Iterative solvers for metamaterials modelling via machine learning

Abstract

The project aims to develop a neural network that predicts the stiffness matrix of a mesh based on the radius of the inclusion in it. This would aim to enhance iterative solvers using preconditioners, reducing the necessity of large libraries of data. To achieve this, data will be obtained by simulating elastic problems with appropriate constraints, testing the data using a solver with a preconditioner. The final step is to create a neural network capable of providing the necessary data with minimal inputs.