Application of evolutionary algorithms in Bayesian multi-objective reliability-based design optimization

Tools

Celorrio, Luis and Patelli, Edoardo; (2019) Application of evolutionary algorithms in Bayesian multi-objective reliability-based design optimization. In: Proceedings of 17th International Probabilistic Workshop (IPW). UNSPECIFIED, GBR.

[thumbnail of Celorrio-and-Patelli-IPW2019-Application-of-evolutionary-algorithms-in-Bayesian-multi-objective-reliability-based-design-optimization]
Preview
 Text. Filename: Celorrio-and-Patelli-IPW2019-Application-of-evolutionary-algorithms-in-Bayesian-multi-objective-reliability-based-design-optimization.pdf
Accepted Author Manuscript
License: Strathprints license 1.0
Download (1MB)| Preview

Abstract

Reliability-Based Design Optimization (RBDO) methods are well known in engineering design. However, these approaches usually require uncertainties to be modelled by statistical distributions. Hence, samples of uncertainty variables with enough size are necessary, so that these variables can be fitted by probabilistic distributions known as aleatory uncertainty. In realistic engineering design, there is a lack of information about design variables or parameters and only a reduced set of samples are available (e.g. physical tests are very expensive). Therefore, design is carried out with incomplete information about input variables and parameters known as epistemic uncertainty. Both types of uncertainties need to be considered in engineering design problems. However, epistemic uncertainty cause the reliability of the system or component to be also a random variable. Therefore, reliability constraints are imposed with a level of confidence specified by the designer posing a significant computational challenge for the design. This paper proposes two effective multi-objective evolutionary algorithms to solve problems of design under uncertainty with incomplete information. The proposed approaches consider the cost and reliability as objective functions. The result is a Pareto front with a trade-off between cost and reliability for different levels of confidence. An analytical example and a structural problem are solved to show the applicability of the approach and how epistemic uncertainty may affect the results.

ORCID iDs

Celorrio, Luis and Patelli, Edoardo ORCID logoORCID: https://orcid.org/0000-0002-5007-7247;
CORE (COnnecting REpositories)