Optimization of composite panels using neural networks and genetic algorithms

W. Ruijter, R. Spallino, Laurent Warnet, Andries de Boer

Research output: Contribution to conferencePaperAcademic

5 Citations (Scopus)
90 Downloads (Pure)

Abstract

The objective of this paper is to present first results of a running study on optimization of aircraft components (composite panels of a typical vertical tail plane) by using Genetic Algorithms (GA) and Neural Networks (NN). The panels considered are standardized to some extent but still there is a wide scope of discrete and continuous design variables that can be adjusted to increase performance or reduce structural weight. A NN is trained for every panel configuration using a backpropagation algorithm with data sets taken from finite element analyses spread randomly over the design space. The trained network is then used to predict the values of the constraint functions (strain and buckling multipliers). The approach is formulated in this manner to maintain maximum flexibility regarding the implementation of new variables or models and with the prospect of optimizing the assembly as a whole. Results show that in design problems with high dimensionality the approach becomes more attractive, especially when the optimization has to be run repeatedly for panels under different loading/sizing conditions. The optimization algorithm has proven to be robust though dependent on the smoothness of the network output function. A modified method that feeds back the found optima is proposed to improve accuracy of the NN and decrease preparation time.
Original languageUndefined
Pages2359-2363
DOIs
Publication statusPublished - 2003

Keywords

  • IR-58847
  • Neural Networks
  • Panel buckling
  • Optimization
  • Genetic Algorithms
  • Composites

Cite this

@conference{8e5dff26c4b040049bafbc3e28ad15f4,
title = "Optimization of composite panels using neural networks and genetic algorithms",
abstract = "The objective of this paper is to present first results of a running study on optimization of aircraft components (composite panels of a typical vertical tail plane) by using Genetic Algorithms (GA) and Neural Networks (NN). The panels considered are standardized to some extent but still there is a wide scope of discrete and continuous design variables that can be adjusted to increase performance or reduce structural weight. A NN is trained for every panel configuration using a backpropagation algorithm with data sets taken from finite element analyses spread randomly over the design space. The trained network is then used to predict the values of the constraint functions (strain and buckling multipliers). The approach is formulated in this manner to maintain maximum flexibility regarding the implementation of new variables or models and with the prospect of optimizing the assembly as a whole. Results show that in design problems with high dimensionality the approach becomes more attractive, especially when the optimization has to be run repeatedly for panels under different loading/sizing conditions. The optimization algorithm has proven to be robust though dependent on the smoothness of the network output function. A modified method that feeds back the found optima is proposed to improve accuracy of the NN and decrease preparation time.",
keywords = "IR-58847, Neural Networks, Panel buckling, Optimization, Genetic Algorithms, Composites",
author = "W. Ruijter and R. Spallino and Laurent Warnet and {de Boer}, Andries",
year = "2003",
doi = "10.1016/B978-008044046-0/50580-7",
language = "Undefined",
pages = "2359--2363",

}

Optimization of composite panels using neural networks and genetic algorithms. / Ruijter, W.; Spallino, R.; Warnet, Laurent; de Boer, Andries.

2003. 2359-2363.

Research output: Contribution to conferencePaperAcademic

TY - CONF

T1 - Optimization of composite panels using neural networks and genetic algorithms

AU - Ruijter, W.

AU - Spallino, R.

AU - Warnet, Laurent

AU - de Boer, Andries

PY - 2003

Y1 - 2003

N2 - The objective of this paper is to present first results of a running study on optimization of aircraft components (composite panels of a typical vertical tail plane) by using Genetic Algorithms (GA) and Neural Networks (NN). The panels considered are standardized to some extent but still there is a wide scope of discrete and continuous design variables that can be adjusted to increase performance or reduce structural weight. A NN is trained for every panel configuration using a backpropagation algorithm with data sets taken from finite element analyses spread randomly over the design space. The trained network is then used to predict the values of the constraint functions (strain and buckling multipliers). The approach is formulated in this manner to maintain maximum flexibility regarding the implementation of new variables or models and with the prospect of optimizing the assembly as a whole. Results show that in design problems with high dimensionality the approach becomes more attractive, especially when the optimization has to be run repeatedly for panels under different loading/sizing conditions. The optimization algorithm has proven to be robust though dependent on the smoothness of the network output function. A modified method that feeds back the found optima is proposed to improve accuracy of the NN and decrease preparation time.

AB - The objective of this paper is to present first results of a running study on optimization of aircraft components (composite panels of a typical vertical tail plane) by using Genetic Algorithms (GA) and Neural Networks (NN). The panels considered are standardized to some extent but still there is a wide scope of discrete and continuous design variables that can be adjusted to increase performance or reduce structural weight. A NN is trained for every panel configuration using a backpropagation algorithm with data sets taken from finite element analyses spread randomly over the design space. The trained network is then used to predict the values of the constraint functions (strain and buckling multipliers). The approach is formulated in this manner to maintain maximum flexibility regarding the implementation of new variables or models and with the prospect of optimizing the assembly as a whole. Results show that in design problems with high dimensionality the approach becomes more attractive, especially when the optimization has to be run repeatedly for panels under different loading/sizing conditions. The optimization algorithm has proven to be robust though dependent on the smoothness of the network output function. A modified method that feeds back the found optima is proposed to improve accuracy of the NN and decrease preparation time.

KW - IR-58847

KW - Neural Networks

KW - Panel buckling

KW - Optimization

KW - Genetic Algorithms

KW - Composites

U2 - 10.1016/B978-008044046-0/50580-7

DO - 10.1016/B978-008044046-0/50580-7

M3 - Paper

SP - 2359

EP - 2363

ER -