Predict arrival time of acoustic emissions on a laboratory fault without the velocity model

Estimation of arrival time of seismic wave requires the knowledge of the velocity model between the hypocenter to the sensors. The velocity model can be obtained by tomographic imaging in the laboratory, even though in most cases isotropic models are assumed. Our earlier work showed that machine learning using artificial neural network (ANN) can relocate acoustic emission (AE) on a laboratory fault without knowing the velocity model (Zhao and Glaser, 2020, Rock Mech Rock Eng, 53, 2053–2061). In this study, we demonstrate that the same ANN structure can be used in a reverse way, i.e., predicting arrival time for AE events with known locations. This approach estimates arrival time without forward modelling, and it may be used to predict arrival time of events at any locations on the fault surface, providing virtual raypath coverage that may facilitate improved tomographic inversion. Moreover, this method could be applied to natural earthquakes, for situations where arrival time data are missing due to, for example, malfunctioned stations and/or weak seismic signals.