Locating and estimating full moment tensors of acoustic emission events in samples under high pressure and temperature

Accurate locations and moment tensors of acoustic emission (AE) events can reveal micro- and macroscopic faulting and phase transformation in rock samples under high pressure and temperature. To process large amount of waveform data of AE experiments, we used a neural network called Long Short-Term Memory(LSTM) in machine learning to train a computer to pick the first P arrivals of AE events. We first used an experiment data set where the first P arrival times of 591 AE events have been manually picked. 30% of data were used to train the LSTM and the rest of data were used to test the performance. We found that the LSTM can pick the first P arrivals with an error less that 0.1 μs for 85% of events and less than 0.5 μs for all events.We then applied LSTM to waveform data from a recent AE events using the double-difference relocation method. In addition, we determined full moment tensors of the events using the amplitudes of first P arrivals. Since rock samples in high pressure experiments are usually an isotropic, we developed a moment tensor inversion method in vertically transverse isotropic media. We used a grid search method to determine the source parameters in terms of fault plane orientation plus isotropic and CLVD strengths. We found that the magnitude of anisotropy of rock sample affects the moment tensor results greatly. In particular, the isotropic component of the source can vary from explosive to none or weak anisotropic model to implosive in strong anisotropic model. We suggest to measure the in-situ anisotropic property of sample in future AE experiments.