Just another WordPress site - Ruhr-Universität Bochum

Abstract

The possibilities of using machine learning models to support or even replace traditional models in solid mechanics have inspired a plethora of new ideas and research directions. For example, in data-driven mechanics, completely new formulations of mechanical equilibrium under non-linear material responses have been introduced. Other works are seeking to replace constitutive models, which are commonly formulated in terms of closed-form algebraic equations or as ordinary differential equations, by trained machine learning algorithms. Such a trained machine learning model can be considered as a digital material twin as it describes the specific material performance and can be used to adapt processing conditions on-line or to predict the remaining lifetime of a component. It is demonstrated that machine learning models can be successfully trained with history-dependent mechanical data for polycrystals with different crystallographic textures. The required training data is obtained from micromechanical simulations based on the crystal plasticity method. The trained model accurately describes the texture-specific elastic-plastic material response under multiaxial loading conditions. In future work, the dynamics of microstructure evolution under thermal and mechanical loads will be included, to fully integrate the process-microstructure-property conditions in the digital material twin.