Statistical analysis of control parameters in evolutionary map L-systems-based topology optimization
Statistical analysis of control parameters in evolutionary map L-systems-based topology optimization
Map L-systems-based parametrization, also referred to as the cellular division method, is a generative encoding, suitable for topology optimization. The parametrization is compact due to its ability to reuse its elements, and therefore capable of covering a large design space with relatively few design variables. Map L-systems are often evolved using genetic algorithms (GAs). A key implementation detail of such procedures, as with most GA-based geometry searches, is the choice of parameters controlling the operation of the evolutionary process. The optimal choice of these in conventional optimization formulations is highly problem-specific -- far less so, however, when the GA evolves an L-systems encoding and does not act directly on the geometry. This is because the L-system encoding is, itself, independent of the geometry. We study the effects of different control parameters by conducting a statistical test of over 400 parameter combinations on five test cases, for which the global optima are known. The best-performing parameter combinations are reported as a Pareto front of the average number of objective function evaluations and ranking based on the average of optimized fitnesses. Finally, three Pareto-optimal parameter combinations are selected and applied to an optimization problem of maximizing the fundamental natural frequency of an integrally stiffened aluminum panel. The best of the resulting designs has a higher fundamental natural frequency than the baseline design by a margin of 11.2%.
Topology Optimization, Generative Encoding, L-Systems, Genetic Algorithm, Control Parameters
1-17
Ikonen, Teemu, Johannes
d94a7607-5b27-4f95-8030-07e89f232231
Sobester, Andras
096857b0-cad6-45ae-9ae6-e66b8cc5d81b
2018
Ikonen, Teemu, Johannes
d94a7607-5b27-4f95-8030-07e89f232231
Sobester, Andras
096857b0-cad6-45ae-9ae6-e66b8cc5d81b
Ikonen, Teemu, Johannes and Sobester, Andras
(2018)
Statistical analysis of control parameters in evolutionary map L-systems-based topology optimization.
Structural and Multidisciplinary Optimization, .
(doi:10.1007/s00158-018-1943-1).
Abstract
Map L-systems-based parametrization, also referred to as the cellular division method, is a generative encoding, suitable for topology optimization. The parametrization is compact due to its ability to reuse its elements, and therefore capable of covering a large design space with relatively few design variables. Map L-systems are often evolved using genetic algorithms (GAs). A key implementation detail of such procedures, as with most GA-based geometry searches, is the choice of parameters controlling the operation of the evolutionary process. The optimal choice of these in conventional optimization formulations is highly problem-specific -- far less so, however, when the GA evolves an L-systems encoding and does not act directly on the geometry. This is because the L-system encoding is, itself, independent of the geometry. We study the effects of different control parameters by conducting a statistical test of over 400 parameter combinations on five test cases, for which the global optima are known. The best-performing parameter combinations are reported as a Pareto front of the average number of objective function evaluations and ranking based on the average of optimized fitnesses. Finally, three Pareto-optimal parameter combinations are selected and applied to an optimization problem of maximizing the fundamental natural frequency of an integrally stiffened aluminum panel. The best of the resulting designs has a higher fundamental natural frequency than the baseline design by a margin of 11.2%.
Text
manuscript_Ikonen_Sobester_preprint
- Accepted Manuscript
Text
publishedPaper_Ikonen_Sobester
- Version of Record
More information
Accepted/In Press date: 8 February 2018
e-pub ahead of print date: 19 March 2018
Published date: 2018
Keywords:
Topology Optimization, Generative Encoding, L-Systems, Genetic Algorithm, Control Parameters
Identifiers
Local EPrints ID: 418325
URI: http://eprints.soton.ac.uk/id/eprint/418325
ISSN: 1615-147X
PURE UUID: dcec9267-8579-47de-8c3f-8d562fbf9dc8
Catalogue record
Date deposited: 28 Feb 2018 17:30
Last modified: 16 Mar 2024 06:16
Export record
Altmetrics
Contributors
Author:
Teemu, Johannes Ikonen
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics