Martin MBARGA NKOGO : "Time-based interpretation of acoustic emission data : importance of suitable validation criteria and relevance of Bayesian variational methods with and without a priori"

Abstract : The acoustic emission technique is a passive and non-destructive method for structural health monitoring and material characterization. Under stress, structures release energy in the form ofwaves transients called acoustic emissions, which carry the characteristics of their sources. These waves are represented in a common feature space. The number of sources and the relevant features being unknown, so for different subsets of features, a clustering is done by varying the number of clusters. Validation consists of determining the number of clusters and the subset of relevant features. This validation has long been made with criteria based on the shape of the clusters, without taking into account clusters chronology. In this thesis we have defined sequential validation criteria dedicated to acoustic emission. A clustering validation method based on cluster trigger histograms has also been proposed for chronological monitoring of events. Finally, a partially supervised learning model has been proposed to help the clustering with partial information. This model also makes it possible to determine the number of natural clusters automatically, without the intervention of a criterion. 

Jury Composition :

Emmanuel RAMASSO, Maître de conférences, Université Bourgogne Franche-Comté, PhD Director

Patrice LE MOAL, Chargé de recherche, CNRS, PhD Co-supervisor

Michèle ROMBAUT, Professeure des universités, Université Grenoble Alpes, Reviewer

Gilles BOURBON, Ingénieur de recherche, CNRS, PhD Co-supervisor

Valérie KAFTANDJIAN-DOUDET, Professeure des universités, INSA Lyon, Reporter

Frédéric PICHON, Professeur des universités, Université d'Artois, Reporter

Localization : Amphithéâtre du Bâtiment Transfert de l'ENSMM - 26, Rue de l'Epitaphe - Besançon