AUTOMATIC CLUSTERING OF INDUSTRIAL DATA WITH THE CONNECTED COMPONENTS
Résumé
This paper proposes the use of techniques from the field of graph theory to automatically determine and cluster the same types of observations without prior knowledge on their number. The idea is to create a graph where observations are vertices and an edge between two observations exists if and only if one observation is in the nearest neighbourhood of another one. Connected components of the graphs can be then considered as groups of observations of the same type. The approach is particularly useful for multiclass anomaly detection in the context of Industry 4.0.
Domaines
Statistiques [stat]Origine | Fichiers produits par l'(les) auteur(s) |
---|---|
Licence |