Prediction of Flow and Reactive Transport using Physics Informed Neural Networks

Abstract:

Many general deep learning approaches to solving physical phenomena such as fluid flow, thermodynamics, and magnetism are susceptible to disobeying laws of physics and poor predictions without predetermined physical information. Although fluid dynamics simulations provide fundamental solutions to flow and reactive transport processes, fluid dynamics simulations often suffer in computational efficiency. Physics informed neural network approaches can provide machine learning solutions to physical systems while respecting the laws of physics given by general nonlinear differential equations. In this work, we apply physics informed neural networks to predict fluid flow in a constrained geometry and compare our results with analytical and numerical solutions produced by fluid dynamics simulations. We test our models to evaluate various flow and transport problems in 2D domains using the advection-diffusion and Navier Stokes differential equations. Additionally, we test flow and transport problems in the presence of an obstructing cylinder to analyze fluid velocity and concentration distribution from advection-diffusion-reaction. Comparison of results between the physics-informed deep learning approach and fluid dynamics simulations will be discussed to highlight the accuracy and efficiency of physics-informed neural networks. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525.