Recovering Periodic Impulsive Signals Through Skewness Maximization

Maximizing the skewness of a measured signal by adaptive filtering to reveal hidden periodic impulses is proposed as a preprocessing method. Periodic impulsive signals are modeled by harmonically related sinusoids to prove that amplitude and phase distortion from a transfer function, effects of sinusoidal interferences, and noise can be compensated for by a linear filter. The convergence behavior of the skewness maximization algorithm is analyzed to show that it is possible to recover the original harmonic structure with an unknown fundamental frequency by achieving maximum skewness in the given signal. It is shown that maximizing the skewness always results in a subspace containing only a single harmonic family. Defect detection in rolling element bearings is presented as an application example and as a comparative study against kurtosis maximization.