Replacing FEA for sheet metal forming by surrogate modeling

Abstract

This paper presents an innovative approach for employing surrogate modeling in computer-aided design of sheet metal work pieces. The surrogate models replace expensive finite element analysis and thus shorten development time in the early design stages considerably. We experimentally compare neural networks and Kriging models for the task of predicting the sheet metal thickness after cold forming of B-pillars in automotive engineering. We also discuss some aspects of handling large data-sets and exploiting certain structures in the data.