Stochastic models for the evaluation of magnetisation faults

• Peter Offermann ^[1] ; Kay Hameyer ^[1]
  1. [1] Rheinisch-Westfälische Technische Hochschule Aachen University

     Rheinisch-Westfälische Technische Hochschule Aachen University

     Städteregion Aachen, Alemania

     
• Localización: Compel: International journal for computation and mathematics in electrical and electronic engineering, ISSN 0332-1649, Vol. 33, Nº 1-2 (Special Issue: ICEF 2012), 2014, págs. 245-253
• Idioma: inglés
• Enlaces
  • Texto completo
• Resumen

  • Purpose – The introduction of stochastic deviations due to production faults into the finite element (FE) simulation of electrical machines requires suitable error-model. These models should describe the occurring deviations from the ideal case. Permanent magnets, which can be used as rotor excitations in synchronous machines (PMSM), are one out of many possible sources for the aforementioned stochastic production variations. Fitting measured magnet variations to simulation models with the aim of describing the occurring production deviations, however, poses a problem due to two reasons: to begin with, only data of measured flux-densities are available. Second, a solution of the inverse problem is required to obtain data about changes inside the magnet. This paper, therefore, presents two solutions to this problem.

    Design/methodology/approach – Two error-models, one based on knowledge about the magnetisation process, the other one built upon principal component analysis, are presented. Both models are evaluated by parametrising them, using a set of measured flux-density data from magnets. Afterwards, each model's applicability and reproduction quality is assessed.

    Findings – Both models still have some drawbacks. While the first model seems to be too coarse grained for certain variations, the second model lacks applicability for a high reproduction quality.

    Originality/value – The comparison of both methods reveals guidelines, which methodology should be applied for predicting which variations. Furthermore, solutions are shown, how to mitigate the problems of the two presented models.