arrow
Volume 14, Issue 2
A Polynomial Chaos Expansion Trust Region Method for Robust Optimization

Samih Zein

Commun. Comput. Phys., 14 (2013), pp. 412-424.

Published online: 2014-08

Export citation
  • Abstract

Robust optimization is an approach for the design of a mechanical structure which takes into account the uncertainties of the design variables. It requires at each iteration the evaluation of some robust measures of the objective function and the constraints. In a previous work, the authors have proposed a method which efficiently generates a design of experiments with respect to the design variable uncertainties to compute the robust measures using the polynomial chaos expansion. This paper extends the proposed method to the case of the robust optimization. The generated design of experiments is used to build a surrogate model for the robust measures over a certain trust region. This leads to a trust region optimization method which only requires one evaluation of the design of experiments per iteration (single loop method). Unlike other single loop methods which are only based on a first order approximation of robust measure of the constraints and which does not handle a robust measure for the objective function, the proposed method can handle any approximation order and any choice for the robust measures. Some numerical experiments based on finite element functions are performed to show the efficiency of the method. 

  • Keywords

  • AMS Subject Headings

  • Copyright

COPYRIGHT: © Global Science Press

  • Email address
  • BibTex
  • RIS
  • TXT
@Article{CiCP-14-412, author = {}, title = {A Polynomial Chaos Expansion Trust Region Method for Robust Optimization}, journal = {Communications in Computational Physics}, year = {2014}, volume = {14}, number = {2}, pages = {412--424}, abstract = {

Robust optimization is an approach for the design of a mechanical structure which takes into account the uncertainties of the design variables. It requires at each iteration the evaluation of some robust measures of the objective function and the constraints. In a previous work, the authors have proposed a method which efficiently generates a design of experiments with respect to the design variable uncertainties to compute the robust measures using the polynomial chaos expansion. This paper extends the proposed method to the case of the robust optimization. The generated design of experiments is used to build a surrogate model for the robust measures over a certain trust region. This leads to a trust region optimization method which only requires one evaluation of the design of experiments per iteration (single loop method). Unlike other single loop methods which are only based on a first order approximation of robust measure of the constraints and which does not handle a robust measure for the objective function, the proposed method can handle any approximation order and any choice for the robust measures. Some numerical experiments based on finite element functions are performed to show the efficiency of the method. 

}, issn = {1991-7120}, doi = {https://doi.org/10.4208/cicp.260512.260912a}, url = {http://global-sci.org/intro/article_detail/cicp/7166.html} }
TY - JOUR T1 - A Polynomial Chaos Expansion Trust Region Method for Robust Optimization JO - Communications in Computational Physics VL - 2 SP - 412 EP - 424 PY - 2014 DA - 2014/08 SN - 14 DO - http://doi.org/10.4208/cicp.260512.260912a UR - https://global-sci.org/intro/article_detail/cicp/7166.html KW - AB -

Robust optimization is an approach for the design of a mechanical structure which takes into account the uncertainties of the design variables. It requires at each iteration the evaluation of some robust measures of the objective function and the constraints. In a previous work, the authors have proposed a method which efficiently generates a design of experiments with respect to the design variable uncertainties to compute the robust measures using the polynomial chaos expansion. This paper extends the proposed method to the case of the robust optimization. The generated design of experiments is used to build a surrogate model for the robust measures over a certain trust region. This leads to a trust region optimization method which only requires one evaluation of the design of experiments per iteration (single loop method). Unlike other single loop methods which are only based on a first order approximation of robust measure of the constraints and which does not handle a robust measure for the objective function, the proposed method can handle any approximation order and any choice for the robust measures. Some numerical experiments based on finite element functions are performed to show the efficiency of the method. 

Samih Zein. (2020). A Polynomial Chaos Expansion Trust Region Method for Robust Optimization. Communications in Computational Physics. 14 (2). 412-424. doi:10.4208/cicp.260512.260912a
Copy to clipboard
The citation has been copied to your clipboard