Ph. D. Project
Title:
Automatic generation of a Digital Twin for intelligent management of water networks.
Dates:
2024/09/01 - 2027/08/31
Supervisor(s):
Description:
A growing number of theoretical and technical modeling frameworks are emerging in various sectors, providing an organized structure for the design of
digital twins. These frameworks are playing a crucial role in diverse fields such as the energy sector [Yu et al., 2022], smart cities [Mylonas et al., 2021],
healthcare [Zimmermann et al., 2019] and even water management [Torfs et al., 2022], [Wei et al., 2022]. Their essential role lies in simplifying the
process of creating highly accurate and detailed virtual representations of physical objects and industrial processes, thus contributing to a better
understanding and management of these complex systems. Other initiatives, such as ISO 23247 [Shao, 2021] or the work of [Tao et al., 2019] in the
manufacturing sector, also aim to standardize methods for creating digital twins, ensuring greater consistency and interoperability between different
systems.
However, while these frameworks provide a solid foundation, their use alone is generally not sufficient to generate a complete and accurate digital twin.
Instantiation remains a complex task requiring in-depth knowledge of the real system, integrating relevant data, existing models and the expertise of the
players involved. A digital twin must meet the specific needs of end-users, adding tangible value to operations. The instantiation of a digital twin is
therefore both time-consuming and costly. The aim of this thesis is to simplify this process by automating the creation of digital twins, whether for
existing installations or the deployment of new water resource management infrastructures.
Although recent work in the scientific literature addresses the automatic generation of digital twins, these initiatives focus exclusively on the
manufacturing industry domain. For example, Garcia et al. [Garcia et al., 2023] explore the feasibility of automatically generating digital twins from
historical data for complex manufacturing processes, where exhaustive manual modeling of all line assets is particularly challenging. In [Lugaresi et al.,
2021], the authors propose a novel approach aimed at discovering, selecting and extracting relevant features of a production system from data logs,
enabling the creation of a digital twin with a suitable level of detail. These initiatives focus exclusively on exploiting historical data, giving the digital
twin a significant dependency on both data freshness and completeness. As a result, the digital twin's ability to accurately represent current situations
and track system evolutions will be limited.
While the integration of real-time data or the identification of causal relationships using advanced artificial intelligence methods offer a partial answer to
this problem, complementary approaches will be explored in this thesis.
In particular, we propose to integrate the knowledge of experts in the field, which can be of various natures:
- technological, with known physical models of the elements making up the actual process to be modeled,
- methodological, taking into account the structuring (hierarchical, distributed, centralized, etc.) of the process being monitored,
- empirical, with context-specific parameterization provided by the expert according to the needs expressed by the plant operator.
These approaches aim to enrich automatic generation by providing a deeper understanding of causal relationships and complex contexts. The proposed
automatic generation will aim to speed up the process of creating and instantiating the digital twin, reducing the need for intensive manual processes,
particularly when deploying on water resource management facilities.
digital twins. These frameworks are playing a crucial role in diverse fields such as the energy sector [Yu et al., 2022], smart cities [Mylonas et al., 2021],
healthcare [Zimmermann et al., 2019] and even water management [Torfs et al., 2022], [Wei et al., 2022]. Their essential role lies in simplifying the
process of creating highly accurate and detailed virtual representations of physical objects and industrial processes, thus contributing to a better
understanding and management of these complex systems. Other initiatives, such as ISO 23247 [Shao, 2021] or the work of [Tao et al., 2019] in the
manufacturing sector, also aim to standardize methods for creating digital twins, ensuring greater consistency and interoperability between different
systems.
However, while these frameworks provide a solid foundation, their use alone is generally not sufficient to generate a complete and accurate digital twin.
Instantiation remains a complex task requiring in-depth knowledge of the real system, integrating relevant data, existing models and the expertise of the
players involved. A digital twin must meet the specific needs of end-users, adding tangible value to operations. The instantiation of a digital twin is
therefore both time-consuming and costly. The aim of this thesis is to simplify this process by automating the creation of digital twins, whether for
existing installations or the deployment of new water resource management infrastructures.
Although recent work in the scientific literature addresses the automatic generation of digital twins, these initiatives focus exclusively on the
manufacturing industry domain. For example, Garcia et al. [Garcia et al., 2023] explore the feasibility of automatically generating digital twins from
historical data for complex manufacturing processes, where exhaustive manual modeling of all line assets is particularly challenging. In [Lugaresi et al.,
2021], the authors propose a novel approach aimed at discovering, selecting and extracting relevant features of a production system from data logs,
enabling the creation of a digital twin with a suitable level of detail. These initiatives focus exclusively on exploiting historical data, giving the digital
twin a significant dependency on both data freshness and completeness. As a result, the digital twin's ability to accurately represent current situations
and track system evolutions will be limited.
While the integration of real-time data or the identification of causal relationships using advanced artificial intelligence methods offer a partial answer to
this problem, complementary approaches will be explored in this thesis.
In particular, we propose to integrate the knowledge of experts in the field, which can be of various natures:
- technological, with known physical models of the elements making up the actual process to be modeled,
- methodological, taking into account the structuring (hierarchical, distributed, centralized, etc.) of the process being monitored,
- empirical, with context-specific parameterization provided by the expert according to the needs expressed by the plant operator.
These approaches aim to enrich automatic generation by providing a deeper understanding of causal relationships and complex contexts. The proposed
automatic generation will aim to speed up the process of creating and instantiating the digital twin, reducing the need for intensive manual processes,
particularly when deploying on water resource management facilities.
Keywords:
Digital Twin, automatic design, Water network.
Department(s):
| Modeling and Control of Industrial Systems |