New Approach of Parametric Spectral Analysis Used in the Diagnosis of Induction Motors K. Mansouri, M. Benouareth, N. Doghmane * and M. Faouzi Karkat. Electronic Department, University of Annaba Algeria ALGERIA Abstract: Today, in response to the industrial requirements, the diagnosis and monitoring of the electric motors, turning at constant and/or variable speed, are much requested. Several works show that the faults were often studied by the analysis of the supply current. Many schemes have been proposed, all based on the Fourier analysis. However, this analysis is badly adapted to these applications. Indeed, the signals used are strongly no stationary. New tools are thus necessary. In this article, a technique for the faults detection in the induction motors is proposed. It’s based on spectral estimate using the least squares approach. The stator current is a harmonic signal. It is thus necessary to know the various amplitudes, frequencies and phases of its spectral components. Key words: diagnosis, induction motor, spectral analysis, faults detection, least squares. 1. Introduction In this paper, a new approach of parametric spectral analysis is proposed. This technique is used for the faults detection in the induction motors. The fault detection in these motors is based on the analysis of the stator currents. The analysis of these current uses the method of least squares. This, leads to the extraction of parameters vector characterizing the behavior of the machine. The purpose of this spectral analysis is to identify the various harmonics present in the stator currents. The identification is not limited to the only frequencies of these harmonics but also their respective amplitudes and their phases. To determine the amplitudes of various harmonics leads to solve a system of linear equations. But for the frequencies and especially the phases the equations to solve will be nonlinear. The goal of the approach proposed is the fault detection of a bar broken of motor rotor. Indeed, one or more broken bars of the rotor have a direct influence on the evolution of the stator currents. 2. Problem position Digital signal processing involves techniques that improve our understanding of information contained in recorded stotoric currents. Normally, these signals are measured and observed in the time domain. When the harmonics contained in these signals are of interest, it’s necessary to study it in the frequency domain (spectral analysis). The spectral analysis of the statoric currents can be used as a processing technique to permit, then, fault detection, as example: bar Broken in rotors. Generally, the fault detection in the induction motors is done directly in the temporal field (observation of the signal evolution) or by a classic spectral analysis based on the FFT algorithm. However, the no stationary character of this type of signals and also the possible presence of very close frequencies make unusable the techniques of spectral analysis based on the FFT. The purpose of this study is to propose a new approach of the spectral analysis able to estimate the frequencies, the amplitudes and the phases of all harmonics present in the statoric currents. The statoric currents are supposed to be formed by a sum of sinusoids or generally sum exponential complexes. Certain frequencies can be revealing presence of one or more faults. To detect and then isolate these frequencies can lead to a fine and precise diagnosis. 3. Problem solution The signal to be analyzed can take the following complex form: ∑ = ⎟ ⎠ ⎞ ⎜ ⎝ ⎛ = M k n Fe f 2jπ k k e a n s 1 ) ( (1) Where {A k } indicate a succession of M complex amplitudes and {f k } a succession of M different closed frequencies [1]. The signal is sampled at Te=1/Fe M is supposed known a priori. The goal is to estimate the values of A k , f k and the variance of additive noise. This is carried out starting from the observation: x(t), recorded on N samples ( N>M) x(t)=s(n)+b(n) (2) This measurement noise is supposed: Gaussian, centred and white. The adopted approach must to estimate, at the same time, the values of and , which k A ˆ k f ˆ Proceedings of the 5th WSEAS Int. Conf. on SIGNAL, SPEECH and IMAGE PROCESSING, Corfu, Greece, August 17-19, 2005 (pp118-122)