Applications of automatic differentiation in topology optimization

Sebastian A. Nørgaard, Max Sagebaum, Nicolas R. Gauger, Boyan Stefanov Lazarov

Research output: Contribution to journalJournal articleResearchpeer-review

801 Downloads (Pure)

Abstract

The goal of this article is to demonstrate the applicability and to discuss the advantages and disadvantages of automatic differentiation in topology optimization. The technique makes it possible to wholly or partially automate the evaluation of derivatives for optimization problems and is demonstrated on two separate, previously published types of problems in topology optimization. Two separate software packages for automatic differentiation, CoDiPack and Tapenade are considered, and their performance and usability trade-offs are discussed and compared to a hand coded adjoint gradient evaluation process. Finally, the resulting optimization framework is verified by applying it to a non-trivial unsteady flow topology optimization problem.
Original languageEnglish
JournalStructural and Multidisciplinary Optimization
Volume56
Issue number5
Pages (from-to)1135-1146
ISSN1615-147X
DOIs
Publication statusPublished - 2017

Keywords

  • Automatic differentiation
  • Lattice Boltzmann
  • Topology optimization

Fingerprint Dive into the research topics of 'Applications of automatic differentiation in topology optimization'. Together they form a unique fingerprint.

Cite this