Model updating is the process of calibrating a numerical model — typically a finite element model — to match measured data from a real structure, usually by adjusting model parameters such as stiffness, mass, or boundary conditions. It is essential for accurate prediction of structural response and for reliable SHM.
Model updating is the process of adjusting the parameters of a numerical model — typically a finite element (FE) model — so that its predictions match measured data from a real structure. The need for model updating arises because initial FE models, based on design drawings and assumed material properties, are inevitably inaccurate: they do not capture construction tolerances, actual material properties, non-structural elements, soil-structure interaction, or changes caused by aging, damage, or modification. By comparing measured responses — usually modal properties from ambient vibration or forced vibration tests — with model predictions, the model can be calibrated to better represent the real structure.
Model updating is broadly classified into two categories. Direct methods update the mass and stiffness matrices of the model using mathematical procedures (such as optimal matrix update or eigenstructure assignment) to reproduce measured modal properties exactly. These methods are computationally efficient but produce updated matrices that may not preserve physical meaning (e.g., they may introduce non-physical couplings between degrees of freedom). Iterative methods — also called parametric methods — adjust a set of selected physical parameters (stiffness of specific elements, mass distribution, boundary conditions) through an optimization process that minimizes the difference between measured and computed responses. These methods preserve physical meaning but can be computationally intensive and may have multiple local minima. The choice of method depends on the objective, the available data, and the required accuracy.
Model updating is essential for several applications. In performance-based earthquake engineering, accurate models are needed to predict the response of a structure to future earthquakes — a task that is impossible without calibration to measured behavior. In structural health monitoring, model updating is used both for baseline characterization (calibrating the healthy-state model) and for damage detection (identifying changes in the model that correspond to damage). In vibration-based damage detection, the difference between the updated model and the original FE model — the parameter changes required to fit the data — provides a direct indication of damage location and severity. Key challenges in model updating include (1) the ill-conditioning of the problem — many parameter sets may produce similar responses, leading to non-unique solutions; (2) the presence of modeling errors that cannot be corrected by parameter adjustment alone; (3) the influence of environmental and operational variability on measured data; and (4) the computational cost of iterative optimization for large models. Modern practice increasingly uses Bayesian methods, which provide a probabilistic framework for model updating that quantifies uncertainty in the updated parameters. In Iran, model updating is used in SHM projects on bridges, dams, and important buildings, and in performance assessment of existing structures, with growing adoption of Bayesian and machine-learning-assisted methods.