CONCEPT AND APPLICATION OF AUTOMATIC PART-RECOGNITION WITH ARTIFICIAL NEURAL NETWORKS FOR FE SIMULATIONS
Editor: Christian Weber, Stephan Husung, Marco CantaMESsa, Gaetano Cascini, Dorian Marjanovic, Serena Graziosi
Author: Spruegel, Tobias C.; Wartzack, Sandro
Institution: Friedrich-Alexander-Universitaet Erlangen-Nuernberg, Germany
Section: Design Methods and Tools - part 2
Currently available finite element software is consequently getting more and more user-friendly, and simulation knowledge must be expanded to a growing pool of new and less-experienced users; for example designers. These users need to be assisted when performing simulation tasks. A promising way is to check the FE results for plausibility, but therefore, it is essential to know what kind of parts should be simulated and checked for plausibility. This paper presents a concept for the automatic recognition of standard parts within the FE software, using a predefined Artificial Neural Network (ANN). An existing approach is extended to deal with the issues of arbitrarily oriented parts in CAD (computer-aided design) assemblies of differing size in 3D space. Hence a Principal Component Analysis (PCA), and a spherical detector surface for FE nodes in spherical coordinates is used to gain input parameters for an ANN. In the paper, the concept and an application for certain standard components are demonstrated.