Optimizing parameterized CAD geometries using sensitivities based on adjoint functions

Trevor Robinson, Cecil Armstrong, Hung Chua, C. Othmer, T. Grahs

Research output: Contribution to journalArticle

28 Citations (Scopus)

Abstract

An approach is presented for determining which parameters defining the features in a CAD model need to be modified, and by what amount, to optimize component performance. It uses sensitivities computed for the parameters to determine the change required in each to optimize the component. Parametric sensitivity is computed by combining a measure of boundary movement due to a parameter perturbation, known as design velocity, and an adjoint sensitivity map over the boundary. The sensitivity map results from an adjoint analysis and approximates the change in objective function (performance) due to a movement of the boundary. Gradient based optimization is used based on the parametric sensitivities.

This presented method is significantly less computationally expensive than alternative approaches, and has the advantage that optimization is based on the parameters defining the CAD model, allowing it to be integrated into design workflows. The efficiency of the approach allows all of the parameters in the CAD model to be used as optimization variables, potentially offering better optimization. The work is immune to many of the issues hampering existing approaches.
LanguageEnglish
Pages253-268
JournalComputer-Aided Design and Applications
Volume9
Issue number3
DOIs
Publication statusPublished - 09 Aug 2013

Fingerprint

Computer aided design
Geometry
Optimization
Optimise
Parameter Perturbation
Work Flow
Objective function
Model
Gradient
Alternatives
Design
Movement

Cite this

@article{ad8766f0997241ffb74ae46177561f09,
title = "Optimizing parameterized CAD geometries using sensitivities based on adjoint functions",
abstract = "An approach is presented for determining which parameters defining the features in a CAD model need to be modified, and by what amount, to optimize component performance. It uses sensitivities computed for the parameters to determine the change required in each to optimize the component. Parametric sensitivity is computed by combining a measure of boundary movement due to a parameter perturbation, known as design velocity, and an adjoint sensitivity map over the boundary. The sensitivity map results from an adjoint analysis and approximates the change in objective function (performance) due to a movement of the boundary. Gradient based optimization is used based on the parametric sensitivities.This presented method is significantly less computationally expensive than alternative approaches, and has the advantage that optimization is based on the parameters defining the CAD model, allowing it to be integrated into design workflows. The efficiency of the approach allows all of the parameters in the CAD model to be used as optimization variables, potentially offering better optimization. The work is immune to many of the issues hampering existing approaches.",
author = "Trevor Robinson and Cecil Armstrong and Hung Chua and C. Othmer and T. Grahs",
year = "2013",
month = "8",
day = "9",
doi = "10.3722/cadaps.2012.253-268",
language = "English",
volume = "9",
pages = "253--268",
journal = "Computer-Aided Design and Applications",
issn = "1686-4360",
publisher = "CAD Solutions",
number = "3",

}

Optimizing parameterized CAD geometries using sensitivities based on adjoint functions. / Robinson, Trevor; Armstrong, Cecil; Chua, Hung; Othmer, C.; Grahs, T.

In: Computer-Aided Design and Applications, Vol. 9, No. 3, 09.08.2013, p. 253-268.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Optimizing parameterized CAD geometries using sensitivities based on adjoint functions

AU - Robinson, Trevor

AU - Armstrong, Cecil

AU - Chua, Hung

AU - Othmer, C.

AU - Grahs, T.

PY - 2013/8/9

Y1 - 2013/8/9

N2 - An approach is presented for determining which parameters defining the features in a CAD model need to be modified, and by what amount, to optimize component performance. It uses sensitivities computed for the parameters to determine the change required in each to optimize the component. Parametric sensitivity is computed by combining a measure of boundary movement due to a parameter perturbation, known as design velocity, and an adjoint sensitivity map over the boundary. The sensitivity map results from an adjoint analysis and approximates the change in objective function (performance) due to a movement of the boundary. Gradient based optimization is used based on the parametric sensitivities.This presented method is significantly less computationally expensive than alternative approaches, and has the advantage that optimization is based on the parameters defining the CAD model, allowing it to be integrated into design workflows. The efficiency of the approach allows all of the parameters in the CAD model to be used as optimization variables, potentially offering better optimization. The work is immune to many of the issues hampering existing approaches.

AB - An approach is presented for determining which parameters defining the features in a CAD model need to be modified, and by what amount, to optimize component performance. It uses sensitivities computed for the parameters to determine the change required in each to optimize the component. Parametric sensitivity is computed by combining a measure of boundary movement due to a parameter perturbation, known as design velocity, and an adjoint sensitivity map over the boundary. The sensitivity map results from an adjoint analysis and approximates the change in objective function (performance) due to a movement of the boundary. Gradient based optimization is used based on the parametric sensitivities.This presented method is significantly less computationally expensive than alternative approaches, and has the advantage that optimization is based on the parameters defining the CAD model, allowing it to be integrated into design workflows. The efficiency of the approach allows all of the parameters in the CAD model to be used as optimization variables, potentially offering better optimization. The work is immune to many of the issues hampering existing approaches.

U2 - 10.3722/cadaps.2012.253-268

DO - 10.3722/cadaps.2012.253-268

M3 - Article

VL - 9

SP - 253

EP - 268

JO - Computer-Aided Design and Applications

T2 - Computer-Aided Design and Applications

JF - Computer-Aided Design and Applications

SN - 1686-4360

IS - 3

ER -