Aerodynamic Force Reconstruction Using Physics-Informed Gaussian Processes

Abstract

A probabilistic physics-informed machine-learning approach is developed to recover aerodynamic loads from noisy measurements of structural response. The formulation avoids separate regularisation schemes and can combine heterogeneous data of different fidelity. Its application to the Great Belt East Bridge reproduces simulated aerodynamic loads accurately in magnitude, phase, peak values and root-mean-square error, supporting applications in model validation, load estimation and structural prognosis.