Mr. Fukushima from Stanford University will visit DPRI on December 22–23.
He will present his recent studies on estimation of frictional parameters using physics-informed neural network and on data assimilation for dynamic rupture simulation.
We would greatly appreciate your participation.

Date: Monday, December 22, 16:00–17:30
Venue: DPRI, main building, Seminar Room E232D
Speaker:　Rikuto Fukushima (Stanford Univ.)
Title: Variational data assimilation for fault friction from geodetic and seismic observations
Abstract:
The fault slip simulation, formulated as a coupling problem of elasticity and fault friction, has successfully reproduced various observed seismic and aseismic slips, advancing our understanding of fault mechanics. Nevertheless, many previous studies relied on trial and error to obtain dynamic parameters (e.g., frictional properties and initial stress) consistent with observed strong motion and surface displacement, and the application of dynamic inversion, which quantitatively determines dynamic parameters with fault physics and observation, is limited. As an inversion method to estimate fault frictional properties, this presentation focuses on the adjoint method and Physics-Informed Neural Networks (PINNs) as a variational data assimilation. I will discuss (i) PINN-based frictional parameter inversion from geodetically observed Slow Slip Events (real data application of 2010 Bungo SSE), and (ii) Adjoint-based rupture inversion from seismic wavefield observation (numerical experiment with synthetic observation).