• Matthias Faes, Ku Leuven
• Stefano Marelli
• Jean-Marc Bourinet
• Enrico Zio

The safe and reliable design of engineering structures and systems, as well as the online assessment of their reliability and safety, depends largely on high-resolution numerical models that approximate their physical behavior. However, solving these models often takes hours or even days, and in the context of uncertainty quantification, reliability analysis and design optimization, the solution procedure must be repeated multiple times for different values of the input parameters. Their application, then, requires high-performance computing facilities, which are not always available. Machine learning approaches and surrogate modelling techniques can provide a solution to this problem by representing the physical system through a data-driven (mostly black-box) approach and constructing an efficient emulator of the high-resolution numerical model. More recently, physics-informed and so-called ‘grey-box' modelling approaches have been introduced as a new paradigm to improve the generalization and training capabilities of data-driven (black-box) models.

On this topic, this mini-symposium is aimed at gathering expert researchers, academics and practicing engineers to present their recent findings, methodological developments, as well as innovative applications, related to the efficient combination of computationally intensive numerical simulation codes with efficient surrogate models and/or data-driven black-box approaches, as well as grey-box and other hybrid combinations of machine learning and numerical modelling