Integration of Experiments and Simulations to Build Material Big-Data

Abstract

In this paper, a method for extracting stress-strain databases from material test measurements is introduced as one of the potential Integrated Computational Materials Engineering (ICME) tools. Measuring spatially heterogeneous stress and strain evolutionary data during material tests is a challenging and costly task. The proposed method can extract a large volume of spatially heterogeneous stress and strain evolutionary data from experimental boundary measurements such as tractions and displacements. For the purpose, nonlinear finite element models are intrusively implemented with artificial neural network (ANN)-based material constitutive models. Then a specialized algorithm that can auto-progressively train ANN material models guided by experimental measurements is executed. Any complex constitutive law is not presumed. From the algorithm, ANN gradually learns complex material constitutive behavior. The training databases are gradually accumulated with self-corrected stress and strain data predicted by the ANN. Finally, material databases are obtained. For an example, visco-elastoplastic material databases are obtained by the proposed method.