This paper proposes a delta connected IM model that takes the Stator winding Inter-Turn Short Circuit (SITSC) fault into account. In order to detect the fault and evaluate its severity, an observer based FDI method is suggested. It allows the generation of residual using extended Kalman filter (EKF). To overcome the problem of the EKF initialization, the cyclic optimization method is applied to determine its tuning parameters. The advantage of the proposed approach is the real-time quantification of the fault severity and the quick fault detection. Using numerical simulation under both the healthy and the faulty conditions, the proposed IM model and EKF-based FDI approach are confirmed. Experimental results obtained by a real-time implementation on test-bench validate the simulated results.

This paper presents the application of a Metaheuristic optimization algorithm for determining the parameters of a PI controller and the values of the state and measurement noise of Kalman Filter. The particle swarm optimization is a new technique that is used to solve complex problems. It minimizes a cost function under the cooperation of many individuals. Kalman Filter is used here to estimate the stator currents and rotor fluxes of the induction motor. The performances of the extended Kalman Filter and the adaptive Kalman Filter are analyzed. They are applied to estimate stator currents; rotor fluxes and rotor speed of the induction motor, and thus help to overcome the speed sensor, which is expensive and bulky. The extended Kalman Filter requires extending the state vector to rotor speed, which implies to use the linearization of the model. The adaptive Kalman Filter consists of determining the rotor speed adaptation law. The stability of the estimation error is proved using a Lyapunov function.

This article presents two variants of the extended Kalman filter (EKF), applied to the fault detection and isolation (FDI) of the inter-turn short circuits in stator windings of the induction motor (IM). Besides the state variables, these two variants allow estimating some parameters, named fault factors. They are three fault factors, i.e., one by phase, which are to be estimated. The first EKF variant estimates the state vector which is extended, simultaneously, to all the three fault factors. The second one uses three EKF, each one of them is extended to only one fault factor, to estimate them. This three EKF form a multi-model and they are switched sequentially, between extended models, with the same dwell time.

No abstract