?
ОТСЛЕЖИВАНИЕ РАЗВИТИЯ РАЗРУШЕНИЯ С ПОМОЩЬЮ КЛАСТЕРИЗАЦИИ ИМПУЛЬСОВ ТЕРМИЧЕСКИ СТИМУЛИРОВАННОЙ АКУСТИЧЕСКОЙ ЭМИССИИ ПРИ ОТСУТСТВИИ ЛОКАЦИИ
The paper studies the clusterability of acoustic emission pulses during high-temperature heating of sandstone sample preliminarily subjected to mechanical loading. Mechanical loading was applied in uniaxial mode up to load close to destructive with appearance of signs of large cracks on the surface. After that, samples were subjected to thermal treatment up to 650 °C with registration of thermoacoustic emission (TAE) pulses. The clustering of pulses was carried out based on their similarity established by the method of mutual correlation of wave forms. Three clusters were identified, each of which contains about ten TAE pulses and may correspond to a specific source, presumably a separate large crack. The distribution of pulses from clusters in time and by amplitude during heating is different for each cluster. This may indicate both different moments and activation thresholds of the corresponding crack, and different rates of crack growth. A separate control experiment was carried out with thermal treatment on a sandstone sample, which was not mechanically loaded beforehand. It was not possible to identify clusters of TAE pulses for it. Additionally, the parameters of the TAE pulses of the identified clusters were analyzed. It was found out that the clusters do not form compact isolated groups in the parameter space, but are distributed against the background of other pulses. Checking the pulses by parameters based on physically justified criteria showed in two out of three clusters the presence of several pulses that were inadequate to the physical characteristics of the experiment. Hierarchical clustering of all TAE pulses in the parameter space did not allow us to identify groups that were in any way similar to the original clusters. The stability of identifying the original clusters in the parameter space was tested using classification by an ensemble of decision trees. The third cluster was recognized with the simplest learning criteria. The dynamics of this cluster's pulses is most similar to the dynamics of crack growth activated at a certain heating temperature.