Abstract
This talk will describe recent advances in inverse uncertainty quantification for structural dynamics, and current open challenges [1,2]. Particular focus will be given on discussing challenges in dealing with uncertainty in the data, physics and knowledge encountered in real-world engineering applications, especially when hybrid physics-data (i.e. physics-enhanced machine learning) models are developed.
Uncertainty in the “data” will be discussed in terms of small volume of data and limited knowledge on prior distribution [3], multimodal distributions [4], spatial measurements correlation [5], and the need to obtain virtual measurements at critical locations [6].
The problem of using a physics-based model that is not “complete” or correct in a hybrid physics-data model, would lead to a wrong uncertainty quantification of the model parameters. A “disentanglement” strategy [7] is going to be presented to identify and attribute variability observed in response measurements of an engineering system to variability stemming from the modelled physics, domain, and class influences is going to be introduced.
Finally, an approach for dealing with the uncertainty in the knowledge of the force function model to be used to described non-smooth nonlinearities is going to be presented [8].
Real-world engineering applications would span from laboratory setups to bridges and offshore wind turbines.
[1] Cicirello A., Physics-Enhanced Machine Learning: a position paper for dynamical systems investigations, 2024 J. Phys.: Conf. Ser. 2909 012034, XIVth International Conference on Recent Advances in Structural Dynamics 01/07/2024 - 03/07/2024 Southampton, United Kingdom.
[2] Kamariotis A, Vlachas K., Dertimanis V., Koune I., Cicirello A., Chatzi E., On the consistent classification and treatment of uncertainties in Structural Health Monitoring applications, ASME J. Risk Uncertainty Part B, 2024.
[3] Igea F., Cicirello A., An interacting Wasserstein gradient flow strategy to robust Bayesian inference for application to decision-making in engineering, Data-Centric Engineering, 2025.
[4] Igea F., Cicirello A., Cyclical Variational Bayes Monte Carlo for Efficient Multi-Modal Posterior Distributions Evaluation, MSSP, 2023
[5] Koune I., Rozsas I., Slobbe A., Cicirello A., Bayesian system identification for structures considering spatial and temporal correlation. Data Centric Engineering, 2023.
[6] Zou J., Lourens E., Cicirello A., Virtual sensing of subsoil strain response in monopile-based offshore wind turbines via Gaussian process latent force models, MSSP, 2023.
[7] Koune I., Cicirello A., Adversarial Disentanglement by Backpropagation with Physics-Informed Variational Autoencoder. Preprint:
[8] Lathourakis C., Cicirello, A., Physics Enhanced Sparse Identification of Dynamical Systems with Discontinuous Nonlinearities. Nonlinear Dynamics, 2024
Contatto: