Hybrid Eigensensitivity‐Machine Learning Framework for Model Updating of a Generally Damped Base‐Isolated Bridge

Structural model updating of generally damped systems often requires prior assumptions on structural parameters to constrain the solution space when classical eigensensitivity (ES)‐based methods are employed. This study proposes a hybrid ES‐machine learning (ML) framework that integrates physics‐based ES analysis with supervised learning for estimating physically admissible structural parameters in generally damped base‐isolated systems while accounting for variability in known parameters. An iterative ES‐based procedure was implemented to quantify relationships between structural and modal parameters and to generate physics‐informed training data under controlled parameter variability. These data were then used to train ML models to estimate structural parameters corresponding to experimentally identified modal properties. The proposed methodology was implemented on field‐recorded seismic data from a multispan base‐isolated bridge and evaluated using four ML models. During training, Gaussian process regression (GPR), support vector machine (SVM), and bagged tree ensemble (BTE) models captured nonlinear relationships between modal and structural parameters better than the multivariable linear regression (MVLR) model. Structural parameters estimated by the baseline ES and ML models satisfied the physical bounds and were validated using the structural response metrics of the modal residuals ( δ z ), measure of fit in acceleration response ( E ) and modal assurance criterion (MAC). Modal residuals ( δ z ) were within the tolerance limits for the baseline ES and 2 ML models (GPR and SVM), and MAC values were greater than 0.97, while corresponding E values ranged between 0.164 and 0.301. This indicated excellent modal correlation and good structural response accuracy with the recorded parameters. The best‐performing GPR ML model achieved average δ z and E values that were 29.1% and 58.1% better than the baseline ES model, demonstrating the usefulness of adding ML to the ES based model updating process.