A Vision for Advancing Digital Twins Intelligence: Key Insights and

Lessons from Decades of Research and Experience with Simulation

Sanja Lazarova-Molnar

1,2 a

Institute of Applied Informatics and Formal Description Methods, Karlsruhe Institute of Technology, Karlsruhe, Germany

Mærsk Mc-Kinney Møller Institute, University of Southern Denmark, Odense, Denmark

Keywords: Digital Twins, Simulation, Key Considerations, Goal-Oriented, Fusion of Data and Expert Knowledge.

Abstract: Digital Twins have revolutionized the domain of Modeling and Simulation by making use of the growing and

cost-efficient possibilities to extract data from systems, as well as the increasing computational power. At the

same time, Digital Twins have enabled tremendous advances in diverse cyber-physical systems by enabling

better monitoring, predictive maintenance, design optimization, and informed decision-making. As their

popularity evolved, the understanding of what a Digital Twins has become more and more dispersed and

unclear. Here, we offer understanding of what a Digital Twin is based on our experience in research within

its native domain of Modeling and Simulation, with a concrete focus on the key considerations that need to

be made when developing Digital Twins or working with them. We, furthermore, emphasize the need to

include all available knowledge for better-informed Digital Twins. To illustrate our ideas and vision, we use

case studies from our research.

1 INTRODUCTION

Digital Twins have emerged as indispensable assets

for industries seeking to optimize operations and

enhance decision-making processes. At their core,

Digital Twins are dynamic digital counterparts of

physical entities, continuously updated based on real-

world data, simulating and mirroring physical

systems' behavior (Friederich et al., 2022; Grieves,

2014). Digital Twins are typically realized through

computational models like machine learning or

simulation models (Masood & Sonntag, 2020). These

twins, both digital and physical, are interconnected in

near-real-time, facilitating constant information

exchange and synchronization. They are inherently

data-driven, relying heavily on supervised and

unsupervised machine learning techniques

(Friederich et al., 2022).

In this paper, we offer understanding of what a

Digital Twin is based on our decades-long experience

in research within its native domain of Modeling and

Simulation (Friederich & Lazarova-Molnar, 2023;

Lazarova-Molnar & Horton, 2003, 2004), with a

concrete focus on the key considerations that need to

be made when developing Digital Twins or working

https://orcid.org/0000-0002-6052-0863

with them. Drawing from this experience, we

emphasize on the Digital Twins' goal-oriented nature,

vital to comprehend for laying the groundwork for

exploring their Digital Twins’ applications and

operationalization. Thus, we concentrate and

elaborate on the imperative of goal-oriented digital

twin design. Besides other key lessons drawn from

the domain of Modeling and Simulation, we also

present our vision of how to enable better-informed

Digital Twins by enabling systematic inclusion of all

information that we have available. To this point, we

briefly introduce two case studies that illustrate this

vision and draft our future research.

2 KEY CONSIDERATIONS FOR

DIGITAL TWINS

Drawing from our experience in developing methods

for Digital Twins and our work in Modeling and

Simulation (Francis et al., 2021; Hua et al., 2022;

Mohamed et al., 2023), especially in the context of

data-driven simulation (Lazarova-Molnar & Li,

2019), we have developed a number of key

considerations that we elaborate in the following:

Lazarova-Molnar, S.

A Vision for Advancing Digital Twins Intelligence: Key Insights and Lessons from Decades of Research and Experience with Simulation.

DOI: 10.5220/0012884800003758

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 14th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH 2024), pages 5-10

ISBN: 978-989-758-708-5; ISSN: 2184-2841

Figure 1: Goal-oriented Digital Twins.

• Goal-oriented Digital Twins: The first two

steps in devising classic simulation studies are

Problem Specification and Objective

Specification (Lazarova-Molnar & Li, 2019).

Thus, when building simulation models, one

has to know what problem the models should

solve and set the objectives to target that. This

is pretty much the same with Digital Twins,

which also need to be built to solve a specific

problem or a set of problems. There cannot be

a Digital Twin that fully models all aspects of

a given system and answers all possible

questions. As such, we have to talk about goal-

oriented Digital Twins. In this manner, a

system does not have a single Digital Twin, but

a set of Digital Twins that are projections of a

system’s behavior on predefined problems and

objectives. This point is illustrated in Figure 1,

where we see that one Digital Twin is needed

to target energy efficiency of a system, and

another one for reliability analysis, for

instance. These Digital Twins also need

different data streams and have different

underlying models. In additon, the underlying

modeling formalisms might be different as

well. They need to be, however, driven by the

problem specifications.

• Digital Twins necessitate a feedback loop:

Digital Twins are different from simulation

models in that that their underlying models

update in near-real-time, continuously

reflecting changes that occur in the

corresponding real systems. In addtion, Digital

Twins need to deliver simulation-informed

decisions back to the real system. For this,

Digital Twins need to incorporate a feedback

loop that enables the bidirectional

communication process which is the detail that

makes Digital Twins different from traditional

and static simulation models.

• Digital Twins automate simulation modeling:

Digital Twins have emerged as a result of the

large prevalence of data and computing

resources. This provided a chance to automate

some of the most expensive computational

techniques, i.e., in this case modeling and

simulation. Thus, Digital Twins can be seen as

an automation of the classical modeling and

simulation processes, and as such, need to

utilize significant portion of the formal

background that has been developed in this

area. This formal background has been around

for many decades and it has been mainly

targeted at knowledge-driven model

development (Law et al., 2007; Maria, 1997;

Zeigler et al., 2000). This body of knowledge

can only benefit, clarify and support the

understanding and advancement of Digital

Twins.

• Automatically deriving simulation models

from ongoing data collection, as is the case

with Digital Twins, provides opportunity to

simultaneously perform validation, and, thus,

release the need for explicit and separate

validation processes. Moreover, validation of

Digital Twins’ underlying simulation models is

no longer additional optional step. On contrary,

integrated validation becomes a prerequisite

and an integral part of Digital Twins as

ensuring that the underlying model accurately

reflects the real system is necessary for a model

to be quialified as a Digital Twin model

SIMULTECH 2024 - 14th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

(Friederich et al., 2022; Hua et al., 2022; Zare

& Lazarova-Molnar, 2024).

• Integration of all that we know or will know:

Digital Twins need not throw away what we

already know or may know in future. Thus,

Digital Twins need to be built in a way that they

seamlessly integrate all that we know or might

know in future about the systems of interest.

This knowledge can come in many different

forms, i.e., as physics-based models, or as

expert knowledge from experience. This is a

strong point in our line of research, on which

we elaborate more in the following.

To illustrate our understanding and definition of

Digital Twins, we designed and further developed a

framework that we initially presented in (Friederich

et al., 2022). For this contribution, we further iterated

on the framework, to present an updated version that

is more complete and considers the elements that are

used the derive decisions, e.g., Data Analytics,

Optimization, etc. This updated framework is

illustrated in Figure 2.

3 FUSION OF DATA AND

EXPERT KNOWLEDGE FOR

DIGITAL TWINS

Development of Digital Twins to accurately represent

their real-world physical counterparts poses

significant challenges. In academic literature, two

main approaches to modeling Digital Twins emerge,

each adhering to distinct paradigms. The first

approach relies on traditional knowledge-driven

modeling methods, while the second favors newer

data-driven model extraction strategies. Presently,

there is a notable preference for data-driven

techniques, which reduces reliance on human

expertise. However, solely relying on a data-driven

approach neglects the potential benefits of integrating

expert knowledge.

In the last decade, along the same line of thought,

we have been fully immersed in the data-driven

methods and discoveries, slightly ignoring the fact

that we, humans, already do know a lot about these

systems that we model and simulate. For this reason,

it is essential that we develop methodology that can

seamlessly integrate human/expert knowledge in the

models that we extract. An engineer that has been

working with a given system, knows a lot about this

system. We should find a way to utilize that and not

simply override with data. Furthermore, there have

been centuries of development of physics-based

models for various phenomena. Therefore, it is

imperative to develop methods to interface this

knowledge with data, as well as use both information

sources to complement and cross-validate each other.

Expert knowledge complements and enhances

data-driven methodologies by offering insights that

may be challenging to obtain through data alone, or at

least to obtain within a given time range.

Additionally, expert knowledge provides nuanced

understandings of phenomena based on expert

experiences and contexts, thereby addressing other

challenges associated with data-driven Digital Twins

model

extractions, such as data scarcity. As a result,

Figure 2: Framework for Data-Driven Digital Twins, extension from (Friederich et al., 2022).

A Vision for Advancing Digital Twins Intelligence: Key Insights and Lessons from Decades of Research and Experience with Simulation

Figure 3: Fusion DT Framework derived from both expert knowledge and data, introduced in (Michelle Jungmann & Sanja

Lazarova-Molnar, 2024).

combining data and expert knowledge can both

enable more accurate and more efficient Digital Twin

model development.

To enhance the aforementioned Digital Twins'

full potential, a framework that seamlessly integrates

human intelligence with technological innovation is

essential. Thus, we have also recently presented an

initial framework for Digital Twins (Michelle

Jungmann & Sanja Lazarova-Molnar, 2024), which

we illustrate in Figure 3. We intend to further enhance

this framework and methodology to leverage

advanced technologies such as large language models

to streamline digital twin workflows and augment

decision-making capabilities. By emphasizing the

importance of visualization in facilitating knowledge

integration, we outline a roadmap for advancing

digital twin performance.

To illustrate the framework for integration of

expert knowledge in Digital Twins, we developed

case studies that we briefly illustrate in the following

section, and refer to the corresponding sources for

more detailed decsriptions.

In the following, we briefly refer to two of our

latest case studies to illustrate the idea of including

expert knowledge for automated extraction of

simulation models, as a vision toward Digital Twins

that utilize the same information sources. Both

studies focus on reliability modeling and analysis.

3.1 Extraction of Fault Tree Models

from Fusion of Data and Expert

Knowledge

Grounded in our empirical evidence, this case study

offers a demonstration of fusing digital twin efficacy

through data-driven methodologies and human

expertise. We designed the process of fault tree

extraction from a fusion of time series data streams

and expert knowledge. With this case study, we

demonstrated the potential of using expert knowledge

for better-informed digital twins in enhancing system

reliability and safety (Niloofar & Lazarova-Molnar,

2021).

The workflow, that is fully detailed in (Niloofar

& Lazarova-Molnar, 2021), consists of using both

data and expert knowledge statements to extract fault

tree models that can be utilized for assessing

reliability of cyber-physical systems. With this, we

were able to enhance the synergy between human

expertise and sensor data by refining methods to

translate expert insights into data-mergeable formats.

We, furthermore, introduced Bayesian probability

updating, weighted cut sets, and validated expert

knowledge through incremental data gathering. In the

upcoming work, we focus on generalizing systematic

integration of data-driven and knowledge-driven fault

tree analysis.

3.2 Extraction of Petri Net Models

from Data and Expert Knowledge

In a more recent effort that was initiated by the work

in (Michelle Jungmann & Sanja Lazarova-Molnar,

2024) and further expanded in (Michelle Jungmann &

Sanja Lazarova-Molnar, 2024), we extracted

reliability-centred Petri nets by fusing streaming data

from a system and expert knowledge. The results we

obtained were promising, demonstrating the

feasibility of explicitly combining expert knowledge

and data for more accurate reliability models. With

this, we highlighted the value that expert knowledge

SIMULTECH 2024 - 14th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

can bring in faster extraction of more accurate

models.

However, we also encountered challenges in these

fusion processes, where one of the major ones was –

who do we trust more when data and expert

knowledge contrast each other. For fusing data and

expert knowledge, we introduced two main strategies,

termed a priori and a posteriori. We implemented

these fusion algorithms in a case study on reliability

modeling, where we formulated and formalized four

expert knowledge statements that varied in

complexity. Synthetic data was generated for the a

priori fusion algorithms, and the results of each

fusion algorithm were executed and analyzed, leading

to the extraction of a Petri net model. The case study

demonstrated the potential of combining expert

knowledge and data for different types of reliability

information using the proposed strategies. In the

upcoming work, we will involve refining the

approach by improving the quality and integration of

expert knowledge, automating formalization from

natural language, incorporating fuzzy logic, and

addressing data gaps in logs.

4 CONCLUSION

Addressing the evolving landscape of Digital Twins

requires a proactive approach to overcoming

challenges and charting future directions. Drawing

from our long-time experience of research in the

domain of Modeling and Simulation, we identified

and elaborated on key considerations for developing

Digital Twins. We, furthermore, emphasized the

importance of developing goal-oriented Digital

Twins, as well as utilizing all available knowledge in

a systematic way to enhance more accurate

underlying models of Digital Twins, and with this

enhance their intelligence. We illustrated our key

points by case studies from our research, which we

used to show directions for our future research.

In conclusion, enhancing Digital Twins'

usefulness and performance requires a holistic

approach that integrates human intelligence, data

analytics, and practical problem-oriented approach,

which also builds upon the existing body of

knowledge from the well-established domain of

Modeling and Simulation. With this, we can enhance

Digital Twins efficacy and their utilization across

diverse industries and for different goals.

ACKNOWLEDGEMENTS

The authors extend their thanks for the funding

received from the ONE4ALL and DMaaST projects

funded by the European Commission, Horizon

Europe Programme under the Grant Agreements No.

101091877 and No. 101138648, correspondingly.

REFERENCES

Francis, D. P., Lazarova-Molnar, S., & Mohamed, N.

(2021). Towards data-driven digital twins for smart

manufacturing. Proceedings of the 27th International

Conference on Systems Engineering, ICSEng 2020,

Friederich, J., Francis, D. P., Lazarova-Molnar, S., &

Mohamed, N. (2022). A framework for data-driven

digital twins of smart manufacturing systems.

Computers in Industry, 136, 103586.

Friederich, J., & Lazarova-Molnar, S. (2023). A

Framework for Validating Data-Driven Discrete-Event

Simulation Models of Cyber-Physical Production

Systems. 2023 Winter Simulation Conference (WSC),

Grieves, M. (2014). Digital twin: manufacturing excellence

through virtual factory replication. White paper,

1(2014), 1-7.

Hua, E. Y., Lazarova-Molnar, S., & Francis, D. P. (2022).

Validation of digital twins: challenges and

opportunities. 2022 Winter Simulation Conference

(WSC),

Jungmann, M., & Lazarova-Molnar, S. (2024). Fusing

Expert Knowledge and Data for Simulation Model

Discovery in Digital Twins: A Case Study from

Reliability Modeling. Winter Simulation Conference

2024,

Jungmann, M., & Lazarova-Molnar, S. (2024). Towards

Fusing Data and Expert Knowledge for Better-

Informed Digital Twins: An Initial Framework.

Procedia Computer Science.

Law, A. M., Kelton, W. D., & Kelton, W. D. (2007).

Simulation modeling and analysis (Vol. 3). Mcgraw-

hill New York.

Lazarova-Molnar, S., & Horton, G. (2003). An

Experimental Study of the Behaviour of the Proxel-

Based Simulation Algorithm. SimVis,

Lazarova-Molnar, S., & Horton, G. (2004). Proxel-based

simulation of a warranty model. European Simulation

multiconference,

Lazarova-Molnar, S., & Li, X. (2019). Deriving simulation

models from data: steps of simulation studies revisited.

2019 Winter Simulation Conference (WSC),

Maria, A. (1997). Introduction to modeling and simulation.

Proceedings of the 29th conference on Winter

simulation,

Masood, T., & Sonntag, P. (2020). Industry 4.0: Adoption

challenges and benefits for SMEs. Computers in

Industry, 121, 103261.

A Vision for Advancing Digital Twins Intelligence: Key Insights and Lessons from Decades of Research and Experience with Simulation

Mohamed, N., Lazarova-Molnar, S., & Al-Jaroodi, J.

(2023). Digital Twins for Energy-Efficient

Manufacturing. 2023 IEEE International Systems

Conference (SysCon),

Niloofar, P., & Lazarova-Molnar, S. (2021). Fusion of data

and expert knowledge for fault tree reliability analysis

of cyber-physical systems. 2021 5th International

Conference on System Reliability and Safety (ICSRS),

Zare, A., & Lazarova-Molnar, S. (2024). Validation of

Digital Twins in Labor-Intensive Manufacturing:

Significance and Challenges. The 7th International

Conference on Emerging Data and Industry 4.0,

Zeigler, B. P., Praehofer, H., & Kim, T. G. (2000). Theory

of modeling and simulation. Academic press.

SIMULTECH 2024 - 14th International Conference on Simulation and Modeling Methodologies, Technologies and Applications