A neural network based global traveltime function (GlobeNN)

Article Geochemistry & Geophysics

Bayesian seismic tomography using normalizing flows

Xuebin Zhao et al.

Summary: The study presents a new method using normalizing flows to solve Bayesian seismic tomographic problems, providing accurate tomographic results with uncertainty information and significantly reducing computational costs.

GEOPHYSICAL JOURNAL INTERNATIONAL (2022)

Add to Collection

Article Geochemistry & Geophysics

Physics-Informed Neural Networks (PINNs) for Wave Propagation and Full Waveform Inversions

Majid Rasht-Behesht et al.

Summary: We propose a new approach based on physics-informed neural networks (PINNs) to solve wave propagation and full waveform inversions (FWIs). PINNs are tested with forward models and FWI case studies, demonstrating their ability to handle various degrees of structural complexity and automatically satisfy absorbing boundary conditions. The formulation of PINNs allows for the seamless encoding of a priori knowledge of subsurface structures. Results show that PINNs provide good results for both forward models and inversions, with limited computational complexity. PINNs offer exciting perspectives for geophysical inversion problems and can easily be extended to other geophysical datasets and joint inversions.

JOURNAL OF GEOPHYSICAL RESEARCH-SOLID EARTH (2022)

Add to Collection

Article Geochemistry & Geophysics

PINNup: Robust Neural Network Wavefield Solutions Using Frequency Upscaling and Neuron Splitting

Xinquan Huang et al.

Summary: Seismic wave-equation based methods and physics-informed neural network (PINN) have great potential in illuminating the interior of Earth. However, their accuracy and training cost are limited when dealing with high-frequency wavefields. Therefore, a novel approach using frequency upscaling and neuron splitting is proposed to improve the accuracy and convergence speed of wavefield solutions.

JOURNAL OF GEOPHYSICAL RESEARCH-SOLID EARTH (2022)

Add to Collection

Article Geochemistry & Geophysics

Upwind, No More: Flexible Traveltime Solutions Using Physics-Informed Neural Networks

Mohammad Hasyim Taufik et al.

Summary: The eikonal equation is important in various scientific and engineering fields, particularly in seismic wave propagation and travel time modeling. Traditional finite-difference methods suffer from numerical inaccuracies and high computational costs, while physics-informed neural networks offer higher accuracy and scalability. This article demonstrates the flexibility and interpolation-extrapolation capability of PINN solutions, particularly in handling gaps in velocity models.

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

Laplace HypoPINN: physics-informed neural network for hypocenter localization and its predictive uncertainty

Muhammad Izzatullah et al.

Summary: This paper introduces a PINN-based inversion framework called HypoPINN for automatic hypocenter localization and proposes an approximate Bayesian framework for estimating its predictive uncertainties. The effectiveness of this method is demonstrated through numerical examples.

MACHINE LEARNING-SCIENCE AND TECHNOLOGY (2022)

Add to Collection

Article Computer Science, Interdisciplinary Applications

PINNeik: Eikonal solution using physics-informed neural networks

Umair bin Waheed et al.

Summary: The proposed algorithm based on physics-informed neural networks offers a high traveltime accuracy for a wide range of applications in seismology. It leverages machine learning techniques like transfer learning and surrogate modeling to speed up computations for updated velocity models and source locations. The method's flexibility in incorporating various physical properties and improving convergence rate makes it an efficient forward modeling engine for seismological applications.

COMPUTERS & GEOSCIENCES (2021)

Add to Collection

Article Geochemistry & Geophysics

EikoNet: Solving the Eikonal Equation With Deep Neural Networks

Jonathan D. Smith et al.

Summary: EikoNet is a deep learning approach proposed to solve the Eikonal equation, allowing for rapid determination of travel time between any two points in a continuous 3-D domain. By exploiting the differentiability of neural networks, the method calculates spatial gradients analytically without needing solutions from a finite-difference algorithm, and has been rigorously tested on velocity models and sampling methods.

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING (2021)

Add to Collection

Article Multidisciplinary Sciences

Learning the solution operator of parametric partial differential equations with physics-informed DeepONets

Sifan Wang et al.

Summary: The physics-informed DeepONets framework can effectively learn the solution operator of arbitrary PDEs even in the absence of paired training data. It is able to predict the solution of various parametric PDEs up to three orders of magnitude faster compared to conventional solvers, setting a new paradigm for modeling and simulation of nonlinear and nonequilibrium processes in science and engineering.

SCIENCE ADVANCES (2021)

Add to Collection

Article Computer Science, Software Engineering

ttcrpy: A Python package for traveltime computation and raytracing

Bernard Giroux

Summary: ttcrpy is a package for computing travel times and ray tracing of seismic and electromagnetic waves, allowing computations on 2D and 3D grids with three different algorithms implemented. It can run calculations in parallel on multi-core machines and the core computing code is written in C++, wrapped with Cython for practical use.

SOFTWAREX (2021)

Add to Collection

Article Geochemistry & Geophysics

The Community Velocity Model V.1.0 of Southwest China, Constructed from Joint Body- and Surface-Wave Travel-Time Tomography

Ying Liu et al.

Summary: The study aims at building a high-resolution community velocity model (CVM) of crust and uppermost mantle in southwest China by joint inversion of body- and surface-wave travel-time data to address the issues of consistent earthquake locations and velocity models. The resulting VP and VS models show two low-velocity zones in the middle-lower crust and a prominent high-velocity region. These new models serve as the first version of the CVM in southwest China for future studies.

SEISMOLOGICAL RESEARCH LETTERS (2021)

Add to Collection

Article Engineering, Multidisciplinary

A physics-informed deep learning framework for inversion and surrogate modeling in solid mechanics

Ehsan Haghighat et al.

Summary: This study presents the application of Physics Informed Neural Networks (PINN) in solid mechanics, improving accuracy and convergence with a multi-network model and Isogeometric Analysis. The study demonstrates the importance of honoring physics in improving robustness and highlights the potential application of PINN in sensitivity analysis and surrogate modeling.

COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING (2021)

Add to Collection

Review Physics, Applied

Physics-informed machine learning

George Em Karniadakis et al.

Summary: Physics-informed learning seamlessly integrates data and mathematical models through neural networks or kernel-based regression networks for accurate inference of realistic and high-dimensional multiphysics problems. Challenges remain in incorporating noisy data seamlessly, complex mesh generation, and addressing high-dimensional problems.

NATURE REVIEWS PHYSICS (2021)

Add to Collection

Article Geochemistry & Geophysics

Solving the frequency-domain acoustic VTI wave equation using physics-informed neural networks

Chao Song et al.

Summary: Frequency-domain wavefield solutions for the anisotropic acoustic wave equation can help describe the anisotropic nature of the Earth, but inverting the impedance matrix for such solutions can be computationally expensive. To address this challenge, the study proposes using physics-informed neural networks (PINNs) to obtain wavefield solutions for an acoustic wave equation in transversely isotropic (TI) media with a vertical axis of symmetry (VTI). By utilizing automatic differentiation in PINNs, numerical dispersion artefacts can be avoided, providing more accurate solutions.

GEOPHYSICAL JOURNAL INTERNATIONAL (2021)

Add to Collection

Article Geochemistry & Geophysics