Fault detection and isolation from an identified MIMO Takagi-Sugeno model of a bioreactor
Résumé
This work aims at evaluating the use of TS MIMO model for the Detection and Isolation of faults of sensors on a nonlinear process. TS model involves a set of local linear models whose contribution on the output signals is expressed by a weighting function. In a previous work, an identification scheme was proposed for TS MIMO model. It doesn't need for the outputs of the process to be separable, as it is usually the case for the identification methods based on MISO model. A model-based method, for faults detection is proposed from the identified model. This new approach differs from the observers and parity space based approaches, as it only uses a partial knowledge of the model. Outputs are projected into a sub-space that is orthogonal to the sequences of input-output weighted signals to provide a residuals set. The process of fault isolation exploits the structure of the residuals set in order to recognize the acting fault. The sensitivity of the residuals to the occurrence of faults is proved while the incidence matrix of the residuals is deduced from simulation. It shows the ability of the generated set of residuals to isolate the fault of sensors. © 2009 IFAC.
Mots clés
Fault detection and isolation
Fault isolation
Local linear models
Model-based method
Nonlinear process
Output signal
Parity spaces
Partial knowledge
Sub-spaces
T S models
Takagi-Sugeno models
Weighting functions
Faults detection
Identification method
Identification scheme
Incidence matrices
Input-output