Improving FEM Calibration Accuracy Using Bayesian Updating and Surrogates

Abstract

Finite Element Models (FEMs) are pivotal in engineering simulations, yet their accuracy heavily relies on precise calibration to reflect real-world behavior. Traditional calibration methods often struggle with computational complexity, measurement uncertainties, and model inadequacies. This research introduces a hybrid methodology integrating Bayesian updating and surrogate modeling to enhance FEM calibration accuracy while maintaining computational efficiency. Bayesian methods systematically incorporate prior knowledge and observational data, quantifying uncertainty and refining parameter estimates. Meanwhile, surrogate models, such as Gaussian Process Regression (GPR), serve as computationally cheap approximations to high-fidelity FEMs, enabling efficient exploration of the parameter space. Experimental validation using a structural beam benchmark problem demonstrates that the proposed approach significantly reduces the error between model predictions and experimental observations. The hybrid framework also provides probabilistic bounds on calibration parameters, allowing robust and interpretable uncertainty quantification. Results confirm the method's potential for improving FEM reliability across a variety of engineering domains.