Comparison of pattern recognition descriptors for chemical acoustic emission analysis

Abstract

SUMMARYFeatures used to characterize acoustic emission signals from chemical systems are evaluated with regard to their potential for pattern recognition. Eight chemical systems involving phase transitions, hydration, dissolution and effervescence are employed and treated as separate signal classes. These are compared pairwise and the discriminatory capabilities of about 50 features are investigated by computing Fisher weights. Time domain and frequency domain descriptors are examined. Correlations among the features evaluated are also reported. Recommended descriptors are the mean and median frequencies, frequency bandwidth, number of level crossings (0% and 25%), crest factor (time and frequency domains), halfIife, kurtosis and normalized percentiles of the signal and its power spectrum. The effectiveness of the recommended descriptors is demonstrated through the separation of signal classes in two different systems (melting ice and an enzyme-catalyzed gas formation reaction) by principal components analysis.

Evidence of presence of tweed in PbSc 0.5 Ta 0.5 O 3 crystals based on acoustic emission frequency spectrum analysis

“…Figure 1 shows an AE burst, accompanying the T * (a), and the corresponding fast Fourier transform (FFT) spectrum from PST-8 (b). Both the AE burst and the FFT spectrum look similar as those previously reported [27]. The AE burst attenuates slowly and monotonically during of about 0.082 ms.…”

“…It is clear, while AE is able to detect both T * and T * tw [22,26], that the mentioned above differences are expected to appear in their frequency spectrum of AE. Indeed, previously an analysis of the frequency spectrum of the waveform of AE bursts indicates the matching between sizes of entities of the different chemical systems, undergoing the phase transitions, hydration, dissolution etc., and the corresponding spectrum of AE bursts from them [27]. Moreover, it was observed that the maxima of the frequency spectrum of AE bursts shift to lower frequencies when increasing the sizes of the entities, i.e.…”