This presentation outlines a vision and a methodological approach currently being explored through initial research for the consistent development of Digital Twins. It explores the integration of Data Representations with the incremental construction of a system Behavior Model, drawing on initial research activities to bridge the gap between theoretical modeling and practical implementation. The ability to integrate Data and Behavior is considered as a means to further consolidate the specification and implementation of reliable DTs in different problem domains.

The presentation offers a comprehensive exploration of the transition from static data mirroring to autonomous, AI-driven systems. At the heart of this evolution is the strong correlation between the Descriptive Model, which serves as the digital “reflection” focusing on data structure and their actual values, and the Behavior Model, the digital “engine” that formalizes causality and governs how a system reacts and evolves over time. By leveraging these two models, Digital Twins (DTs) gain the ability to not only explain why a physical object transitions from one state to another but also to provide the contextualization necessary for advanced control and simulation.

The session proposes a 4-stage maturity roadmap based on ETSI models, starting with Passive/Descriptive high-fidelity data mirroring before moving into Predictive stages where the Behavior Engine enables real-time simulation. The journey continues into Prognostic DTs, which focus on resilience by predicting a “failure of conduct” rather than just a state change, and finally reaches Prescriptive DTs that function autonomously to optimize the physical system.

A central pillar of the presentation is the ongoing definition of a methodological framework that combines top-down and bottom-up approaches to ensure a comprehensive system representation. In a top-down specification, the design process starts in parallel with the physical system, where significant attention is paid to identifying the necessary functions and API interfaces that enable seamless integration between the physical object and its digital counterpart. This approach ensures that the DT is built with the right “hooks” and functionalities from the start, making the development more effective than attempting to add a DT to an already existing system.

Conversely, the bottom-up approach is utilized when creating a DT for systems that lack initial formal specifications, such as in Smart City environments. In this scenario, the focus shifts to leveraging historical and real-time data to not only predict future states but also to incrementally derive the underlying rules, invariant constraints, and executable logic that characterize the system’s behavior. The objective, pursued through the research activities behind this presentation, is to transform observed evidence from the physical world into formal models (such as Statecharts or activity diagrams) by leveraging techniques like metamodeling and General Purpose Sensing. This allows for an actual Behavior Model to be incrementally constructed, even when the original system design was not explicitly defined.

In addition, the research aims to investigate whether and how Explainable AI can be integrated with DT capabilities to provide transparency in system behavior and decision-making. It addresses the “black-box” nature of Deep Learning (e.g., Reinforcement Learning) and evaluates the feasibility of transforming complex neural weights into human-readable activity diagrams and formal rules.Furthermore, as part of the research objectives, there is the aim to define a DT Development Environment that leverages Generative AI for a twofold purpose: stress-testing and behavioral discovery. By generating synthetic scenarios for cases where data is scarce or non-existent, GenAI facilitates the characterization and verification of how the Behavior Model should react in critical, rare, or complex edge cases.

Speaker bios:
Prof. Noel Crespi holds Masters degrees from the Universities of Paris-Saclay (formerly Orsay) and Kent (UK), a diplome d’ingénieur from Telecom Paris, and a Ph.D and an Habilitation from Sorbonne University. From 1993 he worked at CLIP, Bouygues Telecom and then at Orange Labs in 1995. He took leading roles in the creation of new services with the successful conception and launch of Orange prepaid service (15M+ subscribers), and in standardization (from the rapporteurship of IN standard to the coordination of all mobile standards activities for Orange). In 1999, he joined Nortel Networks as telephony program manager, architecting core network products for the EMEA region. He joined Institut Mines-Telecom, Telecom SudParis in 2002 and is currently Professor and Program Director at Institut Polytechnique de Paris, leading the Data Intelligence and Communication Engineering Lab. He coordinates the standardization activities for Institut Mines-Telecom at ITU-T and ETSI. He was an adjunct professor at KAIST (South Korea), a guest researcher at the University of Goettingen (Germany) and an affiliate professor at Concordia University (Canada). He is the scientific director of AI2, a French-Korean laboratory. As a Principal Investigator or Co-Investigator, he has secured research grants added up to 8M+ euros. His current research interests are in Edge Intelligence, IoT, Digital Twin, Artificial Intelligence and NLP.

Roberto Minerva is an Emeritus Associate Professor at the Institut Polytechnique de Paris – Télécom SudParis. His career bridges decades of research in industrial innovation with academic contributions, focusing on the convergence of Distributed Systems, the Internet of Things (IoT), and advanced platforms driven by Artificial Intelligence (AI) and Digital Twin technology. Dr. Minerva’s academic foundation includes a Master’s Degree in Computer Science from the University of Bari (Italy), a Ph.D. in Computer Science and Telecommunications, and the HDR from Sorbonne University (France). From 1987 to 2016, his industrial tenure was spent at CSELT/TILAB (the advanced research branch of TIM – Telecom Italia), a historical hub for Italian telecommunications innovation. He began as a researcher and served as a Research Manager from 1996 on, leading groups dedicated to pioneering foundational technologies, including SDN/NFV, 5G, Big Data, and IoT. He contributed to international cooperative efforts like TINA-C (Service Architectures) and provided global leadership as Chairman of the IEEE IoT Initiative between 2013 and 2016. In 2017, he transitioned to the academic sector, joining Télécom SudParis before achieving Emeritus status. His current research remains centered on IoT, Digital Twin platforms, and the strategic exploitation of AI techniques. He is the author of numerous influential papers published across leading international venues.