The authors propose a new diagnosis method for on-line broken bars detection by parameters estimation. For predictive detection, Kalman filtering algorithm has been adapted to take into account the on-line parameters deviations in faulty case. Within the framework of the diagnosis of the rotor defects, it is difficult to conduct experimental tests to validate the on-line identification of such default. For this reason, one propose an on-line technique to detection rotor broken bars. This technique was validated by simulation tests using an induction machine simulator. The goal of this identification is to detect a possible variation in the resistance of the bars, signature of a rotor defect. Estimation results show a good agreement and demonstrate the possibility of on-line motor bar's break detection.