On the use of neural networks for full waveform inversion
From MaRDI portal
Publication:6096500
DOI10.1016/J.CMA.2023.116278arXiv2303.03260MaRDI QIDQ6096500FDOQ6096500
Authors: Leon Herrmann, Tim Bürchner, Felix Dietrich, S. Kollmannsberger 
Publication date: 12 September 2023
Published in: Computer Methods in Applied Mechanics and Engineering (Search for Journal in Brave)
Abstract: Neural networks have recently gained attention in solving inverse problems. One prominent methodology are Physics-Informed Neural Networks (PINNs) which can solve both forward and inverse problems. In the paper at hand, full waveform inversion is the considered inverse problem. The performance of PINNs is compared against classical adjoint optimization, focusing on three key aspects: the forward-solver, the neural network Ansatz for the inverse field, and the sensitivity computation for the gradient-based minimization. Starting from PINNs, each of these key aspects is adapted individually until the classical adjoint optimization emerges. It is shown that it is beneficial to use the neural network only for the discretization of the unknown material field, where the neural network produces reconstructions without oscillatory artifacts as typically encountered in classical full waveform inversion approaches. Due to this finding, a hybrid approach is proposed. It exploits both the efficient gradient computation with the continuous adjoint method as well as the neural network Ansatz for the unknown material field. This new hybrid approach outperforms Physics-Informed Neural Networks and the classical adjoint optimization in settings of two and three-dimensional examples.
Full work available at URL: https://arxiv.org/abs/2303.03260
Recommendations
- On obtaining initial approximation for the full wave inversion problem using convolutional neural network
- Combining gradient optimization and machine learning methods for inverse problems in layered heterogeneous media
- Robust Full Waveform Inversion: A Source Wavelet Manipulation Perspective
- SeismicNET: physics-informed neural networks for seismic wave modeling in semi-infinite domain
- Fourier-DeepONet: Fourier-enhanced deep operator networks for full waveform inversion with improved accuracy, generalizability, and robustness
deep learninginverse problemsfinite difference methodfull waveform inversionadjoint optimizationphysics-informed neural networks
Cites Work
- Algorithm 799: revolve
- DGM: a deep learning algorithm for solving partial differential equations
- On the limited memory BFGS method for large scale optimization
- Deep learning
- Title not available (Why is that?)
- The adjoint method in seismology
- The adjoint method in seismology. II: Applications: traveltimes and sensitivity
- Finite cell method. \(h\)- and \(p\)-extension for embedded domain problems in solid mechanics
- Solving high-dimensional partial differential equations using deep learning
- Physics-constrained deep learning for high-dimensional surrogate modeling and uncertainty quantification without labeled data
- Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
- Title not available (Why is that?)
- Temporal-kernel recurrent neural networks
- Finite difference computing with PDEs. A modern software approach
- Immersed boundary parametrizations for full waveform inversion
- Topology optimization of turbulent flows
- Physics-informed neural networks for high-speed flows
- An energy approach to the solution of partial differential equations in computational mechanics via machine learning: concepts, implementation and applications
- Conservative physics-informed neural networks on discrete domains for conservation laws: applications to forward and inverse problems
- Adaptive activation functions accelerate convergence in deep and physics-informed neural networks
- Prediction of aerodynamic flow fields using convolutional neural networks
- Modeling the dynamics of PDE systems with physics-constrained deep auto-regressive networks
- A tutorial on the adjoint method for inverse problems
- Scientific machine learning through physics-informed neural networks: where we are and what's next
- Deep learning in computational mechanics. An introductory course
- Physics-informed neural networks for inverse problems in supersonic flows
- Self-adaptive physics-informed neural networks
- Deep learning of free boundary and Stefan problems
- Shape identification of scatterers using a time-dependent adjoint method
- Solving inverse problems in stochastic models using deep neural networks and adversarial training
- Flow over an espresso cup: inferring 3-D velocity and pressure fields from tomographic background oriented Schlieren via physics-informed neural networks
- Finite basis physics-informed neural networks (FBPINNs): a scalable domain decomposition approach for solving differential equations
- Title not available (Why is that?)
Cited In (7)
- SeismicNET: physics-informed neural networks for seismic wave modeling in semi-infinite domain
- Application of neural networks in nonlinear inverse problems of geophysics
- On obtaining initial approximation for the full wave inversion problem using convolutional neural network
- Deep learning in computational mechanics: a review
- Isogeometric multi-resolution full waveform inversion based on the finite cell method
- Fourier-DeepONet: Fourier-enhanced deep operator networks for full waveform inversion with improved accuracy, generalizability, and robustness
- Forward modeling with forced neural networks for gravity anomaly profile
This page was built for publication: On the use of neural networks for full waveform inversion
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6096500)
