Deep-neural-network-based framework for the accelerating uncertainty quantification of a structural–acoustic fully coupled system in a shallow sea

Abstract

To systematically quantify certain uncertainties within the vibro-acoustic coupling problems, we propose a framework for sampling the acceleration and uncertainty quantification based on a Deep Neural Network (DNN). Coupling the Finite Element Method (FEM) and Boundary Element Method (BEM) with Catmull–Clark subdivision surfaces to generate samples for DNN training and testing. Constructing various DNN surrogate models with different input dimensions to generate abundant samples for uncertainty quantification. Applying a highly structured neural network Stochastic Differential Equation Network (SDE-Net) for uncertainty quantification using these samples. Numerical examples are implemented to verify the validity and effectiveness of the proposed algorithm. The main results are as follows: The numerical results match well with the analytical results, indicating that the FEM–BEM can accurately solve the fully coupled structural–acoustic system and provide high-quality initial samples for the DNN; The statistical metrics of DNN testing results demonstrate extremely high prediction accuracy; The proposed uncertainty quantification framework offers a time advantage and potential for dealing with more complex physical problems.