Publication Date:
2023-05-02
Description:
One of the contemporary trend in seismology is process huge data sets automatically with use of Neural Network (NN) formalism. We present seismogram onsets interpretation obtained both by Convolution NN as well as Recurrent NN approach. We investigated data from Acoustic Emission loading experiment with Westerly Granite. Such data appeared to be suitable for testing of NN approach as they are more homogeneous then data originated from natural earthquakes, but simultaneously they are complex enough not to be of trivial interpretation. We designed NN architecture, learned in and compare the results with biased interpretation. We were searching not only for onsets on individual seismograms but we try to identified the whole events. In addition to automatic onsets identification we (also automatically) determined event location and seismic moment tensor. Comparison with biased data proved that these automatically obtained values can be successfully used as preliminary estimation at least. Problems of multiple events identification are discussed as well. The method has a potential to be applicable on natural earthquake seismograms.
Language:
English
Type:
info:eu-repo/semantics/conferenceObject