Strategies for characterization of chemical acoustic emission signals near the conventional detection limit

Abstract

Two methods are proposed whereby spontaneous acoustic emission from chemical processes may be better characterized. Signals must still have peak amplitudes greater than a threshold voltage but, contrary to normal practice, this is now set low enough such that the background noise signals are deliberately collected. These are used as a training set for “noise models” by which other noise and real signals are then classified. Results suggest that real acoustic signals of amplitudes similar to that of the background noise may now be acquired and differentiated from those due to background noise. In the first (sequential) method, signals due solely to noise are collected just prior to the acoustically active reaction, and characterized by 33 time/amplitude- and frequency-based based descriptors; the descriptor distributions obtained provide an accurate model of the noise, and are represented by non-parametric centre and spread estimators. Onset of reaction must result in a detectable change in emission characteristics (emission rate, frequency, RMS, etc.). Valid candidate chemical signals are those later signals which statistically do not fit the noise model. About 850 signals from the recrystallization of potassium nitrate from a hot concentrated solution were treated in this manner; typical signals and improved acoustic spectra were obtained. In the second (simultaneous) method, noise and real signals are acquired together, but the acquisition threshold is lowered such that background noise signals form a significant majority of the signals collected. Real signals are extracted by their differences from the noise-dominated model. This proved successful for signals collected during stressing of samples of poly(vinyl chloride) and Celeron. Attention was paid to recognized sources of interference. Waveforms were examined both visually and by pattern recognition. Use of descriptor distributions in a rule-based signal classification scheme is proposed.