Acoustic Emission Results and Spectral AnalysisDue to their complementarity in signal description, the fraction of highamplitude peaks in acoustic signal (>5000, in 16 signed bits coding), namelyRp+, and the average amplitude value of the sound recorded (AAS, measureof the sound intensity or “loudness”), computed during the last 10 s of thefirst compression step, were thought to be of particular relevance. Otherpossible indices for acoustic temporal signal description, as number of peaksand maximum amplitude, were highly correlated to the one or the other. Whenmapping the results of Rp+ obtained for all samples against those of AAS(Fig. 10), the acoustic signals of samples J and E showed low average amplitudes(AAS < 120), but more peaks of larger amplitude (Rp+ > 7%), i.e.,louder ones. Conversely, the group of samples I, A and D emitted fewer highamplitude peaks (<6%) and higher AAS (close to 180). Other samples showedsimilar values of Rp+ ratio as the latter group, but intermediate values of AAS,closer to the first group. Large SDs limit sample classification at this stage(Table 3). In line with the commonly admitted acoustic perception of crispness(Duizer 2001), the lower values of AAS of samples J and E are in agreementwith the stability of the size distribution of flakes, meaning that very fewpieces are broken in compression to 100 N. Conversely, the size redistributionduring the compression of other samples is due to numerous fractures reflectedby the larger number of acoustic events of moderate amplitude.Spectral analysis has been performed on every sample. Normalizedpower spectra are displayed for representative samples of each group formed by multidimensional analysis of sensory data (Fig. 11A–D). All samples displaya similar feature with rather uniformly distributed emission in terms offrequency, including a gap for a frequency that is slightly higher than 5 kHz.Only the average normalized power spectrum of sample J shows significantdifferences with the other samples, mainly marked by a large peak at 1.7 kHz(Fig. 11D). For further comparison, the principal component analysis is performedon the acoustic spectra (Fig. 12A). The first dimension accounts for56.7% of the variations, and the corresponding eigenvector is close to theopposite of J spectrum, displaying typical low frequency emission in thespectral domain of interest (Fig. 12B). The second principal componentclearly opposes samples E to L and G, by a contrast in acoustic emission,principally in the first part of the spectral range for E (<3 kHz) and in thesecond part for the other two samples (close to 7.5 kHz) (Fig. 12C). All othersamples were characterized by similar values of spectral dimension. Therelevance of the acoustic signal analysis in the spectral domain relies in theevidence of a different behavior for sample J, which confirms a different
fracture mechanism.
การแปล กรุณารอสักครู่..
