Subject Ph. D.

The Digital Twin is a concept, generalising the notion of Digital Mock-up, which proposes a model, in a digital environment, of an object, system or process from the physical world, called the Physical Twin [Grieves 2005], [Grieves 2015], [Fei et al. 2018]. The model and the object are representative of each other. The targeted applications are multiple in the frameworks of supervision or piloting [Cimino et al. 2019]. Although JN is considered to be a key concept in the digital transition of companies (manufacturers, infrastructure managers, etc.) and institutions (hospitals), it remains costly to implement and must therefore be used by all the players concerned in order to be profitable. In this context, user confidence in the system must be established. A JN in which users do not have confidence would be useless and even counter-productive. The concept of trust is already well studied in the context of scenario management applications of the 'What-if? [Pires et al. 2023]. In these, the JN is constructed using behavioural models that allow events to be simulated in order to predict its reactions. In this case, many colleagues are working on trust in recommendation systems, as was done with expert systems in the 90s. The problems studied are those linked to the initialisation of the model (Cold Start) [Son 2016], lack of data [Reshma et al 2016] and the quality of predictions [Meyer 2012]. Work has been carried out to improve confidence in the model or recommendation system [Pires et al 2023]. On the other hand, in the case of JNs built on relational models (BOM, graphs, tables, etc.), the concept of trust is not as much debated [Issa 2023]. This type of JN is used by manufacturers (CPS, production systems, complex systems, etc.) and infrastructure managers (transport, buildings, heritage, etc.) for Asset Management and to plan changes throughout their lifecycle (maintenance work, technological upgrades, etc.) [Issa 2023]. Here, it is not so much the problem of confidence in the model that is of interest, but rather that of confidence in the data. Indeed, the accuracy of the representativeness of the JN in relation to the JP is important and influences a user's confidence in it. In turn, representativeness is linked to the data used to feed the models that make up the JN. Consequently, confidence in the representativeness of the JN is linked to confidence in the data (are they up to date? are they understandable? are they complete? have they been corrupted? etc.). These issues have been studied by the scientific community for many years and are linked to the digitisation of economic activities and the rise of the Internet [Abdul-Raham et al 1998]. They were even the subject of a special initiative by the US Department of Defence as early as 1985 [DoD 1985]. A number of solutions have been developed to address these issues, including blockchain and data certification [Geisler et al. 2003]. However, the literature lacks a metric that would specifically assess the accuracy of the representativeness of the JN and therefore indirectly the confidence that a user might have in the JN when planning a JP evolution, depending on the evolution scenario envisaged. This metric for qualifying the accuracy of the representation is the subject of the study proposed as part of this PhD thesis funding application. The aim of the thesis work will be to define and evaluate a confidence metric that will be a function of the accuracy of the representation of the Digital Twinning (DT) in relation to the Physical Twinning (PT). A deviation may occur because the JN and JP life cycles evolve in parallel. On the one hand (JP), there is work/evolution, planned or not, documented or not by feedback in the JN and, on the other hand (JN), frequent or event-driven, automatic or manual updating of the data, upstream or downstream of the evolutions in the JP. This discrepancy needs to be studied. Our working hypothesis is that this deviation can be suspected and/or observed by the appearance of different values for the same data in redundant documentary sources. In fact, these redundant sources describe all or part of the JP and contain the information needed to instantiate the JN. They come from different phases of the system's life cycle (design, manufacture, maintenance, etc.) and can be of heterogeneous nature and form (textual reports, 2D/3D models, mathematical models, photographs, etc.). The deviation must be identified and assessed. To do this, it is possible to imagine using graph theory to navigate the JN data structure and an ontology to homogenise the data so that it can be compared (or an ontology to extract metadata). Once this deviation has been identified, defined and modelled, the PhD student will have to propose a confidence metric based on it. To do this, he or she will draw on an international bibliographical study as well as on some relational modelling work from the laboratory [Naanaa 2022] or the supervising team [Issa 2023], [Tchana 2021].