Higher Order Statistics and Independent Component Analysis for Spectral Characterization of Acoustic Emission Signals in Steel Pipes

Higher order statistics (HOS) are used to characterize acoustic emission events in ring-type samples from steel pipes for the oil industry. Cumulants are used twofold. First, diagonal bispectrum allows the separation of the primary (original) deformation from the reflections produced mainly in the suppressed chord. These longitudinal reflections can hardly be extracted via second-order methods, e.g., wavelet packets and power spectra, because they are partially masked by both Gaussian and non-Gaussian noise. Second, a cumulant-based independent component analysis may be used before the bispectrum, as a preprocessing complement, in the case of multiple-source and multiple-channel recordings. This algorithm suppresses the mutual influence of the sources and sensors. Sample registers were acquired by wide-frequency-range transducers (100-800 kHz) and digitalized by a 2.5-MHz, 12-bit analog-to-digital converter.