Skip to Main content Skip to Navigation
Theses

Diagnostic à base de modèles incertains utilisant l'analyse par intervalles: l'approche bornante

Abstract : Fault diagnosis methods generally use a model, which represents the no-fauIty behaviour of the supervised physical system. The fundamental problem then encountered cornes from the inaccuracy associated with this model. To avoid confusing a fault with a modelling error, a new strategy taking uncertainties into account is defined. These ones can vary, affect any model parameter and are represented by bounded variables. The method suggested, called bounding approach, is based on interval analysis. After having noticed the drawbacks of a stochastic model, we justify that the bounding approach satisfies the objectives of diagnosis, by conserving more information than traditional de-coupling methods, by generating in a natural way the thresholds and by allowing the treatment of multiplicative and time-variant uncertainties. Then, while focusing on the applications of interval analysis, the advanta.ges and drawbacks of this tool are clarified. The resolution of the problem of dependence is detailed, in particular in the case of a vector field. A formalism allowing to deal with domains, which are more complex than simple boxes, is then proposed. So as to avoid the wrapping effect, a Parity Space approach is proposed. The objective being to build residuals de-coupled from the unknown state variables, the interest to search an uncertain projection matrix rather than a certain one is then shown. The principles of the consistency tests, allowing the analysis of these residuals, are then detailed. Lastly, some algorithms ensuring the estimation of the characteristics of uncertainties, are developed. This thesis is complete by the implementation of the various developed methods on an example.
Document type :
Theses
File URL :
http://docnum.univ-lorraine.fr/prive/INPL_T_2000_ADROT_O.pdf
Complete list of metadata

https://hal.univ-lorraine.fr/tel-01750473
Contributor : Thèses Ul Connect in order to contact the contributor
Submitted on : Thursday, March 29, 2018 - 12:47:00 PM
Last modification on : Saturday, October 16, 2021 - 11:14:14 AM

Identifiers

  • HAL Id : tel-01750473, version 1

Collections

Citation

Olivier Adrot. Diagnostic à base de modèles incertains utilisant l'analyse par intervalles: l'approche bornante. Autre. Institut National Polytechnique de Lorraine, 2000. Français. ⟨NNT : 2000INPL121N⟩. ⟨tel-01750473⟩

Share

Metrics

Record views

92