In this thesis we propose a new formulation of the procedure for fault detection and isolation from the actuators for dynamic systems. Usually, a control system consists of three parts: the sensors, the industrial process and the actuators. The actuator is a device with its own internal structure and its dynamic characteristics. In most studies, it is considered simply and in an approximate manner as constant coefficients, but this type of actuator formulation is not effective for the accurate identification of faults, from actuators for real systems. In order to approach the reality and also to improve the efficiency of diagnosis, we will consider the actuator as a subsystem connected in cascade with the process. Using this new formulation, actuators faults can be modeled by considering them as a change in the internal parameters of the actuator subsystem. We assume that the output of the actuator which is also the entrance of the dynamic system is not measurable, so we'll get the information on the faults actuators only through the output variable of the process. An interesting question may arise during the implementation of the diagnostic procedure for the faults actuators: is that information on faults actuators can be fully seen in the output of the dynamic system? This informational intake is the primary problem to design the desired diagnostic procedure. To describe this new formulation, we introduce some notions on inverse systems and we have also given a rank condition, for invertibility of left linear systems. The purpose of the study of these notions is to find satisfactory conditions that will allow the detection of faults from the actuator. For fault isolation, we used an approach based on interval observers. Indeed, this method allowed us to isolate faults in a very quick manner. The results presented were illustrated by an example of a nuclear reactor simulation showing the effectiveness of the method for fault isolation.
