Experimental investigation and comparative study of signal processing techniques dedicated for the diagnosis of rotor failures in the induction motors

Purpose The purpose of this paper is to propose applications of advanced signal-processing techniques for the diagnosis and detection of rotor fault in an induction machine. Two techniques are used: spectral analysis techniques and time frequency techniques for the diagnosis of an electrical machine. One is based on the power spectral density estimation techniques, such as periodogram and Welch periodogram. The second method is based on Hilbert transform (HT) to extract the envelope for the stator current. Then, this signal is processed via discrete wavelet transform (DWT) for determining the faulty components in the spectrum of the stator current envelope and identifying the eigenvalues of energies (HDWT). Design/methodology/approach First, this paper focused on theoretical development and a comparative study of these signal-processing techniques, which are based on the periodogram, Welch periodogram, HT and the DWT to extract the envelope for the stator current; it is used to compute the energy stored in each decomposition level obtained by the stator current envelope (HDWT). Moreover, the Welch periodogram is applied to obtain the envelope spectrum. Findings The simulation obtained and the experimental validation results of the proposed methods through MATLAB environment show the effectiveness of the proposed approaches with a good accuracy by power spectral density estimation techniques (periodogram and Welch periodogram). Moreover, the faults are manifested through the appearance of new frequencies components, as well as the envelope for the stator current (HT and DWT). This approach is effective for non-stationary and stationary signal to extract useful information for the detection of broken bar fault. Originality/value The current paper proposes a new diagnosis method for the detection and characterization of broken rotor bars defects early; it is founded primarily on theoretical development, and the comparison is based on the power spectral density technique (periodogram and Welch periodogram) and the computation of the energy stored in each decomposition level (precisely the HT and DWT). Moreover, the Welch periodogram is applied to obtain the envelope spectrum. The main advantages of the proposed techniques increase their reliability and availability.

APPLICATION OF WEIGHTING WINDOWS FOR SPECTRAL DENSITY ESTIMATION

The article discusses the application of some of the most important time windows for spectral density estimation determined by the correlogram method (from correlation function) and the periodogram method (from direct fourier transform).

A new approach to nonparametric spectral density estimation

A novel approach to nonparametric spectral density estimation has been proposed. The approach is based on a new evaluation criterion called autocorrelation mean square error (AMSE) for power spectral density (PSD) estimates of available finite length data. Minimization of this criterion not only provides the optimum segmentation for existing PSDE approaches , but also provides a new optimum windowing within the segments that can be combined additionally to the existing methods of nonparametric PSDE. Furthermore, the problem of frequency resolution in existing PSDE methods for noisy signals has been analyzed. In the existing approaches, the additive noise and the finiteness of data which are the causes of the original loss of the frequency resolution are not treated separately. The suggested new approach to spectrum estimation takes advantage of these two different causes of the problem and tackles the problem of resolution in two steps. First, the method optimally reduces noise interference with the signal via minimum noiseless description length (MNDL). The new power spectrum estimation MNDL-Periodogram of the denoised signal is then computed via conventional indirect periodogram to improve frequency resolution.

A new approach to nonparametric spectral density estimation

A novel approach to nonparametric spectral density estimation has been proposed. The approach is based on a new evaluation criterion called autocorrelation mean square error (AMSE) for power spectral density (PSD) estimates of available finite length data. Minimization of this criterion not only provides the optimum segmentation for existing PSDE approaches , but also provides a new optimum windowing within the segments that can be combined additionally to the existing methods of nonparametric PSDE. Furthermore, the problem of frequency resolution in existing PSDE methods for noisy signals has been analyzed. In the existing approaches, the additive noise and the finiteness of data which are the causes of the original loss of the frequency resolution are not treated separately. The suggested new approach to spectrum estimation takes advantage of these two different causes of the problem and tackles the problem of resolution in two steps. First, the method optimally reduces noise interference with the signal via minimum noiseless description length (MNDL). The new power spectrum estimation MNDL-Periodogram of the denoised signal is then computed via conventional indirect periodogram to improve frequency resolution.

Recognition of Moving Tracked and Wheeled Vehicles Based on Sound Analysis and Machine Learning Algorithms

The paper presents results of a preliminary study on verification of the possibility to establish simple methods to process acquired sound signals that were generated by a vehicle in motion; to determine its characteristic features for classification as a wheeled or tracked one. The analysis covered 220 signals acquired from real experiment and pre-processed with the use of power spectral density estimation (PSD) and linear prediction coding (LPC). The signal processing methods were used to generate features for which applicability in the classification process was assessed using a statistical method. The set of features was then optimised to reduce the dimensionality of data. Results of recognition obtained with the proposed non-iterative procedures for solving linearly separable problems were compared with results from standard methods, including SVM and k-NN. The developed features as well as selected methods of classification were proposed with respect to the possibility to implement them in low computational power computers for embedded applications.