Inner-product optical neural processing and supervised learning

Abstract

Increasing demands for reliable production processes of micro-scale components in large quantities has promoted the research on micro cold forming processes. The need to accelerate product design by the use of simulation has revealed limitations of the classical finite element method (FEM) in the sub-millimeter domain where the local behavior may be dominated by one or a few grains only. As a consequence, average material characteristic values obtained from macroscopic specimens are no longer sufficient to describe the local work-piece behavior. Instead, an uncertainty of the material parameters is encountered, which can be characterized mathematically by probability distributions.The research activities presented here focus on the feasibility of FEM simulations based on distributed material parameters. In order to reduce the computational load, pre-calculation of a sufficient number of load cases in representative volume elements (RVE) for usage in a material data base is proposed. Work-piece simulations are then performed as a multi-scale approach with different levels of interactions between adjacent RVEs.Material characterizations of DC01 sheets and their application to numerical simulation of bulge-tests are presented.