Skip to Main content Skip to Navigation

Multi-fidelity Robust Design Optimization Methods for Organic Rankine Cycles

Abstract : Robust design optimization (RDO) is an important tool for the design of industrial products underuncertainty. It combines optimization algorithms and uncertainty quantification (UQ) techniques. Quantificationof uncertainties is generally too expensive for complex numerical models of engineering systems. With the aimof developing efficient RDO strategies designed for industrial applications, the coupling of parsimonious UQtechniques with a multi-objective genetic algorithm based on surrogate models (SMOGA) was studied. In thisregard, a promising RDO technique was used, based on the coupling of two nested surrogate models: the firstis used for UQ, while the response surface of the second is used to accelerate optimization; an infill criterion isused to update the surrogate model during optimizer convergence. Several UQ methods using information onthe gradients of the solution with respect to the uncertain variables were implemented and compared in termsof precision and computational cost. We then selected a so-called “low fidelity” UQ method, i.e. inexpensivebut not very accurate, and a “high fidelity” method in order to build a multi-fidelity surrogate model for robustoptimization. This model allows to have an accuracy close to the high fidelity model for a much lower computationcost. The methods under investigation were applied to the RDO of organic Rankine cycles (ORC) and to theshape optimization of an ORC turbine blade grid, with very promising results.
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Friday, March 12, 2021 - 5:29:19 PM
Last modification on : Friday, August 5, 2022 - 2:54:00 PM
Long-term archiving on: : Sunday, June 13, 2021 - 7:17:57 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03168172, version 1


Aldo Serafino. Multi-fidelity Robust Design Optimization Methods for Organic Rankine Cycles. Fluid mechanics [physics.class-ph]. HESAM Université, 2020. English. ⟨NNT : 2020HESAE055⟩. ⟨tel-03168172⟩



Record views


Files downloads