A surrogate model based on Physics Informed Neural Networks for predicting wake flow past a cylinder

This paper discusses a surrogate model based on Physics Informed Neural Networks (PINNs) for complete data reconstruction from limited labeled data. A benchmark problem of 2D unsteady laminar flow past a cylinder is chosen for comparing the accuracy of existing PINN models to the proposed surrogate model. The proposed surrogate model is an improvement on the existing Backward Compatible PINN (BC-PINN) methodology. We have demonstrated that the performance of the existing Vanilla BC-PINN methodology in solving Navier Stokes equation is at a similar level to that of Standard PINN methodology using mini-batches. However, the proposed methodology resulted in a significant increase in accuracy in comparison with the vanilla BC-PINN.