A data-driven vibration response reconstruction for damping uncertainties when generalizing from the numerical model to the experimental problem

In many engineering applications, Finite Element (FE) models are calibrated in a reference structural state but must also simulate other states. For example, vibration responses of a damaged condition may be simulated using a model calibrated on the intact state. In such cases, damping changes due to damage are generally uncertain, leading to reliability issues in the simulated responses. In this work, we present a methodology that transforms vibration responses to constant damping ratios, invariant to the structural state, using data-driven reconstruction techniques. The aim is to mitigate uncertainties related to damping caused by structural changes. We find that when vibration responses are reconstructed using this methodology, more robust features can be learned from FE data, and the simulated structural conditions generalize better to the experimental domain. The reconstruction methodology, along with a sensitivity analysis and discussion of limitations, is provided. The present work can contribute to enhancing the reliability of physics-based Digital Twin (DT) models and improving cross-domain generalization between FE numerical data and experimental structures under uncertainty. • Α novel structural response reconstruction methodology. • Reliability of state estimation using FE simulation data, for DT and SHM. • Uncertainty in damping and its changes in time-response FE simulations. • Population based scheme that is not affected by damping uncertainties. • ANN trained on the numerical population enhances its generalization capability.