Inverting the Subsurface Fracture Density by Detecting the Phase Difference of Various Seismic wave with Machine Learning

The fracture density of rock could create the applying shock waves phase difference at the laboratory scale. For using this technology in determining the stress status along the wave propagating path, removing unnecessary noise is the most crucial task. In this study, a machine learning method with Long short-term memory (LSTM) algorithm to retreat the signals from sophisticated seismograms is proposed. The primary analyzing target is data across the 2022 Taitung, Eastern Taiwan seismic event and another micro-seismic data set associated with a surface crack on a hill of Ping-Tong, southern Taiwan. It is found that there is no phase difference among vertical and horizontal components from the same record, when comparing the difference between two various records then the result is distinct. The detecting sub-surface crack density via phase difference has increased in some seismic data pairs of eastern Taiwan after the rupture of the 2022 Taitung earthquake. The machine learning method with LSTM helps to elevate the data retrieval accuracy which cannot be done by conventional Fast Fourier Transformation (FFT). Records from stations adjacent to the hypocenter offer better agreement in phase difference measurement, the higher signal possibly causes it to noise ratio (SNR) in the such neighborhood.

 

Keywords: phase difference, machine learning, LSTM, crack density, stress field