0093-9994 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2016.2521829, IEEE Transactions on Industry Applications The Use of the Modified Prony’s Method for Rotor Speed Estimation in Squirrel Cage Induction Motors M. Sahraoui, Member, IEEE, A. J. M. Cardoso, Senior Member, IEEE, K. Yahia, Member, IEEE, and A. Ghoggal Abstract— It is well known that rotor speed estimation assumes a paramount importance for the correct diagnosis of bearings, air-gap eccentricities or rotor bar defects. In this paper, a new technique for rotor speed estimation using a modified Prony’s method is proposed. The algorithm developed for this purpose, is based on tracking the frequencies of the rotor slot harmonics (RSH) which exist in stator currents of most squirrel cage induction motors. High order RSH are used in order to avoid the possible effect of harmonics stemming from other sources. The proposed modified Prony’s method shows a great ability for tracking RSH frequencies and then the rotation speed. In addition, this technique can deal with noisy and non stationary signals and it requires only few data samples, which reduce considerably the computational time and data storage requirements. Consequently, the proposed algorithm is suitable for online implementation. The method’s effectiveness is verified by simulation and experimental tests. Index Terms--Prony’s method; Speed estimation; MCSA; Frequency resolution, Advanced signal processing I. INTRODUCTION It is well known that squirrel cage induction motors are widely used in industry due to their reliability and relatively low cost. Until now, the detection of motor speed without sensors is still an open research topic. Indeed, the rotor speed estimation has a crucial importance for a large variety of electrical engineering applications, including machine condition monitoring and motor drive applications. There are, basically, four approaches for measuring the motor rotor speed. In the first approach, the speed information is obtained directly using speed transducers, such as resolvers, tachogenerators, digital shaft position encoders, etc. However, this approach has many drawbacks. Indeed, mounting a speed sensor can be considered as an additional cost factor that reduces considerably the ruggedness and the simplicity of induction motors. In addition, for low power machines, the cost of the sensor becomes comparable to the cost of the motor itself, which justifies the importance of sensorless speed estimation methods for low power applications. The second approach uses observers and adaptive schemes deduced from motor mathematical models to estimate the rotor speed. The techniques belonging to this approach are based on stator current and voltage measurements, such as model reference adaptive systems (MRAS) [1], Extended Kalman Filter (EKF) [2], adaptive flux observer [3] and sliding mode current observer [4], [5]. Despite the good performances obtained from these methods, they suffer from their dependency on the machine parameter variations. Consequently, the parameter estimation errors can affect the accuracy of rotor speed estimation, which represents their main drawback. The third approach for rotor speed estimation uses probing signals injected into the motor stator currents and/or voltages to detect the rotor flux and consequently the rotor speed [6]. However, injecting probing signals into the motor stator terminals has many drawbacks. For instance, it can induce high frequency torque pulses leading to speed ripple effects. In addition, the useful data may be distorted due to interference with high frequency probe signals. In [7], a method based on high-frequency injection, both at zero and low frequencies, was used for sensorless speed and position control of an induction motor. The algorithm of this method uses a set of synchronous filters to identify the disturbance waveforms and it allows the rejection of saturation and nonlinear inverter effects. The fourth approach for rotor speed estimation uses the signal processing to exploit the existing machine saliencies. The saliencies may be generated by some means of modification to the rotor structure or may occur due to magnetic saturation. As far as the squirrel cage induction motors are concerned, the saliencies exist due to the rotor slotting [8]. The harmonics stemming from rotor slotting (RSH) are usually used on the one hand for fault detection in induction motors [9]. On the other hand, some studies have used the RSH for speed estimation [10], [11] since their frequencies are inherently correlated with the motor speed. The effectiveness of speed measurement methods using rotor slot spectral components strongly depends on the kind of spectral analysis technique applied. The main attention should be focused on the method efficiency for non- stationary signals. In this context, several techniques have M. Sahraoui is with LGEB, University of Biskra, Algeria, and also with CISE - Electromechatronic Systems Research Centre, Covilhã, Portugal (s.m.sahraoui@ieee.org). A. J. M. Cardoso is with the Department of Electromechanical Engineering of the University of Beira Interior, Covilhã, Portugal, and also with CISE - Electromechatronic Systems Research Centre, Covilhã, Portugal (ajmcardoso@ieee.org). K. Yahia is with the Laboratoire de Génie Energétique et Matériaux, LGEM, University of Biskra, Biskra, Algeria and also with CISE – Electromechatronic Systems Research Centre, Covilhã, Portugal (e-mail: kdyahia@gmail.com). A. Ghoggal is with LGEB, University of Biskra, Algeria (ghoetudes@yahoo.fr).