Signal Processing Tools for Non-Stationary Signals Detection

In this paper we aim to compare the abilities and performances of signal processing tools to detect non-stationary signals coming from condition monitoring of electrical machines. From the vast amount of available tools, we focus on existing signal processing methods suitable for real applications for non-stationarities tracking and quantification over time which is particularly interesting in fault diagnosis. First, we assess the spectral kurtosis, a tool that gained much attention because of his capability to characterize transients masked by strong noises. In order to detect non-stationarities, other methods are evaluated such as the spectral subtraction through the short time Fourier transform or the Wiener filtering which can remove stationary components. The analytical framework of each tool is first presented. Non-stationary tests signals based on properties of vibration signals of bearings are proposed to compare effectiveness, advantages and drawbacks of each methods for non-stationarities detection. The purpose is to select a method that is best suited for each type of non-stationarity in order to improve the reliability of the detection.