Fluid flow diffusion modelling with convolutional neural network aided method.

Here is presented a machine learning approach for 2D steady-state fluid gas diffusion field calculations based on Poisson equation for Dirichlet boundary problem. The approach is implemented by a multilayer convolutional neural network (CNN) architecture with output classification layer used as minimization acceptance criterion. Developed CNN architecture is presented, with field image stacks of training and testing data. CNN convergence dynamics is observed and estimated with relative error field differences. Possible applicability of the method for different physical fields modeling is discussed. Presented CNN machine is a step forward in advancement of machine learning methods in computational physics and numerical field modeling. [ABSTRACT FROM AUTHOR]