Keynote 1: Structural Dynamics as a principal pillar to AI-enhanced Digital Twinning
 | Eleni Chatzi is a Full Professor and Chair of Structural Mechanics and Monitoring at the Institute of Structural Engineering of the Department of Civil, Environmental and Geomatic Engineering of ETH Zürich. She currently serves as the President of the European Academy of Wind Energy (EAWE). Her research interests include the fields of Structural Health Monitoring (SHM), hybrid modelling for digital twinning, and data-driven decision support for engineered systems. She is an author of over 350 papers in peer-reviewed journals and conference proceedings, and further serves as an editor for several international journals in the domains of Dynamics and SHM. She led the recently completed ERC Starting Grant WINDMIL on the topic of “Smart Monitoring, Inspection and Life-Cycle Assessment of Wind Turbines”. Her work in the domain of self-aware infrastructure was recognized with the 2020 Walter L. Huber Research prize, awarded by the American Society of Civil Engineers (ASCE). She is further recipient of the 2020 EASD Junior Research Prize in the area of Computational Structural Dynamics, the JM Ko Award and the 2024 SHM Person of the Year award. Abstract: This talk will offer a view to the role of dynamics, often characterised by structured inference schemes, such as Modal Analysis, as a cornerstone to advancing AI-enhanced Digital Twins. Given the highly individual nature of structural assets, dynamical signatures can offer foundational insights into the behavior of complex infrastructures, which operate under highly fluctuating real-world conditions. This talk explores the integration of Structural Health Monitoring (SHM) with physics-enhanced machine learning (PEML), where domain-specific physical principles enrich machine learning models, increasing their robustness and predictive accuracy. Dynamics will be showcased as a main ingredient to such a goal, through i) fusion of related equations and properties into ML-based learning schemes, ii) use of formal grammars as vocabularies that admit dynamics principles as “rules” for model discovery, iii) graph architectures, which can infer dynamic interdependencies at system level, on the basis of simulation and monitoring information. The fusion of data-driven AI with physical laws paves the way for next-generation infrastructure management, where resilient, adaptive systems contribute to sustainable urban and industrial growth. |
Lecture of Honor: Time series based robust damage and fault diagnosis for structures and systems under uncertainty
 | Spilios Fassois is Professor and Director of the Stochastic Mechanical Systems and Automation SMSA Laboratory at the University of Patras Greece. His research interests are on stochastic mechanical and aeronautical systems, statistical time series methods, data-based modeling, diagnostics, Structural Health Monitoring, and Machine Learning with applications on structural, vehicular, aeronautical, and other types of systems. He is the recipient of the 2023 Evangelos Papanoutsos Excellence in Teaching Award at the University of Patras, the 1990 Excellence in Teaching Award of the College of Engineering at the University of Michigan – Ann Arbor, and various other awards and distinctions. He is Editor-in-Chief for the Journal of Mechanical Systems and Signal Processing, Associate Editor and Editorial Board Member for various other international journals, and Scientific Committee member for numerous international conferences. He has given numerous invited presentations, has organized 4 Thematic Issues for esteemed international journals, and published over 320 articles in prestigious technical journals, conference proceedings, and technical encyclopedias. Abstract: The presentation focuses on statistical time series based damage and fault diagnosis for structures and engineering systems operating under uncertainty. The various versions of the problem formulation are reviewed, and a concise, yet critical, overview of the main principles, underlying assumptions, and available approaches is presented. The need for robustness, arising from the necessity for counteracting the effects of uncertain Environmental and Operational Conditions (EOCs), but also those associated with populations of similar structures and systems, is demonstrated. The main concepts and approaches of robust methods are then critically reviewed, with emphasis on conceptual and practical simplicity, ease of use, operation with a low number of sensors and limited numbers of training signals, physical interpretability, and the achievement of high-performance levels for even `minor’ fault levels. The novel and holistic Functional Model (FM) based method, within which the subproblems of damage/fault detection, precise localization, and level estimation may be seamlessly addressed, is subsequently introduced. Its various forms, including those based on measurable EOCs and recent ones capable of eliminating this requirement, are discussed. Application case studies, pertaining to damage diagnosis for engineering structures of various types and on-board fault diagnosis for railway suspension systems under uncertainty, are then presented, with diagnostic performance systematically assessed via Receiver Operating Characteristic curves and related metrics. The presentation concludes with remarks on current achievements, the technology’s status and limitations, and perspectives on the way forward. |
