| 초록 |
Here, we introduce a deep learning method in which dynamics solutions from the nonlinear partial differential equations, including the Navier-Stokes equation, are inferred using the tailored physics-informed neural network architecture for chemical reactors. We incorporate the first-principle governing equation such as species transport equations with finite-rate volumetric reaction and conservative equations for mass, momentum, and energy into the deep neural network architecture. Thus, the trained networks effectively infer both data output responses and physics behavior simultaneously, which leads to the ideal regularization and the ability of extrapolation. Long-term dynamics of three-dimensional continuous stirred tank reactor and von Kármán vortex simulations are exemplified to validate the performance of our surrogate models. |
