Rotor asymmetries in induction machines perturb many components such as, flux patterns and electromagnetic torque. The supervision of these signals enables early detection of such faults and help to machine diagnostic. This paper studies the detection of rotor imperfection by spectral analysis of the electromagnetic torque computed by two rotor flux estimators. In the first approach, the Kalman filter is used assuming to be known the mechanical velocity. The second approach uses Extended Kalman Filter (EKF) for speed estimation. Experimental results show the great capability of these methods to detect this type of faults.