Bayesian optimization-based calibration of 3D discrete element models from shake table test data

Discrete Element (DE) models effectively capture the dynamic behaviour of blocky structures, yet calibrating their parameters remains challenging. This paper presents a Bayesian optimization-based method to calibrate DE models using measured data. The approach is applied to a scaled prototype of a Roman Obelisk tested on a shake table. Stiffness and friction angle of non-linear springs at contacts, and mass-proportional Rayleigh damping, were calibrated to match experimental displacements and frequencies. The method efficiently derived mechanical parameters for the structural model, which accurately predicted the seismic response of the obelisk tested in the laboratory. Results highlight the potential for expanding this technique to field applications, where periodically updated digital twins could support targeted inspection and assessment.