Switzerland: Springer International Publishing.
Hua, E. Y., S. Lazarova-Molnar, and D. P. Francis. 2022. “Validation of Digital Twins: Challenges and Opportunities”. In
Proceedings of 2022 Winter Simulation Conference, edited by B. Feng, G. Pedrielli, Y. Peng, S. Shashaani, E. Song, C. G.
Corlu, L. H. Lee, E. P. Chew, T. Roeder, and P. Lendermann, 2900–2911. Piscataway, New Jersey: Institute of Electrical
and Electronics Engineers, Inc.
ISO 23247:2021 2021. “Automation Systems and Integration — Digital Twin Framework for Manufacturing — Part 1: Overview
and General Principles”. Standard, International Organization for Standardization, Geneva, Switzerland.
Jimenez, J. I., H. Jahankhani, and S. Kendzierskyj. 2020. Health Care in the Cyberspace: Medical Cyber-Physical System and
Digital Twin Challenges, 79–92. Cham, Switzerland: Springer International Publishing.
Karagiannis, V., P. Chatzimisios, F. Vazquez-Gallego, and J. Alonso-Zarate. 2015. “A Survey on Application Layer Protocols
for the Internet of Things”. Transaction on IoT and Cloud computing 3(1):11–17.
Kritzinger, W., M. Karner, G. Traar, J. Henjes, and W. Sihn. 2018. “Digital Twin in Manufacturing: A Categorical Literature
Review and Classification”. IFAC-PapersOnLine 51(11):1016–1022.

59
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.
 Matta and Lugaresi

Kurose, J., and K. Ross. 2013. “Multimedia networking”. Computer networking: a top-down approach, 6th edn. Pearson, New
Jersey:588–592.
Leng, J., D. Yan, Q. Liu, H. Zhang, G. Zhao, L. Wei, D. Zhang, A. Yu, and X. Chen. 2021. “Digital Twin-driven Joint
Optimisation of Packing and Storage Assignment in Large-scale Automated High-rise Warehouse Product-Service System”.
International Journal of Computer Integrated Manufacturing 34(7-8):783–800.
Liu, X., D. Jiang, B. Tao, F. Xiang, G. Jiang, Y. Sun, J. Kong, and G. Li. 2023. “A Systematic Review of Digital Twin About
Physical Entities, Virtual Models, Twin Data, and Applications”. Advanced Engineering Informatics 55:101876.
Lugaresi, G., V. V. Alba, and A. Matta. 2021. “Lab-scale Models of Manufacturing Systems for Testing Real-time Simulation
and Production Control Technologies”. Journal of Manufacturing Systems 58:93–108.
Lugaresi, G., S. Gangemi, G. Gazzoni, and A. Matta. 2022. “Online Validation of Simulation-Based Digital Twins Exploiting
Time Series Analysis”. In Proceedings of 2022 Winter Simulation Conference, edited by B. Feng, G. Pedrielli, Y. Peng,
S. Shashaani, E. Song, C. G. Corlu, L. H. Lee, E. P. Chew, T. Roeder, and P. Lendermann, 2912–2923. Piscataway, New
Jersey: Institute of Electrical and Electronics Engineers, Inc.
Lugaresi, G., and A. Matta. 2021. “Automated Manufacturing System Discovery and Digital Twin Generation”. Journal of
Manufacturing Systems 59:51–66.
Mills, B., and A. Redford. 1983. “Manufacturing of Engineering Materials”. Applied Science:20.
Monostori, L., B. Kádár, T. Bauernhansl, S. Kondoh, S. Kumara, G. Reinhart, O. Sauer, G. Schuh, W. Sihn, and K. Ueda.
2016. “Cyber-physical Systems in Manufacturing”. Cirp Annals 65(2):621–641.
Mykoniatis, K., and G. A. Harris. 2021. “A Digital Twin Emulator of a Modular Production System Using a Data-driven
Hybrid Modeling and Simulation Approach”. Journal of Intelligent Manufacturing:1–13.
Naik, N. 2017. “Choice of Effective Messaging Protocols for IoT Systems: MQTT, CoAP, AMQP and HTTP”. In 2017 IEEE
international systems engineering symposium (ISSE), 1–7. IEEE.
Negri, E., L. Fumagalli, and M. Macchi. 2017. “A Review of the Roles of Digital Twin in CPS-based Production Systems”.
Procedia Manufacturing (11):939–948.
Robinson, S. 2014. Simulation: the Practice of Model Development and Use. London, United Kingdom: Bloomsbury Publishing.
Son, Y. H., K. T. Park, D. Lee, S. W. Jeon, and S. Do Noh. 2021. “Digital Twin–based Cyber-physical System for Automotive
Body Production Lines”. The International Journal of Advanced Manufacturing Technology 115:291–310.
Tan, B., and A. Matta. 2022. “Optimizing Digital Twin Synchronization in a Finite Horizon”. In Proceedings of 2022 Winter
Simulation Conference, edited by B. Feng, G. Pedrielli, Y. Peng, S. Shashaani, E. Song, C. G. Corlu, L. H. Lee, E. P. Chew,
T. Roeder, and P. Lendermann, 2924–2935. Piscataway, New Jersey: Institute of Electrical and Electronics Engineers, Inc.
Tao, F., and M. Zhang. 2017. “Digital Twin Shop-floor: a New Shop-floor Paradigm Towards Smart Manufacturing”. Ieee
Access 5:20418–20427.
Tavakoli, S., A. Mousavi, and A. Komashie. 2008. “A Generic Framework for Real-time Discrete Event Simulation (DES)
Modelling”. In Proceedings of the 2008 Winter Simulation Conference, edited by S. J. Mason, R. R. Hill, L. Mönch,
O. Rose, T. Jefferson, and J. W. Fowler, 1931–1938. Piscataway, New Jersey: Institute of Electrical and Electronics
Engineers, Inc.
Tocher, K. D. 1967. The art of simulation. English Universities Press.
VanDerHorn, E., and S. Mahadevan. 2021. “Digital Twin: Generalization, Characterization and Implementation”. Decision
Support Systems:113524.
Wang, J., X. Niu, R. X. Gao, Z. Huang, and R. Xue. 2023. “Digital Twin-driven Virtual Commissioning of Machine Tool”.
Robotics and Computer-Integrated Manufacturing 81:102499.
Williams, T. J. 1989. “A Reference Model for Computer Integrated Manufacturing (CIM)”. International Purdue Works.
Xiong, M., and H. Wang. 2022. “Digital Twin Applications in Aviation Industry: a Review”. The International Journal of
Advanced Manufacturing Technology 121(9-10):5677–5692.
Zhang, M., F. Tao, and A. Nee. 2021. “Digital Twin Enhanced Dynamic Job-shop Scheduling”. Journal of Manufacturing
Systems 58:146–156.
Zhou, J., Y. Wang, and Y. Q. Chua. 2021. “Real-Time OEE for Industry 4.0 Learning and Practice Training”. In Proceedings
of the 2021 Conference on Learning Factories (CLF).

AUTHOR BIOGRAPHIES
ANDREA MATTA is Full Professor at Politecnico di Milano, where he currently teaches integrated manufacturing systems
and manufacturing processes. His research area includes analysis, design, and management of manufacturing and healthcare
systems. He is Editor in Chief of the Flexible Services and Manufacturing Journal. His email address is [email protected].
GIOVANNI LUGARESI is Assistant Professor at the Department of Mechanical Engineering of KU Leuven (Belgium). He
works on production planning and control, process mining, and robust optimization. His email is [email protected].

60
Authorized licensed use limited to: Zhejiang University. Downloaded on December 24,2024 at 20:18:47 UTC from IEEE Xplore. Restrictions apply.