Developing a data-driven model for unsteady prediction of the flow features using the attention-based approach

We propose a deep learning framework designed to predict the space-time evolution of complex flow problems across various parametric regimes. In this study, we utilize a data-driven model enhanced with a design gate to forecast the space-time evolution of these complex flow scenarios. To generate training data for our approach, we extract unsteady flow solutions around deforming objects from numerical simulations spanning the design space. This dataset serves as the foundation for building our data-driven model, which is based on a transformer architecture enhanced by the design gate. This attention-based structure enhances the neural network's capacity to learn from complex design spaces. We demonstrate that our data-driven model can accurately predict the evolution of the solutions for different objects. This capability allows us to integrate the model with a structural framework in the future works, enabling potential applications in two-way fluid-structure interaction problems.