Open AccessParametric Shape Optimization of Stretch Webs in a Progressive Die Process using a Neural Network Surrogate Model
Author(s) -
Shreeram Athreya,
A. Weinschenk,
Florian Steinlehner,
D. Budnick,
M. J. Worswick,
Wolfram Volk,
Stefan Huhn
Publication year - 2021
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/1157/1/012081
Subject(s) - finite element method , artificial neural network , surrogate model , parametric statistics , process (computing) , computer science , die (integrated circuit) , bending , automation , mechanical engineering , artificial intelligence , structural engineering , engineering , machine learning , mathematics , statistics , operating system
Progressive die stamping provides a solution for producing sheet metal parts in large quantities. These parts are connected to carriers by stretch webs. As the part undergoes bending and forming operations, the stretch webs are exposed to translational and rotational deformation. A suitable design of these entities is crucial to avoid failure caused by splits or excessive thinning. A common way to evaluate such designs is to use finite element (FEM) simulation. Since it is not efficient to run FEM based optimization studies for the design optimization and to enable further automation of the stretch web design, this paper is proposing the use of machine learning (ML) technologies. A surrogate model based on an artificial neural network is used as a predictor in the presented study. This neural network is used to optimize the geometric parameters of the stretch web to obtain a quality result. The model is trained using FEM results and the study shows that it was possible to obtain an accurate model with a prediction error of 5%. The trained surrogate model can be used for the optimization study. This approach is computationally inexpensive and can provide very good results.