Sensitivity driven model updating: a multi-step procedure for structural assessment

Automatic model updating, or model calibration procedures, have significantly streamlined the creation of digital twins, i.e., virtual models strongly informed by experimental data. This methodology supports real-time monitoring and predictive maintenance by ensuring the digital twin accurately reflects the physical structure. However, when the parameters to be inferred are modal parameters, such as natural frequency or mode shape, additional challenges arise. This is especially true in the case of complex monumental buildings. A key issue is ensuring the model not only predicts these parameters correctly but also in the same order as seen in Operational Modal Analysis campaigns. Discrepancies in predicted versus experimental modal parameters often indicate significant modeling errors, resulting in inversions of the predicted vibrational modes. These inversions can invalidate automatic model updating results, slowing the calibration process and leading to potentially incorrect results. Several factors can cause these mode inversions, including inaccuracies in model parameters or modeling simplifications. This work introduces a procedure to mitigate the mode inversion problem by leveraging sensitivity analysis to select the most impactful mechanical parameters and vibration modes for calibration. The proposed methodology involves initial mechanical testing to reduce parameter uncertainties, followed by sensitivity analysis to identify key parameters. Preliminary model assessment and mode matching are then conducted to establish baseline discrepancies. Sensitive parameters are adjusted to align the model more closely with experimental data, followed by a recursive calibration process. The updated model is validated and verified against Operational Modal Analysis results, ensuring correct ordering and reduced errors of modal parameters.