This study focuses on developing a deep learning-based prediction framework for fluid-structure interaction (FSI) problems involving large deformation, with particular emphasis on flexible membrane structures. They exhibit strong coupling between fluid forces and structural deformation, posing significant challenges for conventional CFD–CSD simulations due to mesh distortion and high computational costs. To address this issue, we aim to construct a Masked Deep Neural Network (MDNN) that combines convolutional layers (CNN) for spatial feature extraction, ConvLSTM for temporal dynamics, and a masking mechanism to concentrate computation on highly deforming regions. The proposed model is expected to reduce simulation time while maintaining high accuracy and generalizability. As a first step, we plan to apply the MDNN to a representative case of a flexible membrane under dynamic pressure loading.
Liu et al. (Wed,) studied this question.