Aspects of accuracy and uniqueness of solutions in data-driven mechanics

Abstract

Data-driven methods provide great potential for future applications in engineering, for example in terms of more efficient simulations. Conventional material models and the associated constitutive equations are substituted by a minimization of a distance between so-called material and mechanical states, which, however, leads to non-unique solutions. The aim of this paper is to analyze the influence of the chosen initial values on the accuracy of the obtained results. Furthermore, Mixed Integer Quadratic Programming (MIQP) is implemented and its applicability to data-driven mechanics is assessed.