Model based fault diagnosis aims at detecting and isolating at an early stage the degradations in the operation of a technical system. It relies on the computation of fault indicators, also called residuals. The latter characterize the discrepancy between the data measured on-line on the supervised system and a mathematical model of this system describing its normal behavior or certain faulty behaviors. These indicators are processed in a so-called decision system in order to determine whether some of the residuals are significantly different from zero, and to deduce the most likely fault(s) from the pattern of zero and non-zero residuals.