The purpose of this thesis is to provide solutions for the fault détection and isolation in electrical drives and especialîy induction drives. The considered faults may affect the machine, the converter, the mechanical part and the sensors. The variable speed drive is controlled by a rotor flux vector control loop and a speed loop. Using the results of the fault characterisation, obtained from simulations or experiments, two approaches are proposed. The first one is based on signal processing tools. The variable speed problem is overcome : measures are sampled synchronously with the mechanical position or the rotating field position. The second approach is based on analytical redundancy tools for fault détection and localisation purpose. Extended state observers (extended Kalman filter, model référence adaptive system) are studied and tested experimentally. In order to carry out the fault isolation, unknown input observers are proposed. Approaches are introduced to constitute a global monitoring System combining the developed tools. Some are based on logicat combinations and others on neural networks.
