Physics-informed neural network for seismic wave inversion in layered semi-infinite domain

TL;DR

This paper introduces a physics-informed neural network framework for seismic wave inversion in layered semi-infinite domains, effectively integrating physical laws and boundary conditions to estimate Earth's subsurface material distribution.

Contribution

The paper presents a novel PINN architecture tailored for 1D semi-infinite domains, incorporating absorbing boundary conditions and demonstrating its effectiveness through experiments.

Findings

Accurately estimates subsurface material distribution from sparse data.

Effectively incorporates physical laws and boundary conditions.

Validated through experiments on seismic wave propagation models.

Abstract

Estimating the material distribution of Earth's subsurface is a challenging task in seismology and earthquake engineering. The recent development of physics-informed neural network (PINN) has shed new light on seismic inversion. In this paper, we present a PINN framework for seismic wave inversion in layered (1D) semi-infinite domain. The absorbing boundary condition is incorporated into the network as a soft regularizer for avoiding excessive computation. In specific, we design a lightweight network to learn the unknown material distribution and a deep neural network to approximate solution variables. The entire network is end-to-end and constrained by both sparse measurement data and the underlying physical laws (i.e., governing equations and initial/boundary conditions). Various experiments have been conducted to validate the effectiveness of our proposed approach for inverse modeling of seismic wave propagation in 1D semi-infinite domain.