Spatially defined optimization of FEA using nodal surrogate models

Abstract

Optimization of the structural response for a part relies upon computationally expensive simulations such as finite element analysis (FEA). Surrogate models are able to make cheap predictions of the results; however, traditionally they can only predict a single value (SV) such as a maximum stress or weight value. Recently, a surrogate modeling method has been developed for predicting the full field (FF) of nodal responses in an FEA simulation. This research applies FF surrogate models to optimization, and explores various techniques that are uniquely enabled by these cheap—yet detailed—predictions. Because the FF surrogate models predict the response at every node, constraints and objectives can be spatially defined on a part rather than act on a single value. Regional constraints allow for control over a subset of the nodes in areas of interest on the part. Location-based objectives find designs that either draw a response closer to or drive a response away from a particular node. Pattern-matching objectives find designs in the design space that have a response pattern across the part surface that is as similar as possible to a pre-defined response pattern. These techniques extend the usefulness of FF surrogate models as well as optimization of FEA results for exploring a design space and improving a design.