Topologieoptimierung mittels Deep Learning ohne voroptimierte Trainingsdaten
DS 106: Proceedings of the 31st Symposium Design for X (DFX2020)
Year: 2020
Editor: Dieter Krause; Kristin Paetzold; Sandro Wartzack
Author: Halle, Alex; Campanile, Flavio L.; Hasse, Alexander
Series: DfX
Institution: Professorship Machine Elements and Product Development; Chemnitz University of Technology
Section: Lightweight Design
Page(s): 101-110
DOI number: https://doi.org/10.35199/dfx2020.11
Abstract
Here a method for topology optimization is presented which is able to obtain optimized geometries without iterative optimum search. The optimized geometries are provided by an artificial neural network, the predictor, on the basis of boundary conditions and degree of filling as input data. In the training phase, geometries generated on the basis of random input data are evaluated with respect to given criteria and the results of those evaluations flow into an objective function which is minimized. Other than in state-of-the-art procedures, no pre-optimized geometries are used during training. The trained predictor supplies geometries which are similar to the ones generated by conventional topology optimizers, but requires only a small fraction of the computational effort.
Keywords: deep learning, topology optimization, artificial neural networks, AI in design