Application of Data-Driven and Physics-Embedded Neural Networks in Wake Dominated Flows

In this paper, data-driven and physics-embedded deep neural networks (DNN) are analyzed as potential candidates for flow field reconstruction in fluid mechanics applications. The data-driven and physics-based algorithms are employed to model the canonical case of flow past a circular cylinder. It is shown that both types of DNNs model the flow fields well, with the physics-based alternative having a higher reconstruction accuracy. Moreover, it is seen that the data-driven model provides a reasonably accurate solution with a lower computational expense as compared to the physics-embedded model. Finally a discussion on the strengths and drawbacks of each model is presented compared to mainstream numerical methods.