Direct data-driven computational mechanics for anelastic material behaviours

点击翻译按钮获取中文摘要

Due to the volatile development of storage capacities as well as suitable soft- and hardware, the amount of available data has increased by many orders of magnitude over the last decades. This influences almost all parts of practical life but also science and technology. In particular, the area of engineering and many fields of applied sciences have a significant role in it, since they require to provide, in addition to fundamental conservation principles, relations between fields of interest, for example stress and strain. Traditionally, these relations are derived from constitutive models, relying on a number of assumptions, and prone to significant epistemic uncertainty. Additionally, these models involve parameters, which might be difficult to be identified, especially with relatively simple experiments such as uniaxial tests. The objective of this project is thus to develop methods allowing to perform numerical simulations of the behavior of structures directly from available data (either experimental or coming from fine scale computations), eliminating the necessity to formulate phenomenological constitutive models and the uncertainty associated with them. The methodology has already been demonstrated for elastic materials. The project is based on the research hypothesis that the framework can be extended to inelastic material behaviors, such as elasto-plasticity or visco-elasticity, overcoming the challenge of the resulting increased dimen-sionality of phase space.A general view of a data-driven approach to inelasticity has already been given by the proposers in a recent paper. The objectives of this project are to implement and assess the methodology in a framework able to handle cases oriented towards industrial applications: use of actual experimental and eventually incomplete data, complex 3D geometries and loading, efficiency and robustness of solvers, uncertainty quantification. An additional objective is to provide an online platform allowing to share the methodology and the associated data with the scientific community.