Uncertainty quantification of vibro-acoustic coupling problems for robotic manta ray models based on deep learning

This study proposes a deep learning framework to perform uncertainty quantification of vibro-acoustic coupling problems for robot manta rays. First, Loop subdivision surfaces are used to build the geometric models of robot manta rays. Next, by incorporating the geometric modelling basis functions for numerical simulation, we couple isogeometric finite element and boundary element methods to calculate the sound pressure in the exterior domain of the structure, which generates initial samples for surrogate modelling. Then, deep neural networks are trained as surrogate models with multi-dimensional random inputs to expand the dataset for uncertainty quantification. Finally, the SDE-Net is employed to use the expanded data to quantify uncertainties of system responses caused by geometric and material parameters. Numerical experiments are given to demonstrate the accuracy and effectiveness of the present method.