A study on Bayesian spectrum estimation based diagnostics in electrical rotating machines

Predictive maintenance philosophy is fast becoming a norm in industry, where prognostics and diagnostics in electrical machines are essential. The efficiency and reliability of the technique being utilized depend profoundly on measurement accuracy and analysis. Frequency analysis is commonly used in the interpretation of measurements for condition monitoring purposes. This paper presents a study of techniques in frequency analysis in condition monitoring of electrical rotating machines. Different performance characteristics of various spectral estimation techniques are compared for application in incipient fault diagnosis. The study includes an evaluation of a Bayesian spectral estimation method together with more conventional practices such as the standard periodogram, Welch and Music methods. The investigation uses an example of shaft voltage based condition monitoring in machines for a specific case of eccentricity. Results of the study indicate that the Bayesian method, although unconventional in fault diagnostics, is exceptionally robust and exhibits qualities well-suited to the application.