Application of Physics-Informed Neural Networks to a 3D Fault Slip Model

Forward and Inverse Problems on Slow Slip Events

Abstract

Various fault slip modes, including earthquake and slow slip events (SSEs), have been observed in many subduction zones. This diversity of slip modes can be explained by the heterogeneous frictional properties. Thus, it is crucial to constrain the frictional properties based on the geodetic observation and fault slip model. Physics-Informed Neural Networks (PINNs) have been developed as a new machine-learning based method to solve partial differential equations. Focusing on its potential for physics-based inversion, we extend the PINN-based method to a 3D fault slip model toward the application of real observation. We conducted the fault slip simulations (forward problem) and the estimation of frictional parameters from fault slip observations (inverse problem). We verify that PINNs can work well in a 3D SSE fault model through numerical experiments.