Radial output space mapping for electromechanical systems design - Université Pierre et Marie Curie Accéder directement au contenu
Article Dans Une Revue COMPEL: The International Journal for Computation and Mathematics in Electrical and Electronic Engineering Année : 2014

Radial output space mapping for electromechanical systems design

Résumé

Purpose – The purpose of this paper is to set a relation through adaptive multi-level optimization between two physical models with different accuracies; a fast coarse model and a fine time consuming model. The use case is the optimization of a permanent magnet axial flux electrical machine. Design/methodology/approach – The paper opted to set the relation between the two models through radial basis function (RBF). The optimization is held on the coarse model. The deduced solutions are used to evaluate the fine model. Thus, through an iterative process a residue RBF between models responses is built to endorse an adaptive correction. Findings – The paper shows how the use of a residue function permits, to diminish optimization time, to reduce the misalignment between the two models in a structured strategy and to find optimum solution of the fine model based on the optimization of the coarse one. The paper also provides comparison between the proposed methodology and the traditional approach (output space mapping (OSM)) and shows that in case of large misalignment between models the OSM fails. Originality/value – This paper proposes an original methodology in electromechanical design based on building a surrogate model by means of RBF on the bulk of existing physical model.
Fichier non déposé

Dates et versions

hal-01099602 , version 1 (04-01-2015)

Identifiants

Citer

Maya Hage Hassan, Ghislain Remy, Guillaume Krebs, Claude Marchand. Radial output space mapping for electromechanical systems design. COMPEL: The International Journal for Computation and Mathematics in Electrical and Electronic Engineering, 2014, 33 (3), pp.965 - 975. ⟨10.1108/COMPEL-05-2013-0192⟩. ⟨hal-01099602⟩
129 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More