Higher-order spectra and independent component analysis used for identification and SNR enhancement of acoustic emission signals

Abstract

Higher order statistics are used twofold: to characterize, and to enhance and to de-noise primary signals and their echoes in acoustic emission events occurring in samples of steel pipes under transversal strain. First, the characterization by diagonal bi-spectra is performed and it allows the spectral separation of the primary event from the secondary ones (echoes). Secondly, a cumulant-based independent component analysis is applied for blind sources separation in a low-SNR scenario. The method is first validated considering a synthetic of acoustic signals. Then, the developed algorithm is applied to a sequence of quartets of primary bursts and their first three echoes. The denoising capability of ICA is assessed by comparing the power spectra of the sources vs. the separated signals. Data were acquired by wide frequency-range transducers (100-800 kHz) and digitalized by a 2.5 MHz, 8-bit ADC