Article
Water Resources
Mohammad Afzal Shadab, Dingcheng Luo, Eric Hiatt, Yiran Shen, Marc Andre Hesse
Summary: In this work, a deep learning technique called Physics Informed Neural Networks (PINNs) is used to study steady ground-water flow in unconfined aquifers. PINNs utilize both physics information represented by partial differential equations (PDEs) and data obtained from physical observations. The training of PINNs involves steady-state analytical solutions and laboratory based experiments to predict phreatic surface profiles and estimate the hydraulic conductivity. The results show that PINNs can overcome the limitations of the Dupuit-Boussinesq equation and produce better predictions by incorporating physics information from more complete flow models like the one derived by Di Nucci.
ADVANCES IN WATER RESOURCES
(2023)
Article
Automation & Control Systems
Jingzhi Tu, Chun Liu, Pian Qi
Summary: Crack is a critical factor in degrading the performance of machinery manufacturing equipment. Physics-informed neural networks (PINNs) have shown great potential for solving physical problems, such as predicting crack paths and simulating crack propagation. However, the commonly used posteriori adaptive refinement techniques for obtaining refined meshes require computationally expensive pretest calculations. To address this issue, a PointNet-based adaptive refinement method is proposed to avoid precalculation when constructing the discrete domain, providing efficient and reliable results when using the PINN framework.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Construction & Building Technology
Yutong Wan, Wenzhong Zheng, Ying Wang
Summary: The corrosion of steel caused by chloride is an important factor in the deterioration of reinforced concrete structures. This study demonstrates the use of physics-informed neural networks (PINNs) to accurately determine the chloride diffusion coefficient in non-steady-state immersion tests and accelerated chloride migration tests, considering different chloride binding capacities. PINNs show robustness when trained with noisy data and exhibit excellent generalization performance for interpolation and extrapolation.
CONSTRUCTION AND BUILDING MATERIALS
(2023)
Article
Biochemistry & Molecular Biology
Sara Ibrahim Omar, Chen Keasar, Ariel J. Ben-Sasson, Eldad Haber
Summary: The inverse protein folding problem seeks to predict an amino acid sequence that folds into a specific structure and performs a specific function. Machine learning techniques have been successful in generating functional sequences, but lack interoperability and robustness for proteins that function under non-ambient conditions. To address this, we propose a new Physics-Informed Neural Networks (PINNs)-based protein sequence design approach that combines molecular dynamics simulations and relaxation of binary programming. Our design framework demonstrates effectiveness in designing proteins that can function under non-ambient conditions.
Article
Engineering, Multidisciplinary
Kevin Linka, Amelie Schafer, Xuhui Meng, Zongren Zou, George Em Karniadakis, Ellen Kuhl
Summary: Understanding real-world dynamical phenomena is challenging, and machine learning has become the go-to technology for analyzing and making decisions based on these phenomena. However, traditional neural networks often ignore the fundamental laws of physics and fail to make accurate predictions. In this study, the combination of neural networks, physics informed modeling, and Bayesian inference is used to integrate data, physics, and uncertainties, improving the predictive potential of neural network models.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2022)
Review
Engineering, Electrical & Electronic
Bin Huang, Jianhui Wang
Summary: The advances of deep learning techniques have brought new opportunities to power systems. However, there are challenges in applying deep learning in power systems, such as the requirement for high-quality training data, production of physically inconsistent solutions, and low interpretability. Physics-informed neural networks (PINNs) can address these concerns by integrating physics rules into deep learning methodology. This survey provides a systematic overview of PINN in power systems, summarizing different paradigms and investigating their applications and relevant research.
IEEE TRANSACTIONS ON POWER SYSTEMS
(2023)
Article
Mechanics
Hamidreza Eivazi, Mojtaba Tahani, Philipp Schlatter, Ricardo Vinuesa
Summary: This article introduces the application of Physics-informed neural networks (PINNs) in solving and identifying partial differential equations. By applying PINNs to solve boundary layer problems of the Navier-Stokes equations and simulate various turbulent flow cases, it is demonstrated that PINNs have good applicability for both laminar and turbulent flows.
Article
Computer Science, Interdisciplinary Applications
Mohammad Amin Nabian, Rini Jasmine Gladstone, Hadi Meidani
Summary: PINNs are deep neural networks trained to compute the response of systems governed by PDEs using automatic differentiation. Although successful, they still need improvements in computational efficiency, which is why this paper studies the performance of an importance sampling approach for efficient training of PINNs.
COMPUTER-AIDED CIVIL AND INFRASTRUCTURE ENGINEERING
(2021)
Article
Mathematics, Applied
Zhiwei Gao, Liang Yan, Tao Zhou
Summary: In this work, an adaptive strategy called FI-PINNs is proposed for solving PDE problems. The strategy defines failure probability based on the residual and improves numerical accuracy by adding more training points. Similar to adaptive finite element methods, the approach uses failure probability as a posterior error indicator to generate new training points.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2023)
Article
Computer Science, Interdisciplinary Applications
Emilio Jose Rocha Coutinho, Marcelo Dall'Aqua, Levi McClenny, Ming Zhong, Ulisses Braga-Neto, Eduardo Gildin
Summary: Physics-informed Neural Network (PINN) is a promising tool for physical phenomena described by partial differential equations (PDE), but it struggles with stiff problems that involve shocks in their solutions. Previous studies manually adjusted an artificial viscosity (AV) value to address this, but this paper proposes three methods that do not rely on predefined AV values. These methods successfully learn AV values and shock locations, and improve the approximation error.
JOURNAL OF COMPUTATIONAL PHYSICS
(2023)
Article
Computer Science, Interdisciplinary Applications
Jared O'Leary, Joel A. Paulson, Ali Mesbah
Summary: This study proposes a framework for training artificial neural networks to learn the hidden physics within stochastic differential equations (SDEs). The framework propagates stochasticity through the known structure of the SDE and utilizes automatic differentiation and mini-batch gradient descent to establish the parameters of the neural networks. The results demonstrate the potential of this method in unraveling the hidden physics of multivariate stochastic dynamical systems.
JOURNAL OF COMPUTATIONAL PHYSICS
(2022)
Article
Computer Science, Interdisciplinary Applications
Umair bin Waheed, Ehsan Haghighat, Tariq Alkhalifah, Chao Song, Qi Hao
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)
Article
Computer Science, Interdisciplinary Applications
Ehsan Taghizadeh, Helen M. Byrne, Brian D. Wood
Summary: In this study, a combination of formal upscaling and data-driven machine learning was used to explicitly close nonlinear transport and reaction process in multiscale tissues. The neural network trained to model the closure problem exhibited good generalizability and high fidelity in predicting the effectiveness factor for tissues with different scale and complexity. This approach not only resulted in an upscaled nonlinear PDE but also identified important source terms for closure and improved the accuracy of the models predicting correction factors.
JOURNAL OF COMPUTATIONAL PHYSICS
(2022)
Article
Computer Science, Interdisciplinary Applications
Khang A. Luong, Thang Le-Duc, Jaehong Lee
Summary: Exact boundary conditions imposition becomes challenging when dealing with complex geometric domains or important BCs selection. To overcome this limitation, an unified physics-informed neural network (UPINN) model is introduced, with trial functions provided by deep neural networks (DNNs). The UPINN combines two phases: the first phase finds DNN-based trial functions satisfying essential BCs, and the second phase solves BVPs using exact BCs imposition procedure to constrain network outputs. The UPINN demonstrates improved prediction accuracy and training cost for solid mechanics problems with various BCs, even in the presence of complex restrictions.
ENGINEERING WITH COMPUTERS
(2023)
Article
Computer Science, Interdisciplinary Applications
Levi D. McClenny, Ulisses M. Braga-Neto
Summary: Physics-Informed Neural Networks (PINNs) are a promising application of deep neural networks for the numerical solution of nonlinear partial differential equations (PDEs). In this paper, a fundamentally new way to train PINNs adaptively is proposed, where fully trainable adaptation weights are applied to each training point individually, enabling the neural network to autonomously learn the difficult regions of the solution and focus on them. The self-adaptation weights provide a soft multiplicative soft attention mask resembling mechanisms used in computer vision. In numerical experiments, the proposed SA-PINN outperforms other state-of-the-art PINN algorithms in L2 error with fewer training epochs.
JOURNAL OF COMPUTATIONAL PHYSICS
(2023)
Article
Mathematics, Applied
Kent-Andre Mardal, Jarle Sogn, Stefan Takacs
Summary: In this paper, an optimization problem with limited observation governed by a convection-diffusion-reaction equation is analyzed. Continuous norms are derived to analyze well-posedness and derive error analysis and a robust preconditioner for the problem parameters. Conditions for inf-sup stable discretizations are provided, and a discretization for box domains with constant convection is presented.
SIAM JOURNAL ON NUMERICAL ANALYSIS
(2022)
Article
Mathematics, Applied
Wietse M. Boon, Timo Koch, Miroslav Kuchta, Kent-Andre Mardal
Summary: This paper presents mesh-independent and parameter-robust monolithic solvers for the coupled primal Stokes-Darcy problem, considering three different formulations and their discretizations in various finite element and finite volume methods. Robust preconditioners are derived using a unified theoretical framework, utilizing operators in fractional Sobolev spaces. Numerical experiments demonstrate the parameter-robustness of the proposed solvers.
SIAM JOURNAL ON SCIENTIFIC COMPUTING
(2022)
Article
Biotechnology & Applied Microbiology
Martin Hornkjol, Lars Magnus Valnes, Geir Ringstad, Marie E. Rognes, Per-Kristian Eide, Kent-Andre Mardal, Vegard Vinje
Summary: In this paper, a computational model was used to study the clearance of a tracer driven by the circulation of cerebrospinal fluid (CSF) produced in the choroid plexus (CP). It was found that the circulation in the choroid plexus can accelerate the clearance of CSF both in the subarachnoid space and in the brain parenchyma.
FRONTIERS IN BIOENGINEERING AND BIOTECHNOLOGY
(2022)
Review
Multidisciplinary Sciences
Tomas Bohr, Poul G. Hjorth, Sebastian C. Holst, Sabina Hrabetova, Vesa Kiviniemi, Tuomas Lilius, Iben Lundgaard, Kent-Andre Mardal, Erik A. Martens, Yuki Mori, U. Valentin Nagerl, Charles Nicholson, Allen Tannenbaum, John H. Thomas, Jeffrey Tithof, Helene Benveniste, Jeffrey J. Iliff, Douglas H. Kelley, Maiken Nedergaard
Summary: This article reviews theoretical and numerical models of the glymphatic system, which plays a role in solute transport and has clinical applications in drug delivery, stroke, and neurodegenerative disorders. The authors categorize existing models based on their anatomical functions and highlight the need for future work, including new models and experiments to improve our understanding of the system.
Article
Mathematics, Applied
Qingguo Hong, Johannes Kraus, Miroslav Kuchta, Maria Lymbery, Kent-Andre Mardal, Marie E. Rognes
Summary: This paper introduces a class of three-field finite element formulations for the generalized Biot-Brinkman equations and demonstrates their robustness in different parameter regimes through theoretical analysis and numerical investigation.
JOURNAL OF SCIENTIFIC COMPUTING
(2022)
Article
Clinical Neurology
A. Sperre, I. Karsrud, A. H. S. Rodum, A. Lashkarivand, L. M. Valnes, G. Ringstad, P. K. Eide
Summary: This study found that intrathecal gadobutrol in doses up to 0.50 is safe, based on observations of 196 patients who received intrathecal contrast-enhanced imaging of the cerebrospinal fluid.
AMERICAN JOURNAL OF NEURORADIOLOGY
(2023)
Review
Radiology, Nuclear Medicine & Medical Imaging
Nivedita Agarwal, Laura D. Lewis, Lydiane Hirschler, Leonardo Rivera Rivera, Shinji Naganawa, Swati Rane Levendovszky, Geir Ringstad, Marijan Klarica, Joanna Wardlaw, Costantino Iadecola, Cheryl Hawkes, Roxana Octavia Carare, Jack Wells, Erik N. T. P. Bakker, Vartan Kurtcuoglu, Lynne Bilston, Maiken Nedergaard, Yuki Mori, Marcus Stoodley, Noam Alperin, Mony de Leon, Matthias J. P. van Osch
Summary: Neurofluids refer to all fluids in the brain and spine, including blood, cerebrospinal fluid, and interstitial fluid. Neuroscientists have identified various fluid environments that interact harmoniously to support optimal brain function. Animal studies have been crucial in understanding the dynamics of neurofluids, while human studies are limited due to the lack of noninvasive imaging techniques. The future development of noninvasive MRI techniques holds promise in imaging neurofluid dynamics and identifying pathological processes.
JOURNAL OF MAGNETIC RESONANCE IMAGING
(2023)
Article
Multidisciplinary Sciences
Per Kristian Eide, Aslan Lashkarivand, Are Pripp, Lars Magnus Valnes, Markus Herberg Hovd, Geir Ringstad, Kaj Blennow, Henrik Zetterberg
Summary: In this study, associations between plasma neurodegeneration biomarker concentrations and measures of glymphatic and meningeal lymphatic functions were examined in individuals with neurological disorders. The results suggest that plasma concentrations of neurodegeneration biomarkers are associated with CSF clearance functions.
NATURE COMMUNICATIONS
(2023)
Article
Clinical Neurology
Radek Fric, Geir Ringstad, Per Kristian Eide
Summary: In patients with Chiari malformation type 1 (CMI) and low intracranial compliance (ICC), a preoperative assessment of ICC is performed and patients with low ICC are treated with ventriculoperitoneal shunt (VPS) before foramen magnum decompression (FMD). This study assesses the outcome of patients with low ICC compared to those with high ICC treated with FMD alone.
WORLD NEUROSURGERY
(2023)
Article
Biology
Erik Melin, Geir Ringstad, Lars Magnus Valnes, Per Kristian Eide
Summary: Human MRI data suggests that the parasagittal dura volume may not have a major role in cerebrospinal fluid clearance, but instead functions as a neuro-immune interface. The volume of the parasagittal dura is not influenced by any single variable, but the levels of tracer in the dura are strongly correlated with the levels in the cerebrospinal fluid (CSF) and brain. Furthermore, the peak levels of tracer in the dura occur much later than in the blood, indicating that it is not a major route for CSF clearance.
COMMUNICATIONS BIOLOGY
(2023)
Letter
Multidisciplinary Sciences
Geir Ringstad, Per Kristian Eide
NATURE COMMUNICATIONS
(2023)
Article
Medicine, Research & Experimental
Erik Melin, Are Hugo Pripp, Per Kristian Eide, Geir Ringstad
Summary: This study showed that a CSF tracer substance administered to the lumbar thecal sac can access the parenchyma of the upper cervical spinal cord and brain stem. Clearance from the CSF was delayed in patients with iNPH compared with younger reference patients, suggesting that clearance assessment may be used to tailor intrathecal treatment regimes.
Meeting Abstract
Endocrinology & Metabolism
Alexandra Vallet, Laura Bojarskaite, Daniel Bjornstad, Miroslav Kuchta, Rune Enger, Kent Mardal
JOURNAL OF CEREBRAL BLOOD FLOW AND METABOLISM
(2022)