Influence of non-Gaussian noise on the effectiveness of cyclostationary analysis – Simulations and real data analysis

Abstract Cyclostationary analysis is a useful approach in diagnostics of the machinery with rotating components. It allows indicating the cyclic modulations in the signal via analysis of a bi-frequency map called Cyclic Spectral Coherence (CSC). A non-zero CSC value at two frequencies means the presence of the cyclic process. Unfortunately, we have found that for some cyclostationary signals CSC provides difficult to interpret information. These disturbances in the CSC map have been linked to the presence of non-Gaussian noise. To prove it an original procedure has been proposed. Using simulations covering the model of signal, α -stable distribution, and Monte Carlo simulations it has been shown that indeed increasing presence of non-Gaussian noise makes worse the quality of diagnostic information extracted from CSC map. It has been recalled that the Cyclic Spectral Coherence is based on the autocovariance function of a given signal, thus it is properly defined for data coming from the distribution with the finite second moment. Finally, the authors selected three real examples that confirm the simulation-based findings. The main conclusion is before using the CSC analysis for cyclostationary signal one should validate the type of the noise. If noise is Gaussian - the CSC will bring optimal results. For the increasing level of impulsive non-cyclic noise, the CSC map becomes more and more disturbed and the detection of periodic excitation is difficult. Performed simulations on a very generic model some guidelines have been formulated regarding the acceptable level of non-Gaussian noise.