Fault detection and isolation with robust principal component analysis

Yvon Tharrault (1) , Gilles Mourot (1) , José Ragot (1) , Didier Maquin (1)

1 CRAN - Centre de Recherche en Automatique de Nancy

Principal component analysis (PCA) is a powerful fault detection technique which has been widely used in process industries. However, a main drawback of PCA is that it is based on least squares estimation techniques and hence fails to account for outliers which are common in physical processes. This paper is concerned with the fault detection and isolation problem. The proposed method does not require a data matrix without outliers for a PCA model design. Indeed, the approach directly uses the eventually corrupt database to elaborate a robust PCA model allowing fault detection. Then reconstruction principle and fault signatures analysis are used for fault isolation.

Mots clés: Principal component analysis; robustness; fault signature; fault detection and isolation; outliers

Domaines: Automatique / Robotique

Yvon Tharrault, Gilles Mourot, José Ragot, Didier Maquin. Fault detection and isolation with robust principal component analysis. 8th Conference on Diagnostics of Processes and Systems, DPS'07, Sep 2007, Slubice, Poland. pp.CDROM. ⟨hal-00155273⟩