Development of a Neural Network based Finite Element Method for Thermo-Mechanically coupled Problems
At a glance
The research project aims to develop a neural network based finite element (FE) approach for thermo-mechanically coupled problems. Such an approach holds the large potential to improve the computational performance of conventional FEM especially for large structures containing small local features (e.g. notches, bores) that are critical for their mechanical integrity under complex mechanical and thermo-mechanical loading conditions (multi-scale problem). Additionally, the approach is promising to more efficiently and more effectively predict the response in systems undergoing complex multi-physical loading. This is especially relevant e.g. for the fast and accurate prediction of distortion during or of the material properties of structures as produced by the additive manufacturing process. The application of machine learning methods for mechanical problems is further a strategically important topic for IMES and offers promising perspectives for future applied R&D projects e.g. for the design of efficient digital twins of mechanical systems, neural network based material and damage models and process-structure-property models allowing for the prediction of mechanical properties from manufacturing parameters using mechanical and material physical multi-scale and multi-physics models.